Migration and Culture

MIGRATION AND CULTURE FRONTIERS OF ECONOMICS AND GLOBALIZATION 8 Series Editors: HAMID BELADI University of Texas at...

Author: Gil S. Epstein

68 downloads 1674 Views 4MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

MIGRATION AND CULTURE

FRONTIERS OF ECONOMICS AND GLOBALIZATION 8

Series Editors: HAMID BELADI University of Texas at San Antonio, USA E. KWAN CHOI Iowa State University, USA

FRONTIERS OF ECONOMICS AND GLOBALIZATION VOLUME 8

MIGRATION AND CULTURE Edited by

Gil S. Epstein Department of Economics, Bar-Ilan University, Ramat Gan, Israel

Ira N. Gang Department of Economics, Rutgers University, New Brunswick, NJ, USA

United Kingdom – North America – Japan India – Malaysia – China

Emerald Group Publishing Limited Howard House, Wagon Lane, Bingley BD16 1WA, UK First edition 2010 Copyright r 2010 Emerald Group Publishing Limited Reprints and permission service Contact: [email protected] No part of this book may be reproduced, stored in a retrieval system, transmitted in any form or by any means electronic, mechanical, photocopying, recording or otherwise without either the prior written permission of the publisher or a licence permitting restricted copying issued in the UK by The Copyright Licensing Agency and in the USA by The Copyright Clearance Center. No responsibility is accepted for the accuracy of information contained in the text, illustrations or advertisements. The opinions expressed in these chapters are not necessarily those of the Editor or the publisher. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN: 978-0-85724-153-5 ISSN: 1574-8715

Emerald Group Publishing Limited, Howard House, Environmental Management System has been certified by ISOQAR to ISO 14001:2004 standards Awarded in recognition of Emerald’s production department’s adherence to quality systems and processes when preparing scholarly journals for print

To Ayelet, Laura my mother and to my precious children Oﬁr, Noga, Inbal, and Eytan, with love, Gil. To Gail and to my precious children, Joshua and Eli, with love, Ira. To our ancestors, who understood the forces of migration and culture.

ABOUT THE SERIES: FRONTIERS OF ECONOMICS AND GLOBALIZATION

This series is aimed at economists and ﬁnancial economists worldwide and will provide an in-depth look at current global topics. Each volume in the series will focus on specialized topics for greater understanding of the chosen subject and provide a detailed discussion of emerging issues. The target audiences are professional researchers, graduate students, and policy makers. It will offer cutting-edge views on new horizons and deepen the understanding in these emerging topics. With contributions from leading researchers, each volume presents a fresh look at today’s current topics. This series will present primarily original works, and employ references appropriate to the topic being explored. Each volume will bring a set of highly concentrated articles that will provide in-depth knowledge to a target audience, while the entire series will appeal to a wide audience by providing them with deeper knowledge on a broad set of emerging topics in the global economy. The Frontiers of Economics and Globalization series will publish on topics such as: – – – – – – – – – – – – – – – –

Frontiers Frontiers Frontiers Frontiers Frontiers Frontiers Frontiers Frontiers Frontiers Frontiers Frontiers Frontiers Frontiers Frontiers Frontiers Frontiers

of of of of of of of of of of of of of of of of

Trade Negotiations Derivative Pricing International Lending and Debt Problems Economics Integration Trade and Environment Foreign Exchange International Finance Growth of Open Economies Futures Pricing International Financial Markets Investment Banking Mergers and Acquisitions Government Policy and Regulations Multi-Sector Growth Models Intellectual Property Rights Fragmentations and Outsourcing Hamid Beladi E. Kwan Choi Series Editors

ABOUT THE VOLUME: MIGRATION AND CULTURE

About the editors Gil S. Epstein is a professor of economics at the Department of Economics at Bar-Ilan University, Israel. He is an associate editor of the Journal of Population Economics and e-conomics. In addition, he is a research fellow in IZA and CReAM. He is the author of Endogenous Public Policy and Contests (with Shmuel Nitzan); his papers on migration, public policy, public choice, political economy, and labor economics have been published in leading journals in economics. Ira N. Gang is a professor of economics at the Department of Economics at Rutgers University. Ira Gang is an associate editor of the Journal of Population Economics and on the board of editors of the Journal of International Trade and Economic Development and one of the founding editors of Review of Development Economics. In addition he is a research fellow in IZA and CReAM. His papers on tax reform, development trade liberalization, corruption, migration, and lobbying have been published in the leading journals in economics. About this volume Culture is not new to the study of migration. It has lurked beneath the surface for some time, occasionally protruding openly into the discussion, usually under some pseudonym. The authors of the chapters in this volume bring culture into the open. They are concerned with how culture manifests itself in the migration process for three groups of actors: the migrants, those remaining in the sending areas, and people already living in the recipient locations. The topics vary widely. What unites the authors is an understanding that though actors behave differently, within a group there are economically important shared beliefs (customs, values, attitudes, etc.), which we commonly refer to as culture. Culture and identity play a central role in our understanding of migration as an economic phenomenon, but what about them matters? Properly, we should be looking at the determinants of identity and the determinants of culture (prices and incomes, broadly deﬁned). But this is not what is done. Usually, identity and culture appear in economics articles as a black box. The authors of the chapters in this volume begin to break open the box.

LIST OF CONTRIBUTORS

Randall K. Q. Akee

Department of Economics, Tufts University, Medford, MA, USA

Arnab K. Basu

Department of Economics, College of William and Mary, Williamsburg, VA, USA

Michele Battisti

Department of Economics, Simon Fraser University, Burnaby, BC, Canada

Howard Bodenhorn

The John E. Walker Department of Economics, Clemson University, Clemson, SC, USA

O¨rn B. Bodvarsson

Department of Economics, Department of Management, St. Cloud State University, St. Cloud, MN, USA; Institute for the Study of Labor (IZA), D-53113 Bonn, Germany

Nancy H. Chau

Department of Applied Economics and Management, Cornell University, Ithaca, NY, USA

Barry R. Chiswick

Department of Economics, University of Illinois at Chicago, IL, USA; IZA-Institute for the Study of Labor, Bonn, Germany

Sarit Cohen-Goldner

Department of Economics, Bar-Ilan University, Ramat Gan, 52900, Israel; Institute for the Study of Labor (IZA), Bonn, Germany

Joseph Deutsch

Department of Economics, Bar-Ilan University, Ramat Gan, 52900, Israel

Don DeVoretz

Department of Economics, Simon Fraser University, Burnaby, BC, Canada

Gil S. Epstein

Department of Economics, Bar-Ilan University, Ramat Gan, 52900, Israel; CReAM, London, UK; Institute for the Study of Labor (IZA), Bonn, Germany

Giovanni Facchini

Dipartimento di Scienze Economiche, Aziendali e Statistiche, Universita´ degli Studi di Milano, Milano, Italy; Department of Economics, Erasmus University Rotterdam, Rotterdam, The Netherlands

xii

List of Contributors

Riccardo Faini (Deceased)

Faculty of Economics, Centre for International and Economic Studies, University of Rome Tor Vergata, Rome, Italy; CEPR-Center for Economic and Policy Research, London, UK; IZA-Institute for the Study of Labor, Bonn, Germany

Michael Fertig

Institut fu¨r Sozialforschung und Gesellschaftspolitik, Ko¨ln, Germany; RWIRheinisch-Westfa¨lisches Institut fu¨r Wirtschaftsforschung, Essen, Germany; Institut zur Zukunft der Arbeit (IZA), Bonn, Germany

Ira N. Gang

Department of Economics, Rutgers University, New Brunswick, NJ, USA; Institute for the Study of Labor (IZA), Bonn, Germany; CReAM, London, UK

T. H. Gindling

Department of Economics, University of Maryland Baltimore County, Baltimore, Maryland, USA

Yitchak Haberfeld

Department of Labor Studies, Tel-Aviv University, Tel Aviv, Israel

Robert Kaestner

Institute of Government and Public Affairs, University of Illinois at Chicago, IL, USA

Martin Kahanec

Department of Economics, Central European University (CEU), Budapest, Hungary Institute for the Study of Labor (IZA), Bonn, Germany

Shirit Katav-Herz

School of Management and Economics, Tel-AvivYaffo Academic College, Tel Aviv, Israel

Neeraj Kaushal

School of Social Work, Columbia University, New York, NY, USA

Melanie Khamis

Institute for the Study of Labor (IZA), Bonn, Germany

Sajal Lahiri

Department of Economics, Southern Illinois University Carbondale, Carbondale, IL, USA

Anna Maria Mayda

Department of Economics and School of Foreign Service, Georgetown University, Washington, DC, USA

Yosef Mealem

The School of Banking and Finance, Netanya Academic College, Netanya, Israel

List of Contributors

xiii

Paul W. Miller

School of Economics and Finance, Curtin University, Perth, WA, Australia

Carolyn M. Moehling

Department of Economics, Rutgers University, New Brunswick, NJ, USA

Kusum Mundra

Department of Economics, Rutgers University, Newark, NJ, USA

Anne Morrison Piehl

Department of Economics, Rutgers University, New Brunswick, NJ, USA

Matloob Piracha

School of Economics, University of Kent, Canterbury, Keynes College, Kent, UK

Sara Z. Poggio

Department of Modern Languages and Linguistics, University of Maryland Baltimore County, Baltimore, Maryland, USA

Francisco L. Rivera-Batiz

Economics and Education, Teachers College and International and Public Affairs, Teachers College, Columbia University, New York, NY, USA

Christoph M. Schmidt

Rheinisch-Westfa¨lisches Institut fu¨r Wirtschaftsforschung, Essen, Germany; Ruhr-Universita¨t Bochum

John G. Sessions

Department of Economics, University of Bath, Bath, UK

Moshe Semyonov

Department of Sociology, Tel-Aviv University, Ramat Aviv, Tel Aviv, Israel

Erez Siniver

School of Economics, The College of Management Academic Studies, Rishon Letzion, Israel

Florin Vadean

Centre for Economic and International Studies University of Rome Tor Vergata, Rome, Italy; School of Economics, University of Kent, Canterbury Kent, UK

Alessandra Venturini

IZA-Institute for the Study of Labor, Bonn, Germany; Department of Economics, University of Torino, Torino, Italy; European university Institute, Florence 50014 Fiesole, Italy

Yan Xing

Department of Sociology, University of Illinois, Chicago, IL, USA

xiv

List of Contributors

Mutlu Yuksel

Department of Public Policy, Dalhousie University, Halifax, NS, Canada

Myeong-Su Yun

Institute for the Study of Labor (IZA), Bonn, Germany; Department of Economics, Tulane University, New Orleans, LA, USA

CONTENTS

LIST OF CONTRIBUTORS

xxvii

PREFACE CHAPTER 1 MIGRATION AND CULTURE Gil S. Epstein and Ira N. Gang 1 2 3 4 5

Enclaves and location choice Production, earnings, and competition Assimilation struggles Family issues and the effects of remittances Selection, attitudes, and public policy Acknowledgment References

PART I:

xi

ENCLAVES AND LOCATION CHOICE

1 2 6 8 12 14 16 16 23

CHAPTER 2 INFORMATIONAL CASCADES AND THE DECISION TO MIGRATE Gil S. Epstein

25

1 Introduction 2 The model 2.1 The background 2.2 A one-signal model 2.3 An illustration 2.4 Multiple signaling 3 Network externalities 4 Concluding remarks Acknowledgment Appendix. Proof of Proposition 1 References

25 28 28 29 31 32 33 39 41 41 43

xvi

Contents

CHAPTER 3 THE MEASUREMENT OF INCOME POLARIZATION BY ETHNIC GROUPS: THE CASE OF ISRAEL POPULATION Joseph Deutsch 1 Introduction 2 Measuring polarization when income groups do not overlap 2.1 The case of two groups of equal size 2.2 The case of three nonoverlapping income groups 3 Measuring polarization when income groups do overlap 4 An empirical illustration 4.1 The case of nonoverlapping groups 4.2 The case of overlapping groups Appendix A. On the concept of Shapley decomposition References

CHAPTER 4 THE EFFECTS OF SCHOOL QUALITY IN THE ORIGIN ON THE PAYOFF TO SCHOOLING FOR IMMIGRANTS Barry R. Chiswick and Paul W. Miller 1 2 3 4

Introduction Methodology Country-level data Empirical assessment 4.1 Aggregate-level analyses 4.2 The role of age at migration 4.3 Reference education, overeducation and undereducation, and PISA scores 5 Conclusion Acknowledgments Appendix A. Deﬁnitions of variables Appendix B. Analyses using the Hanushek and Kimko data B.1 Analyses of Hanushek and Kimko using full sample of 73 countries B.2 Analyses of Hanushek and Kimko indices using subset of countries with both PISA and Hanushek and Kimko measures B.3 Analyses of PISA scores using subset of countries with both PISA and Hanushek and Kimko measures Appendix C. Supplementary results References

45 46 46 46 48 50 53 53 57 62 65

67 68 70 72 78 80 85 86 92 93 93 95 95

96

98 102 102

Contents

CHAPTER 5 DEVELOPMENT AND MIGRATION: LESSONS FROM SOUTHERN EUROPE Riccardo Faini and Alessandra Venturini 1 Introduction 2 The pervasiveness of home bias 3 A simple migration model 3.1 The ‘‘home bias’’ model 3.2 The role of ﬁnancial constraints 4 Trends in Southern European migrations 5 Econometric analysis 5.1 The estimating equation 5.2 The data 5.3 Estimation methods and the results 6 Conclusions and policy implications Acknowledgments Appendix. Data and variables appendix A1 Methodology A2 More information on European migration References CHAPTER 6 GEOGRAPHIC DISPERSION AND INTERNAL MIGRATION OF IMMIGRANTS Neeraj Kaushal and Robert Kaestner 1 Introduction 2 Theoretical considerations 3 Empirical models 3.1 Current location choice 3.2 Internal migration 4 Data 5 Results 5.1 Current location choice: Descriptive analysis 6 Current location choice: Multivariate analysis 6.1 Internal migration: Descriptive analysis 6.2 Internal migration: Multivariate analysis 6.3 Decomposition analysis 6.4 Dispersion due to changes in immigrant composition 7 Conclusion Appendix A. Estimates of the effect of location attributes on the current location choices of immigrants, by country of birth Appendix B. Estimates of the effect of location attributes on the current location choices of immigrants, by country of birth and year

xvii

105 106 107 110 111 113 115 119 119 121 123 129 130 131 131 132 133

137 138 139 142 142 145 146 147 147 149 155 157 160 164 166 168

169

xviii

Contents

Appendix C. Logitistic estimates of the effect of individual characteristics and location attributes on the inter-state migration of foreign-born persons, by country of birth References PART II:

PRODUCTION, EARNINGS AND COMPETITION

CHAPTER 7 UNDERSTANDING THE WAGE DYNAMICS OF IMMIGRANT LABOR: A CONTRACTUAL ALTERNATIVE Christoph M. Schmidt 1 2 3 4 5

Introduction The orthodoxy: Country-speciﬁc human capital The contractual model Optimal contracts Policy implications Acknowledgments References

CHAPTER 8 INTERACTIONS BETWEEN LOCAL AND MIGRANT WORKERS AT THE WORKPLACE Gil S. Epstein and Yosef Mealem

170 171

175

177 177 179 181 184 187 189 189

193

1 Introduction 2 The model 3 Concluding remarks Acknowledgment Appendix References

193 195 200 201 201 202

CHAPTER 9 ETHNIC COMPETITION AND SPECIALIZATION Martin Kahanec

205

1 Introduction 2 The model 2.1 Demand 2.2 Supply 2.3 The equilibrium 3 Specialization of ethnic groups 4 Discussion and conclusions Acknowledgments Appendix A.1 Derivation of equilibrium properties using speciﬁc functional forms References

205 209 209 212 213 215 222 223 224 224 226

Contents

CHAPTER 10 NATIONALITY DISCRIMINATION IN THE LABOR MARKET: THEORY AND TEST + B. Bodvarsson and John G. Sessions Orn 1 Introduction 1.1 Nationality discrimination: Meaning and previous literature 2 A theory of nationality discrimination 2.1 The problem setting 3 A test case: Major League Baseball 3.1 Description of the test case 3.2 Empirical analysis 3.3 Decomposition analysis 4 Concluding remarks Acknowledgments Appendix. Summary of studies providing information about ceteris paribus native/immigrant earnings differences References CHAPTER 11 CULTURE, INVESTMENT IN LANGUAGE AND EARNINGS Erez Siniver 1 2 3 4 5 6

xix

231 232 233 238 238 245 245 246 257 261 262 263 266

269

Introduction Review of the literature Data Method Results The effects of networks on the decision to invest in learning the host country’s language 7 Summary and conclusion References

284 289 290

PART III:

293

ASSIMILATION STRUGGLES

CHAPTER 12 IMMIGRATION: AMERICA’S NINETEENTH-CENTURY ‘‘LAW AND ORDER PROBLEM?’’ Howard Bodenhorn, Carolyn M. Moehling and Anne Morrison Piehl 1 2 3 4 5 6

The ﬁrst major wave of immigration Pennsylvania prison data Immigrant arrivals and prison commitments Aggregate incarceration experience: immigrants and natives Exploring the differences in immigrant and native incarceration Variation across immigrant groups: British, Irish, and Germans

269 271 272 272 274

295

296 301 303 306 311 316

xx

Contents

7 Concluding remarks Acknowledgments References CHAPTER 13 A POLITICAL ECONOMY OF THE IMMIGRANT ASSIMILATION: INTERNAL DYNAMICS Gil S. Epstein and Ira N. Gang 1 Introduction 2 The model 2.1 The absolute ranking 2.2 The relative ranking 2.3 Comparing the investment of effort of the groups under both situations 3 Conclusion Acknowledgment Appendix References CHAPTER 14 ASSIMILATING UNDER CREDIT CONSTRAINTS: PUBLIC SUPPORT FOR PRIVATE EFFORTS Sajal Lahiri 1 2 3 4

Introduction The theoretical framework Public support and private assimilation Conclusion Acknowledgment References

CHAPTER 15 IMMIGRANT NETWORKS AND THE U.S. BILATERAL TRADE: THE ROLE OF IMMIGRANT INCOME Kusum Mundra 1 Introduction 2 Immigrant and the Heckscher–Ohlin model 2.1 Assumptions 2.2 Analysis 2.3 Sufﬁcient condition 3 Immigrants’ income and demand 4 Empirical model 4.1 Data 5 Results

320 320 321

325 325 328 330 331 334 335 336 336 337

341 341 344 348 354 355 355

357 358 360 360 361 363 364 365 367 368

Contents

xxi

6 Concluding remarks Appendix A Appendix B References

370 371 371 371

CHAPTER 16 THE SOCIETAL INTEGRATION OF IMMIGRANTS IN GERMANY Michael Fertig

375

1 2 3 4 5

Introduction Economic and societal integration Empirical strategy and data Results Conclusions Acknowledgments Appendix References

CHAPTER 17 WHO MATTERS MOST? THE EFFECT OF PARENT’S SCHOOLING ON CHILDREN’S SCHOOLING Ira N. Gang 1 2 3 4

Introduction Data Empirical Results Conclusions References

CHAPTER 18 INTERGENERATIONAL TRANSFER OF HUMAN CAPITAL UNDER POST-WAR DISTRESS: THE DISPLACED AND THE ROMA IN THE FORMER YUGOSLAVIA Martin Kahanec and Mutlu Yuksel 1 Introduction 2 Literature review 3 Background on internally displaced people and Roma population in Europe 4 Data and descriptive statistics 5 The results 5.1 Income and employment 5.2 Education and intergenerational transfer of human capital 6 Conclusions and policy recommendations Acknowledgment References

375 378 380 383 391 392 392 399

401 401 403 405 412 413

415 416 417 419 422 432 432 438 441 442 442

xxii

PART IV:

Contents

FAMILY ISSUES AND THE EFFECTS OF REMITTANCES

445

CHAPTER 19 HOUSEHOLD STRUCTURE OF RECENT IMMIGRANTS TO ISRAEL Sarit Cohen-Goldner

447

1 Introduction 2 Data analysis 3 Conclusions Acknowledgments References

447 451 464 465 465

CHAPTER 20 CIRCULAR MIGRATION OR PERMANENT RETURN: WHAT DETERMINES DIFFERENT FORMS OF MIGRATION? Florin Vadean and Matloob Piracha 1 2 3 4 5 6

Introduction Framework for analysis Background and data Econometric speciﬁcation Empirical results Conclusions Acknowledgments References

CHAPTER 21 LABOR MIGRATION, REMITTANCES, AND ECONOMIC WELL-BEING: A STUDY OF HOUSEHOLDS IN RAJASTHAN, INDIA Yan Xing, Moshe Semyonov and Yitchak Haberfeld 1 2 3 4 5

Introduction Labor migration and the role of remittances Remittances in India Data and variables Analysis and ﬁndings 5.1 The multiple use of remittances 5.2 Descriptive overview – comparing households with and without labor migrants 5.3 Multivariate analysis 6 Conclusions

467 467 470 472 478 480 491 494 494

497

497 498 500 500 502 502 504 506 512

Contents

Appendix. Consumption assets and the mean of SDLV of households with previous overseas workers, with current overseas workers and with no overseas workers References CHAPTER 22 PROMOTING THE EDUCATIONAL SUCCESS OF LATIN AMERICAN IMMIGRANT CHILDREN SEPARATED FROM PARENTS DURING MIGRATION Sara Z. Poggio and T.H. Gindling 1 Introduction 2 Literature review 3 Policy recommendations from parents and teachers 3.1 Parents 3.2 Teacher survey 4 Conclusions Acknowledgments Appendix References CHAPTER 23 CULTURAL DIFFERENCES IN THE REMITTANCE BEHAVIOUR OF HOUSEHOLDS: EVIDENCE FROM CANADIAN MICRO DATA Don DeVoretz and Florin Vadean 1 Introduction 2 Theoretical considerations 2.1 The demand system 2.2 Demographic controls, immigration entry and assimilation 2.3 Weak separability 3 Data and descriptive statistics 3.1 Family expenditure survey (FAMEX) 3.2 Prices 4 Empirical results 4.1 Homogeneity and symmetry 4.2 Weak separability 4.3 Expenditure elasticities 4.4 Demographic controls 4.5 Immigration entry and assimilation effects 5 Conclusions Acknowledgments References

xxiii

513 514

517 518 518 523 523 524 530 533 534 539

543 543 545 546 547 548 548 548 551 553 553 555 558 563 570 572 573 573

xxiv

PART V:

Contents

SELECTION, ATTITUDES AND PUBLIC POLICY

CHAPTER 24 FSU IMMIGRANTS IN CANADA: A CASE OF POSITIVE TRIPLE SELECTION? Don DeVoretz and Michele Battisti 1 Introduction 2 Literature review 3 Data 3.1 Data source 3.2 Construction of our dataset 3.3 Data selection 4 Regressions results 4.1 OLS results 4.2 Labour force activity 5 Two-stage models 5.1 FSU vs. Canadian born 5.2 USSR/FSU immigrants versus all immigrants 6 Simulations 6.1 Decomposition analysis 7 Conclusions Acknowledgments Appendix A References CHAPTER 25 WHAT DRIVES IMMIGRATION POLICY? EVIDENCE BASED ON A SURVEY OF GOVERNMENTS’ OFFICIALS Giovanni Facchini and Anna Maria Mayda 1 Introduction 2 Political economy model of migration policy 2.1 What drives individual attitudes toward immigration? 2.2 From individual preferences to migration policy 3 Governments’ views and policies toward immigration 3.1 Governments’ views toward immigration 3.2 Governments’ policies toward immigration 4 Individual attitudes toward immigrants 5 Individual opinions and immigration policy 6 Conclusions Acknowledgments Appendix References

577

579 579 583 587 587 588 588 589 589 593 594 594 596 598 598 601 602 603 604

605 606 607 608 611 614 614 616 622 630 637 638 639 646

Contents

CHAPTER 26 CHANGES IN ATTITUDES TOWARD IMMIGRANTS IN EUROPE: BEFORE AND AFTER THE FALL OF THE BERLIN WALL Ira N. Gang, Francisco L. Rivera-Batiz and Myeong-Su Yun 1 2 3 4 5 6

xxv

649

Introduction The determinants of anti-immigrant attitudes The eurobarometer survey and the empirical model Results Changes in attitudes: a decomposition analysis Summary and conclusions Acknowledgments References

649 653 655 660 667 672 674 674

CHAPTER 27 THE IMPLICATIONS OF SOCIAL NORMS ON IMMIGRATION POLICY Shirit Katav-Herz

677

1 Introduction 2 The model 2.1 The choice of the median voter in a one-period model 2.2 Anti-immigrant actions 2.3 The number of immigrants in the second period 2.4 A far-sighted median voter 3 The distribution of immigration over time 4 Conclusions Acknowledgment Appendix References

677 680 682 683 684 684 685 687 688 688 688

CHAPTER 28 ETHNIC FRAGMENTATION, CONFLICT, DISPLACED PERSONS AND HUMAN TRAFFICKING: AN EMPIRICAL ANALYSIS Randall K.Q. Akee, Arnab K. Basu, Nancy H. Chau and Melanie Khamis

691

1 2 3 4

Introduction Data Empirical methodology and results Conclusion Appendix Acknowledgment References

692 697 701 709 709 714 714

AUTHOR INDEX

717

SUBJECT INDEX

725

PREFACE

For some time we have been concerned with the use of the concept of culture in the economics migration literature and in our own work. Migrants, the places they leave and the places they go, are complex elements that are linked in multiple and multidimensional ways. Yet in our search for simple models, we often summarize this information in single categorical variables such as country of origin, just as in the discrimination literature we often summarize the differences in gender or ethnic/racial achievements solely by a shift in the intercept. Occam’s razor is sometimes too sharp. We know culture matters; but what about it matters? Usually culture appears in economics articles as a black box. While useful, this is not satisfying. This is why when the editors of this series offered us the opportunity to edit this volume, we jumped. We invited a range of scholars who in their work had touched on this theme to submit papers. Each submission was subjected to two referees and our own readings. Each accepted paper was accordingly revised. What we ended up with is an excellent heterogeneous collection of papers that address the links between migration and culture – some directly, some indirectly. This book has two alternative functions. It can be used as a text for a course on migration and culture for advanced undergraduate or graduate students. It can also be used (partly or as a whole) as part of an advanced undergraduate or graduate course in migration or labor economics. This volume consists of 28 chapters, broken into an introduction and ﬁve parts. There is a mixture of both theoretical and empirical studies presented in this volume. The introduction provides a brief overview of the literature and of the chapters presented in this volume. Part I deals with enclaves and the locational choices migrants make, part II deals with production, earnings, and competition between migrants and local population, part III deals with the assimilation struggles among migrants and the reaction of the local population, part IV deals with family issues and the effect remittances have on the population left in the home country, and ﬁnally part V deals with the issues of selection, attitudes of the local population, and public policy. The ideas presented in this book are new and fresh and are those of the authors of each chapter. We thank the series editors, Hamid Beladi and E. Kwan Choi, for offering us this opportunity; and the people at Emerald for helping us carry it out. We thank, of course, the contributors to this volume and those who helped us referee and comment. We especially thank Geoffrey Williams, a

xxviii

Preface

rare gem of a graduate student for taking on the task of reading every chapter and offering unique and superior insights. Part of this project was completed while we were on one of our frequent visits to the IZA. We are most grateful to the institute for their warm hospitality and supporting research environment.

CHAPTER 1

Migration and Culture Gil S. Epsteina,b,c and Ira N. Gangb,c,d a

Department of Economics, Bar-Ilan University, Ramat Gan, 52900, Israel E-mail address: [email protected] b Institute for the Study of Labor (IZA), Bonn, Germany c CReAM-Center for Research and Analysis of Migration, London, UK d Department of Economics, Rutgers University, New Brunswick, New Jersey, 08901-1248, USA E-mail address: [email protected]

Abstract Culture is not new to the study of migration. It has lurked beneath the surface for some time, occasionally protruding openly into the discussion, usually under some pseudonym. The authors bring culture into the open. They are concerned with how culture manifests itself in the migration process for three groups of actors: the migrants, those remaining in the sending areas, and people already living in the recipient locations. The topics vary widely. What unites the authors is an understanding that though actors behave differently, within a group there are economically important shared beliefs (customs, values, attitudes, etc.), which we commonly refer to as culture. Culture and identity play a central role in our understanding of migration as an economic phenomenon; but what about them matters? Properly, we should be looking at the determinants of identity and the determinants of culture (prices and incomes, broadly deﬁned). But this is not what is done. Usually identity and culture appear in economics articles as a black box. Here we try to begin to break open the black box. Keywords: Migration, Culture, Discrimination, Assimilation, Mobility Jel classiﬁcations: F22, O15, R23, J61, J71 Migrants are quite diverse. The discussion here is on the distinctions in culture among migrants, the families they left behind, and the local population in the migration destination. The new interactions directly affect all three groups. Assimilation is one result; separation is also a possibility. Location choice, workplace interaction, enclave size, the opportunity for the migrant obtaining credit in their new country, the local

Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008007

r 2010 by Emerald Group Publishing Limited. All rights reserved

2

Gil S. Epstein and Ira N. Gang

population’s reaction to migrants, the political culture of the migrants and local population, links to the country-of-origin, and the economic state of the host country, all contribute to the classic conﬂict between assimilation and separation. Papers examining the working of the assimilation process on the migrants themselves, on the local population, on the families left at the home country, and others can be divided into ﬁve nonexclusive areas: (1) enclaves and location choice; (2) production, earnings, and competition; (3) assimilation struggles; (4) family issues and the effects of remittances; (5) selection, attitudes, and public policy. 1. Enclaves and location choice A characteristic of international migration is the clustering of immigrants in ethnic communities. Prominent examples are the concentration of Turks in Germany, Tamils in Switzerland, Moroccans in the Netherlands and Belgium, Italians in Argentina, Greeks in Australia, and Ukrainians in Canada. Clustering may be very narrow, such as when immigrants from a town or region are concentrated in a speciﬁc foreign town or region. For example, Macedonians from Skopje have come to make up a notable part of the population of Gothenburg, Sweden. In the United States, noticeable clusters of Mexican immigrants exist in California, Texas, Florida, and Chicago. Three-fourths of migrants from Guanajuato, the Mexican state with the highest emigration rate to the United States, go to California or Texas. The prevailing explanation for immigrant clusters is the existence of beneﬁcial network externalities when previous immigrants provide shelter and work, assistance in obtaining credit, and/or generally reduce the stress of relocating to a foreign culture (see Gottlieb, 1987; Grossman, 1989; Marks, 1989; Church and King, 1983; Carrington et al., 1996; Chiswick and Miller, 1996; Epstein, 2003; Munshi, 2003). Ethnic networks, however, might also be associated with negative externalities. Disadvantageous network externalities may arise if immigration is subject to adverse selection, or if increase in immigrant concentration increases competition for jobs and lowers immigrants’ wages. Under certain conditions the tendency to cluster may lower incentives to learn the language of the host country, which in turn may ‘‘trap’’ migrants in poverty (Bauer et al., 2009). These negative network externalities limit the beneﬁts immigrants can obtain from clustering. A growing literature investigates the determinants of location choice by immigrants. The ﬁrst signiﬁcant study on this, Bartel (1989), ﬁnds that post-1964 migrants to the United States tend to locate in cities with a high concentration of immigrants of similar ethnicity. She further shows that highly skilled migrants are less geographically concentrated and rely less on the location of fellow compatriots. Similarly, Jaeger (2007), who differentiates between immigrants of different admission statuses, ﬁnds that immigrants tend to locate where former immigrants of the same ethnicity are concentrated.

Migration and Culture

3

Migration enclaves may be naturally limited in size. Migrants often choose to live together in enclaves, and to carry out a relatively large share of their transactions with other parties of the same enclaves – people who share a language, origin, and history. Such an enclave gives the migrants a clear beneﬁt, particularly if they are more likely to encounter a cooperative environment in such a setting. However, enclaves can feed xenophobia and make natives hostile toward the migrants. Such hostility can be expected to increase as the minority grows in size. This mitigates the beneﬁt from the enclave, as the hostility harms the migrants. Thus, we expect to see numerous enclaves of migrants spread throughout the receiving country rather than concentrated in a single location (Weiss and Rapoport, 2003). Migrant and local populations interact. Each can invest in activities promoting or hindering assimilation. Migrants may want to assimilate, or they may want to hold onto their cultural identities. The local population may welcome or not. A major site for these interactions is within the ﬁrm – the proverbial ‘‘shop ﬂoor.’’ As with enclaves, here also the size of the groups is important. Migrants consider several factors in making their decisions about where to move, including the clustering of compatriots and similar folk in various localities. Ties of kinship, friendship, and village link migrants, former migrants, and nonmigrants in the home and host country. Stock factors measure the degree to which migrants may view a location as (ethnically) hospitable and the availability of information about speciﬁc locations. Stock factors may have an ethnic goods component and include village migration history. Flow factors measure the tendency of migrants to follow the paths of very recent migrants from their own villages. These factors offer different information to a potential migrant. The ethnic goods component sends signals to the migrant about the possibility of living in a culturally similar environment, that is, speaking one’s native language, listening to his music, reading his own newspapers, and eating ethnic food. The ethnic goods factor reduces the monetary and psychic costs of migrating. The village migration history component largely captures information about the host region received in the home village. This includes, for example, information on the labor and housing market, and information on speciﬁc employers in a region. In addition, the migrant may be able to count on contacts in a speciﬁc location established by former migrants from the same village. This factor reﬂects the probability of receiving help from compatriots. The ﬂow factor represents potential herd behavior by migrants, a sort of ‘‘peer emulation effect.’’ Following the argument by Epstein (2010), migrants may choose a location on the supposition that recent migrants had information that he does not have. Until the appearance of the paper by Polachek and Horvath (1977) much of migration theory treated migration as an individual investment

4

Gil S. Epstein and Ira N. Gang

decision. Family members other than the household head are not always explicitly considered. However, other members are clearly inﬂuential in migration decisions. Polachek and Horvath (1977) established the foundations for models of location choice that take into consideration all the different type of considerations. They do so by adopting a life-cycle approach used in human capital theories of earnings accumulation, accounting for household considerations in both a general theoretical and empirical model. More importantly, migration is analyzed within a nonstochastic framework and remigration is endogenously explained. Bauer et al. (2009) examine the determinants of a current migrant’s location choice emphasizing the relative importance and interaction of migrant stocks and ﬂows. They show that both stocks and ﬂow have signiﬁcant impacts on the migrant’s decision of where to locate. The signiﬁcance and size of the effects vary according to legal status and whether the migrant is a ‘‘new’’ or a ‘‘repeat’’ migrant. A different aspect of locational outcomes considers how extensive is polarization based on wages and other economic indicators. Deutsch (2010) takes a multidimensional approach to the measurement of wellbeing, checks whether there has been a change in the degree of (group) polarization in the distribution of well-being in Israel. Deutsch (2010) shows how it is possible to decompose by population subgroups the polarization index. This polarization index is related to the Gini index and its components so that previous results on the decomposition of the Gini index may be incorporated. Two main cases are examined, that of nonoverlapping groups and overlapping groups. Using Israeli data he shows decreasing polarization from 1990 to 2004. Polarization has many aspects; one is education. The payoff to schooling among the foreign-born in the United States is only around one-half of the payoff for the native-born. Chiswick and Miller (2010) examines whether this differential is related to the quality of the schooling immigrants acquired abroad. They use the over-education/required education/under-education speciﬁcation of the earnings equation to explore the transmission mechanism for the origin-country school quality effects. They also assess the empirical merits of two alternative measures of the quality of schooling undertaken abroad. Their results suggest that a higher quality of schooling acquired abroad is associated with a higher payoff to schooling among immigrants in the U.S. labor market. This higher payoff is associated with a higher payoff to correctly matched schooling in the United States, and a greater (in absolute value) penalty associated with years of under-education. A set of predictions is presented to assess the relative importance of these channels, and the over-education channel is shown to be the more inﬂuential factor. This channel is linked to greater positive selection in migration among those from countries with better quality school. In other words, it is the impact of origin-country school quality on the immigrant selection process, rather than the quality

Migration and Culture

5

of immigrants’ schooling per se, which is the major driver of the lower payoff to schooling among immigrants in the United States. Another aspect of locational choice is migrant mobility. Policy-makers in OECD countries appear to be increasingly concerned about growing migration pressure from developing countries. At the same time, at least within Europe, they typically complain about the low level of internal labor mobility. Faini and Venturini (2010) try to shed light on the issues of both internal and external labor mobility. They investigate the link between development and migration and argue, on both theoretical and empirical grounds, that it is likely nonlinear. More precisely, they ﬁnd that, in a relatively poor sending country, an increase in income has a positive impact on the propensity to migrate, even if we control for the income differential with the receiving country, because the ﬁnancial constraint of the poorest become less binding. Conversely, if the home country is relatively better off, an increase in income may be associated with a fall in the propensity to migrate even for an unchanged income differential. Econometric estimation for Southern Europe over the period 1962–1988 provides substantial support to this approach. They estimate ﬁrst the level of income for which the ﬁnancial constraint is no longer binding, around $950, and then the level of income for which the propensity to migrate declines, which is around $4,300 in 1985 prices. They, therefore, predict a steady decline in the propensity to migrate from Southern European countries. Similarly, their results highlight the possibility that the pressure to migrate from Northern African countries and other developing countries may increase with further growth. Taking a broader view Kaushal and Kaestner (2010) study the correlates of immigrant location and migration choices to address the following questions: What location-speciﬁc, economic, and demographic factors are associated with these choices? Does the inﬂuence of these factors differ by immigrant characteristics? What explains the observed increase in immigrant geographic dispersion during the 1990s? Their analysis suggests that: (1) there is signiﬁcant heterogeneity in the correlates of immigrant location and migration choices; associations vary by immigrant birthplace, age, gender, education, and duration of residence in the United States. (2) Economic factors are, for the most part, weakly associated with immigrant location decisions. (3) Immigrants appear to be more attracted to states with large (growing) populations; less attracted to states with a high proportion of other foreign-born persons; more attracted to states with high unionization, and less attracted to states with high crime. (4) The association between location-speciﬁc characteristics and immigrant location choices changed between 1990 and 2000 for some immigrant groups and this explains most of the increase in geographic dispersion during the nineties. In contrast, changes in location attributes and changes in immigrant composition explain relatively little of the increase in dispersion.

6

Gil S. Epstein and Ira N. Gang

2. Production, earnings, and competition The analysis of immigrants’ contributions to the economy has concentrated on immigrants’ impact on native’s employment and wages (Baldwin-Grossman, 1982; Gang and Rivera-Batiz, 1994; Friedberg and Hunt, 1995; Borjas, 2003; Card, 2005; Ottaviano and Peri, 2008). Immigration affects relative supply of workers with different characteristics and effects workers differently depending on their characteristics. The debate has generally turned on the degree of substitutability or complementarity of immigrants and the native-borns: if immigrants tend to cluster into jobs requiring mostly manual work and little education or experience, and the native-borns hold jobs requiring higher levels of education and/or experience, how would increased immigration affect the wages of the native-borns? The answer is, of course, directly related to whether low-skilled and high-skilled labors are substitutes or complements. This is very nicely laid out in Bodvarsson and Van den Berg (2009). The chapters in this book push beyond the scope of the received tradition. The classic confrontation between immigrants and the local population takes place in the labor market. While many papers deal with labor market concerns, the chapters in this book tackle key issues head on, providing new insights to well-worn subject matter. For example, it is very clear that otherwise similar-looking immigrants and locals earn different amounts and have different jobs. The question is whether these differences constitute discrimination or is something else going on. If it is discrimination, what is at the root of it? In part, immigrant earnings are the outcome of the friction between the migrants and the local population. The willingness of the local population to accept the migrants also plays a role here. In terms of assimilation, the effect of the borrowing constraint facing new immigrants on the process of their assimilation in the new society is important. Those who succeed enjoy a higher level of productivity and therefore wages in the future. The level of investment is endogenously determined. Thus, an important assimilation issue is the possibility of borrowing. On this issue, migrants and the local population differ. Empirical evidence on the labor market performance of immigrants shows that migrant workers suffer from an initial disadvantage compared to observationally equivalent native workers, but that their wages subsequently tend to increase faster than native earnings. Economists usually explain these phenomena by spot markets for labor and investments into human capital. By contrast, Schmidt (2010) proposes a contract theoretic model. This alternative has important implications for integration policy, since it suggests investing into the transparency of foreign educational credentials. Also contrasting human capital theory, the model suggests that permanent migrants never earn higher wages than equally skilled temporary migrants.

Migration and Culture

7

One should not ignore the interaction between the local and foreign workers. Epstein and Mealem (2010) consider the interaction between local workers and migrants in the production process of a ﬁrm. Both local workers and migrants can invest effort in assimilation activities in order to increase the assimilation of migrants into the ﬁrm, and by doing so increase their interaction and production activities. They consider the effect, the relative size (in the ﬁrm) of each group, and the cost of activities has on the assimilation process of the migrants. One of the outcomes of this model is specialization in production. If this is the outcome then the question that comes to mind is: are ethnic specialization and thus a downward sloping labor demand curve fundamental features of labor market competition among ethnic groups? In a general equilibrium model, Kahanec (2010) argues that spillover effects in skill acquisition and social distances between ethnic groups engender equilibrium regimes of skill acquisition that differ in their implications for ethnic specialization. Speciﬁcally, fundamental relationships through which relative group sizes determine whether ethnic specialization arises and to what degree are established. Thus, his paper theoretically justiﬁes a downward sloping labor demand curve and explains why some ethnic groups earn more than others, ethnic minorities underperforming or outperforming majorities. As presented above, migrants are many times paid differently than the local population. Bodvarsson and Sessions (2010) focus on immigrant workers paid differently than their equally productive native-born counterparts (‘‘nationality discrimination’’). Constructing a theory and test of nationality discrimination is particularly challenging because: (a) foreign- and native-born workers in the same occupation are very likely to be imperfect substitutes in production, owing to the former group’s imperfectly transferable human capital; but (b) the literature offers models only where majority and minority workers are perfect substitutes. In the theory section, a generalized Leontief production function where native and immigrant workers are distinct inputs is articulated. In the empirical section, a U.S. test case is available: Major League Baseball (MLB). The dataset consists of 1,093 hitters and 1,204 pitchers for four seasons during 1992–1998, a period during which the industry expanded. Salary, experience, player performance, and team performance data come from the Lahmann Baseball Database and race and nationality are inferred from Topps baseball cards. Estimates of nationality discrimination against immigrant players in both job categories are obtained. Culture is intimately linked to pecuniary incentives – to earnings and productivity. This is brought out by Siniver (2010) who shows immigrants do, in fact, respond to economic incentives in acquiring proﬁciency in the language of the host country, particularly immigrants with 13þ years of schooling.

8

Gil S. Epstein and Ira N. Gang

3. Assimilation struggles Some migrants stay in their new country and some go back home. Those who return home bring with them experience and, perhaps, higher human capital. To what extent do the socioeconomic characteristics of circular/repeat migrants differ from migrants who return permanently to their home country after their ﬁrst trip (i.e., return migrants)? What determines each of these distinctive temporary migration forms? What happens to those who do not return, though they continue sending remittances home? What effect does this have on the migrants and those left at home? Minority ethnic group participation in labor markets is quite complex and in many ways different from that of citizens belonging to a nation’s majority ethnicity. Studies of minorities around the world show, with few exceptions, that they tend to earn wages substantially below those of comparable majority workers (Altonji and Blank, 1999; Blau and Kahn, 1997, 2006, 2007; Smith and Welch, 1989; Bhaumik et al., 2006). Partly, this reﬂects a failure on the part of the minority group to undertake the effort to assimilate with the majority (Constant et al., 2009). ‘‘Lack of effort’’ can arise from the desire to maintain a cultural heritage or separate identity that would be lost or reduced if the group assimilated. The failure to take active steps to assimilate can also arise in the face of high adjustment costs, such as inadequate language skills, intergenerational familial conﬂicts, and, in the case of immigrants, lack of knowledge about the host country labor market (Chiswick and Miller, 1995, 1996; Bauer et al., 2005). Yet for immigrants and their descendants, as length of time in the host country increases, assimilation generally creeps in and various immigrant labor market indicators approach those of comparable majority workers. On occasion, minority workers outperform majority workers (Chiswick, 1977; Deutsch et al., 2006). Efforts made to assimilate, and time, are two elements working to bring minorities into line with the majority. A third element, the degree to which the majority welcomes the minority, also plays a role. Often, the majority is less than welcoming, blaming the minority for depressing wages and displacing majority workers – that is, causing majority unemployment. This presumption has very strong policy implications and is implicit, for example, in the calls for increased regulation of immigration heard worldwide. Yet, there is mixed evidence on the impact of minorities on majority wages and employment – it depends on whether they are substitutes or complements with respect to the skills and other attributes they bring to the labor market (Gang and Rivera-Batiz, 1994; Gang et al., 2002). Whether minorities actually lower the wages and increase employment or not, the perception exists that they do so. Because of this perception, the majority may take active steps to discourage minority assimilation – discrimination, isolation, and so on.

Migration and Culture

9

Often the efforts of the minority and the majority are mediated through political institutions. These institutions exist in both the minority and majority worlds. They could be, for example, political parties, trade organizations, unions, or thugs. These are organizations that are able to overcome the free-rider problem individual members of each group have in moving from the actions they desire to take, to actually taking the actions. Yet, while an organization’s purpose may be to represent the members of their group, the interests’ of the organization and that of its members do not always coincide. The work here adds to the blossoming literature on majority–minority conﬂict and resolution, assimilation, and the reestablishment of cultural identity (see, e.g., Alesina and La Ferrara, 2000; Anas, 2002; Bisin and Verdier, 2000; Dustmann et al., 2004; Kahanec, 2006, and Lazear, 1999). Epstein and Gang (2009) are interested in why minorities are so often at a disadvantage compared to the majority, the circumstances under which their status changes or stagnates over time, and role public policy can play. Assimilation efforts by the minority, harassment by the majority and time are the three elements that determine how well the minority does in comparison to the majority. They examine the consequences for these increases in the numbers of members of the minority, time, and the role of the political entity. They construct a model in which there are four actors: the members of the majority and the organization that represents them, and members of the minority and the organization that represents them. Over time, the political entity representing the minority and the members of the minority exhibit different interests in assimilating and in maintaining their cultural identity. They discuss how this affects the minority’s position over time and discuss the public policy implications of the model. Some view migration and crime as dependent. The Bodenhorn et al. (2010) study provides a fresh look at the question of immigration and crime by looking at mid-19th century data created from the records of Pennsylvania’s state prisons from the 1830s to the 1870s. These records provide information on the birthplace, age, prior occupation, country of conviction, crime, and sentence of all individuals entering the prisons. With these data we can examine the share of immigrants in prison commitments as well as in the prison population on a given date. These data, when combined with data on the general population, allows them to determine whether immigrants were disproportionately incarcerated in general and for violent crimes in particular, and whether immigrant incarceration patterns changed over time as immigrants assimilated to life in the United States. The use of micro-level data that allows analysis by type of crime and age provides a much tighter and much richer understanding of immigrant participation in crime. Impressions of immigrants as a source of violence and disruption are longstanding. Furthermore, they underlie many of the theories of culture conﬂict and

10

Gil S. Epstein and Ira N. Gang

assimilation. Modern empirical methods and detailed population data allow revisiting these age-old research questions with a sharper focus. Within immigrant society there is often a conﬂict between those arguing for assimilation and those demanding an independent identity for the group. Of course there are many shades to this discussion; immigrant societies are multilayered and multidimensional with many viewpoints. One point of view may come into conﬂict with others because of the development of rivalrous strategies, at least partly overlapping followers, and/or the necessity of laying claim to having the bigger impact. Supporters of each point of view invest resources and effort into convincing the general body of immigrants of the virtue of their point of view and, therefore, having an effect. Epstein andGang (2010) develop economic theory that considers how such a competition affects the resources invested by the supporters and how beneﬁcial it is to the immigrant group. Fertig (2010) investigates whether and to what extent immigrants in Germany are integrated into German society by utilizing a variety of qualitative information and subjective data collected in the 1999 wave of the German Socio-Economic Panel (GSOEP). To this end, leisure-time activities and attitudes of native Germans, ethnic Germans, and foreign immigrants of different generations are compared. The empirical results suggest that conditional on observable characteristics the activities and attitudes of foreign immigrants from both generations differ much more from those of native Germans than the activities/attitudes of ethnic Germans. Furthermore, the attitude of second-generation immigrants tends to be characterized by a larger degree of fatalism, pessimism, and self-doubt than those of all other groups, although their activities and participation in societal life resemble more those of native Germans than those of their parents’ generation. Whose role in helping the second generation to assimilate and get along in their new country is more important, the mother’s or the father’s? Gang (2010) examines the differential effects of mother’s schooling and father’s schooling on the acquisition of schooling by their offspring. The context is ‘‘cross-cultural,’’ comparing results across three countries: Germany, Hungary, and the Former Soviet Union (FSU). Within these countries, it looks at differences by gender and by different ethnic subgroups. The evidence is, generally, that father’s schooling is more important than mother’s schooling, but this does vary by ethnic group. Moreover, mother’s schooling plays a relatively larger role for females. Kahanec and Yuksel (2010) investigate the effects of vulnerability on educational outcomes in Croatia, Bosnia and Herzegovina, Montenegro, and Serbia using a unique 2004 UNDP dataset. Treating the collapse of the former Yugoslavia as a natural experiment, they compare educational achievement and intergenerational transfer of human capital for three groups that have been differently affected by the wars and postwar distress: the majority as the benchmark, the ex-ante and ex-post vulnerable

Migration and Culture

11

Roma people, and the ex-ante equal but ex-post vulnerable internally displaced people (IDPs). Their ﬁndings reveal signiﬁcant negative effects of vulnerability on educational attainment. IDPs seem to be more negatively affected than Roma and both groups exhibit signiﬁcant inertia in intergenerational transfer of human capital. They ﬁnd evidence that this inertia is stronger for the Roma. Their ﬁndings highlight the need for policies that not only tackle vulnerability as such but also address the spillover effects of current vulnerability on future educational attainment. In the struggle for assimilation credit markets may play an important role. Lahiri (2010) examines the effect of borrowing constraints facing new immigrants in the process of their assimilation in their new society. He does so in two-period model. In period 1, the immigrants invest, with some costs to them, in trying to assimilate. The probability of success in this endeavor depends on the amount invested and on the level of the provision of a ‘‘public’’ good paid for by lump-sum taxation of the ‘‘natives.’’ Those who succeed enjoy a higher level of productivity and therefore wages in period 2. The level of investment is endogenously determined. Given this framework, Lahiri (2010) characterize the optimal level of the public good provision. This is done under two scenarios regarding the credit market facing new migrants. In the ﬁrst, they can borrow as much as they want in period 1 at an exogenously given interest rate. In the second scenarios, there is a binding borrowing constraint. Lahiri (2010) compares the equilibrium level of ‘‘assimilation’’ under the two scenarios. There is a well-established high quality literature on the role of networks, particularly ethnic networks, in international trade. Ethnic networks are a way of overcoming informal barriers (information costs, risk, and uncertainty) to trade by building trust and substituting for the difﬁculty of enforcing contracts internationally. Networks form between migrants and natives in the host country and between migrants and their home country. Ethnic networks exist when assimilation is not complete. Epstein and Gang (2006) consider the struggle of migrants to assimilate and, at the same time, the struggle of the local population to prevent such assimilation. These activities affect trade possibilities. Moreover, they show that it may well be in the interest of migrants who specialize in trade to, at some point in time, turn from investing in assimilation activities and instead invest in anti-assimilation activities in order to preserve immigrants’ preferences for home-country goods. There is increasing evidence in empirical trade that the immigrant population provides the social and co-ethnic networks that facilitate trade with their home country by removing some informal trade barriers and lowering transactions cost to trade. Immigrants’ carry home-country information that helps in matching buyers and sellers and enforcement of trading contacts (information effect) and immigrants affect imports by demanding goods from their home countries (demand effect). Usually, the size of immigrant

12

Gil S. Epstein and Ira N. Gang

stock – both older cohorts and new entrants – captures network size. However, as immigrants stay longer in their host country their information and demand effects may weaken or strengthen. This varies across immigrant groups and type of goods. Trade ﬂows between the host and the home country change in response. Mundra (2010) focuses on the role of immigrants’ economic assimilation on the U.S. bilateral trade using a panel data for 63 trading partners as well as immigrant sending countries over the period 1990–2000. She examines whether the immigrants’ assimilation effect on trade varies across the homogenous goods and differentiated products.

4. Family issues and the effects of remittances Migration is not generally a purely individual decision; most frequently it takes place in a family context. One or two members of the family migrate; the others stay in their home country. For example, for those from Central America and Mexico it is not uncommon for a mother or father (or both) to migrate to the United States and leave their children behind. After the parent(s) have achieved some degree of stability in the United States, the children follow. There are many important questions. Are children separated from parents during migration more likely to fall behind others their age in school? Are they more likely to drop out of high school? Does the impact of separation for children differ when separated from their mothers or fathers? Migration may change family structure in the host country as they interact with the local economy and new culture. This may have strong and important effects on migrant identity and socialization and their willingness to assimilate (Gang and Zimmermann, 2000). The growth perspectives of European Union member countries are seen to be crucially related to the challenge of mobilizing people to work. One issue is that noneconomic migrants have more difﬁculties in economic performance and labor market integration, and are a larger potential burden to the social security systems than economic migrants. Recent work in Denmark and Germany (see Tranaes and Zimmermann, 2004; Schultz-Nielsen and Constant, 2004; Constant and Zimmermann, 2005; Constant et al., 2009) provides new evidence indicating that an ever-rising number of immigrants are unavailable to the labor force. Instead, migrants arrive as refugees, asylum seekers, or for family reuniﬁcation purposes. Differences in labor market attachment might be due to differences in individual characteristics across ethnicities and within ethnicities. The effect of migration and remittances on nonmigrating family members has long attracted attention. Migration and remittances can increase investment in human and physical capital (Cox Edwards and Ureta, 2003; Hildebrand and McKenzie, 2005; Mesnard, 2004), reduce poverty and alter inequality in the home country (Adams, 1992;

Migration and Culture

13

Taylor and Wyatt, 1996). It can also induce chain migration (Dimova and Wolff, 2009). Recent research links migration, transfers, and child labor, showing in the aftermath of migration and the transfers sent by emigrating parents may enable the children and other family members to stop working (Epstein and Kahana, 2008). In recent years, both the structure of families and household composition changed dramatically. For example, more and more young people leave the house of their parents before the establishment of their own family, more and more young couples live together without marriage, etc. Cohen Goldner (2010) explores immigrant family structure in Israel and follows the dynamics of immigrants’ households as a function of time in the new country and labor market performance. Upon arrival a typical immigrant household consists of more than one family. This pattern reﬂected the economic constraints that immigrant faced upon arrival and the need to save additional costs, as well as a sociological need of immigrants to ‘‘stick together.’’ However, as immigrants are integrated in the labor market and time passes, the share of households consisting of more than one family diminishes. To what extent do the socioeconomic characteristics of circular/repeat migrants differ from migrants who return permanently to the home country after their ﬁrst trip (i.e., return migrants)? What determines each of these distinctive temporary migration forms? Piracha and Vadean (2010) using Albanian household survey data and both a multinomial logit model and a maximum simulated likelihood (MSL) probit with two sequential selection equations ﬁnd that education, gender, age, geographical location, and the return reasons from the ﬁrst migration trip signiﬁcantly affect the choice of migration form. Compared to return migrants, circular migrants are more likely to be male, have primary education, and originate from rural, less developed areas. Moreover, return migration seems to be determined by family reasons, a failed migration attempt and also the fulﬁllment of a savings target. Remittances have long been viewed as a means to combat poverty, to improve consumption, to raise standard of living. Remittances, however, can also enable investment in human capital resources (especially education) of the next generation. Haberfeld et al. (2010) examines the impact of remittances sent by labor migrants from India on the standard of living (as a proxy of consumption) and on the education of young children (as a proxy of investment in human capital) on nonmigrating family members. The analysis is conducted on a randomly selected representative sample of households in Rajasthan. Three types of households are distinguished: 575 having labor migrants, 162 without current migrants, and 232 not having migrants at present but sent migrants in the past. Analysis of the data reveals meaningful differences among the types of households. Those having current labor migrants are characterized by the highest standard of living but at the same time by a low level of children’s education. Further analyses suggest that remittances are likely

14

Gil S. Epstein and Ira N. Gang

to increase consumption and improve standard of living but have very little effect on children’s education. Earlier research found that children separated from parents during migration are more likely to lag behind others their age in school and are more likely to drop out of high school. The negative impact of separation during migration on educational success is largest for children separated from their mothers (in contrast to fathers), for those whose parents have lived in the United States illegally, and for those who reunited with parents as teenagers (rather than at younger ages). Poggio and Gindling (2010) suggest public policies to help immigrant children separated from parents during migration to succeed in U.S. schools. The policies are based on focus group discussion with parents separated from their children during migration, interviews with psychologists and school administrators, and an online survey of elementary and high school teachers. DeVoretz and Vadean (2010) analyze the effect of cultural differences among ethnic groups on the remittance behavior of native and immigrant households in Canada. In contrast to the literature that examines remittance motivation in the framework of extended family agreements, they embed remittances in a formal demand system, suggesting that they represent expenditures on social relations with relatives and/or friends and contribute to membership in social/religious organizations, respectively. The results indicate strong ethnic group cultural differences in the remittance behavior of recent Asian immigrant households and highlight the importance of differentiating with respect to cultural background when analyzing the determinants of remittances.

5. Selection, attitudes, and public policy Cost, beneﬁts, and the local population’s reaction affect public policy. We see this in the different policies toward migration as reported by governments to the United Nations Department of Economic and Social Affairs between 1976 and 2007. Preliminary evidence shows that most governments have policies aimed at either maintaining the status quo or at lowering the level of migration. The UN dataset also allows us to document variation in migration policies over time and across countries of different regions and incomes. Battisti and DeVoretz (2010) investigate the economic performance of immigrants from the FSU countries in Canada. The contribution of their paper lies in its use of a natural experiment to detect possible differential labor market performances of Soviet immigrants prior to and after the collapse of the Soviet Union. In short, the collapse of the FSU allows an exogenous supply change in the number and type of FSU immigrants potentially destined to enter Canada. For this purpose, Census micro-level data from the 1986, 1991, 1996, and 2001 Canadian Census are utilized to

Migration and Culture

15

estimate earnings and employment outcomes for pre- and post-FSU immigrants. The ﬁrst goal of Facchini and Mayda (2010) is to measure the restrictiveness of policies toward migration as reported by governments to the United Nations Department of Economic and Social Affairs between 1976 and 2007. Preliminary evidence shows that most governments have policies aimed at either maintaining the status quo or at lowering the level of migration. The UN dataset also allows them to document variation in migration policies over time and across countries of different regions and income levels. Finally, it makes it possible to examine patterns in different aspects of destination countries’ migration policies, such as policies toward family reuniﬁcation, temporary versus permanent migration and highly skilled migration. This analysis leads to an investigation of the politicaleconomy determinants of destination countries’ migration policy. Facchini and Mayda (2010)’s goal is to develop a framework in which voters’ attitudes represent a key component and to examine the link between these attitudes and governments’ policy decisions. To that end, they merge the information contained in the UN migration-policy dataset with crosscountry data on individual attitudes toward immigrants. They use data on public opinion from the International Social Survey Programme, National Identity Module, for the years 1995 and 2003. The merged datasets allow us to investigate whether – within a median voter framework (Benhabib, 1996; Ortega, 2005; Facchini and Testa, 2009) – voters’ migration attitudes are consistent with migration-policy decisions as reported by governments. The link between ethnic conﬂicts and international trafﬁcking is an issue that has recently received a surge in international attention. The main argument is that internal conﬂicts encourage the internal displacement of individuals from networks of family and community, and their access to economic and social safety nets. These same individuals are particularly vulnerable to being trafﬁcked, by the hopes of better economic prospects elsewhere. Akee et al. (2010) take this link between ethnic fragmentation and international trafﬁcking to the data for the ﬁrst time, making use of a novel dataset of international trafﬁcking. They conduct a two-stage estimation, which highlights the ultimate impact of ethnic fragmentation and conﬂict on international trafﬁcking, both directly and indirectly, through their impacts on the scale of internal displacements. From a different angle, Gang et al. (2010) explores the determinants of the attitudes of European citizens toward non-European Union foreigners using samples from the Eurobarometer Surveys. They carry out a probit analysis of some of the key factors inﬂuencing the attitudes of European Union citizens toward foreigners and their changes over time. They study the roles of labor market, concentration of immigrants in neighborhoods, racial prejudice, and education on anti-foreigner sentiment. Implementing the Oaxaca-type decomposition analysis based on probit estimates show a generally rising trend toward greater racial prejudice, and the decline in the

16

Gil S. Epstein and Ira N. Gang

strength of educational attainment in reducing negative attitudes toward foreigners, contributes to the increased anti-foreigner attitudes. Along the same line, Katav-Herz (2010) examines how social norms affect a local population’s attitudes toward immigration. A model is set out showing how a trade-off can arise between the contribution of immigration to the welfare of the local population and the concerns about changes in social norms. The chapter addresses three questions. The ﬁrst question concerns the determination of immigration policy through majority voting when a population differs in attitudes to changes in social norms. The second question concerns how social norms can impede the realization of the beneﬁts of immigration as a solution for ﬁnancing intergenerational transfers to retired people in an ageing population. The third question concerns the timing of immigration when immigration affects social norms. Acknowledgment Financial support from the Adar Foundation of the Economics Department of Bar-Ilan University is gratefully acknowledged. References Adams, R. (1992), The impact of migration and remittances on inequality in rural Pakistan. Pakistan Development Review 31 (4), 1189–1203. Akee, R.K.Q, Basu, A.K., Chau, N.H., Khamis, M. (2010), Ethnic fragmentation, conﬂict, displaced persons and human trafﬁcking: an empirical analysis. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 691–716. Alesina, A., La Ferrara, E. (2000), Participation in heterogeneous communities. Quarterly Journal of Economics 847–904. Altonji, J.G., Blank, R.M. (1999), Race and gender in the labor market. In: Ashenfelter, O., Card, D. (Eds.), Handbook of Labor Economics Vol. 3C. Elsevier Science B.V., Amsterdam, pp. 3143–3259. Anas, A. (2002), Prejudice, exclusion and compensating transfers: the economics of ethnic segregation. Journal of Urban Economics 52 (3), 409–432. Baldwin-Grossman, J. (1982), The substitutability of natives and immigrants in production. The Review of Economics and Statistics 4, 596–603. Bartel, A.P. (1989), Where do the new U.S. immigrants live? Journal of Labor Economics 7 (4), 371–391. Battisti, M., DeVoretz, D. (2010), FSU immigrants in Canada: a case of positive triple selection? In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 579–604. Bauer, T., Epstein, G.S., Gang, I.N. (2005), Enclaves, language, and the location choice of migrants. Journal of Population Economics 18 (4), 649–662.

Migration and Culture

17

Bauer, T., Epstein, G.S., Gang, I.N. (2009), Measuring ethnic linkages between immigrants. International Journal of Manpower 30 (1/2). Benhabib, J. (1996), On the political economy of immigration. Economic European Review 40, 1737–1743. Bhaumik, S.K., Gang, I.N., Yun, M.-S. (2006), Ethnic conﬂict and economic disparity: Serbians and Albanians in Kosovo. Journal of Comparative Economics 34 (4), 754–773. Bisin, A., Verdier, V. (2000), Beyond the melting pot: cultural transmission, marriage, and the evolution of ethnic and religious traits. Quarterly Journal of Economics, 955–988. Blau, F.D., Kahn, L.M. (1997), Swimming upstream: trends in the gender wage differential in the 1980s. Journal of Labor Economics 15 (1), 1–42. Blau, F.D., Kahn, L.M. (2006), The U.S. gender pay gap in the 1990s: slowing convergence. Industrial and Labor Relations Review 60 (1), 45–66. Blau, F.D., Kahn, L.M. (2007), The gender pay gap. The Economists’ Voice 4 (4), (http://www.bepress.com/ev/vol4/iss4/art5). Bodenhorn, H., Moehling, C.M., Piehl, A.M. (2010), Immigration: America’s nineteenth century ‘law and order problem?’ In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization (Vol. 8). Emerald, Bingley, UK, pp. 295–323. Bodvarsson, O¨.B., Van den Berg, H.F. (2009), Economics of Immigration: Theory and Policy. Springer-Verlag, Berlin & Heidelberg. Bodvarsson, O¨.B., Sessions, J.G. (2010), Nationality discrimination in the labor market: theory and test. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization (Vol. 8). Emerald, Bingley, UK, pp. 231–268. Borjas, G.J. (2003), The labor demand curve is downward sloping: reexamining the impact of immigration on the labor market. The Quarterly Journal of Economics 118 (4), 1335–1374. Card, D. (2005), Is the new immigration really so bad? Economic Journal 115 (507), F300–F323. Carrington, W.J., Detragiache, E., Vishwanath, T. (1996), Migration with endogenous moving costs. American Economic Review 86 (4), 909–930. Chiswick, B.R., Miller, P.W. (1995), The endogeneity between language and earnings: international analyses. Journal of Labor Economics 13, 246–288. Chiswick, B.R. (1977), Sons of immigrants: are they at an earnings disadvantage? American Economic Review: Papers and Proceedings 67 (1) 376–380. Chiswick, B.R., Miller, P.W. (1996), Ethnic networks and language proﬁciency among immigrants. Journal of Population Economics 9 (1), 19–35. Chiswick, B.R., Miller, P.W. (2010), The effects of school quality in the origin on the payoff to schooling for immigrants. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 67–103.

18

Gil S. Epstein and Ira N. Gang

Church, J., King, I. (1983), Bilingualism and network externalities. Canadian Journal of Economics 26 (2), 337–345. Cohen-Goldner, S. (2010), Household structure of recent immigrants to Israel. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 447–465. Constant, A., Zimmermann, K.F. (2005), Immigrant performance and selective immigration policy: a European perspective. National Institute Economic Review 194, 94–105. Constant, A., Gataullina, L., Zimmermann, K.F. (2009), Ethnosizing immigrants. Journal of Economic Behavior and Organization 69 (3), 274–287. Cox Edwards, A., Ureta, M. (2003), International migration, remittances and schooling: evidence from El Salvador. Journal of Development Economics 72 (2), 429–461. Deutsch, J. (2010), The measurement of income polarization by ethnic groups: the case of Israel population. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 45–66. Deutsch, J., Epstein, G.S., Lecker, T. (2006), Multi-generation model of immigrant earnings: theory and application. Research in Labor Economics, 217–234. DeVoretz, D.J., Vadean, F.P. (2010), Cultural differences in the remittance behaviour of households: evidence from Canadian micro data. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 543–575. Dimova, R., Wolff, F.-C. (2009), Remittances and Chain Migration: Longitudinal Evidence from Bosnia and Herzegovina. IZA Discussion Papers 4083, Institute for the Study of Labor (IZA), Bonn, Germany. Dustmann, C., Fabbri, F., Preston, I. (2004). Ethnic Concentration, Prejudice and Racial Harassment of Minorities, CReAM Discussion Paper 05/04 (www.econ.ucl.ac.uk/cream/). Epstein, G.S. (2003), Labor market interactions between legal and illegal minorities. Review of Development Economics 7 (1), 30–43. Epstein, G.S., Gang, I. (2006), Ethnic networks and international trade. In: Federico Foders, I., Langhammer, R.J. (Eds.), Labor Mobility and the World Economy. Springer, Berlin Heidelberg, pp. 85–103. Epstein, G.S., Gang, I.N. (2009), Ethnicity, assimilation and harassment in the labor market. Research in Labor Economics 79, 67–90. Epstein, G.S., Gang, I.N. (2010), The political economy of the immigrant assimilation: internal dynamics. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 325–339.

Migration and Culture

19

Epstein, G.S., Kahana, N. (2008), Child labor and temporary emigration. Economics Letters 99 (3), 545–548. Epstein, G.S., Mealem, Y. (2010), Interactions between local and migrant workers at the workplace. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 193–203. Epstein, G.S. (2010), Informational cascades and the decision to migrate. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 25–44. Facchini, G., Mayda, A.M. (2010), What drives immigration policy? Evidence based on a survey of government ofﬁcials. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 605–648. Facchini, G., Testa, C. (2009), Who is against a common market? Journal of the European Economic Association 7, 1068–1100. Faini, R., Venturini, A. (2010), Development and migration: lessons from Southern Europe. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 105–136. Fertig, M. (2010), The societal integration of immigrants in Germany. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 375–400. Friedberg, R.M., Hunt, J. (1995), The impact of immigrants on host country wages, employment, and growth. Journal of Economic Perspectives 9 (2), 23–44. Gang, I.N. (2010), Who matters most? The effect of parent’s schooling on children’s schooling. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 401–414. Gang, I.N., Zimmermann, K.F. (2000), Is child like parent? Educational attainment and ethnic origin. Journal of Human Resources 35, 550–569. Gang, I.N., Rivera-Batiz, F. (1994), Labor market effects of immigration in the United States and Europe: substitution vs. complementarity. Journal of Population Economics 7, 157–175. Gang, I.N., Rivera-Batiz, F., Yun, M.-S. (2002), Economic Strain, Ethnic Concentration and Attitudes Towards Foreigners in the European Union, IZA Discussion Paper 578 (www.iza.org). Gang, I.N., Rivera-Batiz, F.L., Yun, M.-S. (2010), Changes in attitudes towards immigrants in Europe: before and after the fall of the Berlin Wall. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 649–676. Gottlieb, P. (1987), Making their own way: shorthorn blacks’ migration to Pittsburgh, 1916–30. University of Illinois Press, Urbana.

20

Gil S. Epstein and Ira N. Gang

Grossman, J.R. (1989), Land and hope: Chicago, black Southerners, and the Great Migration. University of Chicago Press, Chicago. Hildebrand, N., McKenzie, D. (2005), The effects of migration on child health in Mexico. Economia 6, 257–289. Jaeger, D.A. (2007), Green cards and the location choices of immigrants in the United States, 1971–2000. Research in Labor Economics 27, 131–184. Kahanec, M. (2006), Ethnic Specialization and Earnings Inequality: Why Being a Minority Hurts but Being a Big Minority Hurts More, IZA Discussion Paper 2050, (www.iza.org). Kahanec, M. (2010), Ethnic competition and specialization. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 205–229. Kahanec, M., Yuksel, M. (2010), Intergenerational transfer of human capital under post-war distress: the displaced and the Roma in the Former Yugoslavia. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 415–443. Katav-Herz, S. (2010), The implications of social norms on immigration policy. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 677–689. Kaushal, N., Kaestner, R. (2010), Geographic dispersion and internal migration of immigrants. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 137–173. Lahiri, S. (2010), Assimilating under credit constraints: public support for private efforts. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 341–356. Lazear, E.P. (1999), Culture and language. Journal of Political Economy 107 (6, pt. 2), S95–S126. Marks, C. (1989), Farewell – we’re good and gone: the Great Black Migration. Indiana University Press, Bloomington. Mesnard, A. (2004), Temporary migration and capital market imperfections. Oxford Economic Papers 56, 242–262. Mundra, K. (2010), Immigrant networks and the U.S. bilateral trade: role of immigrant income. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 357–373. Munshi, K. (2003), Networks in the modern economy: Mexican migrants in the U.S. labor market. Quarterly Journal of Economics 118, 549–599. Ortega, F. (2005), Immigration quotas and skill upgrading. Journal of Public Economics 89, 1841–1863.

Migration and Culture

21

Ottaviano, G.I., Peri, G. (2008), Immigration and national wages: clarifying the theory and the empirics. NBER Working Papers 14188, National Bureau of Economic Research (NBER), Cambridge, Massachusetts. Available at http://www.nber.org/papers/w14188 Poggio, S., Gindling, T.H. (2010), Promoting the educational success of Latin American immigrant children separated from parents during migration. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 517–541. Polachek, S.W., Horvath, F.W. (1977), A life cycle approach to migration: analysis of the perspicacious peregrinator. In: Ron Ehrenberg (Ed.), Research in Labor Economics, Vol. 1. JAI Press, Greenwich, Conn., 103–149. Schmidt, C.M. (2010), Understanding the wage dynamics of immigrant labor: a contractual alternative. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 177–191. Schultz-Nielsen, M.L., Constant, A. (2004), Employment trends for immigrants and natives. In: Tranaes, T., Zimmermann, K.F. (Eds.), Migrants, Work, and the Welfare State. University Press of Southern Denmark, Odense, pp. 119–146. Siniver, E. (2010), Culture, investment in language and earnings. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 269–292. Smith, J.P., Welch, F.R. (1989), Black economic progress after Myrdal. Journal of Economic Literature 27 (2), 519–564. Taylor, J.E., Wyatt, T.J. (1996), The shadow value of migrant remittances, income and inequality in the household-farm economy. Journal of Development Studies 32 (6), 899–912. Tranaes, T., Zimmermann, K.F. (2004), Migrants, work, and the welfare state: an introduction. In: Tranaes, T., Zimmermann, K.F. (Eds.), Migrants, Work, and the Welfare State. University Press of Southern Denmark, Odense. Vadean, F., Piracha, M. (2010), Circular migration or permanent return: what determines different forms of migration? In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 467–495. Weiss, A., Rapoport, H. (2003), The optimal size for a minority. The Journal of Economic Behavior and Organization 51 (1), 27–45. Xing, Y., Semyonov, M., Haberfeld, Y. (2010), Labor migration, remittances, and economic well-being: a study of households in Rajasthan, India. In: Epstein, G.S., Gang, I.N. (Eds.), Migration and Culture: Frontiers of Economics and Globalization, Vol. 8. Emerald, Bingley, UK, pp. 497–516.

PART I

Enclaves and Location Choice

CHAPTER 2

Informational Cascades and the Decision to Migrate Gil S. Epsteina,b,c a

Department of Economics, Bar-Ilan University, Ramat-Gan 52900, Israel CReAM-Center for Research and Analysis of Migration, London, UK c Institute for the Study of Labor (IZA), Bonn, Germany E-mail address: [email protected] b

Abstract We introduce the idea that informational cascades can explain the observed regularity that emigrants from the same location tend to choose the same foreign location. Thus, informational cascades generate herd behavior. Herd behavior is compared with the network-externalities explanation of the same phenomenon of migration clustering. Keywords: Migration, informational cascade, herd behavior, xenophobia, network-externalities JEL classiﬁcations: F22, J61

1. Introduction Consider yourself living in a small town in a low-income country, and you have decided to emigrate. Where would you go? You might prefer a country because of your familiarity with its language. That may leave you with a number of alternatives. Or you may choose a foreign location because of the presence of people there from your own home community. You may have a relative or a friend of the family in the foreign location. Or, at least the name and address of somebody who knows your family and who will treat you sympathetically, assisting you with housing and ﬁnding a job, and perhaps in explaining the rules of neighborhood.1 As with language proﬁciency, in all likelihood a number of foreign locations can provide such networks of people to help you. Language itself, and the 1 See Church and King (1983), Gottlieb (1987), Grossman (1989), Marks (1989), and Chiswick and Miller (1996).

Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008008

r 2010 by Emerald Group Publishing Limited. All rights reserved

26

Gil S. Epstein

presence of people ready and willing to help, are not sufﬁcient to provide a basis for a decision. A choice remains to be made from among the available alternatives where language and/or personal connections are present. Help in adjusting to a new environment such as ﬁnding jobs and accommodations are provided by network externalities. The ability to speak his language in a foreign country is in itself helpful. The network allows the immigrant to preserve his traditions and history in the new environment. Network externalities are not always positive. An increase in the number of foreigners in the host country increases competition for jobs that the immigrants can work at, thus decreasing the immigrants’ wages. Moreover, as the number of immigrants increase, the local population may become xenophobic. There are, therefore, two different contradicting effects to network externalities. The immigrant must weigh one against the other and decide which dominates. If full information were available about local conditions, migrants would choose the location where there are net beneﬁts from network externalities. If such full information is not available, a choice is made under conditions of uncertainty. If you have imperfect information, which decision rule should you adopt? In the face of uncertainty, a common decision rule is to randomize, but here you confront an indivisible location decision. You may not know all that much about life in a particular location. You observe, however, that other people who are like you have recently been favoring this location. You might have a personal feeling that the location people have been choosing is not the best from among the available alternatives. You might, however, decide to discount this feeling that is based on your private information, and to proceed on the assumption that others have been making decisions based on better information than you have. That is, you may take the position that so many other people cannot be wrong. If you behave in this way and discount your private information or your feelings to follow the decisions of others, you are adopting a decision rule that gives rise to herd behavior. In order for a population of immigrants to produce network externalities that will attract other migrants, the population of immigrants must be sufﬁciently large. In many situations it is not clear how this critical mass of people arrived at a certain location. Informational cascades – herd behavior – help us understand the creation of the critical mass that creates network externalities. In order to create a herd in a certain location, the number of immigrants needed is relatively small. Thus, herd behavior may be an explanation for the creation of the mass of immigrants that is sufﬁcient to attract others to join and enjoy the positive externalities of the network. Informational cascades also help us understand why we observe immigrants deciding to emigrate to destinations where the negative externalities are stronger than the positive externalities of the network. The reason for this phenomenon is that individuals are uncertain regarding the effect of the network externalities and decide to follow the ﬂow of

Informational Cascades and the Decision to Migrate

27

immigrants rather than the stock of immigrants. Finally, herd behavior enables us to understand how an individual makes a decision when there is more than one country that provides the immigrant with the same level of network externalities. The immigrants will decide to follow the ﬂow (informational cascade or herd behavior) rather than the stock (network externalities) of previous immigrants’ destinations. Several empirical studies investigate the determinants of location choice of immigrants in the United States.2 Bartel (1989) ﬁnds that post-1964 US immigrants tend to locate in cities with a high concentration of immigrants of similar ethnicity. Bauer et al. (2007) distinguish between two types of networks and herd effects, using data from the Mexican Migration Project with individual-level data on Mexican-US migration (based at the University of Pennsylvania and the University of Guadalajara). One of their network variables captures the general variables that describe the type of origin-speciﬁc consumption products that migrants wish to consume in a US location. This variable is generally called ethnic goods, which describes the availability of products that are unique to speciﬁc groups from speciﬁc origins. Their other network variables capture origin, village connections, and the history of a village in US locations. Using these two variables helps to distinguish a generalized network effect from the village-speciﬁc links. The herd variable describes the ﬂow of migrants during the year. Their empirical results show that both network externalities and herds have signiﬁcant effects on the migrant’s decision about where to migrate. Moreover, the signiﬁcance and size of the effects vary according to the legal status of the migrant and whether the migrant is a ‘‘new’’ or a ‘‘repeat’’ migrant. Bauer et al. (2009), based on the same data, use three different measures of ethnic networks in order to investigate different channels through which ethnic networks affect the location choice of migrants. The empirical evidence shows that the availability of ethnic goods, the information provided by return migrants to potential migrants in their origin village, and the number of current migrants from an origin village living in the host location compared to other locations are signiﬁcant and important for the location choice of migrants. Using Israeli data, Epstein and Cohen (2006) investigated the case where migrants from the Soviet Union decided to live in Israel. They show that both network and herd behavior determined the location choice of the migrants. Moreover, it was shown that the network effect has an inverted U shape and the herd effect is linear. It was also discovered that there is an inverted U-shape effect of the size of the network on language proﬁciency and the probability of ﬁnding a job. In this paper we set out a formal framework introducing informational cascades and generating a theory of herd behavior as an explanation of

2

For a more detailed description of empirical results and the literature, see Epstein (2008).

28

Gil S. Epstein

migration. In the model emigrants may have some private information but are imperfectly informed about the attributes of alternative foreign locations, and they observe previous emigrants’ decisions. Behavior is rational on the supposition by impending emigrants that previous emigrants had information that they do not have. The outcome is that emigrants discount private information and duplicate a location that previous emigrants have been observed to choose.3 The consequence is immigrant clustering in foreign locations, but based on a decision rule that does not internalize all true information. Since individuals are discounting private information that may be accurate and making decisions based on the perception that other people’s information is accurate when others are also likewise discounting private information, we can have no expectation that outcomes will have desirable properties. The paper proceeds as follows. We set out the model of herd behavior in the following section, and then analyze herd behavior together with network externalities.

2. The model 2.1. The background We consider a country where potential emigrants are identical other than in age and information, and are uncertain about conditions in the rest of the world. We do not wish to attribute aspects of behavior to risk aversion, and so take emigrants to be risk neutral (although we realize in practice they may not be). An emigrant’s utility U(.) is increasing in income, and in other parameters that we shall subsequently introduce. From among the alternative foreign locations for emigration (legal or illegal), one location objectively offers better conditions than others. Emigrants do not know the identity of this best foreign location. They have a uniform prior over foreign locations. An individual may decide not to emigrate, which is encompassed by viewing one of the locational options as the home country. Individuals may receive information regarding the standard of living and their opportunities in a host country. This information could be obtained from watching TV, reading newspapers, etc. This information is hereafter referred to as a signal. After observing the different information, the individual must make a decision if and where to immigrate. If the individual cannot make a decision based on this information (signal), then there is no difference between an individual who did not receive a signal and this individual – this 3

The theory of information cascades or herd effects has been applied to the explanation of behavior in a number of contexts. See Scharfstein and Stein (1990), Banerjee (1992), and Bikhchandani et al. (1992).

Informational Cascades and the Decision to Migrate

29

individual is, therefore, seen as an individual who didn’t receive any information. We now demonstrate the formal structure of emigration decisions that follow herd behavior using one- and two-signal models. 2.2. A one-signal model Imperfect private information provides a signal, with probability p, regarding the identity of the best foreign country. With probability q, the signal providing this private information is true. If a signal is false, it does not provide information regarding the true signal. Also, to simplify, we assume that, for two locations, qW0.5 (or qW1/m where m is the number of foreign locations). Otherwise, there is a better chance of choosing the preferred country by randomizing than by using the information provided by the signal. Emigration decisions are made sequentially, with people contemplating emigration at a given age or stage in their lives. In the sequential decision process, people of different ages make decisions regarding immigration at different times. Someone may have received a signal, and he or she can also observe the behavior of previous emigrants. Potential emigrants cannot however observe the information signal that was the basis for previous emigrants’ decisions. While potential new immigrants know the choices made by past emigrants, they do not have to know the latter’s position in the queue. Given the information available, each person chooses a location to which to emigrate. The structure of the game and Bayesian rationality are common knowledge. Three assumptions govern individuals’ actions4: (a) A person who does not receive a signal and observes that everybody else has chosen to stay home will also choose not to emigrate. (b) Someone who is indifferent between following his or her own signal and copying someone else’s choice will follow his or her own signal. (c) Someone who is indifferent between copying previous emigrants’ decisions will make a decision by randomizing with equal probabilities assigned to the different alternatives. These assumptions, which minimize the likelihood of herd behavior, give rise to the following different possibilities: The ﬁrst person making a decision: This person fails to receive a signal with probability (1p) and receives a signal with probability p. In the ﬁrst case, by assumption a, individual will not emigrate. In the second case, individual will follow his signal, and will emigrate. The probability that emigration is to the correct country is q. 4

See Epstein (2008).

30

Gil S. Epstein

The second person: If person 2 has received no signal, then she follows person 1. If only person 2 has a signal, she of course will follow her signal. If the two people have different signals (person 1 chose to emigrate and thus had a signal), then person 2 is indifferent between following her own signal and copying person 1, as the both persons’ signals have the same probability of being true. In this case, by assumption b, person 2 will follow her own signal. The third person: If neither of the two previous persons chose to emigrate, this means that neither received a signal. Person 3 will copy them if and only if he does not receive a signal, otherwise will follow the signal he receives. If one of the previous persons chose not to emigrate and the other chose to emigrate, person 1 did not receive a signal and person 2 did receive a signal. If person 3 then receives a signal that indicates emigration to the country to which the second person has emigrated, person 3 will join the second emigrant. Otherwise, if a signal different from that of person 2 is received, person 3 follows his own signal. If persons 1 and 2 have chosen to emigrate to different countries, and person 3 does have a signal, then person 3 will base his emigration decision on his own private information as conveyed by the signal he receives. This can be shown formally in the following way: Assume that person 1 emigrated to country j, person 2 emigrated to country k, and person 3 has a signal to emigrate to country j. Using the Bayesian rule, person 3 can calculate the probability that the true signal is j out of m possible countries5: Prðjj j; k; jÞ ¼

p3 q2 ð1 qÞ1=m Prðk; j; jÞ

(1)

In the same way, person 3 could calculate the probability that the true signal is k: Prðkj j; k; jÞ ¼

p3 qð1 qÞ2 1=m Prðk; j; jÞ

(2)

For qW0.5, Prðjj j; k; jÞ4Prðkj j; k; jÞ

(3)

from which it follows that person 3 will choose to follow his own signal. There is one further possibility: the ﬁrst two persons choose to emigrate to country j and the third person receives a signal to emigrate to country k. This last possibility brings us to a general proposition. First, however, 5

By deﬁnition, the probability q is normalized with regard to the two different locations.

Informational Cascades and the Decision to Migrate

31

to simplify, we add the following assumption: 1p 2 Assumption d : qð1 qÞ40:5 p That is, we place a lower bound on the probability that an individual receives a signal. The assumption is relaxed in Proposition 2. PROPOSITION 1. If at a point in time the number of emigrants in country j is greater than emigrants in all the other countries by at least two persons, then from that time on, all persons, regardless of their signal, will emigrate to country j, and so we have herd behavior. For the proof, see the Appendix. The proposition is true for any number of countries, as the choice is always whether to follow one’s signal or to follow the herd; that is, the problem is always a binomial decision. In order for herd behavior to occur after a difference between two individuals, we require a bound on the probability that a signal is received: p2 qð1 qÞ40:5ð1 pÞ2 Thus, as q increases, in order for herd behavior to occur, a higher value of p is required. As the probability of receiving a signal decreases, more emigrants are required to create herd behavior, and we can conclude that: PROPOSITION 2. For a given probability q that a signal is true, as the probability p that an individual receives a signal decreases, the number of emigrants required to evoke herd effect increases. A person who has chosen to emigrate does not immediately know the quality of life in the new location. Suppose that a person has emigrated, and after some time a clustering of immigrants occurs in a different country. As the emigrant continues to confront uncertainty regarding future income and the future quality or standard of life in the new country, he or she once again calculates the probability regarding the best country. The propositions above indicate that such a person will decide against going to the country of initial choice and join the herd. 2.3. An illustration We now present an illustration. We have established that if the ﬁrst two persons emigrate to the same location, all subsequent persons will emigrate to this same location. The probability that the two ﬁrst persons will emigrate to the same location is (where we assume that j is

32

Gil S. Epstein

the best location): Prðclustering in one countryÞ ¼ Prðclustering in country jj j is the right countryÞPrðj is the right countryÞ þPrðclustering in country jj k is the right countryÞPrðk is the right countryÞ þPrðclustering in country kj k is the right countryÞPrðk is the right countryÞ þPrðclustering in country kj j is the right countryÞPrðj is the right countryÞ (4)

Using the values of the different probabilities, we obtain: Prðclustering in one countryÞ ¼ ðp2 q2 þ pð1 pÞqÞ1 þ ðp2 ð1 qÞ2 þ pð1 pÞð1 qÞÞ0 þðp2 q2 þ pð1 pÞqÞ0 þ ðp2 ð1 qÞ2 þ pð1 pÞð1 qÞÞ1 ¼ p2 q2 þ pð1 pÞq þ p2 ð1 qÞ2 þ pð1 pÞð1 qÞ

(5)

In the case where q ¼ 0.51 and p ¼ 1 (all people obtain a signal), we calculate this probability to be 0.5002. More generally, as q increases, for any p, the probability of clustering in one of the countries increases: @PrðclusteringÞ ¼ 2p2 ð2q 1Þ40 @q

(6)

Herd behavior thus occurs with positive probability. Simple Markovian reasoning tells us that, with an inﬁnite population size, the probability of any event occurring is 1. Thus, if the population size is inﬁnite, after some point in time, with probability 1, there will be clustering of immigrants in one location.6 2.4. Multiple signaling In a multiple-signaling version of the model, a person can receive two types of signals: a general signal, a speciﬁc signal from previous emigrants, and also can observe the behavior of previous emigrants. Again, he or she cannot however observe the signal that was the basis of the decisions of past emigrants, and, given the information available, each person proceeds to choose a country of emigration. We retain assumptions a, b, c, and d and add the assumption e: individuals value a speciﬁc signal from former emigrants, qi, more than a general signal, q, that is, q qi 8 i.7 6

For a similar result, see Bikhchandani et al. (1992). One of the key determinants of the location of immigration is past colonial relationships. The general signal can be interrupted accordingly. We can view the speciﬁc signal as evidence that an immigrant has gone to the wrong country and chooses another location at the next period. Here the individual’s change of his/her decision can be seen as a speciﬁc signal telling the individual not to immigrate to a location. The signal is not true with probability 1 as it is not clear why the emigrant changed his decision, and it may well be that the location may not suit him while this is the correct choice for other emigrants.

7

Informational Cascades and the Decision to Migrate

33

Notice that a person can receive a speciﬁc signal to go to a particular country only if there has been a prior emigrant to that country. An immigrant who receives opposite general and speciﬁc signals must determine which to follow. It is clear that the probability that an individual will choose to emigrate against his speciﬁc signal is smaller than against his general signal. However, the presence of a greater number of emigrants already located in the host country against the speciﬁc signal increases the probability that the emigrant follows the herd. We summarize the results in the following proposition8: PROPOSITION 3. With multiple signals, if two initial persons have emigrated to the same country, subsequent emigrants copy them regardless of their own signals; otherwise, herd behavior will occur when the difference between the number of emigrants in two countries is large enough. As the probability that a person’s own signal is true increases, the difference decreases between the number of emigrants in alternative locations required for herd behavior. 3. Network externalities As we observed in the introduction, herd behavior is conceptually different and distinguishable from migration that is motivated by network externalities (see also the concluding section).9 There is also no reason why herd effects and network externalities should not be simultaneously present to inﬂuence emigration location decisions. When there is simultaneous presence, there is also interaction. In this section we place net externalities within our model of herd behavior and show the nature of the interactions between the two phenomena.10 To introduce network externalities, we follow the representation of Carrington et al. (1996). A potential emigrant calculates the present discounted value of income for staying at home versus emigrating. The present value of income for emigrants is a function of the number of immigrants in the host country. The number of immigrants thus has an effect on the migration decision. Wages in the host country are assumed to decrease with the number of immigrants, while the wage in the home country is an increasing function of the number of people who emigrate. With endogenous moving costs that decrease with the size of the network, the impetus for emigration develops gradually over time. Emigration, once 8

The proof is available on request. When a migrant immigrates as a result of network externalities, he wishes to beneﬁt from the externalities that other migrants can provide him with. On the other hand, in migration as a result of herd behavior, the migrant wishes to maximize the probability of choosing the correct destiny and has uncertainty regarding the information available to him and others. 10 See also Choi (1997). 9

34

Gil S. Epstein

it begins, gains momentum, and the number of people who migrate can well increase even as differences in wages between the country of emigration and immigration decline. In a network-externalities model, costs of relocation thus decrease with the number of immigrants, which encourages more emigration, and leads to immigrant clustering – but some immigrant clustering must already have been present to provide the externalities. Let us assume for now full information. The utility of an immigrant, Uj (.), is a function of two main variables: ﬁrst is the relative wage (relative to his home country) that the immigrant will receive from immigrating to the new location wf (wf is the immigrants earnings at the new location), and second is the number of immigrants from the same origin that previously immigrated to that location, N. From the above discussion, the immigrant’s utility increases with the immigrant’s earnings and increases with the number of immigrants that have already immigrated before him to the same location. Thus, ð@U j ðwj ; NÞ=ð@wj ÞÞ40 and ð@U j ðwj ; NÞ=ð@NÞÞ40. For a given utility level, an iso-utility locus (indifference curve) is described by the following: dU j ðwj ; NÞ ¼

@U j ðwj ; NÞ @U j ðwj ; NÞ dwj þ dN ¼ 0 @wj @N

(7)

which has a negative slope as there is a trade-off between the wage level (earnings) in the host country and the total number of immigrates from the same origin country that have already immigrated before immigrant j: dwj ð@U j ðwj ; NÞ=ð@NÞÞ ¼ o0 dN ð@U j ðwj ; NÞ=ð@wj ÞÞ

(8)

In other words the iso-utility is downward sloping. Moreover, as we increase the wage and the number of previous immigrants from the same origin, the utility of the new immigrant increases. Assume a normal downward sloping demand function for workers in the host country: qd ðwf Þ such that ðqd ðwf Þ=ð@wf ÞÞo0, and an upward sloping supply function of workers: qs ðN L ; NÞ, where NL is the size of the local population such that ðqs ðN L ; NÞ=ð@NÞÞ40. In equilibrium qd ðwf Þ ¼ qs ðN L ; NÞ, thus we obtain that the wages in equilibrium are given by wf ðNÞ. In other words, the wage in equilibrium is a function of the number of immigrants in the country. More speciﬁcally we can easily show that ð@wf ðNÞ=ð@NÞÞo0; namely, as the number of immigrants increase, the equilibrium wage decreases. To illustrate this let us look at speciﬁc demand and supply functions for immigrants and the immigrants’ utility function: Denote demand for immigrants in country j by qD j ¼ b0 b1 wj where qj denotes the number of immigrants and wj the wage. The supply function is qsj ¼ a0 þ a1 wj þ N j ,

Informational Cascades and the Decision to Migrate

35

where Nj is the number of immigrants in the country.11 In equilibrium s qD j ¼ qj and the equilibrium wage is: wj ¼

b0 a0 N j . a1 þ b1

(9)

It is clear that, as the number of migrants increases, the wage decreases. Let wjWwk for all k 6¼ j. Denote utility of a representative emigrant by UðC; N; LÞ ¼ Cd1 N d2 Ld3 where C is consumption and L is the size of the local population, with d1 ; d2 ; d3 o1. Externalities are reﬂected in the size of the local population and in the number of immigrants. All income is spent on consumption, so that C ¼ wj, and d b 0 a 0 N 1 d2 d3 N L (10) UðC; N; LÞ ¼ a1 þ b1 The condition for utility to be increased by more immigrants is: Noðb0 a0 Þ

d2 ¼ N0 d1 þ d2

(11)

which shows that the wage and consumption decrease as immigration increases, but that the loss in utility is offset by network-externality beneﬁts.12 Thus, as the number immigrants in the host country increase until N0, the probability of a new immigrant choosing the same country increases. Increasing the number of immigrants beyond N0 will decrease the probability of an individual choosing that host country – see Figure 1. Now to herd behavior: let one person emigrate to a designated country when a second person receives a positive signal indicating emigration to a different country. If this latter person chooses to follow the ﬁrst migrant, then she knows that all successors will follow, for informational and payoff reasons (herd behavior and positive externalities). If she chooses the other country, there is a positive probability that she will end up alone. So, while she may think that the basic payoff or utility from moving to the alternative country is as good as for the ﬁrst country, the awareness of the positive network payoff will induce her to choose the location chosen by this ﬁrst emigrant. Herd behavior is therefore more pronounced than when externalities are absent, and with high probability the ﬁrst emigrant will be followed by everyone. In disregarding network externalities to focus on herd behavior, we took it to be the case that an emigrant had no information regarding expected utility, and received signals regarding the probability that a

11

See Brezis and Krugman (1996) for an argument that this is so in the short run, but not necessarily in the long run. 12 Notice that ðb0 a0 Þ40 and ðd2 =ðd1 þ d2 ÞÞo1.

36

Probability Of Immigration

Gil S. Epstein

0

No Number of Immigrants

Fig. 1.

N

Network externalities.

particular country offered the best location.13 In the presence of beneﬁcial externalities, the utility from emigration to a country depends on (1) the number of immigrants who have previously immigrated and (2) how many people will immigrate in the future. So even if the wage in a country is relatively low, the positive externalities may make that country an attractive location. For example, suppose n people have emigrated to country j and one person has emigrated to country k, and that utility of an immigrant in country j is higher than that of an immigrant in country k. It could however be that if n immigrants had immigrated to country k and one immigrant to country j, utility in country k would have been higher than in country j (if n immigrants had immigrated to that country). With herd behavior, the probability that a signal received by an individual is true is a function of the number of both previous immigrants who have immigrated to the same country and immigrants who have chosen other countries. We can deﬁne the probability in the following way: Suppose an individual has received a signal that country j is best, and has to choose between country j and country k. Given the number of individuals who have already emigrated to country j and k, the probability that this signal 13

If an individual were able to calculate expected utility in the foreign country, the combined herd effects and positive externalities could be easily established.

Informational Cascades and the Decision to Migrate

37

is true is given by: q0j ð:Þ ¼ q0j

nj nk

while

q0j ðnj =nk Þ 40 @ðnj =nk Þ

(16)

The probability q0j ð:Þ represents the normalized probability that the right thing to do is follow the signal. Thus q0j ð:Þ is a function of all the information, that is, the number of emigrants who have already emigrated to the different countries, and the basic probability that the signal is true q (q0j ð:Þ) is calculated in a similar way as in (1). Thus, the beneﬁts from network externalities inﬂuence the probability that a signal is true via the relative number of immigrants who previously emigrated to the different countries. When we now recompute the probabilities of Section 2, we ﬁnd that herd effects are more pronounced because of the externalities, and we conclude: PROPOSITION 4. The probability of herd behavior increases in the presence of positive externalities. As argued above, given that the immigrant is already in the host country, he prefers that the total number of immigrants will be equal to N 0 ¼ ðb0 a0 Þðd2 =ðd1 þ d2 ÞÞ. However, when this individual makes his decision whether to immigrate to this county, he will compare the expected utility from the different countries and chose the one with the highest value. We therefore may see immigrants deciding to immigrate to a country where the number of immigrants has already exceeded N0. Thus, the probability that an individual will chose to immigrate to a country where the number of immigrants already exceeding N0 is positive. This probability, however, will decrease as the number of immigrants already in the host country increases. We conclude: PROPOSITION 5. Given network externalities, the probability an individual will immigrate to a certain country has an inverse U-shape relationship with regard to the number of immigrants already in the host country. Herd effects are less pronounced when externalities are negative.14 Consider a general signal received by an individual to move to country j rather than country k. The probability associated with this signal increases with the relative number of immigrants who already chose j, only if the total number of past emigrants is less than a bound, determined by the number at which negative externalities set in. 14

When disadvantageous externalities are present, incentives arise to move to new locations, in the course of which individuals tend to reveal private information – as they will only emigrate to another location if warranted by private information.

38

Gil S. Epstein

Denote this number by mj. Then: nj q0j ð:Þ ¼ q0j nk q0j ðnj =nk Þ while 40 for nj omj and nk omk @ðnj =nk Þ and q0j ðnj =nk Þ o0 @ðnj =nk Þ

for nj 4mj and nk omk

ð17Þ

The probability that the private signal is true is independent of the number of emigrants who previously chose a country. If disadvantageous externalities are present, incentives arise to move to new locations, in the course of which individuals tend to reveal private information – as they will only migrate to another location if warranted by private information. Informational herd effects are therefore less pronounced in the case of negative externalities. A migrant may move to a country and ﬁnd out that the marginal positive effect of the externalities is lower than the marginal negative effect of the wage. In other words, the stock of immigrants who have migrated to this host country has exceeded N0. A migrant who is leaving this host country, where the stock of migrants is higher than N0, will now send negative speciﬁc signals to his home-country people who are thinking about migrating to that country. The signals will be saying not to migrate to where he migrated. The local population at the home country receives these negative signals. However, the population at home knows that a lot of individuals have migrated to this country and may even receive other information that this place is the right place to which to immigrate. An individual who has to make the decision will weigh the information he received: the stock of previous individuals who migrated to that country (and to other countries), the general information he received while observing the ﬂow of migrants, and the negative information he received from the migrants who have already immigrated to that country. This individual knows that there is a positive probability that the information he received from the migrants in the host country is true for them as they do not want other migrants to join them. However, it may be optimal for the migrant to join them even if there are negative signals. In order for the individual to follow the ﬂow (herd), the proportion of negative signals relative to the stock of migrants must fall. Thus, if the stock of immigrants is sufﬁciently large in the host country, the new migrants may continue to follow the herd even though the network externalities are negative. Thus, under herd behavior one may observe that the probability of migrating to a certain location will increase over time when it should have decreased if one only takes into account network effects (see Figure 2).

Informational Cascades and the Decision to Migrate

39

Probability Of Immigration

Herd effect

0

Network externalities

No

N

Number of Immigrant

Fig. 2. Network externalities verses Herd behavior. 4. Concluding remarks Our purpose in this paper has been to draw attention to informational cascades and herd behavior as an inﬂuence on where migrants locate. Herd behavior offers an informational perspective on why emigrants from the same location make the same foreign relocation decision. Herd behavior complements network externalities in explaining foreign location decisions. Network externalities may not be sufﬁcient to explain the foreign locational choice, since a number of alternative locations may all offer network externalities. Herd effects can explain which of the alternatives offering network externalities is chosen. There are a number of additional dimensions to differences between network externalities and herd-effect explanations for the choice of emigrants’ destinations. Positive network externalities tell a story of efﬁciency through the internalized beneﬁts provided by the externalities. There are no mistakes. Herd behavior introduces the possibility of economic inefﬁciency through the discounting of accurate private information. Also, a prior critical presence of emigrants with the same cultural background or from the same location is required for network externalities. This is not a prerequisite for a herd-effects explanation of foreign location choice.

40

Gil S. Epstein

When the population of prior immigrants in a foreign location is small, network externalities are of course not present. Still, emigration decisions are made, generally under conditions of uncertainty. In such cases, we can only look to herd effects to explain initial immigrant clustering. After the immigrant population reaches a particular size, relations can become more impersonal, and the arrival of someone from ‘‘back home’’ may not evoke the same feeling of responsibility and benevolence. Network externalities can therefore be subject to diseconomies of size of the immigrant population. After a sufﬁciently long presence, a local individualistic culture can take hold (‘‘let the new arrival work hard and succeed by his own merits like I did’’). Thus, after a certain number of immigrants, it may be beneﬁcial for the emigrant to join a different network. Herd behavior will lead immigrants to continue coming to the same location when network externalities no longer justify this decision. Network externalities appear to be more important than herd effects for illegal immigrants, or when legal immigrants convert to illegality (see Epstein et al., 1998 and Epstein, 2002), because of the greater need of illegal immigrants for surreptitious existence and protection. Herd behavior can however be expected to diminish in signiﬁcance as prospective emigrants have access to more sophisticated and accurate information about conditions in foreign locations, since more weight is then placed by people on private information. Often host countries want migrants (see Epstein and Hillman, 2003). Trying to inﬂuence migration, even a limited migration, may result in a herd of migrants who are not wanted by the host country or by the local population in the host country. This scenario illustrates the complexity of national preferences and xenophobia regarding immigrant composition.15,16 The host country may have a clear objective of what it wants, 15

The national preferences reﬂect electoral outcomes, political popularity, and mass expression in various countries. Local indigenous populations have expressed discontent and uneasiness, and in cases have also become violent, because of immigration issues. In some European countries, parts of the local population have expressed anti-immigration preferences through the polls. Political parties taking explicit anti-immigrant positions have found signiﬁcant support in France, Austria, Switzerland, and the eastern regions of Germany. In Norway, when foreign presence is low, immigration has been a major electoral issue, and also in Denmark, where the foreign presence is higher. There has also been antiimmigration sentiment in Sweden. Xenophobia and national ethnic preference have been found outside of Europe, in Indonesia, for example, the Chinese population suffered in the vast pogroms of the 1960s and again in 1998. Indians were expelled from Uganda. In Fiji the indigenous population revoked democracy when they became a minority. 16 Gang et al. (2002) provide a statistical analysis on the determinants of attitudes toward foreigners displayed by Europeans sampled in the Eurobarometer surveys in 1988 ands 1997. In general they show that those who compete against migrants in the labor market have more negative attitudes toward foreigners, and as the concentration of immigrants in the local population increases, the likelihood of negative attitude increases (a negative networkexternality effect).

41

Informational Cascades and the Decision to Migrate

however, as a result of informational cascades; a herd of migrants may arise and bring unexpected results. Empirically we can distinguish between network externalities and herd effects in stock and ﬂow terms. If the ﬂow of emigrants to different locations is related to the prior stock of emigrants, we can infer network externalities are important. If the ﬂow of emigrants is related to prior ﬂows, herd effects are important. Acknowledgment Financial support from the Adar Foundation of the Economics Department of Bar-Ilan University is gratefully acknowledged. Appendix. proof of Proposition 1 Denote by Prðjj nj; ðn 2Þk; kÞ the probability that j is the best country to which to emigrate, and let it be observed that n individuals have immigrated to country j; (n2) to country k; and an individual receives a signal to immigrate to country k. First consider the case of three persons: The two ﬁrst persons have immigrated to country j and person 3 has received a signal to immigrate to country k. Given assumptions a and b, it is clear that the ﬁrst person has received a signal to immigrate to country j and the second person either did not receive a signal or received a signal to immigrate to country j. We can calculate the probability that j(k) is the true signal. Using the Bayesian rule, given this information, the probability that the j signal is true out of m possible countries is: Prðjj2j; 0k; kÞ ¼

ðp3 q2 ð1 qÞ þ p2 ð1 pÞqð1 qÞÞ1=m Prðk; j; jÞ

(A.1)

In the same way we calculate the probability that k is the true signal: Prðjj2j; 0k; kÞ ¼

ðp3 qð1 qÞ2 þ p2 ð1 pÞqð1 qÞÞ1=m Prðk; j; jÞ

(A.2)

Given that qW0.5, Prðkj2j; 0k; kÞoPrðjj2j; 0k; kÞ. Since the decision is between two different locations (out of a larger set of locations), the conditional probability for q is thus greater than 0.5. Given that the conditional probability qW0.5 it holds that Prðkj2j; 0k; kÞoPrðjj2j; 0k; kÞ. We now consider the case where one person has immigrated to country k, three persons have immigrated to country j, and the ﬁfth person has received a signal to immigrate to country k. Necessarily, the ﬁrst individual received a signal for j and the second receives a signal to country k. The third individual receives a signal for j or randomly selects j. The fourth

42

Gil S. Epstein

individual does not receive a signal for k. Finally, the ﬁfth individual receives a signal for k. So: Prð3j; 1k; kjjÞ ¼ pqpð1 qÞðpq þ 0:5ð1 pÞÞðpq þ ð1 pÞÞpð1 qÞ (A.3) Likewise: Prð3j; 1k; kjkÞ ¼ pqpð1 qÞðpð1 qÞ þ 0:5ð1 pÞÞðpð1 qÞ þ ð1 pÞÞpq (A.4) By Bayes’ rule: Prðjj3j; 1k; kÞ ¼

Prð3j; 1k; kjjÞ Prð3j; 1k; kjjÞ þ Prð3 j; 1k; kjkÞ

(A.5)

which is larger than Prðkj3j; 1k; kÞ ¼ 1 Prðjj3j; 1k; kÞ if and only if Prðjj3j; 1k; kÞ40:5, which is equivalent to Prðjj3j; 1k; kÞ4Prðkj3j; 1k; kÞ. We now see that Prðjj3j; 1k; kÞ4Prðkj3j; 1k; kÞ if and only if ðpq þ 0:5ð1 pÞÞðpq þ ð1 pÞÞpð1 qÞ 4ðpð1 qÞ þ 0:5ð1 pÞÞðpð1 qÞ þ ð1 pÞÞpq

(A.6)

Thus, (A.6) holds if and only if p2 qð1 qÞ40:5ð1 pÞ2

(A.7)

holds by assumption d. The rest of the proof is by induction. We have shown that herd behavior occurs in the two cases. Assume that the country that has the largest number of immigrants, country j, has n (n1) immigrants. Denote by k the country with the second largest number of immigrants, with (n2)(n3) immigrants. We have shown that herd behavior holds true for n ¼ 2 and n ¼ 3. Assume that it holds for n and n1. We will show that it holds for nþ1 and nþ2. Assuming that: Prðjj nj; ðn 2Þk; kÞ4Prðkj nj; ðn 2Þk; kÞ

(A.8)

and Prðjj ðn 1Þj; ðn 3Þk; kÞ4Prðkj ðn 1Þj; ðn 3Þk; kÞ

(A.9)

Our aim is to show that Prðjj ðn þ 1Þj; ðn 1Þk; kÞ4Prðkj ðn þ 1Þj; ðn 1Þk; kÞ

(A.10)

and Prðjj ðn þ 2Þj; nk; kÞ4Prðkj ðn þ 2Þj; nk; kÞ

(A.11)

Using Bayes’ rule, (A.10) and (A.11) hold if and only if Prððn þ 1Þj; ðn 1Þk; kj jÞ4Prððn þ 1Þj; ðn 1Þk; kj kÞ

(A.12)

and Prððn þ 2Þj; nk; kj jÞ4Prððn þ 2Þj; nk; kj kÞ

(A.13)

Informational Cascades and the Decision to Migrate

43

Let us ﬁrst consider the case where nþ1 people have emigrated to country j, n1 people have emigrated to country k, and an individual has received a signal to emigrate to country k: ððn þ 1Þj; ðn 1Þk; kÞ. Given (A.8) and (A.9) it is at most the case that n1 people have emigrated to country j and n1 people have immigrated to country k: ððn 1Þj; ðn 1ÞkÞ, otherwise we would have had a herd behavior when the event ðnj; ðn 2ÞkÞ occurred and the event ððn þ 1Þj; ðn 1Þk; kÞ would have never occurred: Prððn þ 1Þj; ðn 1Þk; kj j Þ ¼ Prððn 1Þj; ðn 1ÞkÞPrð3j; 1k; kj jÞ

1 p2 qð1 qÞ

and Prððn þ 1Þj; ðn 1Þk; kj kÞ ¼ Prððn 1Þj; ðn 1ÞkÞPrð3j; 1k; kj kÞ

1 p2 qð1

qÞ

ðA:14Þ

where Prððn 1Þj; ðn 1Þkj jÞ ¼ Prððn 1Þj; ðn 1Þkj kÞ ¼ Prððn 1Þj; ðn 1ÞkÞ. Given (A.6) and (A.7), it is clear that (A.14) holds. In the same way we prove (A.11). Q.E.D.

References Banerjee, A.V. (1992), A simple model of herd behavior. Quarterly Journal of Economics 107 (3), 797–817. Bartel, A.P. (1989), Where do the new US immigrants live? Journal of Labour Economics 7 (4), 371–391. Bauer, T., Epstein, G.S., Gang, I.N. (2007), The inﬂuence of stocks and ﬂows on migrants’ location choices. Research in Labor Economics 26, 199–229. Bauer, T., Epstein, G.S., Gang, I.N. (2009), Measuring ethnic linkages between immigrants. International Journal of Manpower 30 (1/2), 56–69. Bikhchandani, S., Hirshleifer, D., Welch, I. (1992), A theory of fads, fashion, custom, and culture change as informational cascade. Journal of Political Economy 100 (5), 992–1026. Brezis, E.S., Krugman, P.R. (1996), Immigration, investment, and real wages. Journal of Population Economics 9 (1), 83–93. Carrington, W.J., Detragiache, E., Vishwanath, T. (1996), Migration with endogenous moving costs. American Economic Review 86 (4), 909–930. Chiswick, B.R., Miller, P.M. (1996), Ethnic networks and language proﬁciency among immigrants. Journal of Population Economics 9 (1), 19–35. Choi, P. (1997), Herd behavior, the ‘‘penguin effect’’ and the suppression of informational diffusion: an analysis of informational

44

Gil S. Epstein

externalities and payoff interdependency. Rand Journal of Economics 28 (3), 407–425. Church, J., King, I. (1983), Bilingualism and network externalities. Canadian Journal of Economics 26 (2), 337–345. Epstein, G.S. (2002), Labor market interactions between legal and illegal immigrants. Review of Development Economics 7 (1), 30–43. Epstein, G.S. (2008), Herd and network effects in migration decisionmaking. Journal of Ethnic and Migration Studies 34 (4), 567–583. Epstein, G.S., Cohen, O. (2006), Immigrants during 1990’s from former Soviet Union: herd effect and network externalities. The Economic Quarterly (in Hebrew) 53 (1), 166–201. Epstein, G.S., Hillman, A.L. (2003), Unemployed immigrants and voter sentiment in the welfare state. Journal of Public Economics 87, 1641–1655. Epstein, G.S., Hillman, A.L., Weiss, A. (1998), Creating illegal immigrants. Journal of Population Economics 12, 3–21. Gang, I.N., Rivera-Batiz, F.L., Yun, M.-S. (2002), Economic strain, ethnic concentration and attitudes towards foreigners in the European Union. Mimeo. Gottlieb, P. (1987), Making Their Own Way: Shorthorn Blacks’ Migration to Pittsburgh 1916–30. University of Illinois Press, Urbana. Grossman, J.R. (1989), Land and Hope: Chicago, Black Southerners, and the Great Migration. University of Chicago Press, Chicago. Marks, C. (1989), Farewell – We’re Good and Gone: The Great Black Migration. Indian University Press, Bloomington. Scharfstein, D.S., Stein, J.C. (1990), Herd behavior and investment. American Economic Review 80 (3), 465–479.

CHAPTER 3

The Measurement of Income Polarization by Ethnic Groups: The Case of Israel Population Joseph Deutsch Department of Economics, Bar-Ilan University, Ramat Gan, 52900, Israel E-mail address: [email protected]

Abstract Income polarization is a relatively new concept introduced in the literature of the measurement of income inequality. It has essential properties that may be used to measure relative deprivation and it adds another dimension to the measurement of income inequality concerned mainly with the middle income class (Esteban and Ray, 1994). No study, however, seems to have tried to decompose by population subgroups any of the polarization indices that have appeared in the literature. This study introduces a methodology that decomposes the polarization index recently suggested by Deutsch et al. (2007) by population subgroups. This polarization index is related to the Gini index and its components so that previous results on the decomposition of the Gini index may be applied. Two main cases are examined, that of nonoverlapping groups and overlapping groups. The paper also includes an empirical analysis based on Israeli data for the period 1990–2004, which covers the case of nonoverlapping (income) groups as well as that of overlapping groups, the latter being either Jews of Western and Eastern origin or Jews and Non-Jews. The empirical analysis shows a decrease in polarization over the period 1990–2002 and an increase in polarization during the years 2002–2004. Using the Shapley methodology we analyze the contribution of the different factors to the trend in polarization observed over time. Keywords: Gini index, inequality decomposition, Israel, polarization, population subgroups Jel classiﬁcation: D31

Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008009

r 2010 by Emerald Group Publishing Limited. All rights reserved

46

Joseph Deutsch

1. Introduction Numerous studies have appeared in the literature on the decomposition of income inequality by population subgroups. While certain authors preferred to use entropy-related inequality indices because they may be easily broken down into a between and a within groups inequality component (see, e.g., Bourguignon, 1979; Cowell, 1980, 1984; Mookherjee and Shorrocks, 1987; Shorrocks, 1980, 1984), others preferred using the very popular Gini index, despite the fact that the latter, in addition to between and within groups components, includes also a residual (see Bhattacharya and Mahalanobis, 1967; Dagum, 1980, 1987, 1997; Lambert and Aronson, 1993; Pyatt, 1976; Sastry and Kelkar, 1994; Silber, 1989; Yitzhaki and Lerman, 1991). It turns out, however, that it is possible to give intuitive interpretations to this residual and several have been proposed (see Dagum, 1997; Lambert and Aronson, 1993; Pyatt, 1976; Silber, 1989). No study, however, seems to have tried to decompose by population subgroups any of the polarization indices that have appeared in the literature. The purpose of this paper is precisely to show how it is possible to decompose by population subgroups the polarization index PG recently suggested by Deutsch et al. (2007). This polarization index is related to the Gini index and its components so that previous results on the decomposition of the Gini index may be applied. The paper is organized as follows. Section 2 shows how it is possible to decompose the polarization index PG when there are two nonoverlapping groups of equal size or three nonoverlapping groups, whatever their size. Section 3 extends the analysis to the case of overlapping groups while Section 4 presents an empirical analysis based on Israeli data for the period 1990–2004. This empirical illustration covers the case of nonoverlapping (income) groups as well as that of overlapping groups, the latter being either Jews of Western and Eastern origin or Jews and Non-Jews. 2. Measuring polarization when income groups do not overlap 2.1. The case of two groups of equal size In a recent paper, Deutsch et al. (2007) deﬁned a new index PG of bipolarization as 2PG ¼ ðGB GW Þ=GT

(1)

where GT, GB, and GW refer, respectively, to the Gini index for the whole income distribution and the between and within groups Gini indices. Note that in the case of bipolarization we assume that the population is divided into two groups of equal size, the ‘‘poor’’ who are those whose income is

The Measurement of Income Polarization by Ethnic Groups

47

lower than the median income and the ‘‘rich’’ who are those with an income higher than the median income. Let now GP and GR refer, respectively, to the Gini index among the ‘‘poor’’ and the ‘‘rich.’’ It is then easy to see (cf. Kendall and Stuart, 1969, for a general deﬁnition of the Gini index) that the indices GT, GB, GP, and GR may be expressed as GT ¼ ð1=2ÞðD=yÞ

(2)

GB ¼ ð1=2ÞðDB =yÞ

(3)

GP ¼ ð1=2ÞðDP =yP Þ

(4)

GR ¼ ð1=2ÞðDR =yR Þ

(5)

where D, DB, DP, and DR represent, respectively, the overall mean difference, the between groups mean difference, the mean difference within the group of poor, and the mean difference within the group of rich. yR ; and yP are, respectively, the (arithmetic) mean incomes in Similarly y; the whole population, in the subgroup of rich, and in that of the poor. Finally, in the case of nonoverlapping groups, the overall Gini index may be expressed (see Silber, 1989b) as GT ¼ GB þ GW

(6)

The within groups Gini index GW may, however, be written (see Silber, 1989) as GW ¼ f P sP GP þ f R sR GR

(7)

where fP, fR, sP, and sR refer, respectively, to the population shares of the groups of poor and rich and to the income shares of these two groups. Since we assumed that f P ¼ f R ¼ 1=2

(8)

and since in our case sP ¼ ð1=2ÞðyP =yÞ

(9)

and sR ¼ ð1=2ÞðyR =yÞ

(10)

we end up, combining expressions (7)–(10), with P þ ð1=4ÞðyR =yÞG R GW ¼ ð1=4ÞðyP =yÞG

(11)

The between groups Gini index GB in (3) may, however, be also expressed (see Silber, 1989) as GB ¼ ½ð1=2Þ; ð1=2ÞG½sR ; sP 0

(12)

where [(1/2), (1/2)] is a row vector giving the population shares of the ‘‘rich’’ and the ‘‘poor,’’ ½sR ; sP 0 is a column vector giving the income shares

48

Joseph Deutsch

of these two groups and G is here a two by two square matrix, called G-matrix (see Silber, 1989) whose typical element gij is equal to 0 if i ¼ j, 1 if jWi, and þ1 if joi. Combining (9), (10), and (12) we end up, after simplifying, with GB ¼ ð1=2ÞðyR y P Þ=ðyR þ yP Þ

(13)

Note also that the overall mean income y may be written, in the case of two groups of equal size, as y ¼ ð1=2ÞðyR þ yP Þ

(14)

If we now combine expressions (1), (6), (11), (13), and (14) we will derive PG ¼ f½ð1=2ÞðyR y P Þ=ðyR þ yP Þ ½ðð1=4ÞðyR GR þ y P GP ÞÞ=ðð1=2ÞðyR þ yP ÞÞg=GT

ð15Þ

where GT ¼ GB þ GW ¼ f½ð1=2ÞðyR yP Þ=ðyR þ yP Þ þ ½ðð1=4ÞðyR GR þ yP GP ÞÞ=ðð1=2ÞðyR þ yP ÞÞg

ð16Þ

Combining (15) and (16) we end up, after simplifying, with PG ¼ ½ðyR yP Þ ðyR GR þ yP GP Þ=½ðy R yP Þ þ ðyR GR þ yP GP Þ

(17)

PG ¼ ½ððyR =yP Þ 1Þ ððyR =yP ÞGR þ GP Þ= ½ððyR =yP Þ 1Þ þ ððyR =yP ÞGR þ GP Þ

ð18Þ

so that PG may also be expressed as PG ¼ f ððyR =yP Þ; GR ; GP Þ

(19)

Let now DPG ; DðyR =DyP Þ; DGR ; and DGP refer, respectively, to the changes that took place between two periods, say 0 and 1, in the values of the polarization index PG, the ratio of the mean incomes yR and yP , and the Gini indices GR and GP, we can then derive from (19) that DPG ¼ hðDðyR =DyP Þ; DGR ; DGP Þ

(20)

Using the so-called Shapley decomposition (see Shorrocks, 1999 and Appendix A) it is easy to determine the contributions of the changes DðyR =DyP Þ, DGR, and DGP to the overall change DGP in the value of the polarization index PG. 2.2. The case of three nonoverlapping income groups Since the Gini index will evidently not include any residual term (also called overlap) when income groups do not overlap, there is no reason why the deﬁnition of polarization given in (1) could not apply to the case where there are more than two groups. This is so because the main assumptions

49

The Measurement of Income Polarization by Ethnic Groups

underlying (1) as well as the concept of polarization remain valid, that is, that polarization should increase with the between groups inequality and decrease with the within groups inequality (see, e.g., Esteban and Ray, 1994; Wolfson, 1994, 1997; Wang and Tsui, 2000, or Chakravarty and Majumder, 2001). The present section will therefore derive expressions similar to those given in Section 2, but for the case of three groups. Assume we divide the population into three nonoverlapping income groups. Let fP, fM, and fR refer, respectively, to the shares in the total population of the ‘‘poor,’’ the ‘‘middle class,’’ and the ‘‘rich.’’ Let sP, sM, and sR refer to the corresponding income shares of these three groups. Using the algorithm mentioned in (12) the between groups Gini index GB will now be expressed as GB ¼ ½ðf R ; f M ; f P ÞG½sR ; sM ; sP 0

(21)

where G is now a three by three G-matrix. Let y P ; yM ; and yR refer to the mean incomes of the ‘‘poor,’’ the ‘‘middle class,’’ and the ‘‘rich.’’ Remember now that for any group g (g ¼ R, M, or P), we can deﬁne the income shares sg as sg ¼ f g ðyg =yÞ

(22)

where fg and yg refer, respectively, to the population share and the mean income of group g and that the mean income y in the whole population may be expressed as y ¼ f R yR þ f M yM þ f P yP

(23)

We then end up, combining (21), (22), and (23), after simplifying, with GB ¼ f½f R ð1 f R ÞyR þ ½ð1 f R f P Þðf P f R Þy M ½f P ð1 f P Þy P g=½f R yR þ f M yM þ f P yP GB ¼ fð1=yP Þf½f R ð1 f R ÞðyR =yP Þ þ ½ð1 f R f P Þðf P f R ÞðyM =yP Þ ½f P ð1 f P Þgg=½f R ðyR =yP Þ þ f M ðyM =yP Þ þ f P ð24Þ However, since f M ¼ 1 ðf R þ f P Þ

(25)

GB ¼ kðf R ; f P ; ðyR =yP Þ; ðyM =yP ÞÞ

(26)

Now it is well-known that the within groups Gini index GW (see Silber, 1989) will be expressed in our case as GW ¼ ½ðf R sR GR Þ þ ðf M sM GM Þ þ ðf P sP GP Þ

(27)

50

Joseph Deutsch

Combining expressions (22), (23), (25), and (27), we end up with GW ¼ ½ðf 2R yR GR Þ þ ðf 2M yM GM Þ þ ðf 2P yP GP Þ=½f R yR þ f M yM þ f p yP (28) GW ¼ ½ðf 2R ðyR =yP ÞGR Þ þ ðf 2M ðyM =yP ÞGM Þ þ ðf 2P GP Þ=½f R ðyR =yP Þ þ f M ðyM =yP Þ þ f p

ð29Þ

so that we may also express GW as a function q, that is, as GW ¼ qðf R ; f p ; ðyR =yP Þ; ðyM =yP Þ; GR ; GM ; GP Þ

(30)

Since the polarization index PG, in the case of nonoverlapping groups, may be written as PG ¼ ðGB GW Þ=ðGB þ GW Þ

(31)

we conclude, combining (26) and (30) that PG may be expressed as a function r with PG ¼ rðf R ; f p ; ðyR =yP Þ; ðyM =yP Þ; GR ; GM ; GP Þ

(32)

Using, as in the case of two groups, the sign D to refer to the change of a variable between two periods, we end up expressing the change DPG in polarization as a function s, that is, as DPG ¼ sðDf R ; Df p ; DðyR =yP Þ; DðyM =yP Þ; DGR ; DGM ; DGP Þ

(33)

However, if we assume that there is no change in the relative size of the three groups we may, in fact, write that DPG ¼ sðDðyR =yP Þ; DðyM =yP Þ; DGR ; DGM ; DGP Þ

(34)

Using the so-called Shapley decomposition (see Shorrocks, 1999 and Appendix A), it is then easy to determine the contributions of the changes DðyR =yP Þ; DðyM =yP Þ, DGR, DGM, and DGP to the overall change DPG in the value of the polarization index PG. 3. Measuring polarization when income groups do overlap The deﬁnition of the polarization index PG given in (1) should also apply to the case of overlapping groups, that is, to the case where groups are not deﬁned by their income level but by some other characteristics, for example, ethnicity and educational level. What, however, has to be remembered is that when there is overlap, the denominator of (1) will be expressed as GT ¼ GB þ GW þ OV

(35)

where OV, the third element on the R.H.S. of (35), is a measure of the degree of overlap between the distributions of income of the various

The Measurement of Income Polarization by Ethnic Groups

51

population subgroups. As shown by Silber (1989), OV may be deﬁned as OV ¼ ½e0 Gs ½e0 Gv

(36)

In (36), eu is a 1 by n row vector of the population shares of the various individuals. Therefore, if the observations are at the individual level and there are n individuals, each element of the row vector eu will be equal to (1/n). The letters s and v on the R.H.S. of (36) refer to n by 1 column vectors. The elements of the column vector s are the individual income shares and they are assumed to be ranked by decreasing values of these shares (see Silber, 1989). The column vector v also includes income shares but here the individual shares are ranked ﬁrst by decreasing values of the average income of the group to which the individuals belong, second within each group by decreasing values of the shares (see Silber, 1989). Finally G on the R.H.S. of (36) is an n by n G-matrix of which the deﬁnition was given previously. Using (1) the polarization index PG in the case of overlapping groups will be expressed as PG ¼ ðGB GW Þ=ðGB þ GW þ OVÞ

(37)

Note that as in the case of nonoverlapping groups, the polarization index PG will increase with the between groups inequality GB, decrease with the within groups inequality GW. In addition since the overlapping term OV appears only in the denominator of (37), the polarization index PG will also decrease with the amount of overlap, which also makes sense. It should, however, be stressed that the three components GB, GW, and OV are not independent. To understand this, we have to recall some results derived by Dagum (1980, 1997). Let m represent the number of population subgroups. The overall mean difference D may then be decomposed into the sum of two terms as D ¼ DA þ DW

(38)

where DA refers to what may be called the ‘‘across groups inequality’’ (see Dagum, 1960, 1997), while DW measures the within groups inequality, with DW ¼ ð1=n2 Þ

m XXX X

jyih yjk j

(39)

jyih yjk j

(40)

h¼1 i2h k¼h j2k

and DA ¼ ð1=n2 Þ

m XXX X h¼1 i2h kah j2k

the second subindex (h or k) in (39) and (40) referring to the group to which the individual belongs.

52

Joseph Deutsch

Let us now assume that the groups are ranked by decreasing values of the average of the income in each group so that yh ; the mean income in group h, is higher than yhþ1 , the mean income in group hþ1. Equation (40) may then be written as (41)

DA ¼ Dd þ Dp with Dd ¼ ð1=n2 Þ

m XXX X

jyih yjk j with

yih yjk

(42)

jyih yjk j with

yih yjk

(43)

h¼1 i2h kah j2k

and Dp ¼ ð1=n2 Þ

m XXX X h¼1 i2h kah j2k

Combining (42) and (43) we derive that Dd Dp ¼ ð1=n2 Þ

m XXX X ðyih yjk Þ

(44)

h¼1 i2h kah j2k

" # m X X X X Dd Dp ¼ ð1=n Þ ðnk ðyih ÞÞ ðnh ðyjk ÞÞ 2

h¼1 kah

Dd Dp ¼ ð1=n2 Þ

m X X

i2h

(45)

j2k

½nk nh ðy h y k Þ

(46)

h¼1 kah

where nh and nk represent, respectively, the number of individuals in groups h and k. Since the between groups mean difference DB is obtained by giving each individual the average value of the incomes of the group to which he belongs, we may deﬁne the index DB as DB ¼ ð1=n2 Þ

m X X ½nk nh ðyh yk Þ

(47)

h¼1 kah

and it is easy to observe, when comparing (46) and (47), that DB ¼ ðDd Dp Þ

(48)

Since (41) indicates that DA ¼ ðDd Dp Þ þ ð2Dp Þ

(49)

we conclude, combining (38) and (49), that D ¼ DW þ DB þ ð2Dp Þ

(50)

One should note that (50) indicates that (2Dp), the residual obtained in the traditional decomposition of the mean difference by population

The Measurement of Income Polarization by Ethnic Groups

53

subgroups, is expressed as a simple function of the ‘‘transvariations,’’1 which exist between all pairs of population subgroups. Combining (38), (49), and (50) we may now express the mean difference as D ¼ DW þ Dd þ Dp

(51)

The three elements on the R.H.S. of (51) are now independent from each other whereas those on the R.H.S. of (50) were not, given the deﬁnition of DB that appears in (48). Combining now (2), (3), (35), (37), (38), (41), and (51) we end up with PG ¼ ½Dd ðDW þ Dp Þ=½Dd þ ðDW þ Dp Þ

(52)

It is then easy to observe that PG will increase with Dd but decrease with Dp and Dw. All these three properties make sense and correspond certainly to our intuition of the concept of polarization. Finally using (52), we may express the polarization index PG as a function r with PG ¼ rðDW ; Dd ; Dp Þ

(53)

Let us now use the letter d to indicate a change in a given variable. The change in the degree of polarization between two periods may therefore be written as dPG ¼ rðdDW ; dDd ; dDp Þ

(54)

Applying to (54) the Shapley decomposition procedure (see Appendix A), we will be able to determine the marginal impacts of changes in DW, Dd, and Dp on the overall change DPG in polarization during the period examined. 4. An empirical illustration 4.1. The case of nonoverlapping groups The empirical analysis is based on the income surveys that are conducted each year in Israel. In the ﬁrst stage we have limited our analysis to nonoverlapping groups. Three cases were distinguished. In the ﬁrst one, we assumed that the population was divided in two groups of equal size, 1 Following Gini (1959) we may say that there exists a ‘‘transvariation’’ between two distributions {xi} and {yj} with respect to their (arithmetic, geometric, etc.) means mx and my when among the nxny possible differences (xiyj), the sign of at least one of them is different from that of the expression (mx my), nx, and ny being the number of observations in these two distributions. The importance of such a ‘‘transvariation’’ may be measured in several ways (see DeutschRand Silber, R1997). The reference here is to the moment m1 of order 1, which is deﬁned þ1 y as m1 ¼ 1 gðyÞdy 1 ðy xÞf ðxÞdx where g(y) and f(x) are the densities of y and x.

54

Joseph Deutsch

those with incomes lower and higher than the median income. In the second case, we assumed that the poor were those belonging to the two poorest deciles of the income distribution, the richest those belonging to the two highest deciles of this distribution and the middle class those belonging to the other six deciles. Finally, in the third case, we deﬁned the poor as those belonging to the two poorest deciles, the rich as those belonging to the richest decile, and the middle class those belonging to any of the other seven deciles. Three indicators were computed for each of the years (1990–2004) for which observations were available: the Gini index for the whole income distribution, the polarization PG deﬁned previously and the polarization index KZ proposed by Zhang and Kanbur (2001). In the latter case we assumed that the inequality index selected was Gini rather than an entropy-related index. The results are given in Table 1. Let us ﬁrst examine the case of two nonoverlapping groups of equal size. Table 1 indicates clearly that the polarization indices and the Gini index do not measure the same thing since the indices PG and KZ often move in a direction that is opposite to that observed for the Gini index. There have been ﬂuctuations over time in the value of the polarization indices, but as a whole there seems to have been a decrease in polarization between 1990 and 2001 and an increase between 2002 and 2004.

Table 1. Year

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Yearly values of the Gini index GT and of the two polarization indices PG and KZ Case of two nonoverlapping groups of equal size

Case of three nonoverlapping groups, the poorest and richest each representing 20% of the population and the middle class 60%

Case of three nonoverlapping groups, the poorest representing 20%, the richest 10%, and the middle class 60% of the population

GT

PG

KZ

PG

KZ

PG

KZ

0.4193 0.4250 0.4389 0.4216 0.4441 0.4347 0.4330 0.4407 0.4379 0.4467 0.4413 0.4463 0.4503 0.4409 0.4412

0.3937 0.3845 0.3786 0.3853 0.3697 0.3684 0.3730 0.3603 0.3575 0.3506 0.3560 0.3463 0.3429 0.3580 0.3665

2.2988 2.2492 2.2186 2.2536 2.1733 2.1663 2.1898 2.1265 2.1131 2.0797 2.1056 2.0594 2.0438 2.1151 2.1570

0.5930 0.5987 0.6035 0.6005 0.6007 0.6109 0.6056 0.6066 0.6104 0.6102 0.6110 0.6114 0.6101 0.6030 0.6041

3.9146 3.9833 4.0435 4.0066 4.0087 4.1400 4.0705 4.0842 4.1332 4.1313 4.1419 4.1463 4.1300 4.0384 4.0519

0.3797 0.3832 0.3971 0.3858 0.3985 0.4039 0.3982 0.4176 0.4189 0.4236 0.4174 0.4282 0.4322 0.4217 0.4104

2.2245 2.2428 2.3170 2.2564 2.3248 2.3552 2.3233 2.4339 2.4418 2.4700 2.4330 2.4979 2.5223 2.4587 2.3923

55

The Measurement of Income Polarization by Ethnic Groups

In the case where three nonoverlapping groups are distinguished representing, respectively, 20%, 60%, and 20% of the population polarization seems here to have increased between 1990 and 2001 and decreased afterwards. Finally, in the third case where the share in the total population of the poor, the middle class, and the rich are, respectively, 20%, 70%, and 10% we also observe an increase in polarization between 1990 and 2001–2002 and a decrease between 2002 and 2004. To better understand the differences between the three cases examined, we have presented in Table 2, for each of the three cases, the between and within groups Gini index since the PG as well as the KZ polarization indices are simple functions of the between and within groups Gini indices. In the case where only two groups are distinguished, Table 2 indicates clearly that the increase in polarization observed between 2002 and 2004 is essentially the consequence of the decrease in the within groups component of the Gini index. On the contrary when three groups are distinguished, Table 2 indicates that the trend toward an increasing degree Table 2.

Yearly values of the overall Gini index (GT) and of the between and within groups Gini indices GB and GW

Year

Case of two nonoverlapping groups of equal size

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Case of three Case of three nonoverlapping nonoverlapping groups, the poorest groups, the poorest and richest each representing 20%, the representing 20% of richest 10%, and the the population and middle class 60% of the middle class 60% the population

Overall Gini index GT

Between Within groups Gini groups index GB Gini index GW

Between groups Gini index GB

Within groups Gini index GW

Between groups Gini index GB

Within groups Gini index GW

0.4193 0.4250 0.4389 0.4216 0.4441 0.4347 0.4330 0.4407 0.4379 0.4467 0.4413 0.4463 0.4503 0.4409 0.4412

0.2922 0.2942 0.3025 0.2920 0.3041 0.2974 0.2973 0.2997 0.2972 0.3016 0.2992 0.3004 0.3024 0.2994 0.3014

0.3340 0.3397 0.3519 0.3374 0.3554 0.3501 0.3476 0.3540 0.3526 0.3596 0.3555 0.3596 0.3626 0.3534 0.3539

0.0853 0.0853 0.0870 0.0842 0.0887 0.0846 0.0854 0.0867 0.0853 0.0870 0.0858 0.0867 0.0878 0.0875 0.0873

0.2893 0.2939 0.3066 0.2921 0.3105 0.3051 0.3027 0.3123 0.3106 0.3179 0.3128 0.3187 0.3225 0.3134 0.3111

0.1300 0.1311 0.1323 0.1295 0.1336 0.1295 0.1303 0.1283 0.1272 0.1287 0.1285 0.1276 0.1279 0.1275 0.1301

0.1271 0.1308 0.1364 0.1296 0.1399 0.1373 0.1358 0.1409 0.1407 0.1450 0.1421 0.1459 0.1480 0.1415 0.1398

56

Joseph Deutsch

Table 3.

Correlations coefﬁcient between the Gini index GT and the polarization indices PG and KZ GT

PG

(1) The case of two nonoverlapping groups of equal size GT 1.0000 0.8975 0.8975 1.0000 PG KZ 0.9010 0.9997

KZ

0.9010 0.9997 1.0000

(2) The case of three nonoverlapping groups, the poorest and richest each representing 20% of the population and the middle class 60% GT 1.0000 0.7152 0.7127 0.7152 1.0000 0.9999 PG KZ 0.7127 0.9999 1.0000 3) The case of three nonoverlapping groups, the poorest representing 20%, the richest 10%, and the middle class 70% of the population GT 1.0000 0.8812 0.8761 0.8812 1.0000 0.9997 PG KZ 0.8761 0.9997 1.0000

of polarization between 1990 and 2001 or 2002 is mainly the consequence of an increase in the between groups Gini index, this being true for both cases where three groups have been distinguished. In Table 3, we computed the Pearson correlation coefﬁcients over the years for which data were available between both polarization indices and the Gini index. As expected the correlation between the PG and KZ indices is extremely high (0.99). When only two groups are distinguished there is a strong negative correlation between the overall Gini index and the two polarization indices. However, when three groups are distinguished the correlation between the Gini index and the two polarization indices is positive and quite high (above 0.7). It is specially high when the rich are assumed to represent 10% of the population. In Tables 4–6, we present the results of the Shapley decomposition of the change over time in the value of the polarization index PG, the results being given separately for the 1990–2001 period and for the 2002–2004 period and for the three different divisions of the population (in nonoverlapping group) that have been selected. When only two groups of equal size are distinguished (Table 4), the decrease in polarization between 1990 and 2001 and the increase in polarization between 2002 and 2004 appear to be essentially the result of changes over time in the Gini index among the ‘‘rich.’’ The rise in the dispersion of incomes among the rich during the ﬁrst period led to a decrease in polarization, while the decrease in this dispersion during the second period led to a rise in polarization. The picture is more complex when three groups are distinguished. In Table 5, where the rich and the poor represent each 20% of the population, the increase in polarization that occurred during the period

The Measurement of Income Polarization by Ethnic Groups

Table 4.

57

Shapley decomposition of the change in polarization: The case of two nonoverlapping groups of equal size

(1) The period analyzed: 1990–2001 (a) Ratio of the average income of the rich over that of the poor and Gini indices within the rich and poor subgroups GP GR Year yR =yP 1990 3.8113 0.2616 0.2523 2001 4.0100 0.2479 0.3027 (b) Decomposition of the polarization index PG 1990 2001 Component PG DPG DðyR =yP Þ 0.39364 0.34623 0.04741 0.01171

DGP 0.00443

DGR 0.06354

(2) The period analyzed: 2002–2004 (a) Ratio of the average income of the rich over that of the poor and Gini indices within the rich and poor subgroups GP GR Year yR =yP 2002 4.0603 0.2535 0.3064 2004 4.0359 0.2588 0.2846 (b) Decomposition of the polarization index DPG 2002 2004 Component DPG DðyR =y P Þ DPG 0.34292 0.36644 0.02352 0.00133

DGP 0.00160

DGR 0.02645

1990–2001 was ﬁrst due to an increase in the ratios of the average incomes of the rich and the middle class when compared to that of the poor. There was, however, a counter force, that of the increase in the dispersion of incomes among the rich that per se would have led to a decrease in polarization. During the period 2002–2004, the decrease in polarization was also mainly the consequence of the rise in the dispersion of incomes among the rich. Finally when the rich are assumed to represent 10% and the poor 20% of the population (see Table 6), the rise in polarization during the period 1990–2001 is mainly the consequence of an increased gap between the average incomes of both the rich and the middle class and that of the poor. The decrease in polarization during the period 2002–2004 also seems to be the consequence of what happened to between groups inequality since the main contribution to the change in polarization comes from a decreased gap between the average incomes of the rich and the middle class and that of the poor. 4.2. The case of overlapping groups Two cases have been examined. We ﬁrst limited the analysis to the Jewish population and made a distinction between ‘‘Easterners’’ and ‘‘Westerners.’’ ‘‘Easterners’’ are Jews who were born in Asia or Africa or Jews born

58

Joseph Deutsch

Table 5.

Shapley decomposition of the change in polarization

(1) The period analyzed: 1990–2001 (a) Ratio of the average income of the middle class and the rich over that of the poor and Gini indices within the rich middle class and poor subgroups GP GM GR yM =yP yR =y P 1990 0.17651 3.67907 0.22770 10.34450 0.17692 2001 0.18422 3.40772 0.22306 10.97907 0.23693 (b) Decomposition of the polarization index PG PG0 0.59312 0.61138 PG1 DPG 0.01826 0.01831 Dðy M =yP Þ DðyR =yP Þ 0.01655 0.00026 DGP DGM 0.00492 DGR 0.02126 (2) The period analyzed: 2002–2004 (a) Ratio of the average income of the middle class and of the rich over that of the poor and Gini indices within the rich middle class and poor subgroups GM GR GP yM =yP yR =y P 2002 0.19110 3.50555 0.22368 11.37355 0.24610 2004 0.18482 3.62592 0.22978 11.21478 0.21398 (b) Decomposition of the polarization index PG PG0 0.61013 PG1 0.60414 DGP 0.00599 0.00770 Dðy M =yP ÞM DðyR =yP Þ 0.00372 DGP 0.00020 0.00614 DGM DGR 0.01138

The case of three nonoverlapping groups, the poorest and richest groups representing 20% and the middle class 60% of the population.

in Israel whose fathers were born in Asia or Africa. ‘‘Westerners’’ are Jews born in Europe, America, South Africa, New Zealand, or Australia, or Jews born in Israel whose fathers were born in one of those areas. At the second stage, we divided the population into two groups, Jews and NonJews. Table 7 gives the values of the Gini index and of the polarization indices PG and KZ during the period 1990–2000 when a distinction is made between ‘‘Westerners’’ and ‘‘Easterners’’ and during the period 2001–2004 when a distinction is made between Jews and Non-Jews.2 2

The periods covered in the two cases are not the same because during the 1990–2000 period we could not make a distinction between Jews and Non-Jews, while during the period 2001– 2004 no detailed data were available on the continent of birth of the Jews.

The Measurement of Income Polarization by Ethnic Groups

Table 6.

59

Shapley decomposition of the change in polarization

(1) The period analyzed: 1990–2001 (a) Ratio of the average income of the middle class and of the rich over that of the poor and Gini indices within the rich middle class and poor subgroups GP GM GR yM =yP yR =yP 1990 4.27442 0.26527 12.84149 0.15045 4.27442 2001 4.00204 0.26914 14.38926 0.21793 4.00204 (b) Decomposition of the polarization index DPG PG0 0.37974 0.42824 PG1 DPG 0.04849 0.02007 DðyM =yP Þ DðyR =yP Þ 0.04094 0.00022 DGP DGM 0.00571 DGR 0.00668 (2) The period analyzed: 2002–2004 (a) Ratio of the average income of the middle class and of the rich over that of the poor and Gini indices within the rich middle class and poor subgroups GM GR GP yM =yP yR =yP 2002 0.19110 4.11418 0.26902 14.97818 0.23640 2004 0.18482 4.24924 0.27198 14.43992 0.18649 (b) Decomposition of the polarization index PG PG0 0.43218 PG1 0.41043 DPG 0.02175 0.00975 Dðy M =yP Þ DðyR =yP Þ 0.01289 DGP 0.00017 0.00421 DGM DGR 0.00509

The case of three nonoverlapping groups, the poorest representing 20%, the richest 10% and the middle class 60% of the population.

Table 7 indicates that during the period 1990–2000 the Gini index (when the sample is limited to the Jewish population) ﬁrst increased, then decreased. No important change took place in the Gini index between 2001 and 2004 when its computation is based on the whole population. Table 7 also indicates that the polarization index PG was always negative, whatever the case and the year examined. This clearly indicates (see (54)) that the sum of the within groups mean difference (DW) and the component of the across groups mean difference that reﬂects overlapping (Dp) was greater in absolute value than the component of the across groups mean difference that does not correspond to overlapping (Dd). In the ﬁrst case, when the analysis is limited to the Jewish population (‘‘Easterners’’ versus ‘‘Westerners’’), it appears that as a whole polarization was less important at the end than at the beginning of the period since the index PG is more

60

Joseph Deutsch

Table 7. Year

Gini index GT

The case of overlapping groups

Polarization index PG

Component Dd

Component Dp

Component Dw

Case 1: Two nonoverlapping groups are distinguished: Jews of Western origin (‘‘Westerners’’) versus Jews of Eastern origin (‘‘Easterners’’). Period analyzed: 1990–2000. 1990 0.4234 0.4153 762 530 1,316 1991 0.4278 0.4255 883 626 1,564 1992 0.4427 0.4331 1,048 723 1,925 1993 0.4252 0.4816 919 811 1,815 1994 0.4477 0.4781 1,291 1,088 2,567 1995 0.4427 0.4596 1,568 1,205 3,028 1996 0.4412 0.4662 1,737 1,382 3,390 1997 0.4452 0.4768 2,161 1,803 4,296 1998 0.4382 0.4697 2,309 1,883 4,516 1999 0.4493 0.4495 2,638 1,930 5,018 2000 0.4387 0.4742 2,583 2,144 5,098 Case 2: Two nonoverlapping groups are distinguished: Jews versus Non-Jews. analyzed: 2001–2004. 2001 0.4463 0.7029 1,528 608 2002 0.4503 0.7114 1,506 504 2003 0.4409 0.6996 1,489 583 2004 0.4412 0.6989 1,511 506

Period 8,149 8,424 7,844 8,022

Yearly values of the Gini index GT, of the polarization index PG, and of the components Dd, Dp, and Dw.

negative around 2000 than around 1990. This is so because, as Table 7 shows, the increase between 1990 and 2000 in the components Dp and Dw was greater than that of the component Dd. In the second case, when a distinction is made between Jews and Non-Jews no clear conclusion may be drawn as far as changes in the polarization index PG are concerned. In Table 8, ﬁnally we give the ‘‘Shapley contributions’’ of the changes over time in the three determinants Dd, Dw, and DW to the overall change in the polarization index PG. Let us ﬁrst examine the case where a distinction is made, within the Jewish population, between ‘‘Easterners’’ and ‘‘Westerners.’’ It appears that polarization decreased between 1990 and 2000 (the polarization index PG decreased by 0.059), but this decrease appears to be the consequence of two conﬂicting forces. On one hand the change over time in the component Dd would per se have led to an increase in polarization but variations in the components DW and Dp would have led to a decrease in polarization. This seems to indicate that all the mean differences increased and in particular the overlapping component, hence the decrease in polarization. In the second case, where the population is divided into Jews and NonJews, the changes during the period 2001–2004 were not very important but here again the important role is played by the components Dp and DW rather than by the component Dd.

Value of the polarization index in the initial year (P0G )

Value of the polarization index in the ﬁnal year (P1G )

Change in the value Contribution of of the polarization the change in the index during the component d(CDd) period examined ðDPG Þ

Contribution of the change in the component p(CDp)

Contribution of the change in the within groups inequality GW(CDW)

The case of overlapping groups: Shapley decomposition of the change in polarization

Case 2: Two nonoverlapping groups are distinguished: Jews versus Non-Jews. Period analyzed: 2001–2004. 2001 2002 0.7029 0.7114 0.0085 0.0036 2002 2003 0.7114 0.6996 0.0118 0.0028 2003 2004 0.6996 0.6989 0.0007 0.0038 2001 2004 0.7029 0.6989 0.0040 0.0027

0.0029 0.0023 0.0023 0.0030

0.0078 0.0168 0.0054 0.0038

Case 1: Two nonoverlapping groups are distinguished: Jews of Western origin (‘‘Westerners’’) versus Jews of Eastern origin (‘‘Easterners’’). Period analyzed: 1990–2000. 1990 1991 0.4153 0.4255 0.0102 0.0602 0.0198 0.0506 1991 1992 0.4255 0.4331 0.0077 0.0697 0.0165 0.0609 1992 1993 0.4331 0.4816 0.0485 0.0519 0.0131 0.0165 1993 1994 0.4816 0.4781 0.0035 0.1303 0.0344 0.0924 1994 1995 0.4781 0.4596 0.0185 0.0758 0.0117 0.0456 1995 1996 0.4596 0.4662 0.0066 0.0403 0.0154 0.0315 1996 1997 0.4662 0.4768 0.0106 0.0847 0.0304 0.0649 1997 1998 0.4768 0.4697 0.0071 0.0257 0.0049 0.0137 1998 1999 0.4697 0.4495 0.0202 0.0526 0.0028 0.0297 0.0119 0.0044 1999 2000 0.4495 0.4742 0.0247 0.0084 0.3431 1990 2000 0.4153 0.4742 0.0590 0.4531 0.1690

Period

Table 8.

The Measurement of Income Polarization by Ethnic Groups 61

62

Joseph Deutsch

Appendix A. On the concept of Shapley decomposition Let an index I be a function of n variables and let ITOT be the value of I when all the n variables are used to compute I. I could, for example, be the R-square of a regression using n explanatory variables, any inequality index depending on n income sources or on n population subgroups. Now, let I k=k ðiÞ be the value of the index I when k variables have been dropped so that there are only (nk) explanatory variables and k is also the rank of variable (i) among the n possible ranks that variable (i) may have in the n! sequences corresponding to the n! possible ways of ordering n numbers. We will call I k=k1 ðiÞ the value of the index when only (k1) variables have been dropped and k is the rank of the variable (i). Thus, I 1=1 ðiÞ gives the value of the index I when this variable is the ﬁrst one to be dropped. Obviously there are (n1)! possibilities corresponding to such a case. I 1=0 ðiÞ then gives the value of the index I, when the variable (i) has the ﬁrst rank and no variable has been dropped. This is clearly the case when all the variables are included in the computation of the index I. Similarly I 2=2 ðiÞ corresponds to the (n1)! cases where the variable (i) is the second one to be dropped and two variables as a whole have been dropped. Clearly I 2=2 ðiÞ can also take (n1)! possible values. I 2=1 ðiÞ gives then the value of the index I, assuming only one variable was eliminated and the variable (i) has the second rank. Here also there are (n1)! possible cases. Obviously I n=ðn1Þ ðiÞ corresponds to the (n1)! cases where the variable (i) is dropped last and is the only one to be taken into account. If I is an inequality index, it will evidently be equal to zero in such a case. But if it is, for example, the R-square of a regression it would give us the R-square when there is only one explanatory variable, the variable (i). Obviously I n=n ðiÞ gives the value of the index I when variable (i) has rank n and abcd variables have been dropped, a case where I will always be equal to zero by deﬁnition since no variable is left. Let us now compute the contribution of Cj(i) of variable (i) to the index I, assuming the variable bcad is dropped when it has rank j. Using the previous notations, we deﬁne Cj(i) as

Cj ðiÞ ¼ ð1=n!Þ

ðn1Þ! X

½I j=ðj1Þ ðiÞ I j=j ðiÞh

(A.1)

h¼1

where the superscript h refers to one of the (n1)! cases where the variable i has rank j.

The Measurement of Income Polarization by Ethnic Groups

63

The overall contribution of variable (i) to the index I may then be deﬁned as CðiÞ ¼ ð1=n!Þ

n X

(A.2)

C k ðiÞ

k¼1

It is then easy to prove that I ¼ ð1=n!Þ

n X

(A.3)

CðiÞ

i¼1

Let us give a simple illustration where the index I depends on four determinants a, b, c, d. Table A1 below gives all the possible ways of ordering these four elements. As indicated previously the so-called Shapley decomposition looks at all possible elimination sequences. As Table A1 indicates, there are six cases where a appears in the ﬁrst position, six where it appears in the second, etc. If we look at the various ways of eliminating a, we can say that if a is eliminated ﬁrst its contribution to the indicator will be equal to the difference between the value of the indicator when all four determinants are different from zero and its value when a is equal to zero. In that case (a eliminated ﬁrst) the contribution of a will be written as Cða=a is eliminated firstÞ ¼ ½Iðaa0; ba0; ca0; da0Þ Iða ¼ 0; ba0; ca0; da0Þ Clearly, looking at the ﬁrst column of the table above, there are six such cases out of 24 possible orderings. The case where a is eliminated second is based on the orderings given in the second column of this table. There are three possibilities: b is eliminated ﬁrst and a second, so that the contribution of a in this case will be written as Cða=a is eliminated second and b firstÞ ¼ ½Iðaa0; b ¼ 0; ca0; da0Þ Iða ¼ 0; b ¼ 0; ca0; da0Þ and this case appears twice in the second column of the table. Table A1.

The 24 ways of ordering four elements

a appears in the ﬁrst position

a appears in the second position

a appears in the third position

a appears in the fourth position

abcd abdc acbd acdb adbc adcb

bacd badc cabd cadb dabc dacb

bcad bdac cbad cdab dbac dcab

bcda bdca cbda cdba dbca dcba

64

Joseph Deutsch

c is eliminated ﬁrst and a second so that the contribution of a in this case will be written as Cða=a is eliminated second and c firstÞ ¼ ½Iðaa0; ba0; c ¼ 0; da0Þ Iða ¼ 0; ba0; c ¼ 0; da0Þ and this case appears also twice in the second column of the table. d is eliminated ﬁrst and a second so that the contribution of a in this case will be written as Cða=a is eliminated second and d firstÞ ¼ ½Iðaa0; ba0; ca0; d ¼ 0Þ Iða ¼ 0; ba0; ca0; d ¼ 0Þ and this case appears also twice in the second column of the table. The case where a is eliminated third is based on the orderings given in the third column of this table. There are again three possibilities: b is eliminated ﬁrst, c second, and a third or c is eliminated ﬁrst, b second, and a third. In both cases the contribution of a will be written as Cða=a eliminated third and b and c firstÞ ¼ ½Iðaa0; b ¼ 0; c ¼ 0; da0Þ Iða ¼ 0; b ¼ 0; c ¼ 0; da0Þ b is eliminated ﬁrst, d second, and a third or d is eliminated ﬁrst, b second, and a third. In both cases the contribution of a will be written as Cða=a eliminated third and b and d first or secondÞ ¼ ½Iðaa0; b ¼ 0; ca0; d ¼ 0Þ Iða ¼ 0; b ¼ 0; ca0; d ¼ 0Þ c is eliminated ﬁrst, d second, and a third or d is eliminated ﬁrst, c second, and a third. In both cases the contribution of a will be written as Cða=a eliminated third and c and d first or secondÞ ¼ ½Iðaa0; ba0; c ¼ 0; d ¼ 0Þ Iða ¼ 0; ba0; c ¼ 0; d ¼ 0Þ Finally, the cases where a is eliminated last (fourth) appear in the fourth column of the table above. There are six such cases but in each of these cases the contribution of a may be expressed as Cða=a eliminated fourth while b; c and d first; second or thirdÞ ¼ ½Iðaa0; b ¼ 0; c ¼ 0; d ¼ 0Þ Iða ¼ 0; b ¼ 0; c ¼ 0; d ¼ 0Þ

The Measurement of Income Polarization by Ethnic Groups

65

Taking all the 24 cases into account, we therefore conclude that the overall contribution of a to the indicator I may be expressed as CðaÞ ¼ fð6=24Þ½Iðaa0; ba0; ca0; da0Þ Iða ¼ 0; ba0; ca0; da0Þ þ ð2=24Þ½Iðaa0; b ¼ 0; ca0; da0Þ Iða ¼ 0; b ¼ 0; ca0; da0Þ þ ð2=24Þ½Iðaa0; ba0; c ¼ 0; da0Þ Iða ¼ 0; ba0; c ¼ 0; da0Þ þ ð2=24Þ½Iðaa0; ba0; ca0; d ¼ 0Þ Iða ¼ 0; ba0; ca0; d ¼ 0Þ þ ð2=24Þ½Iðaa0; b ¼ 0; c ¼ 0; da0Þ Iða ¼ 0; b ¼ 0; c ¼ 0; da0Þ þ ð2=24Þ½Iðaa0; b ¼ 0; ca0; d ¼ 0Þ Iða ¼ 0; b ¼ 0; ca0; d ¼ 0Þ þ ð2=24Þ½Iðaa0; ba0; c ¼ 0; d ¼ 0Þ Iða ¼ 0; ba0; c ¼ 0; d ¼ 0Þ þ ð6=24Þ½Iðaa0; b ¼ 0; c ¼ 0; d ¼ 0Þ Iða ¼ 0; b ¼ 0; c ¼ 0; d ¼ 0Þg

One can naturally derive in a similar way the overall contributions C(b) of b, C(c) of c, and C(d) of d to the value of the indicator I. Moreover, it is also easy to verify that CðaÞ þ CðbÞ þ CðcÞ þ CðdÞ ¼ Iðaa0; ba0; ca0; da0Þ.

References Bhattacharya, N., Mahalanobis, B. (1967), Regional disparities in consumption in India. Journal of the American Statistical Association 62, 143–161. Bourguignon, F. (1979), Decomposable income inequality measures. Econometrica 47, 901–920. Chakravarty, S.R., Majumder, A. (2001), Inequality, polarization and welfare: theory and applications. Australian Economic Papers 40, 1–13. Cowell, F.A. (1980), On the structure of additive inequality measures. Review of Economic Studies 47, 521–531. Cowell, F.A. (1984), The structure of American income inequality. Review of Income and Wealth 30, 351–375. Dagum, C. (1980), Inequality measures between distributions with applications. Econometrica 48, 1791–1803. Dagum, C. (1987), Measuring the economic afﬂuence between populations of income receivers. Journal of Business and Economic Statistics 5, 5–12. Dagum, C. (1997), A new approach to the decomposition of the Gini income inequality ratio. Empirical Economics 22, 515–531. Deutsch, J., Hanoka, M., Silber, J. (2007), On the link between the concepts of Kurtosis and bipolarization. Economics Bulletin (36), 1–5.

66

Joseph Deutsch

Deutsch, J., Silber, J. (1997), Gini’s transvariazione and the measurement of distance between distributions, Empirical Economics 22, 547–554. Esteban, J.-M., Ray, D. (1994), On the measurement of polarization. Econometrica 62, 819–852. Gini, C. (1959), Memorie de Metodologia Statistica: Volume Secondo – Transvariazione, Libreria Goliardica, Roma. Kendall, M.G., Stuart, A. (1969), The advanced theory of statistics. Charles Griffen, London. Lambert, P.J., Aronson, J.R. (1993), Inequality decomposition analysis and the Gini coefﬁcient revisited. Economic Journal 103, 1221–1227. Mookherjee, D., Shorrocks, A.F. (1987), A decomposition of the trend in U.K. income inequality. Economic Journal 92, 886–902. Pyatt, G. (1976), On the interpretation and disaggregation of Gini coefﬁcients. Economic Journal 86, 243–255. Sastry, D.V.S., Kelkar, U.R. (1994), Note on the decomposition of Gini inequality. Review of Economics and Statistics LXXVI, 584–585. Shorrocks, A.F. (1980), The class of additive decomposable inequality measures. Econometrica 48, 613–625. Shorrocks, A.F. (1984), Inequality decomposition by population subgroups. Econometrica 50, 1369–1385. Shorrocks, A.F. (1999), Decomposition procedures for distributional analysis: a uniﬁed framework based on the Shapley value. University of Essex, Mimeo. Silber, J. (1989), Factor components, population subgroups and the computation of the Gini index of inequality. Review of Economics and Statistics 71, 107–115. Wang, Y.Q., Tsui, K.Y. (2000), Polarization orderings and new classes of polarization indices. Journal of Public Economic Theory 2, 349–363. Wolfson, M.C. (1994), When inequalities diverge. American Economic Review, Papers and Proceedings 84, 353–358. Wolfson, M.C. (1997), Divergent inequalities: theory and empirical results. Review of Income and Wealth 43, 401–421. Yitzhaki, S., Lerman, R.I. (1991), Income stratiﬁcation and income inequality. Review of Income and Wealth 37, 313–329. Zhang, X., Kanbur, R. (2001), What difference do polarisation measures make? An application to China. Journal of Development Studies 37, 85–98.

CHAPTER 4

The Effects of School Quality in the Origin on the Payoff to Schooling for Immigrants Barry R. Chiswicka,b and Paul W. Millerc a

Department of Economics, University of Illinois at Chicago, IL 60607-7107, USA IZA-Institute for the Study of Labor, Bonn, Germany E-mail address: [email protected] c School of Economics and Finance, Curtin University, Perth, WA 6845, Australia E-mail address: [email protected] b

Abstract The payoff to schooling among the foreign born in the United States is only around one-half of the payoff for the native born. This paper examines whether this differential is related to the quality of the schooling immigrants acquired abroad. The paper uses the overeducation/required education/undereducation speciﬁcation of the earnings equation to explore the transmission mechanism for the origin-country school-quality effects. It also assesses the empirical merits of two alternative measures of the quality of schooling undertaken abroad. The results suggest that a higher quality of schooling acquired abroad is associated with a higher payoff to schooling among immigrants in the US labor market. This higher payoff is associated with a higher payoff to correctly matched schooling in the United States, and a greater (in absolute value) penalty associated with years of undereducation. A set of predictions is presented to assess the relative importance of these channels, and the undereducation channel is shown to be the more inﬂuential factor. This channel is linked to greater positive selection in migration among those from countries with better quality schools. In other words, it is the impact of origin-country school quality on the immigrant selection process, rather than the quality of immigrants’ schooling per se, that is the major driver of the lower payoff to schooling among immigrants in the United States. Keywords: Immigrants, schooling, school quality, earnings, selectivity Jel classiﬁcations: I21, J24, J31, J61, F22

Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008010

r 2010 by Emerald Group Publishing Limited. All rights reserved

68

Barry R. Chiswick and Paul W. Miller

1. Introduction Studies of immigrant economic adjustment have placed considerable emphasis on the less-than-perfect international transferability of immigrants’ human capital. Starting with Chiswick (1978), this has been linked to the lower payoff to schooling for immigrants than for the native born. Chiswick (1978, p. 919) concluded: The smaller partial effect of schooling on earnings in the United States is an important ﬁnding. y The smaller effect of preimmigration schooling may be ‘‘explained’’ by country-speciﬁc aspects of the knowledge acquired in school, by a lower quality of foreign schooling, or by the poorer information it provides employers who use schooling as a screen y. The weaker partial effect of schooling may in part reﬂect self-selection in migration in which only the most able and most highly motivated of those with little schooling migrate, while those with (or who subsequently acquire) higher levels of schooling came from a broader ability and motivation spectrum. Empirical assessment of this important ﬁnding has proceeded along a number of lines. Chiswick and Miller (2008) use insights from the overeducation/required education/undereducation (ORU) literature (Hartog, 2000) to assess the possible contribution of self-selection in migration and the less-than-perfect international transferability of immigrants’ human capital. This is done indirectly through linking these two aspects of the immigrant adjustment process to the patterns observed in the payoffs to overeducation and undereducation. Chiswick and Miller (2008, p. 1339) argue: ‘‘The analysis also suggests that the two related issues of selectivity in migration and the international transferability of skills are both relevant, but their relative importance will vary by country of origin and educational attainment.’’ Bratsberg and Terrell (2002) and Betts and Lofstrom (2000) provide direct evidence on the effect that characteristics of the immigrants’ country of origin might have on the payoff to schooling in the United States. Bratsberg and Terrell (2002) link the payoff to schooling that the foreign born receive in the United States to measures of the resources devoted to education (namely, the pupil–teacher ratio and relative expenditure per pupil in immigrants’ country of origin), a measure of the commitment to education (namely, years of compulsory education in the country of origin), and a number of other variables that cover differences in the transferability of immigrants’ schooling to the US labor market (e.g., English as an ofﬁcial language in the origin labor market). They report (p.193): that differences in the attributes of educational systems account for most of the variation in rates of return to education earned by immigrants applying their source-country education in the U.S. labor

Effects of School Quality in the Origin on Earnings

69

market. We ﬁnd a particularly robust inverse relationship between the rate of return to education and the pupil–teacher ratio in primary schools in the source country, and similarly robust direct relationships between the rate of return and relative teacher wages and expenditures per pupil in the source country. Similar analyses by Betts and Lofstrom (2000, p. 102) led them to conclude: y the characteristics of the source country affect immigrants’ earnings substantially. Reductions in the pupil–teacher ratio and increases in the average level of educational attainment increase earnings of immigrants signiﬁcantly, but only for the most highly educated workers y. GDP per capita affects earnings positively for all immigrants, although it is the least well educated immigrants for whom the effect is the largest. Sweetman (2004) extends this latter line of inquiry by focusing on an outcome measure, test scores from international standardized tests, rather than on input variables from the education production function. Thus, in his analysis of immigrant earnings in Canada, Sweetman relates the birthplace differences in the payoff to schooling to differences in the country-level average test scores compiled by Hanushek and Kimko (2000). Sweetman (2004) reports that the country of origin differences in the payoff to schooling are related to this measure of school quality, although the R2 in the country-level regressions (of less than 0.2) were much lower than those reported by Bratsberg and Terrell (2002) where multiple input variables were used (of up to 0.84).1 In this chapter we merge the approaches of Chiswick and Miller (2008) and Sweetman (2004). Thus we quantify birthplace differences in the payoff to schooling in the United States using both conventional and ORU models of earnings determination. These birthplace differentials are then related to measures of the quality of the immigrant source-country human capital provided by the OECD Programme for International Student Assessment (or PISA) and the Hanushek and Kimko (2000) data previously used by Sweetman (2004) in his analysis of immigrants’ earnings in Canada.2 1 Hanushek and Kimko (2000) impute the majority of their country scores using educational input variables, and hence utilizing both the country-level average test scores and input variables in a single estimating equation has little merit. 2 This relates standardized partial effects of education to standardized test scores. The partial effects of education are standardized in the sense noted by Bratsberg and Terrell (2002, p. 179) ‘‘because the index is constructed on the basis of returns to education in a single market economy, it supplies a productivity-based estimate of the quality of educational institutions in foreign countries.’’

70

Barry R. Chiswick and Paul W. Miller

The structure of this chapter is as follows. Section 2 provides a brief account of the methods that are employed in the statistical analysis. Section 3 reviews the PISA and Hanushek and Kimko (2000) data. Empirical ﬁndings are presented in Section 4. A summary and conclusion are provided in Section 5. 2. Methodology Analyses of the birthplace differentials in the payoff to schooling have estimated both the conventional schooling and experience earnings equation and the ORU earnings equation. The conventional earnings equation relates the natural logarithm of a measure of earnings (hourly, weekly, annual) to years of schooling (EDUC), years of labor market experience (EXP) and its square, and other variables that are generally held to affect earnings, such as marital status, ofﬁcial language skills, and location, and, among the foreign born, years since migration and citizenship. That is: ln Y i ¼ b0 þ b1 EDUCi þ þ ni

(1)

The ORU modiﬁcation of this earnings equation disaggregates the measure of years of schooling into three terms, namely a term for the years of education which are usual or standard in the worker’s occupation, a term for any years of overeducation possessed by the worker, and a term for any years of undereducation. These terms for years of over- and undereducation are measured relative to the central tendency for education in the respondent’s occupation, which is what is referred to in the literature as the required, usual, or standard level of schooling. For simplicity, occupation is treated as exogenous. Speciﬁcally: ln Y i ¼ a0 þ a1 Over_Educi þ a2 Req_Educi þ a3 Under_Educi þ þ ui (2) where Over_Educ ¼ years of surplus or overeducation, Req_Educ ¼ the usual or reference years of education, Under_Educ ¼ years of deﬁcit or undereducation, and EDUC ¼ Over_EducþReq_EducUnder_Educ. Note that for each individual, ‘‘Over_Educ’’ and ‘‘Under_Educ’’ cannot both be positive.3 Either one or both must be zero. There are various ways of compiling a measure of ‘‘Req_Educ’’ (see Hartog, 2000; Chiswick and Miller, 2008). The measure used below is the modal educational attainment of workers in each of the approximately 500 occupations identiﬁed in the 2000 US Census. 3

It will be apparent that the standard earnings equation in (1), ln Y i ¼ b0 þ b1 EDUCi þ þ ui , forces a1 ¼ a2 ¼ |a3|. As this condition does not hold, the ORU speciﬁcation results in a higher R2 and a2Wb1.

Effects of School Quality in the Origin on Earnings

71

When (1) and (2) are estimated on separate samples of the native born and foreign born, considerable interest had been focused on differences by nativity in the estimates of the payoff to schooling and the coefﬁcients of the ORU variables. For the simple foreign-born/native-born dichotomy, the payoff to actual years of schooling for the foreign born is usually much less than the payoff to actual years of schooling for the native born. For example, in analyses of 2000 US Census data, Chiswick and Miller (2008) report that the payoff to schooling for the native born was 10.6 percent, while that for the foreign born was only 5.2 percent. They also show that this payoff varies appreciably by country of origin, being relatively high for immigrants from developed, English-speaking countries, and relatively low for immigrants from less developed, non-English-speaking countries. For example, the payoff to schooling was just 1.8 percent for immigrants from Mexico, 4.3 percent for immigrants from Cuba as well as those from Eastern Europe, but as high as 11 percent for immigrants from Canada. Chiswick and Miller (2008) also report that the payoffs to the ORU variables, though particularly the earnings effects of the undereducation and overeducation variables, also vary by country of origin. In the analyses that follow, these variations are linked to direct measures of the quality of schooling in the immigrants’ country of origin provided by the PISA and Hanushek and Kimko (2000) data. The country-level information on the quality of schooling is incorporated into the study of immigrants’ earnings using Card and Krueger’s (1992) two-step approach. This involves augmenting the usual regression model with birthplace-schooling interaction terms, and then relating the estimated birthplace differentials in the payoff to schooling to the PISA scores and Hanushek and Kimko’s (2000) human capital quality index in a second step or supplementary regression. The supplementary regressions may contain other country-level information, such as GDP per capita. This approach can be represented by two equations (for simplicity only the conventional schooling earnings equation and the PISA scores are considered here), namely: ln Y i ¼ b0 þ

J X

½b1j ðI j EDUCij Þ þ þ ni

i ¼ 1; . . . ; n

(3a)

j¼1

b1j ¼ a0 þ a1 PISAj þ þ Zj

j ¼ 1; . . . ; J

(3b)

where Ij is a vector of dichotomous variables with a value of 1 for each birthplace j, and zero otherwise, and b1j are the separate birthplace effects on the payoff to schooling. This model can be generalized through the inclusion of birthplace intercept shifts. That is, b0 can be replaced by PJ ½b I j¼1 0j j . The estimates in (3b) can be obtained using weighted least squares, where the weights are given by the sample sizes of workers from each country in the ﬁrst-step regression, or the inverse of the variances of

72

Barry R. Chiswick and Paul W. Miller

the slope estimates in the ﬁrst-step regression.4 Further details are provided in Section 5.

3. Country-level data Two measures of school quality are employed in the analyses that follow. The ﬁrst is provided by the reading, mathematics, and science scores for countries in the PISA. The second is the human capital quality indices compiled by Hanushek and Kimko (2000). The PISA is an international standardized assessment, coordinated by the OECD, to measure the outcomes of education systems. This assessment mechanism is administered every three years (ﬁrst conducted in 2000) to 15 year olds in schools of participating countries. Initially the assessment framework of the PISA covered performance only in reading, mathematics, and science.5 However, problem-solving skills, designed to assess cross-curriculum competencies, were also covered in the 2003 survey. The PISA covers both OECD (e.g., France, UK, Australia, and United States) and non-OECD (e.g., Brazil, Chile, Peru, and Thailand) countries. Further details are available from the PISA web site: www.pisa.oecd.org. The reading, mathematics, and science literacy scores from the 2000 PISA survey form the basis of the main set of analyses presented below. Reading literacy in the PISA is deﬁned as the ability to understand, to use, and to reﬂect on written texts in order to fulﬁll one’s goals, to develop one’s knowledge and potential, and to use written information to function or participate effectively in complex modern societies. Mathematical literacy is deﬁned in the PISA as the capacity to identify, understand, and engage in mathematics, as well as to use mathematical knowledge and skills in one’s life. These skills incorporate simple calculations, posing and solving mathematical problems in various situations, and being able to take a point of view and appreciate things expressed numerically. Scientiﬁc literacy is deﬁned in the PISA as the capability to use scientiﬁc knowledge, to identify questions/issues, and to draw evidence-based 4

The use of weighted least squares in the second step mimics the more formal random parameters model, a single equation representation of which is: ln Y ij ¼ b0 þ a0 EDUCij þ a1 PISAj EDUCij þ m1j EDUCij þ nij The random parameters model can be estimated using maximum likelihood methods. 5 The PISA also collects information on a wide range of factors thought to have a bearing on student performance, namely: (i) characteristics of individual students (e.g., their home background and learning approach); (ii) characteristics of schools (e.g., school/classroom atmosphere and school resources); and (iii) characteristics of school systems (e.g., the degree to which individual schools are given autonomy within the education system).

Effects of School Quality in the Origin on Earnings

73

conclusions in order to understand and help make decisions about the natural world and human interactions with it. Table 1 lists information on the mean reading, mathematics, and science literacy scores by country from the 2000 PISA. This table also includes an average score for the OECD. This score is computed using a simple average of the scores for all OECD countries. These scores have been normalized so that the OECD average is 500. The mean reading score for the United States, at 504, is only slightly above the 500-benchmark average across the OECD countries in the survey. There is considerable variation in the reading scores, with the standard deviation of the scores in Table 1 being 54. The reading literacy scores range from below 400 (Peru has a score of 327, Albania 349, Indonesia 371, Macedonia 373, and Brazil 396) to values over 525 (Finland has a score of 546, Canada 534, New Zealand 529, Australia 528, and Ireland 527). The reading score for Mexico, which is the largest source region for immigrants in the United States, is a relatively low 422. The mathematics literacy score for the United States is 493, below the OECD average, while the score for Mexico is 387, which represents a relatively weaker position in mathematics than that reported for reading literacy. The mathematics scores listed in Table 1 are characterized by greater variation than is the case for the reading score: The lowest mathematics score is the 292 for Peru and the highest is Hong Kong’s 560. The range in the scores is thus 268 points, compared to the range of 219 points for reading literacy. Brazil also has a relatively low mathematics score (334), as does Indonesia (367). Countries other than Hong Kong with relatively high mathematics scores are Japan (557) and Korea (547). The standard deviation of the PISA mathematics scores across countries is 65, which is somewhat higher than the standard deviation of the PISA reading scores across countries of 54. There is, however, a very high correlation between the reading and mathematics scores, with the Pearson correlation coefﬁcient between the values in Table 1 being 0.95. The science literacy scores range from Peru’s value of 333 through to the 552 for Korea. Other countries with relatively low scores are Brazil (375), Albania (376), Indonesia (393), and Argentina (396). Other countries with relatively high scores are Japan (550), Hong Kong (541), Finland (538), the UK (532), Canada (529), Australia (528), and New Zealand (528). Thus, the range for the science literacy scores is 219, which is the same as for the reading literacy scores. The standard deviation of the science literacy scores in Table 1, at 53, is also similar to that for the reading scores. The science literacy scores are highly correlated with each of the other measures, with pair-wise correlation coefﬁcients of 0.97 in each instance. The science literacy score for the United States is 499, close to the OECD average. The science literacy score for Mexico is 422, the same distance from the OECD average as for the reading literacy score.

74

Barry R. Chiswick and Paul W. Miller

Table 1.

Mean PISA scores, 2000

Country

Reading

Mathematics

Science

Albania Argentina Australia Austria Belgium Brazil Bulgaria Canada Chile Czech Republic Denmark Finland France Germany Greece Hong Kong Hungary Iceland Indonesia Ireland Israel Italy Japan Korea Latvia Liechtenstein Luxembourg FYR Macedonia Mexico New Zealand Norway Peru Poland Portugal Russian Federation Spain Sweden Switzerland Thailand United Kingdom United States OECD average

349 418 528 507 507 396 430 534 410 492 497 546 505 484 474 525 480 507 371 527 452 487 522 525 458 483 441 373 422 529 505 327 479 470 462 493 516 494 431 523 504 500

381 388 533 515 520 334 430 533 384 498 514 536 517 490 447 560 488 514 367 503 433 457 557 547 463 514 446 381 387 537 499 292 470 454 478 476 510 529 432 529 493 500

376 396 528 519 496 375 448 529 415 511 481 538 500 487 461 541 496 496 393 513 434 478 550 552 460 476 443 401 422 528 500 333 483 459 460 491 512 496 436 532 499 500

Source: Literacy skills for the world of tomorrow – further results from PISA 2000 (OECD and UNESCO Institute for Statistics, 2003).

Effects of School Quality in the Origin on Earnings

75

These country data on student performance in reading, mathematics, and science are positively correlated with typical indicators of economic progress or educational status. For example, the correlation of the country test scores with GDP per capita is between 0.61 (science) and 0.68 (reading). The correlation of the country test scores with educational expenditure per student is between 0.70 and 0.79, for the subgroup of 29 countries for which the educational expenditure data are available. Note, however, that while these correlation coefﬁcients are quite high, the correlations are far from perfect, suggesting that the average test scores may have information content on the school-quality differences across countries that varies from the information in the input variables used in previous studies.6 Hanushek and Kimko (2000) base their measure of human capital quality on six international tests of student achievement in mathematics and science undertaken between 1965 and 1991.7 A total of 26 performance series were collected, and converted to a common scale. Country averages were then obtained for the scores available for each country. Scores for 39 countries were compiled this way. Then these scores were related to a number of input variables, including the primary school enrollment rate, pupil–teacher ratio in primary school, and expenditure on education, and the estimates of this educational quality production function used to infer quality scores for a further 51 countries.8 The data for Hanushek and Kimko’s (2000) preferred human capital quality series are presented in Table 2. This table also contains information on whether the data for a particular country were imputed using the procedure described above. The mean score on the Hanushek and Kimko (2000) quality index is 45.18. There is considerable variation across countries in the scores. There are scores below 25 and scores above 65, and the standard deviation is 13.25. Countries with scores below 25 are Iran (18.26), Kuwait (22.50), Papua New Guinea (22.58), Bahrain (23.19), Chile (24.74), and Central Africa (24.77). Countries with scores above 65 are Singapore (72.13), Hong Kong (71.85), New Zealand (67.06), and Japan (65.50). Thus the score for the United States, at 46.77, is slightly above the overall mean.

6

Random measurement error could also result in the correlation coefﬁcients being less than 1. Four of these tests were administered by the International Association for the Evaluation of Educational Achievement and two by the International Assessment of Educational Progress. 8 Hanushek and Kimko (2000) use the human capital quality variable in cross-country growth regressions. Estimation of models based only on countries with observed human capital quality indicators, and with the broader sample that includes countries where the variable is imputed, led Hanushek and Kimko (2000, p.1196) to conclude ‘‘The estimates using this augmented sample conﬁrm the appropriateness of projection to the expanded set of countries.’’ They also compare a number of their imputed scores with evidence from independent tests, and again conﬁrm the appropriateness of the imputation procedure. 7

76

Barry R. Chiswick and Paul W. Miller

Table 2.

Hanushek and Kimko’s (2000) human capital quality index

Country

Imputed score

Score

Country

Imputed score

Score

Algeria Argentina Australia Austria Bahrain Barbados Belgium Bolivia Botswana Brazil Cameroon Canada Republic of Central Africa Chile China Colombia Congo Costa Rica Cyprus Denmark Dominican Republic Ecuador Egypt El Salvador Fiji Finland France West Germany Ghana Greece Guyana Honduras Hong Kong Hungary

| |

28.06 48.50 59.04 56.61 23.19 59.80 57.08 27.47 31.71 36.60 42.36 54.58 24.77

Kenya Republic of Korea Kuwait Lesotho Luxembourg Malaysia Malta Mauritius Mexico Mozambique Netherlands New Zealand Nicaragua

|

29.73 58.55 22.50 51.95 44.49 54.29 57.14 54.95 37.24 27.94 54.52 67.06 27.30

24.74 64.42 37.87 50.90 46.15 46.24 61.76 39.34 38.99 26.43 26.21 58.10 59.55 56.00 48.68 25.58 50.88 51.49 28.59 71.85 61.23

Nigeria Norway Panama Papua New Guinea Paraguay Peru Philippines Poland Portugal Singapore South Africa Spain Sri Lanka Swaziland Sweden Switzerland Syria Taiwan Thailand Togo Trinidad and Tobago Tunisia Turkey Uruguay UK USA USSR Venezuela Yugoslavia Zaire Zambia Zimbabwe

Iceland India Indonesia Iran Iraq Ireland Israel Italy Jamaica Japan Jordan

| | | | | | |

| | | | | | | | | |

| | | |

| | |

|

51.20 20.80 42.99 18.26 27.50 50.20 54.46 49.41 48.62 65.50 42.28

Source: Hanushek and Kimko (2000); Table C.1.

| | | | | |

|

| | | |

| |

|

| | | | |

| | | | |

38.90 64.56 46.78 22.58 39.96 41.18 33.54 64.37 44.22 72.13 51.30 51.92 42.57 40.26 57.43 61.37 30.23 56.31 46.26 32.69 46.43 40.50 39.72 52.27 62.52 46.77 54.65 39.08 53.97 33.53 36.61 39.64

77

Effects of School Quality in the Origin on Earnings

Table 3.

Correlation coefﬁcients between 2000 PISA scores and PISA scores for 2003 and 2006

Score for 2000

Score for 2003

Score for 2006

Reading Mathematics Science

0.955 0.979 0.948

0.927 0.970 0.943

Note: Correlations based on 29 observations for both the 2000–2003 and 2000–2006 comparisons.

The score for Mexico, at 37.24, is about one-half of a standard deviation below the mean. The Hanushek and Kimko (2000) quality index, being based on standardized tests undertaken between 1965 and 1991, appear to have an advantage over the PISA scores for 2000 in that they relate to a period when many of the immigrants in the US labor market in 2000 would have been enrolled in school in their country of origin. The extent of this advantage will depend on the magnitude of the across-country variation in the intertemporal changes in school quality. Where such variation is modest, the PISA data might be preferred, as these data relate to single tests for a speciﬁc age group, whereas the Hanushek and Kimko (2000) data are averages for a number of age groups, test types, and years of test assessment. There are two pieces of evidence that may be advanced on this. First, PISA scores are also available for 2003 and 2006, and one can therefore look at the relatedness of the scores for 2000 and those for these later years, although this is a short time span. Correlation coefﬁcients between the PISA scores for 2000 and 2003/2006 (listed in Table 3) indicate that there is a very high degree of stability in the PISA scores across time, at least for the six years covered in this presentation. Second, Hanushek and Kimko (2000, Fig. 1) present a visual display of test scores for various countries across time, ranging from 1965 to 1991 (a time span of 26 years). This also conveys the clear impression of stability in the relative standing of various countries with respect to student achievement. As Hanushek and Kimko (2000, p. 1186) state in relation to their Fig. 1: The test performance in Fig. 1 provides some evidence about the stability (over time) of scores. The United States and United Kingdom participate in all six testing programs. Throughout the period, the United Kingdom consistently performs a little better than the United States. Further, with a few exceptions, countries that outperform either the United States or United Kingdom on one test also tend to do so when they participate in other tests and vice versa.

78

Barry R. Chiswick and Paul W. Miller

There is no a priori way of evaluating the relative merits of the two data series, and hence both are used in the analyses below. There are 32 countries for which there are both PISA scores and a value for the Hanushek and Kimko (2000) human capital quality index.9 The simple correlation coefﬁcients between the Hanushek and Kimko (2000) index (covering 1965 to 1991) and the PISA reading, mathematics, and science scores (for 2000) for this group of countries are 0.774, 0.765, and 0.777, respectively. This, like the correlations for the PISA scores for 2000, 2003, and 2006, suggests only modest across-country variation in intertemporal changes in school quality. In other words, the standardized tests of 15 year olds in 2000 should provide an extremely useful measure of across-country differences in student achievement as far back as 1965. To minimize any unintended consequence associated with the use of the contemporary school-quality data, they are entered into the second step of the model along with per capita GDP data for each country. These per capita GDP information are deﬁned with respect to 1980. The use of a 20-year lag in this analysis follows Bratsberg and Terrell (2002. p. 182) who argue ‘‘We lag the educational quality data by 20 years to better capture differences in school quality at the time immigrants undertook their schooling y.’’10 The changes in the estimated effects of the PISA variables as the per capita GDP data are included in the model will inform on whether the contemporary PISA scores are a proxy for origin-country characteristics linked to school quality 20 years ago. Finally, as a further way of ascertaining the nature of the effects captured by the PISA data for 2000, the sample used in the statistical analysis can be restricted to the one-quarter (or other fraction) of immigrants with the most recent exposure to the origin-country school system.11 Results from this extension are discussed in the following section. 4. Empirical assessment The estimating equation used in the ﬁrst step of the assessment of the reasons behind the differences by country of origin in the payoff to schooling in the United States is a standard human capital earnings equation ((1) above). In particular, using data from the 2000 US Census, the natural logarithm of annual earnings in 1999 for males aged 25–64 who had nonzero earnings in that year is related to educational attainment, 9

Only nine of these countries have imputed values in the Hanushek and Kimko (2000) index. Betts and Lofstrom (2000), who use a single-equation approach in which origin-country information is interacted with the immigrant’s preimmigration level of education, reference their variables to the time when the immigrant would have been 10 years old. 11 This sample selection is based on the gap between the immigrant’s age and an assumed school-leaving age associated with their highest grade of secondary or primary schooling. 10

Effects of School Quality in the Origin on Earnings

79

potential labor market experience (computed using the proxy Age Years of Schooling 6), the natural logarithm of weeks worked, dummy variables for married (spouse present), race, US armed forces veteran status, resident of a metropolitan area, resident of a southern state, and English language skills, and, among the foreign born, variables for duration of residence in the United States and citizenship. The data are described in detail in Chiswick and Miller (2010). For the foreign born, the main set of analyses are based on immigrants aged 18 or more at the time of arrival in the United States. This is to ensure that the individuals will typically have completed secondary school in their country of origin, as this is the level that the school-quality data refer to. Deﬁnitions of the variables are presented in Appendix A. The Card and Krueger (1992) two-step approach was applied using both the PISA scores in Table 1 and the larger number of countries (73) with information on the Hanushek and Kimko (2000) index (Table 2). These separate estimates suggested that the PISA scores had far greater information content for understanding the variation in the payoff to schooling that immigrants receive in the United States. For example, the R2 in the second step of the Card and Krueger (1992) two-step approach in aggregate-level models based on the Hanushek and Kimko (2000) data were very low: they were even lower than the values reported by Sweetman (2004), and only one-eighth of the R2 in some of the models based on the PISA scores. However, when the analyses were based on the smaller group of 32 countries for which there are both PISA and Hanushek and Kimko scores, the results from the alternative measures are comparable12: in models where the PISA scores are statistically insigniﬁcant, the Hanushek and Kimko (2000) index is also statistically insigniﬁcant. Where the alternative origin school-quality measures are both statistically signiﬁcant, the coefﬁcients are of the same sign. Moreover, the relative magnitudes of the estimated effects on the various payoffs (to actual years of schooling, years of required schooling, years of undereducation, and years of overeducation) are similar, regardless of whether the analysis is based on the Hanushek and Kimko (2000) index or the PISA reading, mathematics, or science literacy scores. This similarity in ﬁndings presumably follows from the high simple correlation (above 0.76) between the alternative measures noted in Section 3. Given the similarity in statistical ﬁndings, any preference between the measures can be made on other grounds. As the standardized PISA scores 12

The difference in the results from analyses for this smaller group of countries and from analyses for all countries with Hanushek and Kimko scores may be associated with either the greater prevalence of imputed values of the Hanushek and Kimko index when using the larger sample (see footnote 9), or simply different roles for origin-country school quality for the purged countries.

80

Barry R. Chiswick and Paul W. Miller

are for speciﬁc tests for 15 year olds in 2000, whereas the Hanushek and Kimko index is based on results from different tests, conducted on various age groups, and in various years, and the majority of which were imputed, ease of interpretation suggests a preference for the PISA scores. The remainder of this chapter, therefore, is based on the PISA scores. Selected ﬁndings from the analysis using the Hanushek and Kimko (2000) data are reported in Appendix B.

4.1. Aggregate-level analyses There is information in Table 1 on the PISA scores for 40 countries other than the United States. However, the sample of males aged 25–64 years who worked in the United States during 1999 does not contain any immigrants from Iceland, Liechtenstein, or Luxembourg. Hence the analyses below are based on the remaining 37 countries. Only ﬁndings from the second step of the model (i.e., estimation of (3b)) are presented here. There are two sets of results in Table 4 for each PISA score (Reading, Mathematics, Science). The ﬁrst, in column (i), is based on the payoff to schooling across birthplace groups without country ﬁxed effects in the ﬁrst-step regression (i.e., the intercept is simply b0). The second, in column (ii), is for the analogous set of analyses where the ﬁrst-step model takes P account of birthplace ﬁxed effects (i.e., the intercept is generalized to Jj¼1 ½b0j I j ). Table 4. Variable

Estimates from second step of two-step model, aggregate-level analyses Reading literacy Mathematics literacy Science literacy (i)

Constant

(ii)

(i)

0.100 0.173 0.079 (6.28) (4.73) (8.53) PISA/100 0.018 0.029 0.014 (4.42) (3.07) (5.45) 1980 GDP per capita/10,000 0.018 0.031 0.018 (3.88) (2.88) (4.42) Country ﬁxed effects in ﬁrst step No Yes No R2 0.744 0.598 0.785 Sample size 37 37 37

(ii)

(i)

(ii)

0.145 (6.81) 0.024 (4.07) 0.029 (3.04) Yes 0.655 37

0.091 0.162 (6.44) (5.05) 0.016 0.026 (4.36) (3.16) 0.020 0.034 (4.68) (3.43) No Yes 0.741 0.604 37 37

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the PJ ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by j¼1 ½b0j I j . The dependent variable for each model is the estimated partial effects of education for the countries for which there are PISA scores. Absolute values of t-statistics in parentheses.

Effects of School Quality in the Origin on Earnings

81

The precision of the estimates of the payoff to schooling will vary across countries. Therefore, weighted least squares is used to compute the secondstep equations, where the weights are the number of workers for each country of origin in the ﬁrst-step regressions. Hence, important immigrant source countries such as Mexico, Canada, and Korea are assigned relatively more weight than minor source countries such as Denmark and Latvia. An alternative set of weights that was investigated involved the inverse of the variances of the estimates of the birthplace interaction terms in the ﬁrst step. This alternative gives more weight to birthplace effects that are precisely estimated (e.g., for Mexico, Korea, and Russia) and less weight to birthplace effects that are estimated less precisely (e.g., for Belgium, Denmark, and New Zealand). The two sets of weights are highly correlated (correlation coefﬁcient of 0.983 for the column (i) speciﬁcation) and so similar results emerge. For simplicity, only those using the country sample sizes are reported here. For the ﬁrst-step regression for speciﬁcation (i), the payoffs to schooling for the 37 countries that are the focus of this analysis range from 2.7 percent (for Mexico) to 7.9 percent (for Japan), a range of 5.2 percentage points. The standard deviation of the differentials in the payoff to schooling across the 37 countries is 1.4 percent. According to the Table 4 column (i) results, the birthplace differences in the payoff to schooling are positively associated with both the country-level average PISA scores and with 1980 GDP per capita. Up to 79 percent of the variation in the payoffs to schooling is accounted for by the two regressors, with the level of explanation being highest for mathematical literacy and lowest for science literacy. In alternative estimations (not shown here), the 1980 per capita GDP variable was omitted from the model: this change to the model was associated with an increase in the partial effects of the PISA variables by between 50 and 56 percent. This suggests that the effects of the PISA variables in Table 4 are net of the effects of the level of economic development in the country of origin when many immigrants would have been attending school. Each 100-point increase in the PISA scores is associated with between 1.4 and 1.8 percentage points increase in the payoff to schooling in the United States. Hence a 200-point change in the PISA, which is about the range in the data, is associated with around 3.5 percentage points increase in the payoff to schooling. These relationships are described in Fig. 1 in the case of the PISA reading literacy scores.13 In the column (ii) results in Table 4, the ﬁrst-step regression has been augmented with 37 country ﬁxed effects. This less-restrictive speciﬁcation is associated with a greater spread in the estimated payoffs to schooling. 13 Given the broad similarity of the ﬁndings for reading, mathematics, and science, Fig. 1 contains information only on the relationship across countries between the payoff to schooling and reading literacy.

82

Barry R. Chiswick and Paul W. Miller (a) Without country fixed effects in first-step regression 0.09

Returns to Schooling

0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0 300

350

400

450

500

550

600

Reading Literacy (b) With country fixed effects in first-step regression 0.14

Returns to Schooling

0.12 0.1 0.08 0.06 0.04 0.02 0 300

350

400

450

500

550

600

Reading Literacy

Fig. 1. Relationship between the payoff to schooling and PISA reading literacy. (a) Without country ﬁxed effects in ﬁrst-step regression. (b) With country ﬁxed effects in ﬁrst-step regression. For example, the payoff for Mexico is now estimated to be 1.6 percent (compared with 2.7 percent with the common intercept) and that for Japan 8.8 percent (compared with 7.9 percent with the common intercept). The standard deviation of the estimates of the payoff to schooling is 3.2, over two times that when it is assumed that there is a common intercept, as in column (i). While this greater variation in the dependent variable in the second-step regression is associated with a smaller explanatory power of

Effects of School Quality in the Origin on Earnings

83

the model, the PISA scores and 1980 per capita GDP variable both remain highly signiﬁcant, with partial effects that are – following the greater range in the dependent variable – appreciably greater than under the ﬁrst speciﬁcation. Speciﬁcally, the effect of changes in the PISA scores range from 0.024 to 0.029, with the smallest and largest impacts again being associated with mathematics and reading literacy, respectively. The ﬁrst-step regression results were also derived in an alternative way to examine the robustness of the ﬁndings. Thus, the models were estimated without the approximately 51 percent of the data where there are no PISA scores. This change in the sample was also associated with a widening of the range in the estimated payoffs to schooling. It was also associated with a reduction in the explanatory power of the second step of the model compared to the results in Table 4, of around 15 percentage points for speciﬁcation (i) and by 2 to 4 percentage points for speciﬁcation (ii). The partial effects of the PISA variables (not reported here) following this change to the sample, however, were larger than in the benchmark models of Table 4. The analyses were also conducted on subsamples formed using the years since the immigrant would have attended school in the country of origin. Two subsamples were formed: the 25 percent of the original sample with the most recent exposure to the origin-country school system, and the remaining 75 percent. Some of the ﬁndings from this disaggregated analysis (particularly those based on the column (i) speciﬁcation in Table 4) showed that the models had greater explanatory power for immigrants with the most recent exposure to the origin-country school system, whereas other results from the disaggregated analysis (those based on the column (ii) speciﬁcation in Table 4) were contrary to this. This ambiguity presumably follows from the PISA scores offering a very useful measure of the across-country differences in school quality up to four decades ago. One issue that needs to be addressed in this preliminary set of aggregatelevel analyses relates to the role of Mexico. Mexico is the dominant source of immigrants in the United States. In the sample of adult males used above, 29.2 percent are from Mexico. Among immigrants from countries where there are PISA scores, 60.7 percent of the sample is from Mexico. Accordingly, the analyses can be dominated by this group, particularly where the second-step results are weighted by the size of the birthplace groups.14 There are various ways this issue can be assessed, for example, through conducting the analyses of Table 4 for the 36 countries other than Mexico, or undertaking the analyses without weights (so that Mexico counts the same as any other country). The latter approach is adopted here, as this will also provide the opportunity to illustrate the impact that

14

Antecol et al. (2003) have previously drawn attention to the important role that immigrants from Mexico can have in aggregate-level analyses for the foreign born.

84

Table 5. Variable

Barry R. Chiswick and Paul W. Miller

Estimates from second step of two-step model, aggregate-level analyses, without weights Reading literacy Mathematics literacy Science literacy (i)

(ii)

(i)

Constant

0.047 0.110 0.036 (3.55) (2.31) (3.48) PISA/100 0.010 0.023 0.007 (3.01) (1.97) (2.82) 1980 GDP per capita/10,000 0.012 0.008 0.013 (3.92) (0.73) (4.41) Country ﬁxed effects in ﬁrst step No Yes No R2 Sample size

0.699 37

0.268 37

0.691 37

(ii)

(i)

0.087 (2.32) 0.018 (1.90) 0.010 (0.96) Yes

0.040 0.106 (3.15) (2.32) 0.008 0.021 (2.59) (1.97) 0.014 0.011 (4.84) (1.11) No Yes

0.262 37

0.681 37

(ii)

0.267 37

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the PJ ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by j¼1 ½b0j I j . The dependent variable for each model is the estimated partial effects of education for the countries for which there are PISA scores. Absolute values of t-statistics in parentheses.

weighting has on the analyses. Table 5 replicates Table 4 for this set of analyses. The results in Table 5 are broadly the same as those reported in Table 4. The PISA scores remain as a statistically signiﬁcant determinant of the across-country variation in the payoff to schooling among immigrants in the United States. The 1980 GDP per capita variable, however, while having a positive impact in each equation, is signiﬁcant only for the ﬁrst-step equation without country ﬁxed effects; that is, the equation has a common intercept for all countries. In the model where the across-country variation in the payoff to schooling is obtained from the ﬁrst-step P equation with country ﬁxed effects (i.e., the intercept is generalized to Jj¼1 ½b0j I j ), these ﬁxed effects apparently capture all of the inﬂuence of the different stages of economic development of the origin on the earnings of immigrants in the United States (that is, this effect applies to immigrants of all levels of schooling). The analyses were also undertaken with the estimating equation for the second step augmented with a dummy variable for Mexico. This enables the distance of the data for Mexico from the regression line to be assessed. In these analyses, whether conducted using the PISA scores or the Hanushek and Kimko (2000) index, the variable for Mexico was associated with a signiﬁcant negative coefﬁcient, of around 2 percentage points. In other words, given the quality of the schooling in Mexico (as measured in this study), and the relative level of economic development of Mexico, immigrants from Mexico would need to gain an extra 2 percentage points payoff to their education in the US labor market (that is, it should be around ﬁve percent rather than three percent) in order to conform to the

Effects of School Quality in the Origin on Earnings

85

estimated pattern for other countries. The 2-percentage-point shortfall in the payoff to schooling for immigrants from Mexico may be associated with the illegal status in the United States of many from that country. These preliminary results provide strong support for the hypothesis that origin-country school quality, as proxied by the PISA scores, affects the payoff to schooling for immigrants in the United States. The evidence derived using the Hanushek and Kimko (2000) index, reported in Appendix B, reinforced this conclusion. This suggests that the lower payoff to schooling for immigrants in the United States reﬂects, in part, a lower quality of education acquired in the country of origin.

4.2. The role of age at migration Sweetman (2004) conducts analyses of the links between indicators of origin-country school quality and the payoff to immigrants’ schooling in Canada on subsamples deﬁned using age at migration. Sweetman (2004, p. 30) argues ‘‘If it is the quality of the education system that is driving these results, and not other factors, such as discrimination, then immigrants educated primarily in the Canadian system should not be affected by the source country school quality index.’’ He shows that the payoffs to schooling are greatest for those educated primarily in Canada, and smallest for the foreign-born educated abroad. The payoffs to schooling for those with a mix of preimmigration and postimmigration schooling were of intermediate size. Origin-country school quality had no impact on the payoffs to schooling in Canada among immigrants educated primarily in Canada, whereas the payoff to schooling in Canada for immigrants mostly educated abroad was positively related to origin-country school quality.15 In the current study the analyses were repeated for several child immigrant groups. Selected results by age at immigration are presented in Table 6. The ﬁrst set of results presented in this table is the benchmark set of ﬁndings for adult immigrants, from Tables 4 and 5. The other sets of results are for the two samples of child immigrants, namely those who arrived before their tenth birthday, and the more restrictive deﬁnition of those who arrived before their sixth birthday. Two sets of analyses are presented in this table: the ﬁrst (on the left-hand side) is based on the second-step regression models that are weighted according to the number of workers in each country of origin, and the second (on the right-hand side) is from unweighted regressions. The weighted regressions (where considerable weight is given to Mexico) indicate that the school-quality effects are at least as strong 15

Bratsberg and Terrell (2002) focus only on those who were likely to have obtained their education abroad.

86

Table 6.

Barry R. Chiswick and Paul W. Miller

Estimate of PISA effect from second step of two-step model, by age at migration, weighted and unweighted regressions

Variable

With weights Reading Mathematics (i) (ii)

Without weights Science (iii)

Age at migration 18 or more (from Tables 4 and 5) PISA/100 0.029 0.024 0.026 (3.07) (4.07) (3.16) R2 Age at migrationr10 PISA/100 R2 Age at migrationr5 PISA/100 R2 Sample size

Reading (iv)

Mathematics (v)

Science (vi)

0.023 (1.97)

0.018 (1.90)

0.021 (1.97)

0.598

0.655

0.604

0.268

0.262

0.267

0.037 (3.25)

0.031 (4.48)

0.034 (3.68)

0.010 (0.81)

0.013 (1.23)

0.016 (1.36)

0.653

0.714

0.675

0.064

0.087

0.095

0.024 (1.95)

0.023 (2.90)

0.023 (2.25)

0.004 (0.27)

0.005 (0.39)

0.009 (0.59)

0.594 37

0.638 37

0.607 37

0.055 37

0.058 37

0.063 37

Notes: The ﬁrst-step regression is the ﬂexible speciﬁcation where the intercepts are given PJ by j¼1 ½b0j I j . Second-step regression also includes 1980 GDP per capita variable. The dependent variable for each model is the estimated partial effects of education for the countries for which there are PISA scores. Absolute values of t-statistics in parentheses.

among child immigrants as they are among adult immigrants (Table 6, columns i, ii, iii). This suggests that factors other than pure school-quality effects must also be playing a role. Below, we consider one of these, selectivity in migration among less-well educated immigrants (many of whom will be from Mexico). In the unweighted regressions, however, the PISA variables are statistically insigniﬁcant (Table 6, columns iv to vi). The PISA variable is also insigniﬁcant for these two ‘‘child immigrant’’ samples if weighted regressions are estimated on the 36 countries other than Mexico. That is, when Mexico is excluded from the sample, there is evidence that school-quality effects on the payoff to schooling dissipate as younger age-at-migration cohorts are considered. Thus, school quality in the origin is not relevant for the payoff to schooling in the United States for those who migrate as young children and therefore have little or no exposure to school quality in the origin. 4.3. Reference education, overeducation and undereducation, and PISA scores It has been shown here that immigrants from countries that perform poorly on standardized tests are associated with lower payoffs to schooling

Effects of School Quality in the Origin on Earnings

87

in the United States. Chiswick and Miller (2008) link the low payoff to schooling among the foreign born in the United States to a lower payoff to immigrants’ schooling that is surplus to the standard in their occupations, and to a lower penalty to years of undereducation among immigrants compared to the native born. This section examines the links between the returns to immigrants’ overeducation and undereducation and school quality, as measured by the PISA scores. Chiswick and Miller (2008) show that the payoff to schooling in the conventional earnings equation can be linked to the estimated effects on earnings of the education variables in the ORU model. In particular, greater estimated partial effects of the reference education and overeducation variables are shown to be associated with a higher payoff to education in the conventional earnings equation. A more negative earnings effect of undereducation is also associated with a higher payoff to schooling in the conventional human capital earnings model. To quantify the link between the ORU and conventional earnings equations in the current study of origin-country school-quality effects, it is ﬁrst necessary to estimate the ORU model (i.e., estimate (2) as the ﬁrst step in the two-step approach). Then the analyses reported above need to be repeated, replacing in the second step the payoff to schooling from the conventional (ﬁrst step) earnings function with the payoffs to ORU from the ORU speciﬁcation of the earnings function. Table 7 presents results from the second step of the model, where the variations across birthplaces in the payoffs to years of overeducation are related to the PISA scores. The structure of this table is the same as Table 4. These results show that the payoffs to overeducation are not affected by the quality of the origin-country schooling, as measured by the PISA scores.16 The insigniﬁcance of this relationship implies that years of surplus schooling among immigrants are relatively poorly rewarded in the US labor market, irrespective of the quality of the origin-country schooling system. Perhaps this arises because most of the years of surplus schooling were done at an age older than the age at which the PISA scores are measured. Years of surplus schooling among the native born are also poorly rewarded in the US labor market (see Chiswick and Miller, 2008).17 Table 8 presents information on the links between the payoff to the reference levels of schooling and the quality of immigrants’ origin-country schooling, as indexed by the PISA variables. In this instance the estimated

16 As shown in Appendix B, the payoffs to years of overeducation are also not related to the Hanushek and Kimko (2000) index, or to the PISA scores in an alternative sample considered in Appendix B. 17 In Chiswick and Miller’s (2008) aggregate-level analysis, the payoff to years of surplus schooling was 5.6 percent for the native born and 4.4 percent for the foreign born. For each birthplace group the payoff to years of schooling that were usual in the occupation of employment was around 15.5 percent.

88

Table 7. Variable

Barry R. Chiswick and Paul W. Miller

Estimates from second step of two-step model, aggregate-level analyses, focus on overeducation Reading literacy Mathematics literacy Science literacy (i)

(ii)

(i)

Constant

0.034 0.003 0.029 (1.38) (0.09) (1.87) PISA/100 0.004 0.007 0.003 (0.63) (0.94) (0.69) 1980 GDP per capita/10,000 0.023 0.031 0.023 (3.29) (3.93) (3.46) Country ﬁxed effects in ﬁrst step No Yes No R2 Sample size

0.415 37

0.370 37

0.416 37

(ii)

(i)

0.012 (0.71) 0.005 (0.96) 0.031 (4.06) Yes

0.028 0.000 (1.32) (0.02) 0.003 0.007 (0.46) (1.21) 0.024 0.031 (3.68) (4.29) No Yes

0.370 37

0.412 37

(ii)

0.380 37

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by PJ j¼1 ½b0j I j . The dependent variable for each model is the estimated partial effects of education for the countries for which there are PISA scores. Absolute values of t-statistics in parentheses.

Table 8. Variable

Estimates from second step of two-step model, aggregate-level analyses, focus on required education Reading literacy Mathematics literacy Science literacy (i)

(ii)

(i)

Constant

0.063 0.137 0.048 (4.88) (3.62) (6.24) PISA/100 0.012 0.014 0.009 (3.50) (1.42) (4.07) 1980 GDP per capita/10,000 0.013 0.033 0.013 (3.34) (2.98) (3.69) Country ﬁxed effects in ﬁrst step No Yes No R2 Sample Size

0.664 37

0.456 37

0.693 37

(ii)

(i)

0.127 (5.48) 0.013 (1.96) 0.031 (3.01) Yes

0.057 0.129 (4.93) (3.88) 0.010 0.012 (3.39) (1.39) 0.014 0.034 (4.01) (3.38) No Yes

0.483 37

0.659 37

(ii)

0.455 37

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by PJ j¼1 ½b0j I j . The dependent variable for each model is the estimated partial effects of education for the countries for which there are PISA scores. Absolute values of t-statistics in parentheses.

89

Effects of School Quality in the Origin on Earnings

Table 9. Variable

Estimates from second step of two-step model, aggregate-level analyses, focus on undereducation Reading literacy Mathematics literacy Science literacy (i)

(ii)

(i)

Constant

0.079 0.081 0.069 (2.44) (2.55) (3.47) PISA/100 0.015 0.016 0.013 (1.77) (1.99) (2.45) 1980 GDP per capita/10,000 0.009 0.008 0.007 (0.92) (0.89) (0.77) Country ﬁxed effects in ﬁrst step No Yes No R2 Sample size

0.238 37

0.265 37

0.292 37

(ii)

(i)

0.067 (3.52) 0.014 (2.68) 0.006 (0.76) Yes

0.078 0.079 (2.75) (2.87) 0.014 0.016 (2.00) (2.24) 0.009 0.009 (1.09) (1.07) No Yes

0.323 37

0.255 37

(ii)

0.285 37

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the P ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by Jj¼1 ½b0j I j . The dependent variable for each model is the estimated partial effects of education for the countries for which there are PISA scores. Absolute values of t-statistics in parentheses.

partial effect of the PISA scores on the differentials in the payoffs to schooling is signiﬁcant in the majority of the models.18 Hence, a 200-point increase in a speciﬁc PISA score is associated with an increase in the payoff to the reference years of schooling of up to 2.6 percentage points. The partial effects in Table 8 are, however, smaller than the partial effects in Table 4 for actual years of schooling. Recall that the payoff to a year of reference schooling is a payoff to the acquisition of that year of schooling and to moving to an occupation where the extra year of schooling is the usual or reference level. The relatively smaller partial effects in Table 8 suggest that the effect on earnings of the occupational mobility is hardly enhanced by the quality of schooling acquired abroad. Table 9 presents the estimated relationships between the wage effects of years of undereducation across birthplaces and the quality of the schooling acquired abroad. When interpreting these effects it is useful to bear in mind what the negative estimated coefﬁcient on the undereducation variable means. It indicates that a worker who obtains a job in an occupation that has a usual or reference level of education greater than the worker’s actual level of schooling receives a lower wage than the workers in the same occupation who have the usual or reference level of education.

18

The Hanushek and Kimko (2000) index is a statistically signiﬁcant determinant of the variation across countries in the payoff to required years of education (see Appendix B). The PISA scores are also statistically signiﬁcant in each of the models of the determination of the variation in the payoff to years of required education in the alternative sample considered in Appendix B.

90

Barry R. Chiswick and Paul W. Miller

Table 9 results indicate that the wage disadvantage to these undereducated workers rises with the PISA score.19 That is, a foreign-born worker who obtained his schooling abroad in a lower quality school system has a smaller earnings disadvantage than a foreign-born worker who obtained his schooling abroad in a higher quality school system. Undereducated native-born workers are shown by Chiswick and Miller (2008) to have a greater earnings disadvantage than the comparable foreign born.20 Hence, Table 9 results indicate that undereducated foreignborn workers educated abroad in a higher quality school system are more like undereducated native-born workers than are undereducated foreignborn workers educated abroad in a lower quality school system. Chiswick and Miller (2008) link the differential between the native born and foreign born in the earnings effects of undereducation to self-selection in immigration. This argument drew upon Chiswick (1978, p. 912), who suggested that ‘‘Suppose that among those with little schooling only the most able and most highly motivated migrate, while among those with high levels of schooling the immigrants are drawn more widely from the ability distribution.’’ The ﬁndings here in relation to the quality of schooling suggest a generalization of Chiswick’s (1978) argument, to ‘‘Suppose that among those from countries with a poorer quality of school system only the most able and most highly motivated migrate, while among those from countries with a higher quality school system the immigrants are drawn more widely from the ability distribution.’’ The variations in the earnings effects of each of the ORU variables are related to the PISA scores in ways that will lead to the payoff to actual years of schooling being positively related to the PISA scores. The relative importance of the relationships summarized in Tables 7 to 9 in this regard can be assessed using a method based on Chiswick and Miller (2008). This involves using the estimates from the ORU model to predict earnings for workers, and then relating the means of these predictions at each level of actual education to the years of actual education in a linear regression model, weighted by the number of workers at each level of education. The coefﬁcient on the years of actual education variable in this later regression is an estimate of the conventional payoff to schooling. The estimated earnings effects of the ORU variables in Tables 7 to 9 are ﬁrst evaluated at values of the PISA scores that generate an implied payoff to schooling that is the same as the actual payoff for the foreign born who migrated at age 18 or over (4.9 percent).21 The estimated effect of the ORU 19 Similar ﬁndings arise when the Hanushek and Kimko (2000) index is used, or the PISA scores are applied in alternative samples – see Appendix B. 20 The estimated partial effects of the undereducation variable in the aggregate-level analysis in Chiswick and Miller (2008) were 0.067 for the native born and 0.021 for the foreign born. 21 The payoff to schooling for all the foreign born (i.e., including those who immigrated before age 18) is 5.2 percent (see Chiswick and Miller, 2008).

91

Effects of School Quality in the Origin on Earnings

Table 10.

Implied payoffs to schooling, adjusting for effects of ORU variables at various PISA reading scores

i. Native born Foreign born: ii. No adjustment iii. Adjustment only to the earnings effects of reference education for the foreign born iv. Adjustment only to the earnings effects of overeducation for the foreign born v. Adjustment only to the earnings effects of undereducation for the foreign born vi. Adjustment to all three ORU variables

100 PISA points

Benchmark

þ100 PISA points

10.5

10.5

10.5

– 4.6

4.9 4.9

– 5.2

5.0

4.9

4.8

3.9

4.9

5.9

3.7

4.9

6.1

variables can then be evaluated at other values of the PISA scores (e.g., benchmark 7100 points, which will yield a 200-points range, similar to that in the PISA scores) and the simulation exercise described above repeated to assess how the PISA scores impact the payoff to schooling in the conventional earnings equation through each of the ORU variables. Table 10 presents ﬁndings from this analysis based on the PISA reading scores. The ﬁrst row of Table 10 contains the implied payoff to schooling for the native born. This does not vary with the PISA score, and so is recorded at 10.5 percent in each column. The second row presents the implied payoff to schooling for the foreign born. This has been computed from the predictions of the ORU model, calibrated to produce the actual payoff to schooling for this birthplace group of 4.9 percent. The payoff to schooling for the foreign born who immigrated at age 18 or older is thus less than one-half that for the native born. The third row of Table 10 explores the impact of variation in the PISA scores through the estimated effects of the reference years of education in the ORU model. A change up (down) in the PISA reading score of 100 points is associated with an increase (decrease) of around 0.3 percentage point in the payoff to schooling. As shown in the fourth row of the table, adjustment for the estimated effects of the overeducation variable has minimal effect on the payoff to schooling (the effect is just 0.1 percentage point). However, with the adjustment for undereducation, as seen from the ﬁfth row of the table, a change up (down) in the PISA reading score of 100 points is associated with an increase (decrease) in the payoff to schooling of about 1 full percentage point. The far greater effect of the PISA scores via the undereducation variables is consistent with Chiswick and Miller’s (2008) inference that the earnings effects of undereducation are

92

Barry R. Chiswick and Paul W. Miller

the more important contributor to the lower payoff to schooling for immigrants in the US labor market. This effect is linked in their analysis to more intense selection in migration among those with lower levels of schooling. In the ﬁnal row of Table 10 the roles of changes in the PISA scores via all the ORU variables are considered simultaneously. These show that at 100 higher PISA scores the implied payoff to schooling is 6.1 percent compared to 4.9 percent at the immigrant benchmark, but still less than the 10.5 percent for the native born. Thus, these ﬁndings show that the quality of schooling acquired abroad matters to the payoff to the schooling that immigrants receive in the United States. However, while some of the effects appear to operate in the expected way – by increasing the payoff to correctly matched schooling – the most important effect appears to operate by altering the selectivity of immigrants at low levels of schooling where undereducation is relatively more important. Hence, immigrants from countries with higher quality school systems, as proxied by the PISA scores, have a more negative earnings effect associated with undereducation. This leads them to be more like the native born in terms of earnings determination. The interpretation of this offered above is that these relatively less well-educated immigrants from countries with high-quality school systems are less intensely selfselected for migration to the United States. Analyses of the effects that the PISA mathematics and science scores have on immigrants’ payoffs to schooling via the earnings effects in the ORU model were also undertaken. Similar ﬁndings emerge, which demonstrates the robustness of the results. Relevant ﬁndings are presented in Appendix C.

5. Conclusion The payoff to schooling for immigrants in the US labor market is only around one-half of that for the native born. This chapter examines whether this difference is linked to the quality of the schooling acquired abroad by immigrants, and if so, how the school-quality effects are transmitted to earnings in the United States. The analyses offer a comparative assessment of the relative strengths of two measures of the quality of immigrants’ origin-country schooling, the PISA scores, and the Hanushek and Kimko (2000) human capital quality index. As argued above, the Hanushek and Kimko data relate to a period when many of the immigrants in the US labor market in 2000 would have been enrolled in school in their country of origin, whereas the PISA scores relate to testing undertaken in the origin countries in 2000. However, the PISA data relate to single tests for a speciﬁc age group, whereas the Hanushek and Kimko (2000) data are averages for a number of age groups, test types,

Effects of School Quality in the Origin on Earnings

93

and years of test assessment. Yet the two test scores are highly correlated across countries. The results suggest that from the perspective of predicting the payoff to preimmigration schooling among adult male immigrants in the United States, the PISA scores are relevant indicators of origin-country school quality.22 There is a strong, positive relationship between the payoff to schooling for immigrants in the US labor market and the quality of the schooling they acquired prior to immigration, as measured by the PISA reading, mathematics, and science literacy scores. Moreover, the results suggest that a higher quality of schooling acquired abroad is associated with a higher payoff to correctly matched schooling in the United States, a slightly higher payoff to schooling that appears to be surplus of the usual standards in the jobs held by immigrants, and a greater (in absolute value) penalty associated with years of undereducation. The predictions presented suggest that the latter phenomenon is of greater importance to understanding the lower payoff to schooling among the foreign born in the United States. Chiswick and Miller (2008) associate the differential in the earnings penalty for undereducation between the native born and the foreign born with positive selection in immigration among the foreign born. The results in this chapter suggest that immigrants from countries with a poorer quality of school system are associated with more intense selection in immigration, and it is this selection process, rather than the quality of immigrants’ schooling per se, that is the major driver of the lower payoff to schooling among immigrants in the United States. Acknowledgments We thank Derby Voon for research assistance, and Charles Beach for comments on an earlier paper that led to the research reported here. Miller acknowledges ﬁnancial assistance from the Australian Research Council. Appendix A. Deﬁnitions of variables The variables used in the statistical analyses are deﬁned below. Data source: 2000 Census of Population, Public Use Microdata Sample, 5 percent sample of the foreign born, and 0.15 percent random sample of the native born. The foreign-born sample is restricted to those who were at least 18 years of age at the time of immigration.

22

Similar results emerge using the Hanushek and Kimko (2000) index.

94

Barry R. Chiswick and Paul W. Miller

Deﬁnition of Population: Native-born and foreign-born employed men aged 25–64 years who had nonzero earnings in 1999. Dependent variables Earnings in 1999 Explanatory variables PISA

Natural logarithm of earnings in 1999 (where earnings are deﬁned as gross earnings from all sources).

The mean score for the immigrant’s country of origin from the OECD Programme for International Student Assessment. Separate scores for reading, mathematics, and science literacy are used. GDP per capita in 1980 Data on real GDP per capita for 1980 were obtained from Version 6.2 of the Penn World Tables (Alan Heston, Robert Summers, and Bettina Aten, Penn World Table, Version 6.2, Center for International Comparisons of Production, Income and Prices at the University of Pennsylvania, September 2006). These data are denominated in a common currency so that realquantity comparisons can be made across countries. Years of education This variable records the total years of full-time equivalent education. It has been constructed from the Census data on educational attainment by assigning the following values to the Census categories: completed less than ﬁfth grade (2 years); completed ﬁfth or sixth grade (5.5); completed seventh or eighth grade (7.5); completed ninth grade (9); completed tenth grade (10); completed 11th grade (11); completed 12th grade, no diploma (11.5); completed high school (12); attended college for less than one year (12.5); attended college for more than one year or completed college (14); bachelor’s degree (16); master’s degree (17.5); professional degree (18.5); doctorate (20). As with other Census data, the values for educational attainment are self-reported responses. While academic degrees may have required different years of schooling for immigrants educated in some countries of origin, US values are used in the analysis. Usual level of education This variable records the reference years of education. It is constructed using the modal level of education of the native-born workers in the respondent’s occupation of employment based on the Realized Matches procedure. Years of overeducation The overeducation variable equals the difference between the person’s actual years of education and the years of education required for the person’s job where this computation is positive. Otherwise, it is set equal to zero.

Effects of School Quality in the Origin on Earnings

Years of Undereducation

Weeks worked in 1999 Experience Location

Marital status

Veteran

Race

English language proﬁciency

Years since migration Citizenship

95

The overeducation variable equals the difference between the reference years of education in the person’s job and their actual years of education where this computation is positive. Otherwise, it is set equal to zero. This is a continuous variable for the numbers of weeks the individual worked in 1999. Age Years of Education 6 years. The two location variables record residence in a metropolitan area or in the southern states. The states included in the latter are Alabama, Arkansas, Delaware, District of Columbia, Florida, Georgia, Kentucky, Louisiana, Maryland, Mississippi, Missouri, North Carolina, Oklahoma, South Carolina, Tennessee, Texas, Virginia, West Virginia. This is a binary variable that distinguishes individuals who are married, spouse present (equal to 1) from all other marital states. This is a binary variable set equal to one for someone who had served in the US armed forces, and set equal to zero otherwise. This is a dichotomous variable that distinguishes between individuals who are Black ( ¼ 1) and all other races ( ¼ 0). Three dichotomous variables (speaks English very well; well; not well, or not at all) are used to record the English language proﬁciency of the respondents who speak a language other than English at home. The benchmark group is those who speak only English at home. This is computed from the year the foreign-born person came to the United States to stay. This is a dichotomous variable set equal to one for foreign born who hold a US citizenship.

Appendix B. Analyses using the Hanushek and Kimko data B.1. Analyses of Hanushek and Kimko using full sample of 73 countries There are 73 countries for which there is information on the Hanushek and Kimko (2000) human capital quality index and data on workers in paid employment in the 2000 US Census. Table B1 lists results obtained from the second-step regression of the across-country variation in the payoff to schooling against the Hanushek and Kimko (2000) index. Tables B2–B4 report ﬁndings from the second-step regression based on the ORU speciﬁcation of the earnings equation. While the imputed values of the Hanushek and Kimko (2000) index are based, among other variables, on GDP per

96

Barry R. Chiswick and Paul W. Miller

Table B.1. Estimates from second step of two-step model, aggregate-level analyses, based on the Hanushek and Kimko (2000) index Variable

(i)

(ii)

Constant

0.019 (3.69) 0.034 (2.50) 0.008 (2.32) No

0.033 (2.33) 0.068 (1.85) 0.007 (0.72) Yes

0.224 73

0.081 73

HCAP/100 1980 GDP per capita/10,000 Country ﬁxed effects in ﬁrst step R2 Sample size

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the PJ ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by j¼1 ½b0j I j . Absolute values of t-statistics in parentheses.

Table B.2. Estimates from second step of two-step model, aggregatelevel analyses. Focus on overeducation, based on the Hanushek and Kimko (2000) index Variable

(i)

(ii)

Constant

0.025 (3.53) 0.084 (4.65) 0.001 (0.10) No

0.018 (2.20) 0.065 (3.08) 0.002 (0.44) Yes

0.270 73

0.130 73

HCAP/100 1980 GDP per capita/10,000 Country ﬁxed effects in ﬁrst step R2 Sample size

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the PJ ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by j¼1 ½b0j I j . Absolute values of t-statistics in parentheses.

capita (in 1960), the GDP per capita variable is retained in the estimating equation for comparison with the models based on the PISA scores.

B.2. Analyses of Hanushek and Kimko indices using subset of countries with both PISA and Hanushek and Kimko measures There are 32 countries for which there is information on the Hanushek and Kimko (2000) human capital quality index, PISA scores, and data on

97

Effects of School Quality in the Origin on Earnings

Table B.3. Estimates from second step of two-step model, aggregate-level analyses. Focus on required education, based on the Hanushek and Kimko (2000) index Variable

(i)

(ii)

Constant

0.008 (2.20) 0.006 (0.65) 0.010 (3.74) No

0.037 (2.59) 0.031 (0.87) 0.001 (0.15) Yes

0.223 73

0.015 73

HCAP/100 1980 GDP per capita/10,000 Country ﬁxed effects in ﬁrst step R2 Sample size

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the PJ ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by j¼1 ½b0j I j . Absolute values of t-statistics in parentheses.

Table B.4. Estimates from second step of two-step model, aggregate-level analyses. Focus on undereducation, based on the Hanushek and Kimko (2000) index Variable

(i)

(ii)

Constant

Country ﬁxed effects in ﬁrst step

0.005 (0.57) 0.010 (0.46) 0.003 (0.58) No

0.007 (0.61) 0.012 (0.40) 0.005 (0.60) Yes

R2 Sample size

0.013 73

0.012 73

HCAP/100 1980 GDP per capita/10,000

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by PJ j¼1 ½b0j I j . Absolute values of t-statistics in parentheses.

workers in paid employment in the 2000 US Census. Table B5 lists results obtained from the second-step regression of the across-country variation in the payoff to schooling against the Hanushek and Kimko (2000) index for this subset of countries. Tables B6–B8 report ﬁndings from the secondstep regression based on the ORU speciﬁcation of the earnings equation for the same set of countries.

98

Barry R. Chiswick and Paul W. Miller

Table B.5. Estimates from second step of two-step model, aggregate-level analyses. Based on the Hanushek and Kimko (2000) index, 32 countries analyses Variable

(i)

(ii)

Constant

0.056 (13.95) 0.082 (7.41) 0.017 (6.25) No

0.128 (10.66) 0.165 (4.99) 0.040 (5.00) Yes

0.887 32

0.808 32

HCAP/100 1980 GDP per capita/10,000 Country ﬁxed effects in ﬁrst step R2 Sample size

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the PJ ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by j¼1 ½b0j I j . Absolute values of t-statistics in parentheses.

Table B.6. Estimates from second step of two-step model, aggregate-level analyses. Focus on overeducation, based on the Hanushek and Kimko (2000) index, 32 countries analyses Variable

(i)

(ii)

Constant

Country ﬁxed effects in ﬁrst step

0.008 (0.84) 0.016 (0.62) 0.021 (3.28) No

0.015 (1.45) 0.038 (1.36) 0.030 (4.38) Yes

R2 Sample size

0.317 32

0.425 32

HCAP/100 1980 GDP per capita/10,000

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by PJ j¼1 ½b0j I j . Absolute values of t-statistics in parentheses.

B.3. Analyses of PISA scores using subset of countries with both PISA and Hanushek and Kimko measures There are 32 countries for which there is information on the Hanushek and Kimko (2000) human capital quality index, PISA scores, and data on workers in paid employment in the 2000 US Census. Table B9 lists results obtained from the second-step regression of the across-country variation

99

Effects of School Quality in the Origin on Earnings

Table B.7. Estimates from second step of two-step model, aggregate-level analyses. Focus on required education, based on the Hanushek and Kimko (2000) index, 32 countries analyses Variable

(i)

(ii)

Constant

Country ﬁxed effects in ﬁrst step

0.036 (4.38) 0.056 (2.47) 0.012 (2.19) No

0.126 (10.89) 0.110 (3.45) 0.038 (4.85) Yes

R2 Sample size

0.477 32

0.745 32

HCAP/100 1980 GDP per capita/10,000

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the PJ ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by j¼1 ½b0j I j . Absolute values of t-statistics in parentheses.

Table B.8. Estimates from second step of two-step model, aggregate-level analyses. Focus on undereducation, based on the Hanushek and Kimko (2000) index, 32 countries analyses Variable

(i)

(ii)

Constant

Country ﬁxed effects in ﬁrst step

0.054 (4.81) 0.078 (2.50) 0.015 (2.01) No

0.048 (4.44) 0.074 (2.50) 0.014 (1.97) Yes

R2 Sample size

0.463 32

0.458 32

HCAP/100 1980 GDP per capita/10,000

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by PJ j¼1 ½b0j I j . Absolute values of t-statistics in parentheses.

in the payoff to schooling against the three PISA scores for this subset of countries. Tables B10–B12 report ﬁndings from the second-step regression based on the ORU speciﬁcation of the earnings equation for the same set of countries.

100

Barry R. Chiswick and Paul W. Miller

Table B.9. Estimates from second step of two-step model, aggregate-level analyses. Based on PISA scores, 32 countries analyses Variable

Reading literacy Mathematics literacy Science literacy (i)

(ii)

(i)

Constant

0.131 0.245 0.096 (5.86) (6.15) (7.03) PISA/100 0.027 0.047 0.020 (4.97) (4.85) (5.71) 1980 GDP per capita/10,000 0.008 0.002 0.008 (1.39) (1.94) (1.44) Country ﬁxed effects in ﬁrst step No Yes No R2 Sample size

0.597 32

0.618 32

0.649 32

(ii)

(i)

0.188 (7.91) 0.036 (5.86) 0.019 (2.01) Yes

0.121 0.229 (5.90) (6.40) 0.024 0.042 (4.94) (4.96) 0.011 0.025 (1.95) (2.51) No Yes

0.683 32

0.595 32

(ii)

0.625 32

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by PJ j¼1 ½b0j I j . The dependent variable for each model is the estimated partial effects of education for the countries for which there are PISA scores. Absolute values of t-statistics in parentheses.

Table B.10. Estimates from second step of two-step model, aggregate-level analyses. Focus on overeducation, based on PISA scores, 32 countries analyses Variable

Reading literacy Mathematics literacy Science literacy (i)

(ii)

(i)

Constant

0.035 0.035 0.024 (1.60) (1.38) (1.69) PISA/100 0.006 0.003 0.004 (1.14) (0.52) (1.01) 1980 GDP per capita/10,000 0.014 0.019 0.014 (2.38) (2.85) (2.47) Country ﬁxed effects in ﬁrst step No Yes No R2 Sample size

0.298 32

0.304 32

0.292 32

(ii)

(i)

0.030 (1.79) 0.002 (0.49) 0.019 (2.90) Yes

0.029 0.030 (1.45) (1.30) 0.005 0.002 (0.95) (0.36) 0.015 0.020 (2.65) (3.08) No Yes

0.303 32

0.289 32

(ii)

0.301 32

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by PJ j¼1 ½b0j I j . The dependent variable for each model is the estimated partial effects of education for the countries for which there are PISA scores. Absolute values of t-statistics in parentheses.

101

Effects of School Quality in the Origin on Earnings

Table B.11. Estimates from second step of two-step model, aggregate-level analyses. Focus on required education, based on PISA scores, 32 countries analyses Variable

Reading literacy Mathematics literacy Science literacy (i)

(ii)

(i)

Constant

0.072 0.207 0.050 (5.97) (5.50) (6.32) PISA/100 0.015 0.033 0.011 (5.16) (3.60) (5.19) 1980 GDP per capita/10,000 0.004 0.020 0.005 (1.35) (1.97) (1.47) Country ﬁxed effects in ﬁrst step No Yes No R2 Sample size

0.611 32

0.513 32

0.613 32

(ii)

(i)

0.169 (7.25) 0.026 (4.28) 0.018 (1.99) Yes

0.063 0.196 (5.50) (5.74) 0.013 0.030 (4.66) (3.65) 0.006 0.023 (1.95) (2.42) No Yes

0.568 32

0.573 32

(ii)

0.518 32

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the PJ ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by j¼1 ½b0j I j . The dependent variable for each model is the estimated partial effects of education for the countries for which there are PISA scores. Absolute values of t-statistics in parentheses.

Table B.12. Estimates from second step of two-step model, aggregate-level analyses. Focus on undereducation, based on PISA scores, 32 countries analyses Variable

Reading literacy Mathematics literacy Science literacy (i)

(ii)

(i)

Constant

0.102 0.097 0.082 (3.31) (3.31) (4.29) PISA/100 0.020 0.020 0.017 (2.72) (2.86) (3.44) 1980 GDP per capita/10,000 0.006 0.005 0.005 (0.79) (0.70) (0.65) Country ﬁxed effects in ﬁrst step No Yes No R2 Sample size

0.311 32

0.321 32

0.385 32

(ii)

(i)

0.076 (4.19) 0.017 (3.56) 0.004 (0.57) Yes

0.101 0.095 (3.66) (3.65) 0.020 0.020 (3.02) (3.15) 0.008 0.007 (1.03) (0.95) No Yes

0.394 32

0.341 32

(ii)

0.352 32

Notes: Model (i) has a single intercept, b0, in the ﬁrst-step regression. Model (ii) is based on the ﬂexible speciﬁcation in the ﬁrst-step regression, where the intercepts are given by PJ j¼1 ½b0j I j . The dependent variable for each model is the estimated partial effects of education for the countries for which there are PISA scores. Absolute values of t-statistics in parentheses.

102

Barry R. Chiswick and Paul W. Miller

Appendix C. Supplementary results

Table C.1. Implied payoffs to schooling, adjusting for effects of ORU variables at various PISA mathematics scores

i. Native born Foreign born ii. No adjustment iii. Adjustment only to the earnings effects of reference education for the foreign born iv. Adjustment only to the earnings effects of overeducation for the foreign born v. Adjustment only to the earnings effects of undereducation for the foreign born vi. Adjustment to all three ORU variables

100 PISA points

Benchmark

þ100 PISA points

10.5

10.5

10.5

– 4.7

4.9 4.9

– 5.2

5.0

4.9

4.8

4.0

4.9

5.8

3.8

4.9

6.0

Table C.2. Implied payoffs to schooling, adjusting for effects of ORU variables at various PISA science scores

i. Native born Foreign born ii. No adjustment iii. Adjustment only to the earnings effects of reference education for the foreign born iv. Adjustment only to the earnings effects of overeducation for the foreign born v. Adjustment only to the earnings effects of undereducation for the foreign born vi. Adjustment to all three ORU variables

100 PISA points

Benchmark

þ100 PISA points

10.5

10.5

10.5

– 4.7

4.9 4.9

– 5.1

5.0

4.9

4.8

3.8

4.9

5.9

3.7

4.9

6.0

References Antecol, H., Cobb-Clark, D.A., Trejo, S.J. (2003), Immigrant policy and the skills of immigrants to Australia, Canada and the United States. Journal of Human Resources 28 (1), 192–218. Betts, J.R., Lofstrom, M. (2000), The educational attainment of immigrants: trends and implications. In: George, Borjas (Ed.), Issues in the Economics of Immigration. University of Chicago Press for National Bureau of Economic Research, Chicago, pp. 51–115.

Effects of School Quality in the Origin on Earnings

103

Bratsberg, T., Terrell, D. (2002), School quality and returns to education of U.S. Immigrants. Economic Inquiry 40 (2), 177–198. Card, D., Krueger, A.B. (1992), Does school quality matter? Returns to education and the characteristics of public schools in the United States. Journal of Political Economy 100 (1), 1–40. Chiswick, B.R. (1978), The effect of Americanization on the earnings of foreign-born men. Journal of Political Economy 86 (5), 897–921. Chiswick, B.R., Miller, P.W. (2008), Why is the payoff to schooling smaller for immigrants?. Labour Economics 15 (6), 1317–1340. Chiswick, B.R., Miller, P.W. (2009), Educational mismatch: Are highskilled immigrants really working at high-skilled jobs and the price they pay if they aren’t? In: Barry, R. Chiswick (Ed.), High Skilled Immigration in a Globalized Labor Market, American Enterprise Institute, Washington DC. Hanushek, E.A., Kimko, D.D. (2000), Schooling, labor-force quality, and the growth of nations. American Economic Review 90 (5), 1184–1208. Hartog, J. (2000), Over-education and earnings: where are we, where should we go? Economics of Education Review 19 (2), 131–147. Organisation for Economic Co-operation and Development (OECD) and United Nations Educational, Scientiﬁc and Cultural Organization (UNESCO). (2003), Literacy Skills for the World of Tomorrow – Further Results From PISA 2000, OECD and UNESCO Institute for Statistics, Paris. Sweetman, A. (2004), Immigrant Source Country Educational Quality and Canadian Labour Market Outcomes, Ottawa: Statistics Canada, Analytical Studies Branch Research Paper No. 234.

CHAPTER 5

Development and Migration: Lessons from Southern Europe Riccardo Fainia,b,c and Alessandra Venturinic,d,e a

Faculty of Economics, Centre for International and Economic Studies, University of Roma Tor Vergata, Rome, Italy b CEPR-Center for Economic and Policy Research, London, UK c IZA-Institute for the Study of Labor, Bonn, Germany d Department of Economics, University of Torino, Via Po 53, Torino, Italy e European university Institute, Florence 50014 Fiesole, Italy E-mail address: [email protected]; [email protected]

Abstract Policy-makers in OECD countries appear to be increasingly concerned about growing migration pressure from developing countries. At the same, at least within Europe, they typically complain about the low level of internal labor mobility. In this chapter, we try to cast some light on the issues of both internal and external labor mobility. We investigate the link between development and migration and argue, on both theoretical and empirical grounds, that it is likely be nonlinear. More precisely, we ﬁnd that, in a relatively poor sending country, an increase in income will have a positive impact on the propensity to migrate, even if we control for the income differential with the receiving country, because the ﬁnancial constraint of the poorest become less binding. Conversely, if the home country is relatively better off, an increase in income may be associated with a fall in the propensity to migrate even for an unchanged income differential. Econometric estimation for Southern Europe over the period 1962–1988 provides substantial support to this approach. We estimate ﬁrst the level of income for which the ﬁnancial constraint is no more binding, around $950, and then the level of income for which the propensity to migrate declines, which is around $4,300 in 1985 prices. We therefore predict a steady decline in the propensity to migrate from Southern European countries. Similarly, our results highlight the possibility that the pressure to migrate from Northern African countries and other developing countries may increase with further growth. Keywords: Migration, growth JEL classiﬁcation: O15

Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008011

r 2010 by Emerald Group Publishing Limited. All rights reserved

106

Riccardo Faini and Alessandra Venturini

1. Introduction International migrants typically leave their home country in search of better living and working conditions abroad. More rapid development in the sending countries should therefore be associated with falling migration. This proposition is based on simple but compelling economic reasoning. Yet, it is not universally accepted. The US Senate Commission on migration argued that, particularly in the short run, income growth at home may foster rather than discourage migration, to the extent that it uproots traditional modes of production and disseminates information about working opportunities abroad. The recent literature (Lopez and Schiff, 1998; Hatton and Williamson, 2006) has highlighted how faster development at home may relax ﬁnancial constraints on would-be migrants and lead to more rather than less migration (see also Ferenczi and Willcox, 1934; Clark et al., 2002; Hanson and Spilimbergo, 1999a; ILOUNCHR, 1992; Layard et al., 1992; Molle, 1990; Vogler and Rotte, 2000). The empirical evidence is mixed. Schiff (1994) provides convincing evidence that migration costs are indeed quite high and may therefore constrain the mobility choice of ﬁnancially constrained migrants. Hatton and Williamson (2006) argue that the combination of high mobility costs and binding ﬁnancial constraints goes a long way in explaining the historical pattern of international migration skewed toward the relatively well-off countries. However, Lucas (2005) fails to ﬁnd any evidence of a positive relationship between income at home and out-migration rates, even for low-income sending countries. This chapter contributes to this debate in a number of ways. We develop a simple model that highlights the role of income levels in determining the propensity to migrate, while controlling for the income differentials between receiving and sending countries. Much of the literature has failed to control for this latter factor. Second, when assessing the role of the home country income on the propensity to migrate, we introduce a key distinction between the ﬁnancial constraint and the home bias effects. The former simply states that higher income of the poor at home may help relax ﬁnancial constraints and lead to more migration, even with an unchanged income differential. The latter, the ‘‘home bias’’ effect, however, leads to the opposite conclusion, namely that, under the plausible assumption that would-be migrants have an intrinsic preference for their home countries’ social amenities, an increase in home income will lead to a fall in the propensity to migrate, even after controlling for the income differential with the host country. The interaction between the ﬁnancial and the home bias effects generates a complex non-linear relationship between the growth of the sending countries’ income and the propensity to migrate. Our (plausible) conjecture is that, for relatively poor countries, the ﬁnancial constraint effect will dominate. Accordingly, higher income at home will be associated with a rise in the income of the poor and an increase in

Development and Migration: Lessons from Southern Europe

107

migration pressure. For relatively better-off countries, however, the home bias effect is likely to be relatively stronger. Hence, a boost to home income should lead to a fall in the supply of migrants. In the second part of the chapter, we focus on the case of the Southern European countries, which have now completed their migration transition1 from emigration countries to immigration ones and whose experience therefore has some interesting implications for other areas. For the pooled sample, we estimated two-income turning points: the ﬁrst when the ﬁnancial constraint is no longer binding, and the second when the home bias effects start dominating the emigration decision, so that further income growth is associated with a fall in migration. The remainder of this chapter is organized as follows. In the next section, we review the home bias literature in the ﬁnancial and trade ﬁelds and then present a simple model of migration. We then look at the main migration trends from Southern Europe. Countries in Southern Europe have undergone a full migration transition and have now become the destinations of substantial labor ﬂows from Northern Africa, Eastern Europe and many other relatively poor countries. An analysis of their experience can therefore shed some light on the factors that affect the long-run trend in migration from developing countries. Econometric results are presented in Section 4. Concluding comments follow in the last section. 2. The pervasiveness of home bias The inability of economic theory to explain the empirical observation that investors over-invest in domestic equity and consumers over-purchase domestically produced goods and services is known as the ‘‘home bias puzzle.’’ Even in the ﬁnancial market where equity products are more similar, the preference for buying national equity – the home bias – is very strong. Probably, information about foreign assets is less widespread and brokers propose only the better-known products, which are easy to push with clients. In fact, Ahearne et al. (2004) show that the registration of the ﬁnancial product, that is, in the US ﬁnancial market, reduces discrimination in buying foreign equities, but even if the home bias is smaller it still remains. The home bias in trade seems even larger. In his seminal paper McCallun, in 1995, found that in Canada inter-provincial trade was 22 times larger than province-trade. In Europe in the 1980s Nitsch (2000) found that intra-national trade in Europe was 10 times higher 1

The term migration transition is used by demographers to indicate the end of the emigration phase and the beginning of a period of no emigration or of immigration.

108

Riccardo Faini and Alessandra Venturini

than intra-communitarian trade, which clearly shows that frontiers in Europe matter. There are many reasons why natives prefer national products. Ofﬁcial and informal barriers probably exist, but costs are mixed with tastes. Different measures of distance have been implemented in order to consider the average size of the goods market, but in the end an unexplained part still remains. Even among the US states in the absence of trade frictions, the home bias is high. Hillberry and Hummels (2003) found a value that was one-third smaller than that reported by Wolf (2000) (four times larger in state trade that among state trade) but still positive (see also Delgado, 2006; Lewis, 1999; Portes and Rey, 2005). Guiso et al. (2004) go more deeply into the causes of the home bias puzzle and use a measure of cultural biases – namely trust2 in people of different nationality – which is signiﬁcant in explaining trade, portfolio investments and foreign direct investment, even after taking into account different country characteristics, different information sets, historical and cultural variables. What clearly emerges from the survey that they use is that individuals tend to over-trust their fellow citizens and this conditions their economic decisions. Home bias also exists in the labor market. Intra-European migration is very limited, even if the presence of income differentials and of unemployment differentials should encourage more citizens of the European countries to move where wages are higher and where the likelihood of ﬁnding a job is greater. Of course, human beings are not goods: they have mobility costs, which are not only monetary costs. International mobility is reduced by the different languages spoken in the destination and sending countries, a factor that, even when it cannot affect professional life, may impact upon social life. Mobility, however, is also very limited internally to countries, when the linguistic barrier is not an issue but personal costs of the location change remain. People have highly idiosyncratic preferences that have been formed while living in their area of origin and are thus conditioned by the way of life prevailing in the area where they have grown up: for instance, they will be very attached to the food produced and eaten in their region.3 People are attached to their reference groups – friends or relatives – and being without them reduces their utility. As culture bias has effects on economic exchange, it biases labor mobility decisions even more so.

2

The measure of trust is derived from a set of surveys conducted by the Eurobarometer. If you sample people according to their schedule for lunchtime, you can likely guess their location around Europe.

3

Development and Migration: Lessons from Southern Europe

109

To gain evidence on this issue, we may use the data provided by the European panel4 in 1994 for seven countries: France, Belgium, Denmark, Ireland, Italy, Greece, Portugal, and Spain available. There is a question in the survey that asks respondents if they have ever moved to another region in the same country or abroad. And 76.5% of the households surveyed declared themselves to be ‘‘absolute stayers,’’ that is, they had not changed region of residence.5 If we select only the very rich (with incomes in the last 25th% quartile) the ‘‘absolute stayers’’ increase to 77.6%. Fundamentally what emerges is that the residents of the European countries are so well off that they prefer not to move and even remain in their regions of birth. If we analyze with a logit the probability of being an ‘‘absolute stayer,’’ namely the probability of not leaving the region of origin, it increases with income, with age, it is higher for males, lower for singles and different in different countries, with Portugal, Italy, and Belgium showing the highest level of preferences for the home area and Denmark the lowest. This last ﬁnding is easy to explain given the large mobility and similarities among the Northern countries. In addition if we replace the income effect with the level of education – which captures the potential income that could be endogenous to the migration decision – the results presented in Table 1 demonstrate that the higher the level of education, the higher the probability of staying. The European Labour Force Survey is less suited to study long-term mobility because it has only one question on the change of residential position relative to the previous year. However, the results are very similar. In 1994 in the same countries, namely France, Belgium, Denmark, Ireland, Spain, Portugal, Italy, and Greece, 99% of the population had not changed region of residence since the year before. The probability of staying increased with age, was lower for singles and men, and positive to educational variables.6 This is just minor evidence of scant European internal mobility and its relationship with educational level; yet it seems to indicate a strong

4

The ECHP is a large household survey conducted in a number of European countries that yields internationally comparable information on both natives and migrants based on a standardized questionnaire that has a section on the ‘‘Migration trajectory.’’ The survey involves annual interviews of a representative sample of households and individuals in a number of European countries. The total duration of the ECHP is 8 years, running from 1994 to 2001. In the ﬁrst wave (1994), a sample of almost 130,000 people aged 16 years and more was interviewed in the then 12 Member States of the European Union (EU). The little refreshment of the sample makes the ﬁrst year, 1994, the most representative of the entire population. Austria, Finland, and Sweden were added later. For more detailed information see for instance Locatelli et al. (2001). 5 The total amount of people interviewed does not include persons that moved abroad and remained aboard. 6 The European Labour Force Survey has also a question on the mobility abroad but the number of units is just too small to make any consideration.

110

Riccardo Faini and Alessandra Venturini

Table 1.

Probability of remaining in the same region of birth 1994

Regional bias

Coefﬁcient

s.e.

Men Age-base Age-base 2 Education 1 Education 2 Education 3 Single Belgium Denmark France Ireland Spain Portugal Italy Obs.89354

0.07 0.04 0.00068 0.467 0.967 1.31 0.149 0.885 0.85 0.084 0.22 0.246 1.127 0.96 Wald 21609

0.016 4.5 0.00 0.13 0.0013 30 0.00 0.007 0.0003 22 0.00 0.001 0.029 16 0.00 0.075 0.028 33 0.00 0.15 0.027 47 0.00 0.24 0.0199 7 0.00 0.26 0.04 21 0.00 0.12 0.034 24 0.00 0.17 0.028 3 0.003 0.14 0.03 7 0.00 0.038 0.026 9 0.00 0.042 0.035 31 0.00 0.157 0.03 32 0.00 0.144 Log pseudolikelihood 46658

z

PW|z|

Marginal effect

Standard error 0.0028 0.00023 0.00001 0.0043 0.0038 0.005 0.0035 0.0043 0.0081 0.0049 0.0049 0.0043 0.0036 0.0036

Source: Euro-panel dataset.

preference for living in the area of origin which increases with income and education.

3. A simple migration model The determinants of migration decisions have been the object of much research in the literature. Traditionally, it was assumed that the decision to migrate depended on a comparison between income at home and income in the potential host country. The Harris–Todaro model reﬁned this approach by showing that risk-neutral migrants would weigh the wage in the destination country by the probability of ﬁnding a job. The Harris– Todaro model was then extended to allow for nonneutral behavior toward the risk (Banerjee and Kanbur, 1981; Hatton, 1993). The ‘‘new’’ migration literature has focused on several factors that, in addition to wage differentials, may prompt people to migrate, such as the desire to diversify risk, to escape relative deprivation and the presence of imperfect information (Stark, 1991). In the spirit of the new migration literature, our model also emphasizes the role of nonincome factors, in particular the home-bias in locational preferences, in affecting the migration decision. One stylized fact in the empirical literature on migration (Hatton and Williamson, 1993; Wyplosz, 1993) is that very few people migrate, sometimes despite the existence of very large wage differentials. This is particularly true in the case of Europe (Faini and Venturini (1993a, 1993b). In the absence of overwhelming barriers to labor mobility, the puzzle of low migration rates has been attributed to large monetary costs

Development and Migration: Lessons from Southern Europe

111

of migration (Easterlin, 1961), to cost of living differentials, to optimistic expectations (Wyplosz, 1993) as well as to widespread uncertainty about the home country’s prospects (Burda, 1993; Faini, 1995).7 Most of these factors, however, are unable to account for the steady fall in migrations from Southern Europe. Indeed, transportation costs and costs of living differentials between Northern and Southern Europe have if anything fallen in the period under consideration. 3.1. The ‘‘home bias’’ model In the model below, we take a different route. Our starting point is the assumption that people prefer to live in their home countries and that, ceteris paribus, they would rather not migrate so that they can avoid the social, cultural, and psychological costs associated with a move to a different location. More formally, it is assumed that individuals derive utility also from the amenities that they can consume at a given location and that such amenities are more conspicuous in their home country.8 Moving abroad involves a loss of utility because of the need to settle into a new and unfamiliar environment and the loss of social relationships. A home market bias in the locational preference is certainly easier to justify than the corresponding bias in consumption patterns (Venables and Smith, 1986) or in ﬁnancial portfolio allocation (French and Poterba, 1991). As we shall see, one testable implication of this framework is that the wage level in the home country becomes a crucial determinant of the migration decision, even after controlling for the wage differential. Therefore, as in the ‘‘new’’ migration literature, (expected) wage differentials are not all that matters in determining the migratory choice. Formally, we assume that the utility of a potential migrant can be represented as follows: Uðwi ; f i Þ

(1)

where wi and fi denote respectively the wage and the amenities in region i. There are two possible locations, the South (S) and the North (N). The potential migrant initially lives in region S. Following the previous discussion, it is assumed that amenities are larger in the origin country of 7

It may be thought that uncertainty about the home country’s prospects should encourage a risk-averse person to migrate. This is no longer the case, however, once ﬁxed moving costs introduce some irreversibility in the migration choice. 8 A similar hypothesis is made by Djajic and Milbourne (1988) who assume that the marginal utility of consumption at home is always higher than that associated with the same rate of consumption in the host country. This assumption plays a crucial role in their analysis of return migrations. In Courant and Deardorff (1993), each country is endowed with a given level of amenities, but agents have identical preferences. As a result, there is no home market bias in the location choice.

112

Riccardo Faini and Alessandra Venturini

the potential migrant, that is, that fsWfn. For migration to occur, evidently the wage differential, wnws, must be large enough to offset the loss of amenities consequent on moving abroad. Given Equation (1), migration will occur if U(wn, fn)ZU(ws, fs). After taking a simple ﬁrst-order expansion of U(wn, fn) around U(ws, fs), the migration condition becomes: Uðws ; f s Þ þ U w ðwn ws Þ þ U f ðf n f s Þ Uðws ; f s Þ

(2)

or: Uw f fn s Uf wn ws

(3)

where the derivatives of the utility function, Uw and Uf, are evaluated at ws and fs. One crucial consideration is that the right-hand side of Equation (3), that is, the marginal rate of substitution between the real wage (or, more precisely, the goods that such a wage can buy) and the amenities at a given location, will not be generally constant. For instance, if we assume that U(w, f) can be described by a CES function, the migration condition becomes: 1 d f s f n ws 1þr g¼ ¼z (4) d w n ws f s where 1/(1þr) is the elasticity of substitution between w and f,9 while d is the distributional parameter associated with f in the CES function. What Equation (4) suggests is that migration is more likely to occur the larger the wage differential and the smaller the gap in amenities. More crucially, Equation (4) also shows that an increase in the wage in the home country, that is, in ws, will be associated with lower migrations, even with an unchanged wage differential.10 The intuition is simple. In this model, both the wage and the amenities associated with a given location are normal goods. A proportional increase in ws and wn, therefore, has a positive income effect that will prompt consumers to try to consume more of the home country’s amenities. Accordingly, the propensity to migrate will decline. The implications of this result are noteworthy. Increases in the home country income will have a twofold effect on migration, ﬁrst by reducing the wage differential with the host country, secondly by inducing a decline in the propensity to migrate. Clearly, this would enhance the 9 r different from 1, otherwise consumption and amenities are perfect substitutes and the income effect disappears. 10 Notice that this result cannot be derived simply from the concavity of the utility function. To see this one needs only to take a second order expansion of U(w) or, more simply, to consider the case where U(w) ¼ ln w. Clearly, in the latter case, the incentive to migrate depends only on the (relative) wage differential; the wage level in the home country plays no independent role.

Development and Migration: Lessons from Southern Europe

113

effectiveness of those policies that aim at reducing migration by promoting growth in the sending countries. The model developed so far does not allow for heterogeneity among agents. If Equation (4) holds, therefore, all agents would be predicted to migrate. In a general equilibrium framework, wages would adjust so that all agents would be equally indifferent between moving abroad or staying at home. In what follows, we abstract from general equilibrium considerations and allow instead for nonhomogeneous behavior. To this end, we assume that g ¼ (1d)/d is distributed within the home country population according to a Pareto distribution function: y x0 yþ1 (5) x0 g where x0 and y are parameters of the distribution function.11 According to Equation (4), migration will occur if g ¼ (1d)/d is larger than z (the righthand side of the equation). Hence, the migrants’ share in the home country population is equal to: Z1 Probðg zÞ ¼ z

y x0 yþ1 dg ¼ xy0 zy x0 g

(6)

where z ¼ (fsfn)/(wnws) (ws/fs)1þr. 3.2. The role of ﬁnancial constraints Equation (6) deﬁnes the population share of those willing to migrate. However, not all would-be migrants, that is, those for which Equation (4) holds are actually able to move abroad. The presence of minimum educational and wealth requirements may indeed act as a binding constraint for many would-be migrants (Banerjee and Kanbur, 1981). Furthermore, capital markets imperfections may prevent a potential migrant from contracting a loan to pay for the monetary cost of migration.12 Similarly, minimum educational and ability requirements may represent an insurmountable obstacle for many would-be migrants.13 We therefore assume that for someone to be able to migrate, a given characteristic A (say, ﬁnancial wealth) must be greater than a given critical value (‘‘c’’) and therefore satisfy the condition AZc. The number of actual migrants is 11

The Pareto distribution function is deﬁned over the interval (x0, y). Migration costs may be a substantial constraint on the decision to move. See Schiff (1994) for some direct evidence. 13 It may be the case, however, that agents with a higher propensity to migrate strive to acquire the educational achievements necessary to be able to move abroad (see Stark, 1993, for such an approach). 12

114

Riccardo Faini and Alessandra Venturini

then determined by the intersection of the two relevant sets of agents, that is, those for which Equation (4) holds (and are therefore willing to migrate) and those for which the constraint is not binding (and are, as a result, able to move abroad): Z1 Z1 Probðg z; A cÞ ¼

f ðg; AÞdAdg z

(7)

c

where f(g, A) is the joint density function of g and A. In what follows, we assume that A and g are independently distributed and that the characteristic A is distributed among the population according to a Pareto distribution function. It can then be shown that the actual number of migrants (M) as a share of the home country’s population (P) will be equal to: M ¼ xy0 zy xe1 ce P

(8)

In Equation (4), an increase in the home wage was found to discourage desired migration. Plausibly, though, a rise in ws should also relax the ﬁnancial constraint. The distribution of the characteristic A would then shift to the right. We then assume that x1, the lower limit of the support of the distribution of A, is a function of the ‘‘wage rate of the poor’’ (i.e., ﬁrst quintile of the income distribution) in the South: x1 ¼ waqls ðwqls oTÞ

(9)

We expect aW0, the implication being that increase in the home wage will relax the constraint (given aW0), but above a threshold level (T) the budget constraint is no longer binding. Substituting Equation (9) and the expression for z in Equation (8) and taking logs yields after some manipulations the following expression: M ln ¼ y ln x0 þ y lnðwn ws Þ yð1 þ rÞ ln ws þ y lnðf s f n Þ P þ yð1 þ rÞ ln f s þ ea ln wqIs ðwqIs oTÞ e ln c

ð10Þ

Relative amenities and relative wages have the expected impact on the migration rate. We see, however, that even after controlling for the wage differential, the impact of ws is a priori ambiguous. This is because of the contrasting effect of the ﬁnancial constraint and the home bias effects. If y (1þr) is large, the latter dominates, and an increase in ws is associated with a fall in migration. Conversely, if ea is relatively large, the impact of a higher income among the poor (wqIs) may prevail by relaxing the constraint and allowing more would-be migrants to move abroad, and the rate of migration may increase.

Development and Migration: Lessons from Southern Europe

115

4. Trends in Southern European migrations Historically, the Southern European countries have represented a constant source of migrant labor for Northern economies. The Great Depression took a heavy toll on the movement of workers between Southern and Northern Europe. However, the Southern European countries resumed their role as a source of migrant workers for the North after the Second World War. During the second half of the 1950s, inter-continental migration ceased and intra-European migration underwent a massive surge. The main destination countries remained France, Germany, Switzerland, Belgium, The Netherlands, and some Northern European countries. The trend continued unabated until the ﬁrst oil shock, when declining economic opportunities in the receiving countries forced many migrants to return home and discouraged new migrants from trying their luck in Northern Europe. Figure 1 shows how, after a steady increase during the 1960s, migration ﬂows from Southern Europe fell dramatically in the wake of the ﬁrst oil shock. Analysts typically attribute the fall in migration rates after 1973 to the decline in labor demand in the main receiving countries (Salt, 1991). Interestingly enough, however, when in the 1980s economic conditions in Northern Europe showed a clear improvement, migrations from Southern Europe did not resume. There are several possible explanations for this apparent puzzle. First, it could be argued that wage and income differentials between Northern and Southern Europe during the 1980s were no longer providing an adequate incentive for labor to move. But the evidence is simply not there. Figure 2 shows that there was some income convergence between the main sending countries in Southern Europe (Portugal, Spain, and Greece14) and the main destination countries, but the gap remained substantial, with the sole exception of Spain. Moreover, if anything the income gap during the eighties widened, and this should have augmented the incentive to migrate. Neither do we ﬁnd a signiﬁcant improvement in the relative labor market conditions between sending and receiving countries. Figure 3 shows that the increase in unemployment after 1973 did not spare countries in Southern Europe. Second, it is possible that a structural shift in the composition of labor demand toward higher skills meant that employment growth (EG) in the receiving countries no longer had a substantial pull effect on (mainly unskilled) migrations. Nevertheless there was no point system in Europe15

14

Italy is an exception, to the extent that the income gap with Northern Europe declined substantially between 1974 and 1990. For the reasons explained later, however, we do not include Italy in our econometric sample. 15 A points system was not adopted in continental Europe and immigrants in continental Europe were mainly unskilled blue-collar workers. Only migration to the United Kingdom was characterized by skilled migrants, who came from former colonies or were educated in the United Kingdom. Very recently also Ireland has been able to attract skilled immigrants from Eastern Europe.

Fig. 1.

Migration rates from Southern Europe.

116 Riccardo Faini and Alessandra Venturini

Fig. 2.

Income differentials (ratio between host and home country income).

Development and Migration: Lessons from Southern Europe 117

Fig. 3.

Unemployment rate in Southern Europe.

118 Riccardo Faini and Alessandra Venturini

Development and Migration: Lessons from Southern Europe

119

that could help in selecting the migrants by skill. In addition no evidence exists that the destination labor markets changed their labor demand; instead, the available evidence suggests that the demand for unskilled immigrants continued after 1973. When migrations from Southern Europe declined or remained ﬂat, immigration to the traditional destination countries continued, but from other sending countries, in particular ones in Northern Africa and Turkey and Yugoslavia,16 which kept on covering unskilled positions with their workers. Other factors, besides the structural shift in labor demand, must therefore have been at work. One plausible conjecture is that the fall in migration rates from Southern Europe reﬂects supply rather than demand factors. The model developed in Section 2 implies that, even with constant wage differentials, the propensity to emigrate will decline if economic conditions improve in the home country. 5. Econometric analysis 5.1. The estimating equation According to Equation (10), the main determinants of migration are the wage and amenities differentials, the average wage level and the ‘‘wage of the poor’’ in the sending country. For the purpose of estimation, we shall assume the relative level of amenities to be a decreasing function of the number of migrants in the previous three years17 or of the migrant stock in the destination areas, as the new migrants’ social and psychological costs of moving to an unfamiliar location are arguably inversely related to past migrations. In other terms, we expect migration chains to affect not only costs and information but also amenities; the variable capturing this effect will be denoted by MC. Considering that the expected utility of migrant is a weighted function of the utility enjoyed in the state of employment and unemployment, the model can be generalized further to include unemployment rates in both the sending and the receiving countries.18 Even with 16

For an extended description of the migration pattern from Southern Europe by destination see pp. 16–23 and by skill pp. 32–35 in Venturini (2004). 17 In the literature there is a large amount of evidence that a large number of migrants was ‘‘rotating’’ on short-term contracts lasting on average for about three years: see for instance King and Rybaczuk (1993, p. 175). 18 From a formal point of view, unemployment can be introduced into our set-up in a relatively simple manner. Let pi be the probability of being unemployed in region i (with i ¼ N, S) and wi (w i ) the wage rate when employed (unemployed). The migration condition becomes: pn Uðw n ; f n Þ þ ð1 pn ÞUðwn ; f n Þ ps Uðw s ; f s Þ þ ð1 ps ÞUðW s ; f s Þ We only need to take a linear approximation of Uðw n ; f n Þ, U(wn, fn), and Uðw s Þ around U(ws, fs) to ﬁnd an expression analogous to Equation (3), with the relevant wage variable being now ð1 pi Þwi þ pi w i , that is, the expected wage. In the empirical implementation, we assume that the probability of being unemployed in region i is a function of the unemployment rate there.

120

Riccardo Faini and Alessandra Venturini

these modiﬁcations, Equation (10) only reﬂects supply determinants of migration. However, demand (i.e., host country’s) considerations will also play a crucial role in deﬁning immigration policies and determining the evolution of migrations. We follow Hanson and Spilimbergo (1999b) in assuming that policy-makers’ efforts to control immigration are a function of labor market conditions in the host country. This is particularly true for Europe, where the main change in migration policy took place with the introduction by the destination countries of restrictive immigration policies in 1973 just after the ﬁrst oil shock.19 Migration controls, however, can hardly be expected to be fully effective. They rather act like a wire-mesh screen by hindering and slowing down migrations, but also permitting some inﬂows, particularly if the supply is very strong.20 Unfortunately, we have no information on the amount of money spent on, and the extent of migration controls in the destinations of Southern European migrants. Accordingly, we simply assume that the tightness of migration policy is negatively related to the EG rate in receiving countries21 and add this variable to the list of regressors. If home amenities are a normal good, we can expect wages in the sending country to affect migrations in a complex way. More precisely, there will be two thresholds, capturing respectively the willingness and the ability to migrate. Thanks to the amenities effect, residents in the sending country will be willing to migrate only if their income w is smaller than the desired minimum income, say W1; However, only people with income greater than some level W2 will be able to ﬁnance their migration. It follows that if W2WW1 people that can afford to migrate do not wish to do so, and no migration takes place. If instead W1WW2 there may be people that both can afford and are willing to move, namely those whose incomes w are such that W2WwWW1.22

19

For a broader discussion of migration policies in Europe in the period considered see Venturini (2004). 20 We have borrowed this analogy from William Cline’s assessment of protection in the textile and clothing sector (Cline, 1987). 21 Immigration policies may also respond to the unemployment rate in the host country. The coefﬁcient on Un may therefore reﬂect both supply and demand factors. 22 Assuming that home amenities is a normal good, we want to ﬁnd two threshold levels of income: the ﬁrst is the level at which the ﬁnancial constraint is no longer binding (W2), so that all those who want to migrate are able to do so; the second is the level of income which discourages emigration because potential migrants, even if they able, are no longer willing to migrate (W1) because they feel afﬂuent enough to prefer the normal good, that is, home amenities. On the opposite, if the home bias is an inferior good, migration takes place only if woW1, and the ability to migrate arises only if W1WwWW2; thus if woW2oW1, only the former constraint is binding; if it is the reverse, woW1oW2, only the ﬁnancial constraint is binding. If the home amenities are ﬁrst a normal than an inferior good, that is, for a low level of income, both the ﬁnancial constraint (positive) and the willingness to move (negative) matter, while for a high level of income the ﬁnancial constraint is no longer binding and the willingness to move increases with income. If the home amenities are ﬁrst an inferior than a normal good, that is, for a low level of income, the ﬁnancial constraint has a positive effect on migration and a negative

Development and Migration: Lessons from Southern Europe

121

To capture this mechanism, we include in our regression not only the wage (log) differentials, but also average income per capita in the sending country (Ws) and its square (because of the home bias, the willingness to migrate, can be assumed to fall rapidly as income grows) and the income of the ‘‘poorest group of the population’’ in all the countries, measured by average income per capita of the ﬁrst quintile of the population (wqIT). Since the ﬁnancial constraint is no longer binding when income reaches a threshold wqIT*, this variable is not necessarily always included in the regressions and we will discuss in detail the choice of the truncation point below. lnðM=PÞ ¼ a0 þ a1 lnðwn =ws Þ þ a2 ln wqI T þ a3 ln ws þ a4 ðln ws Þ2 þ a5 ln U s þ a6 ln U n þ a7 ln EGn þ a8 lnðMCÞ

(11Þ

5.2. The data We estimated Equation (11) on a sample of Southern European countries that included Greece, Portugal, Spain, and Turkey. Italy was excluded because of a lack of homogeneous conditions in the country, epitomized by the persistent backwardness of the Mezzogiorno area. Whereas the northern part of Italy stopped being a net emigration area many decades ago, the Mezzogiorno was a steady source of migrant workers until at least the early 1980s. The existence of persistent and substantial regional differences within Italy implies that any aggregate analysis of the migration behavior of the country is most likely to be meaningless or even misleading.23 Furthermore, an analysis of migration behavior in the Mezzogiorno has already been conducted by Faini (1989). The migration variable used refers to the gross inﬂows of population from a country of origin to a country of destination. Destination data were used because they are more accurate and do not underestimate the emigration as the data of the country of origin do (the sources are indicated in the Appendix). We used gross ﬂows and not net ﬂows for two reasons: ﬁrst because the decision to emigrate is captured better by the gross ﬂow variable, while the net ﬂow variable is a better proxy for the success of the project, and second because the return ﬂows are not registered correctly in many one on the willingness to move, while for a high level of income, the ﬁnancial constraint is no longer binding and the willingness to move decreases with an increase in income. 23 Admittedly, regional differences in migration behavior and standard of living are also important for other countries in our sample, such as Spain. We believe, however, that the degree of regional inequality is much more pronounced in Italy than in Spain. For instance, in 1988 the ratio between income in the more and in the less developed regions was equal to 1.41 in Spain and 1.78 in Italy.

122

Riccardo Faini and Alessandra Venturini

countries and are frequently better reported in the origin country than in the destination one. We also used population data and not worker data because they are better able to capture the migration decision, which is frequently determined by a search for work. We used ofﬁcial data from host countries, thus capturing only legal migrants. However, speciﬁc surveys show that the number of illegal migrants is strongly correlated with the number of legal ones. In addition, we did not want to study the distribution of migrants among destinations but rather the aggregate effect of income growth on total emigration. Consequently, we summed all the outﬂows in a single measure of total gross emigration rate. As a proxy of wages, we used PPP corrected indicators of income per capita for both the sending and the destination countries. There is considerable discussion on whether income or wage indicators should be included in a migration equation (Hatton and Williamson, 1993). However, for medium and long-run migrations, income data may provide a better indication of the earning potentials of prospective migrants. Empirically, the use of either indicators does not seem to make much difference (Gould, 1979). Hence, given also that wage data, especially for the early years, are of dubious quality, we relied on income data. The income of the ‘‘poor’’ was approximated to the income of ﬁrst quintile of the population, which was derived from the WIID dataset provided by the WIDER on income distribution. Employment and unemployment data were derived from the labor force survey according to the OECD deﬁnition. Finally, the destination variables were weighted averages of relevant variables in each of the destination considered. For instance the income per capita in the destination area was obtained as the geometrical average of income per capita in PP parity of each country of destination weighted by the share of immigrants to that destination as a share of the total outﬂow. The same procedure was adopted for unemployment and employment in destination areas. The weight varies each year if the migrants change the composition of their destinations. This choice was entailed by the type of data available. On deciding whether to move and where to go, the migrant compares the returns on all the possible destinations, so that when emigration from a country is spread across many destinations, the appropriate approach is to include all the destination wages in the emigration equation. This procedure is not feasible with aggregate data, and the correct solution is to combine all the destination variables into a weighted one, which becomes a unique composite destination, while bilateral analyses resents of the changes in the explanatory variables of the other destinations as well.

Development and Migration: Lessons from Southern Europe

123

We used two proxies for the Migratory chain; the sum of the gross migration of the last three years, which according to the sociological and historical literature seems to be the appropriate average length of stay, and the stock of migrants abroad as the share of the native population the year before migration. Whilst on the one hand this second proxy for the migratory chain seems better able to capture the size of the community abroad, on the other hand it is built with data less suited to a time series analysis because it is revised for each censuses, which may therefore under-report the size of the community excluding naturalized foreigners or over-report it with not registered return.

5.3. Estimation methods and the results 5.3.1. Properties of the data We ﬁrst tested for the time series properties of the data. We relied on the Im et al. (2003) test for unit roots in panel data. Table 2 shows that for some of the series (migration, income and its square, foreign unemployment, and especially home unemployment and cumulated ﬂows of emigration) the null hypothesis of a unit root was not strongly rejected. However, further testing showed that for these very same series the

Table 2.

Time series properties: the Im–Pesaran–Shin testa Levels

L(M/P) LW LW2 LDIF Ui Un EGn LWqIb MC L(STOCK/P)

**

1.65 1.67** 1.41*** 2.81* 1.07 1.46*** 3.66* 1.98*** 0.79 1.5***

First differences 3.30* 4.31* 4.32* – 3.26* 5.44* – 3.78* 3.14* 1.4**

M: migrations, P: population, LW: log of income in the home country, LDIF: log of the (relative) income differential, Ui: home country’s unemployment, Un: host country’s unemployment, EGn: employment growth in the host country, LWqI: income of the poor, MC: migratory chain as cumulated last three year ﬂows, STOCK: stock of migrants abroad. * The null hypothesis of a unit root is rejected at a 5% conﬁdence level. ** The null hypothesis of a unit root is rejected at a 10% conﬁdence level. *** The null hypothesis of a unit root is rejected at a 1% conﬁdence level. a The Im–Pesaran–Shin W procedure is a test for unit roots in panel data. It does not impose equal roots in the series. It is distributed as an asymptotically normal variable. b The truncated variable is stationary in ﬁrst differences.

124

Riccardo Faini and Alessandra Venturini

hypothesis of panel cointegration could not be rejected.24 In the worst case, therefore, we were estimating a regression with a mix of stationary and non stationary (but cointegrated) series (Sims et al., 1990). To err on the conservative side, in what follows we present estimates in both levels and in ﬁrst differences. To gain efﬁciency, we pooled the four sample countries together. However, careful testing of the pooling restrictions was indispensable. Both theoretical and Montecarlo evidence (Roberson and Symons, 1992) indicated that forcing the constraint of equal slope coefﬁcients on an heterogeneous panel may result in very large biases. Fortunately, in our case, the pooling restrictions were basically not rejected by the data.25 We therefore relied on a ﬁxed effect framework where the intercept was allowed to differ across countries, but the slope coefﬁcients were assumed to be the same. 5.3.2. Estimation methods of the threshold Because of the presence of an unknown income threshold wqIT*26, above which the ﬁnancial constraint is no longer binding, we could not use OLS. Moreover, nor could we use nonlinear least square, given that wqIT* (simpliﬁed hereafter as T*) entered the regression in a nonlinear and nondifferentiable manner. Fortunately we could follow Khan and Senhadji (2001), who examined a formally similar problem of the relationship between inﬂation and growth, and use the conditional least square. For any T*, we estimated the model by OLS, obtaining the sum of squared residuals as a function of T*. The least squares estimates of T* was found by selecting the value of T*, which minimized the sum of squared residuals and maximized the Student’s t of the variable LWqIT. M ¼ XbT þ eLW qI T ¼ T ¼ T min ; . . . ; T max ln P 24

(12Þ

We relied on the statistics developed by Pedroni (1999). We used his procedure no. 7, which allows for endogenous regressors and heterogeneous dynamics of the error term and is distributed as a standard normal variable. The test value was equal to 20.5. The null hypothesis of no cointegration was clearly rejected (see also Kiviet, 1986). 25 Standard tests (ﬁrst on a pairwise basis and then by adding one country at a time) indicated that pooling was appropriate for Greece, Spain, and Turkey. For instance, the F7,39 test for pooling Spain and Turkey was equal to 2.14. Adding Greece yielded an F7,64 equal to 1.70. The pooling restrictions were (marginally) rejected for Portugal. We therefore estimated the equations in Table 3 also without Portugal. The results, however, did not change in any substantial manner, with the sole exception of the coefﬁcient on the income differential, which lost statistical signiﬁcance. 26 Testing for the actual existence of a threshold effects is a nonstandard problem requiring the use of simulation methods, such as the bootstrap proposed by Hansen (2000). Hence here we shall simply assume the existence of such effects, leaving the issue for further research.

Development and Migration: Lessons from Southern Europe

Table 3.

125

Test results of thresholds

Threshold

WqI

SSR

t student

log Likelihood

Tmax T* Tmin

1812 930 641

17.8 7.8 13.0

1.1 3.4 1

56 14 41

where bT is a vector of parameters (indexed by T to show its dependence on the threshold which ranges from Tmin to Tmax) and X is the corresponding matrix of observations on the explanatory variables. We deﬁned S(LWQIT) and t(LWQIT) respectively as the residual sum of squares with the threshold level of income ﬁxed at T and the t statistic for the low-income variable. The threshold level T* was chosen as T* ¼ arg minT{S(LWQIT)}, with a grid search on T in the range from 641 (minimum for Turkey) to 1812 (maximum for Greece). As shown by Table 3 (where for reasons of space only extreme values are shown; details available on request,) T* is the one with the highest t test and lowest S residuals, and the coefﬁcient has the expected positive sign. Hence, we could move to estimation of model of Equation (12) with the threshold ﬁxed at T* ¼ 930. 5.3.3. Results The econometric results for the pooled sample are reported in Table 4 (columns 1 and 3). First income differentials affect the evolution of migrations: as the wage differential increases, migration grows. Second, labor market conditions in the receiving countries matter considerably. Indeed, both the unemployment rate and the EG rate in the host country play a highly signiﬁcant role in affecting migrations. Third, the level of income in the sending country is a signiﬁcant determinant of migration behavior. As the income of the poorest group of the population increases, the ability to emigrate increases, and the total outﬂows are positively affected by income growth. In our case, the budget constraint stops being binding when the income per capita reaches $950. Instead, if the average income increases after about $3,500 per capita, its effect starts to become negative and we interpret it as the effect of the home bias. This seems to indicate that, for a given wage dispersion, in the early stages of development, increases in the sending country’s economic wellbeing give rise to more rather than less migrations to the extent that they help relax the ﬁnancial and educational constraints which prevented many would-be migrants from moving abroad. A similar pattern, but

126

Riccardo Faini and Alessandra Venturini

Table 4.

The determinants of migration (pooled data)

Dependant variable: ln (M/P)

Constant LDIF U ib Un EGn MC D67 D82 LWqIT LW LW2 R2 SER Sargan w2(25)

OLS

GMM-DIFa

118 (7.3) 2.99 (2.6) 0.01 (0.7) 0.17 (3) 10.7 (4.41) 0.028 (6) 0.70 (6) 0.95 (8) 0.07 (5) 28 (6.8) 1.6 (6) 0.91 0.30 –

– 1.8 (1.6) 0.02 (1.9) 0.14 (3.7) 12.14 (2.9) 0.02 (1.75) 1.2 (2.9) 1.7 (1.7) 0.07 (4.7) 25.2 (3.9) 1.5 (4.0) – 0.31 39

Constant LDIF Uib Un EGn LSTOCK D67 D82 LWQ1T LW LWSQ R2 SER Sargan w2(25)

OLS

GMM-DIFa

194 (4) 3.39 (2.8) 0.01 (0.5) 0.27 (6) 8 (3.6) 0.055 (0.3) .5 (3.6) 0.7 (5.8) 0.09 (2.27) 46 (4.6) 2.7 (4.4) 0.87 0.35 –

– 1.2 (1.3) 0.03 (2.5) 0.14 (4.3) 10.6 (2.4) 0.24 (1.9) 0.9 (10) 0.6 (4.6) 0.047 (1.9) 48.0 (8) 2.9 (9) – 0.26 40

M: migrations, P: population, LW: log of income in the home country, LWSQ, LDIF: log of the (relative) income differential, LwqIT: log income in the ﬁrst quintile in the home country, truncated at the threshold level $950, Ui: home country’s unemployment, Un: host country’s unemployment, EGn: employment growth in the host country, D67: dummy variable (1967 migration stop in Greece), D82: 1982 French regularization for Portugal, LSTOCKP stock of emigrants abroad on native population, MC sum of the three previous gross emigration outﬂows. Notes: Country intercepts have been omitted. T-statistics in parenthesis. The Sargan procedure is a test for the overidentifying restrictions in an instrumental variable context. See Arellano and Bond (1991). First-order serial correlation test has been introduced by taking ﬁrst differences in the original equation. a Dynamic panel data estimation. b Ui in the seventies for Portugal.

in a different context, was identiﬁed by Banerjee and Kanbur (1981).27 For relatively higher levels of income, however, further income growth is associated with lower migrations, even after controlling for the income differential. We used two proxies for the migratory chain: the cumulated ﬂows of the last three years of migration, and the share in the native population of the stock of previous migrants abroad. Not surprisingly, the former performed 27

The main difference between our model and the one of Kanbur and Banerjee is that, in the latter, the downward-sloping portion of the income-migration schedule is simply because an increase in the home country’s income entails a reduction in the income differential with the destination country. By contrast, in our model, the income differential is kept constant and the reduction in the migration rate is due to the effect that greater economic well-being exerts on the propensity to migrate.

Development and Migration: Lessons from Southern Europe

127

better than the latter.28 In principle, the stock variable should be more appropriate because it describes the size of the migration community in the host country; however, the quality of stock data is very poor in many countries (see Appendix for details). Finally, in columns 2 and 4, we follow Arellano and Bond (1991) in allowing for the fact that a ﬁxed effect speciﬁcation may not be appropriate.29 To deal with this problem, we estimated the equation in a ﬁrst-difference form30 and relied on an instrumental variable procedure to allow for the resulting correlation between the new error term and the dependent variable.31 The results again provide strong support for our approach. All coefﬁcients, including domestic unemployment, are quite well determined and bear the right sign. Once again, we ﬁnd that, even after controlling for the wage differential, the level of income in the home country plays a crucial role in inﬂuencing migrations, with a positive effect for relatively poor people and a negative effect after a high level of income. As a further check on the robustness of our results, we introduced both a linear and a quadratic trend term into the equation. Neither of these two variables was statistically signiﬁcant. Furthermore, the size and the statistical signiﬁcance of all the other coefﬁcients were basically unaffected by this modiﬁcation. Cumulated emigration performs better than the stock variable, which is now signiﬁcant but with a negative sign. The negative sign of the stock variable is not unusual, because the change in the stock resents of the return to the home country, which takes place in another phase of the lifecourse and which is negatively correlated with the inﬂows. We also tested for the conjecture that the structural shift in the composition of labor demand away from low-skilled workers meant that EG in the destination countries had a less signiﬁcant impact on migrations after 1980.32 We found little evidence in support of this claim.

28

The stock variable is a very painful one. Even if the amount of people abroad come from the same country, they may have originated from different areas of that country and be very different. In addition new migrants have different backgrounds from previous ones, so that they frequently have nothing in common. For a survey see A Venturini (2004, Chapter 2.6.3, pp. 82). It is not the aggregation of the data which causes a poor performance of the stock variable, because also in the analysis of bilateral ﬂows it performs very poorly. 29 This is because when taking the difference from each country’s mean to calculate the country’s ﬁxed P effect, the error term becomes: eit ð1=TÞ Tt¼1 eit and, for relatively small T, is therefore correlated with the lagged dependent variable. 30 Estimation in ﬁrst difference is also advisable because of the evidence that some series are not stationary, hetheroschedastic s.e. are computed. 31 See Arellano (1987) and Arellano and Bond (1991) for further details on the estimation procedure. 32 Zimmermann (1995) shows that pull factors were much less signiﬁcant in affecting migrations from Southern Europe after 1973.

128

Riccardo Faini and Alessandra Venturini

Table 5. Population share of young adult cohorts (14- to 29-year-aged people)

1960 1965 1970 1975 1980 1985 1988

Portugal

Spain

Greece

Turkey

0.236 0.229 0.206 0.238 0.253 0.253 0.251

0.231 0.223 0.218 0.228 0.230 0.241 0.245

0.251 0.234 0.204 0.216 0.215 0.221 0.219

0.250 0.253 0.249 0.270 0.276 0.282 0.283

The statistical properties of the estimated equations appear to be satisfactory. We tested all equations for residual autocorrelation, stability, and predictive power. In no case did we ﬁnd any indications of signiﬁcant misspeciﬁcation. The ﬁnding that economic growth in the sending country will have a positive impact on migration for relatively poor countries (to the extent that it relaxes existing constraints on migration), but will exert an opposite effect on middle-income countries (given that potential migrants will then be more willing to consume their home countries’ amenities) offers encouraging support to our model. However, demographic considerations may provide an alternative explanation for this ﬁnding. Indeed, demographic transition theories suggest that income growth is initially accompanied by an acceleration in population growth (to the extent that the fall in the death rate predates the decline in the birth rate) and therefore gives rise to an increasing weight of young age cohorts in the population. Given that migration is a (negative) function of age, the larger share of young cohorts will tend to increase migrations. In a second phase, though, the belated decline in the birth rate will induce a decline in the weight of young adult cohorts and a fall in the propensity to migrate. Overall, therefore, demographic factors could fully account for the inverse-U pattern of migrations that we found in our data. We controlled for this factor by introducing into our regressions the share of people aged 14–29 (or 20–29) in the population. Table 5 shows the evolution of the ﬁrst of these two indicators for our sample countries. The share of young adults at ﬁrst declines and then rises after 1970.33 Clearly, it is difﬁcult to reconcile this pattern with the supposedly positive effect of young adult cohorts on migrations. This was indeed conﬁrmed by our regression analysis (not reported here). In no case did the share of young adult cohorts (be it measured by the number of people aged 14–29 or 33

Note, however, that data on the size of population cohorts are available only at ﬁve-year intervals. In the regression analysis, we were therefore forced to rely on a linear interpolation. For a wider debate on the population issue in migration see for instance Coleman (1991).

Development and Migration: Lessons from Southern Europe

129

20–29) in the population prove to be an even nearly signiﬁcant factor in determining migrations. We therefore conclude that, at least for the Southern European countries, demographic factors do not provide a convincing explanation for the hump-shaped pattern of migrations. Overall, our results suggest that the impact of income levels on emigration is rather complex. For relatively richer countries, it will reduce the income differential with the destination countries and also encourage people not to incur the social and psychological costs of migration. Emigration will therefore unambiguously decline. By contrast, for poor countries, the migration impact of higher income should be ambiguous. On the one hand, the income differential with the receiving countries will fall but on the other, the ﬁnancial constraint which prevented many would-be migrants from going abroad will become less binding. The net effect may plausibly be positive, particularly if the sending country is relatively poor to begin with. Our estimates suggest that after 950$ per capita, the ﬁnancial constraint is no longer binding and that the turning point in the migration-income relationship falls within a relatively narrow range, about 4000$ per capita. To sum up, our approach moves some steps forward in explaining two apparent paradoxes in the empirics of migrations. First, it is often found that migrants do not come from the relatively poor areas of countries. It is for instance an established fact among economic historians that in the nineteenth century the ﬂow of intercontinental migrations originated mostly from relatively well-off countries in Europe, namely England ﬁrst and Germany later (Davis, 1984; Razin and Sadka, 1992). Poorer countries in Southern Europe by contrast were relative latecomers as sources of migrant workers. The second puzzle is the fact that often, even in the presence of large and persistent wage differentials, the rate of migration may be very low. The former puzzle is explained by the role of ﬁnancial constraints for would-be migrants. To account for the second puzzle, we rely on the existence of non-monetary costs of migration and the desire by potential migrants to consume more of their home country’s amenities, when their income increases. The empirical relevance of this approach is likely to be more signiﬁcant for international migrations, where cultural, geographical and linguistic barriers matter relatively more.

6. Conclusions and policy implications This chapter has sought to shed light on the issues of internal and external labor mobility. Regarding the former, it has shown that the outlook for internal labor mobility in Europe is rather bleak. Despite sometimes persistent wage and income differentials, there is little evidence that even the full abolition of barriers to internal migrations within Europe may lead

130

Riccardo Faini and Alessandra Venturini

to a resumption of labor ﬂows.34 Our results indicate that the propensity to emigrate from Southern European countries, which used to be the dominant sources of worker migrants within the Community, has fallen dramatically and is not likely to increase again. Indeed, most countries in Southern Europe are well to the right of the migration turning point, meaning that further income growth will further enhance the decline in the propensity to migrate. We have offered a new explanation for this phenomenon, focusing on the impact of income growth, for given wage differentials, on the propensity to migrate. Regarding external migrations, this chapter adds causes of both optimism and pessimism to the traditional view that growth in the sending countries will stem migration pressures. It adds optimism to the extent that it shows that, after a certain point, further growth in the origin countries will lead to lower migration propensity, even with constant wage differentials. Put differently, higher income in the sending countries will lower migrations both through their impact on the income differential and because it will lower the propensity to move abroad. The chapter’s ﬁndings, however, are also a cause for pessimism to the extent that they show that such effects will not work for relatively poor countries, where income growth may be associated with more rather than less migrations. Most sending countries in Northern Africa have still a long way to grow before reaching the migration turning point. In these circumstances, aid and development policies, particularly if geared to egalitarian objectives, may not help much in stemming migration. This is not to say, of course, that aid and development policies should not be encouraged. It is meant instead to emphasize that such policies should not be loaded with ancillary objectives such as the discouragement of migration. Acknowledgments We thank Alberto Bisin, Antonio Spilimbergo, Oded Stark, Stephen Yeo, Stefano Fachin for stimulating discussions, and participants at seminars at the University of Munich, Brescia, Cagliari, MIT, Yale, Tufts, Pompeu Fabra for their comments. We are also grateful to Juan Dolado, Nicholas Glytsos, Louka Katseli, and M. Tribalat for supplying some of the data and to Domenico De Palo for superb research assistance. The responsibility for any remaining errors remains ours alone.

34

See Attanasio and Padoa-Schioppa (1991), Eichengreen (1992), and Decressin and Fatas (1995) for further evidence on labor mobility in Europe.

Development and Migration: Lessons from Southern Europe

131

Appendix. Data and variables appendix A1. Methodology A1.1. Data source The data were collected as part of a CNR project. The use of the dataset was restricted to the members of the project. A1.2. Migration data All data derived from national sources have been cross-checked with the OECD publication now published with the title Migration Trends, which was previously not publicly available and obtained as a restricted internal publication of the SOPEMI group composed of national correspondents. Stock and ﬂows in Germany were derived from the Federal Statistical Ofﬁce, Wiesbaden. The data refer to the foreign legal population in Western Germany. Auslaendische Wohnbevoelkerung in der Bundesrepublik. Immigration of foreigners into Western Germany. Zuzuege von Auslandern in die Bundesrepublik. Stock and ﬂows in Sweden were derived from SOS Befolken ngsfo¨ra¨ndnagel obtained from the Centre for Research in International Migration and Ethnic Relations, Stockholm University. Ministry of Labour, based on permits issued or renewed by the host countries also not for work purposes. Stock and ﬂows in Switzerland: Statistics Ofﬁce of the Bundesamt fu¨r Ausla¨nderfragen, Berne. Stock and ﬂows in France: Data received from the INED ‘‘personne entrees’’ (annual source) and foreign population at the Census (interpolated for the missing years). Stock and ﬂows in The Netherlands: CBS Mndstat Bevolking. CBS Jearwerk Buiteniandse Migratie 1977–1985; 1986–1996. CBS, Maandstatistiek Bevolking and Jaarwerk van de Buitenlandse Migrate various years. Stock and ﬂows in Belgium: National Statistical Ofﬁce. Also used have been: Demographic Statistics, 1990, EUROSTAT Statistiques sur la migration 3C, 1994,EUROSTAT Migration Statistics, 3A, 1995 EUROSTAT A1.3. Other data Total population (P): OECD data. Income per capita in Purchasing Power Parity (Y): Summer and Heston data base.

132

Riccardo Faini and Alessandra Venturini

Unemployment rate (U): OECD data. Employment growth (EG): OECD data. Income of the ﬁrst quintile of the population (LYqI), the share of income of the ﬁrst quintile was derived from the surveys reported in the WIID of WIDER, which includes the dataset of Deninger and Squire. Unfortunately, annual surveys are not available, so that high-quality surveys were selected and the missing years interpolated. A2. More information on European migration A2.1. Policy issue The only major policy change that took place during this time span, 1962– 1988, is the restriction imposed after the oil shock in 1973, which however was effective after the increase of the unemployment rate and the slowdown of GNP growth of destination countries. Consequently, the dummy is frequently not signiﬁcant because the policy has been already anticipated by economic changes. Mean

Greece 1960–1988 Log gross migration rate Log income per capita in PP in origin country Log income per capita in PP in destination area Log income differential Log income per capita in PP in origin country I quintile Origin unemployment rate Destination unemployment rate Destination employment growth Log stock of migrants on population Spain 1960–1988 Log gross migration rate Log income per capita in PP in origin Log income per capita in PP in destination area Log INCOME differential Log Income per capita in PP in origin country I quintile Origin unemployment rate

Standard deviation

Max

Min

1.21 8.314

0.81 0.33

2.42 8.66

0.03 7.65

8.982

0.22

9.28

8.58

0.8 7.28

1.36 0.23

1.11 7.50

0.67 6.83

4.46 3.22 0.002

2.12 2.86 0.015

8.143 8.27 0.05

1.68 0.057 0.03

3.35

0.5

3.89

1.9

0.16 8.61

0.89 0.24

1.38 8.91

0.9 8.029

8.97

0.16

9.18

8.65

0.077322 0.086945 7.64 0.33

0.32914 8.16

0.04876 7

8.233436 7.460659

22.01476

1.3466

133

Development and Migration: Lessons from Southern Europe

(Continued ) Mean Destination unemployment rate Destination employment growth Log stock of migrants on population Portugal 1960–1988 Log gross migration rate Log income per capita in PP Log income per capita in PP in destination area Log income differential Log income per capita in PP in origin country I quintile Origin unemployment rate Destination unemployment rate Destination employment growth Log stock of migrants on population Turkey 1960–1988 Log gross migration rate Log income per capita in PP in origin country Log income per capita in PP in destination area Log income differential Log income per capita in PP in origin country I quintile Origin unemployment rate Destination unemployment rate Destination employment growth Log stock of migrants on population

Standard deviation

Max

Min

2.580993 2.00187 0.006 0.015 3.17 0.31

5.94694 0.03 3.4

0.32715 0.03 1.9

1.21 8.11 8.97

2.86 8.57 9.22

0.32 7.45 8.51

0.922 0.35 0.23

1.060453 0.098974 6.88 0.37

1.23239 7.39

0.888 6.2

4.989552 2.643175 4.19888 3.006526 0.04 0.007

8.66922 9.26638 0.015

1.78144 0.84681 0.016

3.9

4.55

0.88

1.7

0.908323 0.676404 7.888957 0.224642

2.00182 8.19919

1.17527 7.46737

9.005283 0.221087

9.32451

8.60204

1.245705 1.245705 6.42 0.28

1.33892 6.85

1.12054 5.9

10.76118 1.337179 3.664498 3.013951 0.02 0.0149 2.6

1.28

12.91423 7.54886 3.013951 0.62247 0.05 0.028 3.67

0.3

References Ahearne, G.A., Griever, L.W., Warnock, E.F. (2004), Information costs and home bias: an analysis of US holdings of foreign equities. Journal of International Economics 62, 313–336. Arellano, M. (1987), Computing robust standard errors for within-groups estimators. Oxford Bulletin of Economics and Statistics 4, 431–434.

134

Riccardo Faini and Alessandra Venturini

Arellano, M., Bond, S. (1991), Some tests of speciﬁcation for panel data: Monte Carlo evidence and an application to employment equations. Review of Economic Studies 58, 277–297. Attanasio, O., Padoa-Schioppa, F. (1991), Regional inequalities, migration and mismatch in Italy, 1960–86. In: Padoa-Schioppa, F. (Ed.), Mismatch and labor mobility. Cambridge University Press, Cambridge. Banerjee, B., Kanbur, S.M. (1981). On the speciﬁcation and estimation of macro rural-urban migration functions: with an application to Indian data. Oxford Bulletin of Economics and Statistics 43 (3), 7–29. Burda, M. (1993), The determinants of East-West German migration. Some ﬁrst results. European Economic Review 37, 452–461. Clark, X., Hatton, T., Williamson, J. (2002), Where do U.S. Immigrants come from and Why? NBER Working Paper No. 8998. Cline, W. (1987), The Future of World Trade in Textiles and Apparels. Institute for International Economics, Washington, DC. Coleman, D. (1991), Demographic Projections: Is There a Need for Immigration? Giovanni Agnelli Foundation, Turin. Courant, P., Deardorff, A. (1993), Amenities, non-traded goods, and the trade of lumpy countries. Journal of Urban Economics 34, 299–317. Davis, K. (1984), The migration of human populations. Scientiﬁc American 231 (3), 89–105. Decressin, J., Fatas, A. (1995), Regional labor market dynamics in Europe and implications for EMU. European Economic Review 39 (9), 1627–1655. Delgado, J. (2006), Single market trails home bias. Bruegel Policy Brief, n. 5. Djajic, S., Milbourne, R. (1988), A general equilibrium model of guestworker migration. Journal of International Economics 25, 335–351. Easterlin, R. (1961), Inﬂuences on European overseas emigration before World War I. Economic Development and Cultural Change 9, 331–351. Eichengreen, B. (1992), Should the Maastricht treaty be saved? Princeton Studies in International Finance, n. 74, Princeton. Faini, R. (1989), Regional development and economic integration. In: da Silva Lopes, J., Beleza, L. (Eds.), Portugal and the Internal Market of the EEC. Banco de Portugal, Lisbon. Faini, R. (1995), Migrations in the integrated EC. In: Baldwin, R., Haaparanta, P., Kiander, J. (Eds.), Expanding European Regionalism: the EC’s New Members. Cambridge University Press, Cambridge. Faini, R., Venturini, A. (1993a), Trade, aid and migrations. Some basic policy issues. European Economic Review 37, 435–442. Faini, R., Venturini, A. (1993b), Italian migrations: the pre-war period. In: Hatton, T., Williamson, J. (Eds.), International Migration and World Development. Routledge, London, pp. 1850–1939. Ferenczi, I., Willcox, W. (1934), International Migrations. United Nations, Geneva.

Development and Migration: Lessons from Southern Europe

135

French, K., Poterba, J. (1991), Investor diversiﬁcation and international equity markets. American Economic Review 81, 221–226. Gould, J.D. (1979), European inter-continental emigration 1815–1914. Patterns and causes. Journal of European Economic History 8, 593–679. Guiso, L., Sapienza, P., Zingales, L. (2004), Cultural biases in economic exchange. NBER Working Paper No. 11005. Hansen, B.E. (2000), Non-parametric data dependent bootstrap for conditional moment model. Econometric Society World Congress 2000 Contributed Papers 1556. Hanson, G., Spilimbergo, A. (1999a), Illegal immigration, border enforcement, and relative wages: evidence from apprehensions at the US-Mexico Border. American Economic Review 89 (5), 1337–1357. Hanson, G., Spilimbergo, A. (1999b). Political economy, sectoral shocks, and border enforcement, NBER Working Paper No. 7315, Cambridge. Hatton, T. (1993), A model of U.K. emigration, 1870–1913. CEPR Discussion Paper No. 771, London. Hatton, T., Williamson, J. (1993), Latecomers to mass migration: the Latin experience. In: Hatton, T., Williamson, J. (Eds.), The age of Mass migration. Oxford University Press, New York. op. cit. Hatton, T., Williamson, J. (2006), Global Migration and the World Economy Two Centuries of Policy and Performance. MIT Press, Cambridge, MA. Hillberry R., Hummels, D. (2003), International bias: some explanations. Review of Economics and Statistics, 85 (4), 1089–1092. ILO-UNCHR. (1992), Informal Summary Record: Joint ILO-UNCHR Meeting on International Aid as a Mean to Reduce the Need for Emigration. ILO, Geneva. Im, K., Pesaran, H., Shin, Y. (2003), Testing for unit roots in heterogeneous panels. Journal of Econometrics 122 (11), 81–126. Khan, S.M., Senhadji, S.A. (2001), Threshold effects in the relationship between inﬂation and growth. IMF Staff papers, vol. 48, no. 1. King, R., Rybaczuk, K. (1993), Southern Europe and the international division of labour: from emigration to immigration. In: King, R. (Ed.), New Geography of European Migrations. Belhaven Press, London, New York: Haltest Press, Wiley. Kiviet, J. (1986), On the rigour of some misspeciﬁcation tests for modelling dynamic relationships. Review of Economic Studies 53, 241–261. Layard, R., Blanchard, O., Dornbusch, R., Krugman, P. (1992), EastWest migration. The alternatives. MIT Press, Cambridge. Lewis, K. (1999), Trying to explain home bias in equities and consumption. Journal of Economic Literature 37, 571–608. Locatelli, M., Moscato, V., Pasqua, S. (2001), The European community household panel (ECHP): Elements for users with special focus on labour and household economics. CHILD Working Paper No. 24. Lopez, R., Schiff, M. (1998), Migration and the skill composition of the labor force: the impact of trade liberalization in LDCs. Canadian Journal of Economics 31 (2), 318–336.

136

Riccardo Faini and Alessandra Venturini

Lucas, R.E.B. (2005), International migration and its economic impact. In: Bourguignon, F., Pleskovic, B., Sapir, A. (Eds.), Annual Bank Conference in Development Economics Europe, Are We on Track to Achieve the Millennium Development Goals. The World Bank, Washington. Molle, W. (1990), The Economics of European Integration. Darmouth, Aldershot. Nitsch, V. (2000), National borders and international trade: evidence from the European Union. Canadian Journal of Economics, Canadian Economics Association 33 (4), 1091–1105. Pedroni, P. (1999), Critical values for cointegration tests in heterogeneous panels with multiple regressors. Oxford Bulletin of Economics and Statistics 61 (November Special Issue), 653–669. Portes, R., Rey, H. (2005), The determinants of cross-border equity ﬂows. Journal of International Economics 65, 269–623. Razin, A., Sadka, E. (1992), International migration and international trade. NBER Working Paper No. 4230, Cambridge. Roberson, J., Symons, R. (1992), Some strange properties of panel data estimators. Journal of Applied Econometrics 7, 175–189. Salt, J. (1991), Current and future international migration trends affecting Europe, 4th Conference of European Ministers responsible for migration affairs, Luxembourg. Schiff, M. (1994), Trade Policy, Aid and Remittances: The Impact of International Migrations. International Economics Department. The World Bank, Washington, DC. Sims, C.A., Stock, J.H., Watson, M. (1990), Inference in linear time series models with some unit roots. Econometrica 58 (1). Stark, O. (1993), Some New Thoughts on Migration and Development. Harvard University, mimeo. Stark, O. (1991), The Migration of Labor. Blackwell, Oxford. Venables, A., Smith, A. (1986), Trade and industrial policy under imperfect competition. Economic Policy 3, 621–672. Venturini, A. (2004), Post-War Migration in Southern Europe. An Economic Approach. Cambridge University Press, Cambridge. Vogler, M., Rotte, R. (2000), The effects of development on migration: theoretical issues and new empirical evidence. Journal of Population Economics 13, 485–508. Wolf, H.C. (2000), Intranational home bias in trade. Review of Economics and Statistics 82 (4), 55–563. Wyplosz, C. (1993), Migration from the East: the role of reform and capital mobility. In: Siebert, H. (Ed.), Migration: A Challenge for Europe. J.C.B. Mohr, Tubingen. Zimmermann, K. (1995), European migration: push and pull, Annual Bank Conference on Development Economics.

CHAPTER 6

Geographic Dispersion and Internal Migration of Immigrants Neeraj Kaushala and Robert Kaestnerb a

School of Social Work, Columbia University. 1255 Amsterdam Avenue, New York, NY 10027, USA E-mail address: [email protected] b Institute of Government and Public Affairs, University of Illinois 815 West Van Buren Street, Suite 525, Chicago, IL 60607, USA E-mail address: [email protected]

Abstract We study the correlates of immigrant location and migration choices to address the following questions: What location-speciﬁc, economic, and demographic factors are associated with these choices? Does the inﬂuence of these factors differ by immigrant characteristics? What are the factors associated with the observed increase in immigrant geographic dispersion during the 1990s? Our analysis suggests that: (1) There is signiﬁcant heterogeneity in the correlates of immigrant location and migration choices; associations vary by immigrant birthplace, age, gender, education, and duration of residence in the United States. (2) Economic factors are, for the most part, weakly associated with immigrant location decisions. (3) Immigrants appear to be more attracted to states with large (growing) populations; less attracted to states with a high proportion of other foreign-born persons; more attracted to states with high unionization, and less attracted to states with high crime. (4) The association between location-speciﬁc characteristics and immigrant location choices changed between 1990 and 2000 for some immigrant groups and this explains most of the increase in geographic dispersion during the 1990s. In contrast, changes in location attributes and changes in immigrant composition explain relatively little of the increase in dispersion. Keywords: Immigrants, Geographic Mobility, residential choices JEL classiﬁcations: J11

Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008012

r 2010 by Emerald Group Publishing Limited. All rights reserved

138

Neeraj Kaushal and Robert Kaestner

1. Introduction Recent years have seen unprecedented geographic dispersion of immigrants within the United States. Although it is true that approximately 70 percent of the foreign-born continue to live in just 6 states, new immigrant communities are growing in regions that did not have any or much foreign-born population even a decade ago. According to the 2000 Census, during the 1990s, the proportion of immigrants more than doubled in 19 states that traditionally did not have many immigrants. Notably, geographic dispersion is not limited to the newly arrived, as immigrants who have been in the country for several years are also spreading out to areas with traditionally few foreign-born persons (Herna´ndez-Leo´n and Zu´n˜iga, 2003; Durand et al., 2000). Geographic dispersion presents challenges for newly emerging immigrant communities. Arrival of immigrant families may create ﬁscal problems if immigrants use local services, for example public education, but pay less than their share of taxes (Smith and Edmonston, 1998; McCarthy and Vernez, 1998). Alternatively, growing immigrant populations may stimulate economic development and revitalize communities (Moore, 1998; Martin and Widgren, 2002). At the same time, dispersion of immigrants may create ethnic and racial tensions in new immigrant communities (Williams 1994). Indeed, public perception about immigrants ‘‘taking jobs’’ from natives and lowering wages is a source of tension between native and immigrant groups even if empirical evidence supporting this perception is inconclusive (Borjas, 2003; Card, 2001). In short, migration trends have important economic and social implications for both immigrants and natives; and ﬁscal implications for local communities. An understanding of the factors that determine the location choices of immigrants expands our knowledge of immigrant behavior and can inform debates on the effects of immigration on natives. It is also useful in making forecasts on immigration trends that can in turn be used to assess future regional and local demands for economic and social services. A number of scholars have examined the location choices of newly arrived immigrants (Bartel, 1989; Zavodny, 1999; Borjas, 1999; Bauer et al., 2002; Kaushal, 2005). In comparison, there is limited research on the factors that inﬂuence the location choices and migration patterns of immigrants who have been living in the United States for several years, or what we refer to as resident immigrants.1 The purpose of this chapter is to investigate this issue. Speciﬁcally, we provide descriptive information 1

Most research on the internal migration of foreign-born population in the US predates the current immigration wave (Bartel, 1989; Bartel and Koch, 1991; Kritz and Nogle, 1994). For research on the location choices of resident immigrants in Europe see Edin et al. (2003), Aslund (2001). For research on the suburbanization of immigrants see Alba et al. (1999) and Logan et al. (2002).

Geographic Dispersion and Internal Migration of Immigrants

139

relating to the following questions: What location-speciﬁc characteristics are important correlates of immigrant location and migration choices? For example, how important are economic factors such as wage and employment levels, and how important is the concentration of foreignborn persons or ethnic density? Do the factors that are associated with immigrant location and migration decisions differ by immigrant characteristics such as age, gender, educational attainment, and length of stay in the United States? Is the inﬂuence of these factors constant over time? And ﬁnally, what ‘‘explains’’ the dispersion of immigrants during the 1990s – was it related to changes in location characteristics, changes in immigrants’ response to these characteristics, or was it due to changing composition of immigrants? 2. Theoretical considerations To study the location decisions of resident immigrants, we begin with a simple behavioral model. Consider a host country with k distinct geographic regions, indexed by j ¼ 1,y, k. Resident immigrants’ choices depend on the net beneﬁts, or utility, of living in a region (Sjaastad, 1962; Greenwood, 1997). Immigrant i’s lifetime utility of living in region j (Uij) is unobserved, but depends on location-speciﬁc attributes (Lj) and individual characteristics (Xi) that determine the beneﬁts and costs of residing in that region. Algebraically, we write utility as follows: U ij ¼ gðLj ; X i Þ Immigrant i chooses to locate in region j if U ij 4U ik

8jak

Location-speciﬁc attributes that inﬂuence the costs and beneﬁts of residing in a region consist of, among other things, economic opportunities and general economic conditions in that region; presence of social and economic networks (e.g., ethnic enclaves and family relationships); provision of public goods and government provided private goods (e.g., social welfare beneﬁts and unemployment beneﬁts); local taxes; psychic costs of living away from the place of birth; and natural (environment and weather) and cultural amenities (Greenwood, 1997; Tienda and Wilson, 1992; Chiswick and Miller, 2005; Bauer et al., 2005; Card and Lewis, 2005). An important location-speciﬁc attribute is the presence of other foreignborn persons. Previous research has shown that immigrants locate where other immigrants with the same nativity live (Massey, 1985; Bartel, 1989; Zavodny, 1999; Alba et al., 1999). Underlying this strong empirical regularity is the fact that ethnic enclaves or communities provide access to and information about the local labor, housing, and credit markets (Zhou and Logan, 1989; Tienda and Wilson, 1992; Bauer et al., 2005).

140

Neeraj Kaushal and Robert Kaestner

Ethnic communities also provide cultural and linguistic afﬁnities that help ease immigrants’ relocation process. On the contrary, locating in ethnic communities may have certain disadvantages. For instance, Chiswick and Miller (2005) ﬁnd that ethnic enclaves slow down acquisition of destination-speciﬁc skills such as English speaking proﬁciency. Previous studies have reached different conclusions about the effect of certain other location-speciﬁc attributes. For example, consider the evidence related to employment. Some studies found that the unemployment rate was positively associated with the probability of living in a state whereas other studies found no or negative link between the unemployment rate and location choices (Greenwood, 1969, 1997; Kritz and Nogle, 1994; Gurak and Kritz, 2000; Aslund, 2001; Jaeger, 2000). Greenwood (1997), in a comprehensive survey of literature on migration, notes, ‘‘one of the most perplexing problems in migration research, at least from the economist’s perspective, was the consistency with which such conﬂicting results were uncovered in connection with the relationship between unemployment rates and migration’’ (p. 682). Similarly, while Borjas (1999) concluded that immigrants were attracted to states with more generous social welfare beneﬁts, other research rejects the welfare magnet hypothesis (Zavodny, 1999; Kaushal, 2005). Costs and beneﬁts of locating in a state also depend on the personal characteristics of immigrants such as age, marital status, education, employment skills, country of birth, and length of stay in the host country. Mincer (1978) argues that family ties have an important bearing on an individual’s location choice. In the case of natives, he argues, family ties deter migration. For immigrants, family and ethnic ties in the host country ease the process of immigration and signiﬁcantly inﬂuence initial location choice; but as in the case of natives, may deter migration within the host country. Kritz and Nogle (1994) ﬁnd that foreign-born men were more likely to migrate within the United States than foreign-born women and that residential mobility declined with age. Several researchers have found that highly educated immigrants are more likely to migrate within the United States as compared to the less educated (Bartel, 1989; Kritz and Nogle, 1994; Gurak and Kritz, 2000). In general, immigrants equipped with hostcountry speciﬁc skills have better labor market opportunities and are likely to feel less alienated as they possess greater human and cultural capital and this may inﬂuence location choices independently of location characteristics (Gurak and Kritz, 2000; Bauer et al., 2005; South et al., 2005). More educated immigrants may also attach less value to living in ethnic enclaves than those without these skills since job-related skills are a substitute for ethnic employment networks (Chiswick and Miller, 2005; Bauer et al., 2005; Kritz and Nogle, 1994). Geographic proximity to the country of birth seems to have an important bearing on where immigrants locate (Schwartz, 1973; Jaeger, 2000; Kaushal, 2005). For example, Kaushal (2005) ﬁnds that in 1995–1996 over 90 percent of the newly arrived legal immigrants to the United States

Geographic Dispersion and Internal Migration of Immigrants

141

from the Caribbean lived on the east coast and an equally large number of newly arrived legal immigrants from Mexico located on the west coast. Length of stay in host country is another crucial determinant of immigrants’ location and migration decisions. Theoretically, length of stay in the United States may encourage or discourage geographic dispersion. Length of residency in host country is a proxy for information (Gurak and Kritz, 2000). The initial choice of locating in ethnic communities may be based on imperfect knowledge, or it may be due to a herd mentality with new immigrants discounting the private information they have in favor of the decisions of recent waves of immigrants without ﬁnding out the underlying factors that led to those decisions (Bauer et al., 2002). In either case, immigrants may move to new locations as they acquire more information. Moreover, length of stay may inﬂuence immigrant preferences and characteristics. With time, immigrants may acquire labor market skills suitable to the host country, which may open up new economic opportunities outside of ethnic enclaves or communities. Over time immigrants’ visa status may change; temporary residents may receive permanent resident status; permanent residents may become naturalized citizens. Change in visa status may change location preferences or make immigrants eligible for jobs they could not obtain as undocumented or temporary residents, in turn creating opportunities to change location. Alternatively, as immigrants develop bonds within the community, the signiﬁcance of ethnic communities may grow (Kritz and Nogle, 1994; Gurak and Kritz, 2000). Dependence on ethnic communities may also make it less proﬁtable to invest in skills that may be helpful in acquiring jobs outside the enclave, in turn, limiting immigrants’ location options. Kritz and Nogle (1994) ﬁnd that immigrants who arrived after 1964 are more likely than earlier arrivals to change locations. In sum, processes explaining the location and migration choices of resident immigrants are complex. These choices may differ depending on immigrants’ age, gender, education, country of birth, and length of stay in the United States. Location-speciﬁc attributes may also inﬂuence these choices, and the inﬂuence of these attributes may differ depending on the characteristics of immigrants. We apply this basic theoretical framework to study the location and migration choices of resident immigrants, and to explain the recent geographic dispersion of some immigrant groups. Mechanically, the recent spatial dispersion of immigrants may be due to changing location characteristics, changing immigrant characteristics, or changes in immigrant responses (associations) to location characteristics. For example, the 1990s may have created better economic opportunities for immigrants outside ethnic enclaves, or immigrants in recent years may have become more responsive to these opportunities. In this chapter, we attempt to identify the important determinants of immigrant location decisions and the contribution of each of these three factors on immigrant dispersion.

142

Neeraj Kaushal and Robert Kaestner

3. Empirical models 3.1. Current location choice Our ﬁrst objective is to study how the location choices of resident immigrants of different nativity, age, sex, educational attainment, and length of stay in the United States depend on location-speciﬁc attributes such as economic conditions and concentration of foreign-born persons (e.g. share of population from the immigrants’ country of birth, share of ‘‘other’’ foreign-born population). As outlined in the theoretical framework, location choices are outcomes of cost–beneﬁt analyses, and these costs and beneﬁts are rooted in location and individual characteristics. We begin with a model based on the assumption that resident immigrants’ current location represents their long-run equilibrium choice, and that individuals adjust to any change in location-speciﬁc attributes instantaneously. The dependent variable for this analysis is the proportion of immigrants from a country or region (pist), who belong to group ‘‘i’’ and live in state ‘‘s’’ in year ‘‘t.’’ Membership in a group is deﬁned by age (20–39, 40–59), sex, education (less than high school, high school and some college, and BA plus), and number of years lived in the United States (o6, 6–10, and W10 years). There are 36 ( ¼ 2 2 3 3) groups in each year. We do separate analysis for immigrants from ﬁve countries/regions of origin that represent 45% of immigrants living in the United States and a large portion of newly arriving immigrants: Mexico, China, Philippines, and India, and for one grouping of countries, Jamaica, and Dominican Republic (Caribbean).2 We assume that the proportion of resident immigrants belonging to group i that live in state s in year t (pist) depends on the following state characteristics: real wage, employment to population ratio, state population (quadratic), proportion of other foreign-born, proportion born in same country (lagged by 10 years), crime rate, per capita income, poverty rate, and proportion of unionized workers in the state. Many of these factors may potentially be affected by immigrant location decisions. To address the issue of reverse causality, albeit in a limited way, we lag most of the location attributes; details are provided in the data section. Let Lstt be a vector representing the location characteristics of a state.3

2

Country-speciﬁc analyses with state-ﬁxed effects control for time-invariant country-state speciﬁc factors (such as geographic proximity, weather, psychic costs of relocation, etc.) in an unconstrained manner. 3 Given the potential for reverse causality, our analysis is descriptive in nature, but still valuable given the relative lack of systematic descriptive information on immigrant location and migration choices.

Geographic Dispersion and Internal Migration of Immigrants

143

The regression model for this analysis is speciﬁed by Equation (1): Pist ¼ ait þ gs þ LsðttÞ P þ ist i ¼ 1; . . . ; 36 ðgroupÞ s ¼ 1; . . . ; 51 ðstateÞ

(1Þ

t ¼ 1990; 2000 ðyearÞ The parameters ait and gs are group-year and state ﬁxed effects, respectively. State ﬁxed effects control for unmeasured, time-invariant location-speciﬁc attributes. Group-year effects are necessary given that we estimate the model by ordinary least squares. For each group ‘‘i,’’ there are 51 observations (one per state) per year, and these observations sum to one. Thus, within each group-year, the dependent variable is a multinomial probability. Inclusion of group-year effects controls for the mechanical correlation between multinomial probabilities within a group in each year. In addition, calculation of standard errors should account for multinomial correlation (i.e., robust standard errors with clustering at group-year level).4 Equation (1) provides estimates of the effect of location-speciﬁc characteristics on immigrant location choices. However, it is likely that the effect of various location-speciﬁc characteristics may differ by immigrant characteristics. Further, these effects may differ over time. Thus, we specify a model that allows for such possibility: Pist ¼ ait þ gs þ

2000 X 36 X

LsðttÞ Pit þ ist

t¼1990 i¼1

i ¼ 1; . . . ; 36 ðgroupÞ s ¼ 1; . . . ; 51 ðstateÞ

(2Þ

t ¼ 1990; 2000 ðyearÞ Note that in Equation (2), we allow the effects of location-speciﬁc characteristics to differ by 72 group-year categories (36 groups * 2 years). To examine the factors associated with the geographic dispersion of immigrants during the 1990s, we decompose the change in geographic location between 1990 and 2000 into three components: dispersion (between 1990 and 2000) due to changes in group (year) effects, dispersion 4

An alternative way to estimate this model is by using the weighted least squares (WLS) with a block diagonal weighting matrix, with each block representing a year (Zellner and Lee, 1965). However, since many proportions in our analysis are zero, the WLS approach is not feasible. Researchers have also employed conditional logit models on individual level data to study current location choices. This model poses two challenges: ﬁrst, this model is cumbersome and difﬁcult to interpret as we have 51 choices and second, marginal effects are generally computed around the mean probability of living in a state and such computation ignores the actual distribution of immigrants across states.

144

Neeraj Kaushal and Robert Kaestner

due to changes in the association (response) between location choice and location characteristics, and dispersion due to changes in location characteristics. To summarize immigrant geographic location and dispersion, we divide states into traditional and nontraditional, where a traditional state is one in which at least 5 percent of the immigrant population from a certain country lived in 1980. We then examine changes in the probability of living in a traditional state between 1990 and 2000. Speciﬁcally, let P90 be the proportion of group i that lived in traditional immigrant states in 1990 and P00 be the proportion of group i that lived in traditional immigrant states in 2000. Total change in geographic dispersion for group i is given by DDi ¼ Ps90 Ps00 ¼ ða90 a00 Þ þ Ls90 P90 Ls00 P00

(3)

The total change in dispersion has three components: changes in group response between 1990 and 2000, denoted by Dai; changes in the association between location choice and location characteristics for group i, denoted by Dd1i (keeping location attributes at the 1990 level), and changes in location attributes, denoted by Dd2i (keeping associations at the 1990 level). These components are shown below for one group: DDi Dai þ Dd 1i þ Dd 2i Dai ¼ a90 a00 X Dd 1i ¼ d 1iL ¼ Ls90 ðP90 P00 Þ

(4Þ

L

Dd 2i ¼

X

d 2iL ¼ P90 ðLs90 Ls00 Þ

L

The three components of the change in total dispersion are partial effects and do not add to total dispersion DDi. We prefer to focus on the partial effects rather than using an Oaxaca-type decomposition because the partial effects reﬂect the fact that the change in dispersion occurs from 1990 to 2000, and therefore the logically appropriate base year to calculate changes between 1990 and 2000 is 1990. The Oaxaca decomposition would use both 1990 and 2000 as the base year. To simplify the presentation of our results, we show the effects computed using Equation (4) for 10 demographic groups deﬁned by age (2 groups), sex (2 groups), education (3 groups), and years in the United States (3 groups). Note that these 10 groups are not mutually exclusive, and that the effects for each of the 10 groups are weighted averages of the effects for the 36 mutually exclusive demographic groups deﬁned by age, sex, education, and number of years lived in the United States. Weights are the share of immigrants from a country in each of the 36 groups.

Geographic Dispersion and Internal Migration of Immigrants

145

3.2. Internal migration One limitation of the current location choice analysis is the assumption of a long-run equilibrium with instantaneous adjustment. To assess the importance of this assumption, we also examine the migration of resident immigrants. We focus on inter-state migration. Like current location choice decisions, migration decisions too are based on cost–beneﬁt analyses. Here, however, we can suspend the assumption of a long-run equilibrium and examine changes in location directly. We consider immigrant i to have migrated if in year t he lived in a different state than in year t5. According to our theoretical model, the probability that immigrant i living in state j migrates depends on the costs and beneﬁts of migration. The costs and beneﬁts of migration can be summarized by differences in lifetime utilities between locations, which are determined by individual characteristics (X) and location-speciﬁc (origin) attributes (L) in year t5. As in the current location analysis, we do not observe all location-speciﬁc attributes that may be affecting location choice decisions. Let the unmeasured differences in lifetime utility between locations j and k (k ¼ 1, y, K) be denoted by mjk. Let M* denote a latent variable indexing the gains from migration and let M be an observed indicator of whether or not a move has occurred between period t 5 and t. The regression speciﬁcation for this model can be written as: M ijt ¼ jj þ jt þ X it jx þ jL Ljt5 þ lf ðmj1t ; . . . ; mjKt Þ þ eijt M ijt ¼ 1

if

M ijt 40

M ijt ¼ 0

if

M ijt 0

(5Þ

i ¼ 1; . . . ; N j ¼ 1; . . . ; 51 t ¼ 1990; 2000 Equation (5) refers to a binomial outcome (Mijt, inter-state move/no inter-state move) and the model includes state-of-origin ﬁxed effects (fj), year-ﬁxed effects (ft), characteristics of the immigrant (Xit), attributes of the origin state in year t5 (Ljt5), and an unspeciﬁed function ( f ) of the unmeasured differences in lifetime utility between location j and other locations (mjkt), which may vary by year. Data limitations prevent us from measuring mjkt and therefore we have an omitted variable problem (Kaestner et al., 2003). One way to address the omitted variable problem is to control for the migration rate among natives in the origin state j.5 This speciﬁcation allows us to adjust for unobserved factors that affect 5

Inclusion of population in the current location choice analysis serves a similar purpose.

146

Neeraj Kaushal and Robert Kaestner

migration from that state.6 The regression equation can be written as: M ijt ¼ jj þ jt þ X it jx þ jL Ljt5 þ ZN jt5 þ eijt

(6)

In Equation (6), Njt5 denotes the inter-state migration rate of natives who lived in state j in year t5. Time-varying location-speciﬁc characteristics (Ljt5) consist of real wage, employment to-population ratio, state population (quadratic), proportion of other foreign-born and proportion of foreign-born with the same ethnicity (measured in year t10); crime rate, per capita income, poverty rate, and proportion of unionized workers. Individual characteristics (Xit) consist of age, education, sex, marital status, other family income, number of children, and number of young children, English speaking ability, number of years lived in the United States, and citizenship status. To estimate the parameters in Equation (6), we use logistic regression with robust standard errors that are clustered on origin states. For this model, we do not obtain separate estimates of the effect of location attributes by immigrant characteristics. The main purpose of this analysis is to assess the restrictiveness of the long-run equilibrium assumption underlying Equation (1) and to address the omitted variable problem present in the current location choice analysis. Here we mainly want to investigate whether estimates from the migration model, which examines changes in location, are consistent with estimates from the current location choice model.

4. Data We use the ﬁve percent samples of the Integrated Public Use Microdata series (IPUMS) of the 1990 and 2000 US Census.7 The analysis is conducted on immigrants from ﬁve of the largest source countries/ country-groups, viz. Mexico, the Philippines, China, India, and the Dominican Republic and Jamaica. In 2000, approximately 45 percent of all foreign-born population living in the United States was from our selected group of countries that are also among the fastest growing groups of immigrants in the United States. A person is deﬁned as an immigrant if he is born abroad, and is a naturalized US citizen or non-citizen. We restrict our analysis to foreignborn population in age group 20–59. State characteristics such as 6

We compute migration rate among natives, by sex, education, marital status, and age. Arguably, immigration, or arrival of new immigrants to a state, may induce outﬂow of natives. Empirical research, however, provides conﬂicting evidence in this regard (Card, 2001; Borjas, 2005). 7 In the inter-state migration analysis, for Mexicans we select a 20 percent random sample and for immigrants from other countries we use the 100 percent sample.

Geographic Dispersion and Internal Migration of Immigrants

147

employment to population ratio, real wage (deﬂated by the consumer price index), and proportion of unionized workforce are computed from the Current Population Survey-Outgoing Rotation Groups ﬁles using both foreign- and native-born persons. Employment to population ratio and wage are computed by age and gender; thus, within each state and year, there are four values for each of these variables. State crime rate is taken from the Federal Bureau of Investigation crime reports, per capita income is from the Bureau of Economic Analysis and population and poverty rate from the Statistical Abstracts. To minimize measurement error, all location-speciﬁc attributes (other than state demographic characteristics) are 3 year averages. In the current location choice analysis, state characteristics are averaged for t2, t1, and t; and in the internal migration analysis, origin state characteristics are averaged over t5, t6, and t7. We use three demographic variables: state population, proportion of state population from immigrant’s country of birth to capture ethnic density, and proportion born in other (other than those with the same nativity) foreign countries to measure immigrant density. To avoid regressing the dependent variable on itself, ethnic density is lagged by 10 years in the current location choice analysis; whereas state population and proportion other foreign-born are measured as of time t. In the internal migration analysis, like other state characteristics, population for the origin state is measured as of t–5; both ethnic and immigrant density are lagged by 10 years since the Census does not have data on the immigrant population in year t5. Census data provide information on demographic characteristics of individuals such as their age, gender, marital status, education, birthplace, when arrived in the United States, which are used to deﬁne immigrant groups in the current location choice analysis and as controls in the internal migration analysis. In the internal migration analysis, we also control for citizenship status, English-speaking proﬁciency, number of children in the family, unearned family income, and state of residence 5 years ago. The variable relating to state of residence 5 years ago is used to deﬁne whether an individual moved inter-state; and to construct inter-state migration rate for natives. For this reason, migration analysis is limited to immigrants who have been in the United States for at least 5 years.

5. Results 5.1. Current location choice: Descriptive analysis Table 1 presents the proportion of immigrants living in the six states with the largest share of immigrants from a country/region in 1980, and shows that the degree of clustering varies by immigrants’ country of origin.

India

0.85 0.78 0.69

0.85 0.81 0.68

0.71 0.82 0.79

0.40 0.22 0.03 0.03 0.04 0.04 0.77

0.82 0.77 0.65

0.65 0.71 0.77

0.33 0.23 0.03 0.04 0.04 0.05 0.72

CA HI IL NY NJ VA

0.80 0.80 0.75

0.78 0.78 0.75

0.46 0.10 0.07 0.06 0.04 0.03 0.77

0.78 0.80 0.79

0.78 0.83 0.78

0.53 0.08 0.05 0.05 0.05 0.03 0.78

0.74 0.75 0.76

0.70 0.73 0.76

0.49 0.07 0.04 0.05 0.05 0.03 0.74

NY CA IL NJ TX PA

0.73 0.65 0.60

0.65 0.64 0.56

0.15 0.14 0.12 0.10 0.07 0.04 0.62

0.79 0.73 0.63

0.65 0.72 0.66

0.15 0.20 0.08 0.13 0.07 0.03 0.66

0.73 0.69 0.61

0.60 0.65 0.66

0.11 0.21 0.07 0.13 0.07 0.03 0.63

NY FL NJ CT MA CA

0.93 0.86 0.80

0.88 0.89 0.89

0.65 0.09 0.07 0.03 0.03 0.02 0.89

Note: Traditional immigrant states are the six states with the largest proportion of immigrants from a country (region) in 1980.

0.79 0.80 0.79

schooling 0.94 0.91 0.91 0.91 0.84 0.86

0.39 0.24 0.04 0.03 0.03 0.03 0.77

Years of o12 12–15 Z16

CA NY IL TX MA NJ

of years in the United States 0.93 0.89 0.69 0.77 0.94 0.90 0.76 0.76 0.93 0.93 0.86 0.77

0.45 0.19 0.06 0.05 0.01 0.02 0.78

Number o6 6–10 W10

0.58 0.21 0.06 0.03 0.01 0.01 0.91

1980

0.94 0.87 0.83

0.90 0.89 0.87

0.52 0.17 0.10 0.03 0.05 0.03 0.88

1990

0.92 0.91 0.80

0.85 0.89 0.86

0.48 0.17 0.11 0.03 0.06 0.02 0.86

2000

Dominican Republic and Jamaica

0.58 0.22 0.08 0.03 0.01 0.01 0.93

Philippines

CA TX IL AZ NM WA Total

China

1980 1990 2000 State 1980 1990 2000 State 1980 1990 2000 State 1980 1990 2000 State

Mexico

Proportion of immigrants living in traditional immigrant states, by country of birth, years in the United States and education (Census data on foreign-born individuals aged 20–59)

State

Table 1.

148 Neeraj Kaushal and Robert Kaestner

Geographic Dispersion and Internal Migration of Immigrants

149

For instance, geographic concentration among immigrants from India, China, and the Philippines is much lower than that among immigrants from Mexico and the Caribbean (Dominican Republic and Jamaica). Between 1990 and 2000, there was a general increase in geographic dispersion for all immigrant groups, especially Mexicans. In 1980 and 1990, a little over 90 percent of Mexicans lived in 6 states; this proportion fell to 78 percent in 2000. The ﬁgures in Table 1 also reveal a systematic change in the relationship between time in the United States and geographic dispersion. In 2000, the longer a person had been in the United States, the less likely it was for him to live outside one of the traditional immigrant states. In 1980 and 1990, this was generally not the case, as there was either little difference in geographic dispersion by time in the United States or older immigrants were more dispersed than the newly arrived. The exceptions are newly arrived immigrants from Mexico and China who increased their dispersion between 1980 and 1990. The relationship between immigrant education and geographic location also changed between 1980 and 2000. In 1980 and 1990, more educated immigrants exhibited greater dispersion than the less educated. During the 1990s, however, Mexicans and Filipinos without a college degree dispersed at the fastest pace. As a result, in 2000, geographic concentration of Mexican and Philippine immigrants did not differ by their years of schooling. In sum, geographic assimilation of immigrants during the 1990s was accompanied by changes in relative dispersion of newly arrived and older immigrants. Although in the earlier decades older immigrants exhibited a greater tendency to disperse, in the 1990s, recently arrived immigrants appeared to be spreading out the most. Among the groups studied, newly arrived, low-educated Mexican immigrants and newly arrived, Chinese immigrants dispersed at the fastest pace. Location choices in 2000 also suggest that among the two largest groups of immigrants – Mexicans and Filipinos – less educated persons were as dispersed as the highly educated.

6. Current location choice: Multivariate analysis Table 2 presents estimates of current location choice model based on Equation (1) with state ﬁxed effects. Estimates from a similar analysis without state ﬁxed effects are in Appendix A. Each column in this table is from a separate regression. The dependent variable is the proportion of immigrants from a country (listed in the column heading) who belong to group ‘‘i’’ and live in state ‘‘s’’ in year ‘‘t.’’ Immigrants from a country are divided into 36 groups deﬁned by age, sex, education, and years lived in the United States. Each regression also controls for 72 group-year effects.

150

Table 2.

Neeraj Kaushal and Robert Kaestner

Estimates of the effect of location attributes on the current location choices of immigrants, by country of birth Mexico

Employment/ 0.020** population (0.009) Real wage 0.032 (in 100 dollars) (0.047) Real per capita income 0.003 (in 10,000 dollars) (0.003) Poverty rate (poverty 0.002*** (0.001) rate/10) Unionization 0.026* (0.013) Population (in millions) 0.033*** (0.005) Population squared 0.001*** (0.000) (in millions) Proportion other 0.231*** (0.075) foreign-born Proportion with same 0.958*** nativity (in t10) (0.200) Crime rate (per 100 0.004*** persons) (0.001)

China

Philippines

India

Dominican Republic and Jamaica

0.004 (0.012) 0.109** (0.049) 0.001 (0.003) 0.002** (0.001) 0.017 (0.016) 0.017*** (0.004) 0.001*** (0.000) 0.193*** (0.069) 1.404 (4.189) 0.003*** (0.001)

0.013 (0.026) 0.005 (0.043) 0.002 (0.007) 0.000 (0.002) 0.009 (0.026) 0.014** (0.005) 0.0004*** (0.0001) 0.054 (0.123) 0.982 (1.794) 0.003** (0.001)

0.013 (0.009) 0.130*** (0.037) 0.010*** (0.003) 0.001 (0.001) 0.001 (0.015) 0.011*** (0.003) 0.0002** (0.0001) 0.097* (0.056) 1.114 (1.132) 0.001 (0.001)

0.032** (0.013) 0.006 (0.048) 0.010* (0.005) 0.004* (0.002) 0.054** (0.026) 0.004 (0.004) 0.0001 (0.000) 0.055 (0.081) 3.279** (1.405) 0.0002 (0.001)

Notes: Estimated coefﬁcients in each column are from a single regression. The dependent variable is the proportion of immigrants from a country (listed in the column heading) who belong to group ‘‘i’’ and live in state ‘‘s’’ in year ‘‘t.’’ Membership in a group is deﬁned by age, sex, education, and number of years lived in the United States. Heteroscedasticity adjusted standard errors clustered on group-year are in parenthesis. Each regression controls for state ﬁxed effects, 36 group effects and 36 group-year interactions. * 0.05opr0.10. ** 0.01opr0.05. *** pr0.01.

Robust (Huber–White sandwich estimates) standard errors clustered on group-year are in parentheses. Estimates in Table 2 suggest that the effect of location characteristics varies depending on immigrants’ country of origin. Economic attributes have modest and sometimes statistically signiﬁcant effects on immigrant location choices. For instance, a one percentage point increase in the employment rate (employment /population) raised the proportion of immigrants from China, India, and the Philippines living in a state by a statistically insigniﬁcant 0.4–1.3 percentage points and signiﬁcantly lowered the proportion of immigrants from Mexico, Jamaica, and Dominican Republic by 2.0–3.2 percentage points. Note that our inconsistent ﬁndings with respect to the effect of employment

Geographic Dispersion and Internal Migration of Immigrants

151

opportunities on location choices, and the counterintuitive ﬁnding that higher employment leads to a lower probability of locating in a state in the case of immigrants from Latin America is in line with previous research that failed to ﬁnd a consistent effect of the employment rate on immigrant migration choices (Greenwood, 1997; Kritz and Nogle, 1994). Estimates in Table 2 also suggests that a one dollar increase in real wage: lowered the probability of living in a state by 0.005–0.1 percentage points for immigrants from Mexico, China, India, and the Caribbean, and raised the probability of living in a state for Chinese immigrants by 0.1 percentage points. The real wage effects are larger and statistically signiﬁcant for immigrants from China and India. A $10,000 increase in state per capita income had no statistically signiﬁcant effect on the location choice of Mexican, Chinese, and Filipino immigrants, but was associated with an increase in the proportion of immigrants from India, Jamaica, and the Dominican Republic by a statistically signiﬁcant one percentage points. State poverty rate too had a modest effect on the location choices of immigrants; a 10 percentage point increase in the poverty rate raised the proportion of immigrants by 0–0.4 percentage points, with the effects being statistically signiﬁcant for Chinese, Mexican, and Caribbean immigrants. In general, immigrants appear to be more likely to live in states with more unionized labor force. A one percentage point increase in the proportion of unionized workforce raised the probability of living in a state by 0.001–0.05 percentage points; estimates of the effect of unionization are relatively large and statistically signiﬁcant for Mexican and Caribbean immigrants. Unlike the effect of economic factors, the impact of state demographic attributes is large and more uniform across various immigrant groups. For instance, a one million increase in population raised the proportion of immigrants 0.5–2.8 percentage points (computed around the mean state population of 5 million) with the effects being statistically signiﬁcant for all groups except immigrants from Jamaica and the Dominican Republic. A one percentage point increase in the proportion of other foreignborn (other than own country of birth) in a state’s population reduced the proportion of immigrants 0.05–0.2 percentage points and 3 of the 5 estimates – the larger estimates – are statistically signiﬁcant. A one percentage point increase in the proportion of state population from the same country of birth (lagged by 10 years) reduced the probability that an immigrant from the same country will locate in a state for all immigrant groups except for the Chinese, who were more likely to live in states with high density of Chinese immigrants. All estimates of the effect of samenativity population are large, although not always statistically signiﬁcant, and imply one-to-one or greater changes in same-nativity population; every one percentage point change in own-nativity population is associated with a one, or greater than one, percentage point change in the probability

152

Neeraj Kaushal and Robert Kaestner

of living in a state.8 The absence of statistical signiﬁcance for associations of such magnitude is not simply a result of the statistical imprecision due to a lack of within-state variation in own ethnic density. In two of the three cases in which the estimate of the effect of own-nativity population is not statistically signiﬁcant, the estimate would remain insigniﬁcant even if we use the standard errors from a model that excluded state ﬁxed effects (see Appendix A). Finally, high crime rates reduced the probability that Mexican, Chinese, and Philippine immigrants would live in a state; the location choices of Caribbean and Indian immigrants, however, were unaffected by crime. Effect sizes are relatively small – a percent increase in crime (a unit increase in crime per 100 people) changed the proportion of immigrants 0.4–0.1 percentage points. To sum up, our analysis so far suggests that the correlates of location choices differ signiﬁcantly across immigrants from different origin countries. However, some uniformity was found. Immigrants appear to be attracted to states with large (growing) populations, and less attracted to states with high proportion of other foreign-born persons. Surprisingly, a higher share of own-nativity population is associated with a lower probability of living in a state for all immigrant groups in our analysis except for the Chinese. This ﬁnding is inconsistent with much past research and is explained by the inclusion of state ﬁxed effects in our analysis. Estimates in Appendix A reveal that in models without state ﬁxed effects, a larger share of own-nativity population is positively associated with the probability of locating in a state. Moreover, the magnitudes of the estimates are extremely large. However, when state ﬁxed effects are included, the large positive association is eliminated and in fact, the sign is reversed in four of the ﬁve cases. It should also be added that most of the previous research was based on immigration trends before the 1990s, and our ﬁnding may partly reﬂect new immigration patterns. In general, the association between economic factors and location choices is quite heterogeneous usually small in magnitude, and often statistically insigniﬁcant. Better economic opportunities such as higher wages and greater employment levels are often (60% of the time) negatively associated with the probability of living in a state, and higher poverty rates are always positively associated with the probability of living in a state. These are counterintuitive ﬁndings if we believe that immigrants are attracted to places with greater economic opportunity and stronger economies. Obviously, there may be an omitted variables problem, but less so in this analysis than in other analyses that fail to include state ﬁxed effects, and these measures may not reﬂect the true economic opportunities

8

The mean value of the share of state population with same nativity varies from 0.1 to 0.6 percent for the 5 ethnic groups.

Geographic Dispersion and Internal Migration of Immigrants

153

for immigrants. However, these results may be accurate and if so, merit further theoretical and empirical investigation. Finally, immigrants tend not to locate in states with high crime and are generally more likely to live in states with high rates of unionization. To examine whether the associations between location characteristics and immigrant location choice changed between 1990 and 2000, we repeat the above analysis with one modiﬁcation – now we allow the effect of location characteristics to differ in 1990 and 2000. The results from this analysis are presented in Table 3 (Appendix B shows models without state ﬁxed effects). Estimates in Table 3 suggest that in 2000, the association between immigrant location choices and several location characteristics were statistically different from similar associations in 1990. These differences are indicated using the symbol (þ). Among Mexican immigrants, signiﬁcant differences were observed in the association between location choices and the following characteristics: employment rate, poverty rate, unionization, population, and crime rate. Some of these differences do not appear to be qualitatively important, for example, differences in the associations related to population but others such as those pertaining to employment, poverty rate, and unionization are more substantial. For example, in 1990 employment rate had a negative and statistically signiﬁcant effect on the location choices of Mexican immigrants; in 2000, the size of the effect is greatly reduced and statistically insigniﬁcant. The other group for which associations between location choices and several state attributes changed from 1990 to 2000 is immigrants from the Dominican Republic and Jamaica; differences in associations are observed for the following attributes: poverty rate, unionization, population, and proportion with same nativity. Most of these differences, while statistically signiﬁcant, are not qualitatively different in that the sign and general magnitudes of the associations do not change; one exception is the poverty rate, which was negatively associated with the probability of living in a state in 1990 and positively associated with this probability in 2000. For the remaining immigrant groups, there were few major differences over time in the associations between location choices and location characteristics. Overall, the associations between economic factors and location choices continue to be small, and often statistically insigniﬁcant, in both the years. Associations between immigrant location choices and demographic attributes of locations become relatively modest when the effect of location attributes is allowed to differ by year (in comparison with the estimates in Table 2). For instance, we ﬁnd that population does not have a statistically signiﬁcant effect on the location choices of immigrants from China, the Philippines, and India in both 1990 and 2000 and the estimated coefﬁcients are much smaller in Table 3 than in Table 2. Population continues to have a positive effect on Mexican immigrants, although the size of the effect is

***

0.014þ (0.013) 0.066 (0.074) 0.012*** (0.004) 0.014***þ (0.003) 0.035*þ (0.018) 0.018***þ (0.004) 0.000***þ (0.000) 0.231* (0.130) 0.722* (0.320) 0.007***þ (0.002)

2000 0.002 (0.016) 0.152** (0.063) 0.020*** (0.006) 0.001 (0.002) 0.002 (0.021) 0.001 (0.010) 0.000 (0.000) 0.197*** (0.057) 7.776 (5.328) 0.003 (0.002)

1990 0.004 (0.012) 0.053 (0.066) 0.013***þ (0.004) 0.009**þ (0.004) 0.019 (0.028) 0.003 (0.009) 0.000 (0.000) 0.123** (0.059) 4.887 (4.320) 0.004þ (0.002)

2000

China

0.006 (0.035) 0.054 (0.047) 0.011* (0.006) 0.002 (0.003) 0.034 (0.026) 0.002 (0.004) 0.000** (0.000) 0.089 (0.083) 1.111 (0.983) 0.002 (0.001)

1990 0.024 (0.031) 0.040 (0.044) 0.010* (0.005) 0.004* (0.002) 0.013 (0.026) 0.000 (0.003) 0.000 (0.000) 0.123 (0.079) 0.879 (0.757) 0.003*þ (0.001)

2000

Philippines

0.015 (0.013) 0.147*** (0.043) 0.012** (0.005) 0.001 (0.001) 0.031** (0.013) 0.002 (0.004) 0.000 (0.000) 0.014 (0.054) 0.901 (5.880) 0.000 (0.001)

1990 0.011 (0.008) 0.095* (0.048) 0.008*þ (0.004) 0.003 (0.003) 0.016þ (0.015) 0.001 (0.004) 0.000* (0.000) 0.012 (0.042) 0.638 (2.603) 0.001 (0.002)

2000

India

*

0.032 (0.018) 0.087 (0.067) 0.004 (0.005) 0.004*** (0.001) 0.051*** (0.014) 0.012*** (0.004) 0.000*** (0.000) 0.192* (0.102) 13.715*** (2.949) 0.000 (0.001)

1990 0.026** (0.010) 0.016 (0.035) 0.010*** (0.003) 0.005*þ (0.003) 0.064***þ (0.018) 0.010***þ (0.003) 0.000***þ (0.000) 0.187** (0.092) 5.900***þ (1.279) 0.001 (0.001)

2000

Dominican Republic and Jamaica

Notes: Estimated coefﬁcients for each country are from a single regression. The dependent variable is the proportion of immigrants from a country (listed in the column heading) who belong to group ‘‘i’’ and live in state ‘‘s’’ in year ‘‘t.’’ Membership in a group is deﬁned by age, sex, education, and number of years lived in the United States. Heteroscedasticity adjusted standard errors clustered on group-year are in parenthesis. Each regression controls for 36 group effects, 36 group-year interactions and state ﬁxed effects. All location characteristics are interacted with the two year dummy variables and reported ﬁgures are estimates based on these interactions. The symbol ‘‘þ’’ indicates that the coefﬁcient for 2000 is statistically different from the coefﬁcient for 1990. * 0.05opr0.10. ** 0.01opr0.05. *** pr0.01.

0.044 (0.009) Real wage (in 100 dollars) 0.014 (0.037) Real per capita income (in 10,000 dollars) 0.011** (0.004) Poverty rate (poverty rate/10) 0.003** (0.001) Unionization 0.006 (0.014) Population (in millions) 0.016*** (0.004) Population squared (in millions) 0.000 (0.000) Proportion other foreign-born 0.326* (0.173) Proportion with same nativity (in t10) 0.882 (0.550) Crime rate (per 100 persons) 0.006*** (0.002)

1990

Mexico

Estimates of the effect of location attributes on the current location choices of immigrants, by country of birth and year

Employment/population

Table 3. 154 Neeraj Kaushal and Robert Kaestner

Geographic Dispersion and Internal Migration of Immigrants

155

smaller. For immigrants from the Dominican Republic and Jamaica, population is negatively associated with location choices, which is opposite of the ﬁnding in Table 2. The variable measuring proportion with the same nativity (own-ethnic density) is associated with a decline in the proportion of Mexicans living in a state in both 1990 and 2000. For all other groups, increased own ethnic density is associated with an increase in the proportion of immigrants living in a state. All estimates are relatively large, but imprecisely estimated. In all cases (except for immigrants from Mexico), the estimated coefﬁcients are larger in 1990 as compared to 2000 suggesting that the appeal of ethnic communities diminished during the 1990s. Finally, the association between location choice and state crime rate changed for some groups during the 1990s; Mexican, Chinese, and Filipino immigrants were less likely to live in states with high crime, but the location choices of Indian, Jamaican, and Dominican immigrants were indifferent to state crime rate in both years. Overall, estimates in Table 3 demonstrate that there were changes in response to location characteristics between 1990 and 2000 for some groups. We return to this point below when we examine and summarize the correlates of the greater geographic dispersion in 2000. At this point, it is clear that changes in immigrant responses to location characteristics are an important element in explaining changes in immigrant location choices between 1990 and 2000.

6.1. Internal migration: Descriptive analysis There are two notable limitations of the current location choice analysis. One, it is based on the assumption that the current location choices of individuals are their long-term equilibrium decisions. Two, there may be unobserved time-varying location attributes correlated with observed attributes that may confound the estimated coefﬁcients. We can address these problems by studying the internal migration of immigrants. Speciﬁcally, we drop the assumption that individuals’ current location choices are their long-term equilibrium and analyze changes in location, and we address the unobserved variable problem by including native migration propensities. Table 4 presents descriptive information of the internal migration of immigrants, by country of birth, years in the US and years of schooling. Due to data limitations, the sample of analysis is restricted to persons who have been in the United States for at least 5 years. We deﬁne a person to have made an inter-state move if his/her current residence is in a different state than 5 years ago. Several aspects of the ﬁgures in Table 4 merit comment. One, inter-state mobility of immigrants differs depending on their country of origin.

0.10 0.10 0.10 52,181

0.12 0.10 0.08

0.10

1990

0.09 0.09 0.09 3,129,743

0.13 0.09 0.07

0.09

2000

Philippines

Note: Move is deﬁned as an interstate change in residence between year t–5 and t.

0.06 0.08 0.24 37,758

0.05 0.06 0.14 67,475

Years of schooling o12 years 12–15 years Z16 years Sample size

0.07 0.07 0.09 34,226

0.22 0.15 0.06

0.12 0.09 0.05

0.05 0.05 0.09 80,306

0.14

0.09

All 0.05 0.07 Number of years in the United States 5–10 years 0.08 0.09 10–15 years 0.04 0.07 W15 years 0.04 0.05

2000

1990

China

2000

1990

Mexico

0.07 0.12 0.18

0.23 0.13 0.12

0.16

1990

0.11 0.12 0.21

0.27 0.18 0.11

0.18

2000

India

0.07 0.13 0.17

0.11 0.12 0.11

0.11

1990

0.07 0.10 0.13

0.10 0.10 0.09

0.09

2000

Dominican Republic and Jamaica

0.09 0.12 0.20

– – –

0.14

1990

0.09 0.11 0.19

– – –

0.13

2000

US

Proportion of inter-state moves, by country of birth, education, and years lived in the United States (Census data for individuals aged 20–59 living in the United States for at least 5 years)

Country/region of birth

Table 4.

156 Neeraj Kaushal and Robert Kaestner

Geographic Dispersion and Internal Migration of Immigrants

157

Immigrants from India are the most mobile, followed by immigrants from China, with Mexican immigrants being the least mobile group. Two, in 2000, the inter-state migration of Indian and Chinese immigrants was higher than that of natives; whereas the inter-state migration of the other three groups was lower. Three, immigrants who had been in the United States for less than 10 years were more mobile than those who had been in the country for 10 or more years, with the sole exception of immigrants from the Dominican Republic and Jamaica. Finally, in general, more educated persons exhibited a higher tendency to make inter-state residential moves, with the exception of immigrants from the Philippines.

6.2. Internal migration: Multivariate analysis Table 5 contains estimates from the regression analysis of migration based on Equation (5). Figures in each cell are changes in the probability or marginal effects derived from a logisitic regression model and robust standard errors clustered within origin states are in parenthesis. Location attributes are for the origin state. Each column represents a separate regression for immigrants from a speciﬁc country (region). Estimates of the effects of personal characteristics on migration vary depending on an individual’s country of birth. Gender appears to have no statistically signiﬁcant effect on the inter-state mobility of immigrants with the exception of Filipino men who are 20 percent (evaluated relative to mean) less likely to migrate, as compared to women. Married Mexican families are less likely to move inter-state, although the effect size is small at approximately 10 percent (of mean); married persons from India, the Dominican Republic and Jamaica are more likely to move inter-state (effect size is 10–16 percent of mean), and marital status has no statistically signiﬁcant effect on the inter-state mobility of Chinese and Filipino immigrants. Estimates in Table 5 also suggest that English speaking ability has a positive and statistically signiﬁcant effect on the migration propensity of persons from China and the Philippines (effect size 12–30 percent of mean), and no statistically signiﬁcant effect on the inter-state mobility of immigrants from Mexico, India, and the Caribbean. In general, more educated immigrants are more mobile with the exception of Filipino immigrants. Mexican immigrants with 12–15 years of schooling are slightly less likely to migrate than those with less than 12 years of schooling; but those with a BA degree are more mobile (than those with less than 12 years of education). Among Chinese and Indian immigrants, higher education has a particularly large effect on migration – persons with 16 or more years of schooling have a 25–50 percent higher migration propensity as compared to those with less than 12 years of schooling. Research pertaining to internal migration of all foreign-born persons

Table 5. Logitistic estimates of the effect of individual characteristics and location attributes on the inter-state migration of foreign-born persons, by country of birth Mexico

Personal characteristics Male

China

Philippines

India

Dominican Republic and Jamaica

0.010 0.002 0.019*** (0.009) (0.007) (0.004) 0.001 0.002 0.007* (0.004) (0.006) (0.004) 0.002 0.014*** 0.033*** (0.001) (0.003) (0.013) 0.004** 0.012*** 0.006 (0.002) (0.004) (0.011) 0.063*** 0.015 0.013* (0.007) (0.008) (0.017) 0.015** 0.036*** 0.009*** (0.007) (0.004) (0.003) 0.008** 0.033** 0.001 (0.004) (0.004) (0.003) 0.009*** 0.034*** 0.016*** (0.002) (0.004) (0.002)

0.003 0.011 (0.010) (0.012) 0.016* 0.016*** (0.010) (0.006) 0.002 0.007 (0.010) (0.007) 0.003 0.024*** (0.008) (0.004) 0.037*** 0.033*** (0.008) (0.005) 0.047*** 0.001 (0.006) (0.005) 0.013*** 0.003 (0.004) (0.003) 0.029*** 0.007*** (0.005) (0.003)

0.015 0.021 0.094*** (0.040) (0.045) (0.021) Real wage (in 100 dollars) 0.028 0.079 0.060 (0.089) (0.102) (0.102) Real per capita income (in 10,000 0.026 0.020 0.008 (0.062) (0.048) (0.023) dollars) Poverty rate (poverty rate/10) 0.011 0.015 0.011 (0.029) (0.028) (0.020) Unionization 0.211 0.241 0.071 (0.282) (0.422) (0.215) Population (t5) (in millions) 0.005 0.021* 0.013* (0.008) (0.012) (0.007) Population squared (t5) 0.0001 0.001*** 0.0003* (0.000) (0.000) (0.00001) 0.102 Proportion other foreign-born 0.857 1.355* (0.656) (0.781) (0.347) (t10) Proportion with same nativity 0.982 2.712 0.439 (t10) (0.637) (4.993) (1.430) Crime rate (per 100 persons) 0.005 0.003 0.003 (0.005) (0.009) (0.005) Inter-state migration of natives 0.094*** 0.089*** 0.187*** (0.031) (0.033) (0.058)

0.067 0.059 (0.064) (0.076) 0.088 0.021 (0.212) (0.081) 0.042 0.094* (0.049) (0.038) 0.061 0.015 (0.039) (0.045) 0.282 0.457 (0.407) (0.485) 0.021 0.018* (0.014) (0.009) 0.0003 0.0001 (0.0003) (0.0003) 0.043 0.149 (0.600) (0.571) 3.341 0.449 (3.564) (2.025) 0.003 0.003 (0.006) (0.007) 0.391*** 0.233*** (0.058) (0.030)

Married Speaks English well Education ¼ 12–15 years Education Z16 years In United States for 5–10 years In United States for 10–15 years Citizen Location attributes Employment/population

Notes: Figures in each cell are marginal effects of logitistic regressions; heteroscedasticity adjusted standard errors clustered on origin states are in parenthesis. Location attributes are for the origin state. Employment/population and real wage are deﬁned by age and gender; employment/population, real wage, crime rate, per capita income, poverty rate, and unionization rate are origin state averages for t5, t6, and t7. Each column is from a separate regression; each regression also controls for age, number of children, number of young children, other family income, state and year ﬁxed effects. * 0.05opr0.10. ** 0.01opr0.05. *** pr0.01.

Geographic Dispersion and Internal Migration of Immigrants

159

during the 1970s and 1980s had a similar ﬁnding (Bartel, 1989; Kritz and Nogle, 1994). Recent arrivals are more likely to move inter-state than those who have been in the United States for more than 15 years, with the exception of immigrants from Dominican Republic and Jamaica, whose probability to migrate is unaffected by years of stay in the United States. Time in the United States has large effects on migration of Chinese and Indian immigrants; for example, those in the United States for 5–10 years have a migration propensity that is 26–31 percent larger than those who have been in the country for more than 15 years. Comparing this ﬁnding with the results presented in Table 1 suggests that the geographic dispersion of more recently arrived immigrants (who have been in the United States for more than 5 but less than 10 years) is a combination of two factors: their initial location choices are more diverse, and they have a greater tendency to migrate as compared to older immigrants. Finally, being a citizen reduces the probability of making an inter-state move by 7–30 percent (of mean), which is perhaps a reﬂection that citizens develop a higher level of social capital in the communities they live, and are less likely to make inter-state moves. Estimates in the bottom panel of Table 5 show the effect of location characteristics on inter-state moves. It suggests that location-speciﬁc economic attributes are not related to the migration decisions of immigrants in a substantial way. The estimates are modest and statistically insigniﬁcant in most cases. These results are qualitatively similar to those for the current location choice model. State population has a negative effect on the inter-state mobility of all immigrants except for those from Jamaica and the Dominican Republic. A one million increase in state population lowered the probability of outmigration by 7–13 percent in the case of immigrants from Mexico, China, the Philippines, and India; and raised the probability of out-migration of immigrants from the Caribbean by 18 percent. These results are generally consistent with the analogous estimates in Table 2 with the exception of the ﬁnding for immigrants from the Caribbean countries. Here population is negatively associated with living in a state whereas in Table 2 population had a small positive association. Proportion of other foreign-born persons in the origin state had no statistically signiﬁcant effect on the out-migration of all immigrant groups except for those from China who were less likely to move out of states with high foreign-born population. These ﬁndings are inconsistent with those in Table 2, which indicated that the proportion of foreign-born persons was negatively associated with the probability of living in a state. Here, the only statistically signiﬁcant association suggests the opposite. The proportion of state population from immigrant’s country of origin also had no statistically signiﬁcant effect on the internal migration of immigrants, although most estimates are positive (except for Indian immigrants) and relatively large; a one percentage point increase in the

160

Neeraj Kaushal and Robert Kaestner

proportion of the state population with the same nativity raises the probability of migrating by 0.4–2.7 percentage points. These effects are consistent with those in Table 2 except in the case of Indians, and suggest that at least part of the dispersion observed in recent years is on account of internal migration to states with low same-ethnic density. The last row of Table 5 presents the effect of inter-state migration of natives on the interstate mobility of immigrants. Inter-state migration of natives is computed by sex, education, marital status, and age and it is intended to control for unmeasured state-speciﬁc factors. As expected, inter-state migration of natives has a positive association with the inter-state migration of immigrants – a one percentage point increase in the inter-state outmigration of the US-born raised the probability that the immigrant population will out-migrate by 0.1–0.4 percentage points. To recapitulate, our analysis in Table 5 suggests that internal migration of immigrants depends on their personal characteristics. Recently arrived immigrants are more mobile than older immigrants. English speaking ability encourages and being a citizen discourages the inter-state migration of immigrants. With the exception of Filipino immigrants, more educated persons have a higher probability to migrate. Several location-speciﬁc attributes such as population and percent of population with same nativity appear to have a similar effect on the internal migration and current location choices of immigrants. The one signiﬁcant exception pertains to proportion of state population that is foreign-born, but not from the same country of origin.

6.3. Decomposition analysis Our next objective is to identify the factors that ‘‘explain’’ the geographic dispersion observed between 1990 and 2000. Here we focus on immigrants from Mexico and China for whom dispersion was substantial. These groups are also two of the largest and fastest growing immigrant groups in the United States. Since the degree of dispersion experienced by the other three groups is relatively modest (Tables 1 and 8), we decide not to present the analysis for those groups. Speciﬁcally, we investigate whether dispersion away from traditional immigrant states was due to changes in immigrant responses to location characteristics, changes in location characteristics, or changes in immigrant propensities to locate in traditional immigrant states (see Equation 4). For this analysis, we deﬁne traditional immigrant states as those where at least 5 percent of the immigrant population from a certain country lived in 1980.9 9

Using this criterion, the traditional immigrant states for Mexican immigrants are: CA, TX, and IL; for Chinese immigrants: CA and NY; for Filipino immigrants: CA, HI, IL, and NY; for Indian immigrants: NY, CA, IL, NJ, and TX and for immigrants from the Dominican Republic and Jamaica: NY, FL, and NJ.

Geographic Dispersion and Internal Migration of Immigrants

161

Dispersion is measured as the difference between the proportion of immigrants belonging to group i that lived in traditional immigrant states in 1990 and 2000. Table 6 presents estimates of the change in dispersion explained by changes in responses to location characteristics (location characteristics are set at the 1990 level). Table 7 presents estimates of the change in dispersion explained by changes in location attributes between 1990 and 2000 (holding constant the response to location attributes at the 1990 level). Note also that while the regression models allow the effect of each location-speciﬁc attribute to differ for the 36 demographic groups in 1990 and 2000, we present the results for 10 groups, which are not mutually exclusive. Tables 6 and 7 have the same format – columns (1) and (2) give the proportion of immigrants living in the traditional immigrant states in 1990 and 2000, respectively; and the column labeled (1)–(2) computes the level of dispersion, or the difference between the proportion living in traditional immigrant states in 1990 and 2000. Column (3) in Table 6 gives the total dispersion due to changes in immigrant responses (Db); columns (4)–(13) show the change in dispersion due to changes in group response (column 4) and due to changes in response to speciﬁc location attributes (columns 5–13). The effect of change in group response measures variation in dispersion due to change in a group’s propensity to locate in a state; for instance, change in the propensity between 1990 and 2000 that young Mexican males with less than a high school degree would locate in traditional immigrant states. The ﬁgures in Table 6 suggest that changes in response to location attributes explain a signiﬁcant amount of the change in geographic dispersion. Estimates in column (3) indicate that changes in responses led to increased dispersion; the modal effect is between 10 and 12 percentage points suggesting a 10–12 percentage point decrease in the probability of living in traditional immigrant states between 1990 and 2000. However, there is signiﬁcant variation within an immigrant group depending on individual characteristics, and between Mexican and Chinese immigrants in the amount of dispersion explained by changes in responses. Estimates in columns (5)–(13) suggest that immigrants even from the same country have varied responses to location characteristics. For instance, real wage, state population, and ethnic density appear to have much more modest effects on the dispersion of older (in terms of years in the United States) Mexican immigrants as compared to the newly arrived. Similarly, real wage, employment/population ratio and per capita income have a much larger effect on the dispersion of Chinese immigrants with a BA degree than of those without it. Overall, changes in the response to real wages, population and density of ‘‘other’’ foreign-born have the largest effects on changes in geographic dispersion for both Mexican and Chinese immigrants, and in most cases these effects are positive indicating that

0.62 0.63 0.59 0.65 0.73 0.69 0.49 0.57 0.70 0.64

China Male Female Age 20–39 Age 40–59 o 12 years education 12–15 years education Z16 years education In United States o6 years In United States 6–10 years In United States W 10 years

0.55 0.56 0.50 0.62 0.72 0.65 0.43 0.47 0.52 0.62

0.68 0.74 0.68 0.77 0.70 0.71 0.68 0.55 0.66 0.78

0.07 0.07 0.09 0.03 0.01 0.04 0.06 0.10 0.18 0.02

0.16 0.13 0.18 0.09 0.15 0.14 0.11 0.27 0.18 0.09

(1)–(2)

Difference

0.07 0.06 0.05 0.09 0.11 0.12 0.02 0.01 0.12 0.08

0.10 0.09 0.11 0.07 0.10 0.10 0.06 0.18 0.12 0.07

0.01 0.02 0.06 0.04 0.11 0.06 0.10 0.00 0.03 0.01

0.02 0.02 0.03 0.03 0.01 0.06 0.04 0.00 0.04 0.03

0.00 0.01 0.04 0.03 0.05 0.04 0.06 0.01 0.02 0.01

0.01 0.01 0.01 0.03 0.02 0.04 0.04 0.01 0.04 0.02

0.06 0.05 0.09 0.02 0.03 0.04 0.08 0.08 0.08 0.04

0.08 0.05 0.07 0.03 0.06 0.05 0.07 0.15 0.06 0.02

0.05 0.06 0.08 0.02 0.04 0.03 0.07 0.06 0.06 0.04

0.02 0.02 0.02 0.01 0.03 0.00 0.04 0.07 0.02 0.01

Total Demographic Employment/ Real Real per Group population wage capita income (3) (4) (5) (6) (7)

0.01 0.01 0.00 0.00 0.09 0.02 0.05 0.00 0.00 0.00

0.03 0.02 0.03 0.03 0.03 0.02 0.04 0.05 0.03 0.01

(8)

0.00 0.01 0.00 0.01 0.01 0.02 0.01 0.00 0.01 0.01

0.01 0.02 0.01 0.02 0.02 0.02 0.01 0.00 0.03 0.03

(9)

Poverty Unionization rate

0.03 0.02 0.00 0.06 0.03 0.08 0.01 0.03 0.06 0.04

0.11 0.08 0.12 0.04 0.09 0.11 0.05 0.22 0.09 0.05

0.02 0.03 0.03 0.01 0.04 0.03 0.01 0.00 0.02 0.03

0.02 0.03 0.02 0.03 0.03 0.03 0.03 0.01 0.02 0.03

Population Proportion other foreign born (10) (11)

0.03 0.03 0.03 0.02 0.01 0.02 0.04 0.04 0.04 0.02

0.03 0.01 0.03 0.01 0.01 0.03 0.02 0.11 0.01 0.01

0.00 0.00 0.00 0.00 0.08 0.01 0.04 0.01 0.00 0.00

0.01 0.01 0.01 0.02 0.02 0.00 0.02 0.04 0.02 0.00

Crime Proportion born in same rate country (12) (13)

Change in predicted probability of living in traditional immigrant states 1990 location characteristics dispersion due to change in responses (Db)

Probability of living in traditional immigrant states

Note: Traditional immigrant states are deﬁned as states where at least 5 percent of the persons with the same nativity lived in 1980.

0.84 0.87 0.86 0.86 0.85 0.85 0.79 0.82 0.84 0.87

(2)

(1)

Mexico Male Female Age 20–39 Age 40–59 o 12 years education 12–15 years education Z16 years education In United Stateso6 years In United States 6–10 years In United States W 10 years

2000

1990

Probability of living in traditional immigrant states

Table 6.

0.62 0.63 0.59 0.65 0.73 0.69 0.49 0.57 0.70 0.64

China Male Female Age 20–39 Age 40–59 o12 years education 12–15 years education Z16 years education In United States o6 years In United States 6–10 years In United States W10 years

0.55 0.56 0.50 0.62 0.72 0.65 0.43 0.47 0.52 0.62

0.68 0.74 0.68 0.77 0.70 0.71 0.68 0.55 0.66 0.78

0.07 0.07 0.09 0.03 0.01 0.04 0.06 0.10 0.18 0.02

0.16 0.13 0.18 0.09 0.15 0.14 0.11 0.27 0.18 0.09

(1)–(2)

Difference

0.01 0.00 0.01 0.01 0.01 0.00 0.00 0.00 0.00 0.00

(4)

0.04 0.00 0.06 0.00 0.02 0.00 0.08 0.00 0.09 0.00 0.09 0.00 0.00 0.01 0.00 0.00 0.07 0.00 0.07 0.00

0.02 0.00 0.01 0.01 0.01 0.00 0.06 0.04 0.03 0.00

(3)

0.02 0.02 0.02 0.02 0.01 0.01 0.01 0.02 0.02 0.02

0.01 0.02 0.01 0.01 0.01 0.02 0.02 0.02 0.02 0.01

(5)

Total Employment/ Real population wage

0.02 0.01 0.02 0.01 0.01 0.01 0.02 0.01 0.01 0.02

0.03 0.02 0.02 0.02 0.02 0.02 0.03 0.03 0.03 0.02

Real per capita income (6)

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.01 0.01 0.00 0.00 0.00 0.01 0.00 0.00

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.01

0.01 0.03 0.00 0.04 0.01 0.05 0.00 0.02 0.03 0.04

0.04 0.03 0.05 0.00 0.03 0.04 0.02 0.08 0.01 0.02

0.01 0.01 0.02 0.02 0.02 0.04 0.01 0.04 0.01 0.02

0.03 0.01 0.04 0.01 0.02 0.03 0.03 0.06 0.01 0.01

Poverty Unionization Population Proportion rate other foreign born (7) (8) (9) (10)

0.05 0.06 0.03 0.08 0.07 0.09 0.01 0.01 0.07 0.07

0.04 0.02 0.03 0.03 0.03 0.03 0.08 0.04 0.04 0.03

0.01 0.01 0.01 0.01 0.02 0.01 0.00 0.00 0.01 0.02

0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.05

Crime Proportion born in same rate country (11) (12)

Change in predicted probability of living in traditional immigrant states 1990 responses to locations dispersion due to change in location characteristics (DX)

Probability of living in traditional immigrant states

Note: Traditional immigrant states are deﬁned as states where at least 5 percent of the persons with the same nativity lived in 1980.

0.84 0.87 0.86 0.86 0.85 0.85 0.79 0.82 0.84 0.87

(2)

(1)

Mexico Male Female Age 20–39 Age 40–59 o12 years education 12–15 years education Z16 years education In United States o6 years In United States 6–10 years In United States W10 years

2000

1990

Probability of living in traditional immigrant States

Table 7.

Geographic Dispersion and Internal Migration of Immigrants 163

164

Neeraj Kaushal and Robert Kaestner

changes in responses to these three attributes between 1990 and 2000 resulted in increased dispersion. Interestingly, changes in group responses do not dominate; i.e. in most cases changes in geographic dispersion was not driven by changes in propensity (conditional on measured characteristics) to locate by some groups. Estimates in Table 7 show change in dispersion due to changes in location characteristics. There has been relatively little change in state characteristics, so it is unlikely that changes in characteristics will explain a large share of the change in dispersion. Results in column (3) of the table suggest that changes in location characteristics (keeping immigrant responses at the 1990 level) explain between 0–6 percentage points (or 0–55 percent) of the actual dispersion experienced by Mexican immigrants, depending on their personal attributes. In the case of immigrants from China, changes in location attributes do not explain any dispersion. Indeed, the analysis suggests that if responses in 2000 were the same as in 1990, and only location characteristics had changed proportion of immigrants living in traditional immigrant states would have increased. In general, no single location characteristic has any substantial effect on dispersion of Mexican and Chinese immigrants. In the case of Mexican immigrants, changes in proportion born in the same country appear to have increased dispersion, and changes in crime rate lowered dispersion. In the case of Chinese immigrants, changes in proportion born in the same country lowered dispersion while most other factors had negligible impact.

6.4. Dispersion due to changes in immigrant composition One possibility that we have not considered yet is that changes in geographic dispersion between 1990 and 2000 may have been due to changes in immigrant composition – for example, a relative increase in the proportion of immigrants who had low propensity to locate in traditional immigrant states. To investigate whether that was the case, we examine the effect of changing demographic composition of immigrants on geographic dispersion. Speciﬁcally, we compute the probability that immigrants from a country/region would live in the traditional immigrant states if only composition were changed (the probability of living in traditional immigrant states for group ‘‘i’’ were kept at the 1990 level). Table 8 has the results of this analysis. Figures in Table 8 show that demographic composition of immigrants varies depending on their country of origin. For instance, in 2000, 70 percent of Mexican immigrants were young (aged 20–39 years) as compared to only 54 percent of the Caribbean. Similarly, 70 percent of Indian immigrants had at least a BA degree in comparison with only 4 percent of the Mexicans; and a third of Indians had been in the country

Philippines

India

0.56 0.44 0.70 0.30 0.61 0.35 0.04 0.23 0.19 0.58 0.71

0.85

0.56 0.44 0.73 0.27 0.67 0.30 0.03 0.24 0.21 0.55 0.85

0.85

0.00

0.00 0.00 0.03 0.03 0.06 0.05 0.01 0.01 0.02 0.03 0.14 0.63

0.50 0.50 0.43 0.57 0.25 0.40 0.35 0.33 0.22 0.45 0.63 0.59

0.04

0.48 0.02 0.52 0.02 0.52 0.09 0.48 0.09 0.19 0.06 0.35 0.05 0.47 0.12 0.27 0.06 0.24 0.02 0.49 0.04 0.56 0.07 0.70

0.41 0.59 0.56 0.44 0.08 0.47 0.44 0.23 0.22 0.54 0.70 0.70

0.00

0.41 0.00 0.59 0.00 0.46 0.10 0.54 0.10 0.06 0.02 0.49 0.02 0.45 0.01 0.13 0.10 0.19 0.03 0.67 0.13 0.65 0.05 0.61

0.55 0.45 0.57 0.43 0.07 0.28 0.64 0.28 0.24 0.48 0.61 0.60

0.01

0.54 0.01 0.46 0.01 0.61 0.04 0.39 0.04 0.06 0.01 0.24 0.04 0.70 0.06 0.34 0.06 0.19 0.05 0.47 0.01 0.58 0.03

Dominican Republic and Jamaica

0.78

0.45 0.55 0.63 0.37 0.31 0.57 0.12 0.22 0.23 0.54 0.78

Note: Traditional immigrant states are deﬁned as states where at least 5 percent of the persons with the same nativity lived in 1980.

Demographic composition Male Female Age 20–39 Age 40–59 o12 years education 12–15 years education Z16 years education In United States o6 years In United States 6–10 years In United States W10 years Proportion in traditional immigrant states Proportion in traditional immigrant states if only composition changed

China

Dispersion due to changes in composition of immigrants

0.78

0.45 0.55 0.54 0.46 0.27 0.59 0.13 0.14 0.20 0.67 0.76

0.00

0.00 0.00 0.09 0.09 0.04 0.02 0.01 0.09 0.03 0.13 0.02

1990 2000 Difference 1990 2000 Difference 1990 2000 Difference 1990 2000 Difference 1990 2000 Difference

Mexico

Table 8.

Geographic Dispersion and Internal Migration of Immigrants 165

166

Neeraj Kaushal and Robert Kaestner

for less than 5 years in comparison with only 14 percent of the Caribbean immigrants. As in the earlier analysis, changes in demographic characteristics of immigrants are not uniform across immigrants from different countries. Overall, composition of Mexican immigrants registered a modest change during the nineties, and these changes fail to explain any dispersion. Demographic characteristics of Chinese immigrants however changed signiﬁcantly during the 1990s – the proportion of young immigrants (age 20–39 years) rose by 9 percentage points (21 percent); the proportion with a BA degree increased by 12 percentage points (34 percent) and the proportion of recently arrived immigrants increased by 6 percentage points (17 percent). These changes in composition of Chinese immigrants explain 57 percent of their total dispersion. The demographic characteristics of immigrants from the other countries also changed during the 1990s, but these changes fail to explain any dispersion in the case of immigrants from the Philippines and the Caribbean, and explain a third of the total dispersion among Indian immigrants.

7. Conclusion In this chapter, we study the correlates of location and migration choices of immigrants from ﬁve countries (regions), and ﬁnd that immigrant response to location-speciﬁc attributes varies depending on their country of birth. Even among immigrants from the same country, associations between location choices and location-speciﬁc attributes vary depending on immigrant characteristics, such as age, gender, education, and years lived in the United States. Other studies based on immigrant location choices during the 1990s have reached similar conclusions (Massey, 2008). We also ﬁnd that the association between several location characteristics and location choices has changed between 1990 and 2000. Most research on immigrants is done with the assumption that foreign-born persons are a homogenous population. Our analysis, however, suggests this not to be the case, and that generalizations based on the assumption of homogeneity could lead to erroneous conclusions. We ﬁnd that economic attributes are associated with a relatively modest and often statistically insigniﬁcant effect on the location choices of immigrants. This ﬁnding is similar to earlier research that failed to ﬁnd a uniform effect of economic factors on migration decisions of natives (Greenwood, 1997) and immigrants (Kritz and Nogle, 1994). We also ﬁnd that immigrants are less likely to live in states with high crime and more likely to live in states with high unionization. Immigrants appear to be more attracted to states with large (growing) populations, and less attracted to states with high proportion of other foreign-born persons.

Geographic Dispersion and Internal Migration of Immigrants

167

Our analysis also suggests that internal migration of immigrants differed depending on individual characteristics with newly arrived, more educated, English speaking, noncitizens exhibiting a higher propensity to migrate. Like the current location choice analysis, the internal migration analysis also suggests that economic factors are associated with modest changes in inter-state moves. We ﬁnd that the geographic assimilation of immigrants during the 1990s was accompanied by changes in relative dispersion of newly arrived and older immigrants. Although in the earlier decades older immigrants exhibited a greater tendency to disperse, during the 1990s, recently arrived immigrants appeared to be spreading out the most. Among the groups studied, the newly arrived, low-educated Mexican immigrants and newly arrived Chinese immigrants exhibited the fastest dispersion. The current location choices in 2000 also suggest that for the two largest groups of immigrants – Mexicans and Filipinos – less educated persons were as dispersed as the highly educated. However, like previous research based on internal migration during the 1970s and 1980s, our research also showed that more educated immigrants are more mobile (Bartel, 1989; Kritz and Nogle, 1994). For Mexican and Chinese immigrants, the two groups that dispersed the fastest during the 1990s, we examine whether dispersion was due to changes in response to location characteristics, changes in group response, or changes in location characteristics; and ﬁnd that in most cases, change in response to location characteristics appears to explain most of the dispersion; changes in group response do not appear to explain much dispersion nor do changes in location characteristics. Our analysis suggests that Chinese immigrants would have been more clustered in traditional immigrant states if their response to location characteristics had not changed during the 1990s. Our study also suggests that changing demographic composition of immigrants does not explain the geographic dispersion of immigrants from Mexico, the Philippines and the Caribbean; it explained 57 percent of the dispersion among Chinese immigrants and a third of the dispersion experienced by Indian immigrants.

168

Neeraj Kaushal and Robert Kaestner

Appendix A. Estimates of the effect of location attributes on the current location choices of immigrants, by country of birth Mexico

China

Philippines

Employment/ 0.008 0.024* 0.033** population (0.015) (0.013) (0.015) Real wage (in 100 0.200*** 0.325*** 0.358*** dollars) (0.040) (0.049) (0.035) Real per capita 0.000 0.004** 0.010*** (0.002) (0.002) (0.002) income (in 10,000 dollars) Poverty rate (poverty 0.011*** 0.017*** 0.012*** rate/10) (0.001) (0.002) (0.002) Unionization 0.063*** 0.039*** 0.002 (0.005) (0.006) (0.005) Population (in 0.003*** 0.001** 0.007*** (0.000) (0.000) (0.000) millions) Population squared 0.0004*** 0.0002*** 0.001*** (in millions) (0.000) (0.000) (0.000) Proportion other 0.193*** 0.036*** 0.040*** (0.010) (0.010) (0.013) foreign-born Proportion with same 1.566*** 11.047*** 1.593*** nativity (in t10) (0.060) (0.917) (0.168) Crime rate (per 100 0.004*** 0.005*** 0.002*** persons) (0.000) (0.000) (0.000)

India

Dominican Republic and Jamaica

0.027*** (0.009) 0.070** (0.034) 0.008*** (0.001)

0.004 (0.012) 0.155** (0.061) 0.066*** (0.003)

0.005*** 0.030*** (0.001) (0.001) 0.055*** 0.077*** (0.006) (0.009) 0.002*** 0.005*** (0.000) (0.000) 0.0001*** 0.0001*** (0.000) (0.000) 0.107*** 0.026*** (0.012) (0.009) 15.405*** 25.075*** (0.542) (0.912) 0.002*** 0.001*** (0.000) (0.000)

Note: Estimated coefﬁcients in each column are from a single regression. The dependent variable is the proportion of immigrants from a country (listed in the column heading) who belong to group ‘‘i’’ and live in state ‘‘s’’ in year ‘‘t.’’ Membership in a group is deﬁned by age, sex, education, and number of years lived in the United States. Heteroscedasticity adjusted standard errors clustered on group-year are in parenthesis. Each regression controls for 36 group effects and 36 group-year interactions. *0.05opr0.10, **0.01opr0.05, and *** pr0.01.

1990

2000

0.029***þ 0.022 0.010 (0.010) (0.018) (0.015) 0.066 0.275*** 0.351*** (0.047) (0.040) (0.048) 0.008***þ 0.009*** 0.008*** (0.001) (0.002) (0.002) 0.015*** 0.042***þ 0.012***þ (0.001) (0.002) (0.005) 0.083***þ 0.001 0.025***þ (0.007) (0.007) (0.007) 0.002***þ 0.001** 0.002***þ (0.001) (0.000) (0.001) 0.000***þ 0.000*** 0.000***þ (0.000) (0.000) (0.000) 0.140***þ 0.051*** 0.029***þ (0.009) (0.015) (0.008) 1.561***þ 12.106*** 10.947*** (0.043) (0.920) (1.086) 0.001***þ 0.006*** 0.011***þ (0.000) (0.000) (0.001)

2000

China

0.024 (0.023) 0.240*** (0.042) 0.002 (0.002) 0.005*** (0.002) 0.043*** (0.006) 0.009*** (0.000) 0.001*** (0.000) 0.109*** (0.026) 1.846*** (0.281) 0.000 (0.000)

1990

1990

2000

India

1990

0.033**þ (0.014) 0.385***þ (0.057) 0.051***þ (0.002) 0.018***þ (0.002) 0.070***þ (0.010) 0.005***þ (0.000) 0.000***þ (0.000) 0.068***þ (0.007) 20.358***þ (0.484) 0.004*** (0.001)

2000

Dominican Republic and Jamaica

0.065*** 0.033*** 0.023*** 0.051*** (0.018) (0.010) (0.006) (0.016) 0.315*** 0.129*** 0.059þ 0.306*** (0.036) (0.031) (0.033) (0.077) 0.011***þ 0.021*** 0.007***þ 0.098*** (0.001) (0.002) (0.001) (0.004) 0.022***þ 0.008*** 0.012*** 0.029*** (0.002) (0.001) (0.002) (0.002) 0.013***þ 0.028*** 0.042*** 0.030*** (0.005) (0.007) (0.008) (0.010) 0.007***þ 0.000 0.001***þ 0.006*** (0.000) (0.000) (0.000) (0.000) 0.001***þ 0.000*** 0.000***þ 0.000*** (0.000) (0.000) (0.000) (0.000) 0.015þ 0.239*** 0.081***þ 0.262*** (0.013) (0.019) (0.010) (0.016) 1.411*** 37.212*** 15.950***þ 38.785*** (0.192) (1.075) (0.562) (1.031) 0.002***þ 0.004*** 0.004*** 0.005*** (0.000) (0.000) (0.000) (0.001)

2000

Philippines

Note: Estimated coefﬁcients for each country are from a single regression. The dependent variable is the proportion of immigrants from a country (listed in the column heading) who belong to group ‘‘i’’ and live in state ‘‘s’’ in year ‘‘t.’’ Membership in a group is deﬁned by age, sex, education, and number of years lived in the United States. Heteroscedasticity adjusted standard errors clustered on group-year are in parenthesis. Each regression controls for 36 group effects and 36 group-year interactions. All location characteristics are interacted with the two year dummy variables and reported ﬁgures are estimates based on these interactions. *0.05opr0.10, **0.01opr0.05, and ***pr0.01. The symbol ‘‘þ’’ indicates that the coefﬁcient for 2000 is statistically different from the coefﬁcient for 1990.

Employment/population

0.017** (0.008) Real wage (in 100 dollars) 0.018 (0.022) Real per capita income 0.017*** (0.001) (in 10,000 dollars) Poverty rate (poverty rate/10) 0.001 (0.001) Unionization 0.031*** (0.004) Population (in millions) 0.006*** (0.000) Population squared (in millions) 0.001*** (0.000) Proportion other foreign-born 0.223*** (0.009) Proportion with same 3.108*** nativity (in t10) (0.110) Crime rate (per 100 persons) 0.001*** (0.000)

1990

Mexico

Appendix B. Estimates of the effect of location attributes on the current location choices of immigrants, by country of birth and year

Geographic Dispersion and Internal Migration of Immigrants 169

Appendix C. Logitistic estimates of the effect of individual characteristics and location attributes on the inter-state migration of foreign-born persons, by country of birth Mexico

Personal characteristics Male Married Speaks English well Education ¼ 12–15 years EducationZ16 years In United States for 5–10 years In United States for 10–15 years Citizen Location attributes Employment/population

China

Philippines India

Dominican Republic and Jamaica

0.007 0.008 (0.011) (0.008) 0.002 0.007* (0.004) (0.006) 0.002 0.015*** (0.001) (0.003) 0.004*** 0.012** (0.002) (0.005) 0.013** 0.059*** (0.006) (0.009) 0.015** 0.037*** (0.007) (0.005) 0.035*** 0.009** (0.004) (0.004) 0.008*** 0.035*** (0.002) (0.004)

0.019*** (0.006) 0.002 (0.004) 0.033*** (0.013) 0.007 (0.011) 0.017 (0.016) 0.009*** (0.003) 0.0001 (0.003) 0.016*** (0.002)

0.017 0.006 (0.013) (0.012) 0.016* 0.017*** (0.009) (0.006) 0.003 0.007 (0.010) (0.007) 0.001 0.023*** (0.009) (0.004) 0.035*** 0.025*** (0.008) (0.006) 0.045*** 0.002 (0.006) (0.005) 0.012*** 0.003 (0.004) (0.003) 0.031*** 0.006*** (0.005) (0.003)

0.147*** (0.055) 0.438*** (0.140) 0.003 (0.019) 0.071*** (0.012) 0.149** (0.064) 0.003*** (0.001) 0.00001 (0.00004) 0.243 (0.168) 1.985 (2.020) 0.007*** (0.003) 0.140*** (0.033)

0.119*** (0.031) 0.239* (0.131) 0.006 (0.018) 0.079*** (0.010) 0.018 (0.058) 0.000 (0.001) 0.0001** (0.00004) 0.034 (0.156) 0.635*** (0.190) 0.011*** (0.003) 0.218*** (0.059)

0.166** 0.007 (0.083) (0.070) 0.014 0.201 (0.216) (0.148) 0.012 0.026 (0.026) (0.027) 0.088*** 0.038** (0.018) (0.020) 0.152 0.198* (0.097) (0.114) 0.002 0.006*** (0.002) (0.002) 0.0001** 0.0003*** (0.0001) (0.00001) 0.080 0.782*** (0.204) (0.231) 18.62*** 0.987 (3.366) (1.052) 0.018*** 0.009** (0.005) (0.004) 0.412*** 0.327*** (0.082) (0.047)

0.046 (0.060) Real wage (in 100 dollars) 0.054 (0.104) Real per capita income (in 0.025 10,000 dollars) (0.025) Poverty rate (poverty rate/ 0.032** (0.015) 10) Unionization 0.108 (0.071) Population, (t5) (in 0.003*** (0.001) millions) Population squared (t5) 0.0001* (0.000) Proportion other foreign0.324 (0.211) born (t10) Proportion with same 0.943*** nativity (t10) (0.331) Crime rate (per 100 persons) 0.009** (0.004) Inter-state migration of 0.113*** (0.032) natives

Notes: Figures in each cell are marginal effects of logitistic regressions; heteroscedasticity adjusted standard errors clustered on origin states are in parenthesis. Location attributes are for the origin state. Employment/population and real wage are deﬁned by age and gender; employment/population, real wage, crime rate, per capita income, poverty rate, and unionization rate are origin state averages for t5, t6, and t7. Each column is from a separate regression; each regression also controls for age, number of children, number of young children, other family income, and year ﬁxed effects. *0.05opr0.10, **0.01opr0.05, and ***pr0.01.

Geographic Dispersion and Internal Migration of Immigrants

171

References Alba, R., Logan, J.R., Stults, B.J., Marzan, G., Zhang, W. (1999), Immigrants groups in the suburbs: a re-examination of suburbanization and spatial assimilation. American Sociological Review 64 (3), 446–460. Aslund, O. (2001), Now and forever? Initial and subsequent location choices of immigrants. IFAU-Ofﬁce of Labour Market Policy Evaluation. Working Paper 2001:11. Bartel, A.P. (1989), Where do the new US immigrants live? Journal of Labor Economics 7 (4), 371–390. Bartel, A.P, Koch, M.J. (1991). Internal migration of U.S. immigrants. In: Abowd, J.M., Freeman, R.B. (Eds.), Immigration, Trade and the Labor Market. A National Bureau of Economic Research Project Report. The University of Chicago Press, Chicago, pp. 121–134. Bauer, T., Epstein, G., Gang, I.N. (2002), Herd effects or migration networks? The location choice of Mexican immigrants in the US. IZA Discussion Paper No. 551. Bauer, T., Epstein, G., Gang, I.N. (2005). Enclaves, language and the location choice of migrants. Journal of Population Economics 18 (4), 649–662. Borjas, G. (1999), Immigration and welfare magnets. Journal of Labor Economics 17 (4), 604–637. Borjas, G. (2003), The labor demand curve is downward slopping: re-examining the impact of immigration on the labor market. Quarterly Journal of Economics 118, 1335–1374. Borjas, G. (2005), Native internal migration and the labor market impact of immigration. NBER Working Paper No. 11610. Card, D. (2001), Immigrant inﬂows, native outﬂows, and the local market impacts of higher immigration. Journal of Labor Economics 19 (1), 22–64. Card, D., Lewis, E. (2005), The diffusion of Mexican immigrants during the 1990s: explanations and impacts. NBER Working Paper No 11552, August. Chiswick, B.R., Miller, P.W. (2005), Do enclaves matter in immigrant adjustment?. City and Community 4 (1), 5–35. Edin, P.-A., Fredriksson, P., Aslund, O. (2003), Ethnic enclaves and the economic success of immigrants: evidence from a natural experiment. Quarterly Journal of Economics 118 (1), 329–357. Durand, J., Massey, D.S., Charvet, F. (2000), The changing geography of Mexican immigration to the United States. Social Science Quarterly 81 (1), 1–15. Greenwood, M.J. (1969), An analysis of the determinants of geographic mobility in the United States. Review of Economics and Statistics 51 (2), 189–194.

172

Neeraj Kaushal and Robert Kaestner

Greenwood, M.J. (1997), Internal migration in developed countries. In: Rosenzweig, M.R., Stark, O. (Eds.), Handbook of Population and Family Economics (Vol. 1B). Elsevier, New York. Gurak, T.D., Kritz, M.M. (2000), The Inter-state migration of US immigrants: individual and contextual determinants. Social Forces 78 (3), 1017–1039. Jaeger, D. (2000), Local Labor Markets, Admission Categories, and Immigrant Location Choice. Hunter College, New York, Unpublished document. Herna´ndez-Leo´n, R., Zu´n˜iga, V. (2003), Mexican Communities in the South and Social Capital: the Case of Dalton, Georgia. Southern Rural Sociology 19 (1), 20–45. Kaestner, R., Kaushal, N., Ryzin, G.V. (2003), Migration consequences of welfare reform. Journal of Urban Economics 53 (3), 357–376. Kaushal, N. (2005), New immigrants’ location choices: magnets without welfare. Journal of Labor Economics 23 (1), 59–80. Kritz, M.M., Nogle, J.M. (1994), Nativity concentration and internal migration among the foreign-born. Demography 31 (3), 509–524. Logan, J.R., Alba, R.D., Zhang, W. (2002), Immigrant enclaves and ethnic communities in New York and Los Angeles. American Sociological Review 67 (2), 299–322. Martin, P., Widgren, J. (2002), International migration: facing the challenge. The Population Bulletin, Vol. 57, No. 1, Population Reference Bureau, Washington, DC. Massey, D. (1985), Ethnic residential segregation: a theoretical synthesis and empirical review. Sociology and Social Research 69, 315–350. Massey, D. (Ed.) (2008). New Faces in New Places: The Changing Geography of American Immigration. Russell Sage Foundation, New York. McCarthy, K.F, Vernez, G. (1998). Immigration in a changing economy: California’s experience-questions and answers. MR-854/1-OSD/CBR/ FF/WFHF/IF/AMF. Mincer, J. (1978), Family migration decisions. Journal of Political Economy 86 (5), 749–773. Moore, S. (1998), Immigration and the rise and decline of American cities. Essays in Public Policy series: Hoover Institution. Sjaastad, L.A. (1962), The costs and returns of human migration. Journal of Political Economy 70 (Suppl.), 80–93. Schwartz, A. (1973), Interpreting the Effects of Distance on Migration, Journal of Political Economy. Smith, J.P., Edmonston, B. (1998). The immigration debate: studies on the economic, demographic, and ﬁscal effects of immigration. Panel on the Demographic and Economic Impacts of Immigration, National Research Council.

Geographic Dispersion and Internal Migration of Immigrants

173

South, S.J., Crowder, K., Chavez, R. (2005), Migration and spatial assimilation among US Latinos: classical versus segmented trajectories. Demography 42 (3), 497–521. Tienda, M., Wilson, F.D. (1992), Migration and the earnings of Hispanic men. American Sociological Review 57, 661–678. Williams, R., Jr. (1994), The sociology of ethnic conﬂicts: comparative international perspectives. Annual Review of Sociology 20, 49–79. Zavodny, M. (1999), Determinants of recent immigrants’ locational choices. International Migration Review 33, 1014–1030. Zellner, A., Lee, T.H. (1965), Joint estimation of relationships involving discrete random variables. Econometrica 33 (2), 382–394. Zhou, M., Logan, J.R. (1989), Returns on human capital in ethnic enclaves: New York City’s Chinatown. American Sociological Review 54, 809–820.

PART II

Production, Earnings and Competition

CHAPTER 7

Understanding the Wage Dynamics of Immigrant Labor: A Contractual Alternative Christoph M. Schmidta,b a

Rheinisch-Westfa¨lisches Institut fu¨r Wirtschaftsforschung, Hohenzollernstr. 1-3, D-45128 Essen, Germany b Ruhr-Universita¨t Bochum E-mail address: [email protected]

Abstract Empirical evidence on the labor market performance of immigrants shows that migrant workers suffer from an initial disadvantage compared to observationally equivalent native workers, but that their wages subsequently tend to increase faster than native earnings. Economists usually explain these phenomena by spot markets for labor and investments into human capital. By contrast, this chapter proposes a contractual model. This alternative has important implications for integration policy, because it suggests investing into the transparency of foreign educational credentials. Also contrasting human capital theory, the model suggests that permanent migrants never earn higher wages than equally skilled temporary migrants. Keywords: Migration, wage dynamics, human capital, implicit contracts Jel classiﬁcations: J31, J41, D83

1. Introduction Economic globalization and the collapse of the former Socialist economies in Eastern Europe are just two important factors that have contributed to the remarkable surge in international migration ﬂows during the past decades (OECD, 2009). Meanwhile, the typical immigration countries, Australia, Canada, and the United States, have been joined by other wealthy economies, foremost by those of Western Europe, as major receiving countries for immigrants. In all cases, the degree of their economic and social integration is an important aspect both of immigrants’ own prosperity and of the economic future of their destination countries. Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008013

r 2010 by Emerald Group Publishing Limited. All rights reserved

178

Christoph M. Schmidt

The extent of integration will determine whether the receiving society is open for further immigration or rather prefers closing its borders (Bauer et al., 2000; Fertig and Schmidt, forthcoming). This is all the more relevant for – old and new – European immigration countries, for which the age and skill distributions have been changing rapidly, as have the countries of origin of incoming migrant cohorts. Economists commonly assess the speed of economic assimilation by comparing the observed wage or earnings histories of previous immigrant cohorts to that of natives. Existing empirical evidence, most of which has been derived for the United States, suggests that immigrants typically suffer from an initial disadvantage relative to observationally equivalent native workers, but that their wages subsequently tend to rise faster. Early work used cross-sectional data and found this difference in growth rates to be remarkably high. As a stylized fact, immigrants seemed to earn even more than comparable (in terms of age and skill) natives, on average, after spending less than two decades in the destination country (Chiswick, 1978). Whereas the general tendency has been conﬁrmed in numerous studies conducted in a broad range of destination countries (see Borjas, 1994; Bauer et al., 2005, for overviews), its genuine extent is heavily debated in the literature. The seminal contribution by Borjas (1985) emphasized large variations in unobservable traits across cohorts of immigrants who entered at different times. As apparently this unobservable ‘‘quality’’ has declined steadily in US post–World War II history, the more recent literature has debated whether this apparent assimilation is genuine or simply reﬂects the inability of cross-sectional data to uncover declining cohort ‘‘quality’’. Yet, the literature never seriously questions the economic interpretation of the empirical evidence regarding assimilation dynamics, whatever their magnitude. Without exception, wage dynamics of immigrant labor are explained by productivity gains that are remunerated one-to-one in perfectly competitive spot markets for labor. The faster wage growth of migrant workers is viewed as a result of their incentives for investing into country-speciﬁc human capital and, correspondingly, their fast and substantial gains in average productivity relative to natives. This perspective supports clear policy implications. Most importantly, low initial earnings might not be a matter of concern, because most migrants enter as young adults, and the bulk of human capital investment takes place in the initial periods of a worker’s life cycle. Secondly, to improve upon the speed of integration, one should mainly support the acquisition of country-speciﬁc human capital. Contractual models of the labor market (Gibbons and Murphy, 1992; Malcomson, 1997) have demonstrated in a more general context, however, that rising age-earnings proﬁles do not necessarily reﬂect productivity growth. Rather, the observation of low initial earnings and comparatively steep immigrant wage proﬁles could be the result of increasing information

Understanding the Wage Dynamics of Immigrant Labor

179

on actual (time-constant) abilities. This chapter formalizes this idea and explores its implications for economic policy. The discussion suggests that an important element of rational migration policy is investing into the transparency of foreign educational credentials. It also contrasts human capital theory (Dustmann, 1993) that argues that observing low initial wages among recent immigrants might be indicating substantial human capital investments. According to the contractual view, this could also signal low average abilities, and thus be a matter of concern. The proposed model is an adaptation of the model by Harris and Holmstrom (1982) to the case in which the true abilities of migrant workers are uncertain. By contrast to the insightful asymmetric-information models of Stark (1991), here the imperfect information is symmetric. Increasing information on true abilities is derived from sequential observations on the workers’ output. In equilibrium, the wages of workers who generate surprisingly high output are bid up by competitive ﬁrms, whereas the wages of all other workers remain constant. On the contrary, risk-aversion leads migrants to accept initial wage disadvantages as an insurance against the detection of low ability. As a consequence, the average wages of migrant cohorts are rising over time. This mechanism complements any increases in productivity. The chapter is organized as follows. Section 2 brieﬂy reviews human capital arguments, Sections 3 and 4 present the contractual model and its equilibrium, respectively, and Section 5 discusses its policy implications.

2. The orthodoxy: Country-speciﬁc human capital Upon entry into their host country, migrants are different from natives in many respects. As their time of residence in their destination country increases, they typically grow closer to the native population in social, cultural, and economic terms. Economists are mainly concerned with the question, whether the labor market outcomes of immigrants tend to improve considerably over time, as they pick up the language and other destination country-speciﬁc skills that make them more productive. Yet, observing their performance over time also reveals, to both the immigrants themselves and their employers, the migrants’ inherent (timeconstant) abilities. The current orthodoxy in the economic migration literature neglects the question of information and emphasizes the acquisition of new skills. Whether these learning processes are sufﬁcient to narrow or even close the initial gap is of crucial importance for the economic future of the respective host country. Will the immigrants invariably belong to an economic underclass and thus become a burden to the welfare system? This question is particularly important for societies with an elaborate system of unemployment insurance and welfare like the Western European

180

Christoph M. Schmidt

countries. In addition, those are exactly the countries whose age structure bends more and more toward the elderly and thus makes it more and more unlikely to sustain the welfare state. Incoming migrants could in principle rejuvenate the society and secure the necessary economic growth, but only if their own economic performance is satisfactory. International research on the labor market performance of immigrants has tended to focus on the earnings dynamics of ﬁrst-generation immigrants. Most of the studies have been conducted for the United States (see Borjas, 1994; Bauer et al., 2005, for overviews). The inﬂuential study by Chiswick (1978) found that immigrants suffer from an initial earnings disadvantage of about 15%, but the earnings proﬁles of immigrants cross those of natives from below after about 14 years of stay and are higher for immigrants thereafter. Due to problems of identiﬁcation, the true magnitude of the growth differential between native and immigrant earnings remains hotly debated until today (Borjas, 1985, 1987, 1995; LaLonde and Topel, 1992). In particular, no clear pattern arises from the scarce evidence on immigrants to Germany (see Dustmann, 1993, and Schmidt, 1997, for early, and Fertig and Schurer, 2007, for recent contributions). Predominantly, economists interpret these wage dynamics as a reﬂection of productivity gains. Human capital theory (Mincer, 1974; Becker, 1975) formalizes the idea that, over time, a worker might acquire general knowledge about the functioning of the labor market or speciﬁc to the job. In the context of immigration, the focus is on country-speciﬁc human capital (Chiswick, 1978; Duleep and Regets, 1999) that is enhancing worker productivity with all potential employers in the destination country. For native workers, this is part of their endowment at the time of labor market entry, immigrant workers have to acquire it through active investment. It is generally argued that migrants face lower opportunity cost of foregone earnings and have higher returns on human capital investment and therefore have an incentive to invest more. Such an investment will mainly take place in the initial periods of stay in the destination country. In a spot market for labor in which workers are paid according to their value marginal product and where information about worker productivities is common knowledge, these productivity increases translate directly into lower initial earnings and a steeper earnings proﬁle. According to human capital theory, the longer the horizon over which the resulting proﬁts arise, the higher is the rate of return and, thus, the higher is the incentive to invest. The planning horizon of an immigrant at the time of entry into the labor market of the host country is determined by two aspects. First, when an older worker enters the country with the same deﬁcit in country-speciﬁc human capital as a young worker, his horizon is shorter. Consequently, he will invest less and his wage proﬁle will be relatively ﬂat. Second, the horizon depends on the intended duration of stay. A worker with a shorter expected duration of residence

Understanding the Wage Dynamics of Immigrant Labor

181

has relatively small incentives to invest into country-speciﬁc human capital. His wage proﬁle would therefore also tend to have a larger intercept and a small slope. As it is likely that most of the investment will happen in the initial periods, the earnings proﬁle of a worker with a long planning horizon should be steeper than that of a comparable worker with a short planning horizon but of equal innate ability. Unfortunately, the planning horizon of immigrants is an unobservable aspect of migrant behavior – and it might even vary over time which could complicate matters considerably. To operationalize the distinction between temporary and permanent migration, Borjas (1987) identiﬁes propensities to return to the home country by the political conditions in the source country. By contrast, Dustmann (1993) and Bauer and Sinning (forthcoming) use interview information on the expected duration of stay at survey time to infer on the expected duration at the time of immigration. Yet, it is inherently difﬁcult to disentangle the wage effects of duration of residence in the destination country from the reverse effects of match quality on duration. This identiﬁcation problem is reminiscent of the empirical literature on the wage effects of job tenure (Abraham and Farber, 1987; Altonji and Shakotko, 1987; Topel, 1991). It seems fair to conclude that the evidence regarding the precise nature of the duration of residence-wage proﬁle and of its variability with (intended) duration of residence is suggestive at best. The next section develops an alternative justiﬁcation for immigrants’ age-earnings proﬁles being steeper than those of native workers, which is completely unrelated to productivity increases. In this contractual model, the revelation of inherent abilities as duration of residence in the destination country is extended is the major force of immigrant wage growth. It will also be demonstrated that this model implies the effects of a variation in return propensities to differ from the human capital view.

3. The contractual model It is one of the undisputed results of labor economics that substantial heterogeneity exists among the members of the labor force. Typically, there are many components of heterogeneity that are revealed to the economic agents themselves only imperfectly and about which information can only be collected gradually over time. In particular, a worker’s inherent ability to perform in the labor market is such a component. Suppose that this ability can be measured by a unidimensional ability index. For native workers, the assessment of their ability is relatively easy. First, in a typical industrialized society, they went through an elaborate system of education and tests of their capabilities starting from early childhood. By the time they enter the labor market, a lot of information about them has already been revealed. Second, because native workers are

182

Christoph M. Schmidt

close to other native workers from previous generations in cultural and social terms, it is easier for all native economic agents to evaluate their abilities. In particular, employers have routinely assessed the abilities of native workers for a long time and this experience is reﬂected in the quality of their evaluation. Gauging migrant abilities is more complicated. Entering migrants themselves cannot use previous information at school or at work the same way native workers would do this. In addition, they will generally ﬁnd fewer cases similar to their own to compare themselves with. This also holds for potential employers who might not be able to assess workers’ ability with the same ease as that of native workers. Only if, over time, a relevant history of migration has been established between a particular sending country and a particular receiving country, these problems are reduced. The situation of uncertainty thus seems speciﬁcally important to Western Europe, as it increasingly receives migrants from a multitude of new countries of origin. During their labor market career in the host country, the output of migrant workers can be observed, and, over this time, increasing information about their true abilities – as compared to the written credentials they bring with them – will be revealed. This gain in information will have effects on migrants’ wages. In this section, a theory of migrant earnings growth is developed based on this uncertainty idea. The model used is an adaptation of Harris and Holmstrom (1982), a seminal contribution that has been the intellectual foundation for a large range of other important contributions (Narayanan, 1985; Thomas and Worrall, 1988; Beaudry and DiNardo, 1991; Farber and Gibbons, 1996; Chiappori et al., 1999; Holmstro¨m, 1999; Grant, 2003). In this model, the employment relation is not taken as a sequential spot exchange, but as a long-term relationship. In particular, risk-averse workers are insured by risk-neutral ﬁrms against random, publicly observed ﬂuctuations in their marginal revenue product, an application of the ideas of implicit contract theory (Parsons, 1986; Azariadis, 1989). As a consequence of this insurance arrangement, average earnings of immigrants are lower initially, but grow faster over time than those of natives, because individual abilities are more uncertain for immigrants than for native workers at the time of labor market entry. In the modeled economy, a large number of inﬁnitely lived identical ﬁrms produce a single good. The only factor of production is labor. There are overlapping and equally sized generations of migrants and natives with ﬁnite life span T. In the following, observational differences and individual subscripts are suppressed. Workers supply their labor inelastically to the ﬁrm making the most favorable offer at each period. They derive no disutility from effort and they can switch employers costlessly. The output yt of a particular worker, either immigrant or native, depends both on this worker’s ability Z and on a disturbance term et that for tractability is

Understanding the Wage Dynamics of Immigrant Labor

183

assumed to follow a standard normal distribution: yt ¼ Z þ t .

(1)

Two independent stochastic processes are modeled to determine the dynamics of individual earnings development for migrants. The ﬁrst process is taken directly from Harris and Holmstrom (1982) and concerns the way that workers and employers learn about individual productivities. At the beginning of the ﬁrst employment relationship after labor market entry, at t ¼ 0, the immigrant worker, the employer, and all other employers in the market share the identical prior belief that the individual’s ability is distributed normally with mean m0 and precision h0. This prior distribution coincides with the ability distribution of the native workers in the market. In each period t, all parties observe the workers output yt and all update their beliefs according to the Bayesian rule: fðytþ1 ZÞ xt ðZÞ xtþ1 ðZytþ1 Þ ¼ R 1 , 0 0 0 1 fðytþ1 Z Þ xt ðZ Þ dZ

(2)

where f(.) denotes the standard normal density function, and xt(.) denotes the normal density function attributed to the unknown ability at the beginning of period tþ1. Consequently, the posterior beliefs of all agents are again identical. After observing the worker’s output for t periods, this belief can be written as follows: P h0 m0 þ tt¼1 yt (3) Zy1 ; . . . ; yt N ; ðh0 þ tÞ1 . h0 þ t In particular, the mean of the individual’s perceived ability follows a martingale, and the precision increases deterministically. Whereas initial perceived mean abilities are the same for every migrant in an incoming immigrant cohort, over time the beliefs become more and more accurate for each individual. As a consequence of revealed individual information, mean perceived abilities spread out over the range of the distribution, until in the limit the belief for each worker is a point mass about his or her true ability. The limiting frequency distribution of perceived abilities equals the frequency distribution of true abilities in the native population. In a competitive spot market for labor, the consequences for observed wages would be trivial. At each point in time, every worker would be paid his or her expected marginal value product, based on the current belief about this worker’s ability. Although individual productivities do not change over time, perceived abilities and therefore individual earnings would, on average, ﬂuctuate around the population mean. In effect, even if uncertainty on abilities was higher for immigrants than for natives, there would not be any difference in average wage growth between migrants and natives.

184

Christoph M. Schmidt

The second stochastic process governs the way in which migrants decide upon return migration to their respective country of origin. At the end of each period, the migrant worker might return to the source country for exogenous reasons. In the theoretical model, the probability of return migration conditional on still residing in the host country in period t is assumed to be constant over time, q(t) ¼ q. The value of q is assumed to be perfectly observable by all parties. Temporary migrants display a high value of q, permanent migrants have a lower value of q. Migrant workers’ utility is normalized to be zero once they have returned to their source country and, while they are staying in the host country, capital markets do not allow them any borrowing or saving. Consequently, migrants are not able to smooth their consumption paths, and wages and consumption coincide in each period they spend in their destination country, ct ¼ wt. Individual preferences are assumed to be time-separable and workers are assumed to be risk-averse, Vðc1 ; . . . ; cT Þ ¼

T X ½b ð1 qÞt1 Uðct Þ,

(4)

t¼1

where UuW0 and Uvo0. Firms discount their proﬁts with the same discount rate bo1 as individuals. However, there is asymmetry in the attitude toward risk. As ﬁrms are risk-neutral, their preferences can be written as follows: T X Y ½b ð1 qÞt1 ðyt wt Þ. ðy1 ; . . . ; yT ; w1 ; . . . wT Þ ¼

(5)

t¼1

The probability of return migration is taken as exogenous by migrants in maximizing their expected utility as well as by ﬁrms in maximizing their expected proﬁts. In effect, different values of q are reﬂected in different discount rates, but those are equal for both sides of the market. 4. Optimal contracts One essential ingredient of the model is the existence of long-term labor contracts that bind employers to pay a ﬁxed wage over the worker’s entire career. On the contrary, workers are allowed to quit whenever a better offer appears, an assumption that prevents involuntary servitude. Whether these assumptions are realistic depends entirely on the labor market to be modeled. For the German economy, for example, this institutional asymmetry has indeed some intuitive appeal: Turnover is low compared to the United States (Schmidt, 2000) and the labor law is designed to protect the individual worker (Schmidt, 2008). In the model, ﬁrms can commit to output-contingent wage contracts, W ¼ ðw1 ðy1 Þ; w2 ðy1 ; y2 Þ; . . . ; wT ðy1 ; y2 ; . . . yT ÞÞ, but workers cannot commit to staying with the ﬁrm at a

185

Understanding the Wage Dynamics of Immigrant Labor

given wage in the future. Instead, wages can be bid up at any time. The labor market is competitive and ﬁrms have to solve the following problem at the time the migrant enters the labor market: T P E ½b ð1 qÞt1 Uðwt Þjm0 ; h0 maxW s:t:

E

ð1:Þ ( ð2:Þ E

T P

t¼1

½b ð1 qÞ

t¼1 T P

t1

ðyt wt Þjm0 ; h0

¼0 )

½b ð1 qÞt1 ðyt wt Þm0 ; h0 ; y1 ; . . . yt

(6) 0

t¼tþ1

where the second condition has to be satisﬁed for all t ¼ 1; . . . ; T and for all possible output histories ðy1 ; :::; yT Þ 2
186

Christoph M. Schmidt

diminished most rapidly at the initial stages of the career and each period a fraction of the wages are bid up, immigrants’ earnings proﬁles will be steeper than those of natives and the initial average gap due to the larger insurance premium will close over time. The truncation of the earnings distribution at the lower end even implies the overtaking observed by Chiswick (1978) without any need for self-selectivity arguments. At any point in time, a worker has a net value to the ﬁrm that depends on his output history, that is, on mean and precision of the current beliefs, on the current contract wage x, and on the time the worker has left in the labor market. This value has to include the probability that the worker will quit in the future, an event that occurs both when the worker’s output in this period will be high enough to have his or her wage bid up the next period and when the migrant returns to the source country: Z ~ h þ 1; xÞdNðm; ~ hÞ, ut ðm; ut1 ðm; h; xÞ ¼ m x þ ½b ð1 qÞ ~ mm t ðhþ1;xÞ

(7) where integration is over the normal distribution relevant for next period’s mean ability m~ up to the point mt(hþ1,x) beyond which the market value at t will exceed the current wage payment. Differences in the expected planning horizon of entering migrants will lead to similar predictions as for human capital theory. When the entering worker is older, his estimated productivity will be population average, but the required insurance premium will be smaller than for a young worker, because the employer expects to be stuck with a possible lemon for a shorter period. The worker’s earnings proﬁle should therefore be relatively ﬂat. When immigrants differ in their return propensities, those with high return propensity will have a higher market value at the time their contract is negotiated. Then the importance of differences in return propensities diminishes over time. Consequently, those immigrants with high return propensities should display higher intercepts and smaller slopes of their earnings proﬁle. However, there is one essential difference to the human capital argument. The earnings proﬁles of those workers with low return propensities will never cross those of immigrants with high return propensities from below, because the latter command a higher market value at any point in their work history: temporary migrants pose a smaller threat to become a burden to their employer when a low ability is revealed. The gap between migrants with different return propensities is narrowing and becomes less and less signiﬁcant toward the end of a migrant’s career, because both there are fewer periods left in the worker’s life, which the employer has to worry about, and beliefs become more and more precise. Consequently, it would be relatively easy to test for the validity of the different economic hypotheses based on information about wage performances and return propensities, if these propensities were observed.

Understanding the Wage Dynamics of Immigrant Labor

187

The estimation of a wage regression specifying an interaction of duration of residence and initial return propensities would then allow distinguishing between the following hypotheses: A low return propensity in the beginning of the career will lead to lower initial earnings and steeper earnings rise thereafter according to both hypotheses; at the end of the career, it will have a positive effect if the human capital idea is valid, but its effect will be insubstantial and still nonpositive, when the contract model is relevant. These differences in interpretation have important implications for migration policy. 5. Policy implications Starting from the apparent empirical fact that immigrants’ earnings proﬁles tend to be steeper than those of native workers, it has been shown in this chapter that several hypotheses could support this phenomenon. The main focus of the discussion was lying on the contrast between the commonly employed human capital arguments and a variant of the contract model by Harris and Holmstrom (1982). If all migration is permanent, both hypotheses lead to the same prediction. Yet, if entering immigrants can be classiﬁed according to their propensity to return to their respective country of origin, at least in principle, the described hypotheses could be distinguished empirically. In both cases, migrants with lower return propensities display steeper earnings proﬁles, but if the existence of long-term contracts drives the results, they will never catch up with the earnings of migrants with higher return propensities. Due to severe identiﬁcation problems, providing clear-cut evidence for or against the validity of a contractual view of the labor market will be difﬁcult, though. What are the policy implications for the Western European countries that expect a large stream of permanent migrants in the decades to come, but whose previous immigrants had largely been temporary migrants? The human capital view implies that new permanent immigrants to Western Europe will invest more in the acquisition of human capital than comparable temporary migrants would have done. Thus, because immigrants earn much less initially than they could to reap the beneﬁts of their investment later, low initial wages may be interpreted as a sign of these immigrants becoming a valuable part of the labor force in the near future. Low initial wages would then give no reason for concern. If the contractual explanation has high empirical relevance, however, the observation of low initial wages of incoming permanent migrants should alarm the governments of their host countries. This circumstance could be evidence for higher information disparities as well as for a low quality mix, because the wages of permanent migrants will at best catch up to those of comparable temporary immigrants. These considerations will tend to gain in importance in future decades. The OECD (2009) sketches various scenarios of future global migration

188

Christoph M. Schmidt

and draws conclusions on the policy options to be realized by the OECD countries. These projections use as their starting point the intensiﬁed migration ﬂows that have emerged both across OECD countries and from the developing to the developed world. Clearly, they have resulted from a complex range of reasons, touching spheres as diverse as economics, politics, demographics, and the society at large. Now and most likely, this will stay the case in the future, the high standard of living and political stability of the rich world exerts a strong pull for migrants from developing countries. This pull will tend to be particularly powerful in the decades to come, because rich world countries are aging rapidly, making the supply of skills and labor effort a very fruitful endeavor. The demand for migrants will confront an ample supply of potential migrants, due to push factors such as environmental degradation or lacking quality of educational systems. Correspondingly, virtually, all OECD countries have started to engage in active policies to attract foreign students and skilled foreign workers. The background for these measures is the increasing importance of skill, talent, and adaptability of human capital in a globalized labor market, which is reﬂected in intensiﬁed investments into (tertiary) education in both developing and developed countries. The net effects of these efforts on the size and structure of the immigration ﬂows to be expected, from the perspective of any single rich-world country, are difﬁcult to predict. What is quite deﬁnite, though, is that competition for skills and talents among potential destination countries will be quite ﬁerce. The OECD (2009) acknowledges that those destination countries that can draw on established networks of migrants within this country and to the sources of their immigration will continue to have an advantage in attracting newcomers. If, as can hardly be disputed, global competition for talents and skills will indeed intensify in years to come, OECD countries face important policy challenges. On the one hand, it will pay for them to invest into the more active recruitment of talents for studying in the destination country, coupled with measures to make a continued stay attractive for those migrants graduating from these courses of study. On the other hand, it will be worthwhile to help skilled workers to ﬁnd good job matches more quickly and more precisely tailored to their personal proﬁles. This could be achieved, for instance, by making the contents of academic credential acquired abroad more transparent. In the light of the contractual considerations presented there, this effort will not only be relevant empirically, but it will also beneﬁt both sides of the labor market: As symmetric uncertainty regarding genuine productivity is resolved early, the insurance premium for preparing for negative realizations will shrink. However, simply letting information disparities be resolved in the market is not a desirable strategy from the viewpoint of an immigration country. Ex post, high-ability immigrants would like to reveal their

Understanding the Wage Dynamics of Immigrant Labor

189

abilities, whereas low-ability immigrants are pleased to have insured against the detection of low abilities initially. In contrast, low-ability native workers will observe that – after true abilities have been revealed – they receive lower wages than comparable immigrants and are likely to view this as socially unjust. Thus, the existence of information about their abilities at labor market entry has hurt the native workers in the bottom of the ability distribution. To avoid this, one would prefer ﬁrms to be able to distinguish between high- and low-ability immigrant workers. In addition, the protracted tolerance of large uncertainties about the abilities of inﬂowing migrants will presumably have repercussions for the quality of future immigrant cohorts. A government should therefore try to alleviate information deﬁciencies through screening of immigrants; ﬁrms could not do this themselves because large numbers would be needed for reliable screening. This is all the more relevant for immigration countries that do not require standardized aptitude tests for immigrant admission. Acknowledgments The author is grateful for constructive comments by Thomas Bauer, David Card, Ani Dasgupta, Christian Dustmann, Rene´ Garcia, Anette Gehrig, Franque Grimard, Tim Guinnane, John Haisken-DeNew, Annamaria Lusardi, Jo¨rn-Steffen Pischke, Rolf Tschernig, Rainer Winkelmann, and Klaus F. Zimmermann. References Abraham, K.G., Farber, H.S. (1987), Job duration, seniority, and earnings. American Economic Review 77, 278–297. Altonji, J.G., Shakotko, R.A. (1987), Do wages rise with job seniority? Review of Economic Studies 54, 437–459. Azariadis, C. (1989), Implicit contracts. In: Eatwell, J., Milgate, M., Newman, P. (Eds.), The New Palgrave: Allocation, Information, and Markets. Norton, New York/London, pp. 132–140. Bauer, T.K., Haisken-DeNew, J.P., Schmidt, C.M. (2005), International labour migration, economic growth and labour markets: the current state of affairs. In: Macura, M., MacDonald, A.L., Haug, W. (Eds.), The New Demographic Regime – Population Challenges and Policy Responses. New United Nations, New York/Geneva, pp. 111–135. Bauer, T.K., Lofstro¨m, M., Zimmermann, K.F. (2000), Immigration policy, assimilation of immigrants and natives’ sentiments towards immigrants: evidence from 12 OECD-countries. Swedish Economic Policy Review 7, 11–53. Bauer, T.K., Sinning, M. (forthcoming), The savings behavior of temporary and permanent migrants in Germany. Journal of Population Economics.

190

Christoph M. Schmidt

Beaudry, P., DiNardo, J.E. (1991), The effect of implicit contracts on the movement of wages over the business cycle: evidence from micro data. Journal of Political Economy 99, 665–688. Becker, G.S. (1975), Human Capital, second ed. University of Chicago Press, Chicago. Borjas, G.J. (1985), Assimilation, changes in cohort quality, and the earnings of immigrants. Journal of Labor Economics 3, 463–489. Borjas, G.J. (1987), Self-selection and the earnings of immigrants. American Economic Review 77, 531–553. Borjas, G.J. (1994), The economics of immigration. Journal of Economic Literature 23, 1667–1717. Borjas, G.J. (1995), Assimilation and changes in cohort quality revisited: what happened to immigrant earnings in the 1980s? Journal of Labor Economics 13, 201–245. Chiappori, P.-A., Salanie´, B., Valentin, J. (1999), Early starters versus late beginners. Journal of Political Economy 107, 731–760. Chiswick, B.R. (1978), The effect of Americanization on the earnings of foreign-born men. Journal of Political Economy 86, 897–921. Duleep, H.O., Regets, M.C. (1999), Immigrants and human-capital investment. American Economic Review 89, 186–191. Dustmann, Ch. (1993), Earnings adjustment of temporary migrants. Journal of Population Economics 6, 153–168. Farber, H.S., Gibbons, R. (1996), Learning and wage dynamics. Quarterly Journal of Economics 111, 1007–1047. Fertig, M., Schurer, S. (2007), Labour market outcomes of immigrants in Germany. Ruhr Economic Papers No. 20. Fertig, M., Schmidt, C.M. (forthcoming), Attitudes towards foreigners and Jews in Germany: identifying the determinants of xenophobia in a large opinion survey. Review of Economics of the Household. Gibbons, R., Murphy, K.J. (1992), Optimal incentive contracts in the presence of career concerns: theory and evidence. Journal of Political Economy 100, 468–505. Grant, D. (2003), The effect of implicit contracts on the movement of wages over the business cycles: evidence from the national longitudinal surveys. Industrial and Labor Relations Review 56, 393–408. Harris, M., Holmstrom, B. (1982), A theory of wage dynamics. Review of Economic Studies 49, 315–333. Holmstro¨m, B. (1999), Managerial incentive problems: a dynamic perspective. Review of Economic Studies 66, 169–182. LaLonde, R.J., Topel, R.H. (1992), The assimilation of immigrants in the U.S. labor market. In: Borjas, G.J., Freeman, R.B. (Eds.), Immigration and the Workforce: Economic Consequences for the United States and Source Areas. NBER, University of Chicago Press, Chicago, pp. 67–92. Malcomson, J.M. (1997), Contracts, hold-up, and labor markets. Journal of Economic Literature 35, 1916–1957.

Understanding the Wage Dynamics of Immigrant Labor

191

Mincer, J. (1974), Schooling, Experience and Earnings. National Bureau of Economic Research, New York. Narayanan, M.P. (1985), Managerial incentives for short-term results. Journal of Finance 40, 1469–1484. OECD (2009), The future of international migration to OECD countries. Paris. Parsons, D.O. (1986), The employment relationship: Job attachment, work effort, and the nature of contracts. In: Ashenfelter, O., Layard, R. (Eds.), Handbook of Labor Economics, vol.2I. North-Holland, Amsterdam, pp. 789–848. Schmidt, C.M. (1997), Immigrant performance in Germany: labor earnings of ethnic German migrants and foreign guest-workers. Quarterly Review of Economics and Finance, 37, 379–397. Schmidt, C.M. (2000), Persistence and the German unemployment problem: empirical evidence on German labor market ﬂows. Economie et Statistique 332–333, 83–95. Schmidt, C.M. (2008), Wettbewerbso¨ffnung oder Kartellkonservierung: Welchen Zielen dient das Arbeitsrecht und welchen soll es dienen? In: Walter-Raymond-Stiftung der B.D.A. (Hrsg.), Perspektiven fu¨r eine moderne Arbeitsmarktordnung, BDA, Berlin, pp. 29–70. Stark, O. (1991), The Migration of Labor. Basil Blackwell, Cambridge, MA. Thomas, J., Worrall, T. (1988), Self-enforcing wage contracts. Review of Economic Studies 55, 541–554. Topel, R. (1991), Speciﬁc capital, mobility, and wages: wages rise with job seniority. Journal of Political Economy 99, 145–176.

CHAPTER 8

Interactions between Local and Migrant Workers at the Workplace Gil S. Epsteina,b,c and Yosef Mealemd a

Department of Economics, Bar-Ilan University, Ramat-Gan 52900, Israel CReAM-Center for Research and Analysis of Migration, London, UK c IZA-Institute for the Study of Labor, Bonn, Germany E-mail address: [email protected] d The School of Banking and Finance, Netanya Academic College, Netanya, Israel E-mail address: [email protected] b

Abstract In this chapter, we consider the interaction between local workers and migrants in the production process of a ﬁrm. Both local workers and migrants can invest effort in assimilation activities to increase the assimilation of the migrants into the ﬁrm and so increase their interaction and production activities. We consider the effect the relative size (in the ﬁrm) of each group and the cost of activities has on the assimilation process of the migrants. Keywords: Assimilation, contracts, ethnicity, market structure, networks, harassment Jel classiﬁcations: D74, F230, I20, J61, L140

1. Introduction Studies of minorities around the world show, with few exceptions, that they tend to earn wages substantially below those of comparable general workers (Altonji and Blank, 1999; Blau and Kahn, 2006, 2007; Bhaumik et al., 2006). In part, this reﬂects a failure on the part of the minority group to undertake the effort to assimilate (Constant et al., 2009). This failure can be caused in the face of high adjustment costs such as inadequate language skills, intergenerational familial conﬂicts, and, in the case of immigrants, lack of knowledge about the host country’s labor market (Chiswick and Miller, 1995, 1996; Bauer et al., 2005; Epstein and Gang, 2009). On occasion, minority workers outperform the other workers (Chiswick, 1977; Deutsch et al., 2006). Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008014

r 2010 by Emerald Group Publishing Limited. All rights reserved

194

Gil S. Epstein and Yosef Mealem

Efforts of the migrants to assimilate and efforts by the local population to accept them and to bring them into line with the local population are made. Often, the locals are less than welcoming, blaming the newcomers for depressing wages and displacing current workers – that is, causing unemployment. This presumption has very strong policy implications and is implicit, for example, in the calls for increased regulations about immigration that are heard worldwide. Yet, there is mixed evidence about the impact of minorities on wages and employment – it depends on whether they are substitutes or complement the current workers, with respect to the skills and other attributes that they bring to the labor market (Gang and Rivera-Batiz, 1994; Gang et al., 2002). Whether minorities actually lower wages and increase employment or not, the perception exists that they do. Because of this perception, the majority may take active steps to discourage minority assimilation – discrimination, isolation, and so on (see Epstein and Gang, 2006, 2009). Often, the efforts of both parties are mediated through political institutions. These institutions exist in both the minority and the majority worlds. They could be, for example, political parties, trade organizations, unions, or thugs. These are organizations that are able to overcome the free-rider problems individual members of each group have, in moving from the actions they desire to take, to actually taking them. Yet, while an organization’s purpose may be to represent the members of their group, the interests’ of the organization and that of its members do no always coincide (see, e.g., Lazear, 1999; Alesina and La Ferrara, 2000; Anas, 2002; Dustmann et al., 2004; Kahanec, 2006; Epstein and Gang, 2009). We are interested in why minorities are so often at a disadvantage relative to the majority. Assimilation efforts by the minority and the local population are elements that determine how well the minority does in comparison to the local population. We examine the consequences of increases in the numbers migrants, the local population, and the relationship in the production function of the ﬁrm where both work. We construct a model in which there are two actors: the local working population and the migrants working at the same ﬁrm and their interaction within the ﬁrm, in terms of production. Our study shows that the structure of the ﬁrm, the number of migrants, and local population are curtailed for the assimilation process. Moreover, the cost of investment is an important component and can be affected by incentives made by the employer or public policy. More speciﬁcally, we show that increasing the number of migrants in a certain ﬁrm will decrease the investment in assimilation activities by all workers, both local and migrants. In general, we show that it is better for both the local worker and the migrant when the local workers will be in separate ﬁrms. However, this is not always the case and many ﬁrms with migrants and locals working together exist. In this chapter, we consider the effects the size of the population of migrants and local workers have on the assimilation efforts

Interactions between Local and Migrant Workers at the Workplace

195

of both types of workers. We also consider the effect the cost of investment in assimilation activities has on the assimilation process of the migrants in the ﬁrm. 2. The model Consider a ﬁrm that has both locals L (LW1) and migrants (foreign workers) F (FW1). For simplicity, we assume that there is only one group of migrants. The efﬁciency/productivity level of the local workers and the migrants may not be identical. We normalize the efﬁciency level of local population workers to unity. The migrants’ productive/efﬁciency level depends on two main factors: (1) the investment made by the migrant to assimilate, a, and (2) the effort invested by local worker to help the migrant assimilate into the working place, b. We assume that the production function has the following form: 2 3 6 a þ bLa þ 1 7 6 7 (1) X ¼ f 6L; F 7, 4 5 b |ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ{zﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ} G

where G ¼ Fðða þ bLa þ 1Þ=bÞ is the effectiveness of F migrant workers. Let us explain this further. To assimilate one migrant worker, each migrant invest a units for himself and each local worker invests b units. bLa means that, despite the fact that each local worker invests b units in one migrant worker, the impact of L local workers on the assimilation of one migrant worker equals to bLa. Note that aW0 is a marginal effect that L local workers have on the effective number of migrant workers. As a increases, the local workers have a stronger impact on the assimilation of the migrants. If both the local workers and the migrants do not invest efforts for the assimilation of the migrants, the effectiveness of one migrant worker equals to 1/b. Thus, the effectiveness of F migrants will equal to ðða þ bLa þ 1Þ=bÞ. Therefore, the term G ¼ Fðða þ bLa þ 1Þ=bÞ represents the effective number of migrants working. It is assumed that the production function has decreasing returns to scale and satisﬁes fGW0, fLW0, fGGo0, and fLLo0. Let us consider a representative of the local workers and of the migrants. Each representative determines the optimal effort invested in the assimilation process. We assume that there is no free-riding and each worker invests according to the investment of the representative worker of their group. Denote by c the cost of investing one unit to assimilate by the migrant. d is the ratio between the costs of investment of the local worker and the migrant for each unit invested. Thus, the cost of one unit invested by the local worker equals: cd. For d ¼ 1, the cost of investment by the local worker and the migrant is identical. If d is smaller (greater) than the

196

Gil S. Epstein and Yosef Mealem

unit, the cost for the migrant is higher (lower) than that of the local worker. As each local worker invests b units to help each migrant assimilate, the total effort invested by a local worker for F migrants would be bF. It is assumed that the utility each worker obtains equals their wages (equaling the marginal productivity) minus the cost of investing in assimilation activities. The utility of a representative migrant will equal: a þ bLa þ 1 F ca (2) U ¼ fG b The utility of a representative local worker will equal: UL ¼ f L þ

f G abLa1 F dcbF b

(3)

Both the migrant and the local worker determine their investment in assimilation activities by maximizing the utility. The ﬁrst-order conditions for maximization of the utility of both the migrants and the local workers with respect to a and b are given by: U Fa ¼

f G f GG Fða þ bLa þ 1Þ þ c¼0 b b2

(4)

f LG La F f G aLa1 F f GG abL2a1 F 2 dcF ¼ 0. þ þ b b b2

(5)

and U Lb ¼

We assume that the second-order conditions hold1 Denote by a* and b* the optimal investment in assimilation activities invested by the foreign workers and the local workers respectively (thus a* and b* are the outcome of the ﬁrst-order condition deﬁned in (4) and (5)). Let us now consider how the investment, of the different type of workers, changes the differing parameters that identify both the production and the cost functions. We start by considering how a change in the number of migrants in the ﬁrm affects their own investment to assimilate. @a a þ 1 ¼ o0 @F F

(6)

This result states that increasing the number of migrants decreases the investment of each worker in assimilation activities. 1 The second-order condition must satisfy: U Faa ¼ ð2f GG F=b2 Þ þ ðf GGG F 2 ða þ bLa þ 1ÞÞ=b3 o0, U Lbb ¼ ðf LGG L2a F 2 Þ=b2 þ ð2f GG aL2a1 F 2 Þ=b2 þ ðf GGG abL3a1 F 3 Þ=b3 o0. Given this, the Hessian H ¼ ðf GGG f GG aL2a1 F 4 ða þ bLa þ 1ÞÞ=b5 þ ð2ðf GG Þ2 aL2a1 F 3 Þ=b4 is positive because we assume U Faa o0 and f GG o0. If we assume that f GGG ¼ f LGG ¼ f LLG ¼ 0, then the secondorder conditions hold. We will be making this assumption later on in the chapter.

Interactions between Local and Migrant Workers at the Workplace

197

The reason for this result is twofold: (1) increasing the number of migrants, against the number of local workers, increases the proportion of immigrants in the ﬁrm and, as a result, the assimilation is not so curtailed with respect to production and wages, and (2) the total effect of assimilation affects the activities of the migrants; thus, as their numbers increase, each can decrease his/her efforts, but the total investment could still increase.2 We would thus expect to see ﬁrms, with a large number of migrants, investing less effort in assimilation activities than a ﬁrm with a small number of migrants. A policy implication, in this case, could be to divide the migrants into as many ﬁrms as possible, to increase assimilation. Let us now consider how an increase in the number of migrants affects the investment of the local population. We can verify that @b b ¼ o0 @F F

(7)

This result states that increasing the number of migrants decreases the local workers’ investment in each of the migrants. The main reason for this result is that increasing the number of migrants increases the local workers’ marginal investment cost. This is true because each local worker invests efforts in assimilating each migrant and thus increases their marginal investment cost. Increasing the marginal cost decreases the investment in each migrant. Let us now consider how an increase in the number of migrants affects the total investment made by each party: ða F þ Lb FÞ. We can see that @ða FÞ=@F ¼ 1 and @ðLb FÞ=@F ¼ Lð@ðb FÞ=@FÞ ¼ 0, which means that increasing F would decrease the total effort made by the migrants (the elasticity of a with respect to F is ð@a =@FÞðF=a Þ ¼ ðða þ 1Þ=a Þo 1), but the total investment the local workers (or worker) invests in all the migrants is unchanged (the elasticity of b with respect to F is ð@b =@FÞðF=b Þ ¼ 1). Thus, even though the effort invested by the migrants may decrease, the total investment of the local population will not change. However, the effect it has on each migrant will decrease, because the number of migrants has increased. Given that the third derivatives equal or are close to zero f GGG ¼ f LGG ¼ f LLG ¼ 03 (see Epstein and Gang, 2009), we obtain that the effect of a change in the number of local workers on the assimilation of

2

On different aspects of the optimal size of minorities and the size affect on society, see Gradstein and Schiff (2006), Gradstein and Justman (2005), and Rapoport and Weiss (2003). 3 As f GG o0, the second-order conditions hold: U Faa ¼ ð2f GG FÞ=b2 o0, U Lbb ¼ ð2f GG aL2a1 F 2 Þ=b2 o0, and the Hessian H ¼ ð2ðf GG Þ2 aL2a1 F 3 Þ=b4 is positive.

198

Gil S. Epstein and Yosef Mealem

the migrants and local workers can be written as follows:4 @a ½f LG La þ bdcða 1ÞFLa1 U Faa ¼ @L bH

(8)

and ððaf GG FÞ=Lb2 Þ½4La1 Fða 1Þðc ðf GG Fða þ 1ÞÞ=b2 Þ @b þ ð2f GG bL2a1 F 2 Þ=b2 þ ðf LG La FÞ=b ¼ @L H

(9)

From (9), we can see that increasing the number of local workers in the ﬁrm will decrease their efforts ðð@b =@LÞo0Þ if the two categories of workers are rivals, fLGo0, and the marginal effect of L local workers on the effective number of migrant workers is less than one, ao1. Let us explain this result. If the two groups of workers are rivals, fLGo0, then increasing L will decrease the marginal productivity of the effective migrant worker (fG decreases), but on the contrary, increasing L will enhance the assimilation process of the migrants (La increases). But if ao1, then the former effect is stronger than the latter so that as a result, the local worker will decrease his/her efforts. Given this, and given the fact that ð@a =@LÞ40 if and only if f LG La þ bdcða 1Þo0 (from (8)), we see that the migrants increase their efforts to compensate for the reduction of the local population. Another sufﬁcient condition for an increase in the local populations’ efforts to assimilate the migrants is doLa1 (the proof is presented in Appendix). The results state that if the marginal effect of L local workers on the number of migrant workers is greater than one, aW1, and the cost of investment by the local population is smaller than that of the migrants, do1, then increasing the local population will force the migrants to divert more efforts into their assimilation activities. 4

When the third derivatives do not equal zero, we get the following: h i 2a f LG L2a1 F þ f LLGbL F þ dcLa1 Fða 1Þ U Faa f LGG Fðaþ1Þ U Lbb b b2 h i h i 2a1 3 2a 2 a þ f GGG aL b4 F ðaþ1Þ f LGGbL3 F f LG þ f LGGbbL F @a ¼ H @L

and f

La2 F 3

GGG b4 ½f LG aLða þ 1Þ þ f G aða 1Þða þ bLa þ 1Þ h i a 2a2 2 a2 f LLGbL F þ f GG abL b2 F ð2a1Þ þ f G L bFaða1Þ U Faa þ f LGG f GG aL

a1

a2

@b ¼ @L

F 3 ðabLa þ1Þ b4

2f G f GG L b3 F

2

aða1Þ

a1

f LG f GGbaL 3 a

þ f LGG f bLG3 L H

F2

F2 2

þ ðf LGG Þ

La F 3 ðaþbLa þ1Þ b4

Interactions between Local and Migrant Workers at the Workplace

199

Let us now consider how the cost of investing in the assimilation efforts affects those made by both parties. We start by analyzing the increasing cost of investment made by the local population only. We thus ask what would happen if d increases. The result is straightforward: 1. The investment by the migrants will increase: @a =@d ¼ ðb2 cL1a =f GG aFÞ40. 2. The investment by the local population will decrease: @b =@d ¼ ðb2 c=ðf GG aL2a1 FÞÞo0. The results show that increasing the cost of the local population’s investment will decrease their efforts (substitution effect). However, because their efforts have decreased, the migrants must increase their efforts to compensate. Now let us analyze the position when the cost of investment increases for both parties (an increase in c): 1. The migrants investment will increase if and only if5 aLa1 od. If the cost of the local population is greater than that of the migrants, dW1, and the marginal effect of L local workers on the effective number of migrant workers is not higher than one, ar1, then increasing both costs will force the migrants to increase their efforts in equilibrium. 2. The local population’s efforts will increase if and only if6 do0:5aLa1 . From the preceding results, it cannot be that both parties will increase their efforts as a result of increasing costs. Moreover, increasing costs may increase or decrease the migrants’ efforts as long as the local workers decrease theirs (if d is ‘‘high,’’ namely 0:5aLa1 od). In that case, the natural effect, the substitution effect, of increasing the investment cost to the local workers would be a decrease in their efforts. However, with regard to the migrants, we get two contradicting effects. On the one hand, as shown earlier, increasing the cost to the local population will increase the effort of the migrants (effect 1). On the other hand, increasing the cost to the migrants will decrease their efforts (effect 2). We have presented earlier the condition that shows the effect that is stronger: if d is ‘‘high enough,’’ aLa1 od, then the increased c has a ‘‘strong’’ effect on the cost to the local population and, as a result, the effect 1 is stronger than 2. If d is ‘‘high,’’ 0:5aLa1 od, but not high ‘‘enough,’’ aLa1 4d, then the 5 ð@a =@cÞ ¼ ðL1a b2 =f GG aFÞ½ðLa1 ðf LGG bL þ 2f GG ab þ f GGG abLa FÞ=ð2f GG b þ f GGG Fðaþ bLa þ 1ÞÞÞ d thus given that the third derivatives equal to zero, f GGG ¼ f LGG ¼ f LLG ¼ 0 we get ð@a =@cÞ ¼ ðL1a b2 =f GG aFÞðaLa1 dÞ. 6 ð@b =@cÞ ¼ ðb2 =f GG aL2a1 FÞ½d ðLa1 ðf LGG bL þ f GG ab þ f GGG abLa FÞ=2f GG b þ f GGG Fðaþ bLa þ 1ÞÞ thus given that the third derivatives equal to zero, f GGG ¼ f LGG ¼ f LLG ¼ 0, we get ð@b =@cÞ ¼ ððb2 Þ=ðf GG aL2a1 FÞÞðd 0:5aLa1 Þ.

200

Gil S. Epstein and Yosef Mealem

increased c has a ‘‘weak’’ effect on the cost to the local population and so the effect 1 is weaker than 2. We consider the result of the change in the parameter a (the marginal effect of the local population on the assimilation of the migrants). Increasing a means that the local workers have a stronger impact on the assimilation of the workers into the workplace. Increasing a: 1. Decreases the efforts invested by the migrants ð@a =@aÞo0.7 2. Has an ambiguous effect on the investment made by the local population.8 The ﬁrst result shows that as a increases, the local population plays a stronger role in the migrants’ assimilation, which depend more on the local workers activities rather than those of the migrants. Thus, the effectiveness of the migrant’s activities decreases, and as a result, they will decrease their efforts to assimilate. The second result demonstrates, that by increasing a, the local population, on the one hand, has to invest less, because their investment has a stronger effect, while on the other hand, each level of investment is more efﬁcient in increasing assimilation. Therefore, it is not clear which of the two effects is stronger. 3. Concluding remarks In this chapter, we have considered the interaction between local workers and migrants in the production process of a ﬁrm. Both local workers and migrants can invest in assimilation activities to increase their interaction and production activities. The investment made by both type of workers increases the assimilation of the workers. Both have an incentive to invest in the assimilation process; however, this causes costs on both sides. Our study shows that increasing the number of migrants in a ﬁrm will decrease the investment of each worker, both local and migrant, in assimilation activities. We have shown some general conditions under which increasing the size of the local population in the ﬁrm will force the migrants to devote more effort to assimilation activities. Increasing the local population’s investment cost will decrease their efforts (substitution effect). However, because these efforts have decreased, the migrants must increase theirs to compensate. On the contrary, it cannot be that both parties will increase efforts because of increasing costs to both local and migrant workers, in the same proportion. Moreover, increasing the cost to both parties, in the same proportion, may increase or decrease @a =@a ¼ ðb2 c f GG Fða þ 1Þ þ b2 dcL1a ln LÞ=ðf GG aFÞ. ð@b =@aÞ ¼ ðb2 cðd ln L þ La1 ÞÞ=ðf GG aL2a1 FÞ þ ða þ 1 abLa ln LÞ=ðaLa Þ: ð@b =@aÞ40 if a þ 14abLa ln L. 7 8

Interactions between Local and Migrant Workers at the Workplace

201

efforts of migrants, as long as local workers decrease their efforts. The last result, concerning the migrants, can be explained by the following two contradicting effects. On the one hand, increasing the cost to the local population will increase the migrants’ efforts. On the other hand, increasing the migrants’ cost will decrease their efforts. Earlier we have presented the condition explaining which effect is stronger. We considered the marginal effect caused to the local population because of the assimilation of migrants, a – increasing the marginal affect means that the local workers have a stronger impact on the assimilation of migrants into the workplace. The ﬁrst result shows that, as a increases, the local population plays a stronger role in the assimilation and depends more on their own activities than on those of the migrants. Thus, the effectiveness of the migrants’ activities decreases and as a result they will decrease their efforts to assimilate. As seen in the chapter, the structure of the ﬁrm, the number of migrants, and local population are curtailed for the assimilation process. Moreover, the cost of investment is an important component and can be affected by incentives made by the employer or public policy. Acknowledgment Financial support from the Adar Foundation of the Economics Department of Bar-Ilan University is gratefully acknowledged. Appendix ð@a Þ=ð@LÞ40 if and only if f LG La þ bdcða 1Þo0. Let us calculate the last expression using the ﬁrst-order conditions U Fa ¼

f G f GG Fða þ bLa þ 1Þ c¼0 þ b b2

(A.1)

f LG La F f G aLa1 F f GG abL2a1 F 2 þ þ dcF ¼ 0 b b b2

(A.2)

and U Lb ¼

From (A.1), we get: f G ¼ bc

f GG F ða þ bLa þ 1Þ b

and from (A.2), we can extract the expression f LG La þ bdcða 1Þ: f bL2a1 F f LG La þ bdcða 1Þ ¼ a bdc f G La1 GG b

202

Gil S. Epstein and Yosef Mealem

Substituting fG into the last equation: f LG La þ bdcða 1Þ f Fða þ bLa þ 1Þ a1 f GG bL2a1 F L ¼ a bdc bc GG b b f GG La1 Fða þ 1Þ a1 f LG L þ bdcða 1Þ ¼ a bdc bcL þ b a

As fGGo0, we can see that f LG La þ bdcða 1Þo0 (which is equivalent to ð@a =@LÞ40) if bdc bcLa1 o0, which is the same condition as doLa1 . References Alesina, A., La Ferrara, E. (2000), Participation in heterogeneous communities. Quarterly Journal of Economics (August), 847–904. Altonji, J.G., Blank, R.M. (1999), Race and gender in the labor market. In: Ashenfelter, O., Card, D. (Eds.), Handbook of Labor Economics, vol. 3C. Elsevier Science B.V., Amsterdam, pp. 3143–3259. Anas, A. (2002), Prejudice, exclusion and compensating transfers: the economics of ethnic segregation. Journal of Urban Economics 52 (3), 409–432. Bauer, T., Epstein, G.S., Gang, I.N. (2005), Enclaves, language and the location choice of migrants. Journal of Population Economics 18 (4), 649–662. Bhaumik, S.K., Gang, I.N., Yun, M.-S. (2006), Ethnic conﬂict and economic disparity: Serbians and Albanians in Kosovo. Journal of Comparative Economics 34 (4), 754–773. Blau, F.D., Kahn, L.M. (2006), The US gender pay gap in the 1990s: slowing convergence. Industrial and Labor Relations Review 60 (1), 45–66. Blau, F.D., Kahn, L.M. (2007), The gender pay gap. The Economists’ Voice 4 (4), Article 5. Available at http://www.bepress.com/ev/vol4/ iss4/art5 Chiswick, B.R., Miller, P.W. (1995), The endogeneity between language and earnings: international analyses. Journal of Labor Economics 13, 246–288. Chiswick, B.R., Miller, P.W. (1996), Ethnic networks and language proﬁciency among immigrants. Journal of Population Economics 9, 19–36. Chiswick, B.R. (1977), Sons of immigrants: are they at an earnings disadvantage? American Economic Review, Papers and Proceedings, 376–380.

Interactions between Local and Migrant Workers at the Workplace

203

Constant, A., Gataullina, L., Zimmermann, K.F. (2009), Ethnosizing immigrants. Journal of Economic and Behavioral Organization 69 (3), 274–287. Deutsch, J., Epstein, G.S., Lecker, T. (2006), Multi-generation model of immigrant earnings: theory and application. Research in Labor Economics 217–234. Dustmann, C., Fabbri, F., Preston, I. (2004), Ethnic concentration, prejudice and racial harassment of minorities. CReAM Discussion Paper 05/04. Available at http://www.econ.ucl.ac.uk/cream/www. econ.ucl.ac.uk/cream/ Epstein, G.S., Gang, I. (2006), Ethnic networks and international trade. In: Foders, F., Langhammer, R.J. (Eds.), Labor Mobility and the World Economy. Springer, Berlin, Heidelberg, pp. 85–103. Epstein, G.S., Gang, I. (2009), Ethnicity, assimilation and harassment in the labor market. Research in Labor Economics 79, 67–90. Gang, I.N., Rivera-Batiz, F. (1994), Labor market effects of immigration in the United States and Europe: Substitution vs. complementarity. Journal of Population Economics 7, 157–175. Gang, I.N., Rivera-Batiz, F., Yun, M.-S. (2002), Economic strain, ethnic concentration and attitudes towards foreigners in the European Union. IZA Discussion Paper 578. Available at http://www.iza.org. Gradstein, M., Justman, M. (2005), The melting pot and school choice. Journal of Public Economics 89, 871–896. Gradstein, M., Schiff, M. (2006), The political economy of social exclusion, with implications for immigration policy. Journal of Population Economics 19 (2), 197–446. Kahanec, M. (2006), Ethnic specialization and earnings inequality: why being a minority hurts but being a big minority hurts more. IZA Discussion Paper 2050. Available at http://www.iza.org. Lazear, E.P. (1999), Culture and language. Journal of Political Economy 107 (6, pt. 2), S95–S126. Rapoport, H., Weiss, A. (2003), The optimal size for a minority. Journal of Economic Behavior and Organization 52, 27–45.

CHAPTER 9

Ethnic Competition and Specialization Martin Kahaneca,b a

Department of Economics, Central European University (CEU), Nador u. 9, H-1051 Budapest, Hungary b Institute for the Study of Labor (IZA), Schaumburg-Lippe-Str. 5-9, 53113 Bonn, Germany E-mail address: [email protected]

Abstract Are ethnic specialization and thus a downward sloping labor demand curve fundamental features of labor market competition between ethnic groups? In a general equilibrium model, this chapter argues that spillover effects in skill acquisition and social distances between ethnic groups engender equilibrium regimes of skill acquisition that differ in their implications for ethnic specialization. Speciﬁcally, fundamental relationships through which relative group sizes determine whether ethnic specialization arises and in what degree are established. Thus, this chapter theoretically justiﬁes a downward sloping labor demand curve and explains why some ethnic groups earn more than others, ethnic minorities underperforming or outperforming majorities. Keywords: Human capital, ethnic specialization, spillover effects

group,

labor

market,

ethnic

Jel classiﬁcations: J15, J24, J70, O15

1. Introduction Social and economic coexistence of different ethnic groups – groups of people that differ in terms of shared cultural heritage, race, religion, language, history, beliefs, customs, values, or morals – is a worldwide phenomenon. African-Americans in the United States, Turks in Germany, and Roma in Central and Eastern Europe are all examples of distinct ethnic groups in larger societies. Their labor market experience exhibits several intriguing features. The scale puzzle that (1) minority ethnic groups on average earn less than the majority population and that (2) this earnings differential is increasing in minority share in population in a given Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008015

r 2010 by Emerald Group Publishing Limited. All rights reserved

206

Martin Kahanec

region is a well-established empirical regularity.1 Another widely documented phenomenon is the occupational segregation of ethnic groups (Blalock, 1957; Brown and Fuguitt, 1972; Hirschman and Wong, 1984). Altonji and Blank (1999) report that minority workers are overrepresented in less skilled jobs and Blacks in the US are overrepresented in speciﬁc kinds of jobs such as public administration. Grant and Hamermesh (1981), Grossman (1982), Borjas (1983, 1987, 2003), and Kahanec (2006a) provide some evidence for imperfect substitutability of ethnic labor and downward sloping demand for ethnic labor. Richman (2006) discusses how community institutions and occupational specialization create economic advantage for Jewish communities. There remain gaps in our understanding of the underlying nature of labor market competition between different ethnic groups, however. Are there fundamental mechanisms that drive ethnic groups to specialize in certain skills and jobs such that their labor is imperfectly substitutable in the labor market and thus the demand for ethnic labor is downward sloping? What conditions determine whether being a member of ethnic minority is an economic disadvantage or advantage, or the long-run outcomes of immigrant adjustment process? Since the groundbreaking work of Becker (1956), different forms of discrimination have been suggested to explain signiﬁcant differentials in labor market performance between ethnic groups.2 This literature offers some answers to the abovementioned questions, viewing ethnic specialization as a result of discrimination constraints imposed on the behavior of some ethnic groups. For example, discrimination may have driven ethnic groups into certain less-attractive sectors of the economy. From a different viewpoint, ethnic specialization could be explained as a consequence of different ‘‘tastes’’ of different ethnic groups for certain skills, jobs, or occupations (Hofstede, 1980). Ethnic specialization could also occur during immigrants’ adjustment in the host society (Chiswick et al., 2003). In this chapter, I explain labor market specialization of people of different ethnicities and the resulting earnings differentials as driven by their choice between heterogeneous skills and social networks – social structures between individual actors that facilitate their social interaction such as schools or families – where these skills are acquired. In particular, I argue that ethnic specialization is a persistent feature of the labor market even in a world free of discrimination and even in the long run.

1

See, for example, Blalock (1956, 1957), Heer (1959), Brown and Fuguitt (1972), Frisbie and Neidert (1977), and Tienda and Lii (1987). 2 The discrimination literature is immense. Major contributions include Welch (1967) and Arrow (1972a, 1972b, 1973), who discuss the so-called taste for discrimination theories; Phelps (1972), Arrow (1972a, 1972b, 1973), Aigner and Cain (1977), Coate and Loury (1993), and Lundberg and Startz (1998) elaborate on the concept of statistical discrimination. Altonji and Blank (1999) summarize this literature extensively.

Ethnic Competition and Specialization

207

Although the deﬁning cultural and social differences between ethnic groups are instrumental in generating different incentives in skill acquisition for different ethnic groups in the transaction costs tradition, they are not assumed to imply any ad hoc taste differentials over different skills. The argument draws on two fundamental properties of social interaction between different ethnic groups in human capital acquisition: spillover effects in social networks where skills are acquired and social distance between ethnic groups. That individuals learn from their peers, friends, and neighbors has been proposed by a number of scholars. As Lucas (1988) points out, ‘‘human capital accumulation is a social activity, involving groups of people in a way that has no counterpart in the accumulation of physical capital.’’ A number of scholars, such as Glaeser et al. (2002), Foster and Rosenzweig (1995), and Lazear (1999), maintain that social interaction in social networks involves positive externalities such that the aggregate resources of a network exceed the naı¨ ve sum of individual contributions. Foster and Rosenzweig (1995) develop a framework in which the efﬁciency of social learning improves in the number of involved individuals.3 Based on this literature, I adopt the premise that skill acquisition exhibits spillover effects such that the beneﬁts (in terms of the efﬁciency of skill acquisition) from social interaction in a given social network increase in the number of members of that network.4 Naturally, beneﬁts from social interaction depend not only on the number of individuals one interacts with but also on who those individuals are. In the context of interethnic social interaction, sociocultural differences between ethnic groups determine the efﬁciency of social interaction in any social network. In line with Poole (1927) and Lazear (1999), I deﬁne social distance to be the measure of subjective and objective dissimilarities between ethnic groups that hinders social interaction between the members of these groups. A natural corollary of this deﬁnition and the second essential assumption of the chapter is that an individual’s ability to beneﬁt from social interaction in a given social

3

Allen (1982), Ellison and Fudenberg (1993, 1995), and Bala and Goyal (1998) investigate the role of social interaction in learning about optimal actions. Valente (1995), Feick and Price (1987), Gladwell (2000), and Foster and Rosenzweig (1995) substantiate such approach and observe that social networks are an important vehicle of information sharing. These authors document that colleagues, friends, or neighbors share information about their discoveries, experiment outcomes, or search results. Conley and Udry (2005), Foster and Rosenzweig (1995), and Munshi (2004) provide evidence that social interactions signiﬁcantly affect farmers’ proﬁtability upon adoption of new technologies, arguing that this ﬁnding implies that farmers learn about the best practices in social interaction with their peers and neighbors, rather than only mimicking their behavior. Glaeser et al. (2003) ﬁnd large social multipliers in social interaction. Goyal (2003) surveys the literature on social learning. 4 Such effects are also known as network externalities.

208

Martin Kahanec

network negatively depends on her social distance to the other members of this network.5 To study specialization of ethnic groups, I explore the character of social networks where skills are acquired.6 Some social networks, including families, kinships, and certain religious groups, expatriate communities, schools, and clubs permit memberships exclusively from a single ethnic group. In contrast, most schools, student societies, workplaces, academic communities, and cyber networks such as the Internet are inclusive, permitting membership from any ethnic group.7 These exclusive and inclusive social networks are typically different with respect to their complexity, objectives, functions, and, as a consequence, the type of skills they support. On the one hand, in exclusive social networks people typically acquire less formal and non-cognitive skills such as verbal and non-verbal communication skills, general social knowledge and socialization skills, and capability of self-motivation, but also particular arts and crafts skills speciﬁc to exclusive social networks.8 On the other hand, inclusive social networks generally facilitate acquisition of more formal and cognitive skills such as those in e.g. mathematics, medicine, metal processing, machine operating, and banking.9 The key insight of this chapter is that in a world where heterogeneous skills are available in skill-speciﬁc social networks, the efﬁciency differentials engendered by spillover effects, social distances, and different sizes of ethnic groups systematically expose individuals of different ethnicities to different incentives as concerns skill choice. In effect, under certain conditions that are shown to depend on relative group sizes, these differing incentives make ethnic groups acquire different (combinations of) skills. The conditionality of the result on ethnic specialization has important consequences for relative performance of ethnic groups in the labor market and helps us understand some of the abovementioned empirical ﬁndings. 5

Note that social distance as deﬁned here is fully symmetric at the individual level and essentially engenders transaction costs in social interaction between members of different ethnic groups. In contrast to Akerlof (1997), who studies endogenous social distance between homogeneous agents, I consider social distance between members of different ethnic groups to be a predetermined variable that reﬂects the deﬁning distinctiveness of ethnic groups. 6 Note the different role of social networks in this chapter from that discussed by Buhai and van der Leij (2006), who study occupational segregation between social groups as resulting from the inbreeding bias in job referral social networks. 7 Thus, exclusive social networks are always segregated. Inclusive social networks may be integrated as well as segregated; the distinction made in this chapter is that exclusiveness (inclusiveness) is understood as exogenous institutional constraint on network membership while segregation (integration) as endogenous variable concerning equilibrium organization of social interaction as discussed later. 8 For example, child care, cooking, and maintenance skills acquired in the family. 9 Cognitive and noncognitive skills are discussed in, for example, Coleman et al. (1966) and Heckman (2000).

Ethnic Competition and Specialization

209

I develop the argument as follows. First, the elementary properties of ethnic competition in the labor market are described in a stylized model. Assuming imperfect substitutability of ethnic labor, the substitution effect whereby an efﬁciency unit of labor of relatively larger ethnic groups sells at relatively lower wage is modeled. Furthermore, I establish that spillovers in skill acquisition and interethnic social distances disadvantage smaller ethnic group in terms of efﬁciency of human capital acquisition. Next, in a full model, I show that these two properties of social interaction between ethnic groups engender several equilibrium regimes of skill acquisition, some of which exhibit imperfect substitutability of labor of different ethnic groups and thus justify a downward sloping labor demand curve. Then, I discuss the results and conclude.

2. The model 2.1. Demand To establish the substitution effect, consider an economy populated by the continua of individuals from two ethnic groups, I and J, with measures IW0 and JW0 and elements i and j, respectively. Assuming I þ J ¼ 1 and I J without loss of generality, one can refer to ethnic group I as the minority and J as the majority, and I and J also denote proportions of the respective groups in the population. Social distance between minority and majority individuals marks the distinction between the two ethnic groups. Individual membership in one of the two ethnic groups is predetermined for each individual. Except for group membership and social distance, all individuals are identical with respect to their preferences and endowments. Individual preferences are represented by a standard utility function u( ) deﬁned on individual consumptions of the consumption good, Ck, where k 2 fi; jg. Let the consumption good be produced by combining labor input of minority and majority individuals measured in efﬁciency units, denoted Hi and Hj, respectively, according to the aggregate production function: C ¼ FðH I ; H J Þ, (1) RI RJ where H I 0 H i di and H J 0 H j dj. This production function is assumed to exhibit standard properties: positive marginal product of each input, concavity, and constant returns to scale (CRS).Note that it does not a priori make any assumptions about the elasticity of substitution between HI and HJ, and it, in particular, does not exclude the possibility of perfect substitution. Assuming that production takes place in a perfectly competitive industry, wages equal marginal productivities, W i ¼ F H I and W j ¼ F HJ .

210

Martin Kahanec

Proposition 1 states that whenever the production technology (1) is symmetric with respect to minority and majority labor inputs, which is a natural baseline assumption, the ethnic group that supplies more labor earns a lower wage per unit of efﬁcient labor and vice versa. This is a version of the elementary economic law of diminishing marginal product that implies that scarcer resources sell at higher prices, ceteris paribus. PROPOSITION 1. In a perfectly competitive industry, whenever the production technology (1) is symmetric such that FðH I ; H J Þ ¼ FðH J ; H I Þ for every HI and HJ and satisﬁes the properties F H I 40, F H J 40, F H I ;H I o0, o and F H J ;H J o0, H I 4 H J implies W I 4 o WJ. PROOF. That FðH I ; H J Þ ¼ FðH J ; H I Þ for every HI and HJ implies o H J , F H I ;H I o0 and F H J ;H J o0 F H I ¼ F H J , whenever H I ¼ H J . If H I 4 4 imply that F H I o F H J and thus, given perfect competition, W I 4 o WJ. Consider now the case when efﬁcient labor of different ethnic groups is imperfectly substitutable in the labor market. In particular, let an increase in the supply of minority (majority) labor decrease minority (majority) wage relatively more than majority (minority) wage. In other words, let the cross-partial elasticity of complementarity be smaller than own partial elasticity of complementarity.10 Formally, FF H I ;H I FF H J ;H I o F HI F HI F HJ F HI

(2)

and FF H J ;H J FF H I ;H J o . F HJ F HJ F HI F HJ

(3)

For the sake of exposition, I denote w W I =W J and h H i =H j and adopt the representative agent hypothesis groupwise, such that H I ¼ H i I and H J ¼ H j J. It follows that H I =H J ¼ hðI=ð1 IÞÞ and thus relative wages are a function of relative labor supplies and group sizes, w wðh; IÞ. In addition, let us for the moment assume that Hi and Hj, and thus h as well, are independent of I. Proposition 2 states the result that relative wages decrease in both the relative minority size I and the minoritymajority ratio of per capita supply of efﬁcient labor h. 10

The Hicks elasticity of complementarity measures the effect on the relative price of a given factor of production of a change in the relative quantity of that factor, holding marginal costs, and the quantities of other factors constant. See Hicks (1970).

211

Ethnic Competition and Specialization

PROPOSITION 2 (THE SUBSTITUTION EFFECT). Whenever the production technology (1) satisﬁes conditions (2) and (3), h and I are independent of each other, and production is perfectly competitive, @w=ðh; IÞ=@ho0 and @wðh; IÞ=@Io0.

PROOF. Conditions (2) and (3) imply that @ðF H I =F H J Þ=@H I ¼ ðF H I ;H I F H J F H I F H J ;H I Þ=ðF H J Þ2 o0 and @ðF H I =F H J Þ=@H J ¼ ðF H I ;H J F H J F H I F H J ;H J Þ=ðF H J Þ2 40. Given that W i ¼ F H I and W j ¼ F H J under perfect competition and that FðH I ; H J Þ is CRS, it follows that @ðW I =W J Þ= @ðH I =H J Þo0. By deﬁnition, @ðH I =H J Þ=@h40 and @ðH I =H J Þ=@I40. The independence of h and I then implies @wðh; IÞ=@ho0 and @wðh; IÞ=@Io0. Intuitively, whenever efﬁcient labor of different ethnic groups is imperfectly substitutable such that an increase in the supply of a production factor depresses its own price more than the price of other production factors, an increase in the relative supply of a production factor depresses its relative price. Proposition 2 brings to light the substitution effect. Through this effect, ceteris paribus, relatively larger ethnic groups are hurt by the relative abundance of their labor in the labor market, as it depresses the relative wage per unit of their efﬁcient labor. Figure 1 depicts the substitution effect. Recalling the result of Proposition 1 and assuming symmetry such that h ¼ 1 whenever I ¼ J, wðh; IÞ is decreasing in I and attains the value of 1 at I ¼ J. On the contrary, under the condition that labor of different ethnic groups is perfectly substitutable such that the conditions (2) and (3) hold

w

w(h,I )

1

0

0.5

Fig. 1.

The substitution effect.

I

212

Martin Kahanec

as equalities, from the proof of Proposition 2, it is clear that @wðh; IÞ=@h ¼ 0 and @wðh; IÞ=@I ¼ 0, that is, the substitution effect is not operative. 2.2. Supply In this section, I characterize the supply side of the model, establishing the relationship between the share of minority (majority) individuals in the labor market and their supply of labor. We start by postulating that individuals maximize a well-behaved utility function U(C). For simplicity, let us assume that each individual is endowed with two units of time, one of which is inelastically supplied in the labor market and the other one is spent in skill acquisition. Thus, each individual faces the time constraints Lk ¼ 1 ¼ Z k , where Lk is the time individual k spends in skill acquisition, spending the rest of her time, Zk, working. To capture the role of local spillover effects and social distance in human capital acquisition in an easily tractable way, assume for the moment that any given individual interacts with all other individuals (I minority and J majority individuals) in a single economy-wide social network. Let the continuous and differentiable function N(.) characterize the spillover beneﬁts from social interaction in this network. Throughout the chapter, I assume that agents take these spillover effects as given, provided the inﬁnitesimal measure of any individual. I formalize skill acquisition technology as follows: S i ¼ SðLi Þð1 þ NðIÞ þ NðJ=ð1 þ dÞÞÞ

(4)

S j ¼ SðLj Þð1 þ NðI=ð1 þ dÞÞ þ NðJÞÞ,

(5)

where Si and Sj denote human capital of minority and majority workers, respectively, and the continuous and differentiable function S(Lk) satisﬁes dSðLk Þ=dLk 40 and d 2 SðLk Þ=dL2k 0. Given the assumptions above, N(.) is monotonously increasing in the numbers of individuals involved in social interaction, I and J.11 The parameter d40 captures the premise that the spillover beneﬁts from social interaction with individuals of different ethnicity decrease in social distance between ethnic groups. Social distance between members of the same ethnic group is normalized to zero. Assuming that efﬁcient labor is the product of labor time and human capital, individual supplies of efﬁcient labor Hi and Hj are,

11

H i ¼ ð2 Li ÞS i ¼ 1 þ NðIÞ þ NðJ=ð1 þ dÞÞ

(6)

H j ¼ ð2 Lj ÞS j ¼ 1 þ NðI=ð1 þ dÞÞ þ NðJÞ,

(7)

Decreasing returns to social interaction would be a natural assumption, but it is not necessary.

Ethnic Competition and Specialization

213

where the normalization Sð1Þ ¼ 1 and the abovementioned assumption Lk ¼ 1 are used.12 Proposition 3 states that the spillover effects in human capital acquisition and the social distance between minority and majority individuals disadvantage smaller ethnic groups in terms of efﬁciency of human capital acquisition, if the spillover function N(.) satisﬁes the condition dNðKÞ=dK4dNðK=ð1 þ dÞÞ=dK for any K 2 fI; Jg and d40. This condition, adopted henceforth, implies that the marginal beneﬁts from social interaction are not decreasing too fast. It is satisﬁed, for example, by any homogeneous function of degree d40. PROPOSITION 3 (THE EFFICIENCY EFFECT). Given a positive social distance d and that N(.) satisﬁes dNðKÞ=dK4dNðK=ð1 þ dÞÞ=dK for any K 2 fI; Jg, technologies (6) and (7) imply that @hðIÞ=@I40 and, because IoJ, H i oH j . PROOF. dH i =dI ¼ ðdNðIÞ=dIÞ þ ðdNðJ=ð1 þ dÞÞ=dJÞðdJ=dIÞ and dH j =dI ¼ ðdNðI=ð1 þ dÞÞ=dIÞ þ ðdNðJÞ=dJÞðdJ=dIÞ: As dNðKÞ=dK4dNðK=ð1 þ dÞÞ= dK, ðdNðIÞ=dIÞ4ðdNðI=ð1 þ dÞÞ=dIÞ, and ðdNðJÞ=dJÞ4ðdNðJ=ð1 þ dÞÞ= dJÞ, for any admissible I, J, and d. Because J ¼ 1 I, dJ=dI ¼ 1. Therefore, dH i =dI4dH j =dI for any I. Noting that if I ¼ J, it holds that H i ¼ H j , dH i =dI4dH j =dI for any I implies H i oH j for any IoJ. It follows that dh=dI ¼ ðdH i =dIH j H i dH j =dI=H 2j 40Þ. Proposition 3 exempliﬁes the second of the two key relationships discussed in this chapter – the efﬁciency effect. Through this effect, larger ethnic groups are relatively more efﬁcient than smaller ones in human capital acquisition. Intuitively, a member of a smaller ethnic group has a relatively smaller pool of members of her own ethnic group with whom she can socially interact without being obstructed by social distance. In effect, the chance that she is disadvantaged in social interaction by the inefﬁciencies engendered by social distance is relatively higher than that of a member of a relatively larger ethnic group. Figure 2 depicts h as a function of I, which is upward sloping due to the efﬁciency effect and reaches unity at I ¼ J. 2.3. The equilibrium Sections 2.1 and 2.2 depict the properties of the relationship between minority-majority wage and labor ratios, w and h, and minority percentage, I, as determined by the demand and supply sides, respectively. In this section, I turn to the equilibrium properties of these relationships. 12

Without any bearing on the argument, these technologies of producing efﬁcient labor can be reinterpreted as the production functions of intermediate goods Hk, which are inputs in the production of the consumption good C.

214

Martin Kahanec w

1

h(I )

0

0.5

Fig. 2.

I

The efﬁciency effect.

As h is independent of w, as apparent from Section 2.2, the equilibrium properties of h as a function of I are fully determined by the supply side and thus not different from those presented in Proposition 3. Therefore, in the equilibrium, h(I) is increasing in I. As concerns the properties of the relationship between the minoritymajority wage ratio and minority percentage in the equilibrium, these are determined by the demand side, as depicted in Proposition 2, but also by the supply side, whereby h is a function of I. We know from the demand side analysis of Section 2.1 that, taking h and I independent of each other, wðh; IÞ is decreasing in each of its arguments. Section 2.2 tells us that h is an increasing function of I, however. Proposition 4 resolves the equilibrium relationship between w and I, establishing that Proposition 2 and Proposition 3 imply that minority-majority wage ratio is decreasing in minority percentage in the equilibrium, as depicted in Figure 1. PROPOSITION 4. Whenever the production technology (1) satisﬁes conditions (2) and (3), d40, and N(.) satisﬁes dNðKÞ=dK4dNðK=ð1 þ dÞÞ=dK for any K 2 fI; Jg, @wðIÞ=@Io0. PROOF. From Proposition 2, given the independence of h and I, conditions (2) and (3) imply that @wðh; IÞ=@ho0 and @wðh; IÞ=@Io0. From Proposition 3, given d40 and that N(.) satisﬁes dNðKÞ=dK4dNðK=ð1 þ dÞÞ=dK for any K 2 fI; Jg, @hðIÞ=@I40. It is straightforward to see that @wðh; IÞ=@ho0, @wðh; IÞ=@Io0, and @hðIÞ=@I40 imply that wðhðIÞ; IÞ ¼ wðI Þ is decreasing in I. This result is intuitive. Due to the substitution effect, minority-majority relative wage decreases in minority share. An increase in the share of

Ethnic Competition and Specialization

215

minority people increases their efﬁciency in human capital acquisition through the efﬁciency effect such that their per capita supply of efﬁcient labor increases relative to the per capita supply of efﬁcient labor of majority people. This increase further depresses minority-majority relative wage through the substitution effect. 3. Specialization of ethnic groups The substitution and efﬁciency effects link the nature of ethnic competition in the labor market to the relative sizes of ethnic groups. In particular, the relative strengths of these effects determine the properties of w(I), h(I), and relative earnings oðIÞ wðIÞhðIÞ.13 While these two effects can generate various patterns of ethnic earnings inequality, there is an important precondition for the substitution effect to be operative, namely, specialization of ethnic groups such that the conditions (2) and (3) are satisﬁed. Otherwise, ethnic earnings inequality is driven solely by the efﬁciency effect, whereby, contrary to the observed scale puzzle, relatively larger ethnic groups outperform smaller ones. This is easily seen in the model, because perfect substitutability of minority and majority labors implies that F H I ¼ F H J and thus W I ¼ W J , which in turn implies that oðIÞ ¼ hðIÞ. Proposition 3 then implies that oðIÞo1 and @oðI Þ=@I40. The fundamental insight of this section, and the whole chapter, is that local spillover effects and minority-majority social distance under some conditions drive members of different ethnic groups to choose different combinations of exclusive and inclusive skills to acquire. Such ethnic specialization engenders the substitution effect and causes the demand for labor of any given ethnic group to be decreasing in this group’s relative size. To see this, let us relax the assumption about the inelastic allocation of time, such that individual is now free to chose how much of the endowed time he spends working and how much acquiring human capital. In addition, introducing the heterogeneity of skills in the model, I let the individual choose between exclusive and inclusive skills. Re-normalizing the time constraint such that each individual has one unit of time we obtain: Lkx þ Lkn þ Z kx þ Z kn 1,

(8)

where Zkm is the time spent by individual k in utilizing skill m 2 fx; ng of, respectively, exclusive or inclusive type, in production. Given that there are skill-speciﬁc social networks where skills are acquired, technologies (4) to (7) need to be reformulated. In particular, denoting social networks correspondingly to the skills they entail, we 13

Kahanec (2006b) discusses the conditions under which these effects explain the scale puzzle.

216

Martin Kahanec

assume that the following functions describe acquisition of skills: S im ¼ SðLim Þð1 þ NðI im Þ þ NðJ im =ð1 þ dÞÞÞ

(9)

S jm ¼ SðLjm Þð1 þ NðI jm =ð1 þ dÞÞ þ NðJ jm ÞÞ,

(10)

where S(.) is a decreasing returns to scale function of time spent in skill acquisition and Ikm and Jkm denote the numbers of members of ethnic groups I and J in network m of which individual k is a member, respectively. With two kinds of skills that increase the efﬁciency of time spent working, it is assumed that efﬁcient labor is a composite of timeempowered exclusive and inclusive skills as follows: H k ¼ HðZ kx S kx ; Z kn Skn Þ

(11)

where H(.,.) is a well-behaved production function increasing in its arguments with decreasing returns to each input.14 Given the difference of exclusive and inclusive skills, the qualitative properties of individual efﬁcient labor are determined by the combination of skills that the individual has. I operationalize this qualitative variation of efﬁcient labor such that efﬁciency units of labor that involve different (combinations of) skills are imperfect substitutes in the labor market. Thus, for example, if the skills of one individual are predominantly exclusive and the skills of another individual are predominantly inclusive, the elasticity of substitution of labor of these two individuals is ﬁnite. Formally, deﬁning sk S kx =S kn , whenever sk ask0 (sk ¼ sk0 ) for individuals k and ku, the elasticity of substitution between Hk and H k0 is ﬁnite (inﬁnite).15 To establish that there are equilibrium regimes of skill acquisition under which people of different ethnicities choose different (combinations of) skills, this section investigates the individual problem of time allocation. Individuals maximize their utility, taking their resource constraints, available technologies, network effects, wages per unit of their efﬁcient labor, and the price level as given. Because the utility function depends only on the amount of consumption good that an individual consumes, it follows that the agents’ problem boils down to: MaxfH k g

(12)

Lkm ;Z km

14 This technology of producing efﬁcient labor Hk can be, without any bearing on the argument of this chapter, reinterpreted as the production function of the intermediate good Hk, which is an input in the production of the consumption good C. 15 Because sk is determined by the organization of human capital acquisition, which is endogenous in the model, the degree of substitutability between minority and majority labor is in this sense endogenous as well. Whether we deﬁne substitutability of ethnic labor as a function of sk Skx/Skn or sk ZkxSkx/ZknSkn has no bearing on the argument, as discussed below in Section 3 and in the Appendix.

Ethnic Competition and Specialization

217

subject to technologies (9)–(11), and the resource constraints (8), Z km 0 and Lkm 0. Clearly, this maximization problem is largely determined by the elasticity of substitution between exclusive and inclusive skills. To illustrate if these skills are good substitutes in producing Hk, the individual may choose to acquire only that skill that she can acquire more efﬁciently. In contrast, if there are strong complementarities between the two types of skills, the individual will strive to acquire both of them. The speciﬁc condition separating these two regimes under speciﬁc production technologies is derived in the Appendix. The key question, however, is under what conditions people of different ethnicities choose different (combinations of) skills. For each individual, this choice is driven by the efﬁciency of skill acquisition across social networks. From Equations (9) and (10) one can see that social distances and group sizes generate different trade-offs for members of different ethnic groups in skill acquisition. Namely, given spillover effects and social distances in skill acquisition, asymmetric sizes of ethnic groups generate asymmetric trade-offs for members of different ethnic groups and thus drive them to acquire different (combinations of) skills. To investigate the effects of such asymmetries on equilibrium regimes of skill acquisition, I adopt the trembling hand perfect version of the Nash equilibrium. Speciﬁcally, I deﬁne stable equilibrium as the state in which no individual has incentives to deviate, that is, to change his or her allocation of time across social networks, even if, with negligible probability, individuals unintentionally play off-the-equilibrium strategies. Given this equilibrium concept, we can state Proposition 5 about stable equilibrium regimes of skill acquisition: PROPOSITION 5. In any stable equilibrium, no agent is involved in more than one network of any given type, exclusive or inclusive. PROOF. An individual is involved in two (or more) social networks of the same type if and only if their efﬁciencies for this individual are the same. Given the strictly increasing spillover effects, in any stable equilibrium this cannot happen, however, because any perturbation of agents’ involvements makes one of the networks less efﬁcient and causes this individual to abandon it. Similarly, if the combination of skills possessed by an individual is not directly observable in the labor market and workers are distinguished only by their ethnicity, which is a standard asymmetric information assumption, Proposition 6 ensues. PROPOSITION 6. In any stable equilibrium all members of a given ethnic group choose the same combination of skills to acquire.

218

Martin Kahanec

PROOF. If individuals take the wage for a unit of their efﬁcient labor as given with respect to their choice of skills, individuals pick that combination of skills (and thus social networks) that they can acquire most efﬁciently. Consider now an equilibrium with two individuals from the same ethnic group that are involved in two different combinations of social networks. It must then be the case that the efﬁciencies of these two combinations of social networks for the two individuals in production of efﬁcient labor are the same. Such equilibrium is, however, unstable. Any marginal deviation from the distribution of individuals across these two different combinations of networks causes their efﬁciencies to differ, given the strictly increasing spillover effects, and all individuals abandon the less efﬁcient one. Proposition 5 and Proposition 6 imply that at most two different types of labor are supplied in the economy, one speciﬁc for the minority and one for the majority ethnic group.16 In this sense, because these propositions do not rest on the particular speciﬁcation of the production function (1), but on the asymmetric information assumption on which Proposition 6 hinges, this production function can be seen as a harmless simpliﬁcation of a more general production technology with an arbitrary number of types of labor H. To develop the argument that there are equilibria that exhibit specialization of ethnic groups, consider the case in which there are strong complementarities between exclusive and inclusive skills such that individuals necessarily acquire both types of skills. Because all agents of a given type choose the same set of networks and thus skills to acquire, as established in Proposition 6, two different equilibria can arise. In the DI equilibrium, ethnic groups acquire exclusive skills in their group-speciﬁc exclusive social networks and inclusive skills in one integrated inclusive social network where both ethnic groups interact. In the DS equilibrium, in contrast, inclusive skills are acquired in two segregated inclusive social networks, one with minority and one with majority members. In this sense, the DI equilibrium is integrated and the DS equilibrium segregated.17 Proposition 7 discusses the stability of these equilibria. PROPOSITION 7. The DI equilibrium is always stable. The DS equilibrium is stable if and only if the condition I 1=ð2 þ dÞ holds.

16

Note that existence of at least one nonempty social network and thus the existence of a stable equilibrium is not an issue in this model, because individuals always acquire some skills (see technology (11)) and thus are members of at least one social network. See also the discussion and propositions later on the stability of speciﬁc equilibria. 17 Note, however, that there is a degree of segregation in the DI equilibrium as well, as the exclusive networks are by deﬁnition segregated.

Ethnic Competition and Specialization

219

PROOF. The only possibility for an individual to deviate in the DI equilibrium is to form his or her own inclusive social network. Because such network would offer zero network beneﬁts, such deviation is never proﬁtable and the DI equilibrium is therefore stable. Under DS DS DS the DS regime, SDS ix ¼ SðLix Þð1 þ NðIÞÞ and S in ¼ SðLin Þð1 þ NðIÞÞ. A minority individual can deviate only to majority inclusive social network, 0 DS0 ¼ SðL facing S DS in in Þð1 þ NðJ=ð1 þ dÞÞÞ. Thus, the DS equilibrium is stable only if 1 þ NðIÞ 1 þ NðJ=ð1 þ dÞÞ. This condition boils down to I 1=ð2 þ dÞ, given that I þ J ¼ 1. Applying the same line of reasoning to majority individuals, we arrive at the condition J I=ð1 þ dÞ, which is always satisﬁed, however. Do ethnic groups specialize in these equilibria? To answer this question, one needs to look at the relative efﬁciencies of inclusive and exclusive social networks for each ethnic group. In the DI equilibrium for the minority ethnic group, these efﬁciencies are characterized by Sin ¼ SðLin Þð1 þ NðIÞ þ NðJ=ð1 þ dÞÞÞ in the inclusive social network and Six ¼ SðLix Þð1 þ NðIÞÞ in the exclusive one. The respective efﬁciencies for the majority ethnic group are S jn ¼ SðLjn Þð1 þ NðI=ð1 þ dÞÞ þ NðJÞÞ and S jx ¼ SðLjx Þð1 þ NðJÞÞ. Therefore, the efﬁciency trade-offs for minority and majority individuals are the same, if 1 þ NðIÞ þ NðJ=ð1 þ dÞÞ 1 þ NðI=ðð1 þ dÞÞÞ þ NðJÞ ¼ 1 þ NðIÞ 1 þ NðJÞ which is equivalent to NðJ=ð1 þ dÞÞ=ð1 þ NðIÞÞ ¼ NðI=ð1 þ dÞÞ= ð1 þ NðJÞÞ. Clearly, this equality never holds for admissible values of I, J, and d and thus minority and majority individuals are never equally efﬁcient in exclusive and inclusive social networks in relative terms. In particular, minority individuals are relatively more (less) efﬁcient in inclusive (exclusive) social networks than majority individuals. As a result, individuals from ethnic groups of different sizes have different incentives as concerns allocation of time between exclusive and inclusive social networks. Whether these dissimilar incentives lead to dissimilar combinations of exclusive and inclusive skills possessed by individuals of different ethnicities is somewhat more involved a question. The reason is that to beneﬁt from the complementarities between exclusive and inclusive skills that are present under the DI equilibrium, individuals may want to compensate for the efﬁciency differentials across social networks by investing more time in acquiring or utilizing skills acquired in relatively less-efﬁcient social networks. The intuition why such compensating time investment does not lead to the same combination of skills possessed by members of different ethnic groups is straightforward. It rests on the fact that such compensating behavior is costly in terms of overall efﬁciency of skill acquisition and utilization, as compensating implies that relatively

220

Martin Kahanec

larger amounts of time are invested in skills that are acquired relatively less efﬁciently. It is these costs that prevent individuals from fully compensating for the efﬁciency differentials between exclusive and inclusive networks, unless H(.,.) exhibits perfect complementarity between skills. Because these efﬁciency differentials vary across ethnic groups but beneﬁts from complementarities between skills by assumption do not, as the function H(.,.) is assumed to be the same for all ethnic groups, people of different ethnicities acquire different combination of skills. In effect, their labor is imperfectly substitutable in the labor market under the DI equilibrium. Because the efﬁciency differentials are a function of the relative sizes of ethnic groups, the degree of specialization and thus substitutability of labor of different ethnic groups is a function of their relative sizes. This intuitive argument about specialization under the DI equilibrium is formalized in the Appendix for speciﬁc functional forms of production technologies. In the DS equilibrium, the efﬁciencies of minority and minority people in skill acquisition are characterized by Sin ¼ SðLin Þð1 þ NðIÞÞ in the inclusive social network and S ix ¼ SðLix Þð1 þ NðIÞÞ in the exclusive one. On the contrary, the respective efﬁciencies for the majority ethnic group are S jn ¼ SðLjn Þð1 þ NðJÞÞ and S jx ¼ SðLjx Þð1 þ NðJÞÞ. Clearly, the efﬁciency trade-offs between different types of skills are the same across ethnic groups and no ethnic specialization occurs in the DS equilibrium. Proposition 8 summarizes these results. PROPOSITION 8. If exclusive and inclusive skills are not perfect complements (but exhibit sufﬁciently strong complementarities such that individuals acquire both types of skills), under the DI equilibrium ethnic groups of different sizes acquire different combinations of skills. No ethnic specialization occurs under the DS equilibrium, however. If exclusive and inclusive skills are good substitutes, acquiring both types of skills is not necessary and individuals choose just one type of skills to acquire in the equilibrium. The reason is that under such condition, any equilibrium in which individuals combine exclusive and inclusive skills is unstable by the same argument as discussed in the proof of Proposition 6. Five equilibria of this type are possible. First, there are three equilibria in which ethnic groups specialize in the same kind of skills, exclusive (EE) or inclusive (II, IIS). The distinction between the II and IIS equilibria is that under the II equilibrium, there is only one integrated social network in the economy in which minority and majority individuals interact. On the contrary, in the IIS equilibrium, minority and majority individuals form two ethnically segregated inclusive social networks. Second, there are two equilibria under which ethnic groups specialize, one ethnic group acquiring exclusive and the other one inclusive skills (EI, IE). Clearly, ethnic specialization occurs under the EI and IE equilibria, minority and majority individuals acquiring different types of skills.

221

Ethnic Competition and Specialization

Table 1. Equilibrium regimes of skill acquisition Equilibrium Skills of minority people

Skills of majority people

Stability condition

Ethnic specialization

DI DS EE EI IE II IIS

Mix of both types Mix of both types Exclusive Inclusive Exclusive Inclusive (Integrated) Inclusive (Segregated)

None I 1=ð2 þ dÞ None I 1=ð2 þ dÞ None None I 1=ð2 þ dÞ

Yes No No Yes Yes No No

Mix of both types Mix of both types Exclusive Exclusive Inclusive Inclusive (Integrated) Inclusive (Segregated)

The EE, IE, and II equilibria are stable without further restrictions. Under the EE and II equilibria, no deviation to a nonempty network is possible. Under the IE equilibrium, deviation of majority people to the inclusive minority network is possible, but it is inefﬁcient by the argument similar to the one in the proof of Proposition 7. By the same argument, ﬁnally, the EI and IIS equilibria are stable only if the condition I 1=ð2 þ dÞ is satisﬁed. Proposition 9 restates the key results on ethnic specialization under good substitutability of exclusive and inclusive skills. Table 1 summarizes all the seven different equilibria that can arise in the economy. PROPOSITION 9. If exclusive and inclusive skills are good substitutes such that no individual acquires both types of skills, the IE, EE, and II equilibria are always stable. The EI and IIS equilibria are stable under the condition I 1=ð2 þ dÞ. Ethnic specialization occurs in the EI and IE equilibria. The stability condition I 1=ð2 þ dÞ plays an important role in determining which equilibria are stable in the economy. In fact, if this condition holds, not only are all the seven different equilibria viable, so are they in the absence of the assumed institutional exclusivity of exclusive social networks.18 In other words, elimination of institutional exclusion in exclusive social networks does not lead to integration, whenever the size of minority or the social distance between ethnic groups is sufﬁciently large. Provided that this condition holds, one can generalize the argument of this chapter to social contexts without institutional exclusion. An important consequence of the dependence of the stability condition on relative group sizes is that substitutability of labor of different ethnic groups across labor markets is a function of relative group sizes. For example, in regions where ethnic minority is large enough, the EI equilibrium may prevail, while this is not be possible in regions with 18

This result follows from the proof of Proposition 6.

222

Martin Kahanec

relatively small number of minority people, because they have strong incentives to integrate and the II equilibrium prevails. It is also worthwhile to note and easy to see from the preceding analysis that the key result of this chapter on the existence of stable equilibria under which ethnic groups specialize is robust with respect to an alternative assumption of a negative social distance, that is, ‘‘taste’’ for inter-ethnic social interaction.

4. Discussion and conclusions This chapter elucidates the nature of labor market competition between different ethnic groups as driven by their relative sizes and classiﬁes the equilibrium regimes of human capital acquisition in the context of ethnic competition. It is shown that the counteracting substitution and efﬁciency effects are driving relative earnings of ethnic groups of different sizes. While the efﬁciency effect is always operative, given that there are spillover effects and interethnic social distances in skill acquisition, whether the substitution effect is present or not depends on whether ethnic groups specialize in different skills. The key result is that, besides equilibria under which no specialization occurs, there are stable equilibrium regimes of skill acquisition under which ethnic groups acquire different (combinations of) skills. This result is driven by the spillover effects in skill acquisition that, given the positive social distance between ethnic groups, expose different ethnic groups to different efﬁciency trade-offs in acquisition of skills acquired in exclusive social networks, such as the family, and inclusive social networks, such as the school. It implies that ethnic specialization is not necessarily a transient phenomenon that withers away after a period of immigrant adjustment. Which of the multiple equilibria possibly arising in the economy prevails depends especially on the degree of complementarity between exclusive and inclusive skills that determines whether individuals acquire both or just one type of skills. Relative group sizes and social distances between ethnic groups are the key determinants of stability of these equilibria. The multiplicity of equilibria is in fact informative in the light of the mixed results of the studies on the substitutability of ethnic labor. In particular, it implies that the nature of ethnic competition may vary across labor markets such that in some labor markets labor of different ethnic groups is perfectly substitutable while imperfect substitutability prevails in others. Speciﬁcally, this study shows that the degree of specialization of ethnic groups and thus the substitutability of labor of different ethnic groups is a function of their relative sizes. This result stems from the size-dependent efﬁciency trade-offs in skill acquisition in the DI equilibrium that imply that time and skill allocations depend on relative sizes of ethnic groups. Another reason why the relative size of ethnic groups matters for ethnic specialization is the dependence of the stability of some equilibria on the

Ethnic Competition and Specialization

223

size-dependent stability condition that may decide between equilibria with perfectly and imperfectly substitutable labor of different ethnic groups. For example, if the size of the minority community (and thus the spillover beneﬁts it generates) decreases, minority individuals previously specialized in skills acquired in exclusive social networks such as their kinship network may ﬁnd it efﬁcient to integrate into inclusive social networks such as the school. Hence, empirical tests on the slope of the labor demand curve need to take into account variation of the substitution effect across (local) labor markets with different character of ethnic competition and distinguish these effects from human capital spillover effects driven by the relative sizes of ethnic groups. As the proposed model is ‘‘orthogonal’’ to the existing theories of ethnic earnings differentials, such as the theories of (statistical) discrimination, any empirical testing would need to control for the established alternative theories and confounding factors. Specialization is thus viable even for a relatively small minority whenever (1) a sufﬁciently large social distance sustains the EI equilibrium, (2) ethnic majority prevents ethnic minority from their exclusive social networks sustaining the IE equilibrium, or (3) strong complementarities between exclusive and inclusive skills result in the DI equilibrium. Under such circumstances, the substitution effect may, depending on the parameters of the model, outweigh the efﬁciency effect and drive the earnings of ethnic minority above those of ethnic majority. This chapter thus also offers an explanation of why and under what circumstances ethnic minorities may attain higher earnings than majorities. From the policy perspective, this chapter shows that one-off policy measures that induce people to switch between exclusive and inclusive social networks may be effective in improving the overall efﬁciency of the economy. In particular, a policy maker may wish to achieve integration to increase the size of social networks and thus the efﬁciency beneﬁts from spillover effects. However, any such policy must be carefully considered for the price effects of integration that may arise in those cases when the policy leads to (de-)specialization of ethnic groups and thus affects the substitution effect. Furthermore, the effects of such policies on aggregate production also depend on the strength of complementarities between human capitals of different ethnic groups. From a different perspective, integration does not necessarily lead to elimination of ethnic specialization, as evidenced by the DI equilibrium.

Acknowledgments I am grateful to Barry R. Chiswick, Carmel U. Chiswick, Sjak Smulders, William J. Baumol, Patrick Francois, and audiences at various seminars at IZA Bonn, Tilburg University, and NAKE for their helpful comments on previous drafts of this chapters. All remaining errors are mine.

224

Martin Kahanec

Financial support from Volkswagen Foundation for the IZA project on ‘‘The Economics and Persistence of Migrant Ethnicity’’ is gratefully acknowledged. Appendix A.1. Derivation of equilibrium properties using speciﬁc functional forms To establish the results discussed in Proposition 8 in a more speciﬁc setup, I introduce speciﬁc functional forms for the aggregate production technology (1) and technologies (9), (10), and (11) to analytically solve the model. Let the consumption good be produced in a perfectly competitive industry according to the constant elasticity of substitution (CES) aggregate production function: Z ðr1Þ=r Z ðr1Þ=r !r=ðr1Þ I

J

H i di

C¼ 0

H j dj

þ

(A.1)

0

with the elasticity of substitution r41. According to this speciﬁcation, labor of any given type has decreasing marginal returns, production exhibits CRS, and no type of labor is essential in production. Furthermore, I assume the CES technology of producing Hk efﬁciency units of labor, H k ¼ ½ðSkx Z kx Þð1Þ= þ ðS kn Z kn Þð1Þ= =ð1Þ ,

(A.2)

where the ﬁnite and positive parameter e denotes the elasticity of substitution between time-empowered exclusive and inclusive skills in production of individual efﬁcient labor and reﬂects their imperfect substitutability. Skills are acquired in social networks according to a decreasing-returns-to-scale technology, S im ¼ Ljim ð1 þ NðI im Þ þ NðJ im =ð1 þ dÞÞÞ

(A.3)

S jm ¼ Ljjm ð1 þ NðI jm =ð1 þ dÞÞ þ NðJ jm ÞÞ

(A.4)

where j 2 ð0; 1Þ is the measure of decreasing returns to time spent in skill acquisition. Solving the individual problem (12) with the technologies (A.1)–(A.4), and the resource constraint (8), one can show that in the equilibrium, individuals divide their time between acquisition and utilization of skills according to the rule Lkm ¼ jZ km .

(A.5)

Thus, agent k allocates j units of her time to acquisition of skill m for each unit of time spent utilizing this skill in production. It also turns out

225

Ethnic Competition and Specialization

that the condition oðj þ 1Þ=j separates two classes of regimes. In particular, if exclusive and inclusive skills are good substitutes such that ðj þ 1Þ=j, individuals acquire only one type of skills. On the contrary, if oðj þ 1Þ=j, individuals choose to acquire both exclusive and inclusive skills, as there are strong complementarities between these skills. The optimal time allocation in this case is governed by the arbitrage condition: Zkx 1 þ NðI kx Þ þ NðJ kx =1 þ dÞÞ 1=ð1þjjÞ Lkx ¼ ¼ l k . (A.6) zk 1 þ NðI kn Þ þ NðJ kn =ð1 þ dÞÞ Z kn Lkn Denoting l k lk zk , conditions (A.5) and (A.6) give rise to equilibrium time allocations Z kx ¼ lk =ð1 þ lk Þð1 þ jÞ and Z kn ¼ 1=ð1 þ lk Þð1 þ jÞ for the time spent working and Lkx ¼ jlk =ð1 þ lk Þð1 þ jÞ and Lkn ¼ j=ð1 þ lk Þð1 þ jÞ for the time spent in skill acquisition. The relative investment of individuals in acquisition of exclusive and inclusive skills is fully determined by the spillover effects Nkm.19 Specifying the functional forms of these spillover effects to be N im ðI im ; J im ; dÞ ¼ I gim þ ðJ im =ðð1 þ dÞÞÞg

(A.7)

N jm ðI jm ; J jm ; dÞ ¼ ðI jm =ð1 þ dÞÞg þ J gjm ,

(A.8)

where the parameter g40 2 captures the returns to network size. The actual spillover effects depend on the equilibrium organization of skill g acquisition. In the DI equilibrium, the spillover effects are N DI ix ð:; :; :Þ ¼ I , g DI g DI g DI N in ð:; :; :Þ ¼ I þ ðJ=ð1 þ dÞÞ , N jx ð:; :; :Þ ¼ J , and N jn ð:; :; :Þ ¼ ðI=ð1 þ dÞÞg þ J g . This yields the minority and majority arbitrage conditions: 1=ð1þjjÞ 1 þ Ig lDI ¼ i 1 þ I g þ ðJ=ð1 þ dÞÞg and lDI j

¼

1 þ Jg 1 þ ðI=ð1 þ dÞÞg þ J g

ð1Þ=ð1þjjÞ ; respectively:

Recalling that IoJ and d40 it is straightforward to observe that 1 þ Ig 1 þ Jg o o1. 1 þ I g þ ðJ=ð1 þ dÞÞg 1 þ ðI=ð1 þ dÞÞg þ J g Given that under the DI equilibrium oðj þ 1Þ=j, from the arbitrage conditions, it follows that each individual spends more time in exclusive networks than in inclusive ones whenever o1. Furthermore, in such case, minority individuals spend relatively more time in exclusive social 19

In particular, it does not depend on wages. The reason is that individuals take wages as given, time has no other value but in skill acquisition, and skill acquisition does not involve any pecuniary exchange.

226

Martin Kahanec

networks than majority individuals. These results arise as the consequence of skill complementarity that forces individuals to compensate for their lower efﬁciency in exclusive networks by the longer times spent in these networks. Similarly, if 41, in the DI equilibrium, each individual spends more time in inclusive networks than in exclusive ones. Finally, if the technology of combining skills is Cobb–Douglas and ¼ 1, individuals spend equal shares of their time in exclusive and inclusive networks. To show ethnic specialization under DI equilibrium, using the skill acquisition technologies and the arbitrage conditions, we see that, g DI j g g sDI i ¼ ðl i Þ ð1 þ I Þ=ð1 þ I þ ðJ=ð1 þ dÞÞ Þ 1=ð1þjjÞ 1 þ Ig ¼ , 1 þ I g þ ðJ=ð1 þ dÞÞg

whereas g DI j g g sDI j ¼ ðl j Þ ð1 þ J Þ=ð1 þ ðI=ð1 þ dÞÞ þ J Þ 1=ð1þjjÞ 1 þ Jg ¼ . 1 þ ðI=ð1 þ dÞÞg þ J g DI By the same argument as above, it follows that sDI i =sj o1, because oð1 þ jÞ=j under DI equilibrium. In particular, even though minority agents under some circumstances spend more time in their exclusive networks in the DI equilibrium, they unambiguously acquire relatively less exclusive skills than majority individuals. It is easy to see that

DI zDI i si ¼

1 þ Ig 1 þ I g þ ðJ=ð1 þ dÞÞg

=ð1þjjÞ

and DI zDI j sj ¼

1 þ Jg 1 þ ðI=ð1 þ dÞÞg þ J g

=ð1þjjÞ .

Therefore, whether we deﬁne ethnic specialization to prevail if DI DI DI sDI i =sj a1 or zi si =zj sj a1 has no bearing on the result that the DI equilibrium exhibits specialization of ethnic groups.

References Aigner, D., Cain, G. (1977), Statistical theories of discrimination in labor markets. Industrial and Labor Relations Review 30 (2), 175–187. Akerlof, G.A. (1997), Social distance and social decisions. Econometrica 65 (5), 1005–1027.

Ethnic Competition and Specialization

227

Allen, B. (1982), Some stochastic processes of interdependent demand and technological diffusion of an innovation exhibiting externalities among adopters. International Economic Review 23 (3), 595–607. Altonji, J.G., Blank, R.M. (1999), Race and gender in the labor market. In: Ashenfelter, O., Card, D. (Eds.), Handbook of Labor Economics, vol. 3c. Elsevier, Amsterdam, pp. 3143–3259. Arrow, K.J. (1972a), Models of job discrimination. In: Pascal, A.H. (Ed.), Racial Discrimination in Economic Life. DC Heath, Lexington, MA, pp. 83–102. Arrow, K.J. (1972b), Some mathematical models of race discrimination in the labor market. In: Pascal, A.H. (Ed.), Racial Discrimination in Economic Life. DC Heath, Lexington, MA, pp. 187–204. Arrow, K.J. (1973), The theory of discrimination. In: Ashenfelter, O., Rees, A. (Eds.), Discrimination in the Labor Markets. Princeton University Press, Princeton, NJ, pp. 3–33. Bala, V., Goyal, S. (1998), Learning from neighbors. Review of Economic Studies 65, 595–621. Becker, G. (1956), The Economics of Discrimination. University of Chicago Press, Chicago. Blalock, H.M., Jr. (1956), Economic discrimination and Negro increase. American Sociological Review 21 (5), 584–588. Blalock, H.M., Jr. (1957), Per cent non-white and discrimination in the south. American Sociological Review 22 (6), 677–682. Borjas, G.J. (1983), The substitutability of black, Hispanic, and white labor. Economic Inquiry 21 (1), 93–106. Borjas, G.J. (1987), Immigrants, minorities, and labor market competition. Industrial and Labor Relations Review 40 (3), 382–392. Borjas, G.J. (2003), The labor demand curve is downward sloping: reexamining the impact of immigration on the labor market. The Quarterly Journal of Economics 118 (4), 1335–1374. Brown, D.L., Fuguitt, G.V. (1972), Percent nonwhite and racial disparity in nonmetropolitan cities of the south. Social Science Quarterly 53, 573–582. Buhai, S., van der Leij, M. (2006), A social network analysis of occupational segregation. Tinbergen Institute, Discussion Paper TI 2006-016/1. Chiswick, B.R., Lee, Y.L., Miller, P.W. (2003), Patterns of immigrant occupational attainment in a longitudinal survey. International Migration 41 (4), 47–69. Coate, S., Loury, G. (1993), Will afﬁrmative action policies eliminate negative stereotypes. American Economic Review 83 (5), 1220–1242. Coleman, J.S., Campbell, E., Hubson, C., McPartland, J., Mood, A., Weinfeld, F., York, R. (1966), Equality of Educational Opportunity. U.S. Government Printing Ofﬁce, Washington, DC. Conley, T., Udry, C. (2005), Learning about a new technology: pineapple in Ghana. Federal Reserve Bank of San Francisco Proceedings.

228

Martin Kahanec

Availabe at http://www.frbsf.org/economics/conferences/0511/7_ LearningNewTechnology.pdf Ellison, G., Fudenberg, D. (1993), Rules of thumb for social learning. Journal of Political Economy 101 (4), 612–643. Ellison, G., Fudenberg, D. (1995), Word-of-mouth communication and social learning. Quarterly Journal of Economics 110 (1), 83–97. Feick, L.F., Price, L.L. (1987), The market maven: a diffuser of marketplace information. Journal of Marketing 51 (1), 83–97. Foster, A.D., Rosenzweig, M.R. (1995), Learning by doing and learning from others: human capital and technical change in agriculture. Journal of Political Economy 103 (6), 1176–1209. Frisbie, P.W., Neidert, L. (1977), Inequality and the relative size of minority populations: a comparative analysis. The American Journal of Sociology 82 (5), 1007–1030. Gladwell, M. (2000), The Tipping Point: How Little things Can Make a Big Difference. Little, Brown, Boston. Glaeser, E.L., Laibson, D., Sacerdote, B. (2002), An economic approach to social capital. The Economic Journal 112 (483), 437–458. Glaeser, E.L., Sacerdote, B.I., Scheinkman, J.A. (2003), The social multiplier. Journal of the European Economic Association 1 (2–3), 345–353. Goyal, S. (2003), Learning in Networks: A Survey, in Group Formation in Economics: Networks, Clubs, and Coalitions, Eds. G. Demange and M. Wooders, Cambridge University Press. Grant, J.H., Hamermesh, D. (1981), Labor market competition among youths, white women, and others. The Review of Economics and Statistics 63 (3), 354–360. Grossman, J.B. (1982), The substitutability of natives and immigrants in production. The Review of Economics and Statistics 64 (4), 596–603. Heckman, J.J. (2000), Policies to foster human capital. Research in Economics 54 (1), 3–56. Heer, D.M. (1959), The sentiment of white supremacy and ecological study. The American Journal of Sociology 64 (6), 592–598. Hicks, J. (1970), Elasticity of substitution again: substitutes and complements. Oxford Economic Papers 22 (3), 289–296. Hirschman, C., Wong, M.G. (1984), Socioeconomic gains of Asian Americans, blacks, and Hispanics: 1960–1976. The American Journal of Sociology 90 (3), 584–607. Hofstede, G.H. (1980), Culture’s Consequences: International Differences in Work-Related Values. Sage, Beverly Hills, CA. Kahanec, M. (2006a), The substitutability of labor of selected ethnic groups in the US labor market. IZA Discussion Paper 1945. Institute for the Study of Labor (IZA), Bonn.

Ethnic Competition and Specialization

229

Kahanec, M. (2006b), Ethnic specialization and earnings inequality: why being a minority hurts but being a big minority hurts more. IZA Discussion Paper 2050. Institute for the Study of Labor (IZA), Bonn. Lazear, E.P. (1999), Culture and change. The Journal of Political Economy 107 (6), 95–126. Lucas, R.E., Jr. (1988), On the Mechanics of Economic Development. Journal of Monetary Economics 22 (1), 3–42. Lundberg, S., Startz, R. (1998), On the persistence of racial inequality. Journal of Labor Economics 16 (2), 292–323. Munshi, K. (2004), Social learning in a heterogeneous population: technology diffusion in the Indian green revolution. Journal of Development Economics 73 (1), 185–215. Phelps, E.S. (1972), The statistical theory of racism and sexism. American Economic Review 62 (4), 659–661. Poole, W.C., Jr. (1927), Distance in sociology. The American Journal of Sociology 33 (1), 99–104. Richman, B.D. (2006), How community institutions create economic advantage: Jewish diamond merchants in New York. Law and Social Inquiry 31 (2), 383–420. Tienda, M., Lii, D.T. (1987), Minority concentration and earnings inequality: Blacks, Hispanics, and Asians Compared. The American Journal of Sociology 93 (1), 141–165. Valente, T.W. (1995), Network Models of the Diffusion of Innovations. Hampton Press, Inc., Cresskill, NJ. Welch, F. (1967), Labor market discrimination: an interpretation of income differences in the rural south. Journal of Political Economy 75 (3), 225–240.

CHAPTER 10

Nationality Discrimination in the Labor Market: Theory and Test + B. Bodvarssona,b and John G. Sessionsb,c Orn a

Department of Economics, Department of Management, St. Cloud State University, St. Cloud, Minnesota, 56301-4498, USA E-mail address: [email protected] b Institute for the Study of Labor (IZA), D-53113 Bonn, Germany c Department of Economics, University of Bath, Bath BA2 7AY, UK E-mail address: [email protected]

Abstract When immigrants experience ‘‘nationality discrimination’’ in the labor market, ceteris paribus their earnings are lower than native-born workers because they were born abroad. The challenge to testing for nationality discrimination is that the native/immigrant earnings gap will very likely also be inﬂuenced by productivity differences driven by incomplete assimilation of immigrants, as well as the possibility of racial or gender discrimination. There is relatively little empirical literature, and virtually no theoretical literature, on this type of discrimination. In this study, a model of nationality discrimination where customer prejudice and native/ immigrant productivity differences jointly inﬂuence the earnings gap is presented. We derive an extension of Becker’s market discrimination coefﬁcient (MDC), applied to the case of nationality discrimination when there are productivity differences. A number of novel implications are obtained. We ﬁnd, for example, that the MDC depends upon relative immigrant productivity and relative immigrant labor supply. We test the model on data for hitters and pitchers in Major League Baseball, an industry with a history of immigration, potential for customer discrimination, and clean detailed microdata on worker productivities and race. Ordinary least squares (OLS) and decomposition methods are used to estimate the extent of discrimination. We ﬁnd no compelling evidence of discrimination in the hitter group, but evidence of ceteris paribus underpayment of immigrant pitchers. While our test case is for a particular industry, our theoretical model, empirical speciﬁcations, and general research design are quite generalizable to many other labor markets.

Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008016

r 2010 by Emerald Group Publishing Limited. All rights reserved

232

+ B. Bodvarsson and John G. Sessions Orn

Keywords: Nationality discrimination, immigrants, skills transferrability, baseball Jel classiﬁcation: J7

1. Introduction It is well established in the literature on immigrant earnings that immigrants, especially those who have been in the host country only for a short time, are paid differently than their native-born peers. In most countries, for example, immigrants have lower wage rates compared to natives. The literature provides two basic explanations for native/immigrant earnings differences. First, there could be productivity differences resulting from immigrants having (a) lower ‘‘standard’’ qualiﬁcations – years of schooling, years of experience, etc. and/or (b) lower levels of host-country human capital, that is, lack of full assimilation. Second, there could be discrimination against immigrants, what is often called ‘‘nationality discrimination.’’1 If immigrants suffer nationality discrimination, they are ceteris paribus paid less than their native-born peers. One reason is prejudice against immigrants by host-country residents. The challenge to estimating nationality discrimination is to separate the effects of productivity differences from the effects of nationality discrimination on the nativeimmigrant earnings gap. As Nielsen et al. (2004) point out, ‘‘y when analyzing immigrant wage gaps in the presence of potential discrimination, it is important to disentangle the assimilation effect from a potential discrimination effect due to ethnicity y’’ (p. 859; our italics for emphasis). The literature has not yet resolved this ‘‘disentangling’’ problem for several reasons. First, the problem has not been analyzed theoretically. The literature on assimilation, beginning with Chiswick (1978) and Borjas (1985, 1987), has been very helpful in clarifying how incomplete assimilation accounts for productivity differences between natives and immigrants. However, this literature has generally not analyzed how assimilation inﬂuences the earnings gap when immigrants also experience nationality discrimination. We lack a theoretical understanding of how productivity differences and nationality discrimination jointly inﬂuence native/immigrant earnings differences. Second, researchers have often chosen test cases where data constraints make the separation of an assimilation effect from a discrimination effect difﬁcult. A common problem is that data on the quality of immigrant human capital is often 1

A third explanation is that there could be native/immigrant differences in labor supply. For example, immigrants may have lower wage elasticities of supply and/or lower reservation wages. The literature has generally not pursued this explanation, though.

Nationality Discrimination in the Labor Market

233

difﬁcult to obtain. Another problem, emphasized by Bodvarsson and Fuess (2008), is that because many U.S. immigrants are nonwhite and because many studies do not control for race, available estimates of nationality discrimination could be biased because of missing controls for race. Consequently, we still know very little about the contribution of nationality discrimination to the native/immigrant wage gap. We contend that to properly address the ‘‘disentangling’’ problem pointed out by Nielsen et al. (2004), two hurdles must be overcome. First, a theoretical model of production where natives and immigrants are distinct inputs and where immigrants experience discrimination must be articulated. This model would be in contrast to the standard model of discrimination where majority and minority workers are perfect substitutes. As the assimilation literature would strongly argue, a presumption of perfect substitution between natives and immigrants is highly inappropriate because of imperfect transferability of human capital across borders. Second, proper empirical veriﬁcation of a nationality discrimination model requires a test case where there are (i) highly accurate and detailed microdata on native and immigrant worker productivities and (ii) one can adequately control for the potentially confounding effects of race (or gender). In this chapter, we strive to overcome both these hurdles. We present a theory of tastes-based nationality discrimination using a model of production where natives and immigrants are imperfect substitutes. We derive a number of novel implications, which are then tested on a data set from an industry that is highly convenient for testing hypotheses about native/immigrant differences in pay – U.S. Major League Baseball (MLB). MLB is an industry with a long history of immigrant employment and where detailed and accurate data on player and ﬁrm performance are available. While this test case is selected because it is a convenient natural experiment for studying nationality discrimination, we wish to emphasize that our model and test are generalizable to many other labor markets. The remainder of this chapter is organized as follows. In the next subsection, we discuss the concept of nationality discrimination in more detail and then do an assessment of what the literature has to say about it. Our theoretical model is presented in Section 2. In Section 3, we present evidence from tests of some of our model’s implications. Section 4 offers concluding remarks.

1.1. Nationality discrimination: Meaning and previous literature There is a voluminous literature on racial and gender discrimination in the labor market, which dates back to the pioneering works of Gary Becker (1971) and Kenneth Arrow (1973).2 It is well known that

234

+ B. Bodvarsson and John G. Sessions Orn

labor market discrimination can take place with respect to other personal attributes such as age, religious afﬁliation, sexual orientation, weight, accent, or speech patterns. Another attribute that could induce discriminatory treatment is a worker’s nativity status. Nationality discrimination occurs when, ceteris paribus, immigrants are paid less because they were born outside of the host country. This type of discrimination could be driven by prejudice (taste discrimination), imperfect information (statistical discrimination), or institutional factors.3 In this study, we will focus on nationality discrimination due to prejudice. There could be many reasons for prejudice against immigrants. Anecdotal information about nationality discrimination has been prevalent for many years. There are age-old jokes about U.S. immigrants from Ireland, Italy, and other countries regarding their mannerisms, attitudes, ways of life, etc. These sorts of jokes may be reﬂective of nationality discrimination. Native-born workers may harbor resentment toward immigrants because they perceive that immigrants take jobs from them. Native-born consumers may harbor biases against particular countries and value the goods made by workers from those countries less.4 Prejudiced employers may have biases against hiring workers from particular countries. Nationality discrimination could occur in tandem with racial or gender discrimination. For example, if nonwhite U.S. immigrants are paid differently from white persons born in the United States, the differential treatment could reﬂect both racial discrimination and nationality discrimination. Furthermore, there could be an interaction between race and nativity status on pay. The marginal effect of nativity status on pay could depend upon the level of racial discrimination.5 Compared to the literature on racial and gender discrimination, nationality discrimination has received very little attention. Two widely cited expository surveys of the discrimination literature, Cain (1986) and Altonji and Blank (1999), make no mention of it.6 The Appendix to this

2

For expository surveys of this literature, see Cain (1986) and Altonji and Blank (1999). For a thorough comparison and contrasting of different theories of labor market discrimination, see Cain (1986). 4 For example, the terrorist attacks of September 11, 2001, may have resulted in consumer discrimination against employers and their employees born in the Middle East. 5 Nationality discrimination needs to be distinguished from ‘‘ethnic’’ discrimination. Ethnic discrimination is discrimination against persons of a particular ethnicity, or heritage, such as being Korean, Mexican, Italian, or having parents, grandparents, or ancestors from another country. One need not be foreign-born to experience ethnic discrimination. Nationality discrimination means that being native-born versus foreign-born makes a difference in how one is paid, all other things being equal. 6 Cain (1986) mentions ‘‘ethnic discrimination’’ only once (p. 1) and never mentions discrimination with respect to place of birth, while Altonji and Blank (1999) don’t mention either type of discrimination. 3

Nationality Discrimination in the Labor Market

235

paper lists 23 studies on native-immigrant earnings differences that relate to nationality discrimination. These studies have three characteristics. First, only half make the identiﬁcation of nationality discrimination the focus of inquiry. Second, nearly all of these studies lack any original theoretical model. We found only two studies, Bucci and Tenorio (1997) and Hayfron (2002), which articulate a theoretical model of native/ immigrant wage differentials based on taste discrimination.7 Those two models are very general and do not produce any novel implications that are carried over to an empirical speciﬁcation. Third, most of the 23 studies listed in the Appendix do not address the potentially confounding effects of race or gender in the estimation of nationality discrimination. The ﬁrst analysis of earnings differences between foreign- and nativeborn workers that indicates possible nationality discrimination is Chiswick’s (1978) pioneering study of U.S. immigrant assimilation. One of Chiswick’s results, overshadowed by his evidence on immigrant assimilation, was a ceteris paribus earnings premium for immigrant males, as well as native/immigrant differences in rates of return to qualiﬁcations.8 While Chiswick did not say that nationality discrimination contributes to native/immigrant differences in earnings,9 he hinted at the possibility when he wrote (p. 908): For the earnings of the foreign born to exceed the native born eventually suggests that the greater ability, work motivation, or investments in training of the foreign born more than offset whatever earnings disadvantages persist from discrimination against them or from their initially having less knowledge and skills relevant in U.S. labor markets (our italics for emphasis). Tandon (1978) reported evidence of structural differences in earnings between native- and foreign-born male residents of Toronto, but made no mention of discrimination as a factor. Similar types of results were obtained by Fujii and Mak (1983), who studied differences in earnings among six different ethnic groups in Hawaii. Neither of these studies

7

We should also mention a theoretical study by Mu¨ller (2003), who uses a dynamic efﬁciency wage model to derive a ceteris paribus wage gap between natives and immigrants. Mu¨ller assumes that migrants differ from natives only because migrants have a positive probability of return migration. Firms anticipate that migrants will have a greater chance of shirking, hence they are paid less. 8 For example, Chiswick found that the rate of return to education for an immigrant male was lower than for a native-born male. 9 Chiswick said in a footnote (p. 899) that ‘‘The [native/immigrant] earnings gap would not close if a relevant knowledge deﬁciency persisted or if there were discrimination against the foreign born in wages, employment, union membership, or occupational licensing. On the other hand, in some jobs there may be discrimination in favor of the foreign born (e.g. the French chef)’’ (our italics for emphasis).

236

+ B. Bodvarsson and John G. Sessions Orn

control for race, though.10 In a study of male U.K. immigrants, Chiswick (1980) found some evidence of a ceteris paribus earnings premium for immigrants. His results indicate that the size of the foreign/native earnings gap depends upon whether race is included as a control. Chiswick did not mention discrimination as a factor accounting for these results, though. Meng (1987) estimated Chiswick’s model on Canadian data and found some evidence of a ceteris paribus earnings disadvantage for Canadian immigrants. However, like Chiswick, he did not draw any conclusions about whether this earnings gap indicated the presence of nationality discrimination. The ﬁrst study to focus on identifying nationality discrimination is Haig’s (1980) analysis of Australian panel data. While Haig’s study has substantial data limitations that put his estimates at high risk of omitted variables bias,11 his study has two important features. First, it is the ﬁrst application of the Oaxaca (1973) and Blinder (1973) decomposition method to the measurement of nationality discrimination. Second, Haig recognized that racial and nationality discrimination may be correlated, hence the importance of controlling for both race and national origin when testing for nationality discrimination. Studying nearly 15,000 Australian males, he found that while immigrants earn about 6 percent more and have larger human capital endowments than natives, had they received the same returns to their human capital endowments as natives immigrants would have earned 9 percent more. Haig took this as evidence of a ceteris paribus earnings differential attributable to nationality discrimination of approximately 3 percent. Reimers (1984) also recognized the importance of controlling for both race and nativity status in a study of native/immigrant earnings differences. She analyzed data for Hispanic men in the United States, ﬁnding no evidence of ceteris paribus earnings differences between foreignborn Hispanics and native non-Hispanics, but some evidence indicating differences in rates of return to education and experience between the two groups. She made no mention of discrimination as a possible reason for these differences. The ﬁrst U.S. study to focus on identifying nationality discrimination is Gabriel and Schmitz’s (1987) investigation of native and immigrant male earnings. Using data from the 1980 Census of Population and applying the decomposition method, they estimated two models that differ with respect to how the foreign-born earnings equation is speciﬁed. In their ‘‘basic’’ model, Gabriel and Schmitz have no controls for earnings assimilation, 10

It should be noted that while they lacked a race dummy in their empirical speciﬁcations, Fujii and Mak did distinguish Caucasians from members of ﬁve different non-Caucasian groups. 11 Haig measured experience by age, where age was only measured in intervals. Furthermore, his data set contains very little information on personal characteristics.

Nationality Discrimination in the Labor Market

237

whereas in their ‘‘reﬁned’’ model the foreign-born equation did include such controls. They found for the basic model that a relatively small proportion (approximately 13 percent) of the native/immigrant mean earnings gap was attributable to discrimination against foreign-born workers. However, in their reﬁned model, Gabriel and Schmitz observed an immigrant earnings premium. These conﬂicting results were taken as ‘‘y rather inconclusive evidence of earnings discrimination against foreign-born workers y’’ (p. 99). Gabriel and Schmitz (1987) did set a landmark in the literature because a larger proportion of subsequent studies made estimating nationality discrimination the focus, and the decomposition technique became the dominant empirical methodology for estimating nationality discrimination. Tran-Nam and Neville (1988), Daneshvary and Weber (1991) and Daneshvary (1993) used the decomposition method to provide evidence of discrimination against Australian immigrants (Tran-Nam and Neville) and U.S. immigrants. Beach and Worswick (1993), Shamsuddin (1998), and Hayfron (2002) estimated earnings equations that control for both gender and nationality. These authors argued that female immigrants can experience a ‘‘double negative effect’’ on earnings from gender discrimination and nationality discrimination. They provided evidence of this double negative effect in Canada (Beach and Worswick and Shamsuddin) and Norway (Hayfron). Kee (1995) studied nationality discrimination against Dutch immigrants, ﬁnding evidence of discrimination against members of several nationality groups. Bucci and Tenorio (1997) used U.S. Census data to provide evidence of nepotism towards native-born workers. In a study of American and foreign-born scientists working in the United States, Goyette and Xie (1999) reported that female immigrants are less likely to get hired and promoted than their immigrant male and native-born female counterparts They suggested that this may be a result of the ‘‘immigration path’’ (p. 407) taken by many female scientists and engineers, as wives of immigrant men. In a study of the Danish labor market, Nielsen et al. (2004) extended the decomposition technique to allow for the disentangling of the three sources of native-immigrant earnings differences discussed earlier. They found evidence of discrimination against female immigrants. Three studies have tested for nationality discrimination in professional sports. Wilson and Ying (2003) tested for the effects of team nationality composition on fan attendance in the world’s ﬁve largest football leagues. They found some evidence suggesting that employer discrimination is responsible for underrepresentation of various player nationalities. Pedace (2008) studied English soccer and found evidence of employer nepotism in favor of some South American players. Bodvarsson and Fuess (2008) studied MLB and found that foreign-born hitters subject to the Reserve Clause (players for whom the team owner has monopsony power) are disadvantaged in the bargaining process relative to American-born

+ B. Bodvarsson and John G. Sessions Orn

238

players, all other things equal. Unlike Pedace, Bodvarsson and Fuess control for the potentially confounding effects of race. Our assessment of the aforementioned literature on nationality discrimination leads us to conclude the following. Nationality discrimination appears across a majority of studies to be an important contributor to immigrant earnings, especially for certain countries, nationality groups, and occupations. However, the literature lacks a theory of nationality discrimination that is capable of producing testable implications regarding the joint inﬂuences of native/immigrant productivity differences and discrimination. Furthermore, many of the data sets used in the previously discussed studies lack the breadth and depth of microdata needed to produce accurate estimates of nationality discrimination. The goal of our research below is to resolve these deﬁciencies.

2. A theory of nationality discrimination 2.1. The problem setting Suppose production is done using three inputs – units of native-born labor, units of foreign-born labor, and capital. Capital is assumed to be ﬁxed. Native- and foreign-born workers are assumed to perform the same job assignment, but are imperfect substitutes. The imperfect substitutability arises from immigrant workers having acquired their skill sets in their home countries, where educational systems and labor markets are different. Differences in schooling, on-the-job training, culture and traditions, etc. all contribute to differences in human capital endowments between natives and immigrants. To quote Ottaviano and Peri (2005), ‘‘Since foreign-born workers receive [at least part] of their education abroad, they are likely to retain different abilities pertaining to language, quantitative skills, and so on. Therefore they should be differentiated enough to be treated as imperfect substitutes for U.S.-born workers, even within the same education and experience group’’ (our italics for emphasis).12 Assume that technology is characterized by the generalized Leontief production function:13 Q¼

XX j

gij ½Li ðDLj Þ1=2

ði; j ¼ N; IÞ

i

h i ¼ gNN LN þ D1=2 gII LI þ 2gNI ðLN LI Þ1=2 , 12 13

See Ottaviano and Peri (2005, pp. 7–8). See Diewert (1971).

ð1Þ

Nationality Discrimination in the Labor Market

239

where Q is output, LN is the quantity of native-born labor, LI is the quantity of immigrant labor, and gij is a technology coefﬁcient. Using an approach similar to Kahn (1991), we include the parameter D, 0oDo1, above as a measure of the strength of customer prejudice against immigrant workers. Customer prejudice may be viewed as a situation where customers discount the marginal revenue product (MRP) of immigrant labor. The lower (higher) the D, the more (less) intense the prejudice and the lower (higher) is the immigrant MRP.14 If D equals 1, there is no prejudice. While a more traditional approach would be to think of customer discrimination as implying a price discount on the output of immigrant workers, the approach above is equivalent. The parameter D reﬂects the idea that immigrant labor input is valued less when customers are prejudiced. Note also that the above production function is constant returns to scale and imposes restrictions on each technology coefﬁcient such that gij ¼ gji. The sign of each technology coefﬁcient indicates whether inputs i and j are substitutes (gijo0) or complements (gijW0). We will assume that the product and labor markets are perfectly competitive and that product price is normalized at unity. Assume that rN and rI are the market wages of native- and foreign-born workers, respectively. The ﬁrm’s proﬁt function is thus XX gij ½Li ðDLj Þ1=2 rN LN rI LI ði; j ¼ N; IÞ. (2) p¼ j

i

If ﬁrms maximize proﬁts and face constant input prices, the labor market will establish the following system of labor demand functions: 1=2 1 LI (3) rN ¼ gNN þ gNI D1=2 LN 2 1=2 1 1=2 LN rI ¼ gII D þ gNI D LI 2

(4)

While (3) and (4) are not reduced form expressions, they provide some very useful implications. The wage paid to workers comprising a particular group depends upon four factors: (i) the productivities of workers in that group; (ii) the strength of customer prejudice against immigrant workers; (iii) the degrees of substitutability or complementarity between the two labor groups; and (iv) the relative supplies of labor in each group. For example, the immigrant wage (in (4)) depends upon immigrant Note that MRP of immigrant labor is @Q=@LI ¼ D1=2 ½gII þ gNI ðLN =LI Þ1=2 . Under the assumption that employers only utilize that quantity of immigrant labor for which MRP is positive, requiring then that gII þ gNI ðLN =LI Þ1=2 40, an increase in prejudice will unambiguously reduce immigrant MRP regardless of whether the two groups are substitutes or complements.

14

240

+ B. Bodvarsson and John G. Sessions Orn

productivity (reﬂected by gII),15 prejudice (D), the degrees of substitutability or complementarity between natives and immigrants (gNI), and the relative supply of native workers (LN/LI). Equations (3) and (4) have immediate implications for how a change in the intensity of prejudice affects each group’s wage. An increase in prejudice will always reduce the immigrant wage, but the effect on the native wage will depend upon whether the two groups are substitutes or complements. According to (3), if natives and immigrants are complements, an increase in prejudice will reduce the native wage. The reason is that increased prejudice has the effect of reducing the beneﬁts from complementarity. Immigrant MRP falls, leading to less employment of immigrants, which in turn reduces native MRP and less employment of the latter group. When natives and immigrants are substitutes, the lower immigrant MRP induces employers to substitute natives for immigrants, stimulating native employment and wages. One particularly important insight from (3) and (4) is that the wage paid to workers of a particular group depends upon the amount of native/ immigrant integration within and across groups. For example, according to (3), the wage paid to a native-born worker is affected by the number of immigrants per native. Furthermore, how natives are affected by an increase or a decrease in the number of immigrants depends upon both the technology coefﬁcients and the intensity of prejudice against immigrants. If natives and immigrants are substitutes (gNIo0), then if more immigrants are hired the wage paid to natives will fall, all other things equal. However, the drop in that wage will be smaller (larger) the greater (lesser) is the degree of customer prejudice against immigrants.16 Thus, if natives and immigrants are substitutes, prejudice attenuates the adverse effects experienced by natives when more immigrants are hired and the degree of attenuation rises with the degree of prejudice. In contrast, suppose that natives are complementary to immigrants (gNIW0). Then an increase in the supply of immigrants will raise the wage paid to natives, all other things equal. However, the increase in that wage will be smaller (larger) the greater (lesser) is the degree of prejudice.17 Natives beneﬁt from having more immigrants, but prejudice effectively ‘‘taxes’’ that beneﬁt.

15

Note that gII is not precisely the marginal productivity of immigrants, but is correlated with it. If gII rises (falls), the marginal productivity curve will shift up (down). For example, an increase in gII could result from a technological advance, an increase in the average human capital endowment of each worker, or some other exogenous change. 16 When natives and immigrants are substitutes, @2 rN =ð@ðLI =LN Þ@DÞ40; o 0 meaning that as customer prejudice rises, the drop in the wage paid to natives resulting from greater employment of immigrants will be less severe. 17 When natives and immigrants are complements @2 rN =ð@ðLI =LN Þ@DÞ40, meaning that as customer prejudice falls, the gain in the wage paid to natives resulting from greater employment of immigrants will be larger.

Nationality Discrimination in the Labor Market

241

There are similar implications for immigrant wages. If natives and immigrants are substitutes, then greater employment of natives reduces the immigrant wage, but that wage falls less the greater is the amount of prejudice. Immigrant wages have less distance to fall in an environment of greater prejudice when relative employment of natives rises. In contrast, if the two group are complements, then greater hiring of natives will raise the wage of immigrants, but the wage rises less the greater is the prejudice. This is an indirect adverse effect of prejudice on immigrants: Immigrants are harmed directly because customers value their output less, and indirectly because prejudice reduces the beneﬁts enjoyed by immigrants from having a complementary relationship in production with natives. The next step in the analysis is to extend Becker’s (1971) market discrimination coefﬁcient (MDC) concept to the case of nationality discrimination when there is imperfect substitutability between the majority group and the minority group. The MDC in this case measures the percentage earnings premium paid to native workers due to being nativeborn. Applying Becker’s (1971, p. 17) general version of the MDC to the case of nationality discrimination, the MDC is:18 MDCN I ¼

rN ðDo1Þ rN ðD ¼ 1Þ rI ðDo1Þ rI ðD ¼ 1Þ

(5)

The ﬁrst term on the right-hand side of (5) is the native/immigrant wage ratio when there is prejudice, whereas the second term is the ratio in the absence of prejudice. The difference between the two ratios measures the ceteris paribus (adjusted for differences in productivity) nationality pay gap. Using expressions (3) and (4), it follows that MDCN I ¼

gNN þ ð1=2ÞgNI D1=2 ðLI =LN Þ1=2 gII D þ ð1=2ÞgNI D

1=2

1=2

ðLN =LI Þ

gNN þ ð1=2ÞgNI ðLI =LN Þ1=2

. gII þ ð1=2ÞgNI ðLN =LI Þ1=2 (6)

According to (6), discrimination depends upon the strength of customer prejudice, native/immigrant productivity differences, the degree of substitutability or complementarity between natives and immigrants, and the relative supplies of native and immigrant labor. Equation (6) provides some novel implications. We begin with a basic, intuitive one, though. (i) Nationality discrimination is larger the greater is the degree of customer prejudice. 18

This expression is identical to Becker’s general expression for the MDC, which he treats as the economy-wide wage gap when there is employment discrimination.

+ B. Bodvarsson and John G. Sessions Orn

242

PROOF. Differentiating (6) with respect to the customer prejudice parameter (D), we obtain @MDCN ð1=4ÞgNI D1=4 ðLI =LN Þ1=2 I ¼ @D ½gII D þ ð1=2ÞgNI D1=2 ðLN =LI Þ1=2 2

ðgNN þ ð1=2ÞgNI D1=2 ðLI =LN Þ1=2 ÞðgII þ ð1=4ÞgNI D1=4 ðLN =LI Þ1=2 Þ gII D þ ð1=2ÞgNI D1=2 ðLN =LI Þ1=2

.

(7)

If the two groups of labor are substitutes (gNIo0), then the ﬁrst ratio to the right of the equal sign in (7) is negative. Since the second ratio is positive, (7) is guaranteed to be negative, meaning that a decrease in D (strengthening of customers’ distaste for immigrants) raises the amount of market discrimination (ð@MDC N I =@DÞo0). If the two groups are complements (gNIW0), then (7) will also be negative. To see this, manipulation of (7) leads to the ﬁnding that 1=2 1 LI if ¼ gNI D1=4 4 LN 1=2 ! 1 1=4 LN . 4ðoÞ rN rI gII þ gNI D LI 4

@MDCN I 4ðoÞ 0 @D

(8)

Assuming that the product of the wages exceeds 1 and since gIIW0, the expression ð1=4ÞgNI D1=4 ðLI =LN Þ1=2 will be less than the expression to the right of it, hence @MDCN I =@Do0. The above prediction provides the foundation for another prediction, which is less intuitive, not implied by the standard discrimination model, and has very important empirical implications: (ii) The marginal effect of increased prejudice on nationality discrimination is less for higher-productivity immigrants. PROOF. The productivity of immigrant labor is reﬂected in the intercept term (gII). If one differentiates (6) with respect to this intercept term, the following expression is obtained: @2 MDCN D½ð1=4ÞgNI D1=4 ðLI =LN Þ1=2 I ¼ @D@gII ½gII D þ ð1=2ÞgNI D1=2 ðLN =LI Þ1=2 3

ðgNN þ ð1=2ÞgNI D1=2 ðLI =LN Þ1=2 ÞðgII þ ð1=4ÞgNI D1=4 ðLN =LI Þ1=2 Þ

D½ðgNN þ ð1=2ÞgNI D1=2 ðLI =LN Þ1=2 ÞðgII þ ð1=4ÞgNI D1=4 ðLN =LI Þ1=2 Þ

gII D þ ð1=2ÞgNI D1=2 ðLN =LI Þ1=2 ½gII D þ ð1=2ÞgNI D1=2 ðLN =LI Þ1=2 2

:

(9)

Nationality Discrimination in the Labor Market

243

Using the guidelines for signing (8), expression (9) will be negative, meaning that the marginal effect of prejudice on the MDC declines as immigrant workers become more productive. What this means is that should there be an intensiﬁcation of customer prejudice against immigrants, higher-productivity immigrants will experience less of an increase in discrimination than lower-productivity immigrants. There are several important implications here. First, race and productivity interact, something not implied by the standard discrimination model. Second, in an empirical speciﬁcation where the goal is to estimate the extent of nationality discrimination, one must include an interaction term between race and productivity to avoid omitted variables bias. There are two other predictions that relate to worker productivity: (iii) If immigrant workers become more productive, then pay discrimination against them falls. PROOF. Differentiating (6) with respect to gII, we obtain @MDC D½gNN þ ð1=2ÞgNI ðLI =LN Þ1=2 ½gNN þ ð1=2ÞgNI ðLI =LN Þ1=2 ¼ þ 1=2 2 @gII ½gII D þ ð1=2ÞgNI D1=2 ðLN =LI Þ ½gII þ ð1=2ÞgNI ðLN =LI Þ1=2 2 (10) which is o; (W) 0 as gNN þð1=2ÞgNI ðLI =LN Þ1=2

½gII þð1=2ÞgNI ðLN =LI Þ1=2 2 oð4Þ DgNN þgNI D3=2 ðLI =LN Þ1=2 ½DgII þð1=2ÞgNI D1=2 ðLN =LI Þ1=2 2

(11)

Note that the ratio on the left-hand side of (11) is less than 1, while the expression on the right-hand side exceeds 1, guaranteeing that (11) is always negative. Regardless of the signs and magnitudes of the technology coefﬁcients and the relative supplies of each labor type, nationality discrimination falls whenever immigrant labor improves its productivity. For example, a technological advance that makes immigrants more efﬁcient results in less discrimination against them. In terms of the two wage ratios in (6), an increase in immigrant productivity causes the relative native wage with prejudice to fall and the wage without prejudice to fall as well, but the former falls proportionately more than the latter and there will be a net reduction in the MDC: (iv) If native workers become more productive, then discrimination against immigrants rises. An increase in native worker productivity is manifested by an increase in the intercept of the native marginal product function (gNN). Differentiating (6) with respect to this parameter, we obtain @MDCN 1 1 I ¼ 40 1=2 1=2 @gNN gII D þ gNI D ðLN =LI Þ gII þ gNI ðLN =LI Þ1=2

(12)

+ B. Bodvarsson and John G. Sessions Orn

244

The numerator in the ﬁrst ratio on the right-hand side of (12) is less than the numerator in the second ratio, guaranteeing that (12) is always positive. When native workers experience increases in productivity, their relative wage when there is prejudice against immigrants rises proportionately more than the wage in the absence of prejudice. The productivity increase has the unintended consequence of exacerbating the amount of discrimination directed toward immigrants. The preceding two comparative static results have an important implication: When immigrants and natives are not perfect substitutes, the amount of market discrimination against immigrants depends upon relative productivity. Immigrants can help themselves overcome discrimination by boosting their productivities, for example, through additional human capital investments, improved health and motivation, etc. There is also a policy implication: Public money used to train and educate immigrants in the host country could lead to reduced discrimination. However, there is an unintended consequence of improved productivity of native-born workers: If natives invest relatively more in their human capital endowments than immigrants, this will lead to more discrimination. (v) An increase in the relative supply of immigrants increases the amount of nationality discrimination. PROOF. An important property of the demand functions that are generated by the generalized Leontief production function is that equilibrium factor prices depend upon relative factor supply. What does an increase in relative immigrant supply do to the MDC? Differentiating (6) with respect to relative immigrant supply, we obtain @MDCN ð1=4ÞgNI D1=2 ðLI =LN Þ1=2 ð1=4ÞgNI ðLI =LN Þ1=2 I ¼ . @ðLI =LN Þ gII D þ ð1=2ÞgNI D1=2 ðLN =LI Þ1=2 gII þ ð1=2ÞgNI ðLN =LI Þ1=2 (13) Assume that natives and immigrants are substitutes. Then, (13) will be negative (positive) if the ﬁrst ratio to the right of the equal sign is greater in absolute value than the second ratio, that is @MDCN I oð4Þ 0 if @ðLI =LN Þ oð4Þ

1 gII D

1=2

þ ð1=2ÞgNI ðLN =LI Þ1=2

1

(14)

gII þ ð1=2ÞgNI ðLN =LI Þ1=2

Since Do1, the expression 1=ðgII D1=2 þ ð1=2ÞgNI ðLN =LI Þ1=2 Þ is the larger one, conﬁrming a negative sign for (14). The implication is that if residents of the host country harbor prejudice toward immigrants and there is an inﬂux of new immigrants, there will be an increase in the amount of nationality discrimination. Conversely, the implication is that restrictive

Nationality Discrimination in the Labor Market

245

immigration policies will help to eradicate the extent of nationality discrimination. 3. A test case: Major League Baseball 3.1. Description of the test case In this section, we test some implications of the above model. We chose an industry where (a) there are very accurate data on salaries and productivities for individual workers; (b) there is a history of immigration and native/ immigrant integration in the ﬁrm’s work force; (c) worker race and gender can be observed; and (d) there is potential for customer discrimination. One industry satisfying all these criteria is MLB) in the USA.19 In MLB, each team requires two distinctly complementary types of player skill – hitting (an offensive skill) and pitching (a defensive skill) – in the production of baseball entertainment. The industry has a long history of both immigration and racial integration. For many years, MLB has recruited players from Latin American countries, Canada, Australia, Japan, and other countries. The ideal way to measure a Major League player’s MRP is by his contribution to the team’s ticket, broadcasting, and merchandise revenues. Because of the team production nature of baseball, however, it is impossible to empirically disentangle one player’s revenue contribution from another. We thus proxy MRP by the player’s years of MLB experience, tenure with his current team, and various career statistics (computed on a game-by-game basis since the beginning of the player’s Major League career) that proxy his ability and skills. The career statistics we use to measure a hitter’s productivity include At Bats, Stolen Bases, Bases on Balls, Total Bases, Slugging Average, and Batting Average.20 19

Discrimination in the professional sports labor market has received considerable attention among labor economists because of the abundant statistical evidence on a player’s personal attributes, compensation, and productivity. Most studies in this area have focused on racial discrimination with respect to pay, hiring, retention, and positional segregation. For an examination of the research, see Kahn’s (2000) expository survey. 20 A player has an At Bats every time he comes to bat, except in certain circumstances, for example, if he is awarded ﬁrst base due to interference or obstruction or the inning ends while he is still At Bats. A hitter is assigned a Stolen Bases (also called a ‘‘steal’’) when he reaches an extra base on a hit from another player. For example, suppose that hitter A is at ﬁrst base when hitter B hits the ball. Hitter B reaches ﬁrst base (he would be assigned a ‘‘single’’), but hitter A reaches third base. Hitter A would be assigned a Stolen Bases because he reached an extra base. A Bases on Balls (also called a ‘‘walk’’) is assigned when the batter receives four pitches which the umpire determines is a ‘‘ball.’’ A ball is any pitch at which the batter does not swing and is out of the ‘‘strike zone’’ (which means it would not qualify to be a strike). When the hitter is assigned a Bases on Balls, he is entitled to walk to ﬁrst base. Total Bases are the number of bases a player has gained through hitting. It is the sum of his hits weighted by 1 for a single, 2 for a double (if he gets to second base as a result of his hit), 3 for a triple (if he gets to third base), and 4 for a home run. A hitter’s Batting Average is the ratio of hits to At Bats; this measures the hitter’s success rate. Slugging Average, a related measure, reﬂects hitting power, which is Total Bases divided by At Bats.

246

+ B. Bodvarsson and John G. Sessions Orn

We distinguish between hitters that are ‘‘designated hitters’’ from those who are not. A designated hitter is a player who is chosen at the start of the game to bat in lieu of the pitcher in the lineup. We also distinguish, using dummies, between hitters that serve other types of positions. These include whether the hitter served as an inﬁelder or a catcher.21 We measure a pitcher’s productivity by use of the following career statistics: Wins, Losses, Games Started, Complete Games, Saves, Homeruns, Walks, Strikeouts, Innings Pitched, Earned Run Average (ERA), and Strikeout Rate.22

3.2. Empirical analysis Our empirical analysis is set out in Tables 1–6. Tables 1 and 2 show descriptive statistics for hitters and pitchers, respectively. Our full sample comprises 1,093 hitters (901 native, 192 immigrant; 549 white and 544 nonwhite) and 1,203 pitchers (1,031 native, 172 immigrant; 941 white and 262 nonwhite). Salary, experience, performance, and position data were drawn from the Lahman Baseball Database (see: www.baseball1.com) over four seasons – 1992, 1993, 1997, and 1998. The salary data do not include 21

An inﬁelder is a defensive player who plays on the ‘inﬁeld,’ the dirt portion of a baseball diamond between ﬁrst and third bases. The speciﬁc inﬁelder positions are ﬁrst baseman, second baseman, shortstop (which is between second and third bases) and third baseman. In contrast, an ‘outﬁelder’ plays farthest from the batter and his primary role is to catch long ﬂy balls. Outﬁelder positions include left ﬁelder, center ﬁelder and right ﬁelder. The catcher crouches behind home plate and receives the ball from the pitcher. Because the catcher can see the whole ﬁeld, he is best positioned to lead and direct his fellow players in play. He typically calls the pitches by means of hand signals and hence requires awareness of both the pitcher’s mechanics and the strengths and weaknesses of the batter. 22 A pitcher is assigned a Wins or a Losses depending on whether he was the pitcher of record when the decisive run was scored. One is the pitcher of record if one is the pitcher at the point when the player who scores the decisive run is allowed to reach a base. Games Started is the number of times the pitcher was given the ball to start a game, whereas games ﬁnished is the number of times the pitcher was throwing on the mound during the ﬁnal out (which is any failed attempt by a hitter to advance to a base). A shutout is a game in which one team does not score any runs. A pitcher earns a Saves if he is able to hold a lead for his team at the end of the game. Pitchers who earn saves, called relievers, tend not to gain Wins, so it is customary to treat saves and Wins equally, especially when studying pitcher salaries. Number of Home runs, which is assumed to be negatively related to salary, is the number of pitches that were hit by batters that were scored as a home run. A pitcher is assigned a Walks, which is assumed to be negatively related to salary, if he allows a batter to reach base after pitching him four balls. He is assigned a Strikeouts if he pitches three strikes (pitched balls counted against the batter, typically swung at and missed or fouled off) in a row. An Innings is one of nine periods in an MLB game in which each team has a turn at bat; Innings Pitched is the number of such periods when the pitcher was working. Earned Run Average is negatively correlated with the pitcher’s ability to prevent the opposing team from scoring. It equals the number of times the pitcher allows a batter to score a run (where the batter scores a point by advancing around the bases and reaching home plate safely) 9, divided by the number of innings pitched. Finally, the Strikeout Rate is the percentage of times the pitcher has succeeded in striking a batter out.

2,506.41 69.746 254.275 1,060.200 0.407 0.267 0.459 0.383 0.116 0.059

7.061 64.957 2.672 0.600 0.296 0.514 0.486 0.073

Professional characteristics MLB Experience MLB Experience-Squared Tenure with Current Club Free Agent Eligible for Final Offer Arbitration American League National League Canadian Team

Performance At Bats Stolen Bases Bases on Balls Total Bases Slugging Average Batting Average Inﬁelder Outﬁelder Catcher Designated Hitter

13.890 30.304 0.502 0.498 0.824 0.176

Mean

All

Personal characteristics Log Annual Salary Age White Nonwhite Native Immigrant

Variable

2,001.58 112.52 247.74 913.52 0.06 0.03 0.50 0.49 0.32 0.24

3.89 69.31 3.00 0.49 0.46 0.50 0.50 0.26

1.13 3.70 0.50 0.50 0.38 0.38

Std. dev

2,514.121 70.825 263.865 1,066.499 0.408 0.267 0.450 0.390 0.122 0.054

7.099 65.900 2.772 0.597 0.302 0.498 0.502 0.060

13.882 30.538 0.600 0.400 – –

2,042.69 119.51 258.01 934.2 0.06 0.02 0.50 0.49 0.33 0.23

3.94 71.49 3.13 0.49 0.46 0.50 0.50 0.24

1.11 3.69 0.49 0.49 – –

Std. dev

Native

2,470.25 64.693 209.271 1,030.641 0.401 0.266 0.505 0.354 0.089 0.078

6.885 60.531 2.203 0.615 0.266 0.589 0.411 0.135

13.923 29.208 0.042 0.958 – –

Mean

1,800.64 71.11 186.42 811.02 0.07 0.02 0.50 0.48 0.28 0.27

3.63 57.96 2.22 0.49 0.44 0.49 0.49 0.34

1.21 3.58 0.20 0.20 – –

Std. dev

Immigrant

Descriptive statistic – hitters

Mean

Table 1.

2,419.738 44.800 253.131 1,016.772 0.404 0.264 0.556 0.217 0.189 0.046

7.062 64.785 3.062 0.597 0.304 0.521 0.479 0.067

13.865 30.596 – – 0.985 0.015

Mean

1,940.51 72.35 233.32 880.39 0.06 0.02 0.50 0.41 0.39 0.21

3.87 70.06 3.38 0.49 0.46 0.50 0.50 0.25

1.10 3.49 – – 0.12 0.12

Std. dev

White

2,593.888 94.925 255.428 1,104.028 0.410 0.269 0.362 0.551 0.042 0.072

7.061 65.131 2.279 0.603 0.287 0.507 0.493 0.079

13.914 30.011 – – 0.662 0.338

Mean

2,059.46 137.54 261.69 944.57 0.07 0.02 0.48 0.50 0.20 0.26

3.91 68.6 2.50 0.49 0.45 0.50 0.50 0.27

1.16 3.88 – – 0.47 0.47

Std. dev

Nonwhite

Nationality Discrimination in the Labor Market 247

0.250 0.235 0.260 0.255 1093

80.507 13.273 10.621 25,562.99 5,514,009

Mean

All

0.43 0.42 0.44 0.44

6.89 6.58 10.65 3,789.65 4,657,988

Std. dev

0.251 0.244 0.255 0.250 901

80.819 13.312 10.559 25,514.71 5,412,875

Mean

0.43 0.43 0.44 0.43

6.72 6.61 10.69 3,733.19 4,595,757

Std. dev

Native

0.245 0.193 0.281 0.281 192

79.047 13.086 10.913 25,789.51 5,988,602

Mean

0.43 0.40 0.45 0.45

7.47 6.47 10.45 4,046.65 4,924,360

Std. dev

Immigrant

0.255 0.248 0.248 0.250 549

80.938 12.959 10.719 25,508.57 5,313,189

Mean

0.44 0.43 0.43 0.43

6.77 6.6 10.8 3,757.99 4,509,095

Std. dev

White

0.244 0.222 0.272 0.261 544

80.073 13.589 10.522 25,617.9 5,716,676

Mean

0.43 0.42 0.45 0.44

6.99 6.56 10.50 3,824 4,799,205

Std. dev

Nonwhite

Source: All variables except race and Greater metro area characteristics (GMAC) extracted from the Lahman Baseball Database (Version 5.0, Release Date: Dec. 15, 2002). Race is derived form observed Topps Baseball Cards, years 1992, 1993, 1994, 1997, 1999 (only years available). GMAC derived from the Statistical Abstract 1997–1999, the BEA, CA1-3, and from Statistical Canada. a Population denotes the greater metro area population.

Year dummies 1992 1993 1997 1998 Sample size

Greater metro area characteristics Percentage white Percentage black Percentage Hispanic Average annual income ($) Populationa

Variable

Table 1. (Continued )

248 + B. Bodvarsson and John G. Sessions Orn

13.409 29.815 0.782 0.218 0.857 0.143

5.988 53.468 1.924 0.467 0.306 0.513 0.487 0.069

0.442 37.446 34.179 74.12 10.15 2.875 19.488

Professional characteristics MLB Experience MLB Experience-Squared Tenure with Current Club Free Agent Eligible for Final Offer Arbitration American League National League Canadian Team

Performance Starter Wins Losses Games Started Complete Games Shutouts Saves

Mean

0.50 44.33 37.05 105.53 22.24 6.08 51.87

4.20 76.64 2.07 0.50 0.46 0.50 0.50 0.25

1.19 4.09 0.41 0.41 0.35 0.35

Std. dev

All

Personal characteristics Log Annual Salary Age White Nonwhite Native Immigrant

Variable

Table 2.

0.436 39.124 35.593 76.83 10.677 2.997 21.257

6.184 56.43 1.955 0.485 0.300 0.524 0.476 0.057

13.441 30.102 0.885 0.115 – –

Mean

0.50 45.39 37.86 107.99 22.85 6.24 54.87

4.27 78.98 2.11 0.50 0.46 0.50 0.50 0.23

1.20 4.06 0.32 0.32 – –

Std. dev

Native

0.477 27.378 25.692 57.855 6.983 2.14 8.878

4.808 35.692 1.738 0.355 0.343 0.448 0.552 0.14

13.213 28.093 0.169 0.831 – –

Mean

0.50 35.82 30.50 87.82 17.86 4.94 25.42

3.56 57.85 1.85 0.48 0.48 0.50 0.50 0.35

1.15 3.85 0.38 0.38 – –

Std. dev

Immigrant

Descriptive statistic – pitchers

0.441 39.007 35.904 77.769 10.981 3.065 20.941

6.158 55.562 1.935 0.482 0.314 0.518 0.482 0.063

13.451 30.19 – – 0.969 0.031

Mean

0.50 45.27 38.37 108.53 23.33 6.32 52.93

4.20 78.38 2.10 0.50 0.46 0.50 0.50 0.24

1.20 4.02 – – 0.17 0.17

Std. dev

White

0.447 31.832 27.973 61.00 7.16 2.191 14.267

5.374 45.939 1.885 0.412 0.279 0.496 0.504 0.092

13.258 28.466 – – 0.454 0.546

Mean

0.50 40.37 31.13 92.93 17.48 5.06 47.62

4.14 69.64 1.98 0.49 0.45 0.50 0.50 0.29

1.17 4.05 – – 0.50 0.50

Std. dev

Nonwhite

Nationality Discrimination in the Labor Market 249

62.57 249.73 514.13 702.42 0.96 0.02

Std. dev 58.619 234.708 456.643 653.705 4.004 0.078

Mean 63.93 255.69 531.38 716.96 0.95 0.02

Std. dev

Native

43.907 172.204 316.628 470.919 4.152 0.079

Mean 52.12 202.97 374.42 585.76 1.00 0.02

Std. dev

Immigrant

58.842 231.782 450.726 655.16 3.995 0.078

Mean 64.46 257.66 530.21 720.78 0.94 0.02

Std. dev

White

48.16 204.195 386.00 528.473 4.133 0.081

Mean

54.54 217.94 448.86 623.3 1.04 0.02

Std. dev

Nonwhite

0.221 0.239 0.264 0.276 1,203

0.42 0.43 0.44 0.45

0.232 0.25 0.258 0.261 1,031

0.42 0.43 0.44 0.44

0.157 0.174 0.302 13.213 172

0.36 0.38 0.46 1.15

0.236 0.248 0.256 0.26 941

0.42 0.43 0.44 0.44

0.168 0.206 0.294 0.332 262

0.37 0.41 0.46 0.47

80.714 6.84 80.647 6.84 81.116 6.84 80.695 6.91 80.782 6.58 13.038 6.46 13.144 6.52 12.399 6.08 12.946 6.49 13.368 6.34 10.975 10.77 10.848 10.42 11.74 12.68 10.899 10.61 11.251 11.35 25,488.15 3,939.85 25,573.86 3,875.25 24,973.87 4,283.28 25,491.51 3,895.3 25,476.06 4,103.68 5,551,948 4,683,874 5,526,000 4,632,204 5,707,635 4,994,077 5,481,401 4,631,793 5,805,594 4,867,179

56.517 225.779 436.641 627.592 4.025 0.078

Mean

All

Source: All variables except race and Greater metro area characteristics (GMAC) extracted from the Lahman Baseball Database (Version 5.0, Release Date: Dec. 15, 2002). Race is derived form observed Topps Baseball Cards, years 1992, 1993, 1994, 1997, 1999 (only years available). GMAC derived from the Statistical Abstract 1997–1999, the BEA, CA1-3, and from Statistical Canada. a Population denotes the greater metro area population.

Year dummies 1992 1993 1997 1998 Sample size

Greater metro area characteristics Percentage white Percentage black Percentage Hispanic Average annual income ($) Populationa

Homeruns Walks Strikeouts Innings Pitched ERA Strikeout Rate

Variable

Table 2. (Continued )

250 + B. Bodvarsson and John G. Sessions Orn

Nationality Discrimination in the Labor Market

251

information about contract length, bonus clauses, or endorsements. Salaries for players on the Canadian teams were converted to U.S. dollars. The experience data were used to set controls for a player’s eligibility for free agency and ﬁnal offer arbitration.23 The player’s race was inferred from inspection of Topps baseball cards for all four seasons. For the U.S. teams, metropolitan area population and per capita income were obtained from the website of the Bureau of Economic Analysis (see: www.bea.gov). For the Canadian teams, similar data were obtained from the Statistics Canada website (see: www.statcan.ca). Per capita income data for the Canadian cities were converted to U.S. dollars. It appears from Table 1 that immigrant hitters (who are overwhelmingly nonwhite) are generally quite similar to their native-born peers in terms of their personal characteristics and in terms of the types of markets in which they play. In terms of their professional characteristics, however, immigrants tend to have less MLB experience, less tenure with their current club, and are less likely to be eligible for ﬁnal offer arbitration than their native-born counterparts. In terms of performance, immigrant hitters record fewer At Bats, Stolen Bases, Bases on Balls, and Total Bases than native-born hitters, are less likely to be catchers, but more likely to be designated hitters. It also appears from Table 1 that there are no major differences between the personal and professional characteristics of white and nonwhite hitters, nor in the characteristics of the greater metropolitan areas in which they play. In terms of career characteristics, however, nonwhite hitters record signiﬁcantly more At Bats, Stolen Bases, and Total Bases than white hitters. They are also less likely to play as an inﬁelder or catcher, but are more likely to play as an outﬁelder or as a designated hitter. In Table 2, the domination of native and white pitchers is immediately apparent. Pitchers are predominately white natives – 78 percent of pitchers are white, 86 percent of pitchers are native, 89 percent of native pitchers are white, and 97 per cent of white pitchers are native. Relative to their nonwhite and immigrant counterparts, white and native pitchers enjoy higher average earnings, are generally older, have greater MLB experience and tenures with their current club, and are more likely to be free agents. In terms of career characteristics, white and native pitchers record 23

In MLB, player salaries are set under two regimes, one competitive, the other monopsonistic. The monopsonistic regime applies to players with fewer than six years of league experience. These players are subject to the reserve clause and are constrained to negotiate their pay with only one team. The competitive regime applies to players with at least six years of league experience. They are eligible to ﬁle for free agency and may negotiate with any team in the league. Monopsony power effectively begins to erode, however, as early as the fourth year because then a player is eligible for ﬁnal offer arbitration. Arbitration rights tend to relieve players of monopsonistic exploitation because arbitrators strive to award competitive salaries. The Major League added new teams (called ‘‘expansion teams’’) since the early 1990s, leading to a reduction in each team’s monopsony power held over reserve clause players.

252

+ B. Bodvarsson and John G. Sessions Orn

signiﬁcantly higher Wins, Losses, Games Started, Complete Games, Shutouts, Saves, Homeruns, Walks, Strikeouts, and Innings Pitched. We tested several implications of our theoretical model using ordinary least squares (OLS), the results of which are presented in Tables 3 and 4, where we set out earnings regressions for Hitters and Pitchers, respectively. Note that for the nativity status dummy (‘‘Native’’), the player is assigned a value of one if native-born and zero otherwise, as well as a one if white (and zero otherwise). We estimate six speciﬁcations: (1) All, (2) Natives, (3) Immigrants, (4) Whites, (5) Non-Whites, and (6) All with interactions between the nativity status dummy and the various productivity indicators. Looking at Hitters ﬁrst (Table 3), it is evident that the regressions are generally well speciﬁed, and that the coefﬁcients on the explanatory variables are generally robust, across all the various speciﬁcations. Earnings are negatively related to Age but positively and concavely related to MLB Experience. It would appear that the negative coefﬁcient on Age is reﬂecting the player’s physical depreciation, while the positive coefﬁcient on experience is reﬂecting rewards to greater human capital – indeed when we experimented with dropping age from our regressions we found that the coefﬁcient on MLB Experience declined by almost exactly the size of the coefﬁcient on Age. Earnings are also positively and signiﬁcantly related to Tenure with Current Club and also to whether the player is a Free Agent or Eligible for Final Offer Arbitration. Career characteristics are dominated by the effects of a player’s Slugging and Batting Average, although At Bats and Stolen Bases exert small, but signiﬁcant effects on earnings. Two implications of our model are that: (a) If there is discrimination against (in favor of) immigrants, the coefﬁcient on the nativity status dummy will be positive (negative) and signiﬁcant; and (b) productivity and nativity status will interact (the marginal effect of nativity status on pay depends upon how productive one is). For hitters, we found no evidence of differences in intercepts between natives and immigrants. Thus, hypothesis (a) is not conﬁrmed for hitters. However, there is some evidence indicating conﬁrmation of hypothesis (b) for the hitter group. Note from speciﬁcation (6) of Table 3 that three of the nativity status productivity interactions are signiﬁcant, as are the interactions between nativity status and tenure with current club and with playing on a Canadian team. Turning to pitchers (Table 4), we ﬁnd – somewhat surprisingly – that Age impacts positively on the earnings of immigrant and nonwhite pitchers, MLB Experience impacts positively on the earnings of all pitchers, and that Tenure with Current Club impacts positively upon the earnings of all but immigrant pitchers. The coefﬁcients on the productivity variables generally accord to a priori expectations, although there are some noticeable discrepancies across the various subsample regressions. For example, the pay of nonwhite and immigrant, but not white or native, pitchers is signiﬁcantly and positively related to Wins, and signiﬁcantly

1.65 0.49

0.063 0.070

0.000 0.001 0.000 0.000 4.240 3.425 0.026 0.108 0.304 0.046

Performance At Bats Stolen Bases Bases on Balls Total Bases Slugging Average Batting Average Inﬁelder Outﬁelder Catcher Designated Hitter

2.55 3.73 0.76 1.42 7.45 3.21 0.14 0.59 1.56 0.31

7.24 10.40 6.55 4.41 4.49

4.12 0.43 – 0.92 –

t-Stat.

0.317 0.021 0.043 0.632 0.366

0.046 0.020 – 0.051 –

Mean

(1) All

Professional characteristics MLB Experience MLB Experience-Squared Tenure with Current Club Free Agent Eligible for Final Offer Arbitration American League Canadian Team

Personal characteristics Age White Nonwhite Native Immigrant

Variable

0.000 0.001 0.000 0.000 4.610 2.714 0.011 0.163 0.249 0.093

0.065 0.216

0.330 0.021 0.041 0.555 0.322

0.047 0.011 – – –

Mean

2.39 3.85 0.07 1.10 7.47 2.33 0.05 0.75 1.11 0.50

1.53 1.39

6.89 9.57 5.71 3.52 3.60

3.86 0.22 – – –

t-Stat.

(2) Native

Table 3.

0.000 0.000 0.001 0.000 1.026 7.949 0.073 0.01 0.384 0.15

0.128 0.279

0.279 0.023 0.098 0.801 0.438

0.029 0.024 – – –

Mean

0.77 0.32 2.60 1.11 0.74 2.78 0.20 0.03 0.98 0.65

1.34 0.89

2.75 4.58 5.47 2.43 2.11

0.99 0.10 – – –

Mean

(3) Immigrant

0.000 0.001 0.000 0.000 4.386 0.913 0.252 0.383 0.02 0.196

0.128 0.238

0.356 0.021 0.034 0.451 0.335

0.065 – – 0.124 –

Mean

1.05 2.18 0.08 1.34 5.11 0.63 0.83 1.25 0.06 0.70

2.38 1.21

6.06 7.68 3.83 2.25 2.86

4.30 – – 0.38 –

t-Stat

(4) White

Log aannual salary – hitters

0.000 0.001 0.000 0.000 3.885 6.002 0.066 0.085 0.495 0.155

0.029 0.062

0.294 0.023 0.049 0.741 0.377

0.023 – – 0.05 –

Mean

2.50 3.11 2.12 0.86 5.46 3.92 0.30 0.38 2.02 0.96

0.51 0.31

5.47 10.05 4.97 4.18 3.72

1.36 – – 0.87 –

t-Stat

(5) Nonwhite

0.000 0.000 0.001 0.000 1.100 7.572 0.053 0.027 0.414 0.163

0.125 0.173

0.276 0.226 0.095 0.868 0.500

0.044 0.014 – 0.193 –

Mean

0.88 0.02 2.77 1.18 0.86 2.86 0.15 0.08 1.13 0.68

1.43 0.94

3.00 4.66 5.54 2.77 2.69

3.94 0.31 – 0.26 –

t-Stat

0.17 1.06 2.53 0.58 2.53 1.70 0.15 0.49 0.39 0.85

0.59 2.17

0.058 0.344

0.000 0.001 0.001 0.000 3.518 4.874 0.063 0.191 0.167 0.254

0.50 0.25 2.94 0.88 0.40

– – – –

t-Stat

0.051 0.001 0.054 0.309 0.207

– – – – –

Mean

(6) All nativity productivity interactions

Nationality Discrimination in the Labor Market 253

a

0.069 1.31 0.129 1.92 0.208 2.83 9.515 12.96 0.7385 184.6228, 1064 0.58452 1093

0.067 0.144 0.240 10.001 0.7280 150.0527, 873 0.5870 901

0.001 0.005 0.003 0.000 0.000

Mean

1.15 1.90 2.84 11.75

0.16 0.84 1.06 0.45 0.35

t-Stat.

(2) Native

Population denotes the greater metro area population.

Year dummies 1992 1993 1997 Constant R-Squared F-Statistic Root mean squared error Observations

0.84 1.57 0.96 0.87 0.32

t-Stat.

(1) All

0.004 0.008 0.002 0.000 0.000

Mean

Greater metro area characteristics Percentage white Percentage black Percentage Hispanic Average annual income ($) Populationa

Variable

0.139 0.162 0.241 8.283 0.8243 53.2427, 164 0.54901 192

0.011 0.014 0.003 0.000 0.000

Mean

1.12 1.13 1.50 5.46

0.92 1.17 0.46 0.44 0.51

Mean

(3) Immigrant

0.21 1.13 1.60 0.47 0.35

0.048 0.61 0.19 2.08 0.215 2.23 11.249 9.68 0.7266 94.2827, 521 0.58971 549

0.002 0.008 0.005 0.000 0.000

t-Stat

(4) White

Mean

Table 3. (Continued )

1.26 0.57 1.64 8.46

0.48 1.01 0.39 0.94 0.03

t-Stat

0.091 0.055 0.176 8.344 0.7686 102.3627, 516 0.57091 544

0.003 0.008 0.001 0.000 0.000

Mean

(5) Nonwhite

0.075 0.142 0.230 9.418

0.003 0.007 0.002 0.000 0.000

Mean

– – – –

– – – – –

Mean

3.14 2.15 3.14 10.37 0.7465 125.2645, 1047 0.58016 1093

0.71 1.43 0.90 0.77 0.23

t-Stat

– – – –

– – – –

t-Stat

(6) All nativity productivity interactions

254 + B. Bodvarsson and John G. Sessions Orn

0.26 0.21

0.009 0.026

0.383 0.009 0.004 0.003 0.001 0.048 0.003 0.005 0.000 0.001

Performance Starter Wins Losses Games Started Complete Games Shutouts Saves Homeruns Walks Strikeouts

7.65 2.10 1.09 2.38 0.38 4.41 3.50 3.71 0.95 1.85

4.46 9.37 7.46 6.05 6.37

0.22 1.15 – 1.69

t-Stat

0.153 0.014 0.068 0.739 0.421

0.002 0.058 – 0.103

Mean

(1) All

Professional characteristics MLB Experience MLB Experience-Squared Tenure with Current Club Free Agent Eligible for Final Offer Arbitration American League Canadian Team

Personal characteristics Age White Nonwhite Native Immigrant

Variable

0.37 0.007 0.005 0.002 0.002 0.054 0.002 0.005 0.000 0.001

0.019 0.036

0.162 0.014 0.068 0.800 0.470

0.013 0.108 – –

Mean

7.10 1.63 1.09 1.54 0.71 4.52 3.33 3.74 1.23 2.11

0.49 0.25

4.57 9.08 7.08 6.17 6.73

1.24 1.90 – –

t-Stat

(2) Native

Table 4.

2.37 3.70 0.99 2.44 0.35 0.84 1.98 1.46 0.08 0.88

0.94 0.67

0.095 0.160

0.360 0.049 0.009 0.011 0.005 0.030 0.005 0.008 0.000 0.001

2.83 7.03 0.59 0.08 0.03

2.06 1.83 – –

Mean

0.252 0.026 0.016 0.022 0.005

0.043 0.197 – –

Mean

(3) Immigrant

0.382 0.006 0.004 0.001 0.001 0.057 0.003 0.006 0.001 0.001

0.028 0.01

0.179 0.014 0.064 0.761 0.47

0.017 – – 0.067

Mean

6.87 1.33 0.82 0.76 0.48 4.65 4.15 3.80 1.54 2.97

0.70 0.07

5.14 9.13 6.44 5.83 6.53

1.67 – – 0.78

t-Stat

(4) White

Log annual salary – pitchers

0.41 0.04 0.01 0.012 0.02 0.003 0.001 0.004 0.000 0.002

0.015 0.208

0.157 0.021 0.064 0.418 0.143

0.054 – – 0.194

Mean

3.95 3.89 1.13 4.37 1.80 0.09 0.33 0.89 0.23 2.39

0.18 0.82

2.00 6.52 3.08 1.51 1.06

2.90 – – 2.50

t-Stat

(5) Nonwhite

0.304 0.045 0.019 0.013 0.011 0.020 0.005 0.003 0.000 0.001

0.097 0.060

1.91 3.53 2.18 2.40 0.93 0.58 2.39 0.07 0.13 1.00

1.15 0.39

3.03 8.51 0.89 0.50 0.09

1.04

0.531

0.223 0.024 0.020 0.127 0.013

0.21 1.00

t-Stat

0.002 0.052

Mean

0.075 0.038 0.024 0.011 0.014 0.034 0.003 0.002 0.001 0.000

0.118 0.035

0.077 0.010 0.048 0.681 0.464

– – – –

Mean

0.45 2.84 2.53 1.96 1.15 0.92 1.17 0.48 0.49 0.08

1.33 0.27

0.95 3.15 1.94 2.38 2.86

– – – –

t-Stat

(6) All nativity productivity interactions

Nationality Discrimination in the Labor Market 255

a

0.012 0.25 0.126 2.10 0.250 3.68 11.275 17.67 0.7848 195.83 31, 1171 0.56049 1203

0.27 0.97 2.66 0.76 0.49

0.41 1.16 2.52 1.09 0.42

2.15 5.57 4.28

t-Stat

0.045 0.89 0.098 1.50 0.184 2.46 11.253 16.1 0.7933 190.83 30, 1000 0.55222 1031

0.002 0.006 0.005 0.000 0.000

0.001 0.138 5.682

Mean

(2) Native

Population denotes the greater metro area population.

Year dummies 1992 1993 1997 Constant R-Squared F-Statistic Root mean squared error Observations

0.001 0.005 0.005 0.000 0.000

2.43 5.32 4.15

t-Stat

(1) All

0.001 0.131 5.590

Mean

Greater metro area characteristics Percentage white Percentage black Percentage Hispanic Average annual income ($) Population a

Innings Pitched ERA Strikeout Rate

Variable

0.18 0.70 1.00 0.29 0.91

1.18 0.33 0.17

Mean

0.251 1.80 0.19 1.30 0.437 2.72 9.817 6.74 0.8396 47.05 30, 141 0.50688 172

0.002 0.008 0.005 0.000 0.000

0.002 0.025 0.498

Mean

(3) Immigrant

0.36 0.96 2.31 0.47 0.90

1.84 6.36 4.14

0.041 0.77 0.127 1.90 0.209 2.74 11.57 16.08 0.7976 186.13 30, 910 0.54665 941

0.002 0.005 0.005 0.000 0.000

0.001 0.150 5.588

t-Stat

(4) White

Mean

Table 4. (Continued )

0.06 0.74 1.69 0.84 0.52

1.90 0.49 2.86

t-Stat

0.099 1.04 0.021 0.15 0.215 1.43 9.171 6.44 0.8092 55.38 30, 231 0.54352 262

0.001 0.008 0.008 0.000 0.000

0.002 0.025 7.846

Mean

(5) Nonwhite

0.021 0.116 0.228 10.60

0.002 0.006 0.005 0.000 0.000

0.002 0.015 4.054

Mean

– – – – –

– – – – –

0.001 0.160 1.415

Mean

0.43 1.94 3.38 12.80 0.796 138.8051, 1151 0.55039 1203

0.37 1.20 2.51 1.07 0.09

1.79 0.23 1.33

t-Stat

– – – – –

– – – – –

0.74 2.29 0.44

t-Stat

(6) All nativity productivity interactions

256 + B. Bodvarsson and John G. Sessions Orn

Nationality Discrimination in the Labor Market

257

and negatively related to Games Started. The pay of white, native and immigrant pitchers, but not nonwhite pitchers, is signiﬁcantly and positively related to Saves while the pay of white and native, but not nonwhite or immigrant, pitchers is signiﬁcantly and negatively related to Shut Outs, Home Runs, and ERAs. In terms of nationality discrimination within the pitcher group, our analysis conﬁrms some implications of our theoretical model. There is evidence from Table 4 of reverse nationality discrimination for the pooled sample and for the subsample of nonwhite pitchers: In the pooled group, native-born pitchers make 10.3 percent less than their immigrant counterparts, all other things being equal; in the nonwhite subsample, native-born pitchers make 19.4 percent less, all other things being equal. This effect for the pooled sample falls away when the nativity status

productivity interactions are added in speciﬁcation (6). Nevertheless, speciﬁcation (6) indicates fairly substantial premia paid to immigrants, even after controlling for race. According to speciﬁcation (6), four of the nativity status productivity interactions are signiﬁcant, again conﬁrming the model’s implication – perhaps more strongly for pitchers than for hitters – that the marginal effect of nativity status varies with productivity. Finally, we note that four of the nativity status professional characteristics interactions are also signiﬁcant.

3.3. Decomposition analysis In this section, we attempt to identify nationality discrimination using another empirical approach. The fact that players of a particular nativity group enjoy a mean wage differential over players of another nativity group could be a reﬂection of the former group’s greater endowment of ‘‘earning characteristics.’’ Native-born pitchers may, for example, be more productive or have more experience on average than immigrant pitchers. Alternatively, native-born pitchers may be better rewarded for the characteristics they do possess, suggesting some form of positive (negative) discrimination against native-born pitchers. To address this issue we perform a Blinder–Oaxaca decomposition to separate the native/immigrant mean earnings differential into an ‘‘endowment component’’ to account for differences in endowments between individuals, and a ‘‘price component’’ which is usually associated with discrimination.24 Assume that the earnings function of players of nativity j in position i is: ln wij ¼ X ij Bij þ ij 24

(15)

This method of decomposition, initially proposed by Oaxaca (1973) and Blinder (1973), and later generalized by Oaxaca and Ransom (1994), has been applied extensively to discrimination on the basis of gender, race, caste, and religion.

+ B. Bodvarsson and John G. Sessions Orn

258

where i ¼ (N, I) and j ¼ ðH; PÞ denote native and immigrant and pitchers and hitters, respectively. Xij denotes a vector of explanatory variables, Bij the corresponding coefﬁcient vector to be estimated, and eij some wellbehaved error term. Thus, the earnings functions of native-born pitchers, immigrant pitchers, native-born hitters, and immigrant hitters may be denoted: ln wNH ¼ XNH BNH þ NH

(16)

ln wIH ¼ X IH BIH þ IH

(17)

ln wNP ¼ X NP B NP þ NP

(18)

ln wIP ¼ X IP B IP þ IP

(19)

The Blinder–Oaxaca decomposition divides wage differentials into a part that is ‘‘explained’’ by group differences in productivity and a residual part that cannot be accounted for by such differences in wage determinants. This latter ‘‘unexplained’’ component is often used as a measure for discrimination. For example, the predicted average native hitter/immigrant hitter (NH-IH) differential may be represented as: NH NH IH IH X B^ D ln wNHIH ¼ ln wNH ln wNNH ¼ X B^

(20)

) IH

D ln wNHIH ¼ B^ ðX

NH

X

IH

Þ þ X

NH

ðB^

NH

IH

B^ Þ

IH NH IH The ﬁrst term, B^ ðX X Þ, represents differences in endowments between members of the two groups while the second term, NH IH NH B^ Þ, represents differences in rewards. Note that if the X ðB^ overall differential is negative (i.e., D ln wNHIH o0) but the second term is NH IH NH B^ Þ40), then it would suggest that immigrant positive (i.e., X ðB^ hitters are discriminated against despite earning, on average, more than native hitters. In other words, immigrant hitters would do even better with the earnings generating function of native hitters than with their own. Speciﬁcation (20) presumes that the immigrant hitter wage structure prevails in the absence of discrimination. But this is a matter of debate. Assuming away any feelings of malevolence or benevolence from one group toward the other, it is equally valid to presume that the native hitter wage structure prevails, thereby requiring (20) to be respeciﬁed as: NH NH IH NH IH IH D ln wNHIH ¼ B^ ðX X Þ þ X ðB^ B^ Þ

(21)

The ﬁrst and second terms on the right-hand side of (21) still represent differences in endowments and rewards, respectively, but they will generally differ from those derived from (21).25 Many authors concede this ambiguity by simply reporting both decompositions. Some researchers, however, have

259

Nationality Discrimination in the Labor Market

attempted to confront the issue head-on by hypothesizing the nondiscri directly.26 Reimers (1983), for example, minatory parameter vector, B, proposes using the average coefﬁcients over both groups as an estimate Neumark (1988) advocates using the coefﬁcients from a pooled of B. In what follows, we follow regression over both groups as an estimate of B. the ‘‘hybrid’’ decomposition technique popularized by Cotton (1988) in which the prevailing nondiscriminatory wage structure is assumed to be a weighted average of the wage structures of the two groups under consideration: D ln wNHIH ¼ X NH

NH

ðB^

NH

þ X BÞ

IH

IH

X ðB B^ Þ þ Bð

NH

X

IH

Þ

(22)

IH

þ ð1 OÞB^ represents the estimated nondiscriminatory where B ¼ OB^ parameter vector, with O denoting the proportion of the sample comprised by native hitters. The ﬁrst right-hand term in the decomposition is the overpayment enjoyed by native hitters, the second term is the underpayment suffered by immigrant hitters, and the third term is the portion of the wage differential that is explained by differences in endowments. We perform the above three decompositions for the native/immigrant hitter and native/immigrant pitcher differentials, and our results, based on the regressions set out in Tables 3–4, are collected in Tables 5 and 6. Considering Table 5, our regression model implies a positive salary premium for native hitters over immigrant hitters ceteris paribus. The ﬁrst two decompositions, which follow (20) and (21), respectively, in presuming that the immigrant hitter and native hitter wage structure would prevail in the absence of any discrimination, suggests that this premium is predominately explained by endowments. Decomposition based on the immigrant hitter wage structure suggests that 95 percent of the differential is attributable to the superior endowments of native hitters, with only 5 percent being attributable to price effects. Decomposition based on the native hitter wage structure suggests that differences in endowments explain 86 percent of the differential, with discrimination against native hitters reducing the wage differential by 14 percent. The hybrid decomposition, derived from (22), echoes the ﬁnding that the differential is almost entirely endowment driven, with native hitter overpayment and immigrant hitter underpayment offsetting the potential native hitter wage premium by 0.75 and 11.6 percent, respectively. Thus, the decomposition results basically conﬁrm the OLS results, namely that there is no compelling evidence of nationality discrimination either against or in favor of immigrant hitters. 25

The point that an undervaluation of one group implies an overvaluation of the other is neatly summarized by Cotton (1988, p. 238): ‘‘y not only is the group discriminated against undervalued, but the preferred group is overvalued, and the undervaluation of the one subsidizes the overvaluation of the other.’’ 26 Oaxaca and Ransom (1994) provide an integrative treatment of the various methods.

+ B. Bodvarsson and John G. Sessions Orn

260

Table 5.

Oaxaca–Cotton decompositions: native hitter/immigrant hitter Dln wNHIH ¼ ln wNH ln wIH Coef.

Native hitter wage structure NH NH IH Endowment effect: B^ ðX X Þ NH IH IH Price effect: X ðB^ B^ Þ

%

6.031

95.75

0.267

4.25

B^ Þ

6.298

100.00

Immigrant hitter wage structure IH NH IH Endowment effect: B^ ðX X Þ NH IH NH Price effect: B^ Þ X ðB^ IH NH IH NH IH NH Total differential: B^ ðX X Þ þ X ðB^ B^ Þ

7.183

114.05

0.885

14.05

6.298

100.00

0.047

0.75

0.729

11.58

6.980

110.83

6.298

100.00

Total differential:

Hybrid wage structure Native hitter overpayment: Immigrant hitter underpayment: Endowment effect: Total differential:

B^

NH

ðX

NH

X

IH

Þ þ X

IH

ðB^

NH

IH

NH NH BÞ X ðB^ IH IH X ðB B^ Þ

X NH X IH Þ Bð NH NH þ X IH ðB B^ IH Þ BÞ X ðB^ X NH X IH Þ þBð

Table 6.

Oaxaca–Cotton decompositions: native pitcher/immigrant pitcher D ln wIP ¼ ln wNPln wIP Coef.

Native pitcher wage structure NP NP IP Endowment effect: X Þ B^ ðX Price effect: Total differential:

IP

NP

IP

X ðB^ B^ Þ NP NP IP NP IP IP B^ ðX X Þ þ X ðB^ B^ Þ

Immigrant pitcher wage structure IP NP IH Endowment effect: X Þ B^ ðX NP IP NP Price effect: X ðB^ B^ Þ Total differential: Hybrid wage structure Native pitcher overpayment: Immigrant pitcher underpayment: Endowment effect: Total differential:

%

0.397

12,527.91

0.400

12,627.91

0.003

100.00

0.412

13,108.88

0.415

13,008.88

IP NP IP NP IP NP B^ ðX X Þ þ X ðB^ B^ Þ

0.003

100.00

NP NP X ðB^ BÞ

0.057

1,679.40

IP IP X ðB B^ Þ

0.355

10,399.52

0.409

11,978.92

0.003

100.00

X NP X IP Þ Bð NP NP þ X IP ðB B^ IP Þ X ðB^ BÞ X NP X IP Þ þBð

Nationality Discrimination in the Labor Market

261

Table 6 focuses on the native/immigrant pitcher differential, and the results here are in stark contrast to what we found from our OLS analyses. The decompositions based on both native pitcher and immigrant pitcher wage structures suggest that the substantially superior endowments of native pitchers are matched by equally substantial discrimination against immigrant pitchers, resulting in a negligible net potential wage differential. The hybrid decomposition suggests that it is the underpayment of immigrant pitchers, rather than the overpayment of native pitchers, that is acting to offset the endowment effect. In fact, the absolute value of the contribution of the immigrant underpayment to the native/immigrant mean earnings differential is almost the same as the contribution of the endowment effect. These results disconﬁrm the evidence from OLS, which indicated reverse discrimination.

4. Concluding remarks Nationality discrimination is a particularly challenging type of discrimination to detect empirically because in most data sets of immigrants and natives, the effects of birthplace on pay will be inﬂuenced by productivity differences between natives and immigrants. This interrelatedness will occur anytime one is studying differences in pay between majority and minority workers when the two groups are imperfect substitutes. A very strong argument can be made for native/immigrant productivity differences because most immigrants need to assimilate to the host economy. If they also experience nationality discrimination, any theory and test of discrimination must account for the joint effects of birthplace and productivity differences. We suspect that most researchers have shied away from the study of nationality discrimination because of this problem of ‘‘disentangling,’’ to use the language of Nielsen et al. (2004). In this chapter, we have attempted to address the problem of how to measure nationality discrimination from both theoretical and empirical points of view. We developed a model of native/immigrant earnings differences that accounts for the joint effects of birthplace and productivity differences. Our key theoretical concept is an extension of Becker’s traditional measure of discrimination, the MDC, to the case of nationality discrimination when majority and minority workers are imperfect substitutes. Our MDC successfully allows one to tease out the contribution of productivity differences from the contribution of prejudice to the native/immigrant earnings wage gap. This has not been done before theoretically, and we contend that laying the theoretical groundwork is very helpful when it comes to designing an empirical speciﬁcation. Our theoretical model produces a number of counterintuitive implications that are easily testable.

262

+ B. Bodvarsson and John G. Sessions Orn

Another factor that has hampered the study of nationality discrimination in the past has been the absence of test cases that lend themselves well to estimating discrimination when there are majority/minority productivity differences. To test our model, we chose a particular industry that’s very amenable to the study of nationality discrimination – MLB. Our examination of that industry yielded some very interesting results. First, we found evidence of nationality discrimination in the pitchers group – OLS results showing reverse discrimination and decompositions showing discrimination against immigrants. Because the decomposition technique is now generally recognized as the stronger, more informative, of the techniques for detecting discrimination, we are inclined to attach more credibility to the decompositions. The discrimination against immigrant pitchers we observe was quite substantial for the 1990s. However, we found no evidence of discrimination for or against hitters. This leads to a future research question: Why, within the same industry, as well as within the same ﬁrm, does one observe discrimination in one occupation or job assignment and not in another? Our empirical work also revealed conﬁrmation of an important implication from our theory: Birthplace and productivity interact in the determination of pay. This is conﬁrmation of a more general implication of our theory, which is that whenever majority and minority workers are imperfect substitutes in production, the personal attribute that is the focus of prejudice (race, gender, birthplace, etc.) will interact with relative productivity in inﬂuencing pay. Elsewhere (Bodvarsson and Sessions (2008) we have found this to be true when studying racial discrimination across job assignments within the ﬁrm. Therefore, future studies of nationality discrimination, as well as studies of discrimination in general where majority and minority workers are not perfect substitutes, must take account of interaction effects between productivity and birthplace. Now that a theory of nationality discrimination has been tested on a particular industry, the next step in this research area is to investigate discrimination for nationwide panel data sets. It will be interesting to see whether results from an interoccupational nationwide study, particularly in countries where immigration is signiﬁcant, will yield similar results to the ones found here. We hope our study spawns additional work in this area. Acknowledgments We thank Bree Dority O’Callaghan and Robert Girtz for assistance with data collection and bear sole responsibility for any errors.

No

No

No

Tandon (1978)

Chiswick (1980)

Tienda and Niedert (1980) Haig (1980)

No

Yes

Meng (1987)

Gabriel and Schmitz (1987) Tran-Nam and Neville (1988) Daneshvary and Weber (1991) Daneshvary (1993)

Yes

Yes

No

No

Reimers (1984)

Fujii and Mak (1983) No

Yes

No

Chiswick (1978)

Decomposition

OLS

OLS

OLS

Decomposition

OLS

OLS

OLS

OLS

Statistical method

Males in Australian 1981–82 Decomposition Income and Housing Survey White males in 1980 U.S. Decomposition Census of Population White male college graduates in Decomposition 1980 U.S. Census of Population

Males in 1976 Survey of Income and Education Males in 1973 Canadian National Mobility Survey 1980 U.S. Census of Population

Males in 1971 Canadian Census of Population Males in 1972 British General Household Survey Hispanic males in 1976 Survey of Income and Education Australia’s 1973 Henderson Inquiry into Poverty Males in 1975 OEO Census update survey for Hawaii

White males in 1970 U.S. Census

Is focus Data and sample nationality discrimination?

Author(s) and year

Yes

Yes

Yes

Yes

Yes

Yes

Not Yes relevant Not Yes relevant

No

Yes

No

Yes

Yes

Not Yes relevant Yes Yes

Yes

Discrimination against nonEnglish-speaking migrants Regional differences in discrimination Discrimination against immigrants

Discrimination against immigrants Inconclusive evidence

Possible discrimination against immigrants Discrimination against immigrants Discrimination in favor of Filipino immigrants, but against Japanese immigrants None

None

None

None

Evidence of Conclusion reg. discrimination discrimination?

Not Yes relevant None Yes

Control for race?

Appendix. Summary of studies providing information about ceteris paribus native/immigrant earnings differences

Nationality Discrimination in the Labor Market 263

Yes

No

Kee (1995)

Bucci and Tenorio (1997) Shamsuddin (1998)

No

Yes

Goyette and Xie (1999)

Denny et al. (2000)

Yes

No

Beach and Worswick (1993)

Males in the 1980 and1990 U.S. Census of Population Canadian Income (1983) Assets and Debts (1984) of Economic Families and Unattached Individuals 1990 U.S. Census of Population and 1982–89 Survey of Natural and Social Scientists and Engineers British General Household Survey for 1974–93

Males in the Dutch Quality of Life Surveys, 1984–85

Females in the 1973 Canadian Job Mobility Survey

Is focus Data and sample nationality discrimination?

Author(s) and year

Appendix. (Continued)

Generalized Lorenz and Generalized Concentration Curve analysis

OLS, Logit

Decomposition

Decomposition

Decomposition

OLS

Statistical method

Yes

Yes

No

Yes

No

No

Control for race?

No

Yes

Yes

Yes

Yes

Yes

Female immigrant scientists less likely to be employed

Some evidence of discrimination against high-skilled immigrants Discrimination against Antillean and Turkish immigrants Nepotism toward native-born workers Discrimination against male immigrants

Evidence of Conclusion reg. discrimination discrimination?

264 + B. Bodvarsson and John G. Sessions Orn

Yes

No

Yes

Nielsen et al. (2004)

Lang (2005)

Pedace (2008)

Bodvarsson and Fuess Yes (2008)

Yes

Hayfron (2002)

Soccer players in England’s Premier and First to Third division clubs 1996–2002 Major League Baseball players, 1997–98

Norwegian National Insurance Administrative database Two register-based data sets on the Danish population Males in the 2000 German Socio-Economic Panel No

No

OLS

Yes

Stochastic earnings No frontier approach OLS No

Decomposition

Decomposition

Yes

Yes

No

Yes

Yes

Discrimination in favor of South American division players Reserve clause immigrants suffer bargaining disadvantages

Discrimination against immigrants Discrimination against female immigrants None

Nationality Discrimination in the Labor Market 265

266

+ B. Bodvarsson and John G. Sessions Orn

References Altonji, J.G., Blank, R.M. (1999), Race and gender in the labor market. In: Ashenfelter, O.A., Card, D. (Eds.), Handbook of Labor Economics, Vol. 3C. Elsevier, Amsterdam, pp. 3143–3259. Arrow, K.J. (1973), The theory of discrimination. In: Ashenfelter, O.A., Rees, A. (Eds.), Discrimination in Labor Markets. Princeton University Press, Princeton, NJ, pp. 3–33. Beach, C.M., Worswick, C. (1993), Is there a double-negative effect on the earnings of immigrant women? Canadian Public Policy 19, 36–53. Becker, G.S. (1971), The Economics of Discrimination, second ed. University of Chicago Press, Chicago. Blinder, A.S. (1973), Wage discrimination: reduced form and structural estimates. Journal of Human Resources 2, 8–22. Bodvarsson, O¨.B., Fuess, S.M., Jr. (2008), Immigration of highly skilled professionals: discrimination in professional baseball. Unpublished paper. Bodvarsson, O¨.B., Sessions, J.G. (2008), The measurement of racial discrimination in pay between job categories: theory and test. IZA Discussion Paper 3748. Borjas, G.J. (1985), Assimilation, changes in cohort quality, and the earnings of immigrants. Journal of Labor Economics 3, 463–489. Borjas, G.J. (1987), Self-selection and the earnings of immigrants. American Economic Review 77, 531–553. Bucci, G.A., Tenorio, R. (1997), Immigrant-native wage differentials and immigration reform. Review of Development Economics 1, 305–323. Cain, G.G. (1986), The economic analysis of labor market discrimination: a survey. In: Ashenfelter, O.A., Layard, R. (Eds.), Handbook of Labor Economics, Vol. 3A. Elsevier, Amsterdan, pp. 693–785. Chiswick, B.R. (1978), The effect of Americanization on the earnings of foreign-born men. Journal of Political Economy 86, 897–921. Chiswick, B.R. (1980), The earnings of white and coloured male immigrants in Britain. Economica 47, 81–87. Cotton, J. (1988), On the decomposition of wage differentials. Review of Economics and Statistics 70, 236–243. Daneshvary, N. (1993), Earnings differentials between natives and immigrants with a college degree. American Economist 37, 46–52. Daneshvary, N., Weber, W.L. (1991), Sources of wage differentials between native and immigrant workers: a regional analysis. Review of Regional Studies 21, 119–135. Denny, K.J., Harmon, C.P., Roche, M.J. (2000), The distribution of discrimination in immigrant earnings – evidence from Britain 1974–93. IFS Working Paper, Institute for Fiscal Studies, London, UK. Diewert, W.E. (1971), An application of the Shephard duality theorem: a generalized Leontief production function. Journal of Political Economy 79 (May/June), 481–507.

Nationality Discrimination in the Labor Market

267

Fujii, E.T., Mak, J. (1983), The determinants of income of nativeand foreign-born men in a multiracial society. Applied Economics 15, 759–776. Gabriel, P.E., Schmitz, S. (1987), The relative earnings of native and immigrant males in the United States. Quarterly Review of Economics and Business 27, 91–101. Goyette, K., Xie, Y. (1999), The intersection of immigration and gender: labor force outcomes of immigrant women scientists. Social Science Quarterly 80, 314–395. Haig, B.D. (1980), Earnings of migrants in Australia. Journal of Industrial Relations 22, 264–274. Hayfron, J.E. (2002), Panel estimates of the earnings gap in Norway: do female immigrants experience a double earnings penalty? Applied Economics 34, 1441–1452. Kahn, L.M. (1991), Customer discrimination and afﬁrmative action. Economic Inquiry 29 (July), 555–571. Kahn, L.M. (2000), A level playing ﬁeld? Sports and discrimination. In: Kern, W.S. (Ed.), The Economics of Sports. W.E. Upjohn Institute, Kalamazoo, MI. Kee, P. (1995), Native-immigrant wage differentials in the Netherlands: discrimination? Oxford Economic Papers 47, 302–317. Lang, G. (2005), The difference between wages and wage potentials: earnings disadvantages of immigrants in Germany. Journal of Economic Inequality 3, 21–42. Meng, R. (1987), The earnings of Canadian immigrant and native-born males. Applied Economics 19, 1107–1119. Mu¨ller, T. (2003), Migration, unemployment and discrimination. European Economic Review 47, 409–427. Neumark, D. (1988), Employers’ discriminatory behavior and the estimation of wage discrimination. The Journal of Human Resources 23, 279–295. Nielsen, H.S., Rosholm, N., Smith, N., Husted, L. (2004), Qualiﬁcations, discrimination, or assimilation? An extended framework for analysing immigrant wage gaps. Empirical Economics 29, 855–883. Oaxaca, R.L. (1973), Male–female wage differentials in urban labor markets. International Economic Review 14, 693–709. Oaxaca, R.L., Ransom, M.R. (1994), On discrimination and the decomposition of wage differentials. Journal of Econometrics 61, 5–21. Ottaviano, G.I.P., Peri, G. (2005), ‘Rethinking the gains from immigration: theory and evidence from the U.S. National Bureau of Economic Research (NBER) Working Paper 11672 (http://www.nber.org/papers/ w11672). Pedace, R. (2008), Earnings, performance, and nationality discrimination in a highly competitive labor market: an analysis of the English professional soccer league. Journal of Sports Economics 9, 115–140.

268

+ B. Bodvarsson and John G. Sessions Orn

Reimers, C.W. (1983), Labor market discrimination against Hispanic and black men. Review of Economics and Statistics 65, 570–579. Reimers, C.W. (1984), The wage structure of Hispanic men: implications for policy. Social Science Quarterly 65, 401–416. Shamsuddin, A.F.M. (1998), The double-negative effect on the earnings of foreign-born females in Canada. Applied Economics 30, 1187–1201. Tandon, B.B. (1978), Earning differentials among native born and foreign born residents of Toronto. International Migration Review 12, 406–410. Tienda, M., Niedert, L.J. (1980), Segmented markets and earnings inequality of native and immigrant Hispanics in the United States. Proceedings of the American Statistical Association, Social Statistics Section 72–81. Tran-Nam, B., Neville, J.W. (1988), The effects of birthplace on male earnings in Australia. Australian Economic Papers 27, 83–101. Wilson, D.P., Ying, Y.-H. (2003), Nationality preferences for labour in the international football industry. Applied Economics 35, 1551–1559.

CHAPTER 11

Culture, Investment in Language and Earnings Erez Siniver School of Economics, The College of Management Academic Studies, Rishon Letzion, 75190, Israel E-mail address: [email protected]; [email protected]

Abstract Cross-sectional studies have shown that immigrants’ earnings tend to rise faster than those of comparable natives. One reason for this is the immigrant’s acquisition of proﬁciency in the host country’s native language. Immigrants can improve their knowledge of the host country language either by interacting with native speakers or by taking a formal language course. Focusing on Jewish immigrants in Israel from the Former Soviet Union (FSU), the present chapter examines an immigrant’s decision to invest in learning Hebrew by participating in a formal government-sponsored course. The chapter estimates immigrants’ lifetime earnings in order to identify those immigrants with the highest potential beneﬁt from taking a Hebrew course and to determine whether they are more likely to attend one. The chapter ﬁnds that immigrants respond to pecuniary incentives to acquire the language of the host country and that this is particularly true for immigrants with 13þ years of schooling. Keywords: Russian immigrants, wage differences Jel classiﬁcations: J31, F22, R23

1. Introduction About 500,000 immigrants arrived in Israel from the Former Soviet Union (FSU) during the period 1989–1993. Their high level of education and experience in high-skilled occupations differentiate them from other immigrant populations. Thus, 40 percent of arriving FSU immigrants had 13–15 years of schooling as compared to only 13 percent of the Israeli population. (The percentage with 16 or more years of schooling, i.e., about 10 percent, was similar.) About 69 percent of them had worked in the FSU and of those about 36 percent had been employed in scientiﬁc and Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008017

r 2010 by Emerald Group Publishing Limited. All rights reserved

270

Erez Siniver

academic occupations and about 40 percent had worked in professional and technical occupations or as skilled workers in industry, transportation, etc. Only 1.7 percent had worked as unskilled workers (The Israel Ministry of Absorption 1995). Hebrew was not the mother tongue of most FSU immigrants. Undoubtedly, this restricted their earning opportunities relative to comparable natives. In order to assist in their absorption, the Israeli government ﬁnanced the cost of basic Hebrew language courses for immigrants, which are taught in a special language school called ulpan.1 Although the course is given free of charge, participation is costly in terms of time and job interruption (since the course lasts for 12 months). The Survey of Recent Immigrants (SRI) data (described below) indicates that immigrants who have participated in ulpan can speak and write Hebrew ﬂuently. In addition, there is a low dropout rate from ulpan since apparently a one-time decision is made whether or not to attend. This chapter estimates the relationship between immigrants’ demographic characteristics and their incentive to participate in ulpan and calculates their lifetime earnings in order to determine who potentially beneﬁts the most from the course and whether they are the ones who indeed choose to attend. This study contributes to the literature by examining whether immigrants respond to pecuniary incentives to acquire skills in the dominant language. Thus, we show a connection between the acquisition of local culture – assimilation – and pecuniary incentives. In other words, if people are lifetime income maximizers, they will invest in education or in improving their knowledge in the host country’s native language up to the point that the beneﬁt in terms of expected discounted increased income equals the direct costs and forgone earnings. In a discrete choice such as we examine, they will participate in ulpan if the difference between the present discounted value of lifetime earnings with and without participating in ulpan exceeds the direct cost of the ulpan. In some ways, looking at the decision of the immigrants from FSU to participate or not participate in ulpan is particularly interesting. It is interesting because we have a clear path to obtain the skills without formal education, and we can measure the speed at which the skills are obtained through employment. In principle it has not been

1 An ulpan is a school for the intensive study of Hebrew. It is designed to teach the basic language skills of conversation, writing and comprehension to adult immigrants in Israel. Most of them also provide basic knowledge in Israeli culture, history and geography. The primary purpose of the ulpan is to facilitate integration into Israeli society. Ulpan is offered free of charge to recent immigrants in Israel. Many are equipped with modern audio-visual teaching aids. Since their establishment in Jerusalem in 1949, they have produced more than 1.3 million graduates.

Culture, Investment in Language and Earnings

271

applied either to acquiring the language of the host country or to acquiring other skills.

2. Review of the literature The research reported in the literature indicates a positive relationship between an immigrant’s knowledge of the native language in the host country and his earnings. Chiswick and Repetto (2001) and Chiswick (1998) examined Israeli census data and concluded that Hebrew language skills increase with level of schooling and duration in Israel and that earnings increase with the acquisition of both Hebrew writing and Hebrew speaking skills. Other studies (such as Beenstock, 1996; Berman et al., 2003; Beenstock et al., 2001) also found that earnings of immigrants in Israel increase with proﬁciency in Hebrew. Cohen-Goldner and Eckstein (2010) estimate the beneﬁt from government-provided training courses for female FSU immigrants in Israel. They found that training has no signiﬁcant impact on mean offered earnings in blue-collar occupations, while it increases them in white-collar occupations by 19 percent. Training also increases job-offer rates in both types of occupation. Eckstein and Weiss (2004) found that upon arrival in Israel FSU immigrants receive no return on imported skills and that the immigrant’s return on schooling converges to 0.027, which is substantially lower than that for natives (0.069). They also found that during the 10 years following arrival, the earnings of high-skilled immigrants grow at an annual rate of 8 percent; however, the average earnings of immigrants do not converge to those of comparable natives, which is primarily due to the low return on imported skills. Weiss et al. (2003) also found that FSU immigrants can expect their lifetime earnings to be 57 percent less than comparable natives, primarily as a result of the only gradual adaptation of imported schooling and experience to the local labor market. Chiswick and Miller (2007a, b) found that based on data from the 2000 US census, there is little reward for the acquisition of English language skills within one’s occupation and that the reward is much more signiﬁcant for moving to an occupation in which English language skills are more important. They also found that proﬁciency in English is a key factor in determining access to high-paying occupations. Espenshade and Fu (1997) found that in the United States additional years of labor market experience improve the English language skills of most immigrants and that they are further enhanced if immigrants obtain some portion of their formal education in the United States. Grenier (1984) found that language skills have a major effect on wages and that they explain a signiﬁcant proportion of the mean wage differential between non-Hispanic and Hispanic workers. Kossoudji (1988) found that not being able to speak English imposes a real cost on some immigrant workers, both by reducing

272

Erez Siniver

observed earnings and by altering occupational opportunities. Based on data for West Germany, Dustmann (1994) found that language abilities, especially writing proﬁciency, considerably improve the earning position of immigrants. Other studies (such as Chiswick, 1991; Chiswick and Miller, 1994, 2002; McManus et al., 1983; Tainer, 1988; Borjas, 1985, 1987, 1989) have also found that language has an important impact on the earnings of immigrant workers. 3. Data The data conducted by the Israeli Center Burean of Statistics (ICBS). The data was obtained from the SRI2 and was based on a sample of nearly 1,200 households of recent immigrants from the FSU. These households contain 2,715 individuals aged 16–65. The SRI provides information on personal characteristics such as gender, marital status, age, years of schooling, and year of immigration to Israel; employment and current earnings (Table 1 shows that 1,359 immigrants were employed and 1,356 were unemployed); and ability to speak and write Hebrew. Respondents were asked to classify their ability to speak Hebrew as ‘‘ﬂuent,’’ ‘‘with difﬁcultly,’’ or ‘‘cannot speak Hebrew at all,’’ which were coded as 1, 2, and 3, respectively. The ability to write Hebrew was treated in a similar manner. Table 1 shows that the average reported ability to speak Hebrew was 1.44 for the entire sample, 1.4 for immigrants who were employed, and 1.48 for immigrants who were unemployed. The average reported ability to write Hebrew was 1.79 for the entire sample, 1.79 for employed immigrants, and 1.79 for unemployed immigrants. There is some difﬁculty with self-reporting since the result is subjective and does not refer to a common proﬁciency standard. Thus, for example, it is unclear whether respondents have taken their accent into consideration and which reference group they are using in their assessment. Of critical interest for this chapter is the survey question, ‘‘Have you participated in a basic Hebrew language course in ulpan?’’ The immigrants’ responses make it possible to estimate the probability of an immigrant attaining a degree of ﬂuency without language training. The SRI data shows that FSU immigrants who had taken a Hebrew course could speak and write Hebrew ﬂuently. 4. Method The main objective of the chapter is to estimate the lifetime earnings of immigrants with different characteristics (i.e., gender, marital status, education, and age) in order to determine who potentially beneﬁts most from participating in ulpan and whether they are more likely to actually 2

Israel Central Bureau of Statistics, 1993. Monthly Bulletin of Statistics, April 1994. Jerusalem.

273

Culture, Investment in Language and Earnings

Table 1.

Descriptive statistics

Russian immigrants aged 16–65

Total Age Years of schooling Experience Years since migration Currently married Male Speak Hebrew Write Hebrew Wage

Russian immigrants aged 16–65 with 13þ years of schooling

Employed Unemployed Total

38.83 (0.25) 13.66 (0.05) 19.18 (0.25) 25.28 (0.07) 0.82 (0.01) 0.53 (0.01) 1.44 (0.01) 1.79 (0.02) –

38.98 (0.33) 13.81 (0.07) 19.17 (0.33) 25.55 (0.06) 0.85 (0.01) 0.52 (0.01) 1.40 (0.02) 1.79 (0.02) 1949.36 (33.74) 0.0506 – 0.4994 – 0.158 0.1339 0.8435 0.8661

Employed Unemployed Participated in ulpan Did not participate in ulpan Observations 2715

1359

38.69 (0.38) 13.51 (0.08) 19.18 (0.37) 25.00 (0.13) 0.79 (0.01) 0.54 (0.01) 1.48 (0.02) 1.79 (0.02) – – – 0.1792 0.8208 1356

Employed Unemployed

41.30 40.62 (0.27) (0.35) 15.35 15.34 (0.04) (0.05) 19.95 19.27 (0.26) (0.34) 25.22 25.53 (0.09) (0.08) 0.91 0.92 (0.01) (0.01) 0.54 0.51 (0.01) (0.02) 1.40 1.36 (0.01) (0.02) 1.73 1.72 (0.02) (0.02) – 1977.61 (42.10) 0.5231 – 0.4769 – 0.192 0.161 0.8082 0.839

42.04 (0.41) 15.36 (0.06) 20.68 (0.40) 24.88 (0.17) 0.90 (0.01) 0.57 (0.02) 1.44 (0.02) 1.74 (0.03) –

1758

839

919

– – 0.2253 0.7747

Note: Data taken from the SRI survey conducted by the Central Bureau of Statistics in Israel. The survey, conducted from January–April 1994, provides demographic data on the households of FSU immigrants. The data is based on the entire SRI sample. Standard errors appear in parentheses. * Signiﬁcance at the 0.01 level.

participate. Thus, we will estimate the present value3 (PV) of lifetime earnings for FSU immigrants who have participated in ulpan and for those who have not. We estimate the probabilities of achieving proﬁciency in Hebrew within a given period of time without participating in ulpan for immigrants with different demographic characteristics. To do so, we estimated two ordered probit regressions, where the dependent variables for the ﬁrst- and secondorder probits are the ability to speak Hebrew and the ability to write

3

The current value of a sum of money to be received in the future given a speciﬁed discount rate. The higher the discount rate, the lower will be the present value of a future cash ﬂow.

274

Erez Siniver

Hebrew. The independent variables include marital status, gender, education, and duration in Israel. We are interested in the employment probability as a function of Hebrew knowledge, the situation in the labor market and other factors. In order to calculate this probability, we estimated a probit regression using employment as the dependent variable. The independent variables included marital status, gender, education, experience, duration in Israel, ability to speak Hebrew, and ability to write Hebrew. There is evidence that immigrants from the FSU who improve their Hebrew language proﬁciency not only improve their earnings, but also the probability of ﬁnding a job (see Cohen-Goldner and Eckstein, 2009). This implies that OLS estimates may be biased. In order to estimate the earnings equation while controlling for self-selection, we use both the inverse Mill’s ratios in a standard two-stage Heckman model and maximum likelihood estimation with Newton–Raphson maximization. In Section 5 in Results, we estimate the present value of earnings with and without participation in ulpan for each immigrant using the estimators obtained in a, b, and c above. In order to estimate the demographic characteristics of immigrants from the FSU who have participated in ulpan, a probit regression was estimated in which the dependent variable is participating in ulpan (a dichotomous variable that equals one if the immigrant participated in ulpan and zero otherwise) and the independent variables are gender, marital status, education, and age. In order to determine whether the earnings maximization model is able to accurately predict which immigrants will participate in ulpan, we estimated who would potentially beneﬁt the most from participating in ulpan and whether they are the ones more likely to actually participate. To this end, we created an independent variable for beneﬁt (a dichotomous variable that equals one for immigrants whose PV2 =PV1 is in the top 15.8 percent for all immigrants and zero otherwise, where PV2 is the PV of lifetime earnings for immigrants who have not participated in ulpan and 15.8 percent is the percentage of immigrants who have participated in ulpan). A probit regression was estimated with ‘‘participated in ulpan’’ as the dependent variable and with the following independent variables: gender, marital status, education, age, and beneﬁt. If the decision to participate in ulpan is driven by beneﬁt, we would expect only its coefﬁcient to be signiﬁcant and those of the other independent variables not to be.

5. Results We ﬁrst examine the probability of achieving proﬁciency in Hebrew without participating in ulpan. Immigrants can choose between two

Culture, Investment in Language and Earnings

275

different ways of improving their language proﬁciency: participation in ulpan or interacting with native speakers. The SRI data shows that immigrants from the FSU who participated in ulpan could speak and write Hebrew ﬂuently. In this section, we determine at what pace immigrants who have not participated in ulpan learn to speak and write Hebrew ﬂuently. In order to estimate the probabilities of achieving proﬁciency in Hebrew within a given period of time for immigrants with different characteristics, two ordered probit estimations were estimated. The dependent variables for the ﬁrst- and second-order probit regressions are the ability to speak Hebrew and the ability to write Hebrew, respectively. Immigrants were asked to classify their ability to speak/write Hebrew as ‘‘ﬂuent,’’ ‘‘with difﬁcultly,’’ or ‘‘cannot speak/write Hebrew at all,’’ which were coded as 1, 2, and 3, respectively. The independent variables are marital status (a dichotomous variable that equals one for married immigrants and zero otherwise), gender (a dichotomous variable that equals one for males and zero for females), education (years of schooling), duration in Israel (in months), and age. Column 1 of Table 2 shows the results for ability to speak Hebrew as the dependent variable, while Column 2 shows the results for ability to write Hebrew. The coefﬁcient for duration in Israel in both columns is negative and signiﬁcant, implying that immigrants can improve their Hebrew language skills by interacting with native Hebrew speakers. The coefﬁcient for the same variable squared is signiﬁcant and positive in both columns. The coefﬁcients for age are positive (0.113 in Column 1 and 0.096 in Column 2) and signiﬁcant implying that the probability of young FSU immigrants learning to speak or write Hebrew ﬂuently without attending ulpan is higher than for older immigrants, other things being equal. (The coefﬁcients for age squared were not signiﬁcant and are not shown.) The coefﬁcients for education are 0.262 in Column 1 and 0.309 in Column 2 and are signiﬁcant (t-test ¼ 15.06 in Column 1 and 17.83 in Column 2), implying that more-educated immigrants have a higher probability of learning Hebrew than less-educated ones. The coefﬁcients for duration in Israel*education (not shown) were not signiﬁcant implying that the duration/ability to speak Hebrew and duration/ability to write Hebrew proﬁles for more-educated immigrants are parallel to those of lesseducated immigrants. The coefﬁcients for marital status are 0.389 in Column 1 and 1.033 in Column 2 and are both signiﬁcant, implying that single immigrants have a higher probability of learning Hebrew than married ones. The probability that male immigrants will speak Hebrew ﬂuently is higher than for female immigrants; however, the situation is reversed for the ability to write Hebrew ﬂuently. The coefﬁcients for gender (male ¼ 1) are 0.056 in Column 1 and 0.109 in Column 2, which implies that male immigrants have a higher probability of learning to speak Hebrew but that females have a higher probability of learning to write Hebrew.

276

Table 2.

Erez Siniver

The probability of achieving proﬁciency in Hebrew without participating in ulpan

Gender Marital status Education Age Duration in Israel (months) Duration in Israel^2 Cutoff1 Cutoff2 Observations Log likelihood

Dependent variable – ability to speak Hebrew

Dependent variable – ability to write Hebrew

(1)

(2)

0.056* (0.022) 0.389* (0.192) 0.262* (0.017) 0.113* (0.005) 0.206* (0.032) 0.002* (0.0008) 2.0139 (0.463) 0.742 (0.453) 2715 1668.546

0.109* (0.041) 1.033* (0.147) 0.309* (0.017) 0.096* (0.004) 0.176* (0.034) 0.003* (0.0008) 2.812 (0.471) 0.707 (0.464) 2715 2258.244

Note: Standard errors appear in parentheses. Source: SRI. Ability to speak/write Hebrew are classiﬁed as ‘‘ﬂuent,’’ ‘‘with difﬁculty,’’ and ‘‘cannot speak/ write Hebrew at all,’’ which are coded as 1, 2, and 3, respectively. Gender: A dichotomous variable with 1 ¼ male and 0 ¼ female. Marital status: A dichotomous variable with 1 ¼ married immigrants and 0 ¼ otherwise. * Signiﬁcance at 0.01 level.

The results obtained in this chapter are very similar to those obtained in Chiswick (1998). Chiswick used the 1983 Censuses of Israel to analyze Hebrew speaking skills and the effect of Hebrew ﬂuency on the earnings of adult male immigrants. The ﬁndings show that Hebrew ﬂuency increases with a longer duration in Israel, but decreases with age at migration. Immigrants from the FSU who improve their proﬁciency in Hebrew increase not only their earnings but also the probability of ﬁnding a job (see Cohen-Goldner and Eckstein, 2008). To calculate an immigrant’s probability of being employed, a probit regression was estimated with employment as the dependent variable (a dichotomous variable that equals one for employed and zero for unemployed) and with the following independent variables: marital status, education, experience, experience squared, duration in Israel, ability to speak Hebrew, and ability to write Hebrew. Table 3 shows that the coefﬁcient for duration in Israel is signiﬁcant and positive, implying that the probability of ﬁnding a job

Culture, Investment in Language and Earnings

277

Table 3. Probability of employment Dependent variable – employment Gender Marital status Education Experience Experience^2 Duration in Israel (months) Ability to Write2 Ability to Write3 Ability to Speak2 Ability to Speak3 Constant Observations Log likelihood

0.120 (0.078) 0.379* (0.140) 0.002 (0.016) 0.023* (0.010) 0.0006 (0.0003) 0.003* (0.0009) 0.065 (0.107) 0.085 (0.147) 0.179* (0.065) 0.515* (0.206) 1.209 (0.365) 2715 1851.465

Note: Standard errors appear in parentheses. Source: SRI. Employment: 1 ¼ employed, 0 ¼ unemployed. Gender: A dichotomous variable which equals one for males and zero for females. Marital status: A dichotomous variable which equals one for married immigrants and zero otherwise. Ability to speak/write Hebrew are classiﬁed as ‘‘ﬂuent,’’ ‘‘with difﬁculty,’’ and ‘‘cannot speak/ write Hebrew at all,’’ which are coded as 1, 2, and 3, respectively. * Signiﬁcance at 0.01 level.

increases with duration in Israel. The coefﬁcients for ability to speak Hebrew are signiﬁcant and negative (0.179 for speak2 and 0.515 for speak3), implying that immigrants who improve their ability to speak Hebrew also improve their probability of being employed. The coefﬁcients for ability to write Hebrew are positive (0.065 for write2 and 0.085 for write3); however, they are not signiﬁcant, implying that immigrants who improve their ability to write Hebrew do not improve their probability of being employed. The coefﬁcients of duration in Israel*education, ability to speak*education and ability to write*education were not signiﬁcant and are not shown. The results obtained in this chapter are very similar to those obtained in Cohen-Goldner and Eckstein (2010).

278

Erez Siniver

Cohen-Golddner and Eckstein examined the effect of vocational training on the earnings of female immigrants and their prospect of ﬁnding a job. They concluded that for women employed in white-collar professions, vocational training increases both earnings and the prospect of ﬁnding a job. With regard to women in blue-collar professions, vocational training has no effect on earnings but does increase the prospect of ﬁnding a job. There is a large body of evidence that ﬂuency in the language of the host country has a positive effect on an immigrant’s earnings. Indeed, Table 3 shows that immigrants who improve their ability to speak Hebrew also improve their probability of ﬁnding a job. There are two sources for the observed outcome: the ﬁrst is, of course, the causal relationship between the variable of interest (i.e., the immigrant’s wage) and the regressors, and the second is self-selection since the observations were not selected randomly. In other words, the data, which was ﬁltered out in the selection process and is therefore unobservable, will differ from the nonselected data. As a result, the OLS estimators may be biased. In our case, it may be that immigrants with low potential earning power chose not to participate in the workforce, which will create an upward bias in the OLS wage equation. Since our primary interest is in the casual effect, the selection problem must be taken into account and therefore OLS is not appropriate. In order to estimate the earnings equation while controlling for selfselection, we used two alternative methods: (1) the inverse Mill’s ratios in a standard two-stage Heckman model and (2) maximum likelihood estimation with Newton–Raphson maximization. Column 1 of Table 4 presents the results for the Heckman model and Column 2 for maximum likelihood estimation. The self-selection effect in both models is signiﬁcantly different from zero, which justiﬁes the use of one or the other. The coefﬁcients for ability to speak2 and ability to speak3 are negative and signiﬁcant: 0.04 for speak2 and 0.149 for speak3 in the Heckman model and –0.038 for speak2 and 0.147 for speak3 in the maximum likelihood model. Thus, if immigrants improve their ability to speak Hebrew, they also improve their earnings. However, the coefﬁcients for ability to write2 and ability to write3 are negative but not signiﬁcant. The coefﬁcient for education is 0.06 and signiﬁcant, implying that the payoff on education (in terms of earnings) is 6 percent per year of schooling. The coefﬁcients for duration in Israel are positive and signiﬁcant. The coefﬁcients for ability to speak Hebrew*education and ability to write Hebrew*education were not signiﬁcant and are not shown. The entire literature conﬁrm that immigrants with good speaking and writing ability do better in the labor market in terms of employment and earnings than immigrants who cannot speak and write well the host country’s language. However, it is hard to compare across the studies the ‘‘ﬂuency penalty.’’ The reasons are (1) differences in methodologies, (2) self-assessed language proﬁciency, and (3) differences in cohort quality.

Culture, Investment in Language and Earnings

Table 4.

Earning equation estimates Two-step Heckman (1)

Gender Marital status Education Experience Experience^2 Duration in Israel (months) Ability to Write2 Ability to Write3 Ability to Speak2 Ability to Speak3

Gender Marital status Education Experience Experience^2 Duration in Israel (months) Ability to Write2 Ability to Write3 Ability to Speak2 Ability to Speak3 Sigma

279

Maximum likelihood estimation with Newton– Raphson maximization (2)

Dependent variable – employment Probit selection equation 0.074 0.077 (0.049) (0.049) 0.235* 0.236* (0.087) (0.087) 0.0009 0.0009 (0.010) (0.010) 0.014* 0.018* (0.009) (0.009) 0.0004* 0.0004* (0.0002) (0.0002) 0.002* 0.002* (0.0006) (0.0006) 0.102 0.081 (0.067) (0.067) 0.140 0.103 (0.092) (0.092) 0.113* 0.113* (0.042) (0.032) 0.324* 0.308* (0.128) (0.128) Dependent variable – Ln wage Probit selection equation 0.009 0.011 (0.237) (0.032) 0.072 0.068 (0.772) (0.059) 0.06* 0.06* (0.008) (0.006) 0.009* 0.009* (0.003) (0.004) 0.0002 0.0002 (0.001) (0.0001) 0.014* 0.013* (0.065) (0.006) 0.017 0.015 (0.317) (0.043) 0.044 0.038 (0.442) (0.058) 0.038* 0.04* (0.003) (0.004) 0.149* 0.147* (0.050) (0.057) 0.619* 0.608* (0.033) (0.036)

280

Erez Siniver

Table 4. (Continued ) Two-step Heckman (1)

Rho Lambda Observations

Log likelihood

0.655* (0.090) 0.405* (0.080) 2715 observations; 1356 censored and 1359 observed 2912.632

Maximum likelihood estimation with Newton– Raphson maximization (2) 0.623* (0.106)

2715 observations; 1356 censored and 1359 observed 2912.574

Note: Standard errors appear in parentheses. Source: SRI. Ability to speak/write Hebrew are classiﬁed as ‘‘ﬂuent,’’ ‘‘with difﬁculty,’’ and ‘‘cannot speak/ write Hebrew at all,’’ which are coded as 1, 2, and 3, respectively. Gender: A dichotomous variable with 1 ¼ male and 0 ¼ female. Marital status: A dichotomous variable with 1 ¼ married immigrants and 0 ¼ otherwise. * Signiﬁcance at 0.01 level.

We now discuss the relationship between an immigrant’s demographic characteristics (gender, years of schooling, age, and marital status) and the incentives for him to participate in ulpan. To start with, we estimated a probit regression, in which the dependent variable is studying the native language (a dichotomous variable that equals one if the immigrant participated in ulpan and zero otherwise). The independent variables are: gender (1 ¼ male, 0 ¼ female), marital status (1 ¼ married, 0 ¼ single), education (years of schooling), age, and age squared. Column 1 of Table 5 presents the ﬁndings, which indicate that participating in ulpan is most common among the following groups: (1) males (the coefﬁcient for gender is 0.0136 and signiﬁcant), (2) moreeducated immigrants (the coefﬁcient for years of schooling is 0.141 and signiﬁcant), (3) older immigrants (the coefﬁcient for age is 0.018 and signiﬁcant), and (4) single immigrants (the coefﬁcient for marital status is 0.533 and signiﬁcant). Column 2 of Table 5 presents the ﬁndings when the education variable takes a value of one for 13þ years of schooling and zero for less. According to the results, participating in ulpan is more common among immigrants with 13þ years of schooling (the coefﬁcient for education is 0.845 and signiﬁcant). We estimate the effect of participation in ulpan on the PV of lifetime earnings for FSU immigrants. If the increase in earnings as a result of the ulpan is higher than foregone earnings, participation in ulpan is worthwhile.

4.095* (0.557) 2715 1138.640

0.0136* (0.050) 0.533* (0.201) 0.141* (0.022) 0.018* (0.0033) 0.00005* (0.00004) –

2.782* (0.565) 2715 1141.268

0.130* (0.060) 0.575* (0.206) 0.845* (0.141) 0.022* (0.0033) 0.00004* (0.00004) –

(2)

(1)

118.536* (20.487) 2715

3.526* (1.040) 29.833* (5.321) 29.951* (5.211) 12.749* (2.252) 0.069* (0.013) –

r ¼ 1% (3)

123.415* (21.497) 2715

3.528* (1.071) 30.819* (5.545) 31.352* (5.516) 13.403* (2.399) 0.074* (0.014) –

r ¼ 2% (4)

Dependent variable – beneﬁt

115.129* (19.765) 2715

3.543* (1.013) 29.169* (5.161) 28.961* (5.001) 12.282* (2.152) 0.067* (0.012) –

r ¼ 4% (5)

120.415* (21.550) 2715

3.728* (1.071) 31.819* (5.545) 32.452* (5.616) 13.8403* (2.499) 0.084* (0.014) –

r ¼ 6% (6)

116.129* (20.765) 2715

3.643* (1.013) 30.179* (5.261) 29.971* (5.011) 13.280* (2.252) 0.077* (0.015) –

r ¼ 10% (7)

0.130 (0.100) 0.444 (0.428) 0.135* (0.022) 0.037 (0.038) 0.00015 (0.0004) 0.267* (0.068) 4.293* (0.595) 2715 1138.142

r ¼ 4% (8)

Dependent variable – participated in ulpan

Test for beneﬁt as a single motivation

Note: Standard errors appear in parentheses. Source: SRI. Beneﬁt: Equals one for immigrants whose (PV2)/(PV1) is in the top 15.8 percent for all immigrants and zero otherwise. Participated in ulpan – equals one for immigrants who participated in ulpan and zero otherwise. Gender: A dichotomous variable with 1 ¼ male and 0 ¼ female. Marital status: A dichotomous variable with 1 ¼ married immigrants and 0 ¼ otherwise. * Signiﬁcance at 0.01 level.

Observations Log likelihood

Constant

Beneﬁt

Age^2

Age

Education

Marital status

Gender

Dependent variable – participated in ulpan

Dependent variable – participated in ulpan

Beneﬁt gained from ulpan

Demographic characteristics of immigrants participating in ulpan and the beneﬁt they derive

Demographic characteristics of immigrants participating in ulpan

Table 5.

Culture, Investment in Language and Earnings 281

282

Erez Siniver

The PV of lifetime earnings for immigrants who have not participated in ulpan is calculated as: P P 65 12 X ½Pt ðeÞ 3i¼1 3j¼1 Pt ðS ¼ i; W ¼ jÞ E t ðS ¼ i; W ¼ jÞ (1) PV1 ¼ ð1 þ rÞt t¼0 where S is the ability to speak Hebrew, W is the ability to write Hebrew, and Pt(e) is the probability that the immigrant is employed. Pt(e) was estimated using the probit regression (see Table 3). After each speciﬁed number of months, Pt(S ¼ i, W ¼ j) is the probability that an FSU immigrant will have achieved S(i) and W(j). These probabilities were estimated using the ordered probit equation described above (see Table 2). Et(S ¼ i, W ¼ j) is earnings given that the immigrants’ ability to speak Hebrew is i and his ability to write Hebrew is j. Expected earnings were estimated by using the maximum likelihood model (see Table 4), where r is the discount rate. The PV of lifetime earnings for immigrants who have participated in a 12-month ulpan is calculated as: PV2 ¼

65 12 X t¼12

½Pt ðeÞ E t ðS ¼ 1; W ¼ 1Þ ð1 þ rÞt

(2)

The data indicates that immigrants who have participated in ulpan can speak and write Hebrew ﬂuently (i.e., S ¼ 1, W ¼ 1). If PV2 PV1WC, where C is the cost of participating in ulpan, then participation is worthwhile. In order to determine whether actual investment reﬂects incentives, the beneﬁt/cost ratio was ﬁrst estimated (with beneﬁt in this case deﬁned as PV2 =PV1 ) for an FSU immigrant, followed by the calculation of a probit equation to determine whether the beneﬁt/cost ratio exceeds some threshold. The threshold was determined according to the actual proportion of immigrants who chose to participate in ulpan (15.8 percent). Thus, a probit regression was estimated in which the dependent variable is beneﬁt (a dichotomous variable that equals one for immigrants whose PV2 =PV1 is in the top 15.8 percent for immigrants and zero otherwise). The independent variables include: gender (1 ¼ male, 0 ¼ female), marital status (1 ¼ married, 0 ¼ single), education (years of schooling), age, and age squared. Columns 3–7 of Table 5 present the results for various discount rates. The results are as follows: (1) The coefﬁcients for gender are positive and signiﬁcant, implying that males beneﬁt more from participating in ulpan than females. (2) The coefﬁcients for marital status are negative and signiﬁcant, implying that single immigrants beneﬁt more from participating in ulpan than married ones. (3) The coefﬁcients for education are negative and signiﬁcant, meaning that less-educated immigrants beneﬁt more from participating in ulpan than more-educated ones.

Culture, Investment in Language and Earnings

283

(4) The coefﬁcients for age are positive and signiﬁcant, implying that older immigrants beneﬁt more from participating in ulpan than younger ones. In order to test whether the decision to participate in ulpan is driven by the beneﬁt derived, we added beneﬁt (a dichotomous variable, which equals one for immigrants whose PV2 =PV1 is in the top 15.8 percent for all the immigrants and zero otherwise) to the probit regression reported in Column 1 of Table 5. If the decision to participate in ulpan is driven by the beneﬁt derived by each immigrant, we expect to ﬁnd only the coefﬁcient for beneﬁt to be signiﬁcant. The results in Column 8 of Table 5 indeed show that the coefﬁcients for gender, marital status, and age are not signiﬁcant while the coefﬁcients for education and beneﬁt are. (Varying the discount rate did not signiﬁcantly change the results. A rate of 4 percent was used in the estimation.) Less-educated immigrants beneﬁt more from participating in ulpan although participation is more common among more-educated immigrants (see Table 5). Berman et al. (2003) found that ability bias accounts for the apparent return on Hebrew proﬁciency for less-educated FSU immigrants in Israel, but that the return appears to be genuine for the more-educated. If that is the case, then the calculated expected PV is too high for both participants and nonparticipants among the less-educated. This may imply that the beneﬁt of ulpan is overestimated for the less-educated. To deal with this problem, the entire analysis was repeated for only immigrants with 13þ years of schooling and the results are presented in Tables 6–9. Column 2 of Table 9 shows that male, older, and single immigrants tend to beneﬁt more from participating in ulpan (the coefﬁcient for gender is 0.354, for marital status is 38.234, and for age is 14.269 and all are signiﬁcant). Column 1 of Table 9 shows that male, older, and single immigrants, who beneﬁt more from ulpan, are more likely to participate (the coefﬁcient for gender is 0.093, for marital status is 0.451, and for age is 0.033, and all are signiﬁcant). Immigrants with more years of schooling beneﬁt more and are more likely to participate in ulpan. This can be seen in the value for the coefﬁcient of education in Column 2 of Table 9 (which is 33.759 and signiﬁcant) and in Column 1 of Table 9 (which is 0.063 and signiﬁcant). In order to test whether the decision to participate in ulpan is driven by the beneﬁt derived for immigrants with 13þ years of schooling we add beneﬁt (a dichotomous variable that is equal to one for an immigrant whose PV2/PV1 is in the top 19.2 percent4 for all immigrants and zero otherwise) to the probit regression presented in Column 1 of Table 9. Column 3 of Table 9 shows that while the coefﬁcients for gender, marital status, education, and age are not signiﬁcant the coefﬁcient for beneﬁt is. As in the earlier estimation, a 4 percent discount rate was used and varying it did not signiﬁcantly change the results.

4

The threshold was determined according to the actual proportion of immigrants with 13þ years of schooling who chose to participate in ulpan (19.2 percent).

284

Erez Siniver

Table 6. Probability of achieving proﬁciency in Hebrew without participating in ulpan immigrants with 13þ years of schooling Dependent variable – ability to speak Hebrew (1) 0.074* (0.012) Marital status 0.227* (0.0734) Education 0.237* (0.025) Age 0.113* (0.006) Duration in Israel 0.156* (0.039) (months) Duration in Israel^2 0.0008* (0.0002) Cutoff1 1.358 (0.026) Cutoff2 1.402 (0.125) Observation 1758 Log likelihood 1122.421

Gender

Dependent variable – ability to write Hebrew (2) 0.076* (0.035) 0.537* (0.205) 0.286* (0.023) 0.092* (0.005) 0.154* (0.033) 0.002* (0.0009) 2.952 (0.009) 0.748 (0.079) 1758 1569.945

Note: Standard errors appear in parentheses. Source: SRI. Gender: A dichotomous variable with 1 ¼ male and 0 ¼ female. Marital status: A dichotomous variable with 1 ¼ married immigrants and 0 ¼ otherwise. * Signiﬁcance at 0.01 level.

McManus (1985) found very similar results to those obtained in this chapter. McManus estimated the return on English language proﬁciency and argued that it is worthwhile for an immigrant to take a course in the language of the host country. He found that the difference between the present discounted value of lifetime earnings with and without the course exceeded the direct cost of the course. However, McManus did not distinguish between the various ways in which immigrants can improve their language proﬁciency, such as interacting with native speakers. 6. The effects of networks on the decision to invest in learning the host country’s language The results obtained in this chapter may be affected by the fact that we have not taken into account the size of the immigrant population in the individual’s community. From the immigrants’ perspective, leaving behind a familiar culture and adapting to a new culture can be challenging.

Culture, Investment in Language and Earnings

Table 7.

285

Probabilities of employment Immigrants with 13þ years of schooling Dependent variable – employment

Gender Marital status Education Experience Experience^2 Duration in Israel (months) Ability to Write2 Ability to Write3 Ability to Speak2 Ability to Speak3 Constant Observations Log likelihood

0.239* (0.096) 0.249 (0.192) 0.033 (0.045) 0.029* (0.009) 0.001* (0.0004) 0.004* (0.001) 0.127 (0.125) 0.164 (0.182) 0.054* (0.018) 0.343* (0.072) 1.236* (0.407) 1758 1195.336

Note: Standard errors appear in parentheses. Source: SRI. Employment: Equals one for employed and zero otherwise. Gender: A dichotomous variable with 1 ¼ male and 0 ¼ female. Marital status: A dichotomous variable with 1 ¼ married immigrants and 0 ¼ otherwise. * Signiﬁcance at 0.01 level.

Moreover, their experience can be more difﬁcult if they do not succeed in integrating into the host country’s culture and it is often associated with the failure to learn the host country’s language. The concern is that certain immigrant groups might live in enclaves and challenge the life-style as well as the formal or informal institutions of the host country’s culture. Moreover, it might reduce the immigrants’ incentive to invest in studying the host country’s language. The immigration literature has focused on the inﬂuence of ethnic enclaves on economic performance and language ﬂuency. Most studies ﬁnd a negative association between ethnic concentration and language proﬁciency. Card (1990) found that in Miami, which has a large Hispanic community (38.3 percent of the local workforce), the lack of ﬂuency in English among the Marials may have a smaller effect than for immigrants in cities with smaller immigrant

286

Erez Siniver

Table 8.

Immigrants with 13þ years of schooling

Maximum likelihood estimation

Gender Marital status Education Experience Experience^2 Duration in Israel (months) Ability to Write2 Ability to Write3 Ability to Speak2 Ability to Speak3 Constant

Gender Marital status Education Experience Experience^2 Duration in Israel (months) Ability to Write2 Ability to Write3 Ability to Speak2 Ability to Speak3 Constant

Dependent variable – employment Probit selection equation 0.154* (0.061) 0.161 (0.119) 0.008 (0.185) 0.019* (0.002) 0.0006* (0.0003) 0.002* (0.0007) 0.051 (0.078) 0.109 (0.115) 0.032* (0.006) 0.187* (0.069) 0.639 (0.374) Dependent variable – Ln wage Outcome equation 0.016 (0.039) 0.009 (0.079) 0.03* (0.012) 0.023* (0.008) 0.0005* (0.0002) 0.015* (0.004) 0.059* (0.059) 0.06 (0.072) 0.039* (0.003) 0.146* (0.011) 6.508* (0.306)

Culture, Investment in Language and Earnings

287

Table 8. (Continued ) Maximum likelihood estimation Sigma Rho Observations

Log likelihood

0.615* (0.042) 0.636* (0.121) 1758 observations; 839 censored and 919 observed. 1921.117

Note: Standard errors appear in parentheses. Source: SRI. Employment: Equals one for employed and zero otherwise. Gender: A dichotomous variable with 1 ¼ male and 0 ¼ female. Marital status: A dichotomous variable with 1 ¼ married immigrants and 0 ¼ otherwise. * Signiﬁcance at 0.01 level.

populations. Carliner (1981) also found that in Montreal, where the economic rewards for English ﬂuency in the workplace were high whereas the rewards for speaking French exist primarily outside the workplace, there were almost twice as many French bilinguals as French monolinguals. However, in Quebec City, where the rewards for English ﬂuency are smaller, the percentage of French bilinguals was lower and the percentage of English bilinguals was higher. Moreover, in English Canada, only 7 percent of English men spoke French, whereas over 85 percent of French men spoke English. Warman (2007), using the 1981–2001 Censuses in Canada, found that enclaves hindered language skills. Lazear (1999) found that immigrants are most likely to learn English when they live in a community with a smaller proportion of compatriots. This is explained by the fact that immigrants who live in ethnic enclaves obtain lower returns for knowing the native language than do immigrants who live in communities with only a small population of compatriots. McManus (1989) found that lower returns to English mean reduced incentives for individuals to acquire English, which means that fewer Hispanics will acquire English skills, thus contributing to long-lived enclaves. Danzer and Yamen (2010) used the German Socio-Economic Panel (GSOEP), which provides detailed immigrant characteristics, to ﬁnd that if the ethnic concentration increases by one standard deviation, the probability that a person is ﬂuent in German decreases by 2.6 percent. Siniver et al. (2009) determined whether, in addition to the level of ﬂuency in the local language, social involvement variables can explain the growth in income among immigrants in Israel. The study used data from a relatively recent survey carried out by the Jewish Agency in 2006 among 501 immigrants. As part of the survey, participants were asked a large number of subjective

288

Erez Siniver

Table 9.

Immigrants with 13þ years of schooling

Demographic characteristics of immigrants who participated in ulpan

Gender Marital status Education Age Age^2 Beneﬁt Constant Observations Log likelihood

Beneﬁt derived from ulpan

Test for beneﬁt as the exclusive motivation

Dependent variable – participated in ulpan

Dependent variable – beneﬁt

Dependent variable – participated in ulpan

(1)

r ¼ 4% (2)

r ¼ 4% (3)

*

0.354 (0.100) 38.234* (1.429) 33.759* (10.639) 14.269* (2.872) 0.770* (0.282) –

3.269* (0.931) 1758 846.653

2.698* (0.902) 1758

0.093 (0.011) 0.451* (0.248) 0.063* (0.032) 0.033* (0.004) 0.0009* (0.0002) –

*

0.101 (0.124) 0.417 (0.348) 0.019 (0.048) 0.065 (0.045) 0.0004 (0.0005) 0.609* (0.214) 3.263* (0.932) 1758 1195.336

Note: Standard errors appear in parentheses. Source: SRI. Beneﬁt: Equals one for immigrants whose (PV2)/(PV1) is in the top 19.20 percent for all immigrants and zero otherwise. Participated in ulpan: Equals one for an immigrant who participated in ulpan and zero otherwise. Gender: A dichotomous variable with 1 ¼ male and 0 ¼ female. Marital status: A dichotomous variable with 1 ¼ married immigrants and 0 ¼ otherwise. * Signiﬁcance at 0.01 level.

questions related to social involvement: their degree of assimilation in Israeli society, their degree of care and respect for Israeli culture, and the extent of their knowledge in Israeli and Jewish history. Similarly, they were asked how many of their friends and neighbors were immigrants. The ﬁndings indeed support the hypothesis that the level of social involvement is positively correlated with the level of income among immigrants. Immigrants who are more successfully assimilated within Israel are 11 percent more likely to achieve an average or higher level of income. This ﬁnding can be explained by more extensive social networking, which is associated with more successful assimilation, in addition to language proﬁciency. Improved social networking, in turn, leads to better job opportunities and raises the level of income. This ﬁnding was also found to be statistically signiﬁcant and consistent: no evidence was found of simultaneity between level of income and degree of assimilation.

Culture, Investment in Language and Earnings

289

In Israel, however, the enclave effect should not affect the results since the Israeli Government deliberately dispersed FSU immigrant to various parts of the country in order to prevent ethnic enclaves from forming. Thus, the geographic distribution of the immigrants is very similar to that of the native population and the immigrants’ share of the labor force is less than 10 percent in most of the larger cities in Israel.

7. Summary and conclusion An important component of an immigrant’s human capital is ﬂuency in the host country’s language. In the case of most immigrants, their mother tongue is not the majority or dominant language spoken in the host country. An immigrant who does not possess skills in the dominant language has fewer earning opportunities than a comparable native since his opportunities for training and job mobility, whether geographic or occupational, are more limited. Assimilation is much more difﬁcult. Furthermore, greater ﬂuency in the dominant language increases both the efﬁciency of job search and on-the-job productivity. The results obtained were in line with those presented in the literature. Thus, immigrants who improve their ﬂuency in Hebrew not only enhance their earnings but also their probability of being employed. As a result, there is a labor market incentive to acquire dominant-language skills. Whether and under what circumstance this incentive justiﬁes the cost of attaining language skills is of prime interest. This study contributes to the literature by examining whether immigrants respond to pecuniary incentives to acquire skills in the dominant language. Our main results are: (1) The probability of achieving proﬁciency in speaking Hebrew within any given period of time without participating in ulpan is higher for young, male, more-educated, and single immigrants while the probability for proﬁciency in writing Hebrew is higher for young, female, more-educated, and single immigrants. (2) Male, less-educated, older, and single immigrant workers beneﬁt more from participating in ulpan. (3) Participating in ulpan is more common among male, moreeducated, older, and single immigrant workers. (4) Male, older, and single immigrant workers potentially beneﬁt more from participating in ulpan and they indeed are more likely to participate. (5) Less-educated immigrant workers beneﬁt more from participating in ulpan than do more-educated immigrants, although participation in ulpan is more common among the latter. However, this may be due to upward bias in the coefﬁcients for the abilities to speak and write Hebrew for lesseducated immigrants. In other words, the beneﬁt derived from ulpan may be overestimated for less-educated immigrants. (6) When only immigrants with 13þ years of schooling are included in the estimation, the results

290

Erez Siniver

show that male, more-educated, older and single immigrant workers beneﬁt more from participating in ulpan and they participate in ulpan at a higher rate as well. Culture is intimately linked to pecuniary incentives, as immigrants do in fact respond to economic incentives in acquiring proﬁciency in the language of the host country, particularly immigrants with 13þ years of schooling.

References Beenstock, M. (1996), The acquisition of language skills by immigrants: the case of Hebrew in Israel. International Migration 34, 3–30. Beenstock, M., Chiswick, B., Repetto, G. (2001), The effect of linguistic distances and country of origin on immigrant language skills: application to Israel. International Migration 39 (3), 33–60. Berman, E., Lang, K., Siniver, E. (2003), Language-skills complementarily: return to immigrant language acquisition. Labour Economics 10, 265–290. Borjas, G. (1985), Assimilation, changes in cohort quality, and the earning of immigrants. Journal of Labor Economics 3 (4), 463–489. Borjas, G. (1987), Self-selection and the earning of immigrants. American Economic Review 7 (4), 531–553. Borjas, G. (1989), Immigrant and emigrant earnings: a longitudinal study. Economic Inquiry 27, 21–37. Card, D. (1990), The impact of the Mariel Boatlift on the Miami labor market. Industrial and Labor Relations Review 43, 245–257. Carliner, G. (1981), Wage differences by language group and the market of language skills in Canada. Journal of Human Resources 16, 385–399. Chiswick, B. (1991), Speaking, reading and earning among low-skilled immigrants. Journal of Labor Economics 9 (2), 149–170. Chiswick, B. (1998), Hebrew language usage: determinants and effects on earnings among immigrants in Israel. Journal of Population Economics 11 (2), 253–271. Chiswick, B., Miller, P.W. (1994), Language choice among immigrants in a multi-lingual destination. Journal of Population Economics 7, 119–131. Chiswick, B., Miller, P.W. (2002), Do enclaves matter in immigrant adjustment. IZA Discussion Paper No. 449, Bonn, Germany. Chiswick, B., Miller, P.W. (2007a), Matching language proﬁciency to occupation: the effect on immigrants’ earnings. IZA Discussion Paper No. 2587, Bonn, Germany.

Culture, Investment in Language and Earnings

291

Chiswick, B., Miller, P.W. (2007b), The international transferability of immigrants’ human capital skills. IZA Discussion Paper No. 2670, Bonn, Germany. Chiswick, B., Repetto, G. (2001), Immigrant adjustment in Israel: literacy and ﬂuency in Hebrew and earnings. International Migration: Trends, Policy and Economic Impact. Routledge, New York, pp. 204–228. Cohen-Goldner, S., Eckstein, Z. (2008), Labor mobility of immigrants: training, experience, language and opportunities. International Economics Review 49 (3), 837–872. Cohen-Goldner, S., Eckstein, Z. (2010). Estimating the return to training and occupational experience: the case of female immigrants. Journal of Economics 156, 86–105. Danzer, M.A., Yaman, F. (2010), Ethnic concentration and language ﬂuency of immigrant in Germany. IZA Discussion Paper No. 4742, Bonn, Germany. Dustmann, C. (1994), Speaking ﬂuently, writing ﬂuently and earnings of migrants. Journal of Population Economics 7 (2), 133–156. Eckstein, Z., Weiss, T. (2004), On the wage growth of immigrants: Israel, 1990–2000. Journal of the European Economic Association 2 (4), 665–695. Espenshade, T., Fu, H. (1997), An analysis of English language proﬁciency among U.S. immigrants. American Sociological Review 62, 288–305. Grenier, G. (1984), The effects of language characteristics on the wages of Hispanic-American males. The Journal of Human Resources 19, 35–52. Kossoudji, S. (1988), English language ability and the labor market opportunities of Hispanic and East Asian immigrant men. Journal of Labor Economics 6 (2), 205–228. Lazear, E.P. (1999), Culture and language. Journal of Political Economy 107 (6), s95–s126. McManus, W. (1985), Labor market costs of language disparity. American Economic Review 75, 818–827. McManus, W. (1989), Labor market effect of language enclaves. The Journal of Human Resources 25 (2), 228–252. McManus, W., Gould, W., Welch, F. (1983), Earning of Hispanic men: the role of English language proﬁciency. Journal of Labor Economics 1 (2), 101–130. The Israel Ministry of Absorption (1995). Employment of the FSU immigrant. Ministry of Absorption. Siniver, E, Arbel, Y., and Tobol, Y. (2009), Social involvement and level of income among immigrants: new evidence from the Israeli experience. Research Paper No. 11, College of Management Academic Studies, Rishon Letzion, Israel.

292

Erez Siniver

Tainer, E. (1988), English language proﬁciency and the determination of earnings among foreign-born men. Journal of Human Resources 23, 108–121. Warman, C. (2007), Ethnic enclaves and immigrant earnings growth. Canadian Journal of Economics 40 (2), 401–422. Weiss, Y., Sauer, Robert, M., Gotlibovski, M. (2003), Immigration, search, and loss of skill. Journal of Labor Economics 21 (3), 557–592.

PART III

Assimilation Struggles

CHAPTER 12

Immigration: America’s Nineteenth-Century ‘‘Law and Order Problem?’’$ Howard Bodenhorna,b, Carolyn M. Moehlingb,c and Anne Morrison Piehlb,c a

The John E. Walker Department of Economics, Clemson University, Clemson, SC, 29634, USA E-mail address: [email protected] b NBER, Cambridge, MA, 02138, USA E-mail address: [email protected] c Department of Economics, Rutgers University, New Brunswick, NJ, 08901, USA E-mail address: [email protected]

Abstract Past studies of the empirical relationship between immigration and crime during the ﬁrst major wave of immigration have focused on violent crime in cities and have relied on data with serious limitations regarding nativity information. We analyze administrative data from Pennsylvania prisons, with high-quality information on nativity and demographic characteristics. The latter allow us to construct incarceration rates for detailed population groups using U.S. Census data. The raw gap in incarceration rates for the foreign and native born is large, in accord with the extremely high concern at the time about immigrant criminality. But adjusting for age and gender greatly narrows that observed gap. Particularly striking are the urban/rural differences. Immigrants were concentrated in large cities where reported crime rates were higher. However, within rural counties, the foreign born had much higher incarceration rates than the native born. The interaction of nativity with urban residence explains much of the observed aggregate differentials in incarceration rates. Finally, we ﬁnd that the foreign born, especially the Irish, consistently have higher incarceration rates for violent crimes, but from 1850 to 1860 the natives largely closed the gap with the foreign born for property offenses. Keywords: Immigration, crime, prisons JEL classiﬁcations: J61, K14, N31 $

The title is a play on a quote from McCaffrey (1976) that the ‘‘Irish were America’s law and order problem’’ after 1850 (p.68).

Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008018

r 2010 by Emerald Group Publishing Limited. All rights reserved

296

Howard Bodenhorn et al.

Between 1847 and 1858, over 3 million immigrants arrived in the United States. The arrival rate, which had averaged less than 4 per 1,000 of the population in the 1830s, jumped into the double digits, reaching a peak of nearly 16 in 1851. This dramatic inﬂow coincided with a number of social and economic changes. One such change was an increase in crime, particularly violent crime. Violent crime in the United States ‘‘surged’’ in the middle of the nineteenth century just as the foreign-born population reached its peak (Gurr 1989; Lane 1989). Previous research has argued that these two phenomena were strongly linked. Monkkonen (1989) found that immigrants accounted for between one-third and two-thirds of the homicides in New York City between 1852 and 1869 (p. 91). Lane (1979) likewise found that the Irish were disproportionately represented in homicide indictments in Philadelphia between 1839 and 1901 (p. 103). The research literature, however, leaves many unanswered questions about how the ﬁrst major wave of immigration affected crime patterns. First, the studies to date have only examined violent crime. Violent crime accounts for a relatively small fraction of total crime and can exhibit different trends, age patterns, and geographic variation than nonviolent crime. Moreover, the studies of violent crime suffer from limited or inferential information on nativity. Most historical data sources on crime do not systematically report data on place of birth. Monkkonen, who used a sample of homicide reports from newspapers, reported estimates of immigrant involvement in murder after dropping records where nativity was not reported. This likely biases his ﬁndings on the immigrant murder rate upwards. Also without information on nativity, Lane used surnames to infer ethnicity in the Philadelphia court records. Our study provides a fresh look at the question of immigration and crime in the mid-nineteenth century using data sets created from the records of Pennsylvania’s state prisons from the 1830s to the 1860s. These records provide information on the birthplace, age, prior occupation, county of conviction, crime, and sentence of all individuals entering the prisons. The administrative prison data contain a census of all inmates during the covered time period, and the quality of the administrative data is very good. These advantages over the data used in previous studies provide more conﬁdence in interpreting the empirical relationships between nativity and crime and also allow us to test a richer set of hypotheses about the level and type of involvement of immigrants in crime and incarceration.

1. The ﬁrst major wave of immigration In the late 1840s, the United States experienced its ﬁrst wave of mass immigration. The annual number of arrivals increased by a factor of 4 and remained high for almost a decade (Figure 1). At its peak, the arrival rate represented an addition of nearly 2 percent of the population each year.

Immigration: America’s Nineteenth-Century ‘‘Law and Order Problem?’’

297

450,000 400,000 350,000 300,000 250,000 200,000 150,000 100,000 50,000 0 1830

1835

1840

1845

1850

1855

1860

Fig. 1. Immigrant arrivals to the United States, 1830–1862. Data underlying this ﬁgure came from Carter and Sutch (2006) in the Historical Statistics of the United States, Earliest Times to the Present: Millennial Edition. Data underlying that work came from public sources. Most of these new arrivals were ﬂeeing the economic and social upheavals in Europe of the period. Irish immigrants were seeking refuge from the devastation brought about by the potato famine, and German immigrants were escaping the violence and disorder of the social revolution, which was preceded by – and, perhaps, caused by – poor food harvests between 1845 and 1847 and an industrial slump in 1848 (Berger and Spoerer, 2001). Thus, a large proportion of mid-nineteenth-century immigrants would now be considered economic refugees, though some were driven to migration by political factors.1 Nearly from the beginning of the ﬁrst big wave of immigrants, contemporary observers sounded warnings about the economic and social consequences of the inﬂux of new arrivals. Many contemporaries worried about how well Irish Catholics would assimilate in Protestant America, and as early as 1847, an article in the Christian Watchman claimed that ‘‘The fact cannot be concealed that [the U.S.] is receiving into its bosom, a vast amount of poverty, ignorance, disease, and crime’’ (Christian Watchman July 2, 1847, p. 106). McCaffrey’s (1976) assessment accords with the 1

Facilitating these mass population movements was a signiﬁcant decline in transport costs over time. In the early 1840s, the Cunard Line began operating trans-Atlantic steamers, greatly reducing the cost and time of the voyage to America. In 1825, trans-Atlantic passage had cost d20 per head, whereas by 1863, steamer passage was only d4. 15s. and passage on a sailing ship was only d2. 17s. 6d. (Thomas, 1954, p. 96).

298

Howard Bodenhorn et al.

Watchman’s. He contended that the new Irish immigrants of the 1840s and 1850s were the ﬁrst immigrant group crowded into urban ghettoes and among the ﬁrst to experience the negative social consequences of living in crowded tenements. Hunger and poverty pushed Irish girls into prostitution and turned Irish boys into muggers and thieves. McCaffrey (1976, p. 68) claimed that the ‘‘Irish were America’s law and order problem’’ after 1850. Roughly two decades after the peak of Irish immigration, Eliott (1869) reported a remarkable ﬁnding: the number of Irish prisoners in Great Britain’s jails had declined from nearly 15,000 in 1851 to fewer than 2,700 in 1865. What disturbed contemporary Americans was Eliott’s claim that the smaller number of Irish in Great Britain’s jails and prisons was matched by a corresponding increase in the presence of Irish in American jails and prisons. Between the mid-1850s and the mid-1860s Irish immigrants made up about half of all arrestees in Boston and Philadelphia (Handlin, 1959, p. 257; Naylor, 1979, p. 53). In 1858 the New York Times reported that of the 12,000 arrests made by the New York Police Department over a three-month period, 8,000 were Irish and over 2,000 were immigrants from other countries (New York Times, February 22, 1858, p. 4).2 The national data available at the time also supported the perception that immigrants were the law and order problem of the day. The 1850 abstract produced by the Census Bureau reported that of the 27,000 persons convicted of crimes in the year ending June 1, 1850, 14,000 were foreign born (U.S. Bureau of the Census, 1853, p. 29). In an article reporting these data, the New York Times noted, ‘‘While we have, therefore, but about one foreign resident to nine native whites, there is a fraction over one foreign-born criminal to every native, including black and white’’ (New York Daily Times September 24, 1853, p. 4; emphasis in original). Calls were made to restrict the inﬂow of new migrants and limit the rights of immigrants already in the United States. Nativist societies sprung up around the nation, and by 1854, this sentiment had become the cornerstone of a new national political party, the Know-Nothing, or American, party. The Know-Nothing party did not advocate stopping the inﬂow of new migrants but rather sought to limit the inﬂuence of the foreign born on American politics and to exclude two groups of immigrants: paupers and criminals (Jones, 1974, p. 157). The KnowNothings did not have to try hard to establish a link between crime and immigration. As the author of the New York Times article on the 1858 arrest records for New York City noted, ‘‘Police records of this City are,

2

The statistic that ﬁve-sixths of all arrestees were immigrants and two-thirds of all arrestees were Irish immigrants does not seem credible at ﬁrst, but nearly identical proportions are reported by Handlin (1959, p. 257) for Boston in 1864. Nearly 10,000 of the 13,000 Boston arrestees were Irish immigrants; nearly 1,000 were immigrants from elsewhere. Only about 2,100 arrestees were native-born Americans.

Immigration: America’s Nineteenth-Century ‘‘Law and Order Problem?’’

299

after all, the strongest argument that any Know-Nothing can advance’’ in support of immigration restrictions (February 22, 1858, p. 4). The political inﬂuence of the Know-Nothing Party declined after its presidential candidate, Millard Fillmore, was roundly defeated in the 1856 elections (Jones, 1974, p. 157). But calls for restrictions on immigration resurfaced with each new wave of immigrants, and the perceived connection between immigration and crime remained an integral part of the American discussion. When the ﬁrst comprehensive immigration law was enacted in 1891, the ‘‘inadmissible classes’’ included persons convicted of crimes or misdemeanors, and the Immigration Act of 1917 included a provision to deport any immigrant who had been in the United States ﬁve or fewer years and had been sentenced to at least one year in prison, and any immigrant no matter the time spent in the United States who had been convicted of a more serious offense or prostitution (Moehling and Piehl, 2009). Although some accused European countries of intentionally exporting their criminals to the United States, most of the discussion attributed the criminal behavior of immigrants to disadvantage, disappointment, and the difﬁculties of adjusting to life in America.3 These arguments foreshowed academic theories about immigration and criminality that were developed largely during the third wave of immigration, in the early twentieth century. Many early criminologists attributed immigrant crime to the other demographic and social characteristics of immigrants, in particular their high rate of poverty (Taft, 1933). Others essentially extended theories of urban life, as immigrants were highly concentrated in the nation’s cities (see Bursik, 2006 for a review). But some scholars argued that it was the special circumstances faced by immigrants that led them to have different criminal patterns than natives. Sellin (1938) emphasized the ‘‘culture conﬂict’’ faced by immigrants as they try to adjust to a new society and a new set of behavioral norms. Handlin (1959) attributed the social disorder created by Irish immigrants in particular, to the combination of cultural conﬂict and economic disadvantage: ‘‘in no group was there an inherent predilection for crime, but among the Irish the combination of poverty and intemperance created a maladjustment expressed by petty infractions of the rules of society strange to them’’ (pp. 121–122). The New York City police records and the 1850 Census data make a strong prima facie case that immigrants had much higher crime rates than natives. However, these data are not as clear-cut evidence of immigrants’ disproportionate involvement in crime as they may ﬁrst 3

In blaming countries for sending their criminals, there are historical parallels with modern debates. Martinez et al. (2003), for example, provide evidence refuting the contention that the Mariel boatlift was an attempt on Cuba’s part to empty its prisons of hardened felons. Criminals were among the Marielitos, but most were political prisoners rather than violent felons.

300

Howard Bodenhorn et al.

appear. Arrests, then as now, are noisy measures of criminal behavior. Arrests are dominated by minor offenses such as disorderly conduct, and even for more serious offenses, can be based on police suspicions rather than the higher evidentiary standards that courts use. In many situations, the decision to arrest rather than otherwise defuse the situation is at the discretion of the ofﬁcer on the scene. Personal prejudices, such as anti-immigrant or anti-Catholic sentiments, are more likely to play a role in such discretionary situations. Even the conviction data from the 1850 Census must be interpreted with some care. Of the 27,000 convictions reported in the Census, over 17,000 were from just two states, New York and Massachusetts (U.S. Bureau of the Census, 1854, p. 165). The large convict populations in these two states reﬂect two features: crime is more common, per capita, in large cities than in smaller towns or rural areas; and these cities, being the two most common ports of entry, had large immigrant populations (Glaeser and Sacerdote, 1999). So the disproportionate presence of the foreign born in the conviction data reﬂects at least in part the disproportionate presence of the foreign born in jurisdictions with high crime rates. Any comparison of crime involvement by nativity must also take into account the disproportionate representation of immigrants in the demographic group with the highest crime rate: young males. Even though immigrants may have accounted for less than 15 percent of the population in 1850, they accounted for a much higher fraction of males aged 18–25. Because the age–crime proﬁle is steep, failing to account for small differences in age distributions can lead to ‘‘aggregation bias,’’ or to large but inaccurate calculations of differences in aggregated crime rates. Moehling and Piehl (2009) have shown that controlling for age effects greatly alters the comparison of native- and foreign-born conviction rates in the early twentieth century. After the National Origins Quota Act of 1924 sharply reduced the inﬂow of new migrants, the immigrant population aged rapidly relative to the native-born population. Analysts at the time concluded that immigrants had a ‘‘2 for 1’’ advantage in incarceration. But using more detailed population data, Moehling and Piehl show that, for violent crimes in 1930, this advantage was wholly explained by the different age distributions. As noted above, previous historical studies of crime in the nineteenth century have found that immigrants were more prone to criminal behavior, but these studies did not adjust for the age and geographic distribution of immigrants (Lane, 1979; Monkkonen, 1989). More problematic though, these studies focused on homicides and were forced to use surnames or unsystematically collected data on birthplace to infer immigrant status. The data used here afford an opportunity to study a wider range of criminal acts, including property crime and morals offenses, and to more accurately account for nativity. Exploiting these data will, therefore, provide a fuller appreciation of the connection between nativity and criminality in the mid-nineteenth century.

Immigration: America’s Nineteenth-Century ‘‘Law and Order Problem?’’

301

2. Pennsylvania prison data Our approach to the question of immigration and criminality in the nineteenth century is to restrict attention to a single jurisdiction, relying on administrative data on incarceration as well as data on the general population from the decennial federal censuses. We have data for the period 1830–1862 from Pennsylvania’s two nineteenth-century state prisons: the Eastern State Penitentiary in Philadelphia and the Western State Penitentiary in Pittsburgh. The extant records include the ‘‘Descriptive Registers’’ and the ‘‘Convict Docket’’ from the Eastern State Penitentiary and ‘‘Descriptive Registers’’ from the Western State Penitentiary (Eastern State Penitentiary, 1829–1857; Western State Penitentiary, 1826–1876). The ledgers include basic information about the convicts, including their names, ages, nativities, preincarceration occupations, the crimes for which they were incarcerated, sentence lengths, prior convictions, court and county of conviction, and the date of and reason for release (completion of sentence, executive pardon or commutation, or death). The Pennsylvania prison data hold several advantages over previously used information in the study of immigrant criminality. One advantage is that they are not limited to one type of crime. In the 30 years of data used here, convicted felons were incarcerated for more than 60 different crimes, ranging from abortion to vagrancy. The most common offenses, not surprisingly, were property crimes, notably burglary, larceny, and horse theft. Pennsylvania’s prisons were also home to violent offenders – murderers, rapists, and robbers – and to those convicted of various morals infractions, including bigamy, incest, and fornication. These data offer an opportunity to study the connection between immigration and crime that covers the entire gamut of nineteenth-century criminal activity. Because we have data on the crimes that led to incarceration, we can examine whether immigrants and natives were convicted of different types of crimes or if one group was more likely to be involved in violent or property crimes. The main advantage of these data for our examination, however, is the systematically collected information on birthplace. We can determine immigration status directly and can even consider differences in experiences across immigrants from different source countries. Despite the many advantages of the Pennsylvania prison data over those used in previous studies, the data are not without shortcomings. Prison incarcerations do not measure criminal activity per se. Rather, they reﬂect criminal activities that have been reported, investigated, prosecuted, and resulted in a sentence of greater than 12 months.4 So although these data result from a sequence of choices, all jail or prison outcomes are ﬁltered by 4

Under Pennsylvania law, the Eastern and Western State penitentiaries were not to accept criminals sentenced to less than 12-month incarceration; criminals sentenced to less than a year were to serve their time in a county jail. There were, however, some prisoners sent to the penitentiaries with shorter than 12-month sentences. No explanation for these exceptions is given in the records.

302

Howard Bodenhorn et al.

the decision to arrest, prosecute, convict, and sentence. One advantage of our data compared to alternative sources, such as police arrest records or county jail records, is that prison incarcerations were not dominated by minor offenses, the prosecution of which varies greatly over time and space. State prison commitments capture more serious crimes and those to which more resources are put toward apprehension, prosecution, and conviction. Like arrest records, however, incarceration data likely also reﬂect the impact of prejudice and discriminatory justice. The sequence of steps, and hence, the number of actors involved, in the process leading to a prison conviction reduced the effect of any one individual’s personal prejudices on the outcome, but the widespread nativist sentiment during the period no doubt inﬂuenced jurors as well as judges. A criminal justice system biased against immigrants, though, potentially has several consequences, not all of which increase the immigrant incarceration rate. Much crime takes place within rather than across communities. The police may have been reluctant to investigate immigrant-on-immigrant crime. Moreover, fear of the police and the courts may have made immigrants less likely to report crimes. There is no way to identify deﬁnitively the impact of discriminatory justice in incarceration or any other type of crime data. Furthermore, biased application of law would be indistinguishable in crime data from other consequences of immigrant status, such as limited access to English translation or naivete´ about the criminal justice process. Because we have both the entry and release dates of prisoners, we can look not only at the ﬂow of inmates into Pennsylvania state prisons but also at the Pennsylvania state prison population on speciﬁc dates. We can therefore look at the prison population around the dates of the federal censuses to compare the composition of the prison population to that of the nonincarcerated population of Pennsylvania. Moreover, we can use the prison and population data together to construct incarceration rates which control for age, gender, race, and even geographic distribution. In our analysis, we exclude those inmates who were referred from U.S. district courts to serve time in Pennsylvania prisons for violations of federal law. These 158 inmates (2.2 percent of the commitments) were convicted of federal crimes, such as mail theft or counterfeiting, rather than state crimes. At the time, there were few federal correctional facilities, so federal inmates were housed wherever federal ofﬁcials could locate space (Friedman, 1993, pp. 261–272). Our ﬁnal sample then contains the universe of those sent to prison for violations of state criminal law.5 5

It is possible that we miss observing some who were executed for their crimes. If someone were convicted of a capital crime and executed shortly thereafter in the local jurisdiction, they would not enter the prison data. We do observe some inmates who were executed after serving some time in one of the state prisons. We have searched for the universe of executions in order to identify the extent to which the prison data overlook the most serious convictions due to execution, but we have not located this information.

Immigration: America’s Nineteenth-Century ‘‘Law and Order Problem?’’

303

We combine the state prison data with data on the general population of Pennsylvania from the decennial federal population censuses taken from the Integrated Public Microdata Series (IPUMS) (Ruggles et al., 2008).6 The census data allow us to test whether immigrants were disproportionately represented in Pennsylvania’s state prisons and allow us to construct incarceration rates controlling for age, gender, race, and nativity. Despite the quality of the source data, we approach such comparisons and calculations with care. Even today, the federal census suffers from underenumeration, that is, some individuals are not counted in the census data. If underenumeration is uniform across the population, it would present few problems for our analysis. The calculated incarceration rates would be slightly higher than the ‘‘true’’ rates, but we could still look at differences in rates across groups as evidence of systematic differences in behavior. The problem is that underenumeration in the census is potentially biased; the likelihood of being missed by the census varied across groups in the population. Immigrants, who may have faced language barriers, been fearful of government ofﬁcials, and lived in crowded living quarters, may have been more likely to have been missed by census takers than natives. Such biased underenumeration would bias our analysis toward ﬁnding that immigrants had higher incarceration rates than natives. Several scholars have attempted to estimate the extent of underenumeration (Steckel, 1991 provides a discussion), but two are particularly useful for our purposes because they speciﬁcally address biased underenumeration of blacks and immigrants. Using evidence of age heaping, Sharpless and Shortridge (1975) estimate that African American underenumeration was about 10 percent greater than that for whites. They also estimate 20 percent immigrant underenumeration. Furstenberg (1979) uses a rare recount in Philadelphia and estimates immigrant underenumeration to be between 5 and 20 percent higher than that for native-born whites. Because our objective is to compare immigrants in Pennsylvania’s state prisons to immigrants in the state’s general population, we need to be sensitive to the issue of biased underenumeration. In our results we use the lower- and upper-bound estimates (5 and 20 percent) of the relative underenumeration of the foreign born to construct approximate error bounds around our point estimates of criminality. 3. Immigrant arrivals and prison commitments The ﬁrst issue we address is the temporal patterns of immigrant arrivals and prison commitments. Figure 2 graphs all commitments to the Pennsylvania prisons as well as commitments for natives and the foreign born. Two features of the commitment series are easily explainable. 6

The IPUMS data and supporting documentation are available online at www.ipums.umn.edu.

304

Howard Bodenhorn et al.

450 400 350

Total prison commitments Commitments of the foreign born Commitments of the native born

300 250 200 150 100 50 0 1830

1835

1840

1845

1850

1855

1860

Fig. 2. Commitments to Pennsylvania state prisons, 1830–1862. Source: Pennsylvania state prison data. See text for details.

First, commitments increase in the early 1830s not from increases in crime, but because the prisons themselves expanded. The original Western State Penitentiary opened in 1826, but the housing facilities were razed just seven years later when they were deemed inadequate. The Eastern State Penitentiary ofﬁcially opened in 1829 with one operating cellblock. The second building opened in 1831; the seventh and ﬁnal pre-Civil War cellblock opened in 1835 (Eastern State Penitentiary, 2010).7 The volume of commitments remains relatively ﬂat between 1835 and 1855 with, perhaps, a slight decline between 1840 and 1850. Second, the sharp decline in commitments after 1860 likely reﬂects the effects of the Civil War. The war pulled large numbers of young men into military service, which expanded employment opportunities for those who did not enlist (Monkkonen, 1981, p. 80; Gallman, 1990, pp. 271–273).8 7

Five additional cellblocks were constructed at the Eastern State Penitentiary between 1877 and 1911 and the Western State Penitentiary relocated to a new site in 1882, but these capacity increases occurred after the period we discuss. The Eastern State records also fail to reveal when the practice of solitary conﬁnement was abandoned. It was ofﬁcially ended in 1913, but it was known to have unofﬁcially ended decades earlier. Given the rapid increase in commitments in the 1850s, solitary conﬁnement may have been abandoned much earlier than previously believed. 8 Philadelphia did not experience the violent riots in the early years of the war that plagued other large Northern cities. Gallman (1990) attributes this relative peace to the city’s mayor, Alexander Henry, and his efforts starting in the late 1850s to improve the police force (p. 192).

Immigration: America’s Nineteenth-Century ‘‘Law and Order Problem?’’

305

The feature evident in Figure 2 that is less easily explained is the spike in prison admissions to well over 300 admissions per year in the late 1850s, and peak of nearly 400 in 1860. This spike seems to mirror the spike in immigration (nationwide), lagged by about seven years. Several possible explanations could link these spikes. First, commitments may have lagged increased immigrant criminality due to long and variable lags in the dispensation of criminal justice. The lag, however, is far too long for the similar patterns in arrivals and commitments to be related. Nineteenthcentury criminal justice, just as in the twentieth century, was subject to procedural delays and continuances, but criminal cases were cleared from the docket at speeds that would astound modern Americans. If the accused was tried before a quarterly court, he might have been held in a county jail for as long as three months before trial, but grand juries worked with dispatch and the typical criminal trial lasted less than an hour. Moreover, if the jury found the defendant guilty, they determined sentence length in the same sitting. Convicted felons arrived at the prison just days after their court date (Langbein, 1978; Rice, 1996). A more plausible connection hinges on different criminal involvement by age at immigration. If those who initiated the crossing of the Atlantic have low crime rates (due to fear of the government or special positive selection), those who were young at the time of immigration may appear more like second-generation immigrants or natives. Thus, the lag may represent the aging into the crime-prone years of the ‘‘1.5 generation.’’ Unfortunately, the Pennsylvania state prison data do not provide information on the age at immigration, so we cannot test this hypothesis directly, but the data provided in Figure 2 (the commitment diagram) are not fully inconsistent with the hypothesis. There is an increase in foreignborn commitments between 1848 and 1856 (relative to the trend between 1832 and 1847), which may be due to an increased recent immigrant pool or the entry of young-age immigrants into the prime offending years. But Figure 2 reveals an interesting feature of annual commitments. There is a modest increase in foreign-born commitments between 1856 and 1860, but it pales in comparison to the increase in native-born commitments over the same interval. The increase in commitments after 1856 is due to increased commitments of nonviolent offenders and may be due, in part, to the Panic of 1857. Although early nineteenth-century ﬁnancial panics are not believed to have notable employment or wage effects, the series is consistent with crime as an anticyclical activity (Calomiris and Schweikart, 1991).9 While the time patterns are consistent with the panic story, it is not immediately obvious why the 1857 panic 9

Fishlow (1965) argues that a recession began in 1856 caused, in part, by a decline in immigrant inﬂows to the Old Northwest, which had previously boosted land prices and fueled a railroad boom. The recession was transmitted to eastern markets with the failure of Ohio Life and Trust, which led to a run on eastern money center banks.

306

Howard Bodenhorn et al.

would have more pronounced effects than the 1839 panic and why the downturn would have more pronounced effects on the native born than on the foreign born. Some evidence suggests that the pre-prison economic circumstances of inmates declined for natives during the panic years. Using the socioeconomic index (SEI) for occupation titles developed by Reiss et al. (1961) using average educational attainment and earnings for occupations listed in early twentieth-century censuses, the average SEI value of native-born commitments declined from 17.9 between 1850 and 1856 to 15.8 between 1857 and 1860. Among foreign-born commitments, the average SEI value declined from 17.7 in the pre-panic years to 16.6 in the post-panic years. These ﬁndings need to be interpreted with some care due to the fact that the index is based on a ranking of occupations 50 years after the panic. Nonetheless, the statistically signiﬁcant decline in the SEI value of the native born suggests that the average occupation declined from a puddler in an iron foundry to a cook or waiter, which may be an economically meaningful change. That the change in the average occupation of new commitments was not signiﬁcantly different for the foreign born suggests that the post-panic recession may have had a more detrimental effect on native-born than foreign-born workers. But these interpretations are tentative in light of how little economic historians know about the labor market effects of early nineteenth-century recessions. 4. Aggregate incarceration experience: immigrants and natives One issue we can address directly with the Pennsylvania prison data is whether immigrants were more likely than natives to be incarcerated. We begin our analysis by comparing the percent foreign born within the prison population to that of the general population on the dates of the federal censuses. To increase sample sizes and gain some statistical power in the prison data, we take a snapshot of the prison population on June 1 in each of the ﬁve years surrounding the Census year.10 That is, data reported for 1840 reﬂect the composition of the two Pennsylvania prisons on June 1 of 1838–1842. Taking ﬁve years rather than a single year gives us sufﬁcient sample size for later analyses by type of crime and should mitigate any short-term idiosyncratic behavior in crime or sentencing.11 Table 1 presents these numbers. In 1840, 18.7 percent of Pennsylvania prison population was foreign born; a decade later it had increased to 10

Note that these snapshots represent the ‘‘stock’’ of inmates around the Census years. Therefore, the numbers are not directly comparable to those in Figure 1, which plots the ﬂow of commitments into prison. 11 For the broader categories of interest, the ﬁve-year averages do not differ greatly from the data constructed from the prison population on the census dates: June 1, 1840; June 1, 1850; and June 1, 1860. We use ﬁve-year averages to deal with the ‘‘small cell’’ problem that arises once we start cutting the data by multiple characteristics such as crime type, nativity, and age.

Immigration: America’s Nineteenth-Century ‘‘Law and Order Problem?’’

Table 1.

Percentage foreign-born prison versus general population Prison population

All 1840 1850 1860 Females 1840 1850 1860 Males 1840 1850 1860 White males 1840 1850 1860

307

18.7% (2624) 26.4 (2105) 27.4 (3278) 7.3 (137) 35.8 (81) 37.1 (116) 19.3 (2487) 26.0 (2024) 27.1 (3162) 27.8 (1681) 32.6 (1579) 30.8 (2753)

General population All ages

Ages 18–44

13.1%

20.4%

15.0

24.5

19.0 23.6

21.7 25.4

22.3 25.8

Notes: The Pennsylvania state prison population ﬁgures were calculated from snapshots of the prison population on June 1 (the date of the federal population census) of each year in the ﬁve-year period centered on the census year. See text for more information. The numbers in parentheses represent the number of observations used in the calculations. The general population data were constructed from the IPUMS samples of the 1850 and 1860 federal census returns for Pennsylvania. The 1840 federal population census did not collect data on individual nativity.

26.4 percent and further rose to 27.4 percent in 1860. The wave in new immigrant arrivals evident in Figure 1 was accompanied by an increase in the foreign-born prison population. Moreover, the foreign born accounted for a much larger fraction of the prison population than they did of the general population in Pennsylvania during this period. Because the 1840 Census did not collect data on birthplace, we do not know the percentage of foreign born in the general population in that year, but the foreign born

308

Howard Bodenhorn et al.

accounted for only 13.1 percent of Pennsylvania’s population in 1850 and 15.0 percent in 1860. Even if the relative underenumeration of immigrants in the census reached the 20 percent upper bound, the foreign born still accounted for 18 percent of the nonincarcerated population in Pennsylvania in 1860, compared to 27.4 percent of those in prison. A simple comparison of the fraction foreign born in the prison population to that of the population overall may be misleading. The age distribution of the prison population differs greatly from the age distribution of the general population, which makes immigrants appear more prone to criminal behavior than they really are. Panel A of Figure 3 presents the age distribution of the Pennsylvania state prison population in 1860. Over 60 percent of the prison population was between the ages of 18 and 29, and over 85 percent was between the ages of 18 and 44. Panel B of Figure 3 presents the age distributions of the native- and foreign-born populations constructed from the IPUMS sample of the 1860 Census. Almost half of the native population was under the age of 16, a group almost absent from the prison rolls, and only 35 percent was in the prime crime ages of 18–44. In contrast, less than 10 percent of the foreign-born population was children, and almost 65 percent was between the ages of 18 and 44. In other words, the foreign born accounted for a much larger fraction of the population in the age range most at risk for incarceration than that of the overall population. Returning to Table 1, if we compare the percent foreign born in prison to that of the general population ages 18–44, the disparity shrinks; in 1860, immigrants accounted for 27.4 percent of the prison population but 24.5 percent of the general population aged 18–44. If actual immigrant underenumeration approached the 5 percent lowerbound estimates, any difference between the fraction foreign born in prison and that in the general population is practically eliminated. The immigrant population during this period also had a higher fraction male than female. If we compare the fraction foreign born among male prisoners to that of Pennsylvania’s male population ages 18–44, the gap between the prison population and the general population narrows even further: 27.1–25.4 percent in 1860. Interestingly, limiting the comparison to just females has the opposite effect. Foreign-born women accounted for 37.1 percent of the female prison population in 1860 but only 23.6 percent of the Pennsylvania’s female population in the relevant 18–44 age range. It is important to note, however, that very few women were sentenced to the Pennsylvania state prisons during this period; for each yearly snapshot, there are only about 20 female prisoners. Women, then as now, are much less likely to be convicted of crimes and when they are, it tends to be for less serious offenses than men, which results in short sentences typically in local jails or other facilities. The group of women found in the state prisons should be viewed as fairly exceptional, even among female criminals. It is notable, though, that the foreign born are disproportionately represented in this exceptional group.

Immigration: America’s Nineteenth-Century ‘‘Law and Order Problem?’’

Fig. 3.

309

Age distributions, 1860.

Adjusting for age and gender greatly narrows the observed gap between the imprisoned foreign born and those in the general population, but if we adjust for race the gap widens. Pennsylvania had a small, but not insigniﬁcant, black and mixed-race population in the pre-Civil War period, and this population was disproportionately represented in the state prisons. Immigrants during this period were, for the most part, white. If the comparison is restricted to white males ages 18–44, the disparity

310

Howard Bodenhorn et al.

Table 2.

Incarceration rates by race, gender, and nativity (per 100,000 persons aged 18–44)

All males All females White males Black males Native-born males White native-born males White foreign-born males

1850

1860

90 4 72 808 86 62 105

114 4 101 861 112 94 121

Note: See notes to Table 1.

between the percent of foreign born in the prison and general population increases. In 1860, the foreign born accounted for 30.8 percent of the white male prison population but only 25.8 percent of the prime-age white males in the general population, a disparity that only disappears if the highest estimate for potential underenumeration bias (20 percent) is used. We need to ask whether a study of nativity effects on incarceration, and crime more generally, is correct to focus on the white population. Modern research on immigration and crime does not restrict itself to any one racial group, but today’s immigrants are of all races. In the nineteenth century, the discussion of immigration and crime concentrated on the differences by nativity within the white population. Although there were West Indian immigrant communities in some North American cities, they were quite small. Northern cities were not attractive to West Indian migrants because existing African American communities in these cities, where most immigrants made their homes, were small. Black criminality, too, was viewed as a distinct phenomenon from crime within the immigrant communities. Finally, Philadelphia had the largest free African American population of any non-Southern U.S. city, so its experience (and that of Pennsylvania more generally) may not be indicative of the racial composition of crime in nonslave states. Any generalizations based on Pennsylvania’s black experience must be drawn with caution. Table 2 reports incarceration rates (per 100,000) by race, gender, and nativity within the crime-prone 18- to 44-year age category. The ﬁrst thing to note is just how high the incarceration rates are for African American (black and mulatto) males; they exceed 800 per 100,000, making them seven to eight times the rate for any other studied group.12 The exceptional African American incarceration rates are potentially inﬂated by census

12

Current incarceration rates are higher, but similarly disproportionate. White men are incarcerated at a rate of 487 per 100,000 and African American men at a rate of 3,161 per 100,000, for a ratio of over 6 to 1 (Sabol et al., 2009).

Immigration: America’s Nineteenth-Century ‘‘Law and Order Problem?’’

311

underenumeration, but no plausible value for the underenumeration bias can account for the disparity. It seems reasonable to conclude that African Americans were much more likely to be incarcerated than were white men.13 Because the black experience was so different than the native-born white experience, including blacks with native whites makes the foreignborn incarceration experience appear more favorable. In 1860, for instance, the calculated incarceration rate for white foreign-born males is 121 and that of native males is very close (112). If we restrict the comparison to white males, the nativity gap is much larger: 94 white native-born incarcerations per 100,000 population versus 121 foreign-born per 100,000 population in 1860. In most of what follows we exclude African Americans from the analysis, not because their experience is unworthy of study, but rather because it is so markedly different from the white experience that it merits study on its own terms. This focus also allows us to situate our results better into the historical debate about immigration and crime as well as to isolate the effect of nativity from that of race. We further restrict ourselves to the incarceration patterns of males as the rates of incarceration of women in Pennsylvania’s prisons were quite low (4 per 100,000 in 1850 and 1860). 5. Exploring the differences in immigrant and native incarceration The aggregate incarceration rates in Table 2 indicate that among primeage white males, immigrants were much more likely to be in a state prison than were natives in both 1850 and 1860. Even if we assumed the upper-bound estimate (20 percent) for the relative underenumeration of immigrants, the incarceration rate for immigrants would be substantively higher than that of natives (100 vs. 94). These aggregate rates, however, control only crudely for differences in the age distributions of the foreign born and natives. As shown in Figure 3, the age–crime curve is rather steep; 60 percent of all prisoners in 1860 were between the ages of 18 and 29, and this age group comprised a larger fraction of the immigrant population than the native-born population. In Table 3, we present the incarceration rates for immigrants and natives by more narrowly deﬁned age groups. Incarceration rates are minimal for those under 16, and quite low for 16- to 17-year olds and those over age 45.14 For both immigrants and the native born, the 13

Du Bois (1899) found comparable racial incarceration differences in late nineteenth-century Philadelphia. 14 At the time, common law held that those aged 14 and older could be sent to state prison. (In our data, there are several inmates younger than 14.) Separate juvenile jurisdiction came later. The ﬁrst juvenile court was started in Illinois in 1899. We include all observations in the aggregate rates, including children and those with age missing.

312

Howard Bodenhorn et al.

Table 3.

Incarceration rates of white males by age and nativity (per 100,000)

Age category

Under 16 16–17 18–19 20–24 25–29 30–34 35–39 40–44 45–49 50–54 W55

1850

1860

Native born

Foreign born

Native born

Foreign born

0.13 19 48 64 56 52 44 40 24 18 16

0 65 87 121 94 57 95 69 58 38 31

0.24 27 83 118 96 59 52 36 28 28 12

0 29 124 167 117 80 71 65 43 60 38

Note: See notes to Table 1.

age–incarceration rate proﬁles exhibit the expected pattern – peaking in the early twenties, and then slowly falling thereafter. Note, however, that the incarceration rates for immigrants were much higher than those for natives for every age group 18 and older. The nativity differences are particularly striking in 1850 when for many age groups the foreign-born rate was twice that of the natives. Between 1850 and 1860, the incarceration rates for men between 18 and 29 grew dramatically for both nativity groups. In general, the rates increased more for natives than the foreign born, but the foreign born still had substantially higher incarceration rates. At every age, immigrants were more likely than natives to be in a state prison. As discussed earlier, the story of immigration has long been intertwined with the development of cities. Immigrants were concentrated in large cities where reported crime rates were higher, perhaps because criminal opportunities and activities were greater in these areas or perhaps because there was greater law enforcement in these areas. In Tables 4 and 5 we investigate the hypothesis that urban and rural counties had different patterns of both immigration and incarceration. We deﬁne ‘‘urban’’ as Philadelphia and Allegheny counties, with the rest of the state as ‘‘rural.’’ The ﬁrst interesting fact is that while nearly half of the population in the urban counties was foreign born, about 35 percent of prisoners convicted from urban courts were foreign born. In the rural counties, 15–17 percent of the general population was foreign born in 1850 and 1860, whereas nearly one-third of prisoners from rural areas were foreign born. In Table 5 we consider ﬁner age-speciﬁc incarceration rates (comparable to those in Table 3) for the urban and rural counties to explore age–crime

Immigration: America’s Nineteenth-Century ‘‘Law and Order Problem?’’

Table 4.

313

Percentage foreign-born prison versus general population by urban/rural, white males

Age category

Prison population

1840 1850 1860

General population ages 18–44

Rural

Urban

Rural

Urban

28.1 30.5 27.5

27.2 35.8 36.6

14.8 17.4

45.1 48.4

Notes: See notes to Table 1. ‘‘Urban’’ deﬁned as Allegheny and Philadelphia counties.

Table 5.

Incarceration rates of white males by age, nativity, and urban/rural (per 100,000)

Age category

Rural

Urban

Native born

Foreign born

Native born

Foreign born

1850 16–17 18–19 20–24 25–29 30–34 35–39 40–44 45–49 50–54 W55

12 28 48 49 35 34 37 19 15 11

33 51 107 97 78 100 139 80 44 29

59 147 153 88 131 85 55 55 46 76

85 129 140 91 41 89 20 32 28 36

1860 16–17 18–19 20–24 25–29 30–34 35–39 40–44 45–49 50–54 W55

15 63 87 82 56 45 34 26 26 12

42 124 209 141 93 103 63 33 67 31

72 163 257 161 70 85 44 45 42 13

21 124 131 97 67 40 68 58 53 45

Note: See notes to Table 4.

proﬁle in greater detail. When we parse the data in this way, several remarkable features of nineteenth-century criminal sentencing appear. First, the disproportionate representation of the foreign-born young adults in Pennsylvania’s prisons is almost entirely a rural, not an urban, phenomenon. Among men in their twenties in 1850, the rural foreign born

314

Howard Bodenhorn et al.

were incarcerated at twice the rate of the native born. Among men between 35 and 44 years old, the rural foreign born were incarcerated at three times or more the rate of the native born. The disparities narrow somewhat by 1860, but the rural foreign born are about twice as likely as rural nativeborn men in the same age group to be incarcerated. Second, among those men in urban counties in both 1850 and 1860, the foreign born generally have lower incarceration rates than natives. In 1860 in particular, the incarceration rates for the foreign born are sometimes appreciably lower, on the order of 50–60 percent of the native rates. Third, when we compare incarceration rates for urban and rural areas in 1850, the incarceration rates for native-born men are much higher for urban counties than for rural ones. In 1850 the incarceration rate for urban nativeborn teens and adult men is two to three times the rate of rural native-born males in the same age group. Although incarceration rates decline with age for both urban and rural native-born men, the urban offending rate exceeds the rural offending rate well past the peak offending years. This result is consistent with Glaeser and Sacerdote’s (1999) explanation for ‘‘why is there more crime in cities?’’ Cities afford more opportunities for crime, often offer greater rewards for criminal activity, and the anonymity of cities reduces apprehension probabilities relative to rural areas.15 It is striking how different the urban–rural comparison is for natives and the foreign born. For immigrants, the incarceration rates are on the same order of magnitude in rural and urban settings. This discrepancy suggests that the phenomena of immigrant and native crime are somewhat different. Recall that our study measures crime from prison records, which include serious felonies. We know from previous studies that immigrants were arrested at much higher rates than native born. The population of arrests will be dominated by more trivial offenses. The relative underrepresentation of the urban foreign born in prisons (especially relative to the arrest ﬁgures cited earlier, in which the foreign born were the vast majority) suggests that their criminal experience was, as Handlin (1959, p. 134) characterized it, more the consequence of poverty and a failure to comply with the barely understood rules of a new society than ‘‘deliberate’’ wrongdoing by men ‘‘at war with society.’’ This interpretation is consistent with Moehling and Piehl’s (2009) ﬁnding, using data from 50 years later, that the foreign born had high rates of prison commitments for minor crimes, and that these high rates continued well into middle age. In summary, Table 5 reveals that in rural counties, the foreign born had much higher incarceration rates than the native born. By 1860, the nativity gap is so large that the rural foreign-born incarceration rates are the same 15

While the pattern of positive correlation between crime rates and city population was longstanding, the dramatic crime declines over the past decade or so have been largest in the largest cities, so that the gradient is no longer as strong as when evaluated by Glaeser and Sacerdote (1999).

Immigration: America’s Nineteenth-Century ‘‘Law and Order Problem?’’

315

order of magnitude or even higher than those for urban foreign born. For natives, the rural rates are generally half the size of those of the urban native born. Clearly the interaction of nativity with urban residence explains much of the observed aggregate differentials in incarceration rates. While it is well beyond the scope of this chapter, it would be interesting to know whether this urban–rural differential was unique to nineteenth-century Pennsylvania or whether it is a generalizable result. It may explain why, as Martinez and Lee (2000, p. 495) note, ‘‘the major ﬁnding of a century of research on immigration and crime is that immigrants y nearly always exhibit lower crime rates than native groups.’’ This major ﬁnding may exist because studies tend to investigate urban crime. Dense ethnic neighborhoods in urban places may provide cultural refuge for new immigrants not found in rural places. In Tables 6 and 7, we disaggregate along a different dimension, speciﬁcally, offense type. Table 6 shows the offense distribution by nativity. Table 6.

Distributions of commitments by type of crime, white males, 1830–1862

Property crime) Crime against a person Homicide Arson Crime against public order Felony, not speciﬁed Moral crime Misdemeanor, not speciﬁed

Native born

Foreign born

80.2% 7.3 4.7 3.0 1.6 1.6 1.6 0.1

72.4% 10.0 9.5 2.9 3.1 1.1 0.9 0.2

Notes: Property crimes include larceny, burglary, counterfeiting, forgery, fraud, receipt of stolen property, and the like. Crimes against a person include assault, rape, attempted murder, and robbery. Crimes against the public order include perjury, resisting arrest, gambling, and obstruction. Moral crimes in this period include prostitution, incest, pornography, and sodomy. Columns may not sum to 100% due to rounding.

Table 7.

Incarceration rates of white males by crime type and nativity (per 100,000 persons ages 18–44)

Crime type

Violent Property Other

1850

1860

Native born

Foreign born

Native born

Foreign born

10 47 5

22 71 11

17 67 10

39 70 12

Note: Violent crimes deﬁned as homicide plus all crimes against persons.

316

Howard Bodenhorn et al.

Here the data are used somewhat differently in order to produce larger sample sizes, which afford a detailed accounting. The statistics in Table 6 are calculated from all commitments to prison over the period 1830–1862. Relative to the point-in-time snapshots reported earlier, commitment data will emphasize crimes with shorter terms of incarceration. Using this measure, 72 percent of the foreign-born inmates were committed for property offenses as were 80 percent of the natives. Nearly 10 percent of the foreign born were committed for homicide, twice the proportion as among the natives. Note that the offense information is not neatly categorized. Some records indicate simply ‘‘felony’’ or ‘‘misdemeanor’’ without designating the type of underlying offense, but the proportion of unspeciﬁed crimes is too small to substantially alter the native–foreign relative commitment rates unless they were systematically applied to only natives or foreign born. The bottom line is that, among those committed, immigrants were more likely to have engaged in a violent act. Table 7 returns to the incarceration rate measure, grouping crimes into ‘‘violent’’ (homicide, person), ‘‘property,’’ and ‘‘other’’ (public order and moral offenses along with misdemeanor and otherwise uncategorized offenses). Here we see that the foreign born consistently have higher incarceration rates for violent crimes, but from 1850 to 1860 the natives largely closed the gap with the foreign born for property offenses. In the next section we look at whether there are different crime patterns across particular immigrant groups.

6. Variation across immigrant groups: British, Irish, and Germans One ‘‘major ﬁnding’’ of research into immigration and crime in the twentieth century is that immigrants exhibit lower rates of criminality than natives. A second is that there is wide variation in criminality across broadly deﬁned immigrant groups (Martinez and Lee, 2000). That is, ‘‘Asian’’ criminality may not be a meaningful measure if offending rates differ between Chinese, Japanese, Vietnamese, and Filipino immigrants. In the same way, immigrant criminality may not be a meaningful idea for a study of the nineteenth century if offending rates or types of crime committed differed across different nationalities. In this section, we consider commitment and incarceration rates across the three largest mid-nineteenth-century immigrant groups, namely the Irish, Germans, and Britons. Our use of these three groups is based on two criteria. First, we can think of the British as ‘‘old’’ immigrants in that they represented the largest and longest running nationality of immigrants over the very long term between settlement in 1607 Jamestown and the mid-nineteenth century. While there was a long tradition of both Irish and German immigration, the immigrants from these countries changed dramatically at mid-century. Before the potato famine, most Irish immigrants were

Immigration: America’s Nineteenth-Century ‘‘Law and Order Problem?’’

Table 8.

317

Percentage Irish, German, and British prison versus general population, white males Prison population

General population ages 18–44

Irish 1840 1850 1860

14.2% 12.5 12.9

German 1840 1850 1860

6.3 11.2 11.2

6.6 8.8

British 1840 1850 1860

4.2 5.8 4.4

3.9 4.0

10.7% 12.0

Note: See notes to Table 1.

reasonably well-to-do Protestants. After the famine, the majority were poor Catholics (Handlin, 1959; McCaffrey, 1976; Ignatiev, 1995; Farrell, 2003 disputes this). Similarly, the political and social upheaval in midcentury Germany changed the nature of German immigration. Second, we investigate these three groups for the very practical reason that they are the most commonly observed immigrants in the prison data and afford large enough sample sizes to make meaningful comparisons. In Table 8 we compare the proportions of the three principal immigrant groups in the prison population to those in the general population of 18- to 44-year-old men. Two features stand out. First, the proportions of German and British immigrants in the prison population exceed the proportion of these two groups in the overall population in both 1850 and 1860, though the German prison overrepresentation is notably larger than for British immigrants. Second, despite deep contemporary social concerns with the big wave of poor Irish immigrants at mid-century, the prisons were not overrun with Irish immigrants. The proportion Irish in the prison mirrors the proportion Irish in the population. This is despite the contemporary diarist George Templeton Strong’s now infamous quote that ‘‘our Celtic fellow citizens are almost as remote from us in temperament and constitution as the Chinese’’ (quoted in Wittke, 1956, p. 40). It turns out, at least as far as their overall participation in serious crime is concerned, that the ‘‘new’’ Irish were not very remote from the native born. Table 9 reports incarceration rates for the three principal immigrant groups using the ﬁner age categories used in Table 5. It is difﬁcult to draw any generalizations about differential criminality across the three

318

Table 9.

Howard Bodenhorn et al.

Incarceration rates of Irish, German, and British male immigrants by age (per 100,000)

Age category

Irish

Germans

British

1850 16–17 18–19 20–24 25–29 30–34 35–39 40–44 45–49 50–54 W55

87 69 115 69 59 42 47 59 23 9

33 129 106 77 73 132 120 68 55 43

0 28 183 135 38 89 74 28 42 27

1860 16–17 18–19 20–24 25–29 30–34 35–39 40–44 45–49 50–54 W55

0 153 154 104 56 69 58 27 65 43

0 69 170 116 97 85 64 77 78 58

180 67 117 150 72 49 62 18 32 18

Note: See notes to Table 1.

immigrant groups in 1850. At some ages, German and British immigrants are more crime prone than the Irish; at others, they are less. The only systematic difference appears to be a somewhat higher criminal propensity among German immigrants aged 30 and above. In 1860, German incarceration rates continued to exceed those of the Irish and the British in most age groups. More noteworthy, perhaps, are the rising rates of criminality among the Irish and the Germans between 1850 and 1860. Except for those 19 years and below, incarceration rates for these two groups rose over the decade, sometimes dramatically. British immigrants display a declining tendency toward incarceration over the decade, especially for those 35 years and older. This may reﬂect the greater ease with which the British assimilated into American society, but assigning any cause to such changes remains speculative without further research. In Table 10, we again consider differential offending by narrow types of crime across the three groups. It may have been the Irish immigrants’ greater propensity toward violent criminal acts – homicide and person crimes – that moved Strong to consider the Irish a breed apart. The Irish were nearly ﬁve times more likely to be committed for homicide than

319

Immigration: America’s Nineteenth-Century ‘‘Law and Order Problem?’’

Table 10.

Distribution of commitments by type of crime, Irish, German, and British male immigrants, 1830–1862

Property crime Crime against a person Homicide Arson Crime against public order Felony, not speciﬁed Moral crime Misdemeanor, not speciﬁed

Irish

Germans

British

59.3% 13.0 16.6 2.9 5.7 1.3 1.1 0.0

84.9% 6.0 3.5 3.1 0.8 1.4 0.4 0.0

80.8% 10.2 3.4 1.9 1.1 0.4 1.9 0.4

Note: See notes to Table 6.

Table 11.

Incarceration rates by crime type and country of birth (per 100,000 persons ages 18–44)

Crime type

Native born

Irish

Germans

British

1850 Violent Property Other

10 47 5

27 46 11

15 102 6

25 72 9

1860 Violent Property Other

17 67 10

55 38 15

21 94 13

27 83 1

Note: Violent crimes deﬁned as homicide plus all crimes against persons.

Germans or Britons. The Irish were nearly twice as likely as the Germans to be committed for a crime against persons (assault mostly). The image of the young Irish tough popularized in Martin Scorsese’s Gangs of New York may have been realistic if not fully accurate. The evidence lends further support to McCaffrey’s (1976) contention that the Irish were the law and order problem of the 1850s. Table 11 reports incarceration rates (per 100,000) for the Irish, Germans, and British and reproduces the values of the native born (from Table 7) for comparative purposes. In 1850 immigrants were incarcerated at much higher rates than the native born for violent crime, generally at two to three times the native-born rate. By 1860 nativeborn violent incarceration rates had doubled over their 1850 value, but they still remained lower than those for immigrants. Violence, or at least incarcerations for violent crimes, more than doubled for the Irish, reaching an astounding 55 per 100,000 in the years around 1860. Differences in

320

Howard Bodenhorn et al.

property and other crimes pale compared to Irish violence, but it is notable that the rate of property incarcerations among German immigrants is substantially higher than those for the Irish or the native born. Thus, the snapshot of incarcerations around the census years yields a conclusion similar to inﬂow of commitments. The Irish were violent; at the very least, they were convicted for violent acts at a much higher rate than any other group, whether immigrant or native born.

7. Concluding remarks Our analysis of prison data from Pennsylvania during the middle of the nineteenth century provides some insight into the empirical relationship between immigration and crime during the ﬁrst major wave of immigration, a time when concern about immigrant criminality was extremely high. We ﬁnd that adjusting for age and gender greatly narrows the observed gap in incarceration rates of the foreign and native born. Adjusting for race broadens the gap. In 1850 and especially in 1860, incarceration rates for immigrants were much higher for immigrants than for natives for every age group 18 and older. Different patterns by urban/ rural geography are particularly striking. Immigrants were concentrated in the urban counties of Philadelphia and Allegheny where reported crime rates were higher. Within rural counties, the foreign born had much higher incarceration rates than the native born. By 1860, the nativity gap is so large that the rural foreign-born incarceration rates are the same order of magnitude or even higher than those for urban foreign born. For natives, the rural rates are generally half the size of those of urban native born. The interaction of nativity with urban residence explains much of the observed aggregate differentials in incarceration rates. Finally, we found that the foreign born, especially the Irish, consistently have higher incarceration rates for violent crimes, but from 1850 to 1860 the natives largely closed the gap with the foreign born for property offenses. The conclusions for this analysis of prison data is largely in accord with the ﬁndings of Monkkonen and Lane regarding high levels of participation of immigrants in urban violent crime during this period of American history. But in contrast to these scholars, the results from using a broader measure of criminality reveal little gap between immigrants and natives for nonviolent crime. In addition, we have uncovered a dramatic difference between how immigrants fared in cities and rural areas.

Acknowledgments We thank Veronica Hart and Jennifer Chang for exceptional research assistance in entering and cleaning the Pennsylvania prison data.

Immigration: America’s Nineteenth-Century ‘‘Law and Order Problem?’’

321

Bodenhorn thanks the Irish American Cultural Institute, Lafayette College, and the National Science Foundation (SES-0109165) for ﬁnancial support.

References Berger, H., Spoerer, M. (2001), Economic crises and the European revolutions of 1848. Journal of Economic History 61 (2), 293–326. Bursik, R.J. (2006), Rethinking the Chicago school of criminology: a new era of immigration. In: Martinez, R., Jr., Valenzuela, A., Jr. (Eds.), Immigration and Crime. New York University Press, New York, pp. 20–35. Calomiris, C., Schweikart, L. (1991), The panic of 1857: origins, transmission, and containment. Journal of Economic History 51 (4), 807–834. Carter, S.B., Sutch, R. (2006), U.S. immigrants and emigrants, 1820–1998, Table Ad1-2. In: Carter, S.B., Gartner, S.S., Haines, M.R., Olmstead, A.L., Sutch, R., Wright, G. (Eds.), Historical Statistics of the United States, Earliest Times to the Present: Millennial Edition. Cambridge University Press, New York. Christian Watchman. (1847), Pauper Immigration. 28 (July 2), p. 106. Du Bois, W.E.B. (1899), The Philadelphia Negro: A Social Study. University of Pennsylvania, Philadelphia. Eastern State Penitentiary. (1829–1857), Population records, descriptive register. Record Group 15. Records of the Department of Justice. Pennsylvania State Library, Harrisburg. Eastern State Penitentiary. (2010), http://www.easternstate.org/history/ Eliott, J.H. (1869), The increase of material prosperity and of moral agents, compared with the state of crime and pauperism. Merchants’ Magazine and Commercial Review 61 (October), 339–365. Farrell, H.F. (2003), The Irish in New Orleans, colonial period to 1860. Unpublished MA Thesis, University of New Orleans. Fishlow, A. (1965), American Railroads and the Transformation of the Ante-Bellum Economy. Harvard University Press, Cambridge, MA. Friedman, L.M. (1993), Crime and Punishment in American History. BasicBooks, New York. Furstenberg, F.F. (1979), What happened when the census was redone: an analysis of the recount of 1870 in Philadelphia. Sociology and Social Research 63, 475–505. Gallman, J.M. (1990), Mastering Wartime: A Social History of Philadelphia during the Civil War. Cambridge University Press, Cambridge, MA. Glaeser, E.L., Sacerdote, B. (1999), Why is there more crime in cities? Journal of Political Economy 107 (6), S225–S258.

322

Howard Bodenhorn et al.

Gurr, T.R. (1989), Historical trends in violent crime: Europe and the United States. In: Gurr, T.R. (Ed.), Violence in America. Volume 1: The History of Crime. Sage Publications, New York, pp. 21–54. Handlin, O. (1959), Boston’s Immigrants: A Study in Acculturation, revised ed Belknap Press of Harvard University Press, Cambridge, MA. Ignatiev, N. (1995), How the Irish became White. Routledge, New York. Jones, M.A. (1974), American Immigration. University of Chicago Press, Chicago. Lane, R. (1979), Violent Death in the City. Harvard University Press, Cambridge, MA. Lane, R. (1989), On the social meaning of homicide trends in America. In: Gurr, T.R. (Ed.), Violence in America. Volume 1: The History of Crime. Sage Publications, New York, pp. 55–79. Langbein, J.H. (1978), The criminal trial before the lawyers. University of Chicago Law Review 45 (2), 263–316. Martinez, R., Jr., Lee, M.T. (2000), On immigration and crime. In: LaFree, G. (Ed.), The Nature of Crime: Continuity and Change. National Institute of Justice, Washington, DC, pp. 485–524. Martinez, R., Jr., Nielsen, A.L., Lee, M.T. (2003), Reconsidering the Marielito legacy: race/ethnicity, nativity, and homicide motives. Social Science Quarterly 84 (2), 397–411. McCaffrey, L.J. (1976), The Irish Diaspora in America. Indiana University Press, Bloomington. Moehling, C., Piehl, A.M. (2009), Immigration, crime, and incarceration in early twentieth century America. Demography 46 (4), 739–763. Monkkonen, E.H. (1981), Police in Urban America, 1860–1910. Cambridge University Press, Cambridge, MA. Monkkonen, E.H. (1989), Diverging homicide rates: England and the United States, 1850–1875. In: Gurr, T.R. (Ed.), Violence in America. Volume 1: The History of Crime. Sage Publications, New York, pp. 80–101. Naylor, T.J. (1979), Criminals, crime and punishment in Philadelphia, 1866–1916. Unpublished PhD Dissertation, University of Pennsylvania. New York Daily Times. (1853), Imported Crime. September 24, p. 4. New York Times. (1858), Foreign Criminals in New York. February 22, p. 4. Reiss, A.J., Duncan, O.D., Hatt, P.K., North, C.C. (1961), Occupations and Social Status. Glencoe Press, New York. Rice, J.D. (1996), The criminal trial before and after the lawyers: authority, law, and culture in Maryland jury trials, 1681–1837. American Journal of Legal History 40 (4), 455–475. Ruggles, S., Sobek, M., Alexander, T., Fitch, C.A., Goeken, R., Hall, P.K., King, M., Ronnander, C. (2008), Integrated Public Use Microdata Series: Version 4.0, (machine-readable database). . Minnesota Population Center, Minneapolis, MN, (producer and distributor). Available online at http://usa.ipums.org/usa/

Immigration: America’s Nineteenth-Century ‘‘Law and Order Problem?’’

323

Sabol, W.J., West, H.C., Cooper, M. (2009), Prisoners in 2008. Bureau of Justice Statistics Bulletin, NCJ-228417, U.S. Department of Justice, Washington, DC. Sellin, T. (1938), Culture, Conﬂict, and Crime. Social Science Research Council, New York, (Bulletin 41). Sharpless, J.B., Shortridge, R.M. (1975), Biased underenumeration in census manuscripts: methodological implications. Journal of Urban History 1 (August), 409–439. Steckel, R.H. (1991), The quality of census data for historical inquiry: a research agenda. Social Science History 15 (Winter), 579–599. Taft, D.R. (1933), Does immigration increase crime? Social Forces 12 (October), 69–77. Thomas, B. (1954), Migration and Economic Growth. Cambridge University Press, Cambridge, MA. U.S. Bureau of the Census (1853), The Seventh Census: Report of the Superintendent of the Census for December 1, 1852. Robert Armstrong, Printer, Washington. U.S. Bureau of the Census (1854), Statistical View of the United States: Compendium of the Seventh Census. Beverley Tucker, Senate Printer, Washington. Western State Penitentiary. (1826–1876), Convict docket & descriptive register. Record Group 15. Records of the Department of Justice. Pennsylvania State Library, Harrisburg. Wittke, C. (1956), The Irish in America. Louisiana State University Press, Baton Rouge.

CHAPTER 13

A Political Economy of the Immigrant Assimilation: Internal Dynamics Gil S. Epsteina,b,c and Ira N. Gangb,c,d a

Department of Economics, Bar-Ilan University, Ramat Gan 52900, Israel E-mail address: [email protected] b Institute for the Study of Labor (IZA), Bonn, Germany c CReAM-Center for Research and Analysis of Migration, London, UK d Department of Economics, Rutgers University, New Brunswick, New Jersey, 08901-1248, USA E-mail address: [email protected]

Abstract Within immigrant society, different groups wish to help the migrants in different ways – immigrant societies are multilayered and multidimensional. We examine the situation where there exists a foundation that has resources and that wishes to help the migrants. To do so, they need migrant groups to invest effort in helping their country folk. Migrant groups compete against one another by helping their country folk and to win grants from the foundation. We develop a model that considers how such a competition affects the resources invested by the groups’ supporters and how beneﬁcial it is to immigrants. We consider two alternative rewards systems for supporters – absolute and relative ranking – in achieving their goals. Keywords: Immigrants, assimilation, competition JEL classiﬁcations: J61, J71

1. Introduction Immigrant societies are multilayered and multidimensional offering many perspectives, some of which may come into conﬂict with others, leading to the development of rivalrous strategies, at least partly overlapping loyalties of supporters, and the necessity of laying claim to having the bigger impact. Supporters of each perspective invest resources and effort into convincing the general body of immigrants of the virtue of their point of view. These rivalries can be characterized as contests – each side struggles, investing substantial effort, pushing their own agenda in helping their immigrant society. Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008019

r 2010 by Emerald Group Publishing Limited. All rights reserved

326

Gil S. Epstein and Ira N. Gang

In our society, there exists a foundation that has resources and wishes to help migrants. To help migrants the foundation works indirectly, offering grants to groups who directly invest efforts to help migrants (e.g., in the United States, we have the MacArthur Foundation and the Ford Foundation that want to help immigrants.) The foundation offers a prize (grant) for which the groups compete. The competition is such that the one that invests more resources in helping migrants has a higher probability of winning and obtaining even more resources. We address how the foundation elicits the most effort from the different ‘‘grass roots’’ groups. Each group – that is, a part of immigrant society that possesses a common perspective and acts to achieve it – wants the authority and rewards for implementing its own plan, believing its proposal will best help its countrymen. The groups may aim to achieve a certain degree of assimilation on the part of immigrants, though each group has its own strategy. They may differ on the degree of cultural identity they want to maintain with their birthplace (see, e.g., Lazear, 1999; Alesina and Eliana La, 2000; Bisin and Verdier, 2000; Gang and Zimmermann, 2000; Anas, 2002; Dustmann et al., 2004; Kahanec, 2006). Each group seeks to lead immigrant society, and capturing the prize rewarded by the foundation. The key to our analysis of who wins the contest is the contest rule structure. Studies of immigrants around the world show, with few exceptions, their earnings are substantially below those of comparable majority workers (Smith and Welch, 1989; Blau and Kahn, 1997, 2006, 2007; Altonji and Blank, 1999; Bhaumik et al., 2006). Partly, this reﬂects a failure on the part of the immigrants to undertake the effort to assimilate with the local community (Constant et al., 2009). ‘‘Lack of effort’’ can arise from the desire to maintain a cultural heritage or separate identity which would be lost or reduced if the group assimilated. The failure to take active steps to assimilate can also arise in the face of high adjustment costs, such as inadequate language skills, intergenerational familial conﬂicts, and, in the case of immigrants, lack of knowledge about the host country labor market (Chiswick and Miller, 1995, 1996; Bauer et al., 2005). Yet for immigrants and their descendants, as length of time in the host country increases, assimilation generally creeps in and various immigrant labor market indicators approach those of comparable majority workers. On occasion, they outperform the native born (Chiswick, 1977; Deutsch et al., 2006). Efforts made to assimilate, and time, are two elements working to bring immigrants onto line with the native born. A third element, the degree to which the local society welcomes immigrants, also plays a role. Often, the local society is less than welcoming, blaming migrants for depressing wages and displacing native-born workers – that is, causing unemployment. This presumption has very strong policy implications and is implicit, for example, in the calls for increased regulation of immigration heard worldwide. Yet, there is mixed evidence on the impact of immigrants on the local’s wages and employment – it depends on whether they are

A Political Economy of the Immigrant Assimilation

327

substitutes or complements with respect to the skills and other attributes they bring to the labor market (Gang and Rivera-Batiz, 1994; Gang et al., 2002; and this volume). Whether immigrants actually lower wages and increase employment, or not, the perception exists that they do so. Because of this perception, the local population may take active steps to discourage immigrants’ assimilation – discrimination, isolation, and so on. For this reason, effort is needed to decrease the barriers between the local population and the migrants. Epstein and Gang (2009a) are interested in why migrants are so often at a disadvantage relative to the native born, the circumstances under which their status changes or stagnates over time, and role public policy can play. Often the efforts of the immigrants and the local population are mediated through political institutions. These institutions exist in both worlds. They could be, for example, political parties, trade organizations, unions, or thugs. These are organizations that are able to overcome the free-rider problem individual members of each group have in moving from the actions they desire to take, to actually taking the actions. Yet, while an organization’s purpose may be to represent the members of their group, the interests’ of the organization and that of its members do not always coincide. Assimilation efforts by migrants, harassment by the native born, and time are the three elements that determine how well migrants do. Epstein and Gang (2009a) examine the consequences for these increases in the numbers of immigrants, time, and the role of politics. They construct a model in which there are four actors: the members of the majority and the organization that represents them and members of the minority and the organization that represents them. Over time, the organization representing the migrants and immigrants themselves may exhibit different interests in assimilating and in maintaining their cultural identity. They discuss how this affects the migrants’ position over time and discuss the public policy implications of the model. In this chapter, we describe and compare two mechanisms for rewarding groups for their efforts. Absolute ranking is a contest between the groups where the winning one receives all the grants – those who put forth the most effort win. In this situation, the simultaneous bidders are the groups, and their bids are the actions/investments they undertake. Those that take the most action, or those that are perceived to have taken the most action, win and acquire all the grants. On the other hand, in relative ranking, the groups compete against each other and obtain grant relative to the amount of effort invested in the contest. This can be seen as a lottery contest in which each obtains grant proportional to the effort invested. In both cases, in equilibrium, the grant obtained is a function of the efforts invested. The structure of the contest can be a key element determining the direction immigrant society goes. The foundation wishes to maximize the efforts made by the groups to help immigrant society. We develop economic theory that considers how such competition affects the resources

328

Gil S. Epstein and Ira N. Gang

invested by each group and the performance of immigrant society. We wish to see how these two alternative rewards systems – absolute and relative ranking – affect the implementation and achievement of the foundations’ goals. For us, the question is in what situations – and for whom – is an absolute ranking of groups desirable, and in what circumstances – and for whom – is such a ranking a detriment vis-a-vis a relative ranking scheme.1 The next section ﬁrst describes the model. It implements the relative and absolute decision rules in the context of the model and compares the implications for each of the concerned parties. A concluding section follows.

2. The model Consider the case where there exist m groups in the immigrant society that have partly overlapping programs. Each group has the same objective in terms of helping and ﬁnding solutions to problems of their society. Each obtains a reward for helping their fellow immigrants. The reward is a grant (or grants) from the foundation. The maximum reward group i (i ¼ 1, 2) receives by helping its countrymen is ni, where ni can be greater or smaller than nj for all j 6¼ i, depending on which group has more to gain. In probabilistic terms, the probability that group i wins the contest and receives a grant of ni is equal to Pri. The expected grant group i receives from this contest is Prini. Alternatively, we can think of Pri as the proportion of the grant (or a proportion of the grants rewarded) this group receives in the competition. We talk generally about proportions of the grant obtained and not probabilities of winning the contest, keeping in mind that the two are equivalent. Groups invest effort trying to help immigrant society. Effort, xi, can be seen as a monetary value, time, effort, etc., and we assume that the cost of each unit of effort is one unit. Own effort, the efforts invested by the other group, and the stakes and the contest success function (CSF) determine the probability of winning the contest. Let w denote the net payoff received by a group. The expected net payoff (surplus) for the risk neutral group is given by Eðwi Þ ¼ Pri ni xi

1

8i ¼ 1; 2; . . . ; m.

(1)

This is an implementation of the chapter on aid allocation developed by Epstein and Gang (2009b) into the type of questions asked and discussed in Epstein and Gang (2009a).

A Political Economy of the Immigrant Assimilation

329

We assume that the proportion of the grant obtained in the contest (or the probability of winning the contest) satisﬁes the following conditions: 1. The sum of the proportions of the grant obtained equals one, Sm i¼1 Pri ¼ 1. This means that the foundation will only give one of the groups the grant. An alternative explanation would be that both groups get credit for what they did and as such obtain a proportion of the grant they could have won if there was only one group winning. 2. As a group i increases its effort, it obtains a higher proportion of the grant, @Pri =@xi 40: 3. As group j, the opponent of group i, increases its effort, the proportion of the grant that group i obtains decreases, @Pri =@xj o0. 4. The marginal increase in the proportion of the grant obtained from the contest decreases with investment in effort, @2 Pri =@x2i o0 (this inequality ensures that the second-order conditions for maximization are satisﬁed). 5. To simplify, we do not discuss the possibility of free riding for the different groups. One could think of a situation under which the actions of one group positively affect the proportion obtained by the other group, as the people do not always know which of the groups was really responsible for the outcome. We overcome this by assuming @Pri =@xi 40 and Sm i¼1 Pri ¼ 1. The function Pri(.) is usually referred to as a CSF. The functional forms of the CSFs commonly assumed in the literature satisfy these assumptions (see Nitzan, 1994). The groups engage in a contest and we assume a Nash equilibrium outcome. Each group determines the level of its activities xi so that its expected payoff, Eðwi Þ 8i ¼ 1; 2; . . . ; m, is maximized. The ﬁrst-order condition for maximization is given by @Eðwi Þ @Pri ¼ ni 1 ¼ 0. @xi @xi

(2)

Equation (2) is satisﬁed if and only if @ Pri 1 ¼ . @xi ni

(3)

Thus, given that the proportion has decreasing marginal utility with respect to the level of effort invested, the group with the higher beneﬁt from the contest will invest more effort in the contest. For example, if group 1 has the higher beneﬁt in the contest compared to group 2, n1Wn2, then group 1 will determine its effort, x1, such that the marginal proportions are @Pr1 =@x1 o@Pr2 =@x2 ; in order to increase its proportion of the grant. The group that has a higher beneﬁt from winning the contest will invest the highest amount of effort.

330

Gil S. Epstein and Ira N. Gang

To simplify and without loss of generality, assume that n1 n2 4n3 nm .

(4)

This assumption simply states that there are two groups that have higher stakes than all the rest of the groups. We now describe two highly stylized (extreme) regimes: (1) absolute ranking and (2) relative ranking. In the absolute ranking, we have one winning group even though both groups helped immigrant society. Here, the winner of the contest obtains all the grants. On the other hand, in relative ranking, the two groups divide the grants relative to their achievements. These situations do not simultaneously coexist. However, comparing their outcomes provides useful insights, and we compare them after fully detailing each of the scenarios.

2.1. The absolute ranking This ranking states that the group investing the largest amount of effort wins the grant. This type of contest is deﬁned by using the all-pay auction, and here thinking in terms of the probability of winning the contest enhances our intuition. In the absolute ranking, the probability of winning is a function of the efforts invested by groups or the efforts perceived by the foundation. (Note that in equilibrium, the efforts will be a function of the grants the groups can obtain). In the absolute ranking, the probability of winning is 8 1 if xi 4xj 8i a j > > > <1 if i ties for the high bid with k 1 others . (5) Pri ¼ > k > > : 0 if xj 4xi 8i a j It can be veriﬁed that there exists a unique symmetric Nash equilibrium as well as a continuum of asymmetric Nash equilibria. In any equilibrium, groups 3 through m invest zero effort in activities with probability one (see Baye et al., 1996) so that only the two groups who make the greatest efforts will participate. We conduct our analysis for two groups, groups 1 and 2. Without loss of generality, assume that n1Wn2; thus, group 1 has greater gain from winning the contest. It is clear, therefore, that group 1 is able to bid more than group 2. However, it is not clear how much each will bid in equilibrium. Based on these ﬁndings, we can obtain equilibrium expected expenditures, equilibrium probabilities, and expected payoffs. Since the efforts of the two groups are random variables, it is clear that the probability that xi ¼ xj equals zero (P(xi ¼ xj) ¼ 0) (see Cohen and Sela, 2007). Thus, in the case of only two groups, the probability

A Political Economy of the Immigrant Assimilation

of winning the grant becomes ( 1 if xi 4xj Pri ¼ . 0 if xj 4xi

331

(6)

The expected activity level for each group is (see appendix for calculations) Eðx1 Þ ¼

n2 n2 and Eðx2 Þ ¼ 2 . 2 2n1

(7)

The equilibrium probability of winning the contest for each group equals Pr1 ¼

2n1 n2 n2 and Pr2 ¼ . 2n1 2n1

(8)

The expected equilibrium payoff for each group equals Eðw1 Þ ¼ n1 n2 and Eðw2 Þ ¼ 0.

(9)

In equilibrium, the total amount of activities carried out by the groups equals EðX Þ ¼ Eðxi þ xj Þ ¼

n22 þ n2 n1 n2 ðn2 þ n1 Þ ¼ . 2n1 2n1

(10)

Notice that if both groups can obtain the same beneﬁt, n1 ¼ n2 ¼ n, the expenditure of each group is Eðx1 Þ ¼ n=2 and Eðx2 Þ ¼ n=2; the probability of winning for each equals one half, Pr1 ¼ Pr2 ¼ 1=2; the expected payoff for each group is zero, Eðw1 Þ ¼ Eðw2 Þ ¼ 0; and the total effort invested equals X* ¼ n. 2.2. The relative ranking Here, we consider the case when groups compete with one another in a contest in which there is no single winner. Later we will compare the two extreme cases with one another: the absolute ranking with the relative ranking. Without a winner taking all the grants, each group ﬁghts to obtain its maximum possible portion. We assume that the contest is characterized by the relative ranking (Lockard and Tullock, 2001), Pri ¼ xri =ðxrj þ xri Þ for r 2. The return to effort in this lottery function is captured by the parameter r. When r approaches inﬁnity, the relative ranking becomes the absolute ranking under which the group that invests in the highest level of activities wins the contest (see Baye et al., 1993, 1996). The idea behind this is that the group with the higher beneﬁt has a weight of inﬁnity and thus

332

Gil S. Epstein and Ira N. Gang

will win with probability one, and the group with the lower stake will lose with probability one. For now we assume that r is known and ﬁxed and rr2. The expected net payoff (surplus) for the risk neutral group is thus given by Eðwi Þ ¼

xri

xri ni xi þ xrj

8i ¼ 1; 2.

(11)

The ﬁrst-order condition, as stated in Equation (2), which ensures that the group maximizes its expected payoff, is given by r rxr1 @Eðwi Þ i xj ¼ r ni 1 ¼ 0 @xi ðxi þ xrj Þ2

8i; j ¼ 1; 2 i a j.

(12)

Denote by xi 8i; j ¼ 1; 2 i a j the Nash equilibrium outcome of the contest. Solving Equation (12) for both groups using a Nash equilibrium, we obtain that the level of activities each group participates equals2 xi ¼

r rnrþ1 i nj

ðnri þ nrj Þ2

8i; j ¼ 1; 2 i a j.

(13)

We can also think of this term of the proportion of the grants obtained from the contest. Therefore, the Nash equilibrium proportion of the grants obtained in the contest equals Pri ¼

nri nri þ nrj

8i; j ¼ 1; 2 i a j.

(14)

The expected equilibrium payoff for each group equals Eðwi Þ ¼

r r rnrþ1 n2rþ1 ðr 1Þnrþ1 nri i nj i i nj n ¼ i nri þ nrj ðnri þ nrj Þ2 ðnri þ nrj Þ2

(15)

8i; j ¼ 1; 2; i a j; ro2. And ﬁnally, we can calculate the total amount of effort invested in the contest by the two groups. In the chapter, this measure is called grant dissipation and usually has a negative connotation, that is, the contest designer tries to decrease the grant dissipation. Here, grant dissipation can be seen in a positive light as it helps the country needing help. 2 We obtain from the ﬁrst-order conditions (Equation (12)) that 8i; j ¼ 1; 2 i a j; ððrxir1 xrj Þ=ðxri þ xrj Þ2 Þni ¼ 1; therefore, it holds that ððrx1r1 xr2 Þ=ðxr1 þ xr2 Þ2 Þn1 ¼ 1 and ððrx2r1 xr1 Þ=ðxr1 þ xr2 Þ2 Þn2 ¼ 1. Using these two equations, we obtain that ðx2 =x1 Þðn1 =n2 Þ ¼ 1 and thus x2 ¼ x1 ðn2 =n1 Þ. Substituting x2 (x2 ¼ x1 ðn2 =n1 Þ) into ðrx1r1 xr2 Þ=ððxr1 þ xr2 Þ2 Þ n1 ¼ 1, we obtain that x1 ¼ ðr n1rþ1 nr2 Þ=ðnr1 þ nr2 Þ2 . In a similar way, we calculate the optimal level of x2.

A Political Economy of the Immigrant Assimilation

333

We denote this total effort in equilibrium by X*: X ¼ xi þ xj ¼

rnri nrj ðni þ nj Þ ðnri þ nrj Þ2

8i; j ¼ 1; 2 i a j.

(16)

In the case where the groups are symmetric, that is, n1 ¼ n2 ¼ n, we would obtain the following: the level of activities of each group equals xi ¼ nðr=4Þ 8i; j ¼ 1; 2 i a j (remember that r is less than or equal to 2 and therefore the total expenditure will be at the maximum when xi ¼ ni =2); the Nash equilibrium proportion of the grants obtained from the contest will be equal to one half, Pri ¼ 1=2; the expected equilibrium payoff to each group equals (2r)n/4 (once again, remember that r is less than or equal to 2),3 and ﬁnally the total effort in equilibrium equals X* ¼ rn/2. Let us consider how changes in r affect the expected equilibrium payoff, Eðwi Þ ¼

r r rnrþ1 n2rþ1 ðr 1Þnrþ1 nri i nj i i nj n ¼ i nri þ nrj ðnri þ nrj Þ2 ðnri þ nrj Þ2

8i; j ¼ 1; 2; i a j; ro2, and how total effort is affected in equilibrium, X ¼ xi þ xj ¼

rnri nrj ðni þ nj Þ ðnri þ nrj Þ2

8i; j ¼ 1; 2 i a j.

To simplify our calculations denote by a the relative beneﬁt of the second group receiving the grant in relationship to that of the ﬁrst group receiving the grant: a ¼ n2/n1. Given a, we recalculate the expected payoff and total effort in equilibrium as Eðwi Þ ¼

ni ð1 ðr 1Þar Þ rnar ð1 þ aÞ aj and X ¼ 8a ¼ ; i ¼ 1; 2; ro2, 2 2 r r ai ð1 þ a Þ ð1 þ a Þ

where ð@Eðwi ÞÞ=ð@rÞ ¼ ðni ð1 þ a2r ð1 þ ðr r2 ÞLnðaÞÞ þ rar ð2 þ ðr2 3Þ LnðaÞÞÞÞ= ðð1 þ ar Þ2 Þ and ð@X Þ=ð@rÞ¼ ðnar ð1 þ aÞð1 þ rLnðaÞ þ ar ð1 rLnðaÞÞÞÞ=ðð1 þ ar Þ3 Þ:

As we can see from the above, the effect of a change in the parameter r has an ambiguous affect on the expected payoff and expenditure of the groups. For example, without pﬃﬃﬃ loss of generality, assume that ao1. Since Ln(a)o0 then for 1oro 3, @Eðwi Þ=@r40 and for a ¼ 1, @Eðwi Þ=@r40 and @X =@r40. For ro1, it holds that @X =@r40. 3

For rW2, the equilibrium differs from this one as it is based on mixed and not pure strategies. This is the case in the all-pay auction that we previously described.

334

Gil S. Epstein and Ira N. Gang

2.3. Comparing the investment of effort of the groups under both situations The groups do not have a choice between the absolute ranking and relative ranking contests we model above. They face what they face. Over time what they face may change, and we are interested in the outcomes of each of the situations. We now compare these two types of contests from the perspectives of the groups, immigrant society, and the foundation. X* gives the aggregate activity of the groups in equilibrium (for the case of stakes that do not depend on the efforts invested by the contestants, see Epstein and Nitzan, 2006a, 2006b, 2007). Under the relative ranking, Pri ¼ xri =ðxrj þ xri Þ for r 2, from Equation (16), we obtain that the total amount of activities carried out is equal to X L ¼ xi þ xj ¼ ðrnri nrj ðni þ nj ÞÞ=ððnri þ nrj Þ2 Þ 8i; j ¼ 1; 2 i a j. In order to simplify our analysis, let us assume that r ¼ 1 (remember that the values that r can take on in this case are between 2 and 0). Under the absolute ranking, from Equation (10), we obtain that the total investment into activities is equal to EðX p Þ ¼ n2 ðn2 þ n1 Þ=ð2n1 Þ. The total amount of expenditure invested in the contest is higher under the relative ranking than under the absolute ranking regime if X L ¼

n2 n1 n2 ðn2 þ n1 Þ 4 ¼ EðX p Þ. n1 þ n2 2n1

(17)

Equation (17) holds if and only if n21 2n1 n2 n2 40.

(18)

From Equation (18), we may conclude that the total amount of expenditure invested in the contest by the different groups is higher under the relative ranking rather than under the absolute ranking regime if pﬃﬃﬃ (19) n1 4n2 ð1 þ 2Þ. Since, by assumption, n1Zn2, the result tells us that in order for the lottery contest to be worse for the receiving, the grant that one of the groups can obtain from such actions must be larger than the other group’s grant (more than twice as large). We summarize this result in the following proposition: If the variance of grants that can be generated pﬃﬃﬃ by helping immigrant society is sufﬁciently large, that is, n1 =n2 41 þ 2, then the foundation – which is interested in maximizing the total effort of the groups – prefers that the absolute ranking contest where the group that invests the most effort wins. If each group has the same stake, that is, n1 ¼ n2, then the foundation prefers the relative ranking. In order to analyze the preferences of the groups, we must compare their expected payoffs under both the relative ranking and the absolute ranking regime. Remember that we assumed, without loss of generality,

A Political Economy of the Immigrant Assimilation

335

that group 1 has at least as large a stake as the second group (n1Zn2). The groups prefer the regime that generates for them the maximum expected equilibrium payoff, Eðwi Þ. Under the relative ranking, and again assuming r ¼ 1, the expected equilibrium payoff for group 2 (the weaker player) equals Eðw2 Þ ¼ n32 =ðn1 þ n2 Þ2 ; while the expected equilibrium under the absolute ranking equals zero, Eðw2 Þ ¼ 0. Therefore, it is clear that The weaker group, the group that has less to gain from helping its countrymen, always prefers the relative ranking system. For the stronger group, the expected equilibrium payoff under the relative ranking equals E L ðw1 Þ ¼ n31 =ðn1 þ n2 Þ2 ; while the expected equilibrium under the absolute ranking equals E P ðw1 Þ ¼ n1 n2 . The expected payoff for group 1 under the relative ranking regime is greater than that obtained under the absolute ranking regime and thus this group prefers the relative ranking regime if E L ðw1 Þ ¼

n31 4n1 n2 ¼ E p ðw1 Þ ðn1 þ n2 Þ2

(20)

Equation (20) holds if and only if n21 2n1 n2 n2 o0.

(21)

From Equation (21), we may conclude that the expected payoff in the contest and efforts made are higher under the relative ranking rather than under the absolute ranking regime if pﬃﬃﬃ (22) 0on1 on2 ð1 þ 2Þ. In other words, The group with the higher stake, with more to gain from helping their countrymen, prefers the relative ranking to an absolute ranking pﬃﬃﬃif the difference between the groups is not sufﬁciently large, n1 =n2 o1 þ 2. Note that the interests of foundation and the strongest group always align. 3. Conclusion In our society, there exists a foundation that has resources and that wishes to help migrants. In order to help migrants, the foundation needs grass roots organizations to invest their efforts in helping migrants (e.g., in the United States, the McArthur Foundation and the Ford Foundation that want to help immigrants.) The foundation announces a prize (grant) for which the groups compete. The competition is such that the one that invests more resources in helping the migrants has a higher probability of winning and obtaining more resources. The question is how the foundation elicits the most effort from the different ‘‘grass roots’’ groups.

336

Gil S. Epstein and Ira N. Gang

In a highly structured and simple model, we characterize and compare two ex ante regimes: (1) the absolute reward scheme presented by an all-pay auction in which the winner takes all available grants and (2) the relative reward scheme in which the grant allocation rule is a lottery and each group obtains a proportion of the grants. In the former regime, the equilibrium is in mixed strategies, the ‘‘stronger’’ group could actually lose the contest and get nothing. However, the expected payoff for the weaker group is zero. The contests we address are the fractious relationships among groups seeking to increase their expected payoff. We are able to derive a very speciﬁc condition allowing us to see when each of the concerned parties wins and when each loses their contests. If the difference between the groups in terms of the rewards they can obtain from helping the country is not sufﬁciently large, all parties – the two groups and the foundation itself – prefer the lottery regime relative ranking to an absolute ranking. However, if the difference between the groups in terms of the rewards that can be obtained is sufﬁciently large, then the group with the low beneﬁt, group 2, prefers the relative ranking regime while the other group and the foundation prefer the absolute ranking. The contests we address are the fractious relationships among groups seeking help for their immigrant society. Aside from the insights we are able to provide about the reward-ranking scheme, our work is further distinguished by accounting for (i) the possibility of recipient activities that can change the groups’ ordering of the regimes and (ii) recipients gain based on reward regime.

Acknowledgment Financial support from the Adar Foundation of the Economics Department at Bar-Ilan University is gratefully acknowledged

Appendix It is a standard result that there are no pure strategy equilibria in all-pay auctions (Hillman and Riley, 1989; Ellingsen, 1991; and Baye et al., 1993, 1996). Suppose group 2 bids 0ox2rn2. Then the ﬁrst group’s optimal response is x1 ¼ x2þeon1 (i.e., marginally higher than x2). But then x2W0 cannot be an optimal response to x1 ¼ x2þe. Also, it is obvious that x1 ¼ x2 ¼ 0 cannot be an equilibrium. Hence, there is no equilibrium in pure strategies. There is a unique equilibrium in mixed strategies given by the following cumulative distribution functions

337

A Political Economy of the Immigrant Assimilation

(see Hillman and Riley, 1989; Ellingsen, 1991; and Baye et al., 1996): G1 ðx1 Þ ¼

x1 n2

for x1 2 ½0; n2 Þ and G2 ðx2 Þ ¼ 1

n 2 x2 þ n1 n1

for x2 2 ½0; n2 Þ.

The equilibrium c.d.f.’s show that group 1 bids uniformly on [0, n2], while group 2 puts a probability mass equal to (1n2/n1) on x2 ¼ 0. The expected investment expenditures are Z

n2

x1 dG1 ðx1 Þ ¼

Eðx1 Þ ¼ 0

n2 and Eðx2 Þ ¼ 2

Z

n1

x2 dG2 ðx2 Þ ¼ 0

n22 . 2n1

Note, we think of the all-pay auction as probabilistic – that is, the stronger group is more likely to win the contest.

References Alesina, A., Eliana La, F. (2000), Participation in heterogeneous communities. Quarterly Journal of Economics (August), 847–904. Altonji, J.G., Blank, R.M. (1999), Race and gender in the labor market. In: Ashenfelter, O., Card, D. (Eds.), Handbook of Labor Economics, vol. 3C. Amsterdam, The Netherlands, Elsevier Science B.V., pp. 3143–3259. Anas, A. (2002), Prejudice, exclusion and compensating transfers: the economics of ethnic segregation. Journal of Urban Economics 52 (3), 409–432. Bauer, T., Epstein, G.S., Gang, I.N. (2005), Enclaves, language and the location choice of migrants. Journal of Population Economics 18 (4), 649–662. Baye, M.R., Kovenock, D., de Vries, C.G. (1993), Rigging the lobbying process: an application of the all-pay auction. American Economic Review 83 (1), 289–294. Baye, M.R., Koveneock, D., de Vries, C.G. (1996), The all pay auction with complete information. Economic Theory 8 (2) 203–291. Bhaumik, S.K., Gang, I.N., Yun, M.-S. (2006), Ethnic conﬂict and economic disparity: Serbians and Albanians in Kosovo. Journal of Comparative Economics 34 (4), 754–773. Bisin, A., Verdier, T. (2000), ‘‘Beyond the melting pot’’: cultural transmission, marriage, and the evolution of ethnic and religious traits. Quarterly Journal of Economics August, 955–988. Blau, F.D., Kahn, L.M. (1997), Swimming upstream: trends in the gender wage differential in the 1980s. Journal of Labor Economics 15 (1), 1–42. Blau, F.D., Kahn, L.M. (2006), The US gender pay gap in the 1990s: slowing convergence. Industrial and Labor Relations Review 60 (1), 45–66.

338

Gil S. Epstein and Ira N. Gang

Blau, F.D., Kahn, L.M. (2007), The gender pay gap. The Economists’ Voice 4 (4), Article 5. Available at http://www.bepress.com/ev/vol4/ iss4/art5 Chiswick, B.R. (1977), Sons of immigrants: are they at an earnings disadvantage? American Economic Review 67 (1), 376–380, Papers and Proceedings. Chiswick, B.R., Miller, P.W. (1995), The endogeneity between language and earnings: international analyses. Journal of Labor Economics 13, 246–288. Chiswick, B.R., Miller, P.W. (1996), Ethnic networks and language proﬁciency among immigrants. Journal of Population Economics 9, 19–36. Cohen, C., Sela, A. (2007), Contests with ties. The B.E. Journal of Theoretical Economics 7 (1), (Contributions), Article 43. Constant, A., Gataullina, L., Zimmermann, K.F. (2009), Ethnosizing immigrants. Journal of Economic and Behavioral Organization 69 (3), 274–287. Deutsch, J., Epstein, G.S., Lecker, T. (2006), Multi-generation model of immigrant earnings: theory and application. Research in Labor Economics 217–234. Dustmann, C., Fabbri, F., Preston, I. (2004), Ethnic concentration, prejudice and racial harassment of minorities. CReAM Discussion Paper 05/04. Available at www.econ.ucl.ac.uk/cream/. Ellingsen, T. (1991), Strategic buyers and the social cost of monopoly. American Economic Review 81 (3), 648–657. Epstein, G.S., Gang, I.N. (2009a), Ethnicity, assimilation and harassment in the labor market. Research in Labor Economics 79, 67–90. Epstein, G.S., Gang, I.N. (2009b), Good governance and good aid allocation. Journal of Development Economics 89 (1), 12–18. Epstein, G.S., Nitzan, S. (2006a), Reduced prizes and increased effort in contests. Social Choice and Welfare 26 (3), 447–453. Epstein, G.S., Nitzan, S. (2006b), Effort and performance in public policy contests. Journal of Public Economic Theory 8 (2), 265–282. Epstein, G.S., Nitzan, S. (2007), Endogenous Public Policy and Contests. Springer, Berlin, Heidelberg. Gang, I.N., Rivera-Batiz, F. (1994), Labor market effects of immigration in the United States and Europe: substitution vs complementarity. Journal of Population Economics 7, 157–175. Gang, I.N., Rivera-Batiz, F., Yun, M.-S. (2002), Economic Strain, Ethnic Concentration and Attitudes Towards Foreigners in the European Union, IZA Discussion Paper 578. Available at www.iza.org Gang, I.N., Zimmermann, K.F. (2000), Is child like parent? Educational attainment and ethnic origin. Journal of Human Resources 35, 550–569. Hillman, A.L., Riley, J.G. (1989), Politically contestable rents and transfers. Economics and Politics 1 (1), 17–39.

A Political Economy of the Immigrant Assimilation

339

Kahanec, M. (2006), Ethnic specialization and earnings inequality: Why being a minority hurts but being a big minority hurts more. IZA Discussion Paper 2050. Available at www.iza.org. Lazear, E.P. (1999), Culture and language. Journal of Political Economy 107 (6, pt. 2), S95–S126. Lockard, A.A., Tullock, G. (Eds.) (2001), Efﬁcient rent-seeking, chronicle of an intellectual quagmire. Kluwer Academic Publishers, Boston, MA. Nitzan, S. (1994), Modelling rent-seeking contests. European Journal of Political Economy 10 (1), 41–60, (also appears in Lockard and Tullock, 2001). Smith, J.P., Welch, F.R. (1989), Black Economic Progress after Myrdal. Journal of Economic Literature 27 (2), 519–564.

CHAPTER 14

Assimilating Under Credit Constraints: Public Support for Private Efforts Sajal Lahiri Department of Economics, Southern Illinois University Carbondale, Carbondale, IL 62901-4515, USA E-mail address: [email protected]

Abstract We examine the effect of borrowing constraint facing new immigrants on the process of their assimilation in the new society. We shall do so in a twoperiod model. In period 1, immigrants invest, with some costs to them, in trying to assimilate. The probability of success in this endeavor depends on the amount invested and also on the level of the provision of a ‘‘public’’ good paid for by lump-sum taxation of ‘‘natives’’. Those who succeed enjoy a higher level of productivity and therefore wages in period 2. The level of investment is endogenously determined. Assimilation also affects remittances by immigrants. Given this framework, we examine the effect of public support on the degree of assimilation and income repatriation. We do so under two scenarios regarding the credit market facing new immigrants. In the ﬁrst, they can borrow as much as they want in period 1 at an exogenously given interest rate. In the second scenarios, there is a binding borrowing constraint. We compare the equilibrium under the two scenarios. Keywords: Immigration, assimilation, culture, credits JEL classiﬁcation: I28, J61

1. Introduction Migration – domestic and international – has been going on since time immemorial. It has very signiﬁcant short- and long-run implications for everyone involved in the process: immigrants, natives, the country of origin of immigrants, and so on. Therefore, all these groups of people respond to waves of migrations. As for the place or country of origin, the early literature expressed concerns for brain drain and suggested ways of compensating the source countries of migration (see, e.g., Bhagwati and Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008020

r 2010 by Emerald Group Publishing Limited. All rights reserved

342

Sajal Lahiri

Wilson (1989)). More recently, the issues have been reassessed and viewed very differently. Stark (2004), for example, has shown the possibility of human capital development in the source country in the presence of emigration possibilities. The volume of remittances by immigrants and their impacts on the source countries have also led to discussions on the beneﬁts from emigration for source countries (see, e.g., OECD, 2005; Fajnzylber and Lo´pez, 2008; Hanson, 2010).1 As for the host country of migration, in spite of protestations from vested interest groups about immigration, most studies look at immigration favorably (see, e.g., Friedberg and Hunt, 1995). According to Tilghman (2003), 20% of the members of the U.S. National Academy of Sciences are of foreign origin. About one-third of Nobel laureates from the United States are foreign-born. However, it should be acknowledged that the effect of immigration on the employment and wages of natives may well depend on the speciﬁc characterisctics of the immigrants as well as the characteristics of the labor market (see, e.g., Gang and Rivera-Batiz, 1994; Gang et al., 2002). As for immigrants themselves, the extent of their well-being depends, inter alia, on the level of their assimilation in the new society (see, e.g., Constant et al., 2008). Perhaps, it is because of the lack of assimilation that immigrants tend to earn a lot less than their comparable natives (see, e.g., Altonji and Blank, 1999; Blau and Kahn, 1997; Bhaumik et al., 2006). Why is there a lack of efforts on the part of many immigrants to assimilate? Epstein and Gang (2009) explain this phenomenon in terms of hostility and harassment from natives in the labor market. Hatton and Leigh (2010) ﬁnd that immigrants tend to assimilate as communities rather than as individuals, and this makes the issue of assimilation much more complex. Epstein and Gang (2006) analyzes the interlinkages between assimilation and networks, and its impact on the level of assimilation. Fan and Stark (2007) show that when assimilation results in immigrants getting ‘‘closer’’ to their richer natives and more ‘‘distant’’ from their fellow immigrants, the efforts for assimilation get muted. Efforts by immigrants does not always imply that they will succeed in their attempts to assimilate. There could be many factors that would determine the rate of success in assimilation for a given level of effort from immigrants. We have already mentioned about hostility from natives, and this will reduce the probability of success. Chiswick and Miller (1996) and Bauer et al. (2005) examine the effect of high adjustment costs (such inadequate language skills or lack of information on the labor market) on the probability of success in assimilation. Public policies can of course help immigrants in overcoming some of these hindrances. For example, publicly provided language schools, information centers, etc. can go a

1

For some source countries such as Bangladesh total remittance from remittances form a very large part of their total foreign exchange earnings.

Assimilating Under Credit Constraints

343

long way in helping immigrants to succeed in assimilating into their new environment. Many of these schemes exist in many of the countries where the inﬂow of immigrants is high. For example, in Canada new immigrants are entitled to settlement assistance such as free language training under provincial government administered programs usually called Language Instruction for Newcomers to Canada (LINC), for which the federal government budgeted about $350 million to give to the provinces for the ﬁscal year 2006–2007.2 The assimilation of immigrants not only has effect on their earnings, it may also affect their preferences in other ways. In particular, assimilation can lead to immigrants caring more about themselves and relatively less about people left behind in the country of their origin. This can, as Fan and Stark (2007) show, lead to less income repatriation by immigrants. From the above discussions it should be clear that private efforts by immigrants to assimilate and remittances by them are interdependent on each other, and public support for assimilation of immigrants can affect both these variables in a signiﬁcant way. However, one aspect of the host country that hitherto has not been considered in the literature in explaining the lack of private efforts in assimilation is the access, or the lack of it, to credits by immigrants. The manner in which credit ratings are normally calculated in most developed countries are by design stacked against newcomers in those countries. Often low-risk skilled immigrants are denied credits because of a lack of records on their credit history, while, as it is now well known, the same ﬁnancial institutions have been bending over backwards to offer credits to high-risk natives resulting in one of the worst ﬁnancial crisis since the Great Depression.3 In fact, because of the lack of credits from the formal credit institutions, many immigrant groups form their own credit institutions. For example, the institution of Rotating Credit and Savings Associations (ROSCA) can be found among many immigrant groups in the United States of America: among Mexican and Cuban immigrants in Southern California (Velez-Ibanez, 1983; Gama et al., 2010), Caribbean immigrants in New York City (Laguerre, 1998), and Korean Immigrants in Los Angeles (Light et al., 1990) to name a few. Although the ROSCAS help the immigrants in many ways particularly in acquiring consumer durables, there are not typically used for investment purposes. Since the costs of private efforts at assimilation are incurred upfront, and beneﬁts in the form of higher wages come in the future, credits have an obvious role to play in the determination of private efforts at assimilation. It is this void in the literature that the present chapter attempts to ﬁll. 2 http://en.wikipedia.org/wiki/Economic_impact_of_immigration_to_Canada; accessed on February 27, 2010. 3 Perhaps, the ﬁnancial institutions have fears of an immigrant defaulting and returning to the home country.

344

Sajal Lahiri

In order to do so, we shall develop a two-period model in which an immigrant spends a certain amount of effort in assimilation, and this has an opportunity cost to the immigrant in terms of time and income. The probability that the immigrant succeeds in assimilation depends on the level of this private effort and the level of public support for assimilation in the host country. If the immigrant succeeds in assimilation, not only that it raises the wage income of the immigrant but it also has an implication for its preference for remittances sent to people back home. The level of private efforts is optimally chosen by the immigrant. We consider two scenarios. In the ﬁrst scenario, the immigrant can borrow as much as it wants at a given interest rate, and in the second the immigrant is subject to a binding borrowing constraint. In this framework, we shall examine if restrictions on borrowing by the immigrants does indeed affect private efforts at assimilation adversely. We shall also examine if the effect of public support on private efforts at assimilation is lower in the presence of a binding borrowing constraint facing the immigrant than in the absence of it. 2. The theoretical framework We develop a two-period model of a small open economy. A number of identical immigrants arrive in the country at the beginning of period 1. We shall treat this number as exogenous and, without any loss of generally, assume it to be unity. The number of hours available to each immigrant is assumed to be exogenous and once again, without loss of any generality, taken to be unity. In period 1, the immigrant makes some effort to assimilate in the adopted country and this costs him/her e ð 1Þ hours. The immigrant succeeds in assimilation with probability p, which depends on e, and the level of public support for the assimilation program, denoted by g. We shall treat g as a public good. p ¼ pðe; gÞ,

(1)

where we assume:4 ASSUMPTION 1. p1 40; p2 40, and p12 40. The assumption p12 40 implies complementarity between private efforts in, and public support of, assimilation. For simplicity and without any loss of generality, we assume that there is one good in each period, the price of which is normalized to unity. Denoting by ci the consumption of the good in period i, the utility of the immigrant, uI , depends not only on his/her consumption of the good,

4 For any function f ð Þ, we denote by f i as its partial derivative with respect to the ith argument.

Assimilating Under Credit Constraints

345

but also on the amount repatriated to its country of origin.5 The levels of repatriations or remittances in the two periods are denoted by T 1 and T 2 , respectively, and the the level of utility from repatriated income (in the absence of assimilation) is f ðT 1 ; T 2 Þ. Implicitly, we assume that the immigrant cares about the family (direct or extended) left behind. We assume that ASSUMPTION 2. f 1 40, f 2 40, f 11 o0, f 22 o0, and f 12 40. The ﬁrst-order effects f 1 and f 2 can be different for a variety of reasons. For example, if the family at the source country is subject of credit constraints, an extra income in period 1 would reduce the family’s demand for loan and therefore reduce the rate of interest it faces. An extra income in period 2 will have exactly the opposite effect.6 Thus, the same amount of (real) income from repatriation would have different effect on the family’s utility. We assume that assimilation has two effects on the immigrant. First, it increases his/her income in period 2. Second, it reduces the immigrant’s ‘‘link’’ with his/her country of origin. The ‘‘link’’ can be reduced for many reasons. For example, an assimilated immigrant may have more commitments in the adopted country and therefore its relative preference (weight) for own consumptions may go up. Denoting by y ðy 1Þ the weight it puts on utility from remittances after assimilation, the immigrant’s expected utility from repatriated income is ð1 pÞf þ pyf . We assume that the utility from consumption and that from remittances to be additively separable. That is, the expected utility uI of the immigrant is given by uI ¼ vI ðc1 ; c2 Þ þ ½1 ð1 yÞpðe; gÞf ðT 1 ; T 2 Þ,

(2)

where vI(c1,c2) is the direct utility from consumption. The choice variables for the migrant are c1 , c2 , T 1 , T 2 , and e. We shall describe the migrants optimization problem a little later. From (2), the expenditure function of the immigrant is derived as E I ð1; 1=ð1 þ rI Þ; uI ½1 ð1 yÞpð Þf ð ÞÞ where rI is the interest rate the immigrant faces. As is well known, the partial derivative of the expenditure function with respect to a price of a good gives the compensated demand for that good, and the partial derivative with respect to the utility is the reciprocal of the marginal utility of income.7 The expenditure function for the natives are denoted by Eð1; 1=ð1 þ rN Þ; uN Þ where rN is the interest rate facing natives and uN their utility level. 5 Gaytan-Fregoso and Lahiri (2000) provide a microfoundation for this formulation by explicitly modeling the source country. 6 See Jafarey and Lahiri (2005) for an explanation, albeit in a different context. 7 It is also true that E 11 o0, E 22 o0, E 12 40, E 33 40 and if the goods are normal, then E 13 40 and E 23 40. For these and other properties of an expenditure function see, for example, Dixit and Norman (1980).

346

Sajal Lahiri

Turning to the production side, the revenue functions in the two periods are R1 ð1; 1 e; gÞ and R2 ð1; 1 p þ lpÞ, respectively, where l ðl 1Þ is the amount of ‘‘effective’’ labor of the immigrant if assimilation succeeds. The partial derivative of the revenue functions with respect to an endowment of factor of production gives the rate of return to that factor (see, Dixit and Norman, 1980). The function R1 ð Þ is in fact the ‘‘restricted’’ revenue function in period 1, representing total value of production in the private sector when the level of public good provision is g. Since all other endowments do not vary, they are omitted from the arguments of the revenue functions. It is well known that the unit cost of production of the public good is R3 .8 For simplicity, we assume that the cost of production of the public good is paid for by a lump-sum taxation of natives and that its production only uses factors that belong to natives. With these assumption, the budget-balance equations for the government, the immigrant, and natives are given, respectively, by R13 g ¼ T, E

I

(3)

1 1; ; uI ½1 ð1 yÞpf 1 þ rI

½1 p þ lpR22 T2 1 þ T , ¼ ð1 1 þ rI 1 þ rI 1 R2 1 ; u þ ð1 eÞR12 E 1; ¼ R N 1 þ rN 1 þ rN

(4)

eÞR12

½1 p þ lpR22 R13 g T. 1 þ rN

(5)

Equation (3) states that the total cost of producing the public good (the left-hand side) is equal to the amount of lump-sum tax levied on natives (right-hand side). The left-hand sides of (4) and (5) are the discounted present value of expenditures on consumption by the immigrant and natives, respectively. The ﬁrst term on the right-hand side of (4) is the wage income of the immigrant in period 1. The second term is the discounted present values of the second-period expected wage income. The third and the fourth terms are the present values of the repatriated amounts sent back to the country of origin. The ﬁrst four terms on righthand side of (5) together give the income of the natives from private sector (the total factor income in the economy minus the factor income of the immigrant). The ﬁfth terms is natives’ income from the public sector, and the last term is the lump-sum tax that is levied on them. 8

For the derivation and properties of a restricted revenue function, please see, e.g., Hatzipanayotou et al. (2002).

Assimilating Under Credit Constraints

347

The borrowing by the immigrant in period 1, B, is given by B ¼ E I1 ð1 eÞR12 þ T 1 ,

(6)

which is the excess of expenditure over income of the immigrant in period 1. We shall make the Heckscher–Ohlin assumption that that factor endowments do not affect factor prices, that is, the factor endowments lie within the cone of diversiﬁcation and there are no factor intensity reversals (see Dixit and Norman, 1980) for details). That is, ASSUMPTION 3. R122 ¼ R133 ¼ R123 ¼ R222 ¼ 0. It now remains to describe how e; T 1 , and T 2 are determined. For this we differentiate (1)–(4) to obtain. ðl 1ÞR22 I E ð1 yÞfp E I3 duI ¼ R12 þ 1 de 3 1 þ rI 2 R ðl 1Þp2 I þ 2 E ð1 yÞfp 2 dg 3 1 þ rI (7) þ ½1 þ f1 pð1 yÞgE I3 f 1 dT 1 1 B I þ þ f1 pð1 yÞgE 3 f 2 dT 2

drI . 1 þ rI 1 þ rI An increase in the efforts to assimilate has two costs: (i) reduction in wage income in period 1 ðR12 Þ, and (ii) a reduction in utility because of caring less for the family back home (the third term in the coefﬁcient for de). It beneﬁts the immigrant by increasing its wage income in period 2 ððl 1ÞR22 Þ. An increase in g also has the same costs and beneﬁts associated with an increase in e, but it does not reduce wage income in period 1. An increase in either T 1 or T 2 has direct costs (the ﬁrst terms in the coefﬁcients of dT 1 and dT 2 ); they also beneﬁt the immigrant by increasing the utility from repatriating income (the second terms in the coefﬁcients of dT 1 and dT 2 ). Finally, an increase in the interest rate reduces the utility of immigrant since it is a net borrower (the so-called intertemporal terms-of-trade effect). The immigrant decides on the levels of e, T 1 , and T 2 by maximizing uI for a given value of the interest rates. The ﬁrst-order conditions for the immigrant’s optimization problem are given by @uI p ðl 1ÞR22 ¼ 0 ) R12 þ E I3 p1 f ð1 yÞ ¼ 1 , @e 1 þ rI

(8)

@uI ¼ 0 ) 1 ¼ E I3 f1 pð1 yÞgf 1 , @T 1

(9)

348

Sajal Lahiri

@uI 1 ¼0) ¼ E I3 f1 pð1 yÞgf 2 . 1 þ rI @T 2

(10)

The marginal costs and beneﬁts associated with the three choice variables have been explained after (7). The left-hand sides of the above three equations are the marginal costs of the three variables, and the right-hand sides are the marginal beneﬁts. From (9) and (10), we ﬁnd f 1 ¼ ð1 þ rI Þf 2 .

(11)

Note that, Equation (11) implies that in equilibrium we must have f 1 4f 2 , a property that is consistent with our discussion after the statement of Assumption 2. This completes the description of our theoretical framework. We shall assume rN to be exogenous. However, we start with the assumption that rI is also constant. But, later we shall assume that the immigrant is subject to a binding borrowing constraint so that its demand for loan, given by (6), that is, is equal to an exogenously given supply of the loan B, B ¼ B.

(12)

When (12) holds, the interest rate facing the immigrant, rI , becomes an endogenous variable. 3. Public support and private assimilation In this section We shall examine the effect of an increase in the provision of the public good on the level of assimilation of the immigrant.9 we shall do so under two scenarios: (i) the immigrant can borrow as much it wants at the given interest rate rI , and (ii) it faces a binding borrowing constraint so that rI is endogenous, and then we shall examine how the existence of the borrowing constraint affects the results. Using the optimality conditions (8)–(10), Equation (7) reduces to E I3 duI ¼

B p R1 drI þ 2 2 dg. I p1 1þr

(13)

We also ﬁnd that duN ¼ 0. That is, an increase in g unambiguously increases the utility of the immigrant when e, T 1 , and T 2 are optimally chosen. The latter three variables do not affect uI directly as these are optimally chosen (the envelope property). The effect of rI on uI is as before. The utility of natives are unaffected as the public good is produced using factors owned by 9

The actual policy is the lump-sum taxation of natives for the public support of assimilation. However, since the unit cost of providing public service is constant in our analysis, there is no analytical difference between the two.

349

Assimilating Under Credit Constraints

natives, and factor prices are assumed not to be inﬂuenced by factor endowments (Assumption 3). Differentiation of (8), (9), (11), and (12), the and use of (8)–(10) and (13), give us 1 R2 p11 f 2 ð1 yÞ2 ðp1 Þ2 E I33 de p1 1 R p R1 ¼ 2 12 f ð1 yÞp2 E I33 f ð1 yÞp1 þ 2I dg p1 E3 (14) 2 p1 ðl 1ÞR2 p1 f ð1 yÞE 32 f ð1 yÞE I33 p1 B I dr þ E I3 ð1 þ rI Þ ð1 þ rI Þ2 ð1 þ rI Þ2 þ p1 ð1 yÞa½f 1 dT 1 þ f 2 dT 2 , 1 E I13 B I I E 12 I dr ¼ d B R12 þ E I13 f ð1 yÞp1 de 2 I E3 ð1 þ r Þ þ E I13 f1 pð1 yÞg½f 1 dT 1 þ f 2 dT 2 E I13 p2

(15)

R12 þ f ð1 yÞ dg p1 E I3

½E I3 f 11 f1 pð1 yÞgðf 1 Þ2 E I33 dT 1 þ ½E I3 f 12 f1 pð1 yÞgf 1 f 2 E I33 dT 2 (16) I p1 f 1 ð1 yÞa p2 f 1 b f1 E 32 E I33 B I de þ dg þ þ I dr , ¼ 1 pð1 yÞ 1 pð1 yÞ 1 þ rI 1 þ rI E3 I

1

2

I

(17)

I

½ð1 þ r Þf 12 f 11 dT þ ½ð1 þ r Þf 22 f 12 dT ¼ f 2 dr , where a ¼ E I3 fE I33 f1 pð1 yÞg, b ¼ E I3 ð1 yÞ f1 pð1 yÞgE I33

R12 . f ð1 yÞ þ p1 E I3

Note that the coefﬁcient of de on the left-hand side of (14) is negative, and this is consistent with the second-order condition for the immigrants optimization problem. There are two opposite effects of an increase in g on e for given levels of T 1 , T 2 , and rI . First, it increases both the marginal cost and the marginal beneﬁt of increasing e (the second term on the left-hand side, and the term on the right-hand side, of (8)), but, at the equilibrium, the increase in marginal beneﬁt dominates the increase in marginal costs. This effect is given by the ﬁrst term in the coefﬁcient of dg on the right-hand side of (14). The second effect of an increase in g on e appears in terms of an income effect: an increase in g increases real income of the immigrants (see (13)) and this reduces their marginal utility of income, and this in turn increases the marginal cost of increasing e

350

Sajal Lahiri

(the second term on the left-hand side of (8)). An increase in either T 1 or T 2 , for a given level of e and rI , increases the marginal cost of increasing e by increasing the value of f ð Þ and this effect will tend to reduce the value of e. However, an increase in either T 1 or T 2 also increases the marginal utility of income and reduces the marginal cost of increasing e. This effect will tend to increase the value of e. The net effect will be positive if the value of a is negative. An increase in rI has two positive and one negative effect on e for given levels of T 1 , and T 2 . The negative effect is due to the fact that an increase in rI reduces the marginal beneﬁt of increasing e (the term on the right-hand side of (14)). The positive effects come via income effects: (i) an increase in rI reduces the utility of the immigrant via the intertemporal terms-of-trade effect (see (13)) and this reduces the marginal cost of increasing e, and (ii) an increase in rI reduces the present value of the second-period price and this increases the marginal utility of income and thus reduces the marginal cost of increasing e. An increase in the borrowing limit reduces the interest rate facing the immigrant by increasing its supply, for given levels of e, T 1 , and T 2 . An increase in e, for given levels of T 1 and T 2 , reduces income in the ﬁrst period and increases that in the second period. Both these effects increases the demand for loan in the ﬁrst period, increasing the interest rate. An increase in g raises the probability of success in attempts to assimilate and thus the expected income in the second period. This will increase the demand for loan in the ﬁrst period, increasing the interest rate. An increase in either T 1 or T 2 , for a given level of e, will reduce the direct utility from consumption (given by the third argument in E I ð Þ) and thus reducing the demand for consumption and loan in the ﬁrst period. This will reduce the interest rate. As for the effects of e, rI , and g on the equilibrium levels of T 1 and T 2 , from (16) and (17) we ﬁnd @T 1 p1 f 1 ð1 yÞa ¼ ½f ð1 þ rI Þ f 12 ; @e 1 pð1 yÞ 22 @T 2 p f ð1 yÞa ¼ 1 1 ½f ð1 þ rI Þ f 11 , D @e 1 pð1 yÞ 12 I @T 1 f1 E 32 E I33 B þ I ½ð1 þ rI Þf 22 f 12 D I ¼ @r 1 þ rI 1 þ rI E3 D

þ f 2 ½E I3 f 12 f1 pð1 yÞgðf 1 Þ2 E I33 , I @T 2 f1 E 32 E I33 B þ I ½f 11 ð1 þ rI Þf 12 D I ¼ @r 1 þ rI 1 þ rI E3 f 2 ½E I3 f 11 f1 pð1 yÞgf 1 f 2 E I33 , @T 1 p2 f 1 b ½ð1 þ rI Þf 22 f 12 ; ¼ 1 pð1 yÞ @g @T 2 p2 f 1 b ¼ D ½f ð1 þ rI Þf 12 , @g 1 pð1 yÞ 11 D

(18)

Assimilating Under Credit Constraints

351

where D ¼ ½E I3 f 11 f1 pð1 yÞgðf 1 Þ2 E I33 ½ð1 þ rI Þf 22 f 12 ½ð1 þ rI Þf 12 f 11 ½E I3 f 12 f1 pð1 yÞgf 1 f 2 E I33 . Note that D must be positive to satisfy the second-order conditions for the immigrant’s optimization problem. An increase in e increases the probability of success in assimilation p and thus reduces the marginal beneﬁt of increasing either T 1 or T 2 . However, an increase in e increases the direct utility of consumption and thus reduces the marginal utility of income. This will raise the marginal beneﬁt of increasing either T 1 or T 2 . The net effect is positive if and only if a is positive. The same argument goes for the effect of an increase in g on either T 1 or T 2 . The net effect this time is positive if and only if b is positive. The effect of an increase in rI on T 1 is a little different than that on T 2 . This is because an increase in rI reduces the marginal cost of increasing T 2 , but not that of increasing T 1 . An increase in rI reduces the marginal beneﬁt of increasing either T 1 or T 2 by increasing marginal utility of income because of a reduction in utility and in the present value of the second-period income. There are other effects that occur indirectly because of the interdependence in T 1 and T 2 . Having discussed the partial effects, we now look at the total effects. For this, we simultaneously solve (14)–(17) to get Aee de ¼ Aeg dg þ Aer drI ,

(19)

Lr drI ¼ d B þ Lg dg,

(20)

where Aee ¼ Aeg ¼ Aer ¼ Lr ¼

Lg ¼

R12 p11 @T 1 @T 2 , þf2 f 2 ð1 yÞ2 ðp1 Þ2 E I33 p1 ð1 yÞaÞ f 1 p1 @e @e R1 p R1 @T 1 @T 2 þf2 , 2 12 þ f ð1 yÞp2 E I33 f ð1 yÞp1 þ 2I þ p1 ð1 yÞa f 1 p1 @g @g E3 p1 ðl 1ÞR22 p1 f ð1 yÞE 32 f ð1 yÞE I33 p1 B @T 1 @T 2 ð1 yÞa f þ f , þ p 1 1 2 @rI @rI E I3 ð1 þ rI Þ ð1 þ rI Þ2 ð1 þ rI Þ2 1 E I B @T 1 @T 2 @T 1 E I12 13I E I13 f1 pð1 yÞg f 1 I þ f 2 I þ I 2 @r @r @r E3 ð1 þ rI Þ 1 2 @T @T Aer þf2 þ R12 þ E I13 f ð1 yÞp1 E I13 f1 pð1 yÞg f 1 , @e @e Aee 1 R2 @T 1 @T 2 @T 1 E I13 p2 þ f ð1 yÞ þ E I13 f1 pð1 yÞg f 1 þf2 I @g @g @g p1 E 3 1 2 @T @T Aeg . R12 þ E I13 f ð1 yÞp1 E I13 f1 pð1 yÞg f 1 þf2 @e @e Aee

Note that Aee has to be negative for the second-order condition of the immigrant’s optimization problem to be satisﬁed. Also Lr is the slope

352

Sajal Lahiri

of the excess demand for loan function and this has to be negative as well for the system to be Walrasian stable. We shall now examine the effect of a change in g on the equilibrium value of e under two scenarios: (i) the immigrant faces no credit constraint, that is, it can borrow as much as it wants at the interest rate at the exogenously given rI , and (ii) the immigrant faces a binding borrowing constraint (12) and the rate of interest rI is endogenous. These scenarios will now be considered in turn. Case 1: No credit constraint: In this case drI ¼ 0 in Equations (19) and (20) is not applicable. Therefore, de ‘0 () Aeg x0. dgdrI ¼0 First of all note that when y ¼ 1, that is, when assimilation does not reduce the immigrant’s degree of altruism toward its family back home, de=dg40. The complementarity between private effort and public support for assimilation implies that an increase in public support for assimilation increases both the marginal cost and marginal beneﬁt of increasing private efforts (see (8)). However, the increase in marginal beneﬁt dominates that in marginal costs and the net effect is positive. When yo1, the effect is generally ambiguous. However, if the marginal utility of income of the immigrant is more or less constant, that is, E I33 ’ 0, then once again de=dg40. If the marginal utility of income is constant, then the marginal beneﬁt of increasing T 1 or T 2 decreases with g and therefore the optimal values of both T 1 and T 2 decrease as g is increased. These reductions in T 1 and T 2 reduces the marginal costs of increasing e and thus reinforcing the positive effect on e of an increase in g because of the complementarity between private effort and public support for assimilation. These results are formally stated in the following proposition. PROPOSITION 1. In the absence of any credit constraint facing immigrants, an increase in the level of public support for assimilation increases the level of private efforts in assimilation if either y ’ 1 or E I33 ’ 0. Turning to the effect of an increase in the interest rate on the equilibrium level of e, it is to be noted that the effect works through many channels: some positive, some negative. However, when either y ’ 1 or E I3 is constant, it is easy to verify that de=d B40. When y ’ 1, a reduction in rI increases the marginal beneﬁt of increasing e (the right-hand side of (8)) and thus raise the equilibrium level of e. When yo1, there is an additional effect via induced changes in T 1 and T 2 . When E I3 is constant, a reduction in rI increases the marginal cost of increasing T 2 (the left-hand side of (10)) and thus reduces the optimal value of T 2 and thus that of T 1 . This reduction in T 1 and T 2 reduces the marginal cost of increasing e

Assimilating Under Credit Constraints

353

(the left-hand side of (8)), reinforcing the increase in marginal beneﬁts. Formally, PROPOSITION 2. A reduction in the interest rate rI increases the equilibrium level of private efforts in assimilation if either y ’ 1 or E I33 ’ 0 and E I32 ’ 0. Case 2: Binding credit constraint: Having identiﬁed sufﬁcient conditions under which an increase in g increases the equilibrium value of e in the absence of any credit constraint facing the immigrant, we shall now examine what the existence of a binding borrowing constraint does to this comparative static result: is it more likely or less likely that an increase in g will increase the equilibrium value of e under a borrowing constraint than in the absence of it? We shall also examine the effect of a relaxation of the borrowing constraint on the equilibrium value of e. Turning to the second issue ﬁrst, a relaxation of the borrowing con straint ðd B40Þ unambiguously reduces the equilibrium level of the interest I rate r . The intuition is very straightforward. An increase in B shifts the supply function to the right and does not affect the demand function, resulting in a reduction of the equilibrium price (interest rate). A reduction in rI , however, affects the equilibrium level of e via many channels: some positive and some negative, as shown before. Therefore, in general, the sign of de=d B is ambiguous. However, using Proposition 2, we ﬁnd PROPOSITION 3. A relaxation of the borrowing constraint facing immigrants increases the level of private efforts in assimilation if either y ’ 1 or E I33 ’ 0 and E I32 ’ 0. Finally, turning to the effect of a change in g on the equilibrium level of e under a borrowing constraint, it is to be noted that g affects the demand for loan in many ways. It affects consumption in the ﬁrst period via an increase in utility, via an induced changes in remittances, and via an induced change in private efforts in assimilation, and via a change in the probability of success in assimilation because of an increase in g. It also affects ﬁrst-period incomes because of induced changes in e and T 1 . The net effect is ambiguous even when either y ’ 1 or E I3 is constant. We need an additional condition to sign drI =dg. Suppose that public support for assimilation, on its own, does not have a signiﬁcant effect on the probability of success in assimilation, but has a signiﬁcant effect on the marginal probability of private efforts, that is, p2 ’ 0, but p12 40. In this case, g will have no effects on either ﬁrst-period consumption directly or via induced changes in T 1 or T 2 , for given level of e. It will only affect ﬁrst-period income via an induced change in e. Under the conditions in Proposition 1, that is, when either y ’ 1 or E I33 ’ 0, we know from that proposition that an increase in g will increase the equilibrium value of e. This increase in turn will reduce the ﬁrst-period income of the immigrant and therefore increase the demand for loan and thus the equilibrium value of rI . Using Proposition 2, we can then conclude that, under the conditions

354

Sajal Lahiri

stipulated, an increase in g will have a lower impact on the equilibrium level of private efforts in assimilation under a borrowing constraint than in the absence of it. Formally, PROPOSITION 4. Suppose that p2 ’ 0. Furthermore suppose that either y ’ 1 or E I33 ’ 0. Then, an increase in g will have a lower impact on the equilibrium level of private efforts in assimilation under a borrowing constraint than in the absence of it. We conclude this section by drawing a number of policy implications of our results. In the preceding analysis we have found that while an increase in public support for immigrant assimilation is likely to increase private efforts by immigrants to assimilate, a binding borrowing constraint reduces this effect. Furthermore, a relaxation of the borrowing constraint is likely to increase private efforts. Therefore, it seems that public support for immigrant assimilation should be strengthened and restrictions for credits for immigrants should be reduced in order for public support to have the necessary impact. 4. Conclusion The assimilation of immigrants, or the lack of it, in their new adopted country has been receiving a lot of attention of late in the literature on international migration. Since with assimilation immigrants are very likely to improve their well-being, the lack of efforts on their part to assimilate is somewhat puzzling. Many factors which stops immigrants from making adequate efforts at assimilation, have been identiﬁed. Hostility of natives because of perceived adverse effect of immigration on their wages and employment is one such factor. A lack of public support in the assimilation process is another factor. The existence of networks among the immigrants is yet another factor. However, one factor that has not been analyzed in the literature is the lack of credits facing newly arrived immigrants. Since the costs of private efforts at assimilation are incurred upfront, and beneﬁts in the form of higher wages come in the future, credits have an obvious role to play in the determination of private efforts at assimilation. It is surprising, therefore, that this issue remains unexamined in the literature. In this chapter, we have tried to ﬁll this void in the literature. We have developed a two-period model in which immigrants make an effort at assimilation in the ﬁrst period, and, if they succeed, enjoy a higher wage rate in the second period. The probability of success at assimilation not only depends on the level of efforts that immigrants make, but also on the level of public support for it. Successful assimilation also affect their preference for remittances to people at their country of origin. In this framework, we examined the effect of public support on the level private efforts at assimilation. We also analyzed if the above effect is smaller when

Assimilating Under Credit Constraints

355

immigrants face a binding a borrowing constraint. We have found that the presence of a binding borrowing constraint can indeed reduce the beneﬁcial effect of public support for private assimilation. The broad policy prescription of this research is that restrictions for credits for immigrants should be reduced in order for the public supports to have the necessary impact.

Acknowledgment I am grateful to an anonymous referee for helpful comments and suggestions.

References Altonji, J.G., Blank, R.M. (1999), Race and gender in the labor market. In: Ashenfelter, O., Card, D. (Eds.), Handbook of Labor Economics, vol. 3C. Elsevier Science B.V., Amsterdam, pp. 3143–3259. Bauer, T., Epstein, G.S., Gang, I.N. (2005), Enclaves, language and the location choice of migrants. Journal of Population Economics 18 (4), 649–662. Bhagwati, J.N., Wilson, J.D. (Eds.), 1989. Income taxation and international mobility. M.I.T. Press, Cambridge, MA. Bhaumik, S.K., Gang, I.N., Yun, M.-S. (2006), Ethnic conﬂict and economic disparity: Serbians and Albanians in Kosovo. Journal of Comparative Economics 34 (4), 754–773. Blau, F.D., Kahn, L.M. (1997), Swimming upstream: trends in the gender wage differential in the 1980s. Journal of Labor Economics 15 (1), 1–42. Chiswick, B.R., Miller, P.W. (1996), Ethnic networks and language proﬁciency among immigrants. Journal of Population Economics 9, 19–36. Constant, A., Gataullina, L., Zimmermann, K.F. (2008), Ethnosizing Immigrants. Journal of Economic and Behavioral Organization 69 (33), 274–287. Epstein, G.S., Gang, I. (2006), Ethnic networks and international trade. In: Foders, F., Langhammer, R.J. (Eds.), Labor Mobility and the World Economy. Springer, Berlin, Heidelberg, pp. 85–103. Epstein, G.S., Gang, I. (2009), Ethnicity, assimilation and harassment in the labor market. In: Polachek, S., Tatsiramos, K. (Eds.), Ethnicity and Labor Market Outcomes: Research in Labor Economics, Vol. 29. Emerald Group Publishing Limited, Bingley, UK, pp. 67–88. Fajnzylber, P., Lo´pez, J.H. (Eds.) (2008), Remittances and development lessons from Latin America, Latin American Development Forum Series, The World Bank, Washington, DC.

356

Sajal Lahiri

Fan, C.S., Stark, O. (2007), A social explanation of the reluctance to assimilate. Kyklos 60 (1), 55–63. Friedberg, R.M., Hunt, J. (1995), The impact of immigrants on host country wages, employment and growth. The Journal of Economic Perspectives 9 (2), 23–44. Gama, R., D. Medrano, L. Medrano, 2010, Tandas and Cundinas: Mexican-American and Latino-American Rotating Credit Associations in Southern California, in Anthropology of Money in Southern California (http://www.anthro.uci.edu/html/Programs/Anthro_Money/ Tandas.htm). Gang, I.N., Rivera-Batiz, F. (1994), Labor market effects of immigration in the United States and Europe: substitution vs. complementarity. Journal of Population Economics 7, 157–175. Gang, I.N., Rivera-Batiz, F., Yun, M.-S. (2002), Economic strain, ethnic concentration and attitudes towards foreigners in the European Union. IZA Discussion Paper 578 (www.iza.org). Institute for the Study of Labor, Bonn, Germany. Gaytan-Fregoso, H., Lahiri, S. (2000), Foreign aid and illegal immigration. Journal of Development Economics 63, 515–527. Hanson, G. (2010), International migration and the developing world. In: Rodrik, D., Rosenzweig, M. (Eds.), Handbook of Development Economics, vol. 5. Elsevier, B.V., North-Holland, The Netherlands, pp. 4363–4414. Hatton, T.J., Leigh, A.L. (2010), Immigrants assimilate as communities, not just as individuals. Journal of Population Economics. Forthcoming. Hatzipanayotou, P., Lahiri, S., Michael, M.S. (2002), Can cross-border pollution reduce pollution? Canadian Journal of Economics 35, 805–818. Jafarey, S.S., Lahiri, S. (2005), Food for education and funds for education quality: policy options to reduce child labor. Canadian Journal of Economics 38, 394–419. Laguerre, M.S. (1998), Rotating credit association and the diasporic economy. Journal of Development Entrepreneurship 3, 23–34. Light, I., Kwuon, I.J., Zhong, D. (1990), Korean Rotating Credit Associations in Los Angeles. Amerasia Journal 16, 35–54. OECD (2005), Migration, Remittances and Development. OECD Publishing, Paris. Stark, O. (2004), Rethinking the brain drain. World Development 32 (1), 15–22. Tilghman, S.M. (2003), Dealing with foreign students and scholars in the age of terrorism: visa backlogs and tracking systems. Testimony presented to the U.S. House of Representatives Committee on Science, March 26, 2003. Velez-Ibanez, C.G. (1983), Bonds of Mutual Trust: The Culture System of Rotating Credit Associations Among Urban Mexicans and Chicanos. Rutgers University Press, New Brunswick, Canada.

CHAPTER 15

Immigrant Networks and the U.S. Bilateral Trade: The Role of Immigrant Income Kusum Mundra* Department of Economics, Rutgers University, Newark, NJ 07901, USA E-mail address: [email protected]

Abstract This chapter examines the role of immigrant networks on trade, particulalry through the demand effect. First, we examine the effect of immigration on trade when the immigrants consume more of the good that is abundant in their home country than the natives in a standard Heckscher–Ohlin model and ﬁnd that the effect of immigration on trade is a priori indeterminate. Our econometric gravity model consisting of 63 major trading and immigrant-sending country for the United States over 1991–2000. We ﬁnd that the immigrants income, mostly through demand effect has a signiﬁcant negative effect on U.S. imports. However, if we include the effect of the immigrant income interacted with the size of the immigrant network, measured by the immigrant stock, we ﬁnd that higher immigrants income lowers the immigrant network effect for both U.S. exports and imports. This we ﬁnd in addition to the immigrants stock elasticity of 0.27% for U.S. exports and 0.48% for U.S. imports. Capturing the immigrant assimilation with the level of immigrant income and the size of the immigrant enclave this chapter ﬁnds that the immigrant network effect on trade ﬂows is weakened by the increasing level of immigrant assimilation. Keywords: Immigrant networks, immigrant income, trade, immigrant demand, immigrant assimilation, Heckscher-Ohlin JEL classiﬁcations: F22, F11, J10, J61

* Earlier version of Section 2 Immigrant and the Heckscher-Ohlin Model has beneﬁtted from discussion with Prasanta Pattanaik. Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008021

r 2010 by Emerald Group Publishing Limited. All rights reserved

358

Kusum Mundra

1. Introduction Beginning from the work by Gould (1994) there is increasing literature examining the effect of immigrant networks on trade with the immigrants’ home country. There is increasing empirical evidence that the immigrant population, particularly stock of immigrants living in a country, provides the social and coethnic networks that facilitate trade with their home country by removing some informal trade barriers and lowering transactions cost to trade.1 The literature has found that the immigrants (or immigrant based networks) have a positive effect on bilateral trade for the United States (Gould, 1994; Dunlevy and Hutchinson, 1999; Rauch, 1996; Herander and Saavedra, 2005; Bandyopadhyay et al., 2008) and for Canada (Head and Reis, 1998). Immigrants ‘‘home-link’’ increases trade with home countries through the immigrants’ home country information (information effect) and through their demand for goods from their home country (demand effect). In previous literature both the immigrant information and the demand effect is measured by the size of the immigrant stock. In addition to the size of the immigrant stock measuring immigrants’ effect on trade in this chapter we explore the role of immigrants’ income on the bilateral trade, particularly through the immigrants’ demand effect. Immigrants carry home-country information that helps in matching buyers and sellers and enforcement of trading contacts (information effect). Immigrants have information on different traders and the type of goods available both in the United States and their home countries. This knowledge helps in promoting bilateral trade between the host and the home country. In addition, immigrants’ information on the legal set up in their country of origin, familiarity with the home-country language, and knowledge on how business is conducted in their home country helps in enforcing trading contacts with their home country. Immigrants also demand goods from their home country increasing their home country exports to the host country – demand effect.2 Light et al. (2002), while exploring the effect of English speaking immigrants on export claims that immigrant entrepreneurs import familiar goods from their home countries since there is a demand for these goods in their host country. This chapter examines the demand effect of the immigrants, particularly the effect of immigrants’ income on trade. In the literature there is no explicit attempt to distinguish the immigrants’ information effect from the demand

1

In international trade Treﬂer (1997) have found a strong evidence of coethnic and social networks in explaining the missing trade links and Grief (1993) and Rauch and Casella (1998) have shown that business and social networks help in alleviating informal trade barriers. 2 There is an extensive literature on the role played by immigrants demand for goods from their home country in generating and sustaining immigrant entrepreneurship. For a good discussion on immigrants demand and growth of ethnic business enclaves see Portes and Rumbaut (1996), Light and Bonacich (1988), and Halter (1995) to name a few.

Immigrant Networks and the U.S. Bilateral Trade

359

effect and hence there are conﬂicting and different ﬁndings regarding the effect of immigrants’ information and demand on trade (Wagner et al., 2002). Head and Reis (1999) ﬁnd that the immigrant elasticity for imports is three times of that of the exports and they argue that if the information effect for both exports and imports is assumed to be of equal magnitude, then the demand effect of immigrants has to be twice that of their information effect. However, Girma and Yu (2002) and Gould (1994) ﬁnd higher immigrant elasticity for exports than for the U.S. imports. In this chapter we include immigrants’ income in the United States as a proxy for immigrants’ level of assimilation and purchasing power and estimate the demand effect of immigrants’ after controlling for the size of the immigrant network. Immigrants’ demand and its effect on the global economy is under studied. In the majority of international trade models goods mobility is analyzed assuming consumers in the two trading partners (or multi trading partners) have identical demand patterns.3 With increasing migration around the world the immigrants demand for different type of goods will be signiﬁcant and may have important effects on the terms of trade and trade ﬂows. The relationship between trade and immigration, whether they are substitutes or complements, is also an important question for bilateral trade agreements and immigration policy. It is often assumed that the goods and the labor ﬂows are substitutes, as was the case with NAFTA. It was expected that relatively freer trade between Mexico and the United States may raise Mexican wages and eventually lower the immigration from Mexico to the United States (also possibly undocumented migration) – making trade and labor ﬂows substitutes. However, Martin (2005) show that there is an evidence of increased migration post-NAFTA from Mexico to the United States and thus post-NAFTA trade and migration were complements instead of substitutes. Different demand patterns of immigrants from natives may have a signiﬁcant effect on the trade between the sending and the receiving country of the immigrants. Typically, when labor is mobile across countries, it is assumed that migration changes the labor supply of the host and the home country. While the effect of migration on the labor supply is crucial, there are other important effects of migration, in particular on the demand side that are neglected both in the migration and in the trade literature and deserve further exploration. In this chapter, in addition to the empirical investigation of the effect of immigrant income on trade, we also examine the effect of immigrants’ different demand from natives on the trade between the immigrants’ host and their home country in the widely used two input-two good standard Heckscher–Ohlin (H–O) model. We distinguish between the immigrants and the natives on the basis of their

3

It is generally assumed that both migrants and natives have identical and homothetic demand.

360

Kusum Mundra

demand patterns and assume that the immigrants on an average consume more of the goods that are available in abundance in their home countries than the natives. For instance, food is an example where immigrants and natives have different demand patterns. Immigrants demand food from their home countries and there are studies identifying that food choices are determined by individual, cultural, social, economical, and historical factors as in Fischler (1988) and Capella and Arnold (1993). The chapter is organized as follows. In Section 2 we discuss the simple H–O model used in this chapter with different demand for immigrants and natives and Section 3 talks about the effect of immigrant income on trade through their demand effect. Section 4 presents the empirical model and we conclude in Section 5. 2. Immigrant and the Heckscher–Ohlin model In this section we explore the effect of immigration on the terms of trade between the country of origin (H) of the immigrants and the country of settlement (F), if the immigrants and natives have different demand patterns, in the most extensively used H–O trade model. We assume on the lines of the demand effect of immigrants on trade that immigrants on an average demand and consume more goods from their home country than the natives. Suppose because of tariffs and other trade barriers, the relative prices of the ﬁnal goods and hence the factor prices are different in the two countries. Given the initial terms of trade before immigration, immigrants in the host country will have a different level of income and will be faced with different product prices. Therefore, at the terms of trade that prevailed in the equilibrium before immigration the aggregate world demand for commodities can change. This change on the demand side together with the change on the production side from changes in factor supplies in the two countries due to immigration, can lead to changes in the terms of trade. In our simple H–O model there are two countries, H (the immigrants country of origin or the home country) and F (the immigrants’ host country or the foreign country), i ¼ H and F. There are two goods, A and B, produced in both the countries, j ¼ A and B. There are two factors of production (labor L and capital K). Lij is the amount of labor employed in sector j in country i; K ij is the amount of capital employed in sector j in country i; wi is the wage in country i, aiLj and aiKj are, respectively, the labor–output ratio and the capital–output ratio in sector j in country i; and Dij is the demand for good j in country i. 2.1. Assumptions (A2.1) A is labor intensive and B is capital intensive, that is, for every faced price ratio ðw=rÞ ¼ o, ðaLA =aKA Þ4ðaLB =aKB Þ.

Immigrant Networks and the U.S. Bilateral Trade

361

(A2.2) There is constant returns to scale in both the sectors A and B with positive and diminishing marginal productivity. (A2.3) Country H is labor abundant and country F is capital abundant, ðK=LÞH oðK=LÞF . (A2.4) Individuals and ﬁrms are price takers. (A2.5) Country F imposes a small tariff at a rate t on its imports. (A2.6) Capital is owned equally in both the countries and is not mobile across countries. (A2.7) Each individual in country H has a continuous locally nonsatiated, strictly quasiconcave utility function U(.). Similarly the individual utility function in country F is given by V(.). At any given prices and income level people in country H buy more of good A and less of good B than people in country F.4 From assumption (A2.2) it follows that aiLj ¼ aiLj ðoÞ and aiKj ¼ aiKj ðoÞ. The requirement of full employment of labor is, aiLA Ai þ aiLB Bi ¼ Li and for capital is aiKA Ai þ aiKB Bi ¼ K i . Unit cost in each industry is equal to the market price: aiLj wi þ aiKj ri ¼ pij . Assume that country F imports A and country H imports B. Let B be numeraire, so that pB ¼ 1. Let the world equilibrium price ratio be p ¼ pA . From (A2.5) it follows that F H F F H pF ¼ pF ¼ ðpH A ¼ ð1 þ tÞ p, where p A , p A =pB Þ and pB ¼ pB ; this H H F F H F H F makes ðw =r Þoðw =r Þ where w ow and, r 4r . The higher wages in country F is an incentive for people to migrate from country H to F. Assumption (A2.6) would be cleared in the next section. 2.2. Analysis Utility maximization subject to the budget constraint gives the demand H H function for good A and good B in country H as DH A ðp ; y Þ and F F F H H DH B ðp ; y Þ, similarly in country F the demand function is DA ðp ; y Þ and F F F i DB ðp ; y Þ, where y is the individual income in country i. Let us assume mnH proportion of the world population move from country H to country F, where nH ¼ LH =ðLH þ LF Þ.5 At unchanged equilibrium price p migration affects world excess demand for good A through the following channels: (1) Effect on the production of the host country: The increase of labor supply in country F (by dLH ) increases the production of good A at unchanged H F equilibrium price, by dAF ¼ ðaF KB =a ÞdL , say X (see Appendix A). (2) Effect on the production of the home country: The fall in the labor supply of country F (by dLH ) due to migration, lowers the production H H of good A by dAH ¼ ðaH KB =a ÞdL , say Y (see Appendix A).

4 5

We assume that there is no demand reversal. In most of the countries migration and immigration is controlled by the government.

362

Kusum Mundra

(3) Effect on the demand of immigrants: (a) Price effect: The immigrants face a higher price in country F at the unchanged equilibrium price and this lowers their demand for good A by mnH DH Ap ðy; pÞdp, say T, where dp ¼ p t is the change in price for good A in terms of good B faced by the immigrants when they move from country H to country F and DH Ap ð:Þ is the partial change in the demand for good A due to the price change. (b) Income effect: The immigrants lose their income out of capital and gain income in the form of higher wages they earn in country F, it can be said that the net effect on the income is positive otherwise the immigrants have no incentive to move to the host country. The immigrants leave their capital ðmnH K H Þ behind and thus the change in the income of the immigrants due to the loss of rental income on the capital is mnH rH ðK H =LH Þ and this lowers the demand for good A by mnH rH ðK H =LH ÞDH Ay . The higher wage earned by the immigrants is given by ðwF wH Þ ¼ dw (see Appendix A). The effect on the demand for good A is given by mnH DH Ay dw, say F. (4) Effect on the demand of the population in country H who do not migrate: The capital left behind by the immigrants is enjoyed by the natives of country H and their rental income goes up by ð1 mÞnH rH ½ðK H =ð1 mÞ LH Þ ðK H =LH Þ, this in turn increases their demand for good A by ð1 mÞnH rH ½ðK H =ð1 mÞ LH Þ ðK H =LH ÞDH Ay , say S. This distribution of income assumes that there is an equal distribution of capital among the population, assumption (A2.6). With the world prices held ﬁxed at the initial equilibrium level the change in the excess demand can be written as ¼X þY þT þF þS H H ¼ dAF dAH þ mnH DH Ap ðy; pÞdp þ mn DAy dw

þ ð1 mÞnH rH ½ðK H =ð1 mÞ LH Þ ðK H =LH ÞDH Ay

(1)

H F F H ¼ mnH ½ðDH aH Ay dw þ DAp dpÞ þ ðaKB =a KB =a Þ

In the present analysis the change in the excess demand given by (1) is a priori ambigous. The effect of immigration on the terms of trade is indeterminate and the indeterminacy in this analysis comes from the demand side combined with the production side. The change in demand owing to a price change and the change in the demand owing to the change in wages work in opposite directions, therefore, the excess demand change for good A at the unchanged world price can go up, remain same or go down after immigration from one country to another. If the excess demand for good A goes up after immigration from country H to country F, then the world prices for good A must go up, moving the terms of trade in favor of country H. But if the excess demand for good A after immigration falls then the terms of trade would move against country H. Thus, this further makes a case for an empirical examination of the effect of immigration on trade.

Immigrant Networks and the U.S. Bilateral Trade

363

2.3. Sufﬁcient condition Given our assumption that stability conditions hold in the international market at the initial equilibrium prices p , if immigration increases the excess demand for good A, then the terms of trade will move in favor of good A. However, we have already shown that when both goods are normal at p , immigration will increase the demand for both goods. Therefore, it is clear that if, at p , immigration reduces the production of A in country H more than it increases the production of A in country F, then the terms of trade will move in favor of A. At ﬁxed p , ðL=KÞFA and ðL=KÞH A are ﬁxed, therefore, a sufﬁcient condition for the terms of trade to move in favor of good A (at the initial or before immigration prices and wages) is that the fall in the production of good A in the country H exceeds the increase in the production of good A in country F. This implies: H dK FA cðL=KÞFA odK H A cðL=KÞA

(2)

where cðlL=KÞ is the average product of capital written as a function of L=K. After substituting for the change in the amount of capital employed in sector A of country F after migration at p , given by H H H dK FA ¼ dLF =fðL=KÞFA ðL=KÞFB g and dK H A ¼ dL =fðL=KÞA ðL=KÞB g in (2) we get ½jdLF j=fðL=KÞFA ðL=KÞFB gcðL=KÞFA o½jdLH j=fðL=KÞH A H ðL=KÞH B gcðL=KÞA

(3)

F H At the initial equilibrium, cðL=KÞFA ocðL=KÞH A and dL ¼ dL . Thus, (3) holds if H ½ðL=KÞFA ðL=KÞFB ½ðL=KÞH A ðL=KÞB

(4)

After some manipulation (4) becomes H H F H H F H E B ðL=KÞH B =ðo ðo o ÞÞ E A ðL=KÞA =ðo ðo o ÞÞ

(5)

where E A and E B are the elasticities of factor substitution in sectors A and B. The inequality in (5) holds iff H E A =E B ðL=KÞH B =ðL=KÞA

(6)

However, the RHS of (6) is always less than 1 because good A is more labor intensive than good B. Hence, if E A E B , then (2) will necessarily hold and the terms of trade move in favor of good A. Similarly it can be shown that when E B 4E A , then the terms of trade move in the favor of good B.

364

Kusum Mundra

3. Immigrants’ income and demand In the previous studies the ﬁndings on the effect of immigrants on trade are puzzling, particularly because the two channels of immigrant links, immigrant information effect and the immigrants demand effect, are not distinguished and immigrant stock is a proxy for both the effects. In this chapter we attempt to distinguish between the immigrant ‘‘information effect’’ and the ‘‘demand or preference effect’’ by including immigrant stock (measuring the size of the immigrant network) as well as the immigrant income levels from various U.S. trading partners. Immigrants demand goods from their home country and this increases the U.S. imports from their home country. For example, Indian immigrants demand spices from India and gradually there are Indian immigrants in the United States as well as traders of non-Indian origin involved in spice trade with India. It is recognized that this will have a positive effect on the U.S. imports and will not affect U.S. exports. Immigrants’ income will signiﬁcantly affect their demand for goods from their home country, in turn affecting more U.S. imports than exports. If the home country goods are more costly in the United States than some local cheaper substitutes, the demand for home country goods will increase as immigrants’ income rises. However, if the goods from immigrants home country are inferior, higher is the immigrants’ income lower will be their demand for these goods. Immigrants demand for goods from their home country via their income will also depend on the immigrant’s enclave and assimilation in the United States. Immigrant income levels are strongly correlated with the levels of education and past studies have shown that education levels are important in determining the degree of immigrant assimilation in the United States (Borjas, 1995; Greenwood and McDowell, 1986). The literature on the immigrants assimilation in the United States have found evidence that immigrants assimilation not only depends on their education levels, but also on the number of immigrants from their home country living in the United States (Borjas, 1995; Chiswick, 1984). Chiswick and Miller (1996, 2002) measuring immigrants’ social networks by the extent of linguistic concentration in the area where the migrant resides ﬁnd that higher the immigrant network lower is immigrants’ incentive to learn English and hence lower is their assimilation into the host society. Immigrants with a large immigrant enclave will maintain their strong demand for home country goods, but will also have all the resources required to invest in import substitution activities. Dunlevy and Hutchinson (1999) ﬁnd that immigrants lower imports from New Europe, and the reason being that the new immigrants have not been in the U.S. long enough to be able to use their home-country information. But they also argue that the falling pro-trade effect of immigrants over

Immigrant Networks and the U.S. Bilateral Trade

365

time is explained by the argument that the immigrants are becoming Americanized and their ‘‘demand effect’’ is falling. Again food is an excellent example here. We do ﬁnd that the extensive varieties of salsa and Mexican hot sauce production in the United States is due to the large Mexican immigrants. With increasing immigrants from Indian subcontinent in the United States one ﬁnds more and more Indian snack that were previously imported from India are now produced by local businesses owned by Indian immigrants. All these are examples where immigrants with higher income levels and larger immigrant enclaves are substituting the imports from their home country with the U.S.-produced substitutes for ethnic home imports. In the literature on the effect of immigrant networks on trade, it is argued that the immigrant income and demand will have a more signiﬁcant effect on import, however, immigrants’ income might have an indirect effect on the strength of immigrant home link and potentially affecting exports. Larger immigrant stock have a more stronger ‘‘home-link’’ effect. With higher income and more economic assimilation the information effect often captured by immigrant stock might also be getting weaker and thus lowering the immigrant effect on exports. However, there is evidence that there might be a reverse effect with a possibility that over time and with higher upward income mobility in the United States immigrants might specialize in the production and exports of goods from the United States to their home countries. As immigrants rise up in their economic status in the United States they are in a better position and have more well developed social networks in the United States to engage in entrepreneurial activities and opening trade in new channels with their home countries. 4. Empirical model The empirical model is based on the ‘‘gravity framework’’ – where the trade between the United States and its trading partners, who are also immigrant-sending countries, is explained by different economic factors in the United States and the home countries. It is very well known in empirical trade literature that gravity model works well in overall explanation of the trade between countries and is consistent with many trade theories.6 We begin our speciﬁcation with Frankel (1997) basic constant elasticity gravity model where the trade is proportional to the product of GNP or GDP of the two countries and is inverse to the

6

Helpman (1987) showed that the bilateral trade between countries is proportional to their GDP levels in the differentiated products and increasing returns framework, whereas Deardorff (1998) has tried to reconcile the gravity models with traditional H–O frameworks.

366

Kusum Mundra

distance, Dij , between the two countries7 F ij ¼

Y iY j Dij X ij

(7)

To this multiplicative gravity model we add product of per capita GNP, which takes into account the diverse stage of development of different countries (Frankel, 1995; Rauch, 1996). The vector Xij includes factors that assist or hinder trade by inﬂuencing the transaction or transportation cost. In addition to the total income capturing the size of the economy and relative income accounting for the similarity between the United States and other countries, we include on the lines of Frankel whether the United States and its trading partners are both English speaking countries. The gravity model in (7) extends to F USj ¼ ðGNPUS GNPj Þa ðPGNPUS PGNPj Þb ðDISTANCEÞg eX USj

(8)

where F USj is U.S. imports from the home country j and exports to the home country; GNPUS GNPj is the product of the U.S. and the home country’s GNP; PGNPUS PGNPj is the product of the per capita GNP of the home country and the United States; DISTANCE is the bilateral distance between the home country and the United States and X usj ¼ ðENGLISH; lnðIMMSTOCKÞjUS ; IncomejUS Þ ENGLISH is a dummy variable measuring whether the immigrant home country is a majority English speaking country, measuring the language similarity with the United States, IMMSTOCK jUS is the stock of immigrants from country j in the United States, and INCOME jus is the average income of the immigrants from country j in the United States. With higher income we might expect that the immigrants might be demanding more of the relatively expensive goods from their home country or with higher income there is a possibility that immigrants are more assimilated within the American society and demand less of the ethnic goods.8 The log gravity model in (7) becomes ln F USj ¼ r þ a lnðGNPUS GNPj Þ þ b lnðPGNPUS PGNPj Þ þ g ln DISTANCE USjt þ dENGLISH

(9)

þ Z1 lnðIMMSTOCKÞjUS þ Z2 INCOME jUS þ USj We will expect that higher the IMMSTOCK, higher will be the positive effect on trade (Z1 40) and if the higher income might have a positive effect on trade (Z2 40) or a negative effect on trade (Z2 o0). To further explore 7 In a recent chapter Disdier and Head (2008) ﬁnd that after controlling for different sample and methods used to estimate gravity models the negative impact of distance on trade is robust. 8 This might possibly not hold for ethnic restaurant food.

367

Immigrant Networks and the U.S. Bilateral Trade

the role of the immigrants assimilation and income on trade we interact the average immigrant income from country j (INCOME) with the immigrant stock from country. Thus, X usj ¼ ðENGLISH; lnðIMMSTOCKÞjUS ; INCOME jUS ; INCOME jUS

lnðIMMSTOCKÞjUS and the model in (9) becomes ln F USj ¼ r þ a lnðGNPUS GNPj Þ þ b lnðPGNPUS PGNPj Þ þ g ln DISTANCE USjt þ dENGLISH þ Z1 lnðIMMSTOCKÞjUS þ Z2 INCOME jUS

(10)

þ Z3 lnðIMMSTOCKÞ INCOME jUS þ USj There is extensive evidence that larger the size of the immigrant enclave less is the immigrants’ incentive to assimilate with the natives and potentially less is the immigrant integration into the host society. What does this mean for the immigrant effect on bilateral trade ﬂows? Possibly that higher is the immigrant stock from country j, higher is the home effect on trade ﬂows and with rising income and large IMMSTOCK US greater will be the effect of immigrants on U.S. trade with their home country, particularly U.S. imports (Z3 40). However, there is a possibility that with larger share of immigrants from their home country the immigrants might be potentially producing the ethnic goods in the United States and substituting their imports with the goods produced in the United States. In this case we will see that the effect of higher income on the trade ﬂow with the immigrants’ home country will be mitigated by the immigrant stock (Z3 o0). For U.S. exports with rising immigrant income, signifying a higher economic assimilation of the immigrants, makes the immigrant home-link weaker (Z3 o0). To further examine the level of income assimilation of immigrants relative to the natives we include the ratio of average immigrant income from country j in the United States relative to the average native income (PINCOMEUS j ). We estimate the model given by (9) and (10) for U.S. exports and imports. 4.1. Data Our sample consists of 63 countries over 1991–2000.9 The list of the countries is given in Appendix A. The U.S. import data is obtained from the extension of the World Trade Database of Statistics Canada, which is a part of the NBER World Trade Database by Feenstra et al. (2005) and the 9

We add El Salvador and Nicaragua and remove Yugoslavia from the sample of countries used in Frankel (1997).

368

Table 1.

Kusum Mundra

Log of Export and Import, Immigrant Network and Income. U.S. Exports U.S. Imports U.S. Exports

lnðGNPUS GNPj Þ lnðPGNPUS PGNPj Þ lnðDISTANCEÞ ENGLISH lnðIMMSTOCKÞ INCOME

0.482*** (0.065) 0.597*** (0.084) 0.311* (0.158) 0.682*** (0.172) 0.266*** (0.070) 0.0001 (8.16e-06)

0.588*** (0.089) 0.490*** (0.114) 0.506** (0.219) 0.839*** (0.237) 0.4847*** (0.0941) 0.00003** (0.00001)

325 50.56 0.000

331 27.93 0.000

INCOME ðln IMMSTOCKÞ Number of observations F-statistic p-value

U.S. Imports

0.579*** 0.688*** (0.062) (0.089) 0.600*** 0.484*** (0.079) (0.110) 0.201 0.397* (0.149) (0.213) 0.979*** 1.142*** (0.166) (0.239) 1.068*** 0.915*** (0.132) (0.189) 0.0004*** 0.0004*** (0.00007) (0.00009) 0.00004*** 0.00004*** (5.84e-06) (8.31e-06) 325 331 56.73 28.70 0.000 0.000

***Signiﬁcant at 1%; **signiﬁcant at 5%; *signiﬁcant at 10%.

nominal GNP and population is from the Penn World tables.10 Annual data on immigrants across occupation is from the Immigration Statistical Yearbook by the Immigration and Naturalization Services (INS), now called Department of Homeland Security. The data on distance and English language is obtained from the Frankel.11 The annual data on average personal income for foreign born from different trading countries is derived from the March Current Population Survey for the years 1994–2000.12

5. Results Table 1 gives the results from estimating (9) and (10) for the aggregate U.S. exports and imports. From columns (1) and (2) we ﬁnd that immigrant stock has a signiﬁcant and positive effect on the U.S. bilateral trade ﬂows. A 1% increase in the immigrant stock increases U.S. exports by 0.27% and U.S. imports by 0.48%.13 However, we ﬁnd that a 1% increase in 10

The trade data is downloaded from the Center for International Data at the UC Davis (http://cid.econ.ucdavis.edu.) and the website for the Penn World Tables is http:// pwt.econ.upenn.edu. 11 Distance is from ‘‘Direct-Line Distances’’, International Edition, Gary L. Fitzpatrick and Marilyn J. Modlin, Scarecrow Press, Inc. Metuchen NJ and London 1986. 12 Foreign born income is missing for 28 countries in 1994 CPS. 13 This is in line with the previous ﬁndings in the literature.

Immigrant Networks and the U.S. Bilateral Trade

369

the average immigrant income level lowers U.S. imports by 0.003%. Thus, higher income levels of the immigrants in the U.S., signifying more assimilation of the immigrants in the U.S., lowers U.S. imports. However, we ﬁnd a similar signiﬁcant negative effect of income on both exports and imports when we interact the income level with the size of the immigrant enclave. From columns (3) and (4) we ﬁnd that a 1% increase in the income level lowers the U.S. exports and imports by 0.005%. This indicates that higher income coupled with a larger size of the immigrant enclave weakens the effect of immigrant networks on trade ﬂows, both for exports and imports. In Table 2 we give the results from estimating the effect of average income of immigrants from country j relative to natives, a better measure of immigrant assimilation than simply the average level of immigrant income from country j. From col columns and (2) in Table 2 we ﬁnd that higher is the PINCOMEUS lower is the effect on U.S. imports. This clearly shows that as the immigrants income levels are closer to that of the natives or rising above the natives, higher is the immigrant assimilation in the United States and lower is their demand for the home country goods. When we interact the level of PINCOMEUS with the level of the immigrant stock, we ﬁnd that for both the U.S. exports and imports higher PINCOMEUS lowers the trade ﬂows. The fall is higher for U.S. exports (around 0.10%) than the imports (around 0.9%). Table 2.

Log of Export and Import, Immigrant Network and Relative Income.

lnðGNPUS GNPj Þ lnðPGNPUS PGNPj Þ lnðDISTANCEÞ ENGLISH lnðIMMSTOCKÞ PINCOMEUS

U.S. Exports

U.S. Imports

U.S. Exports

U.S. Imports

0.490*** (0.064) 0.602*** (0.084) 0.308* (0.158) 0.720*** (0.172) 0.256*** (0.070) 0.004 (0.002)

0.589*** (0.089) 0.494*** (0.114) 0.503** (0.219) 0.839*** (0.237) 0.141 (0.097) 0.005** (0.002)

325 51.59 0.000

331 27.99 0.000

0.615*** (0.061) 0.589*** (0.077) 0.173 (0.145) 0.979*** (0.166) 1.095*** (0.164) 0.095*** (0.012) 0.009*** (0.001) 325 61.97 0.000

0.702*** (0.090) 0.478*** (0.111) 0.386* (0.213) 1.142*** (0.239) 0.984*** (0.203) 0.080*** (0.018) 0.008*** (0.002) 331 28.64 0.000

PINCOMEUS ðln IMMSTOCKÞ Number of observations F-statistic p-value

***Signiﬁcant at 1%; **signiﬁcant at 5%; *signiﬁcant at 10%.

370

Kusum Mundra

Other variables are what we expected. GNP and PGNP are all positive and signiﬁcant. English language dummy have a signiﬁcant positive effect on both U.S. exports an imports. Distance has a negative signiﬁcant effect on trade ﬂows.

6. Concluding remarks The effect of immigrants’ demand in their host country has been neglected when analyzing the effect of immigration. In the literature exploring the effect of immigrants on trade, immigrant stock is a proxy for both the immigrant information effect and the demand effect. In this chapter we propose to include the effect of income in the host country United States over and above the size of the immigrant stock while examining the effect of immigrant networks on trade. Immigrants relative income to the natives will give us some information on the extent of assimilation of the immigrants in the United States and this assimilation will have an important effect on trade ﬂows, a priori more so for imports than exports. In this chapter we emphasize that immigrants are more than laborers and they have different demand for goods from the natives. We assume that immigrants on an average consume more of the goods that are abundant in their home country in a simple H–O model and ﬁnd that at the terms of trade that prevailed in the equilibrium before immigration, the aggregate world demand for commodities can change. Such a change on the demand side, together with the change on the production side that results from immigration across two countries can lead to changes in the terms of trade. Our econometric model consisting of 63 major trading and immigrant sending country for the United States over 1991–2000 show that the immigrants income, mostly through demand effect, has a signiﬁcant negative effect on U.S. imports only. However, if we include the effect of the immigrant income interacted with the size of the immigrant network, measured by the immigrant stock, we ﬁnd that the income has a negative effect on both the U.S. exports and imports. Higher income of the immigrants coupled with the large size of the immigrants stock weakens the immigrants network effect with their home country, lowering the immigrant network effect for both U.S. exports and imports. This we ﬁnd in addition to the immigrants stock elasticity of 0.27% for U.S. exports and 0.48% for U.S. imports. In this chapter we argue that the immigrant network effect on trade ﬂows is weakened by the level of immigrant assimilation. We capture immigrant assimilation by their level of income in the U.S. We ﬁnd a stronger effect of income assimilation on U.S. imports than exports. This chapter is an attempt to raise the question that simply looking at the

Immigrant Networks and the U.S. Bilateral Trade

371

size of the immigrant stock to capture the effect of the immigrant networks on trade might only be a part of the picture, the effect of immigrant assimilation in the host country also needs to be examine in detail while examining the effect of the immigrant networks on trade. Appendix A Preimmigration trade production in both the countries is as follows: i i i i i i i i Ai ¼ ð1=ai ÞðLi ai KB K aLB Þ ¼ ½L ðkB k Þ=½aLA ðkB kA Þ, i i i i i i i i Bi ¼ ð1=ai ÞðK i aH LA L aKA Þ ¼ ½L ðkB k Þ=½aLA ðkB kA Þ i i i i i i H H H where ai ¼ aH LA aKB aLB aKA , kA ¼ ðK=LÞA , kB ¼ ðK=LÞB , k ¼ ðK=LÞ , and i ¼ H; F. The wage–rental ratio in both the countries are: H H H H H H H oH ¼ ðpaH KB aKA Þ=ðaLA aLB pÞ ¼ kB ðp ðaKA =aKB ÞÞ=ððaLB =aLA Þ pÞ,

oF ¼ ðpaFKB aFKA Þ=ðaFLA aFLB pÞ ¼ kB ðpð1 þ tÞ ðaFKA =aFKB ÞÞ=ððaFLB =aFLA Þ pÞ The higher wage income earned by the immigrants is given by: H H F F F H dw ¼ ½aF ðpaH KB aKA Þ a ðpð1 þ tÞaKB aKA Þ=a a

Appendix B The 63 trading partners are Algeria, Argentina, Australia, Austria, Belgium, Bolivia, Brazil, Canada, Chile, China, Colombia, Denmark, Ecuador, Egypt, El Salvador, Ethiopia, Finland, France, Ghana, Greece, Hong Kong, Hungary, Iceland, India, Indonesia, Iran, Ireland, Israel, Italy, Japan, Kenya, Kuwait, Libya, Malaysia, Mexico, Morocco, Netherlands, New Zealand, Nicaragua, Nigeria, Norway, Pakistan, Paraguay, Peru, Philippines, Poland, Portugal, Saudi Arabia, Singapore, South Africa, South Korea, Spain, Sudan, Sweden, Switzerland, Taiwan, Thailand, Tunisia, Turkey, United Kingdom, Uruguay, Venezuela, Germany. References Bandyopadhyay, S., Coughlin, C., Wall, H. (2008), Ethnic networks and U.D. exports. Review of International Economics 16 (1), 199–213. Borjas, G.J. (1995), The economic beneﬁts from immigration. Journal of Economic Perspectives Spring 3–22.

372

Kusum Mundra

Capella, L.M., Arnold, D.R. (1993), Acculturation, ethnic consumers, and food consumption patterns. Journal of Food Products Marketing 1 (4), 61–79. Chiswick, B.R. (1984), Illegal aliens in the United States labor market: an analysis of occupational attainment and earnings. International Migration Review 18, 714–732. Chiswick, B.R., Miller, P.W. (1996), Ethnic networks and language proﬁciency among immigrants. Journal of Population Economics 9 (1), 19–35. Chiswick, B.R., Miller, P.W. (2002), Immigrant earnings: language skills, linguistic concentrations and the business cycle. Journal of Population Economics 15. Deardorff, A. (1998), Determinants of bilateral trade: does gravity work in a neoclassical world? In: Frankel, J.A. (Ed.), The Regionalization of the World Economy. University of Chicago Press, Chicago, pp. 7–28. Disdier, A., Head, K. (2008), The puzzling persistence of the distance effect on bilateral trade. Review of Economics and Statistics 90 (1), 37–48. Dunlevy, J.A., Hutchinson, W.K. (1999), The impact of immigration on American import trade in the late nineteenth and early twentieth century. The Journal of Economic History 59 (4), 1043–1062. Feenstra, R.C., Lipsey, R.E., Haiyan, D., Ma, A.C., Mo, H. (2005), World trade ﬂows: 1962–2000. NBER Working Paper no. 11040. Massachusetts, USA. Fischler, C. (1988), Food, self and identity. Social Science Information 27, 275–292. Frankel, J.A. (1997), Regional Trading Blocs in the World Economic System. Institute for International Economics, Washington, DC, USA. Gould, D.M. (1994), Immigrant links to the home country: empirical implications for U.S. bilateral trade ﬂows. The Review of Economics and Statistics 76, 302–316. Girma, S., Yu, Z. (2002), The link between immigration and trade: an evidence from the United Kingdom. Review of World Economics 138 (1), 115–130. Greenwood, M.J., McDowell, J.M. (1986), The Factor Market Consequences of U.S. Immigration. Journal of Economic Literature 24 (4), 1738–1772. Grief, A. (1993), Contract enforceability and economic institutions in early trade: the Maghribi Traders’ coalition. American Economic Review 83 (3), 525–548. Head, K., Reis, J. (1998), Immigration and trade creation: econometric evidence from Canada. Canadian Journal of Economics 31 (1), 47–62. Helpman, E. (1987), Imperfect competition and international trade: evidence from fourteen industrial countries. Journal of the Japanese and International Economics 1 (1), 62–81.

Immigrant Networks and the U.S. Bilateral Trade

373

Herander, M.G., Saavedra, L.A. (2005), Exports and the structure of immigrant-based networks: the role of geographical proximity. The Review of Economics and Statistics 87 (2), 323–335. Halter, M. (1995), New Migrants in the Marketplace: Boston’s Ethnic Entrepreneurs. University of Massacusetts Press, Amherst, MA. Light, I., Bonacich, E. (1988), Immigrant entrepreneurs: Koreans in Los Angeles. University of California, CA. Light, I., Zhou, M., Kim, R. (2002), Transnationalism and American exports in an english-speaking world. International Migration Review 36, 702–725. Martin, P. (2005), NAFTA and Mexico-US migration. In: Hufbauer, G.C., Schott, J.J. (Eds.), NAFTA Revisited. Institute for International Economics, Washington, DC. Portes, A., Rumbaut, R.G. (1996), Immigrant America: A portrait, second ed. University of California Press, Berkeley, CA. Rauch, J.E., Casella, A. (1998), Overcoming informational barriers to international resource allocation: prices and group ties. NBER Working Paper no. 6628. Rauch, J.E. (1996), Networks versus markets in international trade. Journal of International Economics 48, 7–35. Treﬂer, D. (1997), Immigrants and natives in general equilibrium models. NBER Working Paper no. 6209. Massachusetts, USA. Available at http://www.nber.org/papers/ Wagner, D., Head, K., Reis, J. (2002), Immigration and the trade of provinces. Scottish Journal of Political Economy 49 (5), 507–525.

CHAPTER 16

The Societal Integration of Immigrants in Germany Michael Fertiga,b,c a

ISG-Institut fu¨r Sozialforschung und Gesellschaftspolitik, Ko¨ln, Germany RWI-Rheinisch-Westfa¨lisches Institut fu¨r Wirtschaftsforschung, Essen, Germany c Institut zur Zukunft der Arbeit (IZA), Bonn, Germany E-mail address: [email protected] b

Abstract This chapter investigates whether and to what extent immigrants in Germany are integrated into German society by utilizing a variety of qualitative information and subjective data collected in the 1999 wave of the German Socio-Economic Panel (GSOEP). To this end, leisure-time activities and attitudes of native Germans, ethnic Germans and foreign immigrants of different generations are compared. The empirical results suggest that conditional on observable characteristics the activities and attitudes of foreign immigrants from both generations differ much more from those of native Germans than the activities/attitudes of ethnic Germans. Furthermore, the attitudes of second-generation immigrants tend to be characterized by a larger degree of fatalism, pessimism and self-doubt than those of all other groups, although their activities and participation in societal life resemble more those of native Germans than those of their parents generation. Keywords: Subjective data, ﬁrst- and second-generation immigrants, ethnic Germans Jel classiﬁcations: J15, J61

1. Introduction Together with the enlargement of the European Union and the consequences of demographic change, the integration of immigrant minorities is Europe’s most important challenge over the next decade. These three challenges are intimately related. The enlargement of the European Union to incorporate countries of Central and Eastern Europe Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008022

r 2010 by Emerald Group Publishing Limited. All rights reserved

376

Michael Fertig

will in all likelihood be associated with additional – though probably moderate (see Bauer and Zimmermann, 1999; Fertig, 2001; or Fertig and Schmidt, 2001a) – migration ﬂows toward the current member states. These ﬂows in turn will have effects on overall population growth, and potentially on the relative status of the immigrant communities in each country. At this stage, however, we do not sufﬁciently understand the mechanisms governing the integration of immigrant minorities into society, and the available policies to smooth this process. An illustrative example in this context is Germany. In the period up to the 1970s migrants to Germany were mainly labor migrants from Southern Europe, driven by labor market opportunities in Germany and depressed conditions in the sending regions. Over the past three decades, the ethnic composition of immigration to Germany has changed (see Fertig and Schmidt, 2002), and the geographic and cultural gaps between Germany and the sending countries have widened. Furthermore, due to its citizenship law Germany has a sizeable community of second generation foreigners whose social and economic characteristics and outcomes are a matter of growing concern (see, e.g., the symposium on second-generation immigrants in the Journal of Population Economics, 2003). Figure 1 illustrates the share of foreigners living in Germany for selected years. Before 2000,1 citizenship law in Germany was dominated exclusively by the jus sanguinis principle, that is, citizenship was acquired by descent. Only children born to either a German mother or a German father received German citizenship upon birth. This regulation that dated back to a law from 1913 was reformed in 2000 to some extent. Children born after January 1, 2000, will be German nationals by birth if at least one parent is German or if at least one parent legally lives in Germany for a minimum of eight years. Moreover, naturalization law that used to be rather restrictive was reformed in 2000 as well. Thus, since 2000 it is considerably easier for children born to foreigners in Germany to acquire German citizenship. These reforms are associated with a decline in the share of foreigners in born in Germany that reached its maximum in 2000 (22.1%) and has declined to around 19% in 2009 (Figure 1). Many observers of the situation of immigrants in Germany fear that as migrant integration opportunities remain limited, the risk of increasing economic and cultural isolation rises, setting the stage for the creation of permanent second class citizens. For instance, participants of the European Economic and Social Committee (EESC) conference on the integration of immigrants emphasized the need for increased political rights for migrants, in addition to equal access to welfare, health, and education (see EESC press release No. 64/2002, September 2002). In Germany,

1

The year 2000 is especially relevant in the context of this chapter, since our data refer to the 1999 (see also below).

377

The Societal Integration of Immigrants in Germany 25

Percent

20

15

10

5

0 1995

2000

2005

2009

Fig. 1. Share of foreigners born in Germany (in % of all foreigners). the Federal Ofﬁce for Migration and Refugees encourages the social and societal integration of immigrants by supporting integration projects in cooperation with associations, foundations, initiatives, and other authorities with the explicit aim to communicate values and norms, to establish contacts between immigrants and natives and to promote societal acceptance of immigrants. Despite the growing recognition of this situation, relatively little research has targeted the question of migrants’ integration into society, nor are the potential consequences of different policies regarding the participation of migrants and other minorities in the society and the political process fully understood. Even less is known about the integration of the descendants of the migrants, the so-called secondgeneration immigrants. This chapter aims at contributing to a better understanding of these processes by investigating whether and to what extent immigrants in Germany are integrated into the German society. To this end, we utilize a variety of qualitative information and subjective data collected in the 19992 wave of the German Socio-Economic Panel (GSOEP) and compare native Germans, ethnic Germans and foreign immigrants of different generations along various dimensions. Speciﬁcally, we investigate whether there are differences between these groups regarding their leisure-time activities and their attitudes toward speciﬁc areas of life. Among the latter are areas which are perceived as important for individual well-being and satisfaction and different views on various aspects of life. Finally, we analyze a range of indicators of the 2

This is the only wave of data containing all relevant items, especially the questions on fundamental attitudes.

378

Michael Fertig

societal integration of immigrant groups which are collected for these groups only, like their German language ability or their contact with natives. In this endeavor, we control for a large set of observable characteristics of individual respondents to account for heterogeneity in individual activities and attitudes. The empirical results suggest that conditional on observable characteristics the activities and attitudes of foreign immigrants from both generations differ much more from those of native Germans than the activities/attitudes of ethnic Germans. Since ethnic Germans are (ﬁrst generation) immigrants who receive German citizenship directly upon entry, these results suggest that citizenship status seems be an important constituent of societal integration. This conclusion is also supported by the ﬁnding that the attitudes of second-generation immigrants tend to be characterized by a larger degree of fatalism, pessimism and self-doubt than those of all other groups, although their activities and participation in societal life resemble more those of native Germans than those of their parents generation. The remainder of this chapter is organized as follows. Section 2 provides an overview on the existing literature regarding the economic and social integration of immigrants. In Section 3 the utilized data and the empirical strategy are explained. Estimation results are presented in Section 4 and Section 5 offers some conclusions.

2. Economic and societal integration Economic research concerning migration issues can be conceptualized into three broad ﬁelds: (i) the decision to migrate, (ii) the performance of migrants in the destination country, and (iii) the impact of immigration on the population indigenous to the destination country. All these research areas are intimately related and carry important implications for immigration policy. The integration of immigrants into destination countries’ societies is a central part of the research done under the heading of (ii). Typically, analyses conducted within this ﬁeld investigate whether wages or employment prospects of immigrants converge or diverge as the duration of residence unfolds compared to that of natives and which reasons can be found for these developments. Another aspect of this line of research concerns the degree of discrimination against immigrants as well as the degree and the consequences of geographical and/or occupational segregation, that is, the clustering of immigrants or speciﬁc immigrant groups in certain geographical areas or occupational groups. The received literature for the US-American experience demonstrates that skills play a dominant role for immigrant performance. These do not only comprise human capital acquired formally as secondary or

The Societal Integration of Immigrants in Germany

379

post-secondary schooling and vocational training, but also informally like labor market experience, or cognitive ability and motivation (see, e.g., the seminal papers by Chiswick, 1978, and Borjas, 1985, 1987). Furthermore, these contributions provide evidence that only part of the human capital acquired by immigrants in their origin country can be transferred to the labor market at the destination. Consequently, upon arrival these immigrants possess a lower earnings capacity, and – since their labor supply is typically inelastic – relatively low earnings. Over their time of residence, they tend to acquire the lacking human capital, for example, the language spoken at the destination. Their low initial earnings capacity implies that the opportunity cost of their investment are relatively low, which makes substantial human capital acquisition likely. After some years of residence in the destination country the earnings of immigrants typically catch up to those of natives (Chiswick, 1978). For the case of Germany, several empirical analyses address the issue of wage performance of the so-called guest workers in the German labor market of the 1980s and early 1990s (see, e.g., Dustmann, 1993; Kurthen et al., 1998; Schmidt, 1997). On balance, these papers demonstrate that in the German labor market formal skills play a decisive role for immigrant wage earnings. For instance, Schmidt (1997) concludes that those immigrants who received their schooling and post-secondary education in Germany achieve earnings parity with native workers, while the typical ﬁrst-generation migrant from the ‘‘guest worker’’ countries lags some 20% behind the average native worker in terms of wages. Dustmann (1993) demonstrates that the distinction of permanent and temporary migrants might be important for the question of earnings dynamics. Furthermore, Schmidt (1997) compares migrants from the ‘‘guest worker’’ countries with ethnic German immigrants – concluding that the latter group of immigrants is typically better educated and economically well integrated. Finally, Dustmann and Schmidt (2000) address the wage performance of female immigrants. To date, almost the complete migration literature and certainly all studies of the German case have concentrated on the analysis of the economic performance of male immigrants. In their paper, Dustmann and Schmidt (2000) emphasize the treatment of labor supply issues that plague all analyses of female wage earnings. They conclude that for the relative wages of female immigrants not only their own formal education, but also their family circumstances – most notably the return plans of their family – play an important role. In general, the majority of the received literature in this ﬁeld concentrates on relative economic success. The focus is almost exclusively on measurable differences in economic outcomes (e.g., wages or employment opportunities) that cannot be traced back to observable differences in the determinants of these outcome measures. One exception is Dietz (2003). The author investigates group formation, values, and attitudes of a

380

Michael Fertig

sample of young ethnic German immigrants who entered Germany from the former Soviet Union between 1990 and 1994. Her results indicate that the circle of friends of the majority of these youngsters consists primarily of members of their own group, which they suffer from language problems and reside in rather segregated areas. Furthermore, the values and attitudes of this immigrant group are characterized by a high acceptance of parental authority, rather traditional gender roles and strong orientation toward collective values rather than an individualistic life style. Another exception is Dustmann (1996). The author investigates the determinants of the feeling of national identity for migrants living in Germany. His results suggest that individual demographic characteristics, nationality and indicators for the family context of respondents affect migrant’s social integration. By contrast, indicators for the labor market status do not exhibit signiﬁcant effects. Moreover, almost all studies for the case of Germany concentrate on ﬁrst-generation migrants, whereas the offspring of these immigrants, the so-called second generation, has not attracted a comparable level of attention. There are two notable exceptions. First, Fertig and Schmidt (2001b) provide a detailed characterization of both immigrant generations in Germany by demographic and socioeconomic characteristics. From their analysis it becomes transparent that there exist considerable differences between both immigrants and natives as well as among the different immigrant generations themselves. The paper, furthermore, investigates the welfare dependence of migrants and contrasts the ﬁndings on the determining factors of the moderate risk of migrants to depend on public assistance payments with the perception of immigrants by native Germans using two complementary datasets. And second, Riphahn (2003) investigates the educational attainment of second-generation immigrants in Germany by analyzing school attendance and completed schooling degrees. The author ﬁnds that after controlling for a variety of individual background characteristics statistically signiﬁcant negative differences between second-generation migrants and comparable natives remain. The ultimate aim of this chapter is the provision of a comprehensive portrait regarding various aspects of the societal integration of different immigrant groups in Germany by analyzing a large set of individual-level data for the year 1999. The next section explains the utilized dataset and the pursued empirical strategy.

3. Empirical strategy and data Measuring societal integration is anything but trivial. Since there is no objective scale, this phenomenon is by its very nature relative. That is, a speciﬁc group of individuals might resemble the behavior or the attitudes/values of a chosen reference group relatively more than another

The Societal Integration of Immigrants in Germany

381

group and might therefore be labeled more integrated. However, the reference group is obviously a choice variable and the extent to which the members of the chosen reference group perform an adequate benchmark might be controversial. Furthermore, preferences, tastes, and values clearly vary from one individual to another, inducing the necessity to control for observable heterogeneity between different respondent groups. But even if signiﬁcant differences between certain population groups remain after controlling for socio-demographic characteristics, it is difﬁcult to establish precisely if these differences are large or frequent enough to label them societal disintegration. In the case at hand, German citizens who were born in Germany form the comparison group for all immigrant groups. Furthermore, we pursue a careful examination and interpretation of estimation results to ward off fallacious conclusions given the aforementioned difﬁculties. In our empirical analyses, we utilize individual-level data from the 1999 wave of the GSOEP. The GSOEP is a representative longitudinal study of private households in Germany. It collects information on all household members, consisting of Germans living in the old and new German states, foreigners, who have entered the country in the 1960s and early 1970s, and recent immigrants to Germany. Information collected includes household socioeconomic composition, occupational biographies, employment, earnings, as well as health and life satisfaction indicators. Furthermore, there are different waves with special questionnaires on, for example, social security, education, and training. The 1999 wave contains a special set of questions related to respondents’ views on life and on the importance of different areas of life for satisfaction and well-being. We explicitly consider the following mutually exclusive immigrant groups in Germany: (i) Ethnic German immigrants, (ii) ﬁrst-generation (foreign) immigrants, and (iii) second-generation (foreign) immigrants. These groups are deﬁned as follows. Ethnic German Immigrants: This group of migrants that entered Germany from Eastern Europe during the 1990s and that receives citizenship status immediately upon arrival is not directly observable in the data. However, the data provide information on German citizenship, place of birth, and immigration year. Therefore, all respondents possessing the German citizenship, which were not born in Germany and which did not live in Germany before 1990 were accounted as ethnic German immigrants. Clearly, this deﬁnition is not completely accurate, since it is possible that German citizens who were born outside Germany and entered the country after 1990 are accounted as ethnic German immigrants as well. However, the vast majority of these people should be ethnic Germans who immigrated from Eastern Europe during the 1990s. First-generation (foreign) immigrants: This group contains respondents without German citizenship who were not born in Germany. The majority

382

Michael Fertig

of individuals in this group comprises the so-called guest workers of the 1960s and early 1970s. Second-generation (foreign) immigrants: This group contains respondents without German citizenship which were born in Germany. The majority of individuals in this group comprises the offspring of the so-called guest worker immigrants of the 1960s and early 1970s. The distinction between these three immigrant groups, and especially the differences in treatment with respect to citizenship status, allows us the shed some light on the nexus of societal integration and citizenship status. In our empirical analyses all three groups are compared to respondents possessing the German citizenship who were born in Germany. In these comparisons, we control for various individual characteristics of the respondents. Besides the immigrant group indicators, respondents’ education, marital status, gender, age, employment status, years of residence in Germany, and other characteristics are taken into account.3 Table A1 in the appendix provides a detailed description of all explanatory variables. To analyze the societal integration of different immigrant groups living in Germany, we utilize three large sets of questions: (a) Questions on leisure-time activities, (b) questions on attitudes, and (c) foreigners/ immigrants speciﬁc questions. For the ﬁrst two sets of questions information is collected for native Germans as well as for all immigrant groups. The last set is speciﬁc to the situation of foreigners/immigrants in Germany. Hence, for this set a comparison to Germans is not possible. Tables A2–A4 in the appendix provide detailed descriptions of the various questions and the answer possibilities. The ﬁrst set comprises questions on leisure-time activities that are supposed to measure the degree of immigrant participation in cultural and leisure activities. This does not only entail the extent to which respondents participate in, for example, cultural, religious, or sport events but also how much they engage in social intercourse with friends or neighbors and the degree they are involved in public initiatives or political parties. The second set of questions comprises the attitudes of respondents toward areas that are important for their well-being and satisfaction. These areas encompass the personal sphere – for example, the importance of family, friends, and career success – as well as general areas like environmental protection and the maintenance of peace. Furthermore, this set also comprises the degree of agreement on several statements regarding attitudes toward life and the future. For instance, respondents are asked for their (dis-)agreement to the statements ‘‘I decide the way my 3

Because a substantial share of respondents – especially within the group of second generation immigrants – is rather young and, hence, still in education, we abstain from using income/wages or other job-related characteristics (e.g., occupation or hours worked) as control variables.

The Societal Integration of Immigrants in Germany

383

life is run,’’ ‘‘In comparison to others, I haven’t achieved what I deserve,’’ and ‘‘If I ever hit upon difﬁculties in my life, I doubt my capabilities.’’ Therefore, the extent to which respondents agree to these statements can be interpreted as indicators for the degree of fatalism, self-doubt, and discontent with which they perceive their own life. Finally, this set also contains a question on respondents’ general optimism toward the future, the extent to which they feel connected with the place they live and their willingness to move away from this place. The third set of questions that is conﬁned to immigrants/foreigners only contains data on the proximity between immigrants and natives (existence of contacts and visits) as well as on the language ability of respondents (regarding German and the language of the origin country). Furthermore, respondents are asked which language they typically use in everyday life and how they perceive their acceptance in German society (experience of disadvantages; wish to stay permanently; feeling as German and connection to origin country). Table A5 in the appendix reports some summary statistics for the utilized sample. From this table, it becomes transparent that for many questions there are large (unconditional) differences in the answer distributions for the different groups. However, the last panel of Table A5 reveals that these groups also differ considerably with respect to observable characteristics. Therefore, a multivariate analysis that controls for observed heterogeneity between respondents is indispensable. The results of our (ordered) probit analyses are reported in the next section. 4. Results In this section the estimation results of our empirical application are reported. Owing to the large number of estimations, it is infeasible to report the full set of results. Rather, the following tables contain a summary of the estimation results that indicate the direction and signiﬁcance of coefﬁcient estimates only.4 In these tables a ‘‘þ’’ denotes a statistically signiﬁcant (95% level) positive difference between the estimated group indicators. A ‘‘’’ indicates that this difference is statistically signiﬁcant negative, and a ‘‘0’’ denotes an insigniﬁcant difference between the respective groups. That is, for instance, the information in row 1 of Table 1 has to be interpreted as follows: The ‘‘þ’’ in columns 1 and 2 indicate that Germans (born in Germany) display a higher probability to visit cultural events than foreigners and ethnic Germans. By contrast, the ‘‘’’ in column 5 suggests that the ﬁrst generation of (foreign) immigrants tend to be less likely than ethnic Germans to visit cultural events. The ‘‘0’’ in the last column 4

Full estimation results are available from the author upon request.

German vs. foreign

German vs. ethnic German

Ethnic German vs. foreign

0

0 0

0

0

0

0

0

0

0

0

0

First vs. second

Notes: A ‘‘þ’’ denotes a statistically signiﬁcant positive, a ‘‘’’ a statistically signiﬁcant negative, and ‘‘0’’ an insigniﬁcant difference between estimated group indicators. For a description of the utilized control variables see Table A1 in the appendix.

þ

0

0

0

0 0

0 þ

þ

þ

þ

Second vs. ethnic German

þ

Second vs. German

First vs. Ethnic German

First vs. German

Results of ordered probit estimations for leisure-time activities

Which of the following activities do you participate in during your free-time? Visits to cultural events þ þ þ Cinema visits, visits to pop 0 0 þ concerts, discos, etc. Active sport þ 0 þ Social intercourse with friends, 0 relatives, or neighbors Lend help to friends, relatives, or 0 0 0 neighbors Honorary ofﬁce participation þ þ 0 in clubs, etc. Participation in public 0 0 0 initiatives, etc. Church-going, visits to religious þ events

Question

Table 1.

384 Michael Fertig

The Societal Integration of Immigrants in Germany

385

indicates that there is no difference between the second and the ﬁrst generation of migrants. Table 1 reports a summary of estimation results for the different leisuretime activities. From this table it becomes transparent, that even after controlling for observable differences between respondent groups like age, gender, education, marital status, etc., signiﬁcant differences between natives and foreigners in Germany remain. Estimation results indicate that Germans compared to foreigners display a signiﬁcantly higher probability to visit cultural events and to do sports actively. Furthermore, they are signiﬁcantly more likely than foreigners to participate in clubs etc. as a honorary ofﬁce worker but display a statistically signiﬁcant lower probability to engage in social intercourse with friends or neighbors and to be involved in religious activities. In general, the differences between the ﬁrst generation of immigrants and Germans born in Germany are much more pronounced than those between natives and the second generation. Existing differences between both immigrant groups indicate that the second generation is closer to native Germans than their parents. However, in the majority of cases the differences between both immigrant generations are negligible. By contrast, ethnic Germans and Germans born in Germany tend to behave similarly. For the majority of leisure-time activities estimation results indicate no statistically signiﬁcant difference between these two groups. The only exceptions are, ﬁrst, that ethnic Germans are less likely than native Germans to visit cultural events and to participate in clubs etc. as honorary worker. And second, ethnic Germans display a statistically signiﬁcant higher probability to be involved in religious activities. In general, ethnic Germans are the population group with the highest incidence of religious activity in their leisure-time. Furthermore, leisure-time activities of ethnic Germans tend to be signiﬁcantly different from those of noncitizens. For the most part, these signiﬁcant differences are driven by the discrepancies between ethnic Germans and the ﬁrst generation of (foreign) immigrants, whereas the activities of the second generation are more similar to those of ethnic Germans. Overall, all immigrant groups in Germany are participating in various dimensions of societal life where the second generation of (foreign) immigrants seems to be more assimilated to the activities of native Germans than their parents. In Table 2A the results for ﬁrst part of the attitudes questions are reported. Here respondents are asked which areas of life are important for their well-being and satisfaction. Estimation results indicate that in the majority of cases there are no signiﬁcant differences between Germans and foreigners. In contrast to the leisure-time activities, this result is, however, mainly driven by the similarity in responses of ﬁrst-generation immigrants and Germans, whereas the answers of a typical respondent from the second immigrant generation differs more from those of Germans.

German vs. foreign

German vs. ethnic German

Ethnic German vs. foreign

0

0

0

0 þ

N.A.a 0 0 0 0 0 0

Second vs. ethnic German

0 0 0 0 0 0 0 0 þ 0 0 0

Second vs. German

þ 0 0 0 0 0 0 0 0 0 0 0 0 0

First vs. second

Notes: A ‘‘þ’’ denotes a statistically signiﬁcant positive, a ‘‘’’ a statistically signiﬁcant negative, and ‘‘0’’ an insigniﬁcant difference between estimated group indicators. a Since all respondents in the Group of ethnic Germans have chosen the same answer category, this comparison is not possible. For a description of the utilized control variables see Table A1 in the appendix.

N.A.a þ 0 0 0 0 0

First vs. ethnic German

0 0 0 0 0 0 0 0 0 0 þ 0 þ

First vs. German

Results of (ordered) probit estimations for attitudes

Which of the following areas are important for your well-being and satisfaction? Work þ þ þ N.A.a Family 0 N.A.a Friends 0 þ Income 0 0 0 Housing 0 0 þ Inﬂuence on political decisions 0 þ Career success þ 0 0 Free-time 0 0 0 Health 0 0 0 Protection of the natural environment 0 þ Faith, religion 0 Residential area 0 þ Mobility to get everywhere quickly þ If you think about the future in general, 0 þ are you optimistic? To what extent do you feel connected with 0 0 þ the place and the area that you live in? Would you consider moving away, e.g., þ þ because of family or job?

Question

Table 2A.

386 Michael Fertig

The Societal Integration of Immigrants in Germany

387

Estimation results, furthermore, suggest that Germans have a signiﬁcantly higher probability than foreigners to regard work and career success as important factors for their well-being. On the contrary, religion and mobility tend to be signiﬁcantly less important for them. The importance of religious activities, however, is especially pronounced in the ﬁrst generation of immigrants, whereas the second generation tends to perceive this area as important as native Germans. For the most part, ethnic Germans tend to perceive different areas as important for their well-being and satisfaction than (foreign) immigrants of both generations and Germans born in Germany. Compared to the latter group, ethnic Germans display a signiﬁcantly higher probability to regard inﬂuence on political decisions, environmental protection, the residential area they are living, mobility and religion as important factors for their well-being. The latter ﬁnding supports the results from Table 1 where ethnic Germans display a higher probability to be involved in religious activities during their free-time. Furthermore, ethnic Germans tend to consider work as less important than native Germans. These results suggest that for ethnic Germans collective values like political inﬂuence, environmental protection, residential area, and religion carry more weight than individualistic values like work. Against the background of the poor economic conditions in the countries they emigrated from, this is certainly a surprising result that might be a reﬂection of their upbringing in a socialistic society and supports the ﬁndings of Dietz (2003) for ethnic German youngsters. The lower panel of Table 2A aims at inquiring how optimistic respondents are regarding future. Furthermore, the ﬁnal two questions address the extent to which respondents feel connected with the place they live in and whether they are willing to move away from there. Estimation results suggest that ethnic Germans exhibit the highest probability to look ahead optimistically, whereas both foreign immigrant generations are more pessimistic. Interestingly, ethnic Germans feel more connected to the place or area they are living than native Germans or noncitizens. They are, however, also the most willing to move away for reasons of family or job. The second generation of immigrants is the population group that is the most similar to ethnic Germans with respect to these issues, whereas their parents generation and Germans born in Germany display the lowest willingness to be mobile. Table 2B contains a summary of the estimation results for the second part of the attitudes questions under investigation. These questions try to establish the degree of respondents’ agreement to different views on life, and therefore, try to extract rather fundamental attitudes of respondents. The ﬁrst ﬁve questions can be interpreted as aiming to extract the degree of fatalism with which respondents view their life and its prospects. Interestingly, foreigners unambiguously tend to display a higher degree of fatalism than native Germans and for the vast majority of cases also compared to ethnic Germans. This phenomenon is especially pronounced for the second generation of (foreign) immigrants and manifests itself in

German vs. foreign

German vs. ethnic German

Ethnic German vs. foreign

First vs. German

First vs. ethnic German

þ þ þ þ þ

0

þ þ þ þ 0

0

0

0

0

0

0

0

0 0

First vs. second

Notes: A ‘‘þ’’ denotes a statistically signiﬁcant positive, a ‘‘’’ a statistically signiﬁcant negative, and ‘‘0’’ an insigniﬁcant difference between estimated group indicators. For a description of the utilized control variables see Table A1 in the appendix.

þ

Second vs. ethnic German

þ

Second vs. German

Results of (ordered) probit estimations for attitudes

The following statements express varying attitudes towards life and the future. Do you agree/disagree? I decide the way my life is run þ 0 þ 0 I have little control over the things 0 þ þ that take place in my life One has to work hard to achieve þ 0 þ 0 success What one achieves in life is mainly a 0 þ þ question of luck or fate I often make the discovery that 0 0 þ 0 others inﬂuence my life If I ever hit upon difﬁculties in my þ 0 þ life, I doubt my capabilities In comparison with others, I haven’t 0 þ þ achieved what I deserve The possibilities in my life are 0 0 0 þ determined by the social conditions If one is socially or politically active, 0 0 0 0 0 one can inﬂuence the social conditions

Question

Table 2B.

388 Michael Fertig

The Societal Integration of Immigrants in Germany

389

their signiﬁcantly higher probability to perceive their life as less selfdetermined and their prospects in life as determined by faith or luck rather than their own endeavors. The differences in agreement between groups are not only statistically signiﬁcant, but also quantitatively substantial. For instance, the probability to agree to the statement ‘‘I have little control over the things that take place in my life’’ is on average 11.5 percentage points higher for foreigners than for native Germans and 8.5 percentage points higher for foreigners than for ethnic Germans. Furthermore, second-generation immigrants tend to exhibit a 14.5 and 18 percentage points higher agreement propensity than native German and ethnic Germans, respectively. For comparison, being female is associated with a 2.5 percentage point and being unemployed with an 8 percentage point higher agreement probability. The sixth question can be interpreted as an indicator for the extent to which respondents doubt their own abilities. Estimation results indicate that Germans born in Germany as well as ethnic Germans tend to be less afﬂicted by self-doubts than non-citizens. Again, this result is primarily driven by the difference between citizens and the second generation of immigrants. The next question refers to the degree of respondents’ satisfaction with their life and what they have achieved so far, whereas the last two questions indicate the degree of skepticism or pessimism with which respondents view the level of self-determination of their own life and their inﬂuence on the political and social environment they are living. With the exception of the last question, which exhibits no signiﬁcant differences whatsoever, estimation results for these attitudes conﬁrm the results of the preceding questions. Second-generation immigrants are less satisﬁed with their life compared to German citizens and display a signiﬁcantly higher probability to doubt that their life is self-determined than ethnic Germans. Again, the deviations between population groups are quantitatively substantial. For instance, agreement to the statement ‘‘If I ever hit upon difﬁculties in my life, I doubt my capabilities’’ is 4.5 and 8.5 percentage points higher for foreigners than for Germans born in Germany and ethnic Germans, respectively. The difference between second generation foreigners and native Germans amounts to 8.5 percentage points, whereas agreement is around 16 percentage points higher for the second generation than for ethnic Germans (for comparison: females exhibit a ten percentage point higher and unemployed an 8 percentage point higher agreement propensity). Altogether, estimation results indicate that even after controlling for a large set of socio-demographic characteristics and in stark contrast to their similarity in behavior, the second-generation of immigrants is a deeply unsettled population group that is plagued by self-doubts and a rather fatalistic and pessimistic view on their life and its prospects. Finally, estimation results for the questions to immigrants/foreigners only are summarized in Table 3. These results indicate that typical members of the second generation of (foreign) immigrants tend to have

390

Michael Fertig

Table 3.

(Ordered) probit results for questions to noncitizens only

Variable

First vs. second

First vs. ethnic German

Second vs. ethnic German

Contact to Germans Visits to Germans Visits from Germans Spoken German Written German Write language of origin country Speak language of origin country Mainly use German Mainly use language of origin country Use both equally Disadvantage Wish to stay Feel German Feel connected with origin country

0 0 0 þ 0 0 þ

N.A. N.A. N.A. þ þ þ 0 þ þ

N.A. N.A. N.A. 0 0 þ 0 0 þ

Note: For a description of the utilized control variables see Table A1 in the appendix.

more contact with Germans (including visits from Germans) than their parents generation. Furthermore, self-assessed ﬂuency in (written and spoken) German is higher for this group than for their parents generation. However, ethnic Germans display the highest self-assessed ﬂuency compared to all other immigrant groups. By contrast, ﬁrst-generation immigrants tend to assess a higher ﬂuency in the language of their origin country compared to ethnic Germans, whereas there is no difference in assessment between both immigrant generations. In line with these ﬁndings, ethnic Germans are more likely to use German as the main language at home, whereas ﬁrst-generation immigrants tend toward the language of their origin country and the second generation is, again, in between. Moreover, members of the ﬁrst-generation immigrant group reported a signiﬁcantly higher experience of disadvantage due to their origin than ethnic Germans, whereas estimation results reveal no difference between the second-generation and ethnic Germans. Unsurprisingly, ethnic Germans display the highest willingness to stay permanently in Germany and to feel German, whereas the ﬁrst generation exhibits the lowest likelihood. Second-generation immigrants are again in between both other groups. By contrast, ﬁrst-generation immigrants are the group with the highest feeling of connection to their origin country and ethnic Germans display the lowest association with the country they emigrated from. In general, this last set of estimation results reveals no surprising ﬁndings. In the context of societal integration of immigrant minorities, language ﬂuency and the feeling of connection to the country of residence as well as the origin country are the most interesting pieces of information.

The Societal Integration of Immigrants in Germany

391

With respect to these indicators, ethnic Germans tend to display the highest degree of integration into the German society since their command of the German language and their connection to Germany as their country of permanent residence are higher compared to both noncitizen immigrant groups. However, this is not to say that the language ﬂuency of ethnic Germans is high in absolute terms. Furthermore, self-assessed measures are always susceptible for misclassiﬁcation errors.5

5. Conclusions Over the past three decades, the ethnic composition of immigration to Germany has changed, and the geographic and cultural gaps between Germany and the sending countries have widened. Germany now has a sizeable community of second-generation immigrants whose social and economic characteristics are a matter of growing concern. Yet, despite the growing recognition of this situation, relatively little research has targeted the question of migrants’ integration into society. Furthermore, even less is known about the integration of the so-called second-generation immigrants. Hence, this chapter contributes to our understanding of these processes by providing an analysis of the extent to which immigrants in Germany are integrated into the German society. Speciﬁcally, we utilize a large set of qualitative information and subjective data collected in the 1999 wave of the GSOEP and compare native Germans, ethnic Germans and foreign immigrants of different generations along various dimensions. We investigate whether there are differences between these groups regarding their leisure-time activities and their attitudes toward speciﬁc areas of life. Finally, we analyze various indicators of the societal integration of immigrant groups that are collected for these groups only, like their German language ability or their contacts to natives. In this endeavor, we control for a large set of observable characteristics of individual respondents to account for heterogeneity in individual activities and attitudes. The empirical results suggest that conditional on observable characteristics the activities and attitudes of foreign immigrants from both generations differ much more from those of native Germans than the activities/attitudes of ethnic Germans. These results indicate that citizenship status plays an important role for societal integration of immigrants. Clearly, the cultural background between ethnic Germans on the one and ﬁrst as well as second generation foreigners on the other hand might differ substantially and can exert considerable inﬂuence on their perceptions on life. However, it seems more than plausible that 5 Dustmann and van Soest (2001) demonstrate in the context of language ﬂuency as a determinant of earnings that misclassiﬁcation of self-assessed language command might be a severe problem.

392

Michael Fertig

individuals holding citizenship of their country of birth or permanent residence feel much more welcome their and, thus, perceive much more control of their life. Finally and most importantly, our estimation results for the questions regarding different views on life indicate that even after controlling for a large set of socio-demographic characteristics the second-generation of immigrants is a deeply unsettled population group which is plagued by self-doubts and a rather fatalistic and pessimistic view on their life and its prospects. This ﬁnding stands in stark contrast to the observed similarity in leisure-time activities of this population group compared to native Germans. Since the typical respondent from the second-generation immigrant group is rather young, their pessimistic perception of life and its prospects should be alarming. Whether and to what extent this is the cause or the consequence of their performance on the German labor market is a currently unresolved issue which has to be addressed in future research. In any case, by ignoring the rather gloomy orientation of this immigrant generation, we are running the risk of losing a sizeable fraction of young people as content and productive members of our future society. Acknowledgments The author is grateful to Thomas K. Bauer and Christoph M. Schmidt for helpful comments. Appendix

Table A1. Variable

German First (generation) Second (generation) Foreign

Secondary schooling Intermediary schooling Technical schooling Upper secondary school. Other schooling No schooling degreea Singlea

Description of explanatory variables Description Immigrant group indicators 1 if respondent has German citizenship and is born in Germany; 0 otherwise 1 if respondent does not have German citizenship and is not born in Germany; 0 otherwise 1 if respondent does not have German citizenship but is born in Germany; 0 otherwise 1 if respondent belongs to ﬁrst generation or second generation of immigrants; 0 otherwise Education category indicators 1 if respondent has secondary schooling degree; 0 otherwise 1 if respondent has intermediary schooling degree; 0 otherwise 1 if respondent has technical schooling degree; 0 otherwise 1 if respondent has upper secondary schooling degree; 0 otherwise 1 if respondent has other schooling degree; 0 otherwise 1 if respondent has no schooling degree; 0 otherwise Marital status indicators 1 if respondent is single; 0 otherwise

The Societal Integration of Immigrants in Germany

393

Table A1. (Continued ) Variable

Description

Married

1 if respondent is married or lives with permanent partner; 0 otherwise 1 if respondent is divorced; 0 otherwise 1 if respondent is widowed; 0 otherwise Other control variables 1 if respondent is female; 0 otherwise (Squared) Age of respondent in years 1 if respondent is registered as unemployed; 0 otherwise 1 if respondent is currently in training (school, university etc.); 0 otherwise 1 if children under 15 live in respondent’s household; 0 otherwise 1 if respondent lives in eastern Germany; 0 otherwise Number of years, the respondent lives in Germany

Divorced Widowed Female (Squared) Age Unemployed In training Children under 15 East Time spent in Germany a

Denotes the reference category within the respective indicator groups.

Table A2.

Description of questions on leisure-time activities

Which of the following activities do you participate in during your free-time? Visits to cultural events, e.g., concerts, theatre, presentations Answer possibilities: Cinema visits, visits to pop concerts, dance events, discos, sporting 1 ¼ never, 2 ¼ rarely, events 3 ¼ every month, Active sport 4 ¼ every week Social intercourse with friends, relatives or neighbors Lend help to friends, relatives, or neighbors when something has to be done Honorary ofﬁce participation in clubs, associations or social services Participation in public initiatives, in political parties, local government Church-going, visits to religious events

Table A3.

Description of questions on attitudes

Which of the following areas are important for your well-being and satisfaction? Work Original answer possibilities: very Family important, important, not very Friends important, totally unimportant. Income These are summarized into: 1 ¼ very Housing important and important; 0 Inﬂuence on political decisions otherwise Career success Free-time Health Protection of the natural environment Faith, religion Residential area Mobility to get everywhere quickly

394

Michael Fertig

Table A3. (Continued ) The following statements express varying attitudes toward life and the future I decide the way my life is run Original answer possibilities: totally In comparison with others, I haven’t achieved what I deserve agree, agree slightly, disagree slightly, What one achieves in life is mainly a question of luck or fate totally disagree. These are If one is socially or politically active, one can inﬂuence the summarized into: 1 ¼ totally agree social conditions and agree slightly; 0 otherwise I often make the discovery that others inﬂuence my life One has to work hard to achieve success If I ever hit upon difﬁculties in my life, I doubt my capabilities The possibilities in my life are determined by the social conditions I have little control over the things that take place in my life If you think about the future in general, are you optimistic? Original answer possibilities: optimistic, more optimistic than pessimistic, more pessimistic than optimistic, pessimistic. These are summarized into: 1 ¼ optimistic and more optimistic than pessimistic; 0 otherwise To what extent do you feel connected with the place and the area that you live in? Original answer possibilities: very strong, strong, not very strong, not at all. These are summarized into: 1 ¼ very strong and strong; 0 otherwise Would you consider moving away, e.g., because of family or job? Answer possibilities: 1 ¼ yes; 2 ¼ possibly, cannot exclude the possibility; 3 ¼ no

Table A4. Variable

Contact to Germans Visits to Germans Visits from Germans

Spoken German Written German Write language of origin country Speak language of origin country Mainly use German Mainly use language of origin country Use both equally Disadvantage Wish to stay Feel German Feel connected with origin country

Description of questions to foreigners only Description Contact to Germans Since you have lived in Germany, have you had close contact to Germans? 1 ¼ yes; 0 otherwise In the last 12 months did you visit any Germans in their home? 1 ¼ yes; 0 otherwise In the last 12 months were you visited by any Germans in your home? 1 ¼ yes; 0 otherwise Language ability In your opinion, how well can you speak and write German/the language of your origin country? Answer possibilities: 1 ¼ not at all; 2 ¼ poorly; 3 ¼ fairly; 4 ¼ good; 5 ¼ very well

Language use What language do you speak here in Germany for the most part? Answer possibilities: Mostly German; the language of your origin country; both about equally as much

Perception of acceptance in German society How often have you experienced disadvantages in the last two years because of your origins? 1 ¼ never; 2 ¼ seldom; 3 ¼ often Do you want to stay in Germany forever? 1 ¼ yes; 0 otherwise To what degree do you think of yourself as German? 1 ¼ not at all; 2 ¼ barely; 3 ¼ in some respect; 4 ¼ mostly; 5 ¼ completely To what extent do you feel connected with the country of your or your family’s origin? 1 ¼ not at all; 2 ¼ barely; 3 ¼ in some respect; 4 ¼ mostly; 5 ¼ completely

Mean

Standard deviation

Germans

0.631 0.833 1.085 0.762 0.854 0.601 0.276 1.167 0.327 0.049 0.312 0.154 0.187 0.444 0.463 0.348 0.049 0.306 0.492 0.315 0.332

0.879 0.998 0.891 0.976 0.964 0.268 0.691 0.859 0.998 0.896 0.592 0.889 0.874

Standard deviation

1.583 1.798 1.741 3.315 2.495 1.213 1.060 2.367

Mean

Ethnic Germans

0.785 0.984 0.917 0.958 0.956 0.219 0.611 0.883 0.997 0.839 0.650 0.855 0.838

1.454 1.616 1.588 3.389 2.449 1.242 1.057 2.065

Mean

0.411 0.125 0.276 0.201 0.205 0.414 0.488 0.321 0.058 0.368 0.477 0.353 0.368

0.656 0.798 1.038 0.775 0.935 0.658 0.293 1.069

Standard deviation

First generation

Summary statistics – questions on leisure-time activity and attitudes

Which of the following activities do you participate in during your free-time? Visits to cultural events. 1.845 0.687 Cinema visits, visits to pop concerts, discos, etc. 2.056 0.889 Active sport 2.106 1.240 Social intercourse with friends, relatives, or neighbors 3.116 0.824 Lend help to friends, relatives, or neighbors 2.382 0.803 Honorary ofﬁce participation in clubs, etc. 1.547 0.950 Participation in public initiatives, etc. 1.143 0.468 Church-going, visits to religious events 1.700 0.922 Which of the following areas are important for your well-being and satisfaction? Work 0.836 0.370 Family 0.981 0.137 Friends 0.892 0.310 Income 0.960 0.196 Housing 0.971 0.168 Inﬂuence on political decisions 0.307 0.461 Career success 0.718 0.450 Free-time 0.903 0.296 Health 0.994 0.080 Protection of the natural environment 0.881 0.323 Faith, religion 0.328 0.469 Residential area 0.896 0.306 Mobility to get everywhere quickly 0.878 0.327

Questions

Table A5.

0.844 0.977 0.953 0.941 0.924 0.272 0.785 0.939 0.988 0.836 0.539 0.825 0.904

1.750 2.750 2.548 3.574 2.456 1.304 1.102 1.823

Mean

0.364 0.151 0.211 0.235 0.265 0.446 0.412 0.240 0.107 0.371 0.499 0.380 0.295

0.697 0.927 1.225 0.705 0.924 0.715 0.380 0.926

Standard deviation

Second generation

The Societal Integration of Immigrants in Germany 395

Mean

If you think about the future in general, are you optimistic? To what extent do you feel connected with the place and the area that you live in? Would you consider moving away, e.g., because of family or job?

0.425 0.411 0.793

0.764 0.784 2.215

2.198

0.870 0.666

0.307 0.954 0.248 0.620 0.849 0.180

0.817

0.336 0.472

0.462 0.210 0.432 0.486 0.359 0.385

2.110

0.719 0.597

0.365 0.938 0.329 0.702 0.785 0.323

Mean

0.427 0.193 0.433 0.476 0.419 0.364

Standard deviation

0.856

0.450 0.491

0.482 0.241 0.470 0.458 0.411 0.468

0.411 0.500 0.481 0.488

Standard deviation

First-generation

0.784 0.476 0.638 0.392

Mean

Ethnic Germans

Do you agree/disagree? 0.308 0.862 0.345 0.456 0.336 0.473 0.480 0.511 0.500 0.489 0.392 0.489

Standard deviation

Germans

The following statements express varying attitudes towards life and the future. I decide the way my life is run 0.894 In comparison with others, I haven’t achieved what I deserve 0.295 What one achieves in life is mainly a question of luck or fate 0.360 If one is socially or politically active, one can inﬂuence the social 0.396 conditions I often make the discovery that others inﬂuence my life 0.239 One has to work hard to achieve success 0.961 If I ever hit upon difﬁculties in my life, I doubt my capabilities 0.250 The possibilities in my life are determined by the social conditions 0.652 More important than any endeavors, are your own capabilities 0.772 I have little control over the things that take place in my life 0.158

Question

(Continued)

Summary statistics – questions on attitudes

Table A5.

2.078

0.853 0.715

0.395 0.928 0.380 0.673 0.828 0.291

0.875 0.424 0.557 0.440

Mean

0.857

0.354 0.452

0.490 0.260 0.486 0.470 0.378 0.455

0.331 0.495 0.497 0.497

Standard deviation

Second-generation

396 Michael Fertig

Contact to Germans Visits to Germans Visits from Germans Spoken German Written German Write language of origin country Speak language of origin country Mainly use German Mainly use language of origin country Use both equally Disadvantage Wish to stay Feel German Feel connected with origin country

Question

N.A. N.A. N.A. 4.027 3.718 4.416 4.146 0.583 0.350 0.068 1.499 0.947 4.298 2.430

Mean

N.A. N.A. N.A. 0.742 0.933 0.736 1.011 0.494 0.477 0.252 0.556 0.225 0.917 1.111

Standard deviation

Ethnic Germans

0.895 0.769 0.814 3.546 2.837 4.476 4.115 0.257 0.392 0.351 1.526 0.619 2.441 3.879

Mean

0.307 0.422 0.389 1.012 1.295 0.618 0.969 0.437 0.488 0.478 0.599 0.486 1.108 0.938

Standard deviation

First-generation

Summary statistics – questions to immigrants/foreigners only

0.979 0.938 0.947 4.570 4.386 4.074 3.547 0.507 0.402 0.091 1.469 0.795 3.079 3.291

Mean

0.142 0.241 0.225 0.676 0.868 0.833 1.054 0.501 0.491 0.288 0.586 0.404 1.082 0.990

Standard deviation

Second-generation

The Societal Integration of Immigrants in Germany 397

a

0.416 0.336 0.034 0.159 0.015 0.018 0.250 0.602 0.081 0.067 0.524 45.329 0.066 0.111 0.329 0.317 45.329

Mean

0.493 0.473 0.182 0.366 0.121 0.131 0.433 0.490 0.273 0.250 0.499 17.023 0.248 0.314 0.470 0.465 17.023

Standard deviation

Germans

0.130 0.110 0.012 0.039 0.571 0.100 0.184 0.742 0.041 0.034 0.487 40.364 0.111 0.135 0.564 0.002 8.438

Mean

0.337 0.314 0.110 0.194 0.496 0.301 0.388 0.438 0.198 0.181 0.500 15.435 0.315 0.342 0.497 0.049 1.889

Standard deviation

Ethnic Germans

Denotes the reference category within the respective indicator groups.

Secondary schooling Intermediary schooling Technical schooling Upper secondary school. Other schooling No schooling degreea Singlea Married Divorced Widowed Female Age Unemployed In training Children under 15 East Time spent in Germany

Variable

(Continued)

0.186 0.026 0.017 0.044 0.445 0.277 0.092 0.816 0.064 0.028 0.484 44.679 0.106 0.028 0.465 0.006 23.291

0.389 0.159 0.131 0.206 0.497 0.448 0.289 0.388 0.246 0.164 0.500 13.801 0.308 0.166 0.499 0.076 9.680

Standard deviation

First-generation Mean

Summary statistics – explanatory variables

Table A5.

0.391 0.183 0.046 0.122 0.058 0.104 0.614 0.330 0.049 0.006 0.487 26.023 0.070 0.348 0.429 0.003 26.023

Mean

0.489 0.388 0.210 0.328 0.234 0.306 0.487 0.471 0.217 0.076 0.501 8.857 0.255 0.477 0.496 0.054 8.857

Standard deviation

Second-generation

398 Michael Fertig

The Societal Integration of Immigrants in Germany

399

References Bauer, T.K., Zimmermann, K.F. (1999), Assessment of possible migration pressure and its labour market impact following EU enlargement to central and Eastern Europe. IZA Research Report No. 3, IZA-Bonn. Borjas, G.J. (1985), Assimilation, changes in cohort quality, and the earnings of immigrants. Journal of Labour Economics 3, 463–489. Borjas, G.J. (1987), Self-selection and the earnings of immigrants. American Economic Review 77, 531–553. Chiswick, B.R. (1978), The effect of Americanization on the earnings of foreign-born men. Journal of Political Economy 86, 897–921. Dietz, B. (2003), Post-soviet youth in Germany: Group formation, values and attitudes of a new immigrant generation. In: Horowitz, T., Kotik-Friedgut, B., Hoffmann, St. (Eds.), From Pacesetters to Dropouts. Post-Soviet Youth in Comparative Perspective. University Press of America, New York/Oxford, pp. 253–271. Dustmann, C. (1993), Earnings adjustments of temporary migrants. Journal of Population Economics 6, 153–168. Dustmann, C. (1996), The social assimilation of migrants. Journal of Population Economics 9, 79–103. Dustmann, C., Schmidt, C.M. (2000), The wage performance of immigrant women: full-time jobs, part-time jobs, and the role of selection. IZA Discussion Paper No. 233, IZA-Bonn. Dustmann, C., van Soest, A. (2001), Language ﬂuency and earnings: estimation with misclassiﬁed language indicators. Review of Economics and Statistics 83, 663–674. Fertig, M. (2001), The economic impact of EU-enlargement: assessing the migration potential. Empirical Economics 26, 707–720. Fertig, M., Schmidt, C.M. (2001a), Aggregate level migration studies as a tool for forecasting future migration streams. In: Djajic, S. (Ed.), International Migration: Trends, Policy and Economic Impact. Routledge, London, pp. 110–136. Fertig, M., Schmidt, C.M. (2001b), First- and second-generation immigrants: What do we know and what do people think? In: Rotte, R. (Ed.), Migration Policy and the Economy: International Experiences. Ars & Unitas, Neuried, pp. 179–218. Fertig, M., Schmidt, C.M. (2002), Mobility within Europe – the attitudes of European youngsters. RWI Discussion Paper No. 1, RWI-Essen. Kurthen, H., Fijalkowski, J., Wagner, G.G. (Eds.) (1998), Immigration, Citizenship, and the Welfare State in Germany and the United States: Immigrant Incorporation. Industrial Development and the Social fabric: An International Series of Historical Monographs (Vol. 14, Pt. A). JAI Press, Stamford/London.

400

Michael Fertig

Riphahn, R.T. (2003), Cohort effects in the educational attainment of second generation immigrants in Germany: an analysis of census data. Journal of Population Economics 16, 711–737. Schmidt, C.M. (1997), Immigrant performance in Germany. Labor earnings of ethnic German migrants and foreign guest-workers. Quarterly Review of Economics and Finance 37, 379–397.

CHAPTER 17

Who Matters Most? The Effect of Parent’s Schooling on Children’s Schooling Ira N. Gang Department of Economics, Rutgers University, New Brunswick, NJ 08901-1248, USA Institute for the Study of Labor (IZA), Bonn, Germany CReAM-Center for Research and Analysis of Migration, London, UK E-mail address: [email protected]

Abstract This chapter examines the differential effects of mother’s schooling and father’s schooling on the acquisition of schooling by their offspring. It does this in a ‘‘cross-cultural’’ context by comparing results across three countries: Germany, Hungary, and the Former Soviet Union. It looks within these countries, by gender, at different ethnic subgroups. Evidence is found, generally, that father’s schooling is more important than mother’s, but this does vary by ethnic group. Mother’s schooling plays a relatively larger role for females.

1. Introduction This chapter looks at the effect of parents’ schooling on the schooling attainment of their children. We examine the differential effects of mother’s schooling and father’s schooling on the acquisition of schooling by their offspring. Schooling is examined in a ‘‘cross-cultural’’ context by comparing results across countries, as well as within a country by looking at these effects by ethnic group and gender. The study makes use of three household level data sets: the German Socio-Economic Panel (GSOEP), the Hungarian Household Panel Survey (HHPS), and the Soviet Interview Project (SIP). We analyze each during times of economic change for the participants in each country. Each of these data sets contains information on various subgroups of the population: the GSOEP consists of Germans and immigrants into Germany and their families, the HHPS makes it possible to distinguish Gypsies and Hungarian, non-Gypsies, and the SIP allows us to identify by Republic of the former Soviet Union (FSU) each persons’ place of birth, and for Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008023

r 2010 by Emerald Group Publishing Limited. All rights reserved

402

Ira N. Gang

Russians, whether the person was Jewish or not. We use these data to analyze demographically comparable groups; in particular we examine the schooling attainment of those born after World War II (approximately) and relate it to the schooling attainment of their parents. Of course, the idiosyncrasies of each data set do not allow perfectly comparable samples. Why should parent’s schooling attainment matter in determining children’s schooling attainment? Parental schooling may be a proxy for a host of unobservable determinants, such as parental preferences for education, children’s ability, and assistance given by parents in school work (Gertler and Glewwe, 1992). If parent’s education matters, it is natural to ask which parent’s education matter’s more? The conventional wisdom is that the mother’s education is more important than the father’s education in children’s attainments, including schooling. This arises from a large number of studies in developing economies and in the United States (e.g., Schultz, 1984; Chiswick, 1988; Arai, 1989; Thomas, 1994; Gertler and Glewwe, 1992). Haveman and Wolfe (1995) examined the large literature on the determinants of children’s attainments in the United States, and conclude that the (p. 1855) ‘‘human capital of the mother is usually more closely related to the attainment of the child than is that of the father.’’1 Why is mother’s schooling is more important than father’s? One explanation rests on the time allocation model.2 Time spent in child care and time spent in the labor market both contribute to high quality children, for example, children’s schooling attainment. This raises the question of the role of nonmarket versus market inputs in children’s educational attainments. If we assume that nonmarket inputs are more important, then the parent who engages in relatively greater non-market activity will exert a greater inﬂuence on children’s schooling. Alternatively, if the contribution through market work is more important in determining children’s education than is the input through non-market work, it is the spouse who is relatively more engaged in market activity who will have the greater inﬂuence. The above argument implies, for example, that if women spend relatively more time than men at home versus in the labor market, their inﬂuence will be greater. However, we might expect to see variations in this inﬂuence (1) across countries and subgroups that face different relative prices of market versus nonmarket activity and (2) across countries and

1

The evidence is not unambiguous; Haveman and Wolfe are making a judgment based on the preponderance of evidence. This is true in the development literature as well. For example, Tansel (1997) ﬁnds for Ghana and the Cote d’Ivoire that father’s literacy was more important than mother’s in children’s schooling attainments. See also, Devereux and Salvanes (2005), Rey and Racionero, (2002), Riphahn (2003), and Deutsch et al. (2006). 2 Alternative, complementary models are the socialization/role model perspective, the lifespan development approach, and stress theory and coping strategies. See Haveman and Wolfe (1995) for a discussion of these models and how they relate to the time allocation model.

The Effect of Parent’s Schooling on Children’s Schooling

403

subgroups that have differing elasticities of child-rearing activity with respect to labor force activity. These two elements might lead us to expect a different effect of mother’s versus father’s schooling on children’s schooling by country, by subgroups within a country, and by gender. What is the evidence on country and subgroup effects? Work by Gang and Zimmermann (2000) ﬁnds that in comparing the schooling attainments of Germans and comparable second generation immigrants (the children of the guestworkers), both father’s and mother’s education was important to Germans, with father’s being more important, while for most of the second generation immigrant subgroups neither parent’s educational background substantially inﬂuenced children’s schooling. These results do, however, vary by origin country. Schultz (1984) ﬁnds that in the United States native-born parent’s schooling affects their offspring educational attainments, with the effect of mother’s schooling about twice that of father’s. Among immigrants, the mother’s effect was less and the father’s was more, but generally the relationship was weaker than for the native-born. Here too the results varied by origin country.

2. Data This study uses data from the GSOEP, the HHPS, and the SIP. The GSOEP is described in Wagner et al. (1993). The ﬁrst wave of the GSOEP was drawn in 1984, from which most of our information is taken. In 1986 a question was asked on parent’s education, and that was matched to the 1984 respondents. Our focus here is on the children of the guestworkers, immigrants from Turkey, Yugoslavia, Greece, Italy and Spain who arrived in Germany from the early 1960s until the program was stopped in the early 1970s. From the sample guestworker children we keep those who were born in Germany or who arrived before the age of 16, and who in 1984 were 17–38 years old. These are considered to be the second generation migrants (Kossoudji, 1989, p. 497). From native German households, we examine the same age cohort. The subsample used here is described in greater detail in Gang and Zimmermann (2000) and Gang (1997). The ﬁrst wave of the HHPS was drawn in 1992 and is described in Sik (1995). In 1993 a question on parents’ education was asked, and we draw our data from the 1993 wave. The interviewers were asked whether they thought the respondent was a Gipsy or not, and we use this to identify Gypsies versus those Hungarians who are not Gypsies. We restricted our sample to those who were 17 to 47 years old in 1993. This leaves us with a sample of 2031 individuals. The SIP data provide us with a contemporary sample of e´migre´s who moved from the Soviet Union to the United States in the period from January 1, 1979, to April 30, 1982, and provides us with detailed

404

Ira N. Gang

background information on those who emigrated.3 The SIP data were collected in 1983 and reports on various aspects of household behavior of respondents during their lifetime in the Soviet Union through the end of their last normal period (LNP), the date on which they declared their intention to emigrate from the Soviet Union.4 Our study is based on a subsample of 919 (the Blue supplement) for whom both basic and extended household characteristics are known, and who were born in the various republics of the FSU.5 Of these, we further restricted the sample to the 519 participants who were between 25 and 50 years old in 1983. Each data set has its own deﬁnition of each variable; we tried to make the variables comparable. The critical variables of the study are children’s and parent’s schooling. For children’s schooling, we translated the different degrees into years of schooling for all three data sets. GSOEP provides data for each individual on the type of school attended. To convert these into years of schooling, we followed the procedure outlined in Gang and Zimmermann (2000). Instead of just adding the standard years for the various educational degrees, we use a more conservative measure that adjusts for ‘‘duplicate’’ degrees and discounts alternative post-schooling degrees (vocational training, university, and the like) by one year. A similar procedure was employed in translating the degrees in the HHPS and the SIP into years of schooling. For parent’s schooling, using the HHPS and the SIP, we also directly translated degrees into years of schooling. Although there were signiﬁcant changes in the structure of schooling in the Soviet Union and Hungary, in balance parents went through the same general type of school system as the children (see Dobson, 1984). However, it is difﬁcult to compare schooling in Germany with the schooling levels acquired by migrants in their home countries. For the non-Germans in the GSOEP, the parent’s human capital dummy variable takes value 1, if its education is at least ‘‘mandatory with degree.’’ In the case of a German, the dummy takes value 1, if the parent has at least a high school degree (Realschule or Abitur). Below we perform separate analyses for each subgroup of the population. For the GSOEP, this means we analyze Germans, Turks, Italians, Spaniards, Greeks and Yugoslavs; for the HHPS, Gypsies and 3

The initial SIP consisted of 2,793 respondents aged 21–70 years at time of emigration. A detailed discussion of the original materials can be found in Millar et al. (1987). The ﬁrst major studies from the data base including commentary on the data can be found in Millar (ed.) (1987) while recent analysis of the household can be found in Ofer and Vinokur (1992), Linz (1995), and Gang and Stuart (1996, 1997). 4 The concept of LNP is important. It was assumed that once a family declared its intention to emigrate its circumstances would change, possibly dramatically, due to ofﬁcial hostility. 5 How representative is our sample? The ethnic composition of the original SIP database is known such that weights could be derived to make any sample representative of the entire Soviet population. We did not weight our observations to obtain a representative picture of the entire Soviet population. This does not affect the slope coefﬁcients in our regressions.

The Effect of Parent’s Schooling on Children’s Schooling

405

Hungarian, non-Gypsies; for the SIP, Russians (by place of birth), Ukrainians, people born in the Baltics, Belarus and Moldava, Caucasus and Central Asia. We further break down the analysis into males and females, and for Russians into Jews and non-Jews. This allows us a complete set of interactions of ethnicity and gender with our explanatory variables, rather than forcing everything into the intercept term. The gain is a better picture of the effect of parent’s schooling on children’s schooling by ethnicity and gender. The cost is that some of analyses are performed on very few observations. This suggests that we should be generous in interpreting our results in terms of statistical signiﬁcance, looking instead for general patterns. What is perhaps surprising is that even with small samples we get reasonably strong results.

3. Empirical Results The analysis is performed using OLS with children’s schooling on the left hand side, and mother’s and father’s schooling and other control variables on the right hand side. The exact speciﬁcation varies by data set, and is given in the notes to Tables 1–3. Tables 1–3 have the same structure, with Table 1 from the analysis of the GSOEP, Table 2 from the analysis of the HHPS, and Table 3 from the analysis of the SIP. The tables summarize the results of the analysis, giving the means of children’s, mother’s and father’s schooling and the estimated elasticities of children’s education with respect to mother’s schooling and father’s schooling. The elasticity estimates are shaded if signiﬁcant at the .10 level for ease of presentation. In addition, the tables present the results of two tests (1) whether parent’s education matters at all in children’s schooling and (2) whether mother’s and father’s effects on the child are signiﬁcantly different from each other. Recall that for some subgroups the sample size is very small, and this should be taken into account in interpreting the results. Let us examine the results using the GSOEP in Table 1. First notice that, with regard to years of schooling in Germany, German children have more years than do non-German children. For Germans, the mean is 12.1 years, while for non-Germans it ranges from 7.6 for Turks to 9.3 for Spaniards. Recall that these numbers are for the same post-World War II cohort. As for parents schooling, the numbers mean something different for Germans and non-Germans. For Germans, recall that our measure of German’s parents education is whether or not parents have at least a high school education. This level of education was achieved by 12 percent of the parents. For non-Germans they mean having completed the basic mandatory degree in their country of origin; we ﬁnd this ranges from 9 percent of the Turks to 21 percent of the Yugoslavs. Note that what is ‘‘mandatory’’ varies from country to country.

1920

Female

161

132

73

30

43

Male

Female

Yugoslavs All

Male

Female

293

1920

Male

Turks All

3840

Germans All

Sample size

7.9 (4.8) 8.9 (4.1) 7.2 (5.2)

7.6 (4.2) 8.3 (3.8) 6.7 (4.4)

12.1 (2.5) 12.3 (2.5) 11.9 (2.4)

Children’s education (years of schooling)

Table 1.

.16 (.37) .10 (.31) .21 (.41)

.04 (.19) .04 (.19) .05 (.22)

.08 (.28) .08 (.28) .09 (.28)

Mother’s education

Means

.21 (.41) .13 (.35) .26 (.44)

.09 (.29) .06 (.24) .13 (.33)

.12 (.32) .11 (.32) .13 (.33)

Father’s education

.218* (.034) .121* (.029) .342* (.049)

.017 (.011) .021*** (.012) .020 (.017)

.006* (.001) .005* (.002) .007* (.002)

Children’s education with respect to mother’s education

.173* (.032) .093* (.032) .281* (.042)

.006 (.014) .022* (.009) .008 (.028)

.014* (.001) .015* (.002) .013* (.002)

Yes Yes Yes

Yes Yes

No

Yes

Yes

No

Yes

No

.40

.10

.33

.14

.05

.13

.23

Yes**

Yes

No

.35

.29

Adjusted R-squared

Yes

Yes

Mother’s and father’s education signiﬁcantly different from one another

Yes

Yes

Mother’s and father’s Children’s education education with jointly respect to father’s signiﬁcant education

Estimated elasticities

Inﬂuence of Parent’s Schooling on Children’s for Those Living in Germany

406 Ira N. Gang

68

113

68

45

Female

Spaniards All

Male

Female

9.3 (3.6) 9.3 (3.6) 9.1 (3.8)

8.1 (4.5) 8.2 (4.6) 8.1 (4.3)

8.9 (4.4) 9.3 (4.3) 8.3 (4.5)

.12 (.33) .18 (.38) .04 (.21)

.10 (.30) .08 (.27) .13 (.34)

.09 (.28) .08 (.27) .09 (.29)

.17 (.38) .21 (.41) .11 (.32)

.13 (.34) .09 (.28) .19 (.39)

.10 (.31) .06 (.25) .15 (.36)

.006 (.019) .024 (.029) .001 (.006)

.003 (.026) .044* (.016) .088** (.051)

.005 (.023) .010 (.025) .016 (.047)

.036 (.022) .087* (.032) .005 (.018)

.002 (.031) .001 (.021) .053 (.070)

.031 (.022) .031** (.018) .007 (.054)

No

Yes

No

Yes

No

No

Yes

Yes

No

No

No

No

No

Yes

No

No

No

No

.25

.11

.17

.15

.31

.18

.05

.26

.13

Notes: The OLS regressions have children’s schooling in years on the left-hand side. The children are between 17 and 38 years old in 1984. On the right-hand side is mother’s schooling, father’s schooling, child’s age and age-squared, a dummy variable taking on the value 1 if the child is still in school, and, where appropriate, a dummy variable taking on the value of 1 if the child is male. Note that the measure of mother’s and father’s schooling differs for Germans and non-Germans in the analyses done with the GSOEP. For Germans, the variables take on the value of 1 if at least Realschule or Abitur has been earned, 0 if not. For the non-Germans, the variables take on the value of 1 if at least the mandatory degree in the home country has been earned, 0 if not. Source: Author’s calculations from German Socio-Economic Panel (Wagner et al., 1993). Yes** indicates a p-value between .05 and .10; otherwise yes indicates a p-value below .05. * Indicates a p-value below .05. ** Indicates a p-value between .05 and .10. For ease of reading, elasticities are shaded if p-value is below .10.

91

Male

54

Female

159

62

Male

Italians All

116

Greeks All

The Effect of Parent’s Schooling on Children’s Schooling 407

52

Female

7.5 (2.4) 7.8 (2.1) 7.1 (2.7)

10.7 (2.3) 10.5 (2.1) 10.9 (2.4)

Children’s education (years of schooling)

3.6 (3.3) 3.7 (3.3) 3.6 (3.3)

7.9 (3.0) 8.0 (3.1) 7.9 (3.0)

Mother’s education (years of schooling)

Means

4.6 (3.2) 4.9 (3.2) 4.3 (3.2)

8.6 (3.3) 8.6 (3.3) 8.5 (3.2)

Father’s education (years of schooling)

.023 (.041) .025 (.054) .029 (.063)

.119* (.016) .101* (.023) .138* (.022)

Children’s education with respect to mother’s education

.100** (.055) .183* (.087) .103 (.069)

.120* (.016) .118* (.024) .124* (.020)

Children’s education with respect to father’s education

Estimated elasticities

No

No

Yes No

No

Yes**

No

No

Yes Yes

No

Mother’s and father’s education signiﬁcantly different from one another

Yes

Mother’s and father’s education jointly signiﬁcant

Inﬂuence of Parent’s Schooling on Children’s for Those Living in Hungary

.20

.18

.16

.25

.25

.25

Adjusted R-squared

Notes: The OLS regressions have children’s schooling in years on the left-hand side. The children are between 17 and 47 years old in 1993. On the right-hand side is: mother’s schooling in years, father’s schooling in years, child’s age and age-squared, a dummy variable taking on the value 1 if the child is still in school, a dummy variable taking on the value 1 if the child is now in an urban environment, a dummy variable taking on the value 1 if the child is now the head of a household, and, where appropriate, a dummy variable taking on the value of 1 if the child is male. Note that Hungarian means the non-Gypsies in the sample. Source: Author’s calculations from the Hungarian Household Panel Survey (1995). Yes** indicates a p-value between .05 and .10; otherwise yes indicates a p-value below .05. * Indicates a p-value below .05. ** Indicates a p-value between .05 and .10. For ease of reading, elasticities are shaded if p-value is below .10.

54

Male

990

Female

106

935

Male

Gypsies All

1925

Hungarians All

Sample size

Table 2.

408 Ira N. Gang

105

143

Male

Female

22

80

Female

Baltics All

101

Male

68

Male, Jewish

181

37

Male, not Jewish

Ukrainians All

80

Female, Jewish

Female, not Jewish 63

248

Russians All

Sample size

Table 3.

11.5 (2.3)

11.9 (2.5) 12.3 (2.5) 11.6 (2.4)

13.1 (2.3) 13.6 (2.1) 12.7 (2.4) 12.8 (2.3) 12.6 (2.5) 13.4 (2.2) 13.8 (2.1)

Children’s education (years of schooling)

9.5 (3.5)

8.9 (3.6) 8.9 (3.4) 8.9 (3.8)

10.3 (3.7) 10.7 (3.9) 10.0 (3.6) 9.9 (3.4) 10.1 (3.7) 10.1 (4.1) 11.1 (3.7)

Mother’s education (years of schooling)

Means

9.5 (3.10)

9.3 (3.8) 9.4 (3.8) 9.2 (3.9)

10.8 (4.0) 11.1 (4.1) 10.6 (3.9) 11.3 (4.0) 10.1 (3.8) 11.3 (4.3) 11.0 (4.0)

Father’s education (years of schooling)

.057 (.234)

.122* (.045) .151* (.059) .111 (.073)

.074* (.038) .086 (.053) .083 (.051) .110 (.069) .080 (.074) .126** (.071) .072 (.073)

Children’s education with respect to mother’s education

.019 (.175)

.063 (.042) .069 (.055) .042 (.064)

.126* (.035) .139* (.047) .100* (.050) .147* (.065) .070 (.079) .179* (.072) .101** (.060)

Children’s education with respect to father’s education

Estimated elasticities

No

Yes**

No

No

Yes

No

No

Yes

No

Yes

No

Yes**

No

No

Yes

Yes

No

No

No

.04

.03

.21

.16

.07

.33

.17

.28

.20

.17

.20

Mother’s and Adjusted father’s education R-squared signiﬁcantly different from one another

Yes

Yes

Yes

Mother’s and father’s education jointly signiﬁcant

Inﬂuence of Parent’s Schooling on Children’s for Those Born in the Former Soviet Union

The Effect of Parent’s Schooling on Children’s Schooling 409

35

28

41

Female

Caucasus All

Central Asia All

11.5 (6.1)

12.9 (2.6)

10.9 (2.7) 10.4 (2.8) 11.4 (2.6)

Children’s education (years of schooling)

8.5 (3.6)

10.4 (4.0)

7.5 (3.4) 8.2 (3.7) 6.9 (3.0)

Mother’s education (years of schooling)

Means

9.2 (4.3)

11.4 (3.9)

7.8 (3.9) 7.6 (4.0) 8.0 (3.8)

Father’s education (years of schooling)

.339* (.081)

.151 (.104)

.114** (.062) .088 (.085) .093 (.096)

Children’s education with respect to mother’s education

.022 (.073)

.107 (.120)

.176 (.068) .151 (.092) .221** (.109)

Children’s education with respect to father’s education

Estimated elasticities

Yes

Yes**

Yes

Yes

Yes

Mother’s and father’s education jointly signiﬁcant

Yes

No

No

No

No

.28

.11

.14

.24

.20

Mother’s and Adjusted father’s education R-squared signiﬁcantly different from one another

Notes: The OLS regressions have children’s schooling in years on the left hand side. The children are between 25 and 50 years old in 1983 and living in the United States. On the right-hand side is: mother’s schooling in years, father’s schooling in years, child’s age and age-squared, a dummy variable taking on the value 1 if the child is still in school, a dummy variable taking on the value 1 if the child was born in a city, a dummy variable taking on the value 1 if the child migrated within the Former Soviet Union, and, where appropriate, a dummy variable taking on the value of 1 if the child is male, and a dummy variable taking on the value of 1 if the child is Jewish. Source: Author’s calculations from the Soviet Interview Project (Millar et al., 1987). Yes** indicates a p-value between .05 and .10; otherwise yes indicates a p-value below .05. * Indicates a p-value below .05. ** Indicates a p-value between .05 and .10. For ease of reading, elasticities are shaded if p-value is below .10.

36

Male

Belarus and Moldava All 71

Sample size

Table 3. (Continued )

410 Ira N. Gang

The Effect of Parent’s Schooling on Children’s Schooling

411

What about the effects of parent’s schooling on children’s schooling? For Germans, the estimated elasticities are very small, but signiﬁcantly different from zero. Children’s education is very inelastic with respect to parent’s schooling. Father’s schooling is more important than mother’s, on the order of three times as important for males and twice as important for females. This pattern is not maintained among Germany’s second generation immigrants. For Greeks, parent’s education plays no role in children’s schooling. For Turks, no role is found for women, while for men father’s education plays a signiﬁcant role. Mother’s schooling plays a role for Italian females; for males mother’s schooling actually seems to lower their educational achievement. Father’s schooling matters to male Spaniards, but not mother’s; for females there is no parental effect. Yugoslavs are very different. Mother’s schooling is three times more important for females over males; and the effect of father’s schooling is to lower children’s schooling for both males and females. Parental schooling has an effect on the schooling attainment of the next generation in varying degrees. For Germans, father’s education is a more important inﬂuence on educational attainment than mother’s education. For Yugoslavs the opposite is true. For the other groups parent’s schooling has a weak relationship, certainly a weaker relationship to children’s schooling attainments. Schultz (1984) also found, for the United States, that there is a weaker link between the second generation and their parents than between the children of the native-born and their parents. The shock of immigration weakens the inter-generational transfer of human capital through this mechanism. Let us now turn to the results from using the HHPS. The average non-Gipsy Hungarian born after 1945 had 10.7 years of schooling, while the Gypsy has a much lower 7.5 years. The level of Gipsy education is more compatible to the parental generation of Hungarians. Gipsy mothers averaged 3.6 years, and fathers 4.6 years of schooling. For Hungarians, parent’s schooling matters to children, very slightly more so for females. Mother’s and father’s schooling have the same effects on children. It is a different matter for Gypsies. For females, parent’s schooling has no effect. For males, father’s does and mother’s does not. The effect of father’s is quite strong (though inelastic). Hypothetically, if the Gipsy father had twice their average level of education, the Gipsy son would have approximately 1.5 additional years of schooling. For the cohort we analyze from the FSU, the year of birth may have been as early as 1933. This was necessitated by the age structure of the sample. Still, we ﬁnd the mean level of schooling quite high across all of the Republics, although clearly it is highest among Russians, and it is higher among males than females. Even parents’ schooling is relatively high, although it is as low as 6.9 years for females from Belarus and

412

Ira N. Gang

Moldava. The greatest gains in schooling has occurred outside of Russia and among females. Examining the FSU results, it is quite clear that, except in the Baltic Republics (where we only have 22 observations), parent’s education matters. Furthermore, the effects of mother’s and father’s schooling on children’s are not signiﬁcantly different from one another. However, examining the estimated elasticities presents a slightly different story, especially considering our small sample sizes. For Russians, father’s education is more important, about 1.5 times than mother’s (except for Jewish females, for whom mother and father exert the same inﬂuence). For Ukrainians, mother’s education is about two times more important than father’s. Fathers matter about twice as much for those from Belarus and Moldava; while mothers 1.5 time for those from the Caucasus. For those from Central Asia, mother’s education is extremely important, father’s not at all. Indeed the elasticity estimate for Central Asians of children’s education with respect to mother’s schooling is the largest in all three data sets. Finally, in comparing non-Jewish Russians to Jewish Russians, we ﬁnd the effect of parental schooling on children’s schooling attainment stronger for non-Jews, particularly with respect to father’s schooling attainment.

4. Conclusions We have examined the effects of mother’s and father’s schooling on children’s schooling. Although the conventional wisdom argues that mother’s schooling is more important than father’s in determining children’s schooling, our ﬁndings show a different result. We have found, contrary to the conventional wisdom, that overall the evidence indicates that father’s education is more important than mother’s and that mother’s schooling is relatively more important for females. However, this varies quite a bit for different ethnic groups, and there is a lot of evidence within this study that contradicts this general statement. The effect of parent’s schooling is generally less for second generation immigrants in Germany and for Gypsies in Hungary. This may be because of the low schooling attainments of the parents and the institutionalization of schooling in the children’s generation. Note that all of the results, even though for the most part statistically different from zero, are small, that is, children’s schooling is very inelastic with respect to parent’s schooling. Perhaps this is because of the average parent’s low educational attainment relative to their children, or the institutionalization and increased egalitarianism of the educational systems in all of the countries in the second half of this century. This would tend to place factors other than parent’s schooling attainments as important actors in the determination of children’s schooling attainment.

The Effect of Parent’s Schooling on Children’s Schooling

413

It is, perhaps, most surprising that we ﬁnd parental inﬂuences, and that they vary so widely across countries, subgroups, and gender. We must raise the point that these three data sets encompass countries that loosely can be termed Central Europe. In the context of the discussion of why one might expect mother’s or father’s schooling to matter more, we much raise the question: ‘‘Is there a Central European phenomenon that is different then what we witness in the United States or in the developing economies?’’ This chapter raises many questions. It tells us that we need to further explore the underlying socioeconomic settings that determine the link between children’s schooling attainment and that of their parents. Understanding this link and what it responds to may become especially important in light of changes in developing and in the transition economies, and the commensurate changes in family settings, economic opportunities and politics.

References Arai, K. (1989), A cross-sectional analysis of the determinants of enrollment in higher education in Japan. Hitotsubashi Journal of Economics 30, 101–120. Chiswick, B.R. (1988), Differences in education and earnings across racial and ethnic groups: tastes, discrimination, and investments in child quality. Quarterly Journal of Economics 103 (3), 571–597. Deutsch, J., Epstein, G.S., Lecker, T. (2006), Multi-generation model of immigrant earnings: theory and application. In: Solomon, P., Konstantinos, T. (Eds.), The Economics of Immigration and Social Diversity (Research in Labor Economics, Vol. 24). Emerald Group Publishing Limited, Bingley, UK, pp. 217–234. Devereux, P., Salvanes, K. (2005), Why the apple doesn’t fall far: understanding intergenerational transmission of human capital. American Economic Review 95 (1), 437–449. Dobson, R.B. (1984), Soviet education: problems and policies in the urban context. In: Morton, H.W., Stuart, R.C. (Eds.), The Contemporary Soviet City. M.E. Sharpe, Armonk, NY, pp. 156–177. Gang, I.N. (1997), Schooling, parents and country. DIW-Vierteljahrshefte (Quarterly Journal of Economic Research) 1–97, 180–186. Gang, I.N., Stuart, R.C. (1996), Urban to urban migration: Soviet patterns and post-Soviet implications. Comparative Economic Studies 38 (1), 21–36. Gang, I.N., Stuart, R.C. (1997), What difference does a country make? Earnings of Soviets in the Soviet Union and in the United States. The Quarterly Review of Economics and Finance 37 (Suppl. 1), 345–360.

414

Ira N. Gang

Gang, I.N., Zimmermann, K.F. (2000), Is child like parent? educational attainment and ethnic origin. Journal of Human Resources 35 (3), 550–569. Gertler, P., Glewwe, P. (1992), The willingness to pay for education for daughters in contrast to sons: evidence from Rural Peru. The World Bank Economic Review 6 (1), 171–188. Haveman, R., Wolfe, B. (1995), The determinants of children’s attainments: a review of methods and ﬁndings. Journal of Economic Literature 33, 1829–1878. Hungarian Household Panel Survey 1992–1994 [Computer File] (1995), Conducted by TARKI (Social Research Informatics Centre), Department of Sociology, Budapest University of Economics, Hungary. Kossoudji, S.A. (1989), Immigrant worker assimilation: is it a labor market phenomenon? Journal of Human Resources 24, 495–527. Linz, S.J. (1995), Russian labor market in transition. Economic Development and Cultural Change 43, 693–716. Millar, J. R. (Ed.) (1987), Politics, Work, and Daily Life in the USSR. Cambridge University Press, New York. Millar, J.R., et al. (1987), Soviet Interview project, 1979–1983. Inter-University Consortium for Political and Social Research, Ann Arbor, MI. Ofer, G., Vinokur, A. (1992), The Soviet Household under the Old Regime. Cambridge University Press, New York. Rey, E., Racionero, M. (2002), Optimal education choice and redistribution when parental education matters. Oxford Economic Papers 54, 435–448. Riphahn, R.T. (2003), Cohort effects in the educational attainment of second generation immigrants in Germany: an analysis of census data. Journal of Population Economics 16 (4), 711–737. Schultz, T.P. (1984), The schooling and health of children of U.S. immigrants and natives. In: Schultz, T.P., Wolpin, K.J. (Eds.), Research in Population Economics, Vol. 5. JAI Press, Greenwich CT, pp. 251–288. Sik, E. (1995), Measuring the unregistered economy in post-communist transformation. Eurosocial Report 52, Vienna, Austria. Tansel, A. (1997), Schooling attainment, parental education, and gender in Cote d’Ivoire and Ghana. Economic Development and Cultural Change 45 (4), 825–856 (University of Chicago Press). Thomas, D. (1994), Duncan ‘‘Like father, like son or like mother, like daughter: Parental education and child health’’. Journal of Human Resources 29 (4), 950–989. Wagner, G.G., Burkhauser, R.V., Behringer, F. (1993), The English language public use ﬁle of the German socio-economic panel. Journal of Human Resources 28, 429–433.

CHAPTER 18

Intergenerational Transfer of Human Capital under Post-War Distress: The Displaced and the Roma in the Former Yugoslavia Martin Kahaneca and Mutlu Yukselb a

Department of Economics, Central European University (CEU), Nador u. 9, H-1051 Budapest, Hungary E-mail address: [email protected] b Department of Public Policy, Dalhousie University, Halifax, NS, Canada B3H 3J5 E-mail address: [email protected]

Abstract In this chapter, we investigate the effects of vulnerability on income and employment in Bosnia and Herzegovina, Croatia, Montenegro, and Serbia using a unique 2004 UNDP dataset. Treating the collapse of the former Yugoslavia as a natural experiment, we compare three groups that have been differently affected by the wars and post-war distress: the majority as the benchmark, the ex ante and ex post vulnerable Roma people, and the ex ante equal but ex post vulnerable refugees and internally displaced people (RIDPs). Our ﬁndings reveal signiﬁcant negative effects of vulnerability on income and employment. RIDPs seem to be about as negatively affected as Roma across the four states, which indicate that vulnerability inﬂicted by relatively recent displacement may have similar effects as vulnerability rooted deep in the past. When we look at education as one of the key determinants of socio-economic outcomes, both groups exhibit similarly substandard educational outcomes of children and signiﬁcant inertia in intergenerational transfer of human capital. Our ﬁndings highlight the need for policies that not only tackle vulnerability as such, but address the spillover effects of current vulnerability on future educational attainment. Keywords: Vulnerability, labor market, education, Roma, refugees, internally displaced people, discrimination integration JEL classiﬁcations: I21, I12, J24, N34

Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008024

r 2010 by Emerald Group Publishing Limited. All rights reserved

416

Martin Kahanec and Mutlu Yuksel

1. Introduction It has become customary in the literature to look at the roles that ethnicity and immigrant origin may play for socio-economic outcomes in contexts characterized by a static partition of the studied population by ethnicity or immigrant origin. Vulnerability in terms of inclusion into social and economic relationships and outcomes is then ascribed to these static measures. Constant and Zimmermann (2008) propose a two-dimensional measure of ethnicity whereby the strength of the attachment to the host and own cultures is measured and shown to affect socio-economic outcomes. In some situations, however, ethnicity remains constant but the changing context interacts with ethnicity and engenders vulnerability of some ethnic groups. One such example is the case of former Yugoslavia, where the violent conﬂicts of the 1990s gave rise to new boundaries and displaced people along ethnic lines. Internally displaced persons (IDPs) – ‘‘Persons or groups of persons who have been forced or obliged to ﬂee or to leave their homes or places of habitual residence, in particular as a result of or in order to avoid the effects of armed conﬂict, situations of generalized violence, violations of human rights or natural or humanmade disasters, and who have not crossed an internationally recognized state border’’ exemplify such a situation (UN High Commissioner for Refugees, 1998). Examples include Serbs displaced from Kosovo to Serbia during or shortly after the NATO bombing of Kosovo in 1999; or Muslim Bosnians displaced from Serb-dominated parts of Bosnia and Herzegovina to Bosnia as a consequence of the Bosnian War (1992–1995). A remarkable peculiarity of displacement in former Yugoslavia is that in most cases displacement entailed a changing status from being a minority in the given settlement (e.g. Muslim Bosnian in a predominantly Serb village in Bosnia and Herzegovina) to being a member of ethnic majority (e.g. Muslim Bosnian in Bosnia). This applies to the displaced Serbs as well, who were dominant in former Yugoslavia, but were displaced from non-Serb settlements to those dominated by Serbs. Another speciﬁc feature of the context of former Yugoslavia is that several groups that were a minority before the 1990s have become a majority group in the newly emerged states (e.g. Kosovars in Kosovo). The context of former Yugoslavia thus enables one to study socio-economic outcomes of people that were ex ante fairly integrated and equal, but put in a vulnerable position by the armed conﬂicts of the 1990s that resulted in their displacement and ex post vulnerability that does not match the traditional ethnic minority–majority dichotomy. In this chapter, we evaluate the effect of vulnerability on income and employment in the context of former Yugoslavia. We study the effects of vulnerability for two groups that shared the social, political, and economic developments and the distress caused by the armed conﬂicts, yet their

Intergenerational Transfer of Human Capital under Post-War Distress

417

experience differed in their ex ante and ex post vulnerability: The RIDPs who were put in a vulnerable position by exogenous events – the Yugoslav wars and the Roma who were in a vulnerable position regardless of the wars. Those who were neither displaced nor members of the Roma ethnic minority can serve as a natural control group. As educational attainment is one of the key determinants of socio-economic outcomes, we then study the educational attainment of children and intergenerational transfer of human capital across the three studied groups. This approach enables us to elucidate the long-run effects of vulnerability on socio-economic outcomes. The possibility to benchmark RIDPs’ outcomes, besides the usual control group of those who were not affected or vulnerable, to those of the people who were vulnerable ex ante as well as ex post makes the former Yugoslav context particularly interesting for the study of vulnerability. In fact, being tied to the fall of the Berlin War that marked the end of the bipolar world order that weakened the communist federal regime in SFRY and unleashed the separatist factions in turn, these armed conﬂicts can in fact be interpreted in the present study as a natural experiment. This enables us under certain conditions to interpret the effects of vulnerability on income and employment outcomes as causal. In the next section, we review the literature on the topics studied. We then discuss the context of former Yugoslavia and the fates of IDPs and Roma in particular. The following section introduces and describes the data. We then develop an estimation strategy and present the results. Finally, we conclude and discuss some policy implications.

2. Literature review Extensive literature looks at association between armed conﬂicts and country’s socio-economic performance from a macroeconomic perspective. This strand of the literature mainly ﬁnds that war impacts are limited to the destruction of physical capital, in line with the predictions of the neoclassical economic growth model, which suggests rapid catch-up growth post-war. Among others, using the extensive U.S. bombing campaign in Vietnam as a quasi experiment, Miguel and Roland (2010) show that U.S. bombing did not have had long lasting impacts on poverty rates, consumption levels, infrastructure, literacy and population density 25 years after the war in Vietnam. Studies that focus on U.S. bombing during WWII – including in Japan (Davis and Weinstein, 2002) and Germany (Brakman et al., 2004) – also ﬁnd few if any persistent impacts of the bombing on local population or economic performance. Along these lines, Organski and Kugler (1977, 1980) provided similar evidence on war devastation mainly for European countries suggesting that for both

418

Martin Kahanec and Mutlu Yuksel

capitalist and socialist economies, the economic effects of the two world wars tended to dissipate after only 15–20 years. Owing to data constraints however, only a handful of studies has attempted to provide microlevel evidence on the cost of armed conﬂicts on civilians’ outcomes. Using plausibly exogenous city-by-cohort variation in the intensity of WWII destruction in Germany as a natural experiment, Akbulut-Yuksel (2009) shows that wartime destruction had a substantial negative effect on long-term human capital formation, health, and labor market outcomes of Germans who were at school-age during WWII. Angrist and Kugler (2008) show that an exogenous upsurge in conﬂict activities arising from increase in coca prices and cultivation in Colombia has a negative effect on teenager boys’ school enrollment. Shemyakina (2006) examines the effects of civil conﬂict in Tajikistan and ﬁnds that girls residing in conﬂict areas are less likely to complete secondary school education; however, the civil conﬂict had little, or no, effect on educational attainment of boys. Similarly, using WWII as an instrumental variable to estimate the causal effect of education on earnings, Ichino and Winter-Ebmer (2004) ﬁnd that individuals who were 10 years old during or immediately after WWII acquire less education and earned signiﬁcantly less in adulthood compared to other cohorts within Germany and Austria as well as to individuals of the same cohort born in non-war countries (viz., Switzerland and Sweden). This study closely relates to literature looking the labor market impacts of displacement. Using the 15 years of civil conﬂict in Colombia as a natural experiment, Caldero´n and Iba´n˜ez (2009) studied the impact of forced migration on the labor market outcomes. To address the endogeneity in the location decision, they use an interaction of the number of massacres at the origin and the distance to the state capital as an instrumental-variable for these immigrants ﬁnal destination. They ﬁnd that the labor supply shock induced by the displaced people has negative impacts on wages and employment opportunities of all workers, but these adverse labor market impacts are particularly large for low-skill workers. Kondylis (forthcoming) provides similar evidence from the civil conﬂict in Bosnia and Herzegovina. Using the level of violence in the pre-war residence as an instrument for individual’s displacement, she ﬁnds that there is a positive selection into displacement. However, she shows that displaced Bosnians are less likely to be in work, particularly women, even though they assimilate into the labor market over time. She suggests that the high levels of informality are likely to contribute to the negative effect of displacement. However, the inactivity of Bosnian women after displacement also leaves room for channels such as cultural and sociological factors that play an important role in intra-household allocations. This chapter also contributes to literature looking at Roma and their socio-economic outcomes. Using data from the UNDP/ILO survey conducted in Bulgaria, the Czech Republic, Hungary, Romania and

Intergenerational Transfer of Human Capital under Post-War Distress

419

Slovakia in 2001, Milcher and Zigova´ (2005, 2006) analyze the educational attainment of Roma people and to what extent their human capital is rewarded in the labor market. They test whether the insufﬁcient education is a mediator for their weak attachment to the labor market and high poverty among Roma. First, they ﬁnd that Roma people are more likely to reside in regions with lower economic performance and school enrollment rates. Second, they show that the likelihood of obtaining a regular wage job increases and the probability of being passive beneﬁcial decreases substantially if one household member has higher education. Moreover, they ﬁnd that the propensity to have occasional wage income is similar across different education categories and education has a marginal, negative impact on the probability of households living on loans. Although the education serves as a way out of poverty trap for Roma people, they concluded that education is more important for Roma residing in less developed economies in Central and Eastern Europe. O’Higgins and Ivanov (2006) revisit the same questions using two surveys compiled in 2002 and 2004. Similar to Milcher and Zigova´ (2005), they ﬁnd that the lack of formal education explains the considerable part of the high unemployment among Roma. However, they also ﬁnd that discrimination against Roma in the labor market has an important role for their weaker labor market attachment. Thus, due to the statistical and taste-base discrimination in the labor market, majority of Roma people work in informal sector in low-quality jobs. The authors argue that policy makers should devise labor market programs that are likely to generate opportunities for autonomous income rather than temporary employment programs to improve the labor market outcomes of Roma people. Milcher (2006) provides further evidence on the well-being and vulnerability of Roma people. Using microlevel data on Roma, refugees, IDPs and the majority living in proximity to the Roma, she ﬁrst shows that income and expenditure are highly correlated with individual’s educational attainment, labor market outcomes, and access to secure housing and health care. However, the difference between poor and nonpoor households is less pronounced in Roma sample relative to refugees and internally displaced sample. In other words, she ﬁnds that regardless of education or other individual characteristics, the probability of being poor is substantially higher among Roma people or, to a lesser extent, among refugee or IDP compared to the majority population.

3. Background on internally displaced people and Roma population in Europe One of the venerable groups we analyze in this study is internally displaced people. In the UN report (2006), internally displaced person is deﬁned as ‘‘Persons or groups of persons who have been forced or obliged to ﬂee or

420

Martin Kahanec and Mutlu Yuksel

to leave their homes or places of habitual residence, in particular as a result of or in order to avoid the effects of armed conﬂict, situations of generalized violence, violations of human rights or natural or humanmade disasters, and who have not crossed an internationally recognized state border.’’ According to the Internal Displacement Monitoring Centre (IDMC) in 2005, there were 24 million internally displaced people in 51 countries worldwide. In contrast to other vulnerable groups across the globe, displaced people are not necessarily vulnerable before their displacement per se (UNDP, 2006). However, the conﬂicts and the consequent displacements mostly cost them their wealth, homes, jobs, and networks. In many cases, they have limited ability to transfer their human capital to their new destinations and have hardship in entering the local labor market in their new homes. In this study, we mainly focus on individuals and families in the Central and Eastern Europe who were displaced during the past two decades. Owing to the outbreaks of series of civil conﬂicts and political turmoil in Socialist Federal Republic of Yugoslavia during 1990s, thousands of families were forced to leave their homes and communities without the institutional and organizational infrastructure to accommodate such displacement. In general, inter-ethnic relations in pre-war Socialist Federal Republic of Yugoslavia were cordial, as Tito managed to enforce a strict policy of ‘‘brotherhood and unity’’ by suppressing ethno-nationalism among the various ‘‘nationalities’’ or ‘‘ethnicities.’’ However, shortly after the fall of the Berlin Wall, the communist federal regime in Socialist Federal Republic of Yugoslavia weakened mounting tensions between Federalist (Serbs, Yugoslavs) and Separatist factions (Croats, Slovenes). Subsequently, Croatia, Macedonia and Slovenia declared independence in 1991 and Yugoslavia began to dissolve. Following these events, civil war broke out in Bosnia (1992–1995) between the pro-independence Bosniak–Croat coalition and the Serbs who boycotted the referendum for independence (Swee, 2009). At the same time, the Croatian War of Independence (1991– 1995) broke out between the Croatian army and the Serbia-controlled Yugoslav People’s Army (JNA) and the local ethnic Serbs in Croatia, when the latter announced their secession from Croatia. As a result, the Serb forces in Bosnia and Croatia carried out waves of aggression that marked the earliest events of the Bosnian War and Croatian War of Independence, killing and displacing thousands of Bosnians and Croats (Vulliamy, 1994). In August 1995, the North Atlantic Treaty Organization conducted sustained air strikes against the Serb strongholds, thus internationalizing the conﬂict in its ﬁnal stages (Owen, 1997a; Owen, 1997b). Subsequently, Serbs, Bosnian, and Croats signed the Dayton Peace Agreement in December 1995, concluding the Europe’s deadliest conﬂict since WWII. The agreement partitioned Bosnia by an Inter-Entity Boundary Line (IEBL) into two ethnically divided entities – the Bosniak–Croat Federation of Bosnia and Herzegovina (FBiH) and the Serb Republika Srpska (RS).

Intergenerational Transfer of Human Capital under Post-War Distress

421

Overall, the human cost of the armed conﬂict was tremendous. Reports by the International Criminal Tribunal for the former Yugoslavia (ICTY) estimate that 102,000 people were declared missing or dead. According to 1999 data, the conﬂict in Bosnia and Herzegovina caused 2.2 million people to be displaced from their homes, which is half of the total population of Bosnia and Herzegovina estimated in the 1991 Census. Between 1996 and 2004, over a million of the displaced return back to their pre-war residences both from locations in Bosnia and Herzegovina, from other Yugoslav successor states, and from further abroad. Even though refugee returns have continued since then, it seems that close to a million Bosnians retain some form of displaced status (UNDP, 2006). The second source of major displacements in the territory of former Yugoslavia was the armed conﬂict in Kosovo. During the Socialist Federal Republic of Yugoslavia, Kosovo was given an autonomous status within the Republic of Serbia since the majority of the population in Kosovo was Albanian. However, Kosovo declared independence with the dissolution of Socialist Federal Republic of Yugoslavia, which initiated the years of conﬂict between Yugoslav government and Kosovo Albanian rebel guerillas (and the near conﬂagration in Macedonia in 2001). The conﬂict was resolved after NATO attacked Yugoslavia, and Yugoslav troops were withdrawn from Kosovo. Nevertheless, like the aforementioned conﬂicts in Bosnia and Croatia, war in Kosovo caused a massive displacement of population in Kosovo which is estimated to be close to a million people. Another vulnerable group we examine in this study is Roma people who have been historically subjected to persecution and discrimination (Fraser, 1992). Estimates suggest that there are approximately 5–10 million Roma people worldwide, majority of them residing in the Central and Eastern Europe. Roma is one of the most vulnerable groups in Europe with very low labor force participation, extremely high unemployment (often reaching 50–80%). Even when Roma people are employed, they primarily work in the informal sector in unsecured jobs, especially in Southeastern Europe. A lack of formal education, poor health and discrimination has been put forward as some of the potential reasons for the underrepresentation of Roma in the formal sector. For instance, the UN study in 2006 reports that 2 of the 3 Roma (compared with one in seven in majority communities) do not complete primary school, and 2 of the 5 (compared to 1 in 20 in majority communities) do not attend primary school. The ﬁgures are even more striking when we focus only on Roma women. Estimates in the UN report show that three quarters of Roma women do not complete primary education (compared with one in ﬁve women from majority communities) and almost a third is illiterate (compared with 1 in 20 women from majority communities). The economic and social vulnerability of Roma population seem to prevail over generations. Like their parents, Roma children have lower educational attainment, spend less time at school and are more likely to be

422

Martin Kahanec and Mutlu Yuksel

illiterate. For instance, 38% of Roma children do not complete elementary school, compared to only 4% for children from majority households. On the contrary, only a small fraction of Roma children with elementary education stay on at school to complete either primary or secondary education.

4. Data and descriptive statistics In this chapter, our results are based on the UNDP’s Vulnerable Groups Survey, conducted in Albania, Bosnia and Herzegovina, Bulgaria, Croatia, Macedonia, Montenegro, Romania, Serbia, and Kosovo in 2004. UNDP conducted a comprehensive survey on all the households in Roma settlements and areas with large fraction of Roma population, RIDPs, and residents of majority communities living in proximity to these two vulnerable groups. This survey provides a wide range of information on individual and household characteristics as well as detailed information on community environment, labor market, and discrimination. In the empirical analysis, we mainly use information collected in Croatia, Bosnia and Herzegovina, Montenegro, and Serbia from this survey as displaced people in the sample are residing only in these countries. In UNDP data, the areas with Roma enclaves were determined using countries’ census data. UNDP survey was conducted in areas where percentage of Roma was equal to or higher than the nationwide percentage of Roma obtained from census data. Likewise, communities with large share of refugees/displaced people (RIDPs, thereafter) were deﬁned following the similar methodology. However, in order to construct the national averages for RIDPs, ofﬁcial registries and data provided by relevant institutions dealing with displaced populations were used for the sampling design instead of census data. In addition, UNDP data provides detailed information on majority population deﬁned as ‘‘non-Roma, non-displaced’’ that is living in proximity to these two vulnerable groups. The control groups’ samples were constructed using similar approach as for the two vulnerable groups. For Roma sample, majority population interviewed is representative samples of non-Roma communities living in settlements with Roma communities of ‘‘average and above’’ size. Similarly, the control group for displaced sample is nondisplaced populations living in proximity. In this respect, our data is representative within communities with larger share of Roma and RIDPs. Thus, it is worth noting that the status of majority samples could be worse than national averages as these samples are representative of communities living in proximity to the two vulnerable samples. However, these control groups still provide the ‘‘benchmark’’ needed for evaluation of Roma and displaced persons’ poverty and vulnerability since they live in the same community and face with similar hurdles.

Intergenerational Transfer of Human Capital under Post-War Distress

423

The strengths of this data are multifold. First of all, it is a unique data that allows us to analyze the majority, the ex ante and ex post vulnerable Roma people, and the ex ante equal but ex post vulnerable IDPs within the same dataset. Most of the previous studies were able to study only one of these aforementioned vulnerable groups in Europe; however analyzing Roma and internally displaced people together may help us understand the hurdles these groups encounter in labor market and human capital formation and devise policies to improve the next generations’ economic status and well-being. Second, it provides similar information on Roma and RIDPs for all ex-Yugoslavian countries, which allows us to compare experiences of Roma and RIDPs across different countries. Table 1 presents the characteristics of household heads for majority population, Roma and RIDPs in ex-Yugoslavian countries, respectively. Hence, mostly household heads provide general information about the household and each person residing in the household in the survey, they entail detailed analysis. Table 1 points to substantial differences in characteristics of household heads between majority population and two vulnerable groups. For example, in Bosnia and Herzegovina, Roma household heads are younger, have larger household size, lower employment probabilities, labor market income, as well as educational attainment relative to both majority and RIDPs. On the contrary, household heads in RIDP households are only slightly different from majority household heads in terms of age, marital status, household size, employment probability, and education. Similar to majority, 50% of RIDP household heads have secondary education and their average years of schooling are approximately 10 years. However, Table 1 also shows that their high levels of educational attainment are not rewarded in the labor market of host communities. It seems that RIDP household heads are earning 70% of majority household heads in the labor market even though latter has only 1 year of additional education. We observe similar patterns between groups in Croatia, Montenegro, and Serbia as well, as summarized in columns (4)–(12). Taken together, Table 1 suggests that Roma household heads are less equipped for the labor market; thus they are more likely to have adverse labor market experience both in terms of participation and earnings. On the contrary, it seems that RIDP household heads suffer in the local labor market despite their high levels of educational attainment. Likely mechanisms behind this penalty might be their lack of local labor market knowledge and ethnic networks or taste-based discrimination against them in the local labor market. Another potential explanation might be their limited ability to transfer the skills they acquired in their previous place of residence. Of course it is not possible to provide deﬁnitive proof of any of these stories, and undoubtedly additional mechanisms are at work too, but this seems to be plausible and important mechanisms for RIDP household heads’ negative labor market experience (which we will rigorously elaborate in the empirical analysis).

404

400

398

43.61 48.74 24.50 34.92 4.59 9.76 86.00 37.94 13.75 50.00 0.25 12.06 72.75 64.07 53.00 61.31 4.85 3.47 1.24 0.42 12.25 31.91 132.63 177.56 (116.12) (109.47) 89.87 205.50 (174.61) (200.61) 193.62 281.31 (208.78) (190.50) 258

49.12 41.47 11.35 23.35 54.09 22.57 59.69 30.71 2.81 0.20 54.65 480.00 (406.54) 622.01 (659.82) 789.83 (667.20)

(6)

255

198

37.27 49.94 11.37 26.26 6.32 8.97 81.70 49.74 17.45 46.11 0.85 4.15 87.45 69.19 20.24 17.77 4.97 3.33 1.75 0.47 23.14 33.33 435.15 301.97 (831.65) (258.21) 183.54 249.81 (342.81) (353.88) 435.32 398.05 (377.31) (344.80)

(5)

198

49.71 18.69 12.51 9.64 58.88 31.47 73.74 36.87 3.54 0.26 59.09 285.21 (219.08) 407.57 (317.13) 497.64 (319.30)

(7)

206

Serbia

(9)

204

404

49.20 25.99 12.12 13.12 57.18 29.70 70.05 78.20 3.18 0.21 54.95 217.33 (179.13) 252.91 (243.72) 376.26 (272.27)

(10)

(12)

R&IDPs

400

405

46.40 48.66 16.50 17.78 6.45 10.67 75.19 27.90 22.56 49.88 2.26 22.22 79.75 76.30 79.70 66.50 4.41 3.85 1.02 0.72 24.75 31.36 128.14 140.80 (202.36) (111.94) 103.21 148.54 (169.90) (228.12) 170.28 229.51 (192.19) (268.77)

(11)

R&IDPs Majority Roma

46.97 50.38 22.33 18.63 3.17 11.44 94.50 23.27 5.50 53.96 0.00 22.77 73.79 73.04 23.12 37.25 3.51 3.47 1.06 0.43 17.96 37.75 147.85 195.77 (98.82) (105.48) 117.27 203.14 (102.27) (190.86) 197.02 260.92 (164.55) (229.93)

(8)

R&IDPs Majority Roma

Montenegro

Notes: The table includes percentages, means, and standard deviations for household heads between the ages of 23 and 65 using the 2004 UNDP dataset. Income variables are in Euros.

N

49.30 23.02 11.33 18.56 57.92 23.51 71.78 70.30 3.07 0.18 48.02 237.39 (180.24) Household income-wage 292.12 (267.12) Household income-all sources 355.56 (252.06)

Age Female Years of schooling Primary school Secondary school Tertiary Married Urban Family size Number of children Employed Income

(4)

(3)

(1)

(2)

R&IDPs Majority Roma

Croatia

Summary statistics for household heads

Majority Roma

Bosnia and Hertzegovina

Table 1.

424 Martin Kahanec and Mutlu Yuksel

Intergenerational Transfer of Human Capital under Post-War Distress

425

In the next section, we particularly focus on households with children in analyzing the intergenerational transmission of human capital over generations. Therefore, it is of interest to analyze whether the differences across groups summarized in Table 1 also prevail among household heads with children younger than 22 years of age residing within the same household. Table 2 reports the descriptive statistics for this subsample. Table 2 shows that approximately less than half of the household heads in our sample have young children residing within the same household. As expected, household heads with children are younger, more likely to be married, more likely to be employed, have higher household income and educational attainment relative to household heads without children. However, Table 2 points to even larger discrepancy in returns to education between majority and RIDPs. Table 2 reveals that in all four countries in our sample, RIDP household heads with children earn 40–55% of the majority in the labor market and as a household income even though both groups have comparable educational attainment. This ﬁnding suggests that the earning penalty/loss among RIDPs generated by the displacement is even more striking when we focus only on household heads with children. Thus, this disparity in status of RIDP household heads raises concerns not only for household heads themselves but also for next generations’ economic and social well-being. Having shown the characteristics of household heads, we now turn to analysis of entire sample. Table 3 displays the characteristics of all individuals in our sample between the age of 23 and 65. Table 3 mimics patterns presented in Tables 1 and 2. Similar to previous tables, Table 3 also suggests that RIDPs resemble majority population in terms of marital status, educational attainment, and number of children. Similar to majority population, RIDPs are also more likely to be high school graduate and have none or only one child. However, Roma people are more likely to have primary or elementary education and three children or more. For example, in Montenegro 70% of the Roma population has only primary education, where the average year of schooling for Roma is 3.94 years. In Bosnia and Herzegovina, Croatia and Serbia, the majority of Roma has elementary or secondary education, whereas only 1% of them have university degree compared to 25–30% of majority and 5–20% of RIDPs in these countries. Another striking pattern emerges from Table 3 is differences in employment probabilities between majority and two vulnerable groups. Table 3 shows that even though RIDPs look similar to the majority in terms of education, their labor market attachment shows resemblance to Roma instead of majority population. In all countries, RIDPs are considerably less likely to be employed compared to majority despite the fact that they are more likely to work relative to Roma. A comparison of wage income across groups yields similar conclusion further suggesting that RIDPs indeed face hurdles not only in ﬁnding jobs but also in ﬁnding well-paid jobs in their new destination.

143

238

177

41.39 44.28 23.53 35.59 4.79 10.31 83.61 33.90 16.39 54.24 0.00 11.86 76.47 70.62 48.74 59.89 5.94 4.15 1.78 0.62 13.45 37.85 133.53 191.60 (114.18) (112.66) 100.98 201.98 (200.80) (194.42) 202.98 267.07 (226.83) (168.24) 68

44.16 30.88 11.88 13.24 67.65 19.12 88.24 35.29 4.16 0.46 73.53 530.39 (315.81) 837.93 (615.34) 934.45 (557.33)

(6)

166

76

37.67 42.28 7.23 15.79 6.50 9.93 79.62 37.33 19.75 58.67 0.64 4.00 93.37 85.53 23.49 14.47 5.69 4.57 2.19 1.01 27.11 39.47 482.08 366.01 (903.79) (299.06) 183.02 351.50 (341.53) (402.80) 455.01 513.34 (308.96) (390.37)

(5)

94

47.29 12.77 12.83 6.38 57.45 36.17 87.23 36.17 4.15 0.30 80.85 336.29 (268.66) 491.31 (319.60) 552.37 (332.03)

(7)

70

Serbia

(9)

92

163

46.23 19.02 12.34 9.82 57.06 33.13 85.28 79.75 3.91 0.29 69.94 217.29 (145.79) 292.42 (227.83) 390.87 (261.77)

(10)

(12)

R&IDPs

203

215

41.86 43.39 9.85 14.42 7.06 10.98 72.91 25.58 25.12 51.16 1.97 23.26 87.68 86.98 73.89 66.05 4.94 4.59 1.28 1.08 31.53 36.74 156.82 130.52 (261.90) (98.53) 122.07 148.14 (218.03) (262.33) 181.47 210.36 (239.14) (294.91)

(11)

R&IDPs Majority Roma

43.40 48.03 7.14 14.13 3.94 11.45 92.65 20.88 7.35 57.14 0.00 21.98 92.86 83.70 38.57 38.04 5.67 4.05 2.17 0.58 34.29 45.65 188.67 221.68 (106.02) (125.18) 162.13 230.97 (130.07) (202.92) 249.84 279.86 (186.87) (291.23)

(8)

R&IDPs Majority Roma

Montenegro

Notes: The table includes percentages, means, and standard deviations for household heads between the ages of 23 and 65 using the 2004 UNDP dataset. Income variables are in Euros.

N

45.13 18.88 11.59 13.99 65.73 20.28 86.71 70.63 3.98 0.39 62.24 262.12 (166.38) Household income-wage 365.08 (277.59) Household income-all sources 410.73 (266.43)

Age Female Years of schooling Primary school Secondary school Tertiary Married Urban Family size Number of children Employed Income

(4)

(3)

(1)

(2)

R&IDPs Majority Roma

Majority Roma

Croatia

Summary statistics for household heads with children

Bosnia and Hertzegovina

Table 2.

426 Martin Kahanec and Mutlu Yuksel

87.25 12.51 0.24

27 16.25 20 36.75

Highest degree completed Primary school 17.98 Secondary school 58.62 Tertiary 23.4

Number of children Zero 61.63 One 19.55 Two 14.85 Threeþ 3.97

51.01 18.84 22.11 8.04

32.96 54.1 12.94

52.35 37.32 63.76 29.97

(3)

R&IDPs

64.17 16.93 15.75 3.15

13.58 57.92 28.51

52.87 37.9 63.64 57.58

(4)

Majority

13.1 15.87 23.8 47.23

84.63 13.76 1.61

49.44 31.92 84.82 18.22

(5)

Roma

Croatia

54.31 18.78 16.24 10.67

45.59 49.71 4.71

50.91 37.83 71.58 32.17

(6)

R&IDPs

58.65 23.06 16.79 1.5

11.96 57.95 30.09

51.02 38.89 64.73 52.28

(7)

Majority

29.57 19.05 26.07 25.31

78.12 20.68 1.2

48.04 36.22 79.75 16.46

(8)

Roma

Montenegro

Summary statistics for all adults

41.94 21.34 20.35 16.37

26.18 53.2 20.63

49 37.17 70.55 26.16

(9)

R&IDPs

48.48 24.75 19.7 7.07

8.29 59.68 32.03

50.14 36.56 66.38 53.23

(10)

Majority

50.75 13.07 11.56 24.62

94.58 5.42 0

46.49 34.52 75.52 17.61

(11)

Roma

Serbia

50.98 20.1 19.61 9.31

16.63 61.76 21.62

49.15 37.68 65.64 31.28

(12)

R&IDPs

Notes: The table includes percentages, means, and standard deviations for household heads between the ages of 23 and 65 using the 2004 UNDP dataset. Income variables are in Euros.

49 33.5 69.13 7.07

(2)

(1)

52.42 38.72 68.82 41.97

Female Age Married Employed

Roma

Majority

Bosnia and Hertzegovina

Table 3.

Intergenerational Transfer of Human Capital under Post-War Distress 427

428

Martin Kahanec and Mutlu Yuksel

Tables 1–3 indicate that Roma people in our sample are generally younger than majority and RIDPs. Life Cycle Theory suggests that individuals’ earnings and employment proﬁle exhibit U-shape pattern. At the initial stage of life cycle, both employment and earning increases with age. However, the reverse is true at the later stage, where employment and earning decrease as individual ages. Therefore, it is of interest to analyze whether the aforementioned differences between groups in employment and earnings are driven by the differences in age distribution across groups. Analyses by age groups are summarized in Figures 1–3. Figure 1 presents the average years of schooling by age groups for majority, RIDPs and Rome in four countries in our sample. In Bosnia and Herzegovina, Serbia, and Montenegro, in all age groups, RIDPs have similar educational attainment as majority, which is substantially higher than Roma residing in their community. In addition, Figure 1 suggests that there is a less discrepancy between groups in terms of education in Croatia, where RIDPs older than 55 have similar educational attainment as Roma population. Figure 2 illustrates the employment probabilities by age groups for majority population and the two vulnerable groups. In UNDP Survey, respondents were asked to report whether they are employed in formal or informal sector. This information is important since vulnerable groups are more likely to be employed in informal sector and focusing only on formal sector may yield to a downward estimate of employment among RIDPs and Roma households. Using this information in the survey, we coded individuals as employed if they have reported working in either formal or informal sector. To begin with, U-shape pattern emerged in Figure 2 conﬁrms the life cycle theory. Indeed, probability of employment increases by age until the age of 40–45 and decreases afterwards for all groups. Note however that there are stark differences at the employment probabilities across groups. At all age groups, majority population is substantially more likely to be employed compared to RIDPs and Roma. However, even though employment probabilities of RIDPs lie between majority and Roma, they are more likely to resemble Roma than majority in terms of their employment behavior. Figure 2 shows that both RIDPs and Roma have weaker labor market attachment both in formal and informal sector which may lead to a higher poverty and vulnerability. However, focusing only on employment may be misleading since RIDPs and Roma are more likely to be on welfare and receive transfer payments compared to majority. In the survey, respondent were asked to report their household income from all sources including all kinds of wages, earnings, old age pension, disability pension, state transfer for children, unemployment, poverty and local assistance beneﬁts, remittances or gifts received from friends and relatives and aids from NGOs, charitable or humanitarian contributions. To account for potential differences across households in labor market earnings, we generated a

4

6

8

10

12

14

16

4

6

8

10

Serbia

21
61
Montenegro

61
56
51
31
26
56
51
46
36
31
26
21
Fig. 1. Average years of schooling by age groups. Notes: In ﬁgures, diamond presents the average for Roma, square the averages for RIDPs and circle the averages for majority population in the community. The averages by age groups are calculated using 2004 UNDP dataset.

Years of Schooling

12

36
14

41
Croatia

41
Bosnia & Hertzegovina

46
16

Intergenerational Transfer of Human Capital under Post-War Distress 429

Serbia

Croatia

montenegro

21
0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

21
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Age Groups

21
0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

21
0.55 0.5 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0

Bosnia & Hertzegovina

Fig. 2. Average employment probability share by age groups. Notes: In ﬁgures, diamond presents the average for Roma, square the averages for RIDPs and circle the averages for majority population in the community. The averages by age groups are calculated using 2004 UNDP dataset.

Employment Share

430 Martin Kahanec and Mutlu Yuksel

400

Montenegro

Age Groups

21
100

100

21
200

300

400

150

200

250

300

500

600

450

350

Croatia

21
1100 1000 900 800 700 600 500 400 300 200

700

Serbia

21
150

200

250

300

350

400

450

Bosnia & Hertzegovina

Fig. 3. Average household head’s income by age groups from all sources. Notes: In ﬁgures, diamond presents the average for Roma, square the averages for RIDPs and circle the averages for majority population in the community. The averages by age groups are calculated using 2004 UNDP dataset.

Household Head Income

500

Intergenerational Transfer of Human Capital under Post-War Distress 431

432

Martin Kahanec and Mutlu Yuksel

measure for household income from all sources using the aforementioned information in the survey. Figure 3 presents the average household income from all sources by age groups for all groups. We believe this measure will help us understand better the economic well-being of the vulnerable groups. It is striking that the same picture emerges as in employment when we focus on average household income for all sources. That is, similar to employment, both RIDPs and Roma households have considerably lower household income even when we account for welfare payments and other income sources. Figure 3 clearly illustrates that the household income of RIDPs are virtually similar to that of Roma and in most cases earn almost 50% lower than the majority.1 Taken together, Figures 1–3 show that in all age groups, both Roma and RIDPs suffer in the local labor market despite latter have high levels of educational attainment. Table 4 reports the summary statistics for children. We restrict our analysis to children who are between 6 and 22 years old and residing in the same household with their parents. Table 4 reveals that Roma children are almost a year younger, less likely to be female, have lower educational attainment and more likely to be out of school. Children from RIDPs households, however, appear to be similar to children from majority households in terms of years of education, school attendance, and demographic characteristics. Overall, this table hints that the lower household income in RIDPs households has no or limited adverse effects on children’s human capital formation and vulnerability is likely to be limited to the current generation. In contrast, in Roma households, vulnerability and lack of human capital had been transferred to next generations leaving this vulnerable group in the vicious cycle of poverty trap over generations. 5. The results 5.1. Income and employment In the previous section, we presented descriptive characteristics for majority population, Roma and RIDPs in ex-Yugoslavian countries. As summarized earlier, all these groups differ in terms of their observable characteristics including educational attainment, marital status, number of children, and employment choices. In this section, we compare Roma and RIDPs’ households income from all sources, monthly wage in household, and employment status relative to majority population. We present conditional means on these labor market measures using regression analysis, in which we compare Roma and RIDPs to the 1

Figures are qualitatively similar if we use average individual income instead of average household income.

625

220

N

315

15.14 46.03 7.54 54.92 36.51 8.57 0.00 3.81

(3)

R&IDPs

107

14.13 45.79 7.70 58.16 37.76 4.08 0.00 8.41

(4)

Majority

409

12.88 48.41 4.80 89.92 9.81 0.27 3.91 2.93

(5)

Roma

Croatia

141

13.74 49.65 6.50 66.39 31.09 2.52 0.71 4.96

(6)

R&IDPs

236

16.03 46.19 8.63 50.00 34.32 15.68 1.69 5.51

(7)

Majority

408

13.97 45.34 4.58 90.91 8.35 0.74 5.15 3.43

(8)

Roma

Montenegro

Summary statistics for children

432

14.09 41.90 6.52 63.57 32.25 4.18 0.69 1.39

(9)

R&IDPs

163

15.78 44.17 8.87 39.26 46.63 14.11 0.00 6.13

(10)

Majority

176

14.08 38.07 3.04 92.94 5.88 1.18 3.41 5.68

(11)

Roma

Serbia

Notes: The table includes percentages, means, and standard deviations for children between the ages of 6 and 22 using the 2004 UNDP dataset.

13.67 44.80 3.13 92.48 7.36 0.16 3.04 0.16

14.85 53.64 7.85 50.00 37.27 12.73 0.00 2.27

(2)

(1)

Age Female Years of schooling Primary school Secondary school Tertiary Married Employed

Roma

Majority

Bosnia and Hertzegovina

Table 4.

153

15.35 41.18 7.77 55.56 34.64 9.80 0.00 4.58

(12)

R&IDPs

Intergenerational Transfer of Human Capital under Post-War Distress 433

434

Martin Kahanec and Mutlu Yuksel

majority after controlling for observable characteristics. We report the estimates from our empirical analysis in Tables 5 and 6 using the 2004 UNDP dataset and follow the basic speciﬁcations for all groups’ earnings equations widely applied in the literature: the variable of interest is regressed on individual characteristics such as gender, age, education, marital status, urban indicators, number of children; country dummies to control for ﬁxed differences across countries (in regressions reported in column 1); and dummies for each group. The reported standard errors are clustered by country, accounting for the correlations in outcomes of individuals residing in the same country. The dependent variable is the natural logarithm of household income from all sources (in Table 5, Panel A), natural logarithm of household monthly wage income (in Table 5, Panel B), natural logarithm of individual income from all sources (in Table 5, Panel C), probability of employment (in Table 6 Panel A), and the likelihood of unemployment (in Table 6 Panel B). In all these regressions, the omitted group is ‘‘majority population.’’ In all tables, the ﬁrst row can be interpreted as the mean difference in the outcome of interest of Roma population with respect to majority population, once observable controls are included. Similarly, second row indicates the mean difference in outcomes between RIDPs and majority after controlling for differences in their characteristics. Panel A of Table 5 reports the estimation results where the dependent variable is natural logarithm of household income from all sources. Each column is from a separate regression that controls for female and urban dummies along with marital status, educational attainment, and number of children in the household. Table 5 suggests that Roma earn 50% less than majority population in their community even after controlling for potential differences in educational attainment and family size. The second row presents evidence on whether RIDPs households hold less income than majority population in the same community. Similar to Roma, RIDPs have also substantially lower household income relative majority population. Moreover, in most of the cases, the coefﬁcient for Roma (ﬁrst row) and RIDPs (second row) lie within each other’s 95% conﬁdence interval suggesting that these two coefﬁcients are statistically indistinguishable. Only exception is Bosnia and Herzegovina. It seems that in Bosnia and Herzegovina, even though RIDPs still have lower household income than majority, they fare better in the local market compared to Roma population in this country (column (2)). After showing the estimation results for the household income from all sources, we next turn to analyzing the household income only from wage earnings (labor market earnings). Panel B of Table 5 presents the results when the outcome of interest is logarithm of household income from labor market. This household income measure is appealing to estimate to what extent one’s human capital is rewarded in the local market holding other characteristics constant. It appears that although it is still negative and

435

Intergenerational Transfer of Human Capital under Post-War Distress

Table 5.

Estimates for income

All

Bosnia

Croatia

Serbia

Montenegro

(1)

(2)

(3)

(4)

(5)

0.517*** (0.117) 0.608*** (0.098) 0.015 (0.091) 0.116 (0.093) 0.000 (0.076) 0.512*** (0.093) 0.951*** (0.127) 0.126*** (0.022)

0.458*** (0.078) 0.529*** (0.066) 0.093 (0.089) 0.195** (0.085) 0.044 (0.067) 0.505*** (0.075) 0.845*** (0.092) 0.035 (0.023)

0.542*** (0.144) 0.736*** (0.074) 0.008 (0.133) 0.265** (0.119) 0.014 (0.067) 0.454*** (0.131) 0.599*** (0.138) 0.050* (0.028)

1,038

511

989

0.628*** (0.084) 0.169*** (0.055) 0.130* (0.074) 0.256*** (0.073) 0.091* (0.051) 0.264*** (0.080) 0.688*** (0.088) 0.006 (0.024)

0.343** (0.154) 0.369*** (0.108) 0.022 (0.094) 0.117 (0.101) 0.093 (0.090) 0.510*** (0.122) 0.907*** (0.150) 0.030 (0.036)

0.466*** (0.087) 0.519*** (0.069) 0.096 (0.090) 0.119 (0.092) 0.101 (0.079) 0.471*** (0.083) 0.892*** (0.097) 0.064** (0.028)

741

325

815

0.071 (0.123) 0.405*** (0.097) 0.269*** (0.070) 0.092 (0.094)

0.222*** (0.066) 0.311*** (0.062) 0.223*** (0.048) 0.036 (0.059)

Panel A: Household income-all sources 0.515*** Roma 0.493*** (0.033) (0.079) 0.174*** Refugee & IDPs 0.454** (0.135) (0.052) Female 0.044 0.124* (0.071) (0.069) 0.343*** Married 0.254** (0.049) (0.065) Urban 0.006 0.033 (0.026) (0.046) 0.299*** Secondary school 0.425*** (0.064) (0.069) Tertiary 0.753*** 0.673*** (0.077) (0.083) Number of children 0.012 0.013 (0.030) (0.019) N

3,015

Panel B: Household income-wage Roma 0.559** (0.106) Refugee & IDPs 0.397** (0.117) Female 0.034 (0.078) Married 0.214** (0.057) Urban 0.043 (0.062) Secondary school 0.406*** (0.068) Tertiary 0.782*** (0.074) Number of children 0.026 (0.021) N

2,302

Panel C: Individual income-all sources 0.476*** Roma 0.271* (0.096) (0.070) 0.077* Refugee & IDPs 0.246* (0.080) (0.045) 0.301*** Female 0.258*** (0.023) (0.044) 0.112** Married 0.061* (0.025) (0.046)

477 0.918*** (0.156) 0.601*** (0.076) 0.113 (0.186) 0.413** (0.174) 0.081 (0.071) 0.374*** (0.143) 0.567*** (0.158) 0.052* (0.030) 421 0.210* (0.124) 0.339*** (0.052) 0.300*** (0.048) 0.001 (0.062)

436

Martin Kahanec and Mutlu Yuksel

Table 5. (Continued )

Urban Secondary school Tertiary Number of children N

All

Bosnia

Croatia

Serbia

Montenegro

(1)

(2)

(3)

(4)

(5)

0.026 (0.024) 0.456*** (0.057) 0.882*** (0.105) 0.024 (0.014)

0.036 (0.041) 0.335*** (0.060) 0.761*** (0.073) 0.004 (0.018)

0.128* (0.071) 0.474*** (0.114) 0.926*** (0.138) 0.013 (0.030)

0.022 (0.058) 0.534*** (0.066) 1.043*** (0.076) 0.047** (0.019)

0.017 (0.046) 0.371*** (0.101) 0.589*** (0.114) 0.073** (0.034)

3,887

1,343

711

1,257

576

Notes: The table reports OLS estimates of Roma and R&IDPs on different income variables. Clustered standard errors by country are reported in parenthesis. Regressions control for schooling indicators; age and its square; a marriage, gender, and urban dummies; Roma and R&IDPs dummies; and number of children and country ﬁxed effects. ***1%, **5%, and *10% signiﬁcance level.

statistically signiﬁcant, the magnitude of coefﬁcient for RIDPs decreases once we restrict our analysis to labor market income. By contrast, however the coefﬁcient for Roma becomes more negative when we focus on income from wages. Taken together, these changes in the coefﬁcients hint that one of the reasons for the differences between groups may be that Roma are more likely to receive welfare and other source of income compared to RIDPs that are probably less aware of welfare opportunities in their new destination. As an additional outcome, we also examine the logarithm of individual income from all sources, summarized in Panel C. Since there are differences in household size across groups, a comparison of individual incomes may yield a clearer picture on the vulnerability of Roma and RIDPs. The analysis reported in Panel C mimics the previous ﬁndings on household income, summarized in Panel A and Panel B. Similarly, both individuals from Roma and RIDP households have lower individual income from all sources after differences in demographic characteristics and educational attainment is taken into account. The ﬁrst and second rows show that Roma and RIDPs receive 27% and 25% less individual income, respectively, compared to otherwise comparable majority population in the same community. These coefﬁcients are half the size of the estimates for household income presented in Panel A. Though it is beyond the scope of this chapter to disentangle the underlying mechanisms responsible for this difference; we may suggest several explanations. One potential reason would be the child employment and state transfer for children. Of course we have no deﬁnitive proof, but it might be that the

437

Intergenerational Transfer of Human Capital under Post-War Distress

Table 6.

Estimates for labor market attachment among adults

Panel A: Employment Roma Refugee & IDPs Female Married Urban Secondary school Tertiary Number of children Panel B: Unemployment Roma Refugee & IDPs Female Married Urban Secondary school Tertiary Number of children N

All

Bosnia

Croatia

Serbia

Montenegro

(1)

(2)

(3)

(4)

(5)

0.159*** (0.021) 0.140*** (0.035) 0.175*** (0.013) 0.074*** (0.016) 0.031** (0.012) 0.247*** (0.023) 0.424*** (0.020) 0.032*** (0.009)

0.181*** (0.022) 0.055*** (0.017) 0.141*** (0.017) 0.069*** (0.019) 0.004 (0.017) 0.211*** (0.023) 0.424*** (0.039) 0.011 (0.007)

0.194*** (0.042) 0.161*** (0.033) 0.165*** (0.029) 0.111*** (0.035) 0.021 (0.033) 0.255*** (0.037) 0.384*** (0.054) 0.045*** (0.012)

0.141*** (0.024) 0.193*** (0.018) 0.176*** (0.018) 0.079*** (0.023) 0.026 (0.021) 0.288*** (0.025) 0.452*** (0.035) 0.047*** (0.009)

0.123** (0.053) 0.192*** (0.030) 0.255*** (0.029) 0.005 (0.038) 0.072** (0.031) 0.225*** (0.052) 0.415*** (0.059) 0.014 (0.012)

0.216*** (0.041) 0.192*** (0.049) 0.143*** (0.013) 0.057*** (0.014) 0.027 (0.038) 0.095** (0.042) 0.279*** (0.012) 0.013*** (0.005)

0.338*** (0.031) 0.099*** (0.027) 0.170*** (0.021) 0.027 (0.024) 0.118*** (0.021) 0.023 (0.028) 0.281*** (0.032) 0.001 (0.007)

0.150*** (0.046) 0.173*** (0.040) 0.108*** (0.028) 0.096** (0.038) 0.015 (0.033) 0.137*** (0.036) 0.218*** (0.042) 0.034*** (0.010)

0.204*** (0.030) 0.262*** (0.026) 0.151*** (0.020) 0.066** (0.026) 0.023 (0.023) 0.162*** (0.024) 0.300*** (0.024) 0.010 (0.008)

0.069 (0.050) 0.254*** (0.033) 0.102*** (0.027) 0.060* (0.034) 0.070*** (0.027) 0.102** (0.040) 0.297*** (0.029) 0.013 (0.011)

7,923

2,647

1,301

2,736

1,239

Notes: The table reports marginal effect of Roma and R&IDPs on employment and unemployment. Clustered standard errors by country are reported in parenthesis. Regressions control for schooling indicators; age and its square; a marriage, gender and urban dummies; Roma and R&IDPs dummies; and number of children and country ﬁxed effects. ***1%, **5%, and *10% signiﬁcance level.

majority of the population is more likely to receive state transfer for children, and their working-age children (young adults) are more likely to engage in labor market activities compared to working-age children (young adults) from Roma and RIDPs households. The analysis from Table 4 on working-age child (young adult) employment suggests that this

438

Martin Kahanec and Mutlu Yuksel

would be one of the explanations. Another explanation would be the employment of other household members. It might be that members of majority population are more likely to have spouses who are more educated and more likely to be working compared to the vulnerable groups. Under this scenario, the vulnerable households are likely to experience double disadvantage, one due to discrimination and another due to the household composition. As ﬁnal outcomes in this section, we estimate the employment and unemployment probabilities, which are presented in Table 6. These regressions are at the individual level and reports whether individuals between 23 and 65 years old work either in formal or informal sector. We ﬁnd that adults in Roma households are 16% less likely to engage in labor market activity compared to members of majority population. The corresponding difference between employment probabilities of RIDPs and majority population is 14%. Similar to ﬁndings on household income in Table 5, for employment probability, the coefﬁcients for Roma and RIDPS are virtually similar as well (they lie within each other’s 95% conﬁdence interval). However, Panel B of Table 6 displays the ﬁndings for unemployment. We deﬁne unemployment measure using individuals who are actively looking for job; therefore our measure excludes discouraged workers and individuals that are out of labor force. Findings in Panel B of Table 6 further conﬁrm previous ﬁndings on employment. Both Roma and RIDPs are more likely to be unemployed compared to majority residing in the same community. Thus, when we consider ﬁndings in Table 6 together; we ascertain that members of two vulnerable groups disproportionately suffer in the labor market even when they exhibit the characteristics of majority population. Moreover, our analysis also suggests that these vulnerable groups are not only less likely to ﬁnd jobs in the local labor market due to discrimination or lack of social networks but also less likely to ﬁnd well-paid jobs since the differences between groups are even larger when we focus on income.

5.2. Education and intergenerational transfer of human capital To evaluate the long-run effects of vulnerability, in this section we explore the results from the analysis of educational attainment of children between 6 and 22 years of age residing in the same household. We ﬁrst look at the vulnerability along children’s educational attainment measured by being out of education. As evident from column 1 of Table 7, being a Roma increases the probability of being out of education by approximately 31%. The corresponding probability for RIDPs is 13%. Similar results arise if we look at years of education in column 1 of Table 8. Being a Roma or RIDP reduces educational attainment by 1.68 and 0.30 years, respectively. All these effects are statistically signiﬁcant at 1% conﬁdence level.

439

Intergenerational Transfer of Human Capital under Post-War Distress

Table 7.

Estimates for vulnerability along children’s school attainment

Roma Refugee & IDPs Female Age Age2 Household head’s age Household head’s age2 Household heads years of schooling Household heads income Observations

All

Bosnia

Croatia

Serbia

Montenegro

(1)

(2)

(3)

(4)

(5)

0.307*** (0.040) 0.130*** (0.040) 0.039** (0.019) 0.048*** (0.017) 0.004*** (0.001) 0.007 (0.013) 0.000 (0.000) 0.040*** (0.004) 0.034** (0.015)

0.492*** (0.060) 0.161** (0.069) 0.018 (0.036) 0.087*** (0.030) 0.005*** (0.001) 0.002 (0.021) 0.000 (0.000) 0.042*** (0.006) 0.028 (0.028)

0.022 (0.063) 0.036 (0.049) 0.013 (0.032) 0.058** (0.027) 0.004*** (0.001) 0.026 (0.026) 0.000 (0.000) 0.030*** (0.007) 0.063** (0.031)

0.281*** (0.068) 0.207*** (0.061) 0.027 (0.032) 0.010 (0.032) 0.003** (0.001) 0.003 (0.026) 0.000 (0.000) 0.042*** (0.007) 0.049*** (0.019)

0.309** (0.129) 0.055 (0.102) 0.117** (0.051) 0.007 (0.052) 0.002 (0.002) 0.014 (0.036) 0.000 (0.000) 0.034*** (0.011) 0.044 (0.051)

2,998

1,130

493

973

402

Notes: The table reports marginal effect of Roma and R&IDPs children on schooling indicator. Clustered standard errors by country are reported in parenthesis. Regressions control for children’s and household head’s age and its square and household head’s years of schooling and income. ***1%, **5%, and *10% signiﬁcance level.

Regardless of the outcome we look at, child’s age plays a signiﬁcant role as expected, whereas age of the household head turns insigniﬁcant. Children in households with higher income appear to have better educational prospects. Looking at intergenerational transfer of human capital measured by the effect of the household head’s educational attainment on children’s human capital formation, we ﬁnd a strong positive effect. In particular, every additional year of household head’s schooling reduces the child’s probability of being out of education by 4% and increases the expected educational attainment of children by 0.26 years. This high intergenerational transfer of human capital may hide important differences between the majority and vulnerable groups studied. We investigate this possibility in Panel B of Tables 7 and 8. Although we do not observe any such differences in the model with being out of education as the explained variable, we ﬁnd that intergenerational transfer of human capital is about twice as strong for vulnerable groups than for the majority population when children’s educational attainment is measured by years of education.

440

Martin Kahanec and Mutlu Yuksel

Table 8.

Roma Refugee & IDPs Female Age Age2 Household head’s age Household head’s age2 Household heads years of schooling Household heads income Observations

Estimates for children’s years of schooling All

Bosnia

Croatia

Serbia

Montenegro

(1)

(2)

(3)

(4)

(5)

1.679*** (0.184) 0.297** (0.123) 0.018 (0.086) 1.091*** (0.076) 0.017*** (0.003) 0.108* (0.060) 0.001* (0.001) 0.265*** (0.020) 0.196*** (0.069)

2.239*** (0.331) 0.084 (0.195) 0.149 (0.162) 1.130*** (0.137) 0.021*** (0.005) 0.078 (0.103) 0.001 (0.001) 0.259*** (0.031) 0.028 (0.139)

0.783** (0.325) 0.252 (0.301) 0.074 (0.172) 1.263*** (0.142) 0.023*** (0.005) 0.212 (0.140) 0.003 (0.002) 0.150*** (0.043) 0.343** (0.163)

1.404*** (0.287) 0.496** (0.205) 0.039 (0.130) 1.116*** (0.115) 0.016*** (0.004) 0.052 (0.130) 0.001 (0.002) 0.259*** (0.037) 0.188** (0.084)

2.292*** (0.693) 0.092 (0.283) 0.028 (0.221) 0.987*** (0.238) 0.011 (0.008) 0.147 (0.160) 0.002 (0.002) 0.291*** (0.066) 0.424* (0.225)

3,062

1,130

476

1,014

442

Notes: The table reports OLS estimates of Roma and R&IDPs children on years of schooling. Clustered standard errors by country are reported in parenthesis. Regressions control for children’s and household head’s age and its square and household head’s years of schooling and income. ***1%, **5%, and *10% signiﬁcance level.

This ﬁnding suggests that the vulnerability is likely being transferred over generations, where less educated parents raise children with lower human capital endowments. Pooling the countries and groups of people into a single regression may hide important differences in the effects studied if the true models differ across groups and countries. Looking at Columns (2)–(5) in Tables 7 and 8 where we report the results for children’s years of education, it is immediately obvious that vulnerability manifests itself in all the countries under scrutiny and for both the Roma and RIDPs. In all countries the effect of being Roma is larger than that of being an RIDP but Montenegro. The observed effects for RIDPs are in fact not signiﬁcant in Croatia, Montenegro and Serbia. Intergenerational transfer of human capital is positive across the board, but important differences arise between the studied groups of people. In Bosnia signiﬁcantly positive effects are observed only for Roma. Croatia exhibits similar results, but event the effects for Roma are only marginally signiﬁcant. In Serbia, all groups exhibit about the same positive

Intergenerational Transfer of Human Capital under Post-War Distress

441

and signiﬁcant effects. In Montenegro, positive and signiﬁcant effects are observed only for the two vulnerable groups.

6. Conclusions and policy recommendations This chapter elucidates the current and future prospects of vulnerable groups in the context of former Yugoslavia. We ﬁnd that vulnerability is associated with signiﬁcantly substandard income and employment prospects. As we do not ﬁnd much residual difference between the Roma and the RIDPs, we offer some evidence that vulnerability inﬂicted by relatively recent displacement may have similar effects as vulnerability rooted deep in the past. To the extent that the Yugoslav wars can be treated as exogenous and acknowledging the limitations of our study in that we cannot observe socio-economic outcomes before the wars, our results hint at causal interpretation of the effects of vulnerability. Our results show that being a member of a vulnerable group manifests itself also through the educational attainment of children. It is worrying that vulnerable groups seem to be entrapped in this adverse situation, as the link through which lower educational attainment of parents affects children’s educational attainment seems to be particularly strong for them. Comparing the two different genealogies of vulnerability, it seems that the Roma who have been in a vulnerable position before as well as after the armed conﬂicts of the 1990s in Yugoslavia are in a worse position than the RIDPs who had not been vulnerable ex ante but their displacement as a consequence of the armed conﬂicts have put them in a vulnerable position. This together with the particularly substandard educational outcomes of the Roma parents indicates that this entrapment is more severe for this group and may be an artifact of historically deep-entrenched vulnerability. From the policy perspective, these results show that while violent conﬂict can put large groups of people into a vulnerable position, the prospects of people whose vulnerability is deep-rooted historically are at a signiﬁcantly higher risk of a dynamic trap whereby parents’ outcomes determine children’s outcomes. In the educational domain, this indicates that policy efforts directed at RIDPs’ integration need to focus on childrens’ access to education, whereas for the Roma also the parents and the mechanisms through which parents affect children’s outcomes must be given signiﬁcant attention. This could entail life-long education of the parents, or specialized ﬁeld workers working with the parents and enabling them to e.g. help their children with homework assignments. For the RIDPs the policy efforts to facilitate access to education would need to address the speciﬁc situation of RIDPs households case-by-case, which could include poor infrastructure, lack of social contacts, poor housing and access to health, lack of documents, or insufﬁcient social assistance and services when in need.

442

Martin Kahanec and Mutlu Yuksel

Acknowledgment We are especially grateful to Mevlude Akbulut-Yuksel for very useful comments and discussions. The authors bare the sole responsibility for any errors that may remain.

References Akbulut-Yuksel, M. (2009), Children of war: the long-run effects of largescale physical destruction and warfare on children, IZA. Discussion Papers, No. 4407. IZA, Bonn. Angrist, J., Kugler, A. (2008), Rural windfall or resource curse? Coca, income and civil conﬂict in Colombia. Review of Economics and Statistics 90 (2), 191–215. Brakman, S., Harry, G., Schramm, M. (2004), The strategic bombing of cities in Germany in World War II and it impact on city growth. Journal of Economic Geography 4 (1), 1–18. Caldero´n, V., Iba´n˜ez, A.M. (2009), Labor market effects of migrationrelated supply shocks: evidence from internally displaced populations in Colombia. Research Working Papers 14, MICROCON – A Micro Level Analysis of Violent Conﬂict. Constant, A.F., Zimmermann, K.F. (2008), Measuring ethnic identity and its impact on economic behavior. Journal of the European Economic Association 6 (2–3), 424–433, 04–05. MIT Press. Davis, D., Weinstein, D. (2002), Bones, bombs, and break points: the geography of economic activity. American Economic Review 92 (5), 1269–1289. Fraser, A. (1992), The Gypsies. Blackwell, Oxford. Ichino, A., Winter-Ebmer, R. (2004), The long-run educational cost of World War II. Journal of Labor Economics 22 (1), 57–86. Kondylis, F. (forthcoming), Conﬂict displacement and labor market outcomes in post-war Bosnia and Herzegovina. Journal of Development Economics. Miguel, E., Roland, G. (2010), The long run impact of bombing Vietnam. Journal of Development Economics. Milcher, S. (2006), Poverty and the determinants of welfare for Roma and other vulnerable groups in Southeastern Europe. Comparative Economic Studies 48, 20–35. Milcher, S., Zigova´, K. (2005), Evidence of returns to education among Roma in central and Eastern Europe and their policy implications. International Research Journal 3 (1), spring. O’Higgins, N., Ivanov, A. (2006), Education and employment opportunities for the Roma. Comparative Economic Studies 48 (1), 6–19.

Intergenerational Transfer of Human Capital under Post-War Distress

443

Organski, A.F.K., Kugler, J. (1977), The costs of major wars: the Phoenix factor. American Political Science Review 71, 1347–1366. Organski, A.F.K., Kugler, J. (1980), The War Ledger. University of Chicago Press, Chicago. Owen, R.C. (1997a), The Balkans air campaign study: part one. Airpower Journal 11 (2), 4–25. Owen, R.C. (1997b), The Balkans air campaign study: part two. Airpower Journal 11 (3), 6–27. Shemyakina, O. (2006), The effect of armed conﬂict on accumulation of schooling: results from Tajikistan. Households in Conﬂict Network Working Paper No. 12. Swee, E. (2009), On war and schooling attainment: the case of Bosnia and Herzegovina. Households in Conﬂict Network (HiCN) Working Paper 57. UNDP. (2006), At risk: Roma and the displaced in Southeast Europe. Bratislava Regional Centre. UN High Commissioner for Refugees. (1998), Guiding Principles on Internal Displacement, E/CN.4/1998/53/Add.2, available at: http:// www.unhcr.org/refworld/docid/3c3da07f7.html [accessed 30 July 2010]. Vulliamy, E. (1994), Seasons in hell: understanding Bosnia’s war. St. Martin’s Press, New York.

PART IV

Family Issues and the Effects of Remittances

CHAPTER 19

Household Structure of Recent Immigrants to Israel Sarit Cohen-Goldnera,b a

Department of Economics, Bar-Ilan University, Ramat-Gan 52900, Israel Institute for the Study of Labor (IZA), Bonn, Germany E-mail address: [email protected] b

Abstract The change in household structure is a worldwide phenomenon that reﬂects demographic changes, social and cultural trends, and changing economic conditions. The purpose of this chapter is to explore the prevalence of multigenerational households among recent immigrants from Eastern Europe to Israel. The size of the household among these immigrants is smaller, on average, than the household size among native-Israelis, even though immigrants have a higher tendency to live in extended households. Our analysis shows that the share of multigenerational households declines with duration in Israel among young immigrants, but not so much among elder immigrants who arrived at older age. This difference may reﬂect the better economic integration of younger immigrants in the local labor market and the lower attachment of younger immigrants to cultural habits that existed in the origin country. In addition, there is a great similarity in the prevalence of multigenerational households between cohorts suggesting that immigrants, presumably, do not form a multigenerational household in Israel in order to provide them with a social anchor, but rather to help them overcome economic constraints upon arrival. Keywords: Immigration, Household Structure JEL classiﬁcations: J11, J12

1. Introduction The change in household structure is a worldwide phenomenon that reﬂects demographic changes, social and cultural trends, and changing economic conditions. For example, the rise in divorce rates and in the share of single parents, both contribute to the changing structure of the Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008025

r 2010 by Emerald Group Publishing Limited. All rights reserved

448

Sarit Cohen-Goldner

traditional families and households. Migration can also affect household structure in the receiving country if immigrants adopt a different household structure than natives. In this chapter we study the prevalence of multigenerational households among immigrants who arrived in Israel during the early 1990s from the Former Soviet Union (FSU). The prevalence of multigenerational households among immigrants is a known phenomenon. In the United States, for instance, there is considerable racial and ethnic variation in the prevalence of extended households (multigenerational and other forms), and the rates of multigenerational households among immigrants tend to be higher than among their US-born counterparts. However, the prevalence of this type of extended-family structure varies substantially by country of origin as well as by duration of residence in the United States. Under the assumption that the initial household structure new immigrants adopt in the receiving country is similar to the one they had in the origin country and reﬂects cultural preferences, the ﬁnding that differences in living arrangements between the native population and immigrants decrease with time in the new country is usually taken as evidence of assimilation and a depart from the cultural patterns of the origin country.1 Nevertheless, it is possible that the initial household structure of immigrants in the new country is not a continuation of their household structure in the origin country and results not from cultural preferences, but rather from economic constraints upon arrival. In this case, the gradually growing similarity between household structure among natives and immigrants in the new country reﬂects immigrants’ economic integration, rather than cultural assimilation. Glick and Van Hook (2002) study the impact of immigration on the racial and ethnic diversity of the older population in the United States. Using data from the CPS, they demonstrate that much of this variation is attributable to recent immigration and the relative economic position of immigrant parents. Particularly, Asian and Central and South American immigrant parents are more likely to live in households where their adult children provide most of the household income. However, they ﬁnd that the likelihood of the parents to live in this ‘‘economically-dependent’’ role decreases with duration of residence in the United States. In order to assess the extent to which living arrangements among immigrants are attributed to the migration process per se rather than merely cultural vestiges from the country of origin, Glick and Van Hook (2007) 1

For example, explanations for changes in extended-family living by duration of residence in the United States often begin with the presumption that recent immigrants come to the United States with their own unique values and preferences that inﬂuence their family behaviors, including sharing households with adult children. These preferences contrast with the ﬁndings regarding the US-born population that is more likely to value privacy and independent living (Burr and Mutchler, 1993).

Household Structure of Recent Immigrants to Israel

449

make a bi-national comparison of household structures between Mexico and the United States. Speciﬁcally, they use the 2000 Census data from Mexico and the United States to compare the prevalence and age patterns of various types of extended family and nonkin living arrangements among Mexican-origin immigrants and nonimmigrants on both sides of the US–Mexico border. They ﬁnd that newly arrived immigrants to the United States display unique patterns in the composition and stability of their households relative to nonimmigrants in both countries. Recent immigrants are more likely to reside in an extended family or nonkin household, while among those living with relatives, recent immigrants are more likely to live with extended family from a similar generation (such as siblings and cousins). Further, these households experience high levels of turnover. In the view of the authors, the results suggest that the high levels of co-residence observed among recently arrived Mexican immigrants represent a departure from ‘‘traditional’’ household/family structures in Mexico and are related to the challenges associated with international migration. That is, after arrival in the new country migrants are exposed to social and economic uncertainty, and extended household structure serves them as a social net allowing them to exploit economics of scale within the household. The mass migration from the FSU to Israel started toward the end of 1989 due to a sudden change in immigration policy towards Soviet-Jews in the FSU. The main inﬂux of immigrants occurred during the initial wave of 1989–1991, when more than 330,000 immigrants arrived in Israel. During 1992–2000 the ﬂows were much smaller such that the accumulated number of FSU immigrants who arrived during the 1990–2000 decade reached about 1 million immigrants (the Jewish Israeli population in 1989 was about 4.6 million). The demographic characteristics of FSU immigrants differ from those of the native-Israeli population.2 A substantial difference between immigrants and native-Israelis that clearly affects household structure is the composition of age. FSU immigrants are on average older than natives. The two main sources for this difference are (a) the high share of elder immigrants who arrived in this wave and (b) the lower fertility rate of immigrant women relative to the fertility rate of native females. Another notable difference between immigrants and natives is related to marriage patterns and marital status. FSU immigrants tend to get married earlier than native-Israelis, and the share of divorced individuals is also

2

Since Israel was built on waves of immigrants from a wide range of countries including western countries as well as Arab countries, one may argue that FSU immigrants are not comparable to native-Israelis as a whole. Thus, in the analysis below we present the ﬁgures for the whole Jewish native population, which includes those born in Israel or immigrated prior to 1989 (hereafter ‘‘natives’’) as well as for ‘‘European natives’’ which consists of Jews who were born in Europe/America/Oceania /FSU and immigrated prior to 1989 and those born in Israel and whose father was born in Europe/America/Oceania/FSU.

450

Sarit Cohen-Goldner

higher among male and female immigrants, in comparison to their native counterparts. These unique characteristics of FSU immigrants inﬂuence their family composition and their household structure. In general, the size of the household among FSU immigrants is smaller, on average, than the household size among natives, even though immigrants have a higher tendency to live in extended households (e.g., a household that includes more than one family or multigenerational household). The high prevalence of extended household structure upon arrival may indicate that economic constraints bring FSU immigrants to live together in order to save expenditures as well as alleviate the adaptation to the life in the new country. For example, several studies found that newly arrived FSU immigrants tend to have lower returns to education and labor market experience than natives do (e.g., Eckstein and Weiss, 2004; Cohen-Goldner and Eckstein, 2008, 2010). Failing to take this information into account may exaggerate the degree to which culturally enforced norms for co-residence account for the relatively high levels of extended households among FSU immigrants and the extent to which acculturation accounts for the declining rates of these households with time spent in Israel. Furthermore, in the analysis we distinguish between two cohorts of FSU immigrants: those who came in the initial wave of 1989–1991 and those who arrived in the subsequent wave of 1992–1994. The reason for this distinction is that prior to the 1990s, immigration from the FSU to Israel was rare. Earlier waves of immigration from FSU to Israel occurred in the 1970s and early 1980s, but since these waves were quite small, we can argue that those who came in the initial wave of 1989–1991 did not have a developed social net in Israel, while those who arrived in 1992–1994 had such a net due to the large number of immigrants who came in 1989–1991. Thus, the prevalence of multigenerational households among FSU immigrants may also depend on year of arrival. We start by describing the differences in prevalence of multigenerational households between various subgroups that the Israeli Jewish population consists of. We deﬁne a multigenerational household as a household that includes at least one parent of the head of household (or parent of the spouse of the head) or a household that includes a grandchild aged 15þ. Later, focusing on FSU immigrants, we explore the role of the personal characteristics of the head of the household, who is deﬁned in our data as the main earner in the household regardless of gender and age, and of elderly individuals (aged 60 and above) in establishing multigenerational household in Israel.3 The distinction between heads and elder individuals is especially important for estimating the impact of duration in Israel on household

3

Note that while each household has one head, the number of elderly individuals in the household can be greater than one and is not limited.

Household Structure of Recent Immigrants to Israel

451

structure since immigrants may either arrive as elderly or live into old age while in Israel, creating two very different groups of older immigrants. Older new immigrants have shorter work histories in Israel and presumably face greater economic and social difﬁculties. For example, Strosberg and Naon (1997) report that elder FSU immigrants adopt a multigenerational household in Israel, both due to economic constraints as well as due to familiarity with this type of living arrangement from life in the FSU. For this group, the formation of multigenerational households is an adaptive response to adverse structural conditions including economic disadvantage (Angel and Tienda, 1982; Burr and Mutchler, 1993). On the other hand, immigrants who arrived in Israel as younger adults, and have been aging in Israel, are more attached to the labor force and therefore may become less dependent on their adult children as they age. Katz and Lowenstein (1999) found that the younger generation of FSU immigrants emphasized the economic beneﬁts of living in shared households, but would prefer, in the future, to live in separate households. These different attitudes between elder and younger immigrants toward formation of multigenerational households in Israel may show more persistence of this household structure among elders and less persistence among younger immigrants. Our analysis shows that the share of multigenerational households declines with duration in Israel among young immigrants, but not so much among elder immigrants who arrived at older age. This difference may reﬂect the better economic integration of younger FSU immigrants in the local labor market and the lower attachment of younger immigrants to cultural habits that existed in the origin country. In addition, there is a great similarity in the prevalence of multigenerational households between the two cohorts, suggesting that immigrants, presumably, do not form a multigenerational household in Israel in order to provide them with a social anchor, but rather to help them overcome economic constraints upon arrival. 2. Data analysis The data for this study is taken from the national annual Israeli Labor Force Surveys (LFS) of 1992–2007, which is conducted by the Israeli Central Bureau of Statistics and represents the Israeli population in work ages.4 Figure 1 presents the average size of households by nativity. The average size of native household in Israel in 1992 was 3.35 persons, while among European natives it was 2.92 and among immigrants who arrived during 1989–1991 it was 3.43 persons. The size of households declined over 4

The Labor Force Survey data was provided by the Israel Social Science Data Center (ISDC) at the Hebrew University of Jerusalem.

452

Sarit Cohen-Goldner

3.6 Natives European Natives 1989-91 Immigrants 1992-94 Immigrants

3.4 3.2 3.0 2.8 2.6 2.4 2.2

2.0 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 year

Fig. 1. Mean number of persons in household. For sample and group deﬁnitions, see Table 1. Source: CBS Labor Force Surveys. Data was available to us through Israel Social Science Data Center (ISDC) at the Hebrew University. the years for all the four subgroups, where immigrants experience the sharpest decline, such that in 2007 immigrants’ household size (2.45 persons for 1989–1991 cohort and 2.36 for 1992–1994 cohort) is very similar to that of European natives (2.36). Figure 2 shows that the share of immigrant households that include young children aged 0–4 is lower than that share among natives and is very similar to that share among European native households. While among natives the share of households with young children aged 0–4 declined from 23% in 1992 to 16% in 2007, the share of these households among FSU immigrants from 1989 to 1991 (1992–1994) cohort declined from 16% (18%) in 1992 (1995) to only 10.7% (10.5%). Overall, Figures 1 and 2 show that the two groups of FSU immigrants are more comparable to European natives than to the native Jews as a whole. In Figure 3 we present the share of multigenerational households by nativity. The share of multigenerational households among natives and among European natives was rather constant over 1992–2007. The average share of multigenerational households during this period is 5.65% among natives and 3.35% among European natives. The prevalence of multigenerational households among immigrants from the two cohorts is substantially higher than among natives and European natives, and it decreases with time spent in Israel. In 1992, 29.4% of 1989–1991 immigrant households were multigenerational households, and this share declined to 11.9% in 2007, while among immigrant households

Household Structure of Recent Immigrants to Israel

453

50% Natives European Natives 1989-91 Immigrants 1992-94 Immigrants

40%

30%

20%

10%

0% 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 year

Fig. 2. Share of households including children under 4. For sample and group deﬁnitions see Table 1. Source: CBS Labor Force Surveys. Data was available to us through Israel Social Science Data Center (ISDC) at the Hebrew University. 50% Natives European Natives 1989-91 Immigrants 1992-94 Immigrants

40%

30%

20%

10%

0% 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 year

Fig. 3. Share of multigenerational households (including a parent of head/ spouse, or a grandchild aged 15þ with/without parent). For sample and group deﬁnitions see Table 1. Source: CBS Labor Force Surveys. Data was available to us through Israel Social Science Data Center (ISDC) at the Hebrew University.

454

Sarit Cohen-Goldner

50% 1989-91 Immigrants 1992-94 Immigrants

40%

30%

20%

10%

0% 2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

time in Israel

Fig. 4. Share of multigenerational households (including a parent of head/ spouse, or a grandchild aged 15þ with/without parent), 1992–2007. For sample and group deﬁnitions see Table 1. Years 1995–2007 for 1992–1994 immigrants. Source: CBS Labor Force Surveys. Data was available to us through Israel Social Science Data Center (ISDC) at the Hebrew University. who arrived during 1992–1994 the share of multigenerational households declined from 24% 1995 to 14.4 in 2007.5 In Figure 4 we present the share of heads of multigenerational households among immigrants from the two cohorts as a function of time since migration. As one can see, there is a great similarity between the two cohorts both in the level and trend of this share. Yet, the prevalence of multigenerational households among the later cohort (1992–1994) is usually smaller, conditional on time since migration. Table 1 presents summary statistics of heads of households by nativity. For example, the average age of a native head is 47.44, while the average age of a European native head is 54.77. Immigrant heads (mainly from the ﬁrst cohort) are on average more educated than native and European native heads, and the share of divorced heads among immigrants, from both cohorts, is larger than that among natives and European natives. The share of female heads is quite similar for natives, European natives, and the 1989–1991 cohort and stands on about a third. The share of female heads among 1992–1994 immigrant households is almost 40%. In addition, the share of married immigrants among 5

Note that we follow FSU immigrants who arrived in Israel during 1989–1991 or 1992–1994. Thus, in 1992 (1995), immigrants from the 1989–1991 (1992–1994) cohort have lived 1–3 years in Israel while in 2007 they had been living 16–18 (13–15) years in Israel.

455

Household Structure of Recent Immigrants to Israel

Table 1.

Summary statistics – head of household, 1992–2007

Variable

Nativesa

European 1989–1991 1992–1994 nativesb Immigrants Immigrantsc

Age

47.44 (18.84)

54.77 (18.81)

Females

11.95 (4.37) 32.88%

12.91 (4.31) 35.82%

48.69 (16.59) 39.53 (16.73) 10.17 (4.56) 13.75 (3.48) 33.40%

Marital status Married Single Divorced Widowed Household includes children under 4 Household includes children under 14

58.77% 22.06% 6.93% 12.24% 18.40% 35.30%

58.79% 14.73% 7.17% 19.31% 11.91% 26.81%

64.70% 12.34% 12.50% 10.46% 12.07% 33.22%

50.23% 24.53% 14.77% 10.47% 12.82% 31.60%

Labor market status Employed 62.37% Unemployed 2.62% Out of the labor force 35.01% Nonmarried parent with children under 4 0.88% Nonmarried parent with children under 14 3.35% Multigenerational householdd 5.65% Number of observations 507,525

58.76% 1.61% 39.62% 0.50% 2.23% 3.35% 207,045

67.81% 3.97% 28.23% 1.20% 4.99% 18.69% 39,417

60.46% 3.25% 36.29% 2.21% 7.83% 17.99% 19,176

Age on arrival Time in Israel Schooling

44.34 (18.47) 36.32 (18.72) 9.03 (3.74) 12.92 (3.56) 40.69%

Notes: Dropped observations: Households with no head or with more than one head, heads who are aged under 14, with schooling over 30 or with missing schooling, with unknown residence and immigrants with negative age on arrival. Standard deviation in parentheses. Source: CBS Labor Force Surveys. Data was available to us through Israel Social Science Data Center (ISDC) at the Hebrew University. a Born in Israel or immigrated prior to 1989, excluding Arabs and non-Jews. b Born in Israel and father born in Europe/America/Oceania/FSU, or immigrated prior to 1989 and born in Europe/America/Oceania /FSU, excluding Arabs and non-Jews. c Years 1995–2007. d Including a parent of head/spouse, or a grandchild aged 15þ with/without parent.

the 1992–1994 cohort is substantially lower than the share among the 1989–1991 cohort, mainly due to a high share of singles, but also due to a higher share of divorced. Nonetheless, the share of households with children aged 0–14 in the two cohorts of immigrants (33.22% and 33%, respectively) are very close to the share among natives (35.3%). The combination of all the above features results in a high share of singleparent (mainly single-mother) households among immigrants, in general, and among the 1992–1994 cohort, in particular. The share of single-parent households with children aged 0–14 stands at 7.96% among 1992–1994

456

Sarit Cohen-Goldner

Table 2.

Share of multigenerational households, 1992–2007a

Sample

1989–1991 1992–1994 Nativesb European (%) nativesc (%) Immigrants (%) Immigrantsd (%)

Full sample Tel Aviv Haifa Jerusalem Married Nonmarried (single, divorced, widowed) Employed Nonemployed (unemployed, out of the labor force) Years of schooling 0–12 Years of schooling 13þ

5.65 5.31 4.27 4.07 2.02 10.83

3.35 2.93 2.63 2.55 1.87 5.45

18.69 18.47 17.69 15.50 18.97 18.18

17.99 16.23 13.18 11.55 18.59 17.38

7.40 2.75

4.66 1.48

24.15 7.20

25.31 6.80

6.37 4.45

3.56 3.11

15.60 20.30

15.62 20.10

Notes: Dropped observations: Households with no head or with more than one head, heads who are aged under 14, with schooling over 30 or with missing schooling, with unknown residence and immigrants with negative age on arrival. Source: CBS Labor Force Surveys. Data was available to us through Israel Social Science Data Center (ISDC) at the Hebrew University. a Including a parent of head/spouse, or a grandchild aged 15þ with/without parent. b Born in Israel or immigrated prior to 1989, excluding Arabs and non-Jews. c Born in Israel and father born in Europe/America/Oceania/FSU, or immigrated prior to 1989 and born in Europe/America/Oceania /FSU, excluding Arabs and non-Jews. d Years 1995–2007.

cohort and 5% among 1989–1991 cohort. For comparison, this share among natives (European natives) is 3.35% (2.23%). The average share of multigenerational households over 1992–2007 among FSU immigrants from the two cohorts is above 18% in both cohorts. Table 2 presents the share of multigenerational households among the two native groups and two cohorts of immigrants by various characteristics of the head. Overall, the share does not vary much with marital status and locality. However, it varies with labor market attainment and level of schooling of the head. In particular, it is considerably higher among employed heads and among more educated heads. In Table 3 we report results from two Logit regressions for the likelihood of immigrants to be a head of multigenerational household. These regressions include immigrants from only the two cohorts mentioned above. In the ﬁrst regression we do not allow for any differences between the two cohorts, while in the second regression we allow immigrant heads who arrived in 1992–1994 to differ from those who arrived in 1989–1991, both in the baseline propensity to be a head of multigenerational household and in the effect of time spent in Israel on this likelihood. In both speciﬁcations, the propensity to be a head of a multigenerational household declines with the head’s age at arrival and

457

Household Structure of Recent Immigrants to Israel

Table 3.

Logit regressions – headsa

Variable

Spec. 1

Spec. 2

Constant

2.1464* (0.1565)

0.0026* (0.0012) 0.0258* (0.0037) 0.0759* (0.0277) 0.7373* (0.0438) 0.3717* (0.0482) 0.8955* (0.0728) 0.0437 (0.032) 0.0069 (0.0565) 0.3131* (0.0566) 0.411* (0.0478) 0.4126* (0.0565) 0.1802* (0.0334) 0.079* (0.0296) 1.9643* (0.0435) 1.9535* (0.0637) 0.1272* (0.0174) 0.0112** (0.0067) 0.3151* (0.0832) 1.0006* (0.3127)

Number of observations Log likelihood

58,593 25,062.37

58,593 25,052.88

Age on arrival Schooling Female Married Divorced Widowed Household includes children under 14 Single parent (nonmarried with children under 14) Tel Aviv Haifa Jerusalem North Center Employed Unemployed Time in Israel

*

0.0025 (0.0012) 0.0259* (0.0037) 0.0753* (0.0277) 0.738* (0.0438) 0.3737* (0.0482) 0.8983* (0.0728) 0.0433 (0.032) 0.0081 (0.0565) 0.3136* (0.0566) 0.4051* (0.0478) 0.4136* (0.0565) 0.1804* (0.0334) 0.0771* (0.0296) 1.9677* (0.0435) 1.9554* (0.0636) 0.0619* (0.0085)

Time in Israel for 1992–1994 Immigrants 1992–1994 Immigrants

Notes: Dropped observations: Households with no head or with more than one head, heads who are aged under 14, with schooling over 30 or with missing schooling, with unknown residence and immigrants with negative age on arrival. Omitted categories: Single, south, out of the labor force. Regression also includes year dummies for 1992–2006. Standard errors appear in parentheses. Source: CBS Labor Force Surveys. Data was available to us through Israel Social Science Data Center (ISDC) at the Hebrew University. * Signiﬁcant at 5% level. **Signiﬁcant at 10% level. a Years 1995–2007 for 1992–1994 immigrants.

458

Sarit Cohen-Goldner

education. The likelihood of a female immigrant to be a head of a multigenerational household is higher than that of a male immigrant. Singles are more likely to be a head of a multigenerational household than widowed, married, and divorced immigrants. Children aged 0–14 and single parents do not signiﬁcantly affect the propensity of immigrants to be a head of a multigenerational household, while locality does. The highest propensity to live in a multigenerational household is found in the center of Israel, not including Tel Aviv. Immigrant heads who are out of the labor force are more likely to live in a multigenerational household, but there is no signiﬁcant difference in the likelihood of employed and unemployed immigrant heads to live in such a household. Though at ﬁrst glance this result is somewhat surprising, one must keep in mind that the head of the household is deﬁned as the main earner and that less than 4% of immigrant heads are unemployed (Table 1 columns 3 and 4). The propensity of immigrant heads to live in a multigenerational household declines signiﬁcantly with time spent in Israel, as could also be seen in Figure 4, and the year dummies are not signiﬁcant. Adding a dummy for the 1992–1994 cohort and a dummy of the 1992–1994 cohort interacted with time in Israel (Table 3, speciﬁcation 2) do not change the quantitative effect of age, education, marital status, presence of children, labor market status and locality on the propensity to be a head of a multigenerational household. However, immigrants who arrived in the 1992–1994 cohort have a signiﬁcantly lower propensity to live in a multigenerational household upon arrival than those who arrived in the initial wave of 1989–1991. In addition, the (negative) coefﬁcient of time in Israel is now doubled (0.127 in the 2nd speciﬁcation compared to 0.062 in the 1st speciﬁcation) and the interaction term of 1992–1994 dummy with time spent living in Israel is positive and signiﬁcant. To illustrate the evolution of multigenerational households among the two cohorts, we present in Figure 5 the predicted share of multigenerational households based on the 2nd speciﬁcation.6 The ﬁgure shows that within cohort the differences between comparable male and female are quite small and that overall there are almost no differences in the likelihood of comparable immigrants from the two cohorts to live in a multigenerational household.7 So far we studied the effect of the head’s characteristics on the prevalence of multigenerational households among immigrants. Now we turn to study the prevalence of multigenerational households among the 6

These predictions are calculated for a married (male/female) immigrant with children under 14, who is employed, lives in the center of Israel, has 13 years of schooling, and whose age at arrival was 40. 7 The slight increase in the graphs for immigrants from the ﬁrst cohort is due to the effect of year dummies included in the regression.

459

Household Structure of Recent Immigrants to Israel 50%

Male who arrived to Israel in 1991 Female who arrived to Israel in 1991 Male who arrived to Israel in 1994 Female who arrived to Israel in 1994

40%

30%

20%

10%

0% 2

3

4

5

6

7

8

9

10

11

12

13

14

time in Israel

Fig. 5. Multigenerational household predicted probability – heads. Predictions for a married immigrant with children under 14, who is employed, lives in the center, has 13 years of schooling and arrived to Israel at age 40. Source: Authors’ calculations based on coefﬁcients from Table 3. Data was available to us through Israel Social Science Data Center (ISDC) at the Hebrew University. elders. Figure 6 presents the share of elder people aged 60 or above living in multigenerational household by nativity. This share is quite constant for natives and European natives and stands, on average, at 10.8% and 6%, respectively. Among immigrants from both cohorts, on the other hand, the share is substantially higher and declines over time, though not as fast as among heads. In particular, in 1992 (1995), 46% (39%) of elder immigrants who arrived during 1989–1991 (1992–1994) lived in a multigenerational household and this share declined to 22% (29%) in 2007. These high shares of multigenerational households among elder immigrants may indicate speciﬁc difﬁculties this population faces in the new country.8 Table 4 presents the characteristics of individuals aged 60þ by nativity. The average age of this elder group is 71–72 for both natives and immigrants. Among the elder individuals who belong to the 1989–1991 cohort, the average age at arrival in Israel was 62, while among the 1992–1994 cohort the average age at arrival was 64. Elder immigrants are 8 Lewin-Epstein and Semyonov (2008) document a low share of home ownership among FSU immigrants aged 50þ (30% among new immigrants relative to 80% among natives). In addition, they ﬁnd that a high share of these immigrants have a mortgage, indicating that elder immigrants face substantial difﬁculties in the housing market.

460

Sarit Cohen-Goldner

50% Natives European Natives 1989-91 Immigrants 1992-94 Immigrants

40%

30%

20%

10%

0% 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 year

Fig. 6. Share of individuals aged 60þ in multigenerational households (including a parent of head/spouse, or a grandchild aged 15þ with/without. For sample and group deﬁnitions see Table 4. Source: CBS Labor Force Surveys. Data was available to us through Israel Social Science Data Center (ISDC) at the Hebrew University. slightly more educated than elder European natives and are considerably more educated than the elder native population. The share of females in the elder population of immigrants is 60–61% in both cohorts, while among elder natives it is 55% and among elder European natives it is 57%. The share of divorced among the two immigrant groups is also higher than among the two native groups. The main differences between elder natives and elder immigrants is found with respect to labor market status, the presence of children in the household, and the interaction between single elderly and children in the household. Speciﬁcally, more than 85% of the elder immigrants do not participate in the labor market, relative to 80–81% among natives and European natives. In addition, the share of elder immigrants (from both cohorts) who live in a household with children is considerably higher than the share among natives. This is true both when considering young children aged 0–4 and children aged 0–14. Furthermore, of these elder immigrants who live in households with children, the majority are not married.9 For example, 17.78% of the elder immigrants who came in 1992–1994 lived in a household with children aged 0–14. Almost 58% of these immigrants (that is, 10.29/17.78) are not married.

9

Not married includes single, divorced, and widowed.

461

Household Structure of Recent Immigrants to Israel

Table 4.

Summary statistics – individuals aged 60þ, 1992–2007

Variable

Nativesa

European 1989–1991 1992–1994 nativesb Immigrants Immigrantsc

Age

71.5 (7.77)

72.91 (7.84)

Females

9.6 (5.03) 55.20%

Marital status Married Single Divorced Widowed Household includes children under 4 Household includes children under 14

11.08 (4.38) 57.12%

71.08 (7.19) 62.17 (8.09) 9.91 (4.62) 12.38 (4.22) 60.01%

71.24 (7.29) 63.72 (7.89) 8.51 (3.8) 11.97 (4.53) 61.59%

61.04% 3.27% 4.17% 31.52% 1.29% 3.09%

58.20% 2.82% 3.95% 35.03% 0.55% 1.61%

57.57% 2.08% 6.83% 33.52% 4.05% 15.13%

48.05% 3.92% 9.60% 38.44% 5.06% 16.60%

Labor market status Employed 17.94% Unemployed 0.67% Out of the labor force 81.40% Nonmarried parent with children under 4 0.48% Nonmarried parent with children under 14 1.17% Multigenerational householdd 10.79% Number of observations 228,543

18.84% 0.44% 80.72% 0.23% 0.78% 6.11% 133,847

12.11% 1.08% 86.81% 1.79% 7.71% 33.33% 23,652

8.89% 0.70% 90.41% 3.21% 10.01% 34.60% 11,212

Age on arrival Time in Israel Schooling

Notes: Dropped observations: Individuals who are aged under 14, with schooling over 30 or with missing schooling, with unknown residence and immigrants with negative age on arrival. Standard deviation in parentheses. Source: CBS Labor Force Surveys. Data was available to us through Israel Social Science Data Center (ISDC) at the Hebrew University. a Born in Israel or immigrated prior to 1989, excluding Arabs and non-Jews. b Born in Israel and father born in Europe/America/Oceania/FSU, or immigrated prior to 1989 and born in Europe/America/Oceania /FSU, excluding Arabs and non-Jews. c Years 1995–2007. d Including a parent of head/spouse, or a grandchild aged 15þ with/without parent.

Table 5 reports the results from two Logit regressions for the likelihood of elder immigrants aged 60þ to live in multigenerational household. Like in the analysis for heads, we use two speciﬁcations. The ﬁrst does not allow for cohort effects to affect the formation of a multigenerational household while the second does. In both speciﬁcations the propensity of elder immigrants to live in a multigenerational household increases with age at arrival and declines with education. The likelihood of an elder female immigrant to live in a multigenerational household is higher than that of a male immigrant; and widowed elders are more likely to live in

462

Sarit Cohen-Goldner

Table 5.

Logit regressions – individuals aged 60þa

Variable

Spec. 1

Spec. 2

Constant

4.7349* (0.2707)

0.0372* (0.0022) 0.0076* (0.0033) 0.2459* (0.0309) 1.0104* (0.1214) 1.6543* (0.1274) 1.9365* (0.1203) 3.4431* (0.062) 0.9848* (0.0844) 0.2433* (0.0739) 0.0258 (0.0556) 0.0635 (0.0636) 0.1198* (0.0398) 0.1509* (0.0362) 0.1589* (0.0544) 0.1108 (0.1606) 0.1836* (0.0208) 0.0407* (0.0079) 0.904* (0.0968) 1.8647* (0.4157)

Number of observations Log likelihood

34,864 17,122.59

34,864 17,072.40

Age on arrival Schooling Female Married Divorced Widowed Household includes children under 14 Single parent (nonmarried with children under 14) Tel Aviv Haifa Jerusalem North Center Employed Unemployed Time in Israel

*

0.0366 (0.0022) 0.0071* (0.0033) 0.2533* (0.0308) 1.0293* (0.1212) 1.6624* (0.1272) 1.9384* (0.1201) 3.4351* (0.0619) 0.9723* (0.0842) 0.2486* (0.0737) 0.0308 (0.0555) 0.0708 (0.0635) 0.1174* (0.0398) 0.1427* (0.0362) 0.1601* (0.0544) 0.1204 (0.1605) 0.0163 (0.0103)

Time in Israel for 1992–1994 immigrants 1992–1994 Immigrants

Notes: Dropped observations: Households with no head or with more than one head, heads who are aged under 14, with schooling over 30 or with missing schooling, with unknown residence and immigrants with negative age on arrival. Omitted categories: single, south, out of the labor force. Regression also includes year dummies for 1992–2006. Standard errors appear in parentheses. Source: CBS Labor Force Surveys. Data was available to us through Israel Social Science Data Center (ISDC) at the Hebrew University. * Signiﬁcant at 5% level. **Signiﬁcant at 10% level. a Years 1995–2007 for 1992–1994 immigrants.

463

Household Structure of Recent Immigrants to Israel

a multigenerational household in comparison to single, married, and divorced immigrants. Children aged 0–14 and the interaction between nonmarried elderly and number of children aged 0–14 signiﬁcantly affect the propensity of elder immigrants to live in a multigenerational household. In particular, the coefﬁcient of children is positive while the coefﬁcient of the interaction term is negative and smaller, implying that the presence of children increases this propensity both for married and nonmarried elders, in comparison to no children. Elder immigrants who are out of the labor force are more likely to live in a multigenerational household than those who are out of the labor force, and employed immigrants have a lower propensity to live in such an extended household in comparison to the unemployed. According to the 1st speciﬁcation, time spent in Israel does not have a signiﬁcant effect on the propensity of elder immigrants to live in a multigenerational household. Adding a dummy for the 1992–1994 cohort and a dummy for the 1992–1994 cohort interacted with time in Israel (Table 5, speciﬁcation 2) show that this propensity does decline with duration in Israel for both cohorts, and that elder immigrants from the second cohort have a signiﬁcantly lower propensity to live in a multigenerational household upon arrival than those who arrived in the initial wave of 1989–1991. The evolution of multigenerational households among elder immigrants with duration in Israel is illustrated in Figure 7, which presents the 100% 90% 80% Male who arrived to Israel in 1991 with children under 14

70%

Female who arrived to Israel in 1991 with children under 14 Male who arrived to Israel in 1994 with children under 14

60%

Female who arrived to Israel in 1994 with children under 14 Male who arrived to Israel in 1991 without children under 14

50%

Female who arrived to Israel in 1991 without children under 14

40%

Male who arrived to Israel in 1994 without children under 14 Female who arrived to Israel in 1994 without children under 14

30% 20% 10% 0% 2

3

4

5

6

7

8

9

10

11

12

13

14

time in Israel

Fig. 7. Multigenerational household predicted probability – individuals aged 60þ with and without children under 14. Predictions for a married immigrant, who is out of the labor force, lives in the center, has 13 years of schooling, and arrived to Israel at age 60. Source: Calculations based on coefﬁcients from Table 5.

464

Sarit Cohen-Goldner

predicted probability of different representative immigrants to live in a multigenerational household based on the 2nd speciﬁcation.10 The main striking result is that conditional on personal characteristics, the probability of elder immigrants to live in a multigenerational household does not decline with duration in Israel. This result is derived from the effect of the variable time in Israel as well as the time (year) dummies, and is obtained for immigrants from the two cohorts.11 While there are no apparent differences in the probability of males and females from the same cohort to live in a multigenerational household and almost no differences between immigrants from the two cohorts, there are very signiﬁcant differences in this probability with respect to presence of children ages 0–14. Speciﬁcally, the likelihood of an elder immigrant to live in a multigenerational household is four times higher if there are children aged 0–14 in the household. 3. Conclusions This chapter studied the prevalence of multigenerational households among FSU immigrants to Israel. The formation of multigenerational households and other forms of extended households is a known phenomenon among immigrants and may reﬂect cultural preferences toward co-residing as well as economic constraints upon arrival in the new country. In this chapter we studied the prevalence of multigenerational households among FUS immigrants to Israel. We distinguished between younger and elder immigrants and between immigrants from two cohorts. The analysis shows that the share of multigenerational households declines with duration in Israel among young immigrants, but not so much among elder immigrants who arrived at older age. This difference may reﬂect better economic integration of younger immigrants in the local labor market and lower attachment of younger immigrants to cultural habits that existed in the origin country. In order to detangle the role of social forces from the role of economic conditions, we used two cohorts of FSU immigrants. The ﬁrst cohort that arrived in 1989–1991 did not have a developed social network of previous FSU immigrants, while the second cohort of 1992–1994 had the social infrastructure developed by the ﬁrst cohort. We ﬁnd great similarity in the 10

These predictions are calculated for a married (male/female) immigrant who is out of the labor force, lives in the center of Israel, has 13 years of schooling, and whose age at arrival was 60. We present the predictions for this immigrant both in the case where there are no children in the household and in the presence of children aged 0–14 in the household. 11 We also ran the two regressions without year dummies and found that the coefﬁcient of time in Israel is very small and insigniﬁcant. In the ﬁrst speciﬁcation without year dummies, this coefﬁcient was 0.0045 and its standard error was 0.0035, while in the 2nd speciﬁcation it was 0.0141 and the standard error was 0.004.

Household Structure of Recent Immigrants to Israel

465

prevalence of multigenerational households between the two cohorts, suggesting that immigrants, presumably, do not form a multigenerational household in Israel in order to provide themselves with a social anchor, but rather to help themselves overcome economic constraints upon arrival. Acknowledgments I thank Tali Larom for excellent research assistance. References Angel, R., Tienda, M. (1982), Determinants of extended household structure: cultural pattern or economic need? The American Journal of Sociology 87, 1360–1383. Burr, J.A., Mutchler, J.E. (1993), Ethnic living arrangements: cultural convergence or cultural manifestation? Social Forces 72, 169–179. Cohen-Goldner, S., Eckstein, Z. (2008), Labor mobility of immigrants: training, experience, language and opportunities. International Economic Review 49 (3), 837–872. Cohen-Goldner, S., Eckstein, Z. (2010), Estimating the return to training and occupational experience: the case of female immigrants. Journal of Econometrics 156 (1), 86–105. Eckstein, Z., Weiss, Y. (2004), On the wage growth of immigrants: Israel 1990–2000. Journal of the European Economic Association 2, 665–695. Glick, J.E., Van Hook, J. (2002), Parents’ coresidence with adult children: can immigration explain racial and ethnic variation? Journal of Marriage and Family 64 (1), 240–253. Glick, J.E., Van Hook, J. (2007), Immigration and living arrangements: moving beyond economic need versus acculturation. Demography 44 (2), 225–249. Katz, R., Lowenstein, A. (1999), Adjustment of older Soviet immigrant parents and their adult children residing in shared households: an intergenerational comparison. Family Relations 48 (1), 43–50. Lewin-Epstein, N., Semyonov, M. (2008), Home ownership and living conditions among Israelis 50 and older. Social Security – Journal of Welfare and Social Security Studies 76, 153–174, (Hebrew). Strosberg, N., Naon, D. (1997), The absorption of elderly immigrants from the former Soviet Union: selected ﬁndings regarding housing, social integration, and health. Gerontology 79, 5–15, (Hebrew).

CHAPTER 20

Circular Migration or Permanent Return: What Determines Different Forms of Migration? Florin Vadeana,b and Matloob Pirachab a

Centre for Economic and International Studies, University of Rome Tor Vergata, Rome, Italy E-mail address: [email protected] b School of Economics, University of Kent, Canterbury, Kent, UK E-mail address: [email protected]

Abstract This chapter addresses the following questions: To what extent do the socio-economic characteristics of circular/repeat migrants differ from the migrants who return permanently to the home country after their ﬁrst trip (i.e. return migrants)? And, what determines each of these distinctive temporary migration forms? Using Albanian household survey data and both a multinomial logit model and a maximum simulated likelihood (MSL) probit with two sequential selection equations, we ﬁnd that education, gender, age, geographical location and the return reasons from the ﬁrst migration trip signiﬁcantly affect the choice of migration form. Compared to return migrants, circular migrants are more likely to be male, have primary education and originate from rural, less developed areas. Moreover, return migration seems to be determined by family reasons, a failed migration attempt but also by the fulﬁlment of a savings target. Keywords: Circular migration, return migration, sample selection JEL classiﬁcation: C35, F22, J61 1. Introduction The last two decades have seen a signiﬁcant increase in temporary migration as compared to the more ‘‘traditional’’ long-term/permanent migration which had been prevalent before 1990s. For instance, in 2006 alone nearly 2.5 million individuals were admitted into the OECD countries on temporary contracts, which is over three times the number of legally admitted permanent migrants (OECD, 2008). Most of the temporary migration is repeat or circular in nature (i.e. the repeated back and forth movements between the home country and one or more countries of Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008026

r 2010 by Emerald Group Publishing Limited. All rights reserved

468

Florin Vadean and Matloob Piracha

destination) but since there is no systematic tracking of migrants’ movements, it is often quite difﬁcult to estimate its magnitude.1 With recent migration programmes aiming to encourage short-term contracts not only in the EU but also in other industrialised countries, temporary labour movement is likely to increase even further, especially repeat migration, as the new programmes introduce ‘‘assurances’’ of re-employment upon return to the host country after spending some time in the home country.2 Furthermore, host countries have recognised the necessity to remove certain rules applying for long-term foreign residents that prevent them from returning temporarily to their home countries.3 Given the policy emphasis on circular/repeat nature of temporary migration, it is important to understand the different dimensions of these movements and the characteristics and correlates linked to the varied temporary migration forms. Circular migration is frequently linked to expectations of mutual gains for migrant-sending and -receiving countries and migrants and their families. The general idea is that circularity of skilled workers would allow industrialised countries to ﬁll labour market gaps with the simultaneous compensation of possible ‘‘brain drain’’ in developing migrant-sending countries, through transfers of know-how and technology. Moreover, circular migration at all skill levels should have a positive effect on sustained remittance ﬂows; these private money transfers being often perceived to make an important contribution to poverty alleviation and investment opportunities in the home country. While the socio-economic motivations and determinants of temporary migration have been extensively analysed in the literature (e.g. Djajic and Milbourne, 1988; Stark, 1991; Dustmann, 1995, 1997, 2003; Borjas and Bratsberg, 1996; Mesnard, 2004), most studies focused mainly on the decision of migrants to return to the home country and the amount of time spent abroad, irrespective of the form of temporary migration.4 1

One exception is Constant and Zimmermann (2007) who found using the German SocioEconomic Panel data that more than 60 per cent of the guest workers exited and re-entered Germany at least once between 1984 and 1994. It is, however, difﬁcult to tell whether the guest workers who left actually returned to their countries of origin or spent some time in a third country. 2 For example, France introduced a new type of permit in 2006, targeted at seasonal workers, allowing them to hold a job for less than 6 months during 3 consecutive years, provided they maintain their residence outside France. 3 The European Commission, for instance, is considering amendments to the directive on the status of long-term residents (Directive 109/2003) to allow migrants to return to their home countries for more than 12 months without putting their rights at risk (OECD, 2008). 4 There are a few exceptions. Massey and Espinosa (1997) analyse the re-migration decision of return migrants in Mexico but without taking into account the possible sample selection bias (i.e. return migrants may be a non-random selected group of the total population). Constant and Zimmermann (2007) study the topic from the host country perspective. They analyse the frequency of exits and the amount of time spent outside Germany by guest workers who entered the country before 1984.

Circular Migration or Permanent Return

469

The increased interest in circular migration gives rise, however, to questions about the differences in socio-economic characteristics between circular/repeat migrants and migrants who return permanently to the home country (usually after the ﬁrst trip) and the determinants of these distinctive temporary migration forms. Assessing them could be fundamental in understanding the way in which migration can be more effectively managed for the beneﬁt of both sending and receiving countries. We attempt to ﬁll this gap in the literature by analysing the correlates and determinants of different forms of temporary migration in a systematic way. First, using a multinomial logit (MNL) model, we analyse the choice of individuals from four alternatives: no migration, long-term/ permanent migration, return migration and circular migration.5 Then, using a maximum simulated likelihood (MSL) probit model with two sequential selection equations, we investigate the probability of returnees to re-migrate after their ﬁrst trip, by controlling for sample selection bias into initial migration and return migration. Along with the socio-economic and regional characteristics, we also take into consideration the effect of own migration history (e.g. past migration movements, legal vs. illegal residence, success in ﬁnding work and return reasons) on the re-migration intentions of returnees as previous experience is assumed to strongly affect subsequent migration decisions. Our main research questions are as follows: To what extent do the socio-economic characteristics of circular/ repeat migrants differ from the migrants who return permanently to the home country after their ﬁrst trip? And, what determines each of these distinctive temporary migration forms? We aim to answer these questions using data from the 2005 Albanian Living Standard Measurement Survey (ALSMS). This dataset contains a rich set of information on the past trips of return migrants as well as information on the non-migrant, migrant and temporary migrant population groups, allowing us to conduct a reasonable analysis of the selfselection of individuals into different migration forms.6 To our knowledge, this is the ﬁrst study to analyse circular migration in the context of the European East-West migration experience. Our results show that education, gender, age, geographical location and the return reasons from the ﬁrst migration trip signiﬁcantly affect the choice of migration form. Compared to return migrants, circular migrants are more likely to be male, have only primary education and originate

5

In our analysis, return migration refers to permanent return to the home country after a single migration episode, whereas circular migration refers to multiple (two or more) trips, that is, repeat or seasonal migration. Temporary migration includes both migration forms. 6 Datasets from migrant sending countries usually have information only on non-migrants and return migrants, but not on the characteristics of migrants who are abroad, whereas migrant host country data lack information on the characteristics of the population from which immigrants are selected (i.e. the non-migrants).

470

Florin Vadean and Matloob Piracha

from rural, less developed areas. Moreover, permanent return after the ﬁrst trip seems to be determined by family reasons, a failed migration attempt but also the fulﬁlment of a savings target. The results also conﬁrm the hypothesis that return migration accentuates the type of selection that generated the initial migration ﬂow (see Borjas and Bratsberg, 1996). Moreover, circular migration seems to occur along the same pattern, with circular migrants being signiﬁcantly less educated compared to permanent returnees. The remainder of this chapter is organised as follows. The next section presents a general framework for analysis. Some background information and stylised facts on the different forms of Albanian migration are presented in Section 3. Section 4 presents the econometric speciﬁcation, while Section 5 discusses the empirical results of the multivariate analysis of the determinants of migration forms. The last section concludes this chapter.

2. Framework for analysis Inherent in the concept of temporary migration is the decision to return to the home country after spending a period of time in the host country. However, the idea of return migration is at odds with the perceived notion of migration which is seen as a strategic choice by individuals to move from a low-wage, high-unemployment region/country to the one which has relatively higher wages and employment rates. Since agents make a lifetime, utility maximising decision based on perceived net beneﬁts from migration, migrants should intuitively remain abroad until retirement. However, many recent papers have explored the possibility of return migration before the end of the individual’s active life cycle (i.e. retirement) and despite persistent income differences between the home and host countries. Arguments used for explaining the decision to return are, for example, location-speciﬁc preferences (i.e. higher utility for consumption at home), differences in purchasing power between the host and home country currencies, higher returns at home to the human capital accumulated in the host country or higher returns at home to the capital accumulated abroad in the presence of capital constraints (e.g. Djajic and Milbourne, 1988; Dustmann 1995, 1997, 2003; Mesnard, 2004). Alternatively, return may occur due to a revision of the initial migration decision. For example, a migrant may return as a result of failure in achieving an initial migration target (i.e. does not ﬁnd job or ﬁnds a job only at a lower wage than expected; Borjas and Bratsberg, 1996) or because of ranking higher in the income distribution in the home reference group compared to the reference group in the host country (i.e. relative deprivation; Stark, 1991).

Circular Migration or Permanent Return

471

Long-term/permanent migration Circular/repeat/seasonal migration Return migration (i.e. permanent return after the first trip) Stay put

Decision Tree 1.

Return and re-migration integral to the initial migration decision. Stay abroad (i.e. long-term/permanent migration)

Migrate

Re-migrate (i.e. circular/repeat migration) Return

Stay put

Decision Tree 2.

Settle permanently back (i.e. return migration)

Multiple revisions of the migration decision.

The empirical analysis conducted in this chapter is based on two decision frameworks. On the one hand, as in Hill (1987), the choice of circular migration can be considered integral to the initial migration decision, that is, made before the migrant leaves the home country (see Decision Tree 1). Given higher income opportunities abroad and preference for living in the country of origin, individual utility is assumed to depend on a time path of residence in the home and host country and is maximised by choosing the optimal amount of time spent abroad as well as the frequency of trips. On the other hand, the decision process can be, for example, altered by the presence of uncertainty or imperfect information about the prospects in the destination country (and, while abroad, about the prospects in the home country). In this setup, a migrant decides while abroad, based on the realities he faces, whether he should return or not.7 However, once back home, there is another layer in the decision process regarding re-migration, perhaps due to problems of re-integration, the failure to ﬁnd a suitable job or having to acquire more capital for the business started after return. In this case, the decision process would be of the form given in Decision Tree 2. Another complexity of the migration process comes from the character of the migration decision: is it a choice or an outcome? If we consider return as endogenous then the migrant decides about the form of migration, the duration of stay abroad and the frequency of trips (Radu and Epstein, 2007). Temporary migration might, however, be induced exogenously by host country policies as well. In recent years, there has 7

Note that, for the purpose of our analysis, long-term and permanent migration is treated in the same way. Based on this, we will use the two words interchangeably throughout the text.

472

Florin Vadean and Matloob Piracha

been a proliferation of immigrant employment schemes in industrial countries for sectors with jobs avoided by natives, with strong seasonal ﬂuctuations (e.g. farming, road repairs and construction), and in the service industry (e.g. hotels and restaurants). These employment schemes offer a variety of pre- and post-admission conditions and incentives, designed to keep ﬂows temporary (Dayton-Johnson et al., 2007). Nevertheless, migrants do have the option among different immigration regimes, for example, those which are more open to permanent migration (i.e. US, Canada, Australia and New Zealand), those with temporary migration programmes (i.e. West European countries and the Gulf States) and/or those that are more lax with respect to immigration offences (i.e. irregular migration, overstaying of temporary residence permits; e.g. South European countries). Therefore, in the majority of cases, the form of migration can be assumed to be a choice.

3. Background and data Existing estimates suggest that since 1990 over a million Albanians (i.e. about 30 per cent of the population) have either settled or worked for short time periods abroad, which is by far the highest proportion amongst the Central and East European countries (ETF, 2007; Vullnetari, 2007). Own estimates based on data from the 2005 ALSMS led to similar ﬁgures. Using direct information on the migration history of the individuals surveyed and indirect information on the present migration status and migration history of the spouses and children living outside the household and the siblings of the household head and spouse, we found that in 2005 about 24.6 per cent of the Albanian population aged 15 to 64 was either currently migrant (16.5 per cent) or had a past migration experience (8.1 per cent). In addition, part of the migrants living abroad at the time of the survey will also return, and hence the asserted proportion of one-third temporary migrants should be seen as a lower bound. The main reason for migration is for employment purposes. The collapse of the industrial sector in the early transition years and the absence of a welfare state have pushed many workers outside the labour market and into poverty. By 2004, around 30 per cent of Albanians were estimated to live below the poverty line; half of them in extreme poverty, subsisting on less than US$1 per day (Barjaba, 2004). In face of these harsh realities, many have sought employment abroad, mainly in neighbouring EU countries. Because of their geographical proximity, the main destination countries are Greece and Italy, hosting almost 80 per cent of Albania’s migrants in 2005. About 600,000 worked and/or lived in Greece, about 250,000 in Italy, while another approximately 250,000 were scattered among industrialised countries in Western Europe and North America

Circular Migration or Permanent Return

473

(Vullnetari, 2007). The sector of employment and, thus, the form of migration is varying signiﬁcantly among destinations: seasonal employment in construction, farming and tourism in Greece; temporary employment in manufacturing, construction and services in Italy and predominantly permanent migration of skilled migrants to Western Europe, the US and Canada (Barjaba, 2004; ETF, 2007). The data used for the empirical analysis comes from the 2005 ALSMS, collected by the Albanian Institute for Statistics (INSTAT) with technical support from the World Bank. The dataset is based on a survey of 3,640 households (17,302 individuals) and contains a detailed module on migration.8 We drew the information on migrants from two parts of the migration module. The ﬁrst is on the migration history of the household members present (e.g. country of last migration episode, year of migration, time spent abroad, legal residence abroad, legal work abroad, reasons for returning to Albania and previous migration episodes since turning 15). The second part provides detailed information on the spouse and/or children who are currently abroad, and we added these absent household members to the sample. Since the focus of this chapter is the analysis of the determinants of labour migration movements, we restricted our sample to individuals in the potential labour force (i.e. not enrolled in education, not a housewife/ husband, not retired, not handicapped and not in military service) aged between 20 and 60. Moreover, in order to select the permanent migrants from our second group, we excluded all migrants who were abroad at the time of the survey for 3 years or less (i.e. 539 observations). For the purpose of this analysis, our deﬁnition for a permanent migrant is, hence, an individual who has spent 37 months or more abroad since the last time he/she left the country.9 Given the above screening and after excluding all observations with missing values for the variables included, our sample contains 7,280 individuals: of which 4,756 (65.3 per cent) are non-migrants, 1,430 (19.6 per cent) permanent migrants, 536 (7.4 per cent) return migrants (i.e. individuals who migrated only once and were back in Albania at the time of the survey) and 558 (7.7 per cent) circular migrants (i.e. individuals who migrated more than once in the past and were back in Albania at the time of the survey). Group mean values of the data described above show that Albanian migration, and in particular temporary migration, is predominantly male (see Table 1). Females represent 35 per cent of the permanent migrants,

8

A migrant is deﬁned as a person who migrated abroad for at least 1 month, for non-visits purposes, since turning age 15. 9 Percentile statistics show that 90 per cent of the temporary migrants returned to Albania after spending a maximum of 3 years abroad during their ﬁrst migration episode.

3.571 3.818 12.363 4.859 1.953

0.529 0.250 0.286 0.288 0.176 30,886.23 6.920

Household characteristics HH subjective economic status in 1990 HH subjective economic status in 2005 Log of HH income HH size Number of friends

Community and regional characteristics Urban area Region: Coastal Region: Central Region: Mountain Region: Tirana Average wage at district level (LEK) Number of migrants in community (PSU)

Migration history (ﬁrst migration trip) Age at ﬁrst migration trip Length of ﬁrst migration trip

0.522 39.422 0.485 0.389 0.126 0.050 0.057 0.009 0.799

Mean value

Non-migrants

25.126 92.081

0.566 0.037** 0.165*** 0.415 0.011 0.276 0.138*** 0.150 0.016 0.160 297.60** 30,588.63 3.715*** 10.635

3.476 0.095* 0.200*** 4.018 *** 0.408 11.956 1.681*** 3.178 0.224*** 1.729

0.171*** 0.350 6.623*** 32.799 0.040*** 0.445 0.070*** 0.459 0.030*** 0.096 0.042*** 0.092 0.066*** 0.123 0.051*** 0.059 0.165*** 0.634

Difference Mean value

Permanent migrants Mean value

Return migrants

4.270*** 29.396 70.012*** 22.069

0.011 0.576 0.098*** 0.317 0.010 0.285 0.050*** 0.200 0.198 0.038** 607.68*** 31,196.31 1.822*** 8.813

0.171 3.647 0.038 4.056 *** 0.497 12.452 1.618*** 4.797 0.426*** 2.155

0.268*** 0.082 4.492*** 37.291 0.027 0.418 0.035 0.494 0.008 0.088 0.034** 0.058 ** 0.037 0.086 0.011 0.071 0.165*** 0.799

Difference

Descriptive statistics by form of migration

Individual characteristics Gender (female ¼ 1) Age Education level: primary Education level: secondary Education level: tertiary Speaks English (1990) Speaks Italian (1990) Speaks Greek (1990) Married

Table 1.

3.210 3.762 12.031 5.151 1.833

0.014 35.547 0.557 0.400 0.043 0.020 0.034 0.065 0.806

Mean value

2.919*** 12.610***

26.477 9.459

0.204*** 0.373 0.045 0.272 * 0.333 0.048 0.121*** 0.321 0.124** 0.073 1,743.90*** 29,452.41 0.545** 9.358

0.438*** 0.294*** 0.421*** 0.354*** 0.322***

0.068*** 1.744*** 0.139*** 0.095*** 0.045*** 0.038*** 0.052*** 0.006 0.008

Difference

Circular migrants

474 Florin Vadean and Matloob Piracha

1,430

0.899 0.899 0.160 0.748 0.092 0.277 0.071 0.005 0.555 0.007 0.015 0.071 0.411 0.379 0.210

0.535*** 0.535*** 0.071*** 0.399*** 0.469*** 0.091*** 0.037*** 0.053*** 0.465*** 0.006 0.021*** 0.332*** 0.337*** 0.213*** 0.124*** 0.364 0.364 0.090 0.349 0.562 0.368 0.034 0.058 0.090 0.013 0.035 0.403 0.748 0.166 0.086 31.235 0.160 0.416 0.424 0.216 0.459 0.106 0.218 0.192 0.646 0.162 536

0.125*** 0.181*** 0.029* 0.050* 0.078*** 0.120*** 0.030*** 0.015 0.075*** 0.006 0.025*** 0.031 0.132*** 0.100*** 0.032** 3.970*** 0.073*** 0.021 0.094*** 0.095*** 0.046 0.146*** 0.098*** 0.351*** 0.362*** 0.012 0.238 0.545 0.061 0.299 0.640 0.487 0.004 0.043 0.014 0.007 0.011 0.434 0.880 0.066 0.054 27.265 0.233 0.437 0.330 0.122 0.505 0.253 0.120 0.543 0.283 0.174 558

Notes: The sample included is the potential labour force (i.e. not enrolled in education, not a housewife/husband, not retired, not handicapped and not in military service) aged 20 to 60. HH (household)-subjective economic status: 1 ¼ poor to 10 ¼ rich. The differences are computed between the mean values in the adjoining columns. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

Legal residence during ﬁrst migration trip Legal residence during last migration trip Work during ﬁrst migration trip: no Work during ﬁrst migration trip: legally Work during ﬁrst migration trip: illegally Married: no Married w/o children: migrated with spouse Married w/o children: spouse in Albania Married w/children: migrated with spouse and children Married w/children: migrated with children, spouse in Albania Married w/children: migrated with spouse, children in Albania Married w/children: spouse and children in Albania Migrated to Greece Migrated to Italy Migrated to other country Age at ﬁrst return Occupational choice (2005): not working Occupational choice (2005): wage employment Occupational choice (2005): self-employment Return reason: family/non- economic Return reason: unsuccessful Return reason: temporary/seasonal permit Return reason: accumulated enough savings Re-migration intention: yes Re-migration intention: no Re-migration intention: don’t know Observations 4,756

Circular Migration or Permanent Return 475

476

Florin Vadean and Matloob Piracha

but only 8.2 per cent of the return migrants and just 1.4 per cent of the circular migrant group. Migrants in all groups are on average younger compared to nonmigrants. In order for migration to be ﬁnancially rewarding (i.e. additional income from employment abroad to exceed the migration costs), it has to take place early in the active lifetime. Taking into account that migration costs are highest if resettling permanently to another country, it is not surprising that permanent migrants are on average the youngest at the time of migration with an average age of 25.1 compared to 29.4 in the case of return migrants. Regarding the educational composition of the different groups, permanent and return migrants have the highest secondary education rate: 45.9 and 49.4 per cent, respectively, compared to 38.9 per cent for nonmigrants. It is likely that many secondary-educated workers lost their jobs during the early transition period as uncompetitive state-owned factories were put into administration or were closed. Hence, most of them used migration as a strategy to improve their standard of living. Moreover, 55.7 per cent of circular migrants have at most primary education (which probably explains also why they are on average younger at their ﬁrst migration trip than the return migrants). The majority of them are probably small (subsistence) farmers who supplement their income through seasonal work abroad. Tertiary-educated workers are least likely to migrate mostly because of relatively better job opportunities for this group in Albania. With 12.6 per cent, the tertiary education rate of non-migrants is about 3 percentage points higher compared to permanent and return migrants and 8.3 percentage points higher compared to circular migrants. Migrants were signiﬁcantly more likely to have spoken at least one foreign language in 1990, with the form of migration being related to the language of the destination countries. The main destination country for circular migrants has been Greece (88.0 per cent); for return migrants Greece (74.8 per cent) and Italy (16.6 per cent), while many permanent migrants have also settled, besides Greece (41.1 per cent) and Italy (37.9 per cent), in other West European or North American countries (21.0 per cent). In terms of marital status, permanent migrants had the lowest marriage rate in 2005. Nevertheless, at the time they left the country, they had the highest marriage rate (72.3 per cent) compared to the other migrant groups (63.2 per cent for return and 51.3 per cent for circular migrants). Migrating for longer periods without the spouse imposes, in many cases, considerable strain on the relationship of a couple, often leading to separation and divorce. However, the savings accumulated abroad made it easier for temporary (i.e. return and circular) migrants to start up a family after return. Temporary migrants were signiﬁcantly more likely to have children at the time of their ﬁrst migration, but they were less likely to migrate with them. Return migration seems to be more common among members of relatively richer households. Many in this group are target savers

Circular Migration or Permanent Return

477

originating from middle or upper middle class families who, through migration and investment of the repatriated savings after return, signiﬁcantly improved their economic situation above the Albanian average (see Piracha and Vadean, 2010). Compared to permanent migrants, they might also have decided to return permanently back because of their relatively better social and economic position in Albania (Stark and Taylor, 1991). Contrarily, circular migrants are members of poorer and relatively larger families. Permanent migrants originate from households with less social connections, which probably means they had lower social and emotional relocation costs. However, they left from communities that have more individuals as current or past migrants. As found in other studies, that could be evidence of the fact that migrant networks and/or the culture of migration in the community are important for the migration decision (see Azzarri and Carletto, 2009). Geographically, most permanent and return migrants are from urban areas (56.6 and 57.6 per cent, respectively), while circular migrants originate from rural areas (62.7 per cent) and regions closer to Greece (i.e. the Central and the Mountain regions).10 Regarding the migration history, circular migrants were least likely to have legal residence during their ﬁrst migration trip (only 23.8 per cent of them) but that increased considerably in time to 54.5 per cent for the last migration trip. This is certainly due to the large legalisation programs in Greece and Italy after 1999. As for return migrants, they are also quite likely to have migrated illegally: only 36.4 per cent of them had legal residence abroad. Borjas and Bratsberg (1996) argued that the failure of a migrant to obtain legal residence while abroad might determine his decision to return back permanently. Nevertheless, if a migrant does intend to return to his home country but does not intend to migrate again in the future, he is certainly more likely to overstay a work or tourist visa in order to fulﬁl, for example, his savings target. With paid employment being the main reason for temporary migration, return and circular migrants were signiﬁcantly more likely to work while abroad compared to permanent migrants. Nevertheless, they were also considerably more likely to work illegally. The reason for returning differs notably between the forms of temporary migration. While the majority of return migrants moved back because of failing their migration target (45.9 per cent; i.e. have not found work, have not obtained legal residence or have been deported) or after having accumulated enough savings (21.8 per cent), 25.3 per cent of the circular migrants have returned from the ﬁrst trip because of the expiry of

10

Using 2002 ALSMS data, Carletto et al. (2006) show similar geographical patterns of permanent and temporary migration.

478

Florin Vadean and Matloob Piracha

a seasonal/temporary work permit (compared to only 10.6 per cent in the case of return migrants). Finally, there seems to be quite a strong state dependency in circular migration: in 2005, 54.3 per cent of the individuals who migrated repeatedly in the past (i.e. circular migrants) intend to migrate again during the next 12 months. In contrast, only 19.2 per cent of the return migrants expressed their intention to re-migrate. 4. Econometric speciﬁcation The migration decision processes described in Section 2 lead to alternative econometric models. If assuming a single utility maximisation migration decision over the lifetime (i.e. Decision Tree 1 in Section 2), the form of migration may be determined by a pairwise comparison of the indirect utilities of the given alternatives: no migration : permanent migration :

U N 4U P ; U N 4U R ; U N 4U C ; U P 4U N ; U P 4U R ; U P 4U C ;

return migration :

U R 4U N ; U R 4U P ; U R 4U C ; U C 4U N ; U C 4U P ; U C 4U R ;

circular migration :

(1)

where N, P, R and C stand for no migration, permanent migration, return migration and circular migration, respectively. The unordered choice settings can be motivated by a random utility model (Greene, 2002). For the ith individual faced with k ¼ fN; P; R; Cg choices, the utility of choice j is given by U ij ¼ bj xi þ ij

(2)

where U ij is the indirect utility of choice j for individual i, xi a vector of characteristics which affect the choice of the migration form and bj a vector of choice-speciﬁc parameters. Assumptions about the disturbances (eij) determine the nature of the model and the properties of its estimator. We assume that eij are independent and identically distributed with type I extreme value distribution, which leads to the MNL model (McFadden, 1974; Greene, 2002). The probability of choosing alternative j is speciﬁed as follows: Prðyi ¼ jÞ ¼ P

ebj xi bk xi k¼N;P;R;C e

(3)

Not all bj in Equation (3) are identiﬁed and we normalise by setting bN ¼ 0. The dynamics among the possible choices in the estimation results of the MNL model are illustrated by computing odds ratios. The factor change in the odds of outcome m versus outcome n for a marginal increase in xk and

Circular Migration or Permanent Return

479

the other independent variables in the model held constant is given by Omjn ðx; xk;mjn þ 1Þ ¼ ebk;mjn Omjn ðx; xk;mjn Þ

(4)

The limit of analysing the determinants of the migration form in the framework of a MNL model is that one can control only for variables observed for all alternatives. One problem arising from that is the difﬁculty in some cases to infer the direction of causality. Many of the individuals’ socio-economic characteristics observed for all population groups (e.g. age, marital status, household size or household income) are collected at the time of survey (i.e. in 2005). However, for migrants, these might have been different at the time of their ﬁrst migration episode, their return or the subsequent migration trips. Therefore, some of the observed socioeconomic characteristics may, in fact, be determined by the migration experience and the form of migration chosen. In addition, the MNL model does not allow to control the effect of a previous migration experience (e.g. found work while abroad for the ﬁrst time, legal residence while abroad or reason for returning) on the decision to re-migrate since non-migrants have no such experience. Nevertheless, if we assume that the individual revises his initial migration decision after each migration step (Decision Tree 2 in Section 2), the migration experience should signiﬁcantly inﬂuence future migration movements. Running separate regressions only for migrants will give biased and inconsistent results as migrants might be a non-randomly selected group. A more consistent model is a probit with two sequential self-selection equations: the ﬁrst equation controls for selection into migration while the second – including only migrants – for the selection into return. This model can be estimated stepwise (i.e. the inverse Mill’s ratio – IMR – of the ﬁrst selection probit is introduced as a covariate in the second selection equation and the IMR from the second selection equation is then used as a covariate in the outcome probit) or by maximum likelihood. Relative to the maximum likelihood approach, the stepwise method is often perceived to give inconsistent results (Lahiri and Song, 2000). In particular, this is the case when there is strong multicollinearity between covariates in the outcome equation and the selection controls (i.e. covariates of the selection equations). If there are no overlapping covariates in the outcome and selection equations, then multicollinearity can be assumed insigniﬁcant (see Nawata and Nagase, 1996; Stolzenberg and Relles, 1997). The equations for the probit model with two sequential selections have the following form for each observation:11 Migrant : M ¼ W 0 b þ m; where M ¼ IðM 40Þ 11

(5)

Temporary migration includes circular migration and return migration (i.e. permanent return after the ﬁrst trip).

480

Florin Vadean and Matloob Piracha

Temporary migrant : T ¼ Y 0 d þ t; where T ¼ IðT 40Þ if and missing otherwise

(6)

Circular migrant: C ¼ Z0 y þ c; where C ¼ IðC 40Þ if T ¼ 1 ðand M ¼ 1Þ and missing otherwise.

(7)

The variables denoted by asterisks are the latent outcomes, and those without are binary indicators summarising the observed outcomes. I(.) is the indicator function equal to one if its argument is true, and zero otherwise. We assume the error terms ðm; t; cÞ N 3 ð0; VÞ, where V is a symmetric matrix with typical element rkl ¼ rlk for k; l 2 fm; t; cg and kal and rkk ¼ 1 for all k. The errors in each equation are assumed to be orthogonal to the predictors (elements of the vectors W, Y and Z, respectively). We deﬁne a set of signs variables kt ¼ 2t 1 for t 2 fM; T; Cg. The likelihood contribution for a temporary migrant, that is, with M ¼ 1 and T ¼ 1 is L3 ¼ F3 ðkM W 0 b; kT Y 0 d; kC Z0 y; kM kT rmt ; kM kC rmc ; kT kC rtc Þ,

(8)

the likelihood contribution for a permanent migrant (i.e. M ¼ 1 and T ¼ 1) is L2 ¼ F2 ðkM W 0 b; kT Y 0 d; kM kT rmt Þ,

(9)

while the likelihood contribution for a non-migrant (i.e. M ¼ 0) is L1 ¼ F1 ðkM W 0 bÞ

(10)

It follows that the log-likelihood contribution to be calculated by the evaluator function for each observation is ln L ¼ ð1 MÞ ln L1 þ Mð1 TÞ ln L2 þ MR ln L3

(11)

In order to avoid multicollinearity due to overlapping covariates in the outcome and selection equations, the model is estimated using MSL in Stata. We evaluate multivariate standard normal probabilities with 200 random draws using mvnp( ), a function developed by Cappellari and Jenkins (2006), based on the Geweke–Hajivassiliou–Keane (GHK) smooth recursive conditioning simulator. For the maximisation, we used Stata’s modiﬁed Newton–Raphson algorithm (see Gould et al., 2003).12 5. Empirical results Despite the limits of the MNL model discussed in the previous section, it offers a good starting point for the analysis. The estimation results give 12

We thank Lorenzo Cappellari and Stephen Jenkins for advice on the Stata programming.

Circular Migration or Permanent Return

481

information on variables that affect similarly the choice of all migration forms and variables that only affect the choice of particular forms of migration. Thus, besides theoretical arguments, the estimation results can be used as additional justiﬁcation for the selection instruments used in the probit model with two sequential equations. The estimation results of the MNL model are given in Table 2 and the respective factor changes in odds in Table 3.13 We note that being a female decreases signiﬁcantly the likelihood of being a migrant, in particular a circular migrant. Moreover, as predicted by various migration models and conﬁrmed by empirical ﬁndings, age decreases the odds of all forms of migration versus non-migration. In particular, permanent migration seems to be a decision taken at a younger age (a marginal increase in age decreases the odds of permanent migration vs. non-migration by a factor of 0.90) as social and ﬁnancial relocation costs are lower and the larger time span until the end of the active lifetime allows for higher gains from migration (Radu and Epstein, 2007). The education level has quite a differentiated effect on migration. Having a tertiary education negatively affects the likelihood of being a permanent or circular migrant versus non-migrant, while secondary education decreases the odds of being a return versus non-migrant and increases the odds of being circular versus return migrant. In order to analyse more in-depth the self-selection process into the different types of migration with respect to education, we will include the education level variables in all three equations of the three-variate probit model. A ﬁrst possible candidate as a selection control into migration seems to be the proﬁciency in Greek in 1990 since it positively affects all three forms of migration similarly. A further possible control variable for the selection into migration is the number of migrants in the community. Its positive effect on all migration forms eventually conﬁrms that the culture of migration often inﬂuences the decision to migrate. However, its signiﬁcant negative effect on return and circular versus permanent migration calls for additional tests for the inclusion in the ﬁrst selection equation only (i.e. the decision to migrate or not) or in the second selection equation as well (i.e. if a migrant, the decision to return after the ﬁrst trip or not). Finally, originating from an urban area and/or an area with a higher average wage level seems to positively affect the return decision compared to settling permanently abroad and negatively affect the decision to migrate again after the ﬁrst trip (i.e. circular migration).

13

The Small–Hsiao test for independence of irrelevant alternatives (IIA) holds for all subsets. Furthermore, the likelihood ratio tests for combining alternatives show that no pair of alternatives should be collapsed. Test results are available from the authors upon request.

482

Table 2.

Florin Vadean and Matloob Piracha

Multinomial logit estimation of choice among migration forms

Individual characteristics Gender (female ¼ 1) Age Education level: secondary Education level: tertiary Speaks English (in 1990) Speaks Italian (in 1990) Speaks Greek (in 1990) Married

Permanent migrant vs. non-migrant

Return Circular migrant migrant vs. vs. non-migrant non-migrant

1.16001 [0.13634]*** 0.10814 [0.00729]*** 0.15244 [0.10275] 0.68525 [0.24132]*** 0.40394 [0.31354] 0.50185 [0.32705] 1.72834 [0.34696]*** 0.53196 [0.19997]***

2.96162 [0.19971]*** 0.0647 [0.00533]*** 0.20663 [0.08121]** 0.44185 [0.29003] 0.02481 [0.23567] 0.47912 [0.28007]* 2.03414 [0.18069]*** 1.07557 [0.15940]***

Household (HH) characteristics HH subjective economic status in 1990 0.04367 [0.03793] HH size 0.77753 [0.02711]*** Number of friends 0.02129 [0.02819] Regional characteristics Number of migrants in the community Urban area Log of average wage (district level) Constant Observations Pseudo-R-sq

0.19938 [0.00951]*** 0.16214 [0.10524] 0.40163 [0.23697]* 8.34078 [2.35848]*** 7,280 0.29

4.98761 [0.42093]*** 0.09308 [0.00714]*** 0.01752 [0.08146] 0.57404 [0.28535]** 0.19694 [0.34043] 0.16864 [0.45261] 2.10866 [0.51933]*** 1.60809 [0.20682]***

0.01181 0.02296 [0.02589] [0.04327] 0.06617 0.02224 [0.02489]*** [0.02562] 0.07319 0.03991 [0.02030]*** [0.05393] 0.14632 0.15929 [0.01840]*** [0.02095]*** 0.27318 0.12512 [0.09816]*** [0.11110] 0.64509 2.59168 [0.34892]* [1.49510]* 7.75753 26.64024 [3.61103]** [15.51550]*

Notes: Robust standard errors in brackets, adjusted for 12 clusters (i.e. counties). HH subjective economic status 1990: 1 ¼ poor to 10 ¼ rich. The control group for the education level is ‘‘Primary or less’’. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

483

Circular Migration or Permanent Return

Table 3.

P vs. N R vs. N R vs. P C vs. N C vs. P C vs. R

P vs. N R vs. N R vs. P C vs. N C vs. P C vs. R

Odds ratios for choice among migration forms

Gender

Age

Education Education Speaks level: level: English secondary tertiary (1990)

0.31*** 0.05*** 0.17*** 0.01*** 0.02*** 0.13***

0.90*** 0.94*** 1.04*** 0.91*** 1.02 0.97***

1.16 1.23** 1.06 1.02 0.87 0.83**

Married

HH HH size subjective economic status 1990

1.70*** 2.93*** 1.72*** 4.99*** 2.93*** 1.70*

0.96 1.01 1.06 0.98 1.02 0.97

0.46*** 0.94*** 2.04*** 0.98 2.13*** 1.04

Speaks Italian (1990)

Speaks Greek (1990)

1.65 1.61* 0.98 1.18 0.72 0.73

5.63*** 7.65*** 1.36 8.24*** 1.46 1.08

0.50*** 0.64 1.28 0.56** 1.12 0.88

1.50 1.03 0.68 0.82 0.55 0.80

No. of friends

No. of Urban migrants in area community

Log of average wages (district)

0.98 1.08*** 1.10*** 0.96 0.98 0.89*

1.22*** 1.16*** 0.95*** 1.17*** 0.96** 1.01

0.67* 1.91* 2.85*** 0.07* 0.11 0.04*

1.18 1.31*** 1.12 0.88 0.75* 0.67***

Notes: Odds ratios computed based on the estimation in Table 2. HH subjective economic status 1990: 1 ¼ poor to 10 ¼ rich. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

The variables chosen to describe the selection into migration (ﬁrst equation in Table 4)14 are three language variables (i.e. speaking English, Italian and Greek in 1990), the household subjective economic situation in 1990 and the number of migrants in the community. Since speaking the language of the destination country decreases the costs of migration (e.g. makes it easier to access information about opportunities on foreign labour markets and to ﬁnd a job), language proﬁciency in 1990 should positively affect the likelihood of migration. Nevertheless, only a small number of Albanians spoke a foreign language in 1990 and many migrants learned the language of the host country while abroad. The likelihood of returning and then re-migrating is, hence, less likely to be affected by the language proﬁciency before migration took place. This is also conﬁrmed by the results of the MNL estimation: the odds of return versus permanent migration and of circular versus return migration are insigniﬁcant for proﬁciency in all three languages. 14

Standard errors were adjusted for cluster sampling in the 12 Albanian counties: Berat, Dibe¨r, Durre¨s, Elbasan, Fier, Gjirokaste¨r, Korc- e¨, Kuke¨s, Lezhe¨, Shkode¨r, Tirana and Vlore¨.

484

Table 4.

Florin Vadean and Matloob Piracha

MSL three-variate probit with two selections of the decision to migrate circularly

Migration equation Gender (female ¼ 1) Education level: secondary Education level: tertiary Spoke English in 1990 Spoke Italian in 1990 Spoke Greek in 1990 Economic situation in 1990 Number of migrants in the community Constant

Temporary migration equation Gender (female ¼ 1) Age at 1st migration trip Education level: secondary Education level: tertiary Urban location Log of average wage (district level) Months remained away (1st trip) Obtained legal residence (1st trip) Worked abroad during 1st trip: legally Worked abroad during 1st trip: illegally Married w/o children: migrated with spouse Married w/o children: spouse in Albania Migrated with spouse and children

0.91973 [0.10259]*** 0.12118 [0.03889]*** 0.318 [0.07497]*** 0.30884 [0.02519]*** 0.47507 [0.04518]*** 1.0658 [0.07640]*** 0.01951 [0.01560] 0.10232 [0.00595]*** 0.93551 [0.17973]***

Circular migration equation Gender (female ¼ 1) Age after 1st migration trip Education level: secondary Education level: tertiary Married Occupation choice: wage employment Occupation choice: selfemployment HH size Number of friends

Economic situation in 2005 0.38552 [0.11062]*** 0.01221 [0.00699]* 0.31984 [0.09872]*** 0.78719 [0.14547]*** 0.26868 [0.08172]*** 1.5512 [1.72603] 0.03081 [0.00449]*** 0.64919 [0.10551]*** 0.09404 [0.14898] 0.43439 [0.13281]*** 0.03396 [0.27415] 0.63277 [0.13695]*** 0.70885 [0.14655]***

Urban location Log of average wage (district level) Months remained away (ﬁrst trip) Country of destination (1st trip): Greece Country of destination (1st trip): Italy Return reason: family/ non-economic Return reason: unsuccessful Return reason: accumulated enough savings Constant

Cross-equation correlations r21 r31 r32

0.9797 [0.29066]*** 0.04237 [0.00402]*** 0.12562 [0.06262]** 0.01866 [0.26178] 0.54729 [0.12475]*** 0.29507 [0.12391]** 0.62165 [0.16893]*** 0.02171 [0.01518] 0.11309 [0.03504]*** 0.05299 [0.04534] 0.27238 [0.09893]*** 1.86451 [0.79664]** 0.01378 [0.00212]*** 0.28767 [0.28719] 0.3151 [0.34798] 0.61794 [0.23282]*** 0.52414 [0.19882]*** 0.56279 [0.19982]*** 21.39878 [8.39567]**

0.28083 [0.12835]** 0.27005 [0.15023]* 0.187141 [0.17342]

485

Circular Migration or Permanent Return

Table 4. (Continued ) Migrated with children, spouse in Albania Migrated with spouse, children in Albania Spouse and children in Albania Country of destination (1st trip): Greece Country of destination (1st trip): Italy Constant

0.03286 [0.54167] 0.35861 [0.12125]*** 0.2946 [0.14297]** 1.26645 [0.15638]*** 0.22255 [0.18163] 15.3382 [17.69421]

Total number of observations Number of migrants Number of temporary migrants Number of circular migrants Log of pseudo-likelihood

7,280 2,524 1,094 558 4894.81

Notes: Robust standard errors in brackets, adjusted for 12 clusters (i.e. counties). HH subjective economic status: 1 ¼ poor to 10 ¼ rich. The control group for the education level is ‘‘Primary or less’’; for working abroad during 1st migration trip is ‘‘No’’, for the family structure is ‘‘Single’’ and for the countries of destination is ‘‘Other’’; for occupational choice is ‘‘Not working’’ and for the return reasons is ‘‘Seasonal/temporary migration’’. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

Individuals from poorer households should have had stronger incentives to migrate after 1990 in order to improve their situation. Therefore, the household subjective economic situation in 1990 is expected to negatively affect the probability to migrate. Finally, by decreasing migration costs through network effects, the number of migrants in the community should positively affect the migration decision. Nonetheless, the speciﬁc migration form could be eventually inﬂuenced by the preponderance of migrants of a particular form in the community (i.e. herd effect) but not by the aggregate migration. Most selection instruments are signiﬁcant and have the expected signs (see Table 4). From the three languages considered, speaking at least some Greek in 1990 has the strongest effect on migration. The common border of about 282 km and a shared culture and history made Greece the most important destination. Temporary migration was probably mainly encouraged by the relatively low cost of crossing the Greek border (in particular illegally) during the 1990s, while permanent migration was mainly fuelled by the large exodus at the beginning of the 1990s of ethnic Greeks living in the Southern part of Albania, who were rapidly nationalised in Greece (see Barjaba, 2004). Speaking Italian or English had a positive effect on being a migrant as well but to a lesser extent. This is not surprising because of the relatively greater distance and, thus, higher ﬁnancial migration cost to Italy, Western Europe and North America, compared to Greece.

486

Florin Vadean and Matloob Piracha

The household’s subjective economic situation in 1990 has a negative effect on being a migrant, though not signiﬁcant. It seems, therefore, that migration is used as a strategy to improve the standard of living by individuals across social strata. Finally, the number of migrants in the community is positively and signiﬁcantly correlated with the probability of initial migration. This would conﬁrm the social capital hypothesis and previous empirical ﬁndings, as for example Massey and Espinosa (1997), that the existence of a strong community migrant network proves essential for the reduction in the costs and risks of ﬁnding a good job abroad and, thus, the success of the migration project. For the selection into temporary migration (i.e. being return or circular vs. permanent migrant; the second equation in Table 4), we used instruments observed only for migrants. First, compared to settling permanently abroad, temporary migration should be positively affected by age at migration. As predicted by various migration models and conﬁrmed by empirical ﬁndings, permanent migration should be a decision taken at a younger age as social and ﬁnancial relocation costs are lower and the larger time span until the end of the active lifetime allows for higher gains (see, e.g., Radu and Epstein, 2007). Nevertheless, re-migration should be rather determined by age after return since even if migrated for the ﬁrst time at the same age, the age after return depends on the amount of time spent abroad. Further, having obtained legal residence should give migrants access to legal and better employment and, thus, increasing the probability of staying permanently abroad. Contrarily, ﬁnding no or only illegal employment should increase migration costs (e.g. forgone earnings) and/or income risk and, therefore, the probability to return as well. While the residence status variable is signiﬁcant and has the expected sign, only having worked illegally is signiﬁcantly and positively correlated with the likelihood of returning. Permanent migrants (compared to temporary) seem to either work legally or not participate in the labour market, giving evidence that besides better access to the labour market the legal residence status eventually gave migrants the opportunity to access the host countries’ social security system and stay (at least temporarily) outside the labour market. Finally, individuals who had migrated with close family members (i.e. spouse or children) should be less likely, while those who left close family members behind more likely, to return.15 The estimation results conﬁrm that compared to being single during the ﬁrst migration trip, married migrants without children were signiﬁcantly more likely to return 15

Since successful young migrants would be more likely to marry after return and start a family (i.e. have children), the decision to re-migrate after return would rather depend on the new family structure, and we have tried to capture that by the variables ‘‘marital status in 2005’’ and ‘‘household size in 2005’’ (see third equation in Table 4).

Circular Migration or Permanent Return

487

if they had a spouse back in Albania. However, the direction of causality is not straightforward: the spouse’s decision not to follow the partner abroad might have motivated the migrant to return; but likewise, the spouse’s decision not to migrate could have been inﬂuenced by the migrant’s choice to stay only temporarily abroad. Unsurprisingly, we ﬁnd that having migrated with both spouse and children strongly decreased the likelihood of returning to Albania, conﬁrming that permanent migrants are more likely to reunite with close family members in the host country (Faini, 2007). Nonetheless, having children back home is positively correlated to the decision to return, irrespective of having migrated with or without the spouse. A formal test for whether sample selection is ignorable is based on the null hypothesis that the cross-equation correlations are jointly different from zero. The test results show that the estimation results would have been biased and inconsistent, had we not corrected for selection.16 Mainly, the error terms of the ﬁrst and second equations are signiﬁcantly negatively correlated. This might be due to the unobserved preference for living in the home country that is hypothesised to decrease the likelihood of an initial migration, but, if having migrated, to increase the likelihood of returning to the home country (see Hill, 1987; Djajic and Milbourne, 1988; Dustmann, 1995, 1997, 2003). As expected from the descriptive statistics and the estimation results of the MNL model, being a female decreases signiﬁcantly the probability of being a migrant; if a migrant, the probability to have returned; and ﬁnally, the probability to have re-migrated, if having returned after the ﬁrst migration trip. Given the more traditional gender roles in the Albanian context, women are often in charge of taking care of children and household, while the men are the bread-earners (King et al., 2006). Therefore, it is not surprising that Albanian women often follow their husband in case he settles abroad, but are signiﬁcantly less likely to engage in temporary migration for employment purposes. The gender difference between return and circular migration can be further explained through the gender difference in terms of the type of jobs they engage in, with men taking jobs with a more seasonal character, for example, in construction, farming and tourism (ETF, 2007). Regarding the education level, our estimation results show that secondary education slightly increases the probability of initial migration, while tertiary education strongly decreases it. These conﬁrm the ﬁndings of de Coulon and Piracha (2005) that Albanian migration is not associated with higher educated individuals. They explain this by the fact that more educated individuals would face higher assimilation costs in the foreign labour markets (i.e. problems regarding recognition

16

Test results are available from the authors upon request.

488

Florin Vadean and Matloob Piracha

of diplomas or practising the profession in a foreign language), situation that mainly applies for such professions as medical doctors, lawyers or teachers. Moreover, as hypothesised by Borjas and Bratsberg (1996), we ﬁnd that return migration to Albania accentuates the selection type of the initial migration ﬂow. From the initial middle to low educated migrant population, those with the highest education return to Albania, leaving abroad a permanent migrant group with an even lower average education level. In the framework of Borjas and Bratsberg’s relative returns to skills hypothesis, lower skilled individuals would migrate if the returns to skills are relatively higher in the home compared to the destination country. Moreover, the most skilled in the migrant group being the marginal migrants would also be the ﬁrst to return because the human capital accumulated abroad would give them relatively higher earnings in the home compared to host labour market.17 Additionally, we observe that re-migration of returnees occurs along the same pattern, with the lowest educated from the return migrants engaging in repeat/circular migration, most certainly taking advantage of the relatively higher earnings abroad for their (lower) education level. As observed also from the coefﬁcients of the occupational choice variables, circular migrants are more likely to stay outside the labour market, probably because of the poor opportunities and/or low paid jobs available to them on the Albanian labour market. Only the better-educated returnees seem to settle permanently back, probably enjoying the returns from the human and/or ﬁnancial capital accumulated abroad in higher earning jobs and/or self-employment.18 Social relations have conﬂicting effects on the temporary migration decision. On the one hand, being married is signiﬁcantly and positively related to circular migration movements, giving probably evidence to the fact that a married couple can reduce income risk if one spouse works abroad. But as argued by Hill (1987), migrants seem to prefer to smooth the emotional cost of being parted from their loved ones by splitting the total amount of time spent abroad into several, shorter migration trips. On the other hand, the household size is rather unimportant in the decision process about the type of temporary migration. The re-migration decision is negatively related to the

17

They tested for their hypothesis by proxying the relative returns to skills by the income inequality in the US immigrants’ host countries. Albania’s Gini index was at every point between 1990 and 2005 below that of Greece and Italy (i.e. the main destination countries). However, considering arguments such as ‘‘more educated individuals face relatively higher assimilation costs in foreign labour markets’’ (de Coulon and Piracha 2005), the real returns to education (i.e. netted assimilation costs) could be indeed relatively higher in Albania. 18 For more on occupational choice of return migrants in Albania, see Piracha and Vadean (2010).

Circular Migration or Permanent Return

489

extra-household social capital (i.e. the number of friends), friends being eventually better placed compared to other household members (i.e. housewife and children) to provide information about job and business opportunities at home. The economic conditions and labour market opportunities in the region of origin seem to be an important determinant of the form of migration too. Individuals from rural areas are more prone to choose circular migration. Majority of them are most probably farmers, who add to small incomes from subsistence farming through seasonal work in Greece. Contrarily, migrants from urban areas and districts with higher average wages are more likely to return permanently to Albania as their chances of ﬁnding suitable jobs or to start up a business with the savings accumulated abroad are probably higher. Finally, the return reason has a strong and robust effect on the likelihood of having migrated repeatedly versus having settled permanently in Albania after the ﬁrst migration trip. Failing the migration target is a negative experience that not only determines return migration (Borjas and Bratsberg, 1996) but also seems to act as a deterrent for future migration movements as well. Similarly, everything else being equal, having accumulated enough savings during the ﬁrst migration trip has a strong negative effect on the probability of being a circular migrant. Target savers may have intended from the very beginning to return permanently back after the ﬁrst trip and start a business with the capital accumulated abroad, as argued by Mesnard (2004). Nevertheless, family reasons seem to be equally important in deterring further migration movements. As for circular migration, it seems to be a choice made before leaving the country for the ﬁrst time. Having returned from the ﬁrst trip because of the expiry of a temporary/seasonal work permit signiﬁcantly increases the likelihood of an additional migration episode. The MSL probit with double selection is run under three speciﬁcations of the dependent variable of the outcome equation. The ﬁrst (third equation in Table 4) considers repeat migration movements in the past versus having migrated only once. However, some of the returnees who have migrated only once (i.e. return migrants) may migrate again in the future and could be, in fact, circular migrants, even if we do not observe that. Assuming that individuals in this subgroup of return migrants have characteristics similar to circular migrants, our results could be biased. Therefore, in order to test the robustness of our results, in a second speciﬁcation (third equation in Table 5), we consider the return migrants who intend to re-migrate in the next 12 months as circular migrants as well, while in the third speciﬁcation (third equation in Table 6), they are excluded from the analysed sample. With the exception of the marital status, we ﬁnd all results discussed above to be quite robust.

490

Table 5.

Florin Vadean and Matloob Piracha

MSL three-variate probit with two selections of the decision to migrate circularly

Migration equation Gender (female ¼ 1) Education level: secondary Education level: tertiary Spoke English in 1990 Spoke Italian in 1990 Spoke Greek in 1990 Economic situation in 1990 Number of migrants in the community Constant

Temporary migration equation Gender (female ¼ 1) Age at 1st migration trip Education level: secondary Education level: tertiary Urban location Log of average wage (district level) Months remained away (1st trip) Obtained legal residence (1st trip) Worked abroad during 1st trip: legally Worked abroad during 1st trip: illegally Married w/o children: migrated with spouse Married w/o children: spouse in Albania Migrated with spouse and children Migrated with children, spouse in Albania Migrated with spouse, children in Albania

0.91992 [0.10273]*** 0.11977 [0.03930]*** 0.32307 [0.07595]*** 0.31756 [0.02664]*** 0.47612 [0.04458]*** 1.06505 [0.07700]*** 0.01929 [0.01567] 0.10223 [0.00606]*** 0.93489 [0.18040]***

Circular migration equation Gender (female ¼ 1) Age after 1st migration trip Education level: secondary Education level: tertiary Married Occupation choice: wage employment Occupation choice: selfemployment HH size Number of friends

Economic situation in 2005 0.39951 [0.11192]*** 0.01133 [0.00683]* 0.31844 [0.09832]*** 0.77434 [0.15305]*** 0.27346 [0.08621]*** 1.51915 [1.74863] 0.03096 [0.00449]*** 0.64698 [0.10974]*** 0.08541 [0.14589] 0.4264 [0.13198]*** 0.03055 [0.27377] 0.64162 [0.14515]*** 0.67536 [0.13330]*** 0.05391 [0.55243] 0.35283 [0.11355]***

Urban location Log of average wage (district level) Months remained away (ﬁrst trip) Country of destination (1st trip): Greece Country of destination (1st trip): Italy Return reason: family/noneconomic Return reason: unsuccessful Return reason: accumulated enough savings Constant

Cross-equation correlations r21 r31 r32

0.68564 [0.35213]* 0.03339 [0.00322]*** 0.13474 [0.07522]* 0.10377 [0.27096] 0.05373 [0.13475] 0.4843 [0.17152]*** 0.79218 [0.16110]*** 0.00479 [0.02729] 0.08811 [0.04283]** 0.00384 [0.03855] 0.34499 [0.09344]*** 0.79975 [0.67083] 0.00948 [0.00519]* 0.23958 [0.23667] 0.34379 [0.36970] 0.56235 [0.18798]*** 0.7112 [0.18118]*** 0.84397 [0.19065]*** 10.9935 [7.16252]

0.28681 [0.12131]** 0.03251 [0.13408] 0.22637 [0.22284]

Total number of observations 7,280 Number of migrants 2,524

491

Circular Migration or Permanent Return

Table 5. (Continued ) Spouse and children in Albania Country of destination (1st trip): Greece Country of destination (1st trip): Italy Constant

0.29359 [0.14468]** 1.26557 [0.15346]*** 0.21721 [0.18201] 14.9689 [17.95151]

Number of temporary migrants Number of circular migrants Log of pseudo-likelihood

1,094 661 4891.62

Notes: Robust standard errors in brackets, adjusted for 12 clusters (i.e. counties). Returnees who migrated only once but intend to re-migrate considered also as circular migrants. HH subjective economic status: 1 ¼ poor to 10 ¼ rich. The control group for the education level is ‘‘Primary or less’’; for working abroad during 1st migration trip is ‘‘No’’, for the family structure is ‘‘Single’’ and for the countries of destination is ‘‘Other’’; for occupational choice is ‘‘Not working’’ and for the return reasons is ‘‘Seasonal/temporary migration’’. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

6. Conclusions Theoretical and empirical evidence on the determinants of circular migration is still very limited, and this chapter is an attempt to ﬁll the literature gap. We think the results obtained in this chapter could be used as an aid in understanding the migration patterns and processes in order to design policies to more effectively manage migration for the beneﬁt of both sending and receiving countries. Although the analysis is conducted using Albanian household data, the results could be generalised to other developing migrant-sending countries as well, especially East European countries like Moldova, Bosnia and Herzegovina or Kosovo. The main objective of this chapter is to study the correlates and determinants of different forms of migration with a particular emphasis on circular migration. We chose Albania for our empirical analysis because it is a country of mass emigration and about one-third of its aggregate migration movements are temporary. Furthermore, as in other East European countries, Albanian temporary migration hides different realities: about 50 per cent of the temporary migrants are permanent returnees (i.e. have migrated abroad only once), while the others are circular/repeat migrants. Our empirical results show that the form of migration is determined by gender, age, the labour market returns to speciﬁc education levels, family ties, urban/rural origin and past migration experience. For example, women and tertiary educated are more likely to stay put in Albania. The amount of time spent abroad, legal residence and accompanying family are positively related to permanent migration, while age, secondary education,

492

Table 6.

Florin Vadean and Matloob Piracha

MSL three-variate probit with two selections of the decision to migrate circularly

Migration equation Gender (female ¼ 1) Education level: secondary Education level: tertiary Spoke English in 1990 Spoke Italian in 1990 Spoke Greek in 1990 Economic situation in 1990 Number of migrants in the community Constant

Temporary migration equation Gender (female ¼ 1) Age at 1st migration trip Education level: secondary Education level: tertiary Urban location Log of average wage (district level) Months remained away (1st trip) Obtained legal residence (1st trip) Worked abroad during 1st trip: legally Worked abroad during 1st trip: illegally Married w/o children: migrated with spouse Married w/o children: spouse in Albania Migrated with spouse and children Migrated with children, spouse in Albania Migrated with spouse, children in Albania

0.89881 [0.10181]*** 0.13534 [0.03642]*** 0.29402 [0.07165]*** 0.31384 [0.01982]*** 0.48033 [0.04357]*** 1.07779 [0.07839]*** 0.02055 [0.01397] 0.10418 [0.00556]*** 0.99128 [0.16645]***

Circular migration equation Gender (female ¼ 1) Age after 1st migration trip Education level: secondary Education level: tertiary Married Occupation choice: wage employment Occupation choice: selfemployment HH size Number of friends

Economic situation in 2005 0.45105 [0.13940]*** 0.00546 [0.00825] 0.28045 [0.08206]*** 0.76549 [0.15310]*** 0.35154 [0.09646]*** 1.49579 [1.67047] 0.034 [0.00383]*** 0.66411 [0.13050]*** 0.07271 [0.16837] 0.41373 [0.15935]*** 0.10059 [0.26654] 0.73376 [0.18248]*** 0.45655 [0.15951]*** 0.34431 [0.56918] 0.40121 [0.16767]**

Urban location Log of average wage (district level) Months remained away (ﬁrst trip) Country of destination (1st trip): Greece Country of destination (1st trip): Italy Return reason: family/ non-economic Return reason: unsuccessful Return reason: accumulated enough savings Constant

Cross-equation correlations r21 r31 r32

1.07557 [0.25392]*** 0.04302 [0.00479]*** 0.14049 [0.07297]* 0.05251 [0.29667] 0.36897 [0.12068]*** 0.50982 [0.17797]*** 0.8443 [0.17983]*** 0.0034 [0.02093] 0.12967 [0.04124]*** 0.02628 [0.03953] 0.30528 [0.09541]*** 1.58441 [0.87677]* 0.0171 [0.00299]*** 0.28406 [0.29841] 0.4294 [0.41280] 0.70411 [0.23335]*** 0.75954 [0.20599]*** 0.80683 [0.19300]*** 19.17225 [9.32517]**

0.34605 [0.12729]*** 0.14700 [0.14983] 0.32111 [0.21016]

Total number of observations 7,177 Number of migrants 2,421

493

Circular Migration or Permanent Return

Table 6. (Continued ) Spouse and children in Albania

0.39294 [0.15225]***

Country of destination (1st trip): Greece Country of destination (1st trip): Italy Constant

1.31392 [0.10982]*** 0.24129 [0.15164] 14.5516 [17.06961]

Number of temporary migrants Number of circular migrants Log of pseudo-likelihood

991 558 4654.30

Notes: Returnees who migrated only once but intend to re-migrate excluded from the sample. Robust standard errors in brackets, adjusted for 12 clusters (i.e. counties). HH subjective economic status: 1 ¼ poor to 10 ¼ rich. The control group for the education level is ‘‘Primary or less’’, for working abroad during 1st migration trip is ‘‘No’’, for the family structure is ‘‘Single’’ and for the countries of destination is ‘‘Other’’; for occupational choice is ‘‘Not working’’ and for the return reasons is ‘‘Seasonal/temporary migration’’. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

failed migration or fulﬁlment of a savings target determine permanent return after the ﬁrst trip. Being a male, having a lower education level, originating from a rural area and having a positive temporary migration experience in the past are factors affecting circular migration. Given that majority of the circular migrants are primary educated, their main contribution to development in Albania is probably through increasing the aggregate demand via remittances and repatriated savings. Nevertheless, development gains from transfers of skills and technology could probably be achieved through return migration. As shown by Piracha and Vadean (2010), many successful returnees start up their own businesses and become entrepreneurs after settling back to Albania. Probably the most notable result is the conﬁrmation of the hypothesis and empirical ﬁndings of Borjas and Bratsberg (1996) that return migration accentuates the type of selection – in our case negative selection – that generated the initial migration ﬂow. Additionally, our results provide evidence that re-migration of return migrants (i.e. circularity) occurs along the same pattern: circular migrants being signiﬁcantly less educated compared to migrants who return permanently to Albania after the ﬁrst trip. Given the limits of the data, it was not possible to capture all the important aspects of different migration forms. More research is needed on the selection patterns into circular migration. Of particular interest is the assessment of the possibility that in the case of relative lower returns to skills in the home country, individuals with higher skills/education are motivated to migrate circularly and contribute to the economies of both origin and destination countries, as often expected by policy makers.

494

Florin Vadean and Matloob Piracha

Acknowledgments We thank Artjoms Ivlevs, Nabanita Datta Gupta, Barry Reilly, two anonymous referees, and participants at the 8th Annual GEP Postgraduate Conference as well as at the 12th IZA European Summer School in Labour Economics for helpful comments. An earlier version was part of a report for the ‘‘Managing Labour Migration to Support Economic Growth’’ project coordinated by the OECD Development Centre, whose ﬁnancial support is gratefully acknowledged. The usual disclaimer applies.

References Azzarri, C., Carletto, C. (2009), Modelling migration dynamics in Albania: a hazard function approach. Policy Research Working Paper No. 4945, The World Bank, Washington, DC. Barjaba, K. (2004), Albania: Looking Beyond Borders. Migration Policy Institute, Mimeo, Washington, DC. Borjas, G., Bratsberg, B. (1996), Who leaves? The out-migration of the foreign-born. Review of Economics Statistics 78 (1), 165–176. Cappellari, L., Jenkins, S.P. (2006), Calculation of multivariate normal probabilities by simulation, with applications to maximum simulated likelihood estimation. Stata Journal 6 (2), 156–189. Carletto, C., Davis, B., Stampini, M., Zezza, A. (2006), A country on the move: international migration in post-communist Albania. International Migration Review 40 (4), 767–785. Constant, A., Zimmermann, K.F. (2007), Circular migration: counts of exits and years away from the host country. IZA Discussion Paper No. 2999, Institute for Study of Labor, Bonn. Dayton-Johnson, J., Katseli, L.T., Maniatis, G., Mu¨nz, R., Papademetriou, D. (2007), Gaining from Migration: Towards a New Mobility System. OECD Development Centre, Paris. de Coulon, A., Piracha, M. (2005), Self-selection and the performance of return migrants: the source country perspective. Journal of Population Economics 18 (4), 779–807. Djajic, S., Milbourne, R. (1988), A general equilibrium model of guest-worker migration: the source-country perspective. Journal of International Economics 25 (3–4), 335–351. Dustmann, C. (1995), Savings behavior of migrant workers: A life-cycle analysis. Zeitschrift fu¨r Wirtschafts- und Sozialwissenschaften 115 (4), 511–533. Dustmann, C. (1997), Return migration, uncertainty and precautionary savings. Journal of Development Economics 52 (2), 295–316. Dustmann, C. (2003), Return migration, wage differentials and the optimal migration duration. European Economic Review 47 (2), 353–369.

Circular Migration or Permanent Return

495

ETF (2007), The Contribution of Human Resources Development to Migration Policy in Albania. European Training Foundation, Torino, Italy. Faini, R. (2007), Remittances and the Brain Drain. The World Bank Economic Review 21 (2), 177–191. Gould, W., Pitblado, J., Sribney, W. (2003), Maximum Likelihood Estimation with Stata, second ed. Stata Press, Stata Corp. Greene, W.H. (2002), Econometric Analysis, ﬁfth ed. Prentice-Hall, New Jersey. Hill, J.K. (1987), Immigrant decisions concerning duration of stay and migratory frequency. Journal of Development Economics 25 (1), 221–234. King, R., Dalipaj, M., Mai, N. (2006), Gendering Migration and Remittances: evidence from London and Northern Albania. Population Space and Place 12 (6), 409–434. Lahiri, K., Song, J.G. (2000), The effect of smoking on health using a sequential self-selection model. Health Economics 9 (6), 491–511. Massey, D.S., Espinosa, K.E. (1997), What’s driving Mexico-U.S. migration? A theoretical, empirical, and policy analysis. The American Journal of Sociology 102 (4), 939–999. McFadden, D. (1974), The Measurement of Urban Travel Demand. Journal of Public Economics 3 (4), 303–328. Mesnard, A. (2004), Temporary migration and capital market imperfections. Oxford Economic Papers 56 (2), 242–262. Nawata, K., Nagase, N. (1996), Estimation of sample selection bias models. Econometric Reviews 15 (4), 387–400. OECD, A. (2008), International Migration Outlook: SOPEMI 2008. OECD, Paris. Piracha, M., Vadean, F. (2010), Return migration and occupational choice: evidence from Albania. World Development 38 (8), 1141–1155. Radu, D.C., Epstein, G. (2007), Returns to return migration and determinants of subsequent moves. EALE Conference Paper, EALE Annual Conference, 20–22 September 2007, Oslo. Stark, O. (1991), The Migration of Labour. Basil Blackwell, Oxford. Stark, O., Taylor, J.E. (1991), Migration incentives, migration types: the role of relative deprivation. The Economic Journal 101 (408), 1163–1178. Stolzenberg, R.M., Relles, D.A. (1997), Tools for intuition about sample selection bias ant its correction. American Sociological Review 62 (3), 494–507. Vullnetari, J. (2007), Albanian migration and development: state of the art review. IMISCOE Working Paper No. 18, Institute for Migration and Ethnic Studies (IMES), Amsterdam.

CHAPTER 21

Labor Migration, Remittances, and Economic Well-Being: A Study of Households in Rajasthan, India Yan Xinga, Moshe Semyonova,b and Yitchak Haberfeldc a

Department of Sociology, University of Illinois, Chicago, IL 60607-7140, USA E-mail address: [email protected] b Department of Sociology, Tel Aviv University, Tel Aviv 69978, Israel E-mail address: [email protected] c Department of Labor Studies, Tel Aviv University, Tel Aviv 69978 Israel E-mail address: [email protected]

Abstract Remittances sent by immigrants have long been viewed as a means to combat poverty, to improve consumption, and to raise standard of living. The present study examines the impact of remittances on the economic well-being of Indian households. The analysis is conducted on a randomly selected representative sample of households in Rajasthan. Three types of households are examined: 575 households having current labor migrants, 162 never having migrants, and 232 not having migrants at present but sent migrants in the past. Analysis of the data reveals meaningful differences between the three types of households. Speciﬁcally, those having labor migrants are characterized by the highest household income and standard of living. Further analyses suggest that although remittances are likely to improve economic well-being and to secure a higher standard of living they do not have long lasting effect on the economic well-being of the families when migration ends. Keywords: Inequality, Indian Society, migration, remittances, standard of living JEL classiﬁcations: F24 - Remittances

1. Introduction Labor migration has long been motivated by the growing need of families in poor regions of the world to increase the ﬂow of income to the household in country of origin. Researchers of global migration have not Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008027

r 2010 by Emerald Group Publishing Limited. All rights reserved

498

Yan Xing et al.

only underscored a substantial rise in the number of labor migrants and in ﬂows of remittances but they also demonstrate that remittances are often used to combat poverty and to improve living standard of the family members in the homeland (e.g., Stark, 1984; Massey, 1990, 1994; Massey et al., 1993; Massey and Parrado, 1994; Semyonov and Gorodzeisky, 2004, 2008). Indeed, the growing body of literature on the topic suggests that more and more households in poor countries have begun adopting labor migration as an economic strategy. That is, more persons are leaving their homes in search of higher earnings than the earnings they could possibly have in their country of origin to remit portions of these earnings to family members left behind. In other words, in the global economy, labor migration has become a way of life for many poor households as they increasingly become dependent on remittances for daily survival. In the present research, we study the role played by remittances sent by labor migrants to improve standard of living of their households. We do so by focusing on a representative sample of households in Rajasthan, India, and by comparing indicators of living standard across three types of households: those currently having labor migrants (hereafter CULM), those not having migrants but used to have labor migrants previously (hereafter PRLM), and those with no labor migrants (NOLM). The chapter is organized is as follows: ﬁrst, we review the literature on labor migration and the role of remittances in the global economy in general and in India in particular; second, we introduce the data source and the variables utilized in the analyses; third, we present the ﬁndings; and ﬁnally, we discuss the meaning of the ﬁndings and their implications. By so doing, we are in a position to contribute to a better understanding of the impact that global migration and remittances exert on the well-being of the households in sending societies in general and to advance knowledge on labor migration and its consequences in India in particular.

2. Labor migration and the role of remittances Social scientists view global labor migration, ﬁrst and foremost, as a rational response to the differential distribution of economic opportunities across labor markets. According to this view, individuals are likely to migrate from places with scarcity of capital and abundance of labor where wage returns on human capital resources are relatively low to places with abundance of capital and scarcity of labor where wage returns on human capital resources are relatively high (Borjas, 1987). The direction of migration, thus, is asymmetric – from places with limited economic opportunities where both demand for labor and wages are low to places with abundant economic opportunities where demand for labor and wages are relatively high. It has been well established in the social science literature that decisions about labor migration are usually reached within the household unit as a

A Study of Remittances in Rajasthan, India

499

rational economic strategy (e.g., Stark, 1984; Taylor, 1987; Massey, 1990, 1994; Massey et al., 1993, 1998; Massey and Parrado, 1994). According to this view, the family adopts a strategy aimed at allocating resources rationally and efﬁciently to increase the ﬂow of incomes and to lower economic risks. That is, in many poor countries, members of the household act collectively to consider the cost of the family’s temporary separation and the potential increase in household income when deciding to send one or more members of the household as labor migrants (Epstein and Kahana, 2008). The migrants, in turn, are expected to remit a substantial portion of their earnings back home. Labor migration, thus, should be viewed as a rational economic decision that minimizes risks associated with potential markets failure and at the same time maximizes the potential earnings of the household (e.g., Stark, 1984; Massey, 1990, 1994; Massey et al., 1993; Massey and Parrado, 1994; Semyonov and Gorodzeisky, 2008). Remittances – the portion of migrant workers’ earnings sent back to their home country – exert considerable impact both on the social system and the economy of the home country and on quality of life of individual households (Russel, 1986; Itzigsohn, 1995; Durand et al., 1996; Lu and Treiman, 2006; Singh, 2007; Semyonov and Gorodzeisky, 2005, 2008). At the macro level, remittances give a country a large stable ﬂow of hard foreign currency. Remittances contribute to both foreign exchange balances and the gross domestic product (GDP). ‘‘Remittances remain the second-largest ﬁnancial ﬂow to developing countries after foreign direct investment, more than double the size of net ofﬁcial ﬁnance’’ (Global Development Finance, 2004, p. 169). Furthermore, Dunn (2004) shows that remittances are greater than the combined money given by international foundations, nongovernment organizations and corporate philanthropy and argues that remittances help to correct inequities that had not been resolved neither by the market nor by the government (Dunn 2004). In general, studies show that the majority of money returned to households of labormigrants in poor countries is spent on consumption rather than on investments as it is mostly used to cover basic daily needs and household expenses (e.g., Zlotnik, 1990; Findley, 1994; Seddon, 2004; Cohen, 2005; Orozco et al., 2005). For example, Cohen (2005) found that in the rural communities of Oaxaca, Mexico, approximately 92% of the remittances are spent to meet daily needs and to pay for household expenses. Similar pattern was found in Zambia (Hansen, 2000) where those who migrated to the Congo to work in mines typically spent most of their income on purchase of goods and clothing [see also Madhavan (1985) and Helweg (1983) for India, Adler (1980) for Algeria, and Castano (1988) for Columbia]. While the majority of remittances are used for basic daily needs, studies also ﬁnd that some money is invested in business start-ups and is used for other purposes (Durand et al., 1996; Koc and Onan, 2004; Cohen and Rodriguez, 2004; Cohen, 2005; Semyonov and Gorodzeisky, 2008).

500

Yan Xing et al.

3. Remittances in India The study of the impact of remittances on economic well-being of households within the context of the Indian society is especially illuminating for several reasons. The World Bank Migration and Remittances Factbook reported that remittances sent back to India accessed $45 billion in 2008. According to this report, India continues to retain its position as the leading recipient of remittances in the world, with China and Mexico close behind at $34 billion and $26 billion, respectively. The amount of remittances sent to India has risen steadily during the past 20 years and rather dramatically during the past decade. In 1990–1991, Reserve Bank of India (RBI, which is the central bank of India) reported that remittances from overseas Indians were a modest $2.1 billion. By 2000–2001, this amount had risen to $12.85 billion and since 2000 the remittances have constituted a substantial portion (about 3%) of the Indian GDP.1 Labor migration and money sent back home by migrants have become part of reality for many families in India. In fact, in some parts of India, ‘‘it is unusual for a family group not to have one of its members overseas’’ (Singh, 2007, p. 93). Consequently, overseas remittances become the most important mean for economic survival for many (especially poor) Indian families. Furthermore, remittances were found to exert a signiﬁcant impact on economic well-being of families. Households that receive overseas remittances are likely to attain better economic conditions than households with no migrant workers (Helweg, 1983; Madhavan, 1985; Singh, 2007). Nevertheless, the literature on the role of labor migration in Indian society has not yet informed us in details on how remittances are used by the households and the extent to which remittances improve living conditions among families of labor migrants as compared to other families. Thus, in the analysis that follows, we examine the extent to which the ﬂow of remittances explains disparities in standard of living between households with and without overseas workers. Indeed, on the basis of the literature discussed at the outset of the chapter, we expect households that sent one of their members to work outside the region to enjoy higher living standard than households with NOLM.

4. Data and variables Data for estimating the impact of remittances on economic well-being were obtained from a survey of households. The survey was conducted by 1

Sources: Reserve Bank of India: RBI Bulletin December 1997, December 2004, January 2005, February 2006; ‘‘Invisibles in India’s Balance of Payments.’’ RBI Bulletin, November 2006; Reserve Bank of India: ‘‘Handbook of Statistics on the Indian Economy 2004–05.’’

A Study of Remittances in Rajasthan, India

501

the Department of Geography, Mohanlal Sukhadia University, Udaipur, in 2007. The households included in the survey were randomly sampled from four districts in Rajasthan, the largest state in India: Banswara, Udaipur, Dungarpur, and Chittaurgarh. Rajasthan is located in the northwest of India, with a primarily agricultural and pastoral economy. It is common for rural Rajasthan families to have members of the family working abroad (mostly in Middle Eastern countries). In the present analysis, we study 575 households that currently sent family member(s) to work abroad (CULM) and 394 households that did not send a labor migrant to an overseas labor market. Among the latter group of households, 232 had overseas workers in the past (PRLM) while 162 of them never had overseas labor migrants (NOLM). In what follows we examine disparities in household income and in standard of living among these three types of households. We also examine the extent to which these disparities could be attributed to labor migration and to remittances sent by oversea workers. Two dependent variables are examined in the analysis. They are household income and household standard of living. Household income is measured in terms of rupees per year. Income has two major measured components: earnings from domestic labor market (in rupees) and remittances received from overseas earnings (in rupees). We also included (only among PRLM households) an estimated value of the amount of remittances (in rupees) that were received in the past. Household income is used in the analysis, once as a dependent variable and once as a predictor of standard of living. Standard of living is measured as a cumulative scarcity index of the number of household goods and facilities in the possession of the household (see Semyonov and Lewin-Epstein, 2000; Semyonov and Gorodzeisky, 2008). Eight items of household consumption assets2 were selected for the construction of a standard of living index. Each item was assigned a value of 1 when it was in possession of the household (and 0 otherwise). It is expressed in terms of a cumulative scale, weighted by the rarity of the assets among all households (hereafter, SDLV). The distribution of the items that construct the index across the three types of households is presented in Appendix. The independent variables that are utilized to predict household standard of living include: age of household head (in years), education of household head (years of formal schooling), occupation of household head (three major categories), sector of employment of household head (public or private sector), size of household (number of persons in the household), and household migrant status (three groups of households). 2

The eight items are two wheeler, four wheeler, fridge, washing machine, phone, electricity consumption, tape water connection, and mixers. Since all the households have TV sets, we exclude this item from the measure of SDLV.

502

Yan Xing et al.

Table 1.

Mean (SD) characteristics of households

Variable

Deﬁnition

Family size

Number of persons

Age of household head

In years

Education of household head Number of room in the house

In years Number of rooms

Mean (SD) N ¼ 969 5.65 (2.40) 44.32 (12.20) 8.86 (3.40) 5.34 (3.26)

Person per room

Persons per room

1.45 (1.25)

Occupation of household head

Agricultural work (%) Blue collar job (%) White collar job (%) Household head work in public sector (%)

35.2 35.8 29.0 4.0

Total income ( ¼ domestic In rupees per year earnings þ remittances)

253,134.57 (276,306.81)

Domestic labor market income Remittance

In rupees per year

Previous remittance

In rupees per year

Domestic earnings per capita Total income per capita

In rupees per year In rupees per year

60,274.17 (89,270.86) 352,513.21 (267,632.07) 195,791.50 (162,431.42) 11,084.98 16,976.39) 44,729.32 (44,260.38)

Remittance per capita

In rupees per year

Standard living index (SDLV)

Cumulative scarcity index of the number of household goods and facilities in the possession of the household

In rupees per year

61,550.99 (42,116.01) 1.58 (0.92)

The detailed deﬁnitions of all variables along with their mean value (standard deviation) or percentage are presented in Table 1.

5. Analysis and ﬁndings 5.1. The multiple use of remittances Before comparing the characteristics, income, and standard of living of the subpopulations and before estimating the impact of remittances income on

503

A Study of Remittances in Rajasthan, India

Table 2.

Purposes for which remittances were spent by households with previous overseas workers and current overseas workers CULM (N ¼ 575)

PRLM (N ¼ 232)

Number of family

%

Number of family

%

Daily life expense Regular living expenses Repayment of debt Social obligations

380 347 404

66.09 60.35 70.26

131 174 68

56.47 75.00 29.31

Consumption House construction Live-stock Appliances and vehicles for own use

258 140 299

44.87 24.35 52.00

64 7 49

27.59 3.02 21.12

Material capital investment Purchase of land Machinery Business Vehicles for business

86 159 51 28

14.96 27.65 8.87 4.87

20 11 13 8

8.62 4.74 5.60 3.45

Human capital investment Education

312

54.26

46

19.83

standard of living it seems important to examine, ﬁrst, the ways and purposes for which remittances are being used and spent in Rajasthan. Thus, in Table 2 we list various purposes for which remittances are used by two types of households: those currently having migrants (CULM) and those that previously sent labor migrants (PRLM) since households with no migrants – NOLM – receive no remittances they are not included in this table. The data show that the three most important purposes of remittances use among households with migrant workers are: regular daily living expenses (66% for CULM households, 56% for PRLM), repayment of debt (60% for CULM and 75% for PRLM) and social obligation expenses (i.e., mainly wedding ceremonies, but also community services, and traditional rituals and ceremonies, 70% for CULM and 29% for PRLM), followed by house construction (45% for CULM and 28% for PRLM), and consumption of appliances and vehicles (52% for CULM and 21% for PRLM). It is important to note that while consumption seems to be the immediate goal of remittances use, quite a few households spend remittances also for investment purposes both for the purchase of land and machinery and for investment in enhancement of human capital of the future generation. That is, 54% among households with current migrants indicated that remittances are being used for investment in education.

504

Yan Xing et al.

5.2. Descriptive overview – comparing households with and without labor migrants In Table 3 we compare the characteristics of the three subpopulations for a descriptive overview. Our data reveal that the three types of households differ considerably by standard of living, by their labor market income as well as by their demographic and social compositions. Table 3. Mean(SD) characteristics of households with previous overseas workers, with current overseas workers and without overseas workers Variable

NOLM (N ¼ 162) Mean

Family size Age of household head Education of household head Number of room in the house Person per room Occupation of household head Agricultural work (%) Blue collar job (%) White collar job (%) Household head work in public sector (%)

PRLM (N ¼ 232) Mean

CULM (N ¼ 575) Mean

**

5.09 (1.78) 41.53** (12.10)

**

5.13 (2.07) 42.91** (10.38)

6.03 (2.61) 45.67 (12.71)

11.35** (3.85) 6.07 (3.46)

8.08* (2.65) 3.69** (1.92)

8.47 (3.22) 5.73 (3.46)

1.04** (0.62)

1.72** (1.17)

1.46 (1.39)

37.1 26.5* 36.4* 13.6*

29.8* 41.8 28.4 3.0*

36.6 36.1 27.3 1.4

Total income ( ¼ domestic earnings þ remittances)

106,748.77** (126,539.48)

77,671.55** (80,229.94)

365,172.70 (301,002.08)

Domestic labor market income

106,748.77** (126,539.48) 0.00

77,671.55** (80,229.94) 0.00

40,125.96 (72,216.00) 352,513.21 (267632.07) 0.00

Remittance Previous remittance Domestic earnings per capita Total income per capita Remittance per capita Standard living index (SDLV) *

0.00 22,060.00** (27,496.71) 22,060.00** (27,496.71)

195,791.50** (162,431.42) 15,425.99** (14,479.19) 15,425.99** (14,479.19)

0.00

0.00

1.34** (1.13)

1.45** (0.82)

6,224.48 (11,127.20) 62,971.14 (46,719.73) 61,550.99 (42,116.01) 1.70 (0.87)

A signiﬁcant difference with households with current overseas workers at alpha ¼ 0.05 level. A signiﬁcant difference households with current overseas workers at alpha ¼ 0.01 level.

**

A Study of Remittances in Rajasthan, India

505

The data displayed in Table 3 (see also Appendix A for detailed information on the components of living standard) reveal that on average, standard of living among households in Rajasthan is quite low when compared to the level of standard of living of households in industrialized economically developed countries. On average, slightly over one-third (33% for CULM and 47% for NOLM) households are not connected to electricity, and over two-third (64% for CULM and 72% for NOLM) do not have tap water. Availability of electricity and tap water is higher among households of previous labor migrants (access to electricity and to tap water, respectively, is 91% and 58% among PRLM). Perhaps, households of previous migrants used ﬂows of remittances in the past to connect to both the electric and the water systems. A comparison of home appliances in the possession of the households reveals that those having currently labor migrants are better off than households with previous labor migrants or with no migrants. Around 60% of CULM household have fridge, mixers, and phones in their possession but only about one-third of the other households have fridge in their possession and about half of them have mixers and phones. Only a small number of households have washing machines (10% for CULM, 3% for PRLM, and 12% for NOLM). Households of CULM are more likely to own a vehicle (whether an automobile or two-wheeler). That is, whereas 13% of households with current labor migrants own a four wheeler, only 4% of PRLM and 8% of NOLM households posses a car. The differences in the possession of goods and in living facilities are clearly reﬂected in the value of the cumulative scale of standard of living. Consistent with expectations, households with current labor migrants have the highest value for the standard of living index (1.70), followed by households with previous labor migrants (1.45). Households with NOLM have the lowest score on the cumulative index of living standard (only 1.34). As compared to both CULM and PRLM, they have lower standard of living. The data displayed in Table 3 show that remittances sent by migrant workers compose over 90% of the income of CULM households. Speciﬁcally, when remittances are included in the household income, the average income of CULM reaches 365,172 rupees (about $8,060). By way of comparison, this income is about three times the total income of households without any labor migrants and ﬁve times the income of households with previous labor migrants. The income per capita of CULM households reaches 62,971 rupees (around $1,390) (about three times the income of households without any labor migrants and four times the income of households with previous labor migrant). Apparently, remittances not only have become the major source of income for households with overseas labor migrants, it also contributes greatly to increasing income disparities between households with and without overseas labor migrants. It should be noted, however, that households

506

Yan Xing et al.

without any overseas labor migrants have the highest average income from domestic labor markets and households with current overseas labor migrants have the lowest level of domestic income. It seems reasonable to suggest that CULM are less dependent on the ﬂow of domestic earnings and perhaps have adopted the migration strategy due to lack of ability to reap high earnings in the domestic labor market. By contrast, the relatively high income of NOLM serves as a deterrent of labor migration. It serves as an incentive for staying in the region. The three types of households not only differ in standard of living and income level but they also differ in their demographic and social compositions. Households without any overseas workers are of the smallest household sizes. They are also characterized by heads of households of younger age and with higher educational level. By contrast, the heads of households with current overseas workers are older and are characterized by the lowest level of education. PRLM falls in between. Likewise, households also vary in their occupational composition. Households with no migrants have the highest rate of white collar workers and the highest rate of public-sector workers while households with current oversea workers have the lowest proportion of white collar workers and public-sector employment. These ﬁndings probably point at a selection process into labor migration. Households with high-status and stable jobs tend not send their members as labor migrants. Another possible selection process in this context is related to the immigrants themselves. Who are those members selected by the household to be sent overseas? Are they the ‘‘best’’ members or the ‘‘worst’’ members (however deﬁned) of the household? We cannot answer this question based on our data. Consequently we have to assume that similar decision rules are used by sending households residing in the same geographical region.

5.3. Multivariate analysis Since the data displayed in Table 3 reveal that the three types of households differ by access to remittances, labor market income, and standard of living as well as by their demographic and social compositions, it is important to estimate the net effect of migration and remittances on the two indicators of economic well-being: household income and household standard of living. Therefore, in the analysis that follows, we estimate a series of regression equations predicting household income (presented in Table 4) and household standard of living (presented in Tables 5 and 6). Table 4 displays results of two regression equations predicting the income of the household. Column (1) pertains to total household income while column (2) pertains to household income per capita. The data reveal, rather clearly, net of the household sociodemographic characteristics; labor migration exerts signiﬁcant effects on both total and per capita

507

A Study of Remittances in Rajasthan, India

Table 4.

Regression equation coefﬁcients predicting total household income in logarithm [LN] and total income per capita (LN)

Family size Age of household head Education of household head Occupation of household heada Blue collar job White collar job Household head work in public sector (dummy)b Current migrant (CULM dummy) Previous migrant (PRLM dummy) Constant R2

Total household income

Income per capita

(1)

(2)

0.0647*** 0.0189*** 0.0031

0.0788*** 0.0163*** 0.0022

0.0584 0.0816 0.4935***

0.0397 0.0672 0.4816***

1.2279*** 0.4458*** 10.1919***

1.2110*** 0.4428*** 9.4613***

0.5989

0.5218

*

Po0.10; **Po0.05; ***Po0.01. Omitted category – agricultural work. b Omitted category – private sector. a

income. Other things being equal, the income of CULM households, whether total or per capita, is substantially higher than income of either NOLM or PRLM as evident by the positive and signiﬁcant coefﬁcient for CULM [b ¼ 1.23 in column (1) and b ¼ 1.21 in column (2)]. The income of PRLM households, however, is substantially lower than the other groups as evident by the negative coefﬁcient for PRLM [b ¼ 0.45 in column (1) and b ¼ 0.44 in column (2)]. Apparently, the impact of past migration on income ﬂows is not long lasting. Notwithstanding the impact of labor migration on household income, the ﬁndings reveal that household income is also affected by sociodemographic characteristics of household head. More speciﬁcally, the data indicate that both total household income and income per capita tend to rise with age of head of the household and to be higher among those holding a job in the public sector. Not surprisingly total income is likely to also increase with size of household (due to number of potential earners), and income per capita tends to decrease with size of households (due to number of potential consumers). In Table 5 we examine the net impact of income and remittances on SDLV for each one of the three subgroups. A series of regression equations are estimated, thus, for each group as follows: Columns (1a–4a) for CULM; columns (1b–3b) for PRLM; and columns (1c–2c) for NOLM. In column (1a), we predict standard of living among CULM without accounting for labor market income among the independent variables. In column (2a), we include domestic labor market income among the

(4a)

(1b)

Po0.10; **Po0.05; ***Po0.01. Omitted category – agricultural work. b Omitted category – private sector.

a

*

0.0792

0.0945

0.1505

0.1455

0.0794

(3b)

(1c)

NOLM

0.0977

0.4837

0.0784**

0.0843 0.0337 0.0604 0.0480 0.6622* 0.5567 0.2164*** 0.1045

0.0506

1.2610**

0.0478 0.1794 0.0883

0.0155 0.0188 0.0741 0.0129** 0.0101 0.0007 0.0825*** 0.0689*** 0.0464

(2b)

0.1089

(3a)

R2

(2a)

PRLM

0.2624

(1a)

CULM

Regression equation coefﬁcients predicting Standard Living Index by household migrant status

Family size 0.0160 0.0056 0.0018 0.0069 0.0085 Age of household head 0.0195*** 0.0182*** 0.0147*** 0.0159*** 0.0144** Education of household head 0.0480*** 0.0481*** 0.0540*** 0.0523*** 0.0806*** Occupation of household heada Blue collar job 0.1310 0.1227 0.1536* 0.1394* 0.0943 White collar job 0.0788 0.0815 0.0828 0.0623 0.0778 Household head work in public sector (dummy)b 0.5264* 0.6701** 0.5713* 0.5633* 0.4952 Domestic labor market income (in 100,000 rupees) 0.1531*** 0.0630 Remittance (in 100,000 rupees) 0.0783*** Previous remittance (in 100,000 rupees) Total household income (in 100,000 rupees) 0.0771*** Constant 0.3741 0.4378* 0.3235 0.3414 0.2531

SDLV

Table 5.

0.1056

1.4726*

0.1585 0.0776 0.0408 0.2162***

0.0806 0.0071 0.0313

(2c)

508 Yan Xing et al.

509

A Study of Remittances in Rajasthan, India

Table 6.

Coefﬁcients of pooled regression equations predicting standard living index SDLV (1)

(2)

(3)

Family size 0.0143 0.0064 0.0088 Age of household head 0.0141*** 0.0096*** 0.0110*** Education of household head 0.0531*** 0.0554*** 0.0556*** Occupation of household heada Blue collar job 0.0669 0.0843 0.0794 White collar job 0.0899 0.0782 0.0726 0.1849 0.2425 0.2217 Household head work in public sector (dummy)b 0.1876** Current migrant (CULM dummy) 0.3790*** 0.2370** Previous migrant (PRLM dummy) 0.1668* 0.2352** 0.2180** *** Domestic labor market income 0.1374 (in 100,000 rupees) Remittance (in 100,000 rupees) 0.0830*** Total household income (in 100,000 rupees) 0.0904*** CULM total household income PRLM total household income Constant 0.2204 0.3202 0.3234 R2

0.0705

0.1284

0.1237

(4) 0.0097 0.0104*** 0.0545*** 0.0853 0.0654 0.2550 0.2903*** 0.2314*

0.1700*** 0.0847 0.0160 0.2739 0.1273

*

Po0.10; **Po0.05; ***Po0.01. Omitted category – agricultural work. b Omitted category – private sector. a

predictors of SDLV, and in column (3a), we introduce both domestic labor market income and remittances sent by current oversea workers in the set of predictors. In column (4a), total income replaces the two components of income. The regression coefﬁcients demonstrate that among CULM, remittances are the foremost important determinant of household standard of living. Although domestic earnings exerts a positive and signiﬁcant effect on SDLV in equation 2a (implying that SDLV tends to rise with increase in domestic earnings), the effect of domestic earnings on SDLV becomes statistically insigniﬁcant in column (3a) (where both remittances and domestic earnings are included among the predictors). In other words, the impact of remittances on SDLV in column (3a) is positive and signiﬁcant while the impact of domestic earnings is statistically insigniﬁcant. Apparently, among CULM households, standard of living is likely to rise, ﬁrst and foremost, with the receipt of overseas remittances but not as much due to ﬂow of earnings that are produced in the domestic market. Indeed, total income (which takes into consideration both remittances and earnings) exerts positive inﬂuence on living standard – the higher the income ﬂow from all resources the higher the standard of living of the household.

510

Yan Xing et al.

Columns (1b, 2b, and 3b) of Table 5 pertain to the sample of household with previous overseas workers (PRLM). In column (1b), we let the index of standard of living be a function only of the sociodemographic attributes of the household. In column (2b), a measure of earnings generated in the domestic labor market income is added to the set of predictors, and in column (3b), we also include a variable representing the (estimated) amount of past remittances to explore whether the of remittances received in the past on standard of living at the present time. The data reveal, rather clearly, that among household with previous labor migrants, remittances that were received in the past are the most important determinant of household standard of living. The positive and signiﬁcant coefﬁcient for domestic earnings in column (2b) indicates that earnings ﬂows produced in the domestic labor market inﬂuence standard of living – the higher the earnings the higher the standard of living. However, when past remittances are also included among the independent variables [column (2b)], the data reveal that past remittances exerts signiﬁcant impact on SDLV but domestic earnings have no signiﬁcant impact on household standard of living. Apparently, number of goods and appliances in the possession of the household as an indicator of standard of living is likely to rise due to remittances received in the past and much less so due to earnings attained in the local labor market. The higher were the remittances sent in the past the higher is the current standard of living. Once again, the data support the hypothesis that past remittances have a signiﬁcant impact on household standard of living. The last two columns in Table 5 [columns (1c and 2c)] pertain to the subgroup of households with no immigrants (NOLM). Thus, standard of living among NOLM is predicted in column (2c) as a function of household sociodemographic attributes plus only domestic earnings (since this is their one and only source of income, domestic earnings capture the total earnings of NOLM). The positive and signiﬁcant coefﬁcient for domestic earnings in column (2c) reveals that standard of living of NOLM households is signiﬁcantly inﬂuenced by domestic earnings – the higher are the domestic earnings, the higher is the standard of living. Notwithstanding the impact of remittances and earnings on standard of living, the results displayed in Table 5 suggest that household standard of living is also affected by some of the sociodemographic characteristics of the head of the household. First, among households with either current or previous oversea labor migrants, standard of living is likely to rise with age and education of household head (the effects of those two variables are positive and signiﬁcant in almost all equations). Likewise, standard of living tends to be lower among those having a job in the public sector (among CULM and PRLM). Public sector job, however, has no signiﬁcant effect on SDLV among households without any migrants. This is a somewhat surprising and unexpected ﬁnding since public sector job in this region represent a stable and relatively lucrative employment.

A Study of Remittances in Rajasthan, India

511

In Table 6 we pooled the three subpopulations to examine the extent to which income and remittances are responsible for variations between groups in their current standard of living. Thus, in Table 6 we display a series of regression equations predicting standard of living for all households as a function of the type of household (dummy variables representing CULM, PRLM versus NOLM) household sociodemographic characteristics and amount of current income from different sources (i.e., domestic and remittances). In column (1), standard of living (SDLV) is predicted as a function of household sociodemographic characteristics plus two dummy variables that distinguish among three types of household (i.e., CULM, PRLM, and NOLM). In column (2), sources of household income, including both domestic earnings and current remittances are added to the set of predictors. In column (3), the two sources of household income are replaced by total income (the combination of earnings from both domestic labor market and overseas remittances). In addition, column (4) introduces the interaction term between labor migration status and total income. The coefﬁcients in column (1) suggest that, other things being equal, households with current oversea labor migrants enjoy the highest standard of living than expected on basis of their sociodemographic attributes. The effects of migrant status in column (1) is positive and signiﬁcant for CULM (b ¼ 0.379) but is not signiﬁcant for PRLM (b ¼ 0.167), implying that, other things being equal, households with oversea labor migrants are able to purchase more goods and facilities than households without oversea lab migrants. The results displayed in columns (2 and 3) (where income is also included in the equations) suggest that although income ﬂows strongly affect SDLV, disparities in standard of living among different types of households could not be fully attributed to differences in the amount of the income ﬂows they report receiving. In fact, the analysis reveals that once differences in income are taken into consideration the standard of living of both CULM and PRLM is signiﬁcantly higher than that expected on basis of their characteristics and incomes. It appears that the former group (CULM) is likely to increase consumption due to potential future ﬂows of remittances they anticipate at present and the latter group (PRLM) was able to enhance standard of living due to remittances received in the past. The insigniﬁcant interaction terms between migration status and income in column (4) indicate that income is similarly used by all groups to increase standard of living. Finally, it is also worth noting that similar to ﬁndings reported in Table 5, SDLV is also affected by characteristics of the household regardless of income. Speciﬁcally, the number of goods and facilities in the possession of the household (as an indicator of standard of living) is likely to rise not only with the ﬂow of remittances and of domestic labor market income but also with age and education of the household heads.

512

Yan Xing et al.

6. Conclusions The present study attempts to explore the impact of remittances sent by oversea labor migrants on the economic well-being of households in India. The data support the argument that remittances enable households to improve economic well-being and to secure a higher standard of living. Other things being equal, households that send labor migrants and households that sent migrants in the past have higher standard of living (as measured by possession of household goods and access to basic home appliances) than households that do not send labor migrants. The ﬁndings of the present study provide ﬁrm support to the thesis that labor migration is a rational economic strategy adopted by poor families in poor regions of the world to improve their living conditions. The Indian case studied here suggests that households with relatively higher socioeconomic status and those holding better and stable jobs are less likely to leave home in search of work and earnings outside their country. The data reveal, rather clearly, that households that do not send any workers are characterized by higher educational level and by higher status and tend to hold more secure jobs (mostly jobs in the public sector of the local labor market). Furthermore, the size of their family is smaller, and they tend to reside in larger houses. By contrast, households that rely on remittances sent by labor migrants are characterized by lower education and by a larger family size. It is highly possible that, due to limited opportunities in the local labor market, the relatively disadvantaged households resort to labor migration as a rational economic strategy that ensures greater ﬂows of income to the household and improvement of living conditions. The lower income of families that previously had someone working overseas relative to those having current labor migrants further supports the argument that remittances help in improving household economic wellbeing. The return of migrant workers back home brings down household income signiﬁcantly, while overseas labor migration and remittances signiﬁcantly raise household income. There might be different reasons for families’ decision to stop sending workers. One possibility is related to the life cycle. Households with previous overseas workers are found to be headed by younger men relative to heads of households with present overseas workers. Younger heads might need to work locally to take care of younger children, while older household heads prefer to work overseas and to send remittances to support older children’s education. This is partly supported by the purposes for which remittances were spent by households. The most important purpose for households with previous overseas workers was payment of debt, while households with current overseas workers have multiple important purposes to spend remittances on, including children’s education. Further research is needed to explore how families make the decision whether to send workers abroad or to call them back home.

513

A Study of Remittances in Rajasthan, India

The present study focuses on the impact of remittances on income ﬂows and economic well-being of the family members in Rajasthan, India. It does not focus, however, on other aspects of household quality of life. Nor does it examine the impact of labor migration on family relations and on children left behind. The ﬁndings revealed by our analysis suggest that poor households in this region are quite likely to send their members overseas in search of better earnings opportunities. The inﬂows of extra earnings due to remittances enable households to purchase goods for consumption and to improve quality of life. However, a comparison between households with past and present migrants suggests that the impact of remittances on standard of living, although substantial, is not long lasting but rather of short duration. The comparison reveals that standard of living of households with migrants is signiﬁcantly higher than the standard of living of households that previously sent migrants abroad (although both enjoy higher standard of living than households with NOLM). It is quite possible that previous migrants used remittances mostly for immediate consumption purposes without investing a substantial portion in business ventures or in human capital of household members. Therefore, impact of past migration is relatively short and does not last over a long time span. Indeed, these issues and similar issues that were not addressed by our study should be further examined by future research.

Appendix. Consumption assets and the mean of SDLV of households with previous overseas workers, with current overseas workers and with no overseas workers

SDLV Item in detail Electricity consumption Tap water connection Refrigerator Washing machine Phone Mixers Two wheeler Four wheeler

NOLM (N ¼ 162)

PRLM (N ¼ 232)

CULM (N ¼ 575)

1.34 Number of yes 86

53.09

1.45 Number of yes 212

91.38

1.70 Number of yes 385

45

27.78

134

57.76

209

36.35

59 19 83 92 86 13

36.42 11.73 51.23 56.79 53.09 8.02

70 6 103 118 133 10

30.17 2.59 44.40 50.86 57.33 4.31

323 57 379 359 391 73

56.17 9.91 65.91 62.43 68.00 12.70

%

%

% 66.96

Notes: For each item, 1 is coded for the household that has it, while 0 is coded for the household that does not has it. SDLV ¼ sum[1 or 0 (1 the % of the households that have this item among total 969 households)].

514

Yan Xing et al.

References Adler, S. (1980), Swallows’ Children: Emigration and Development in Algeria. International Labor Ofﬁce, International Migration for Employment Branch, Geneva. Borjas, G.J. (1987), Self-selection and the earnings of immigrants. The American Economic Review 77, 531–553. Castano, G.M. (1988), Effects of emigration and return on sending countries: the case of Colombia. In: Stah, C.W. (Ed.), International Migration Today. UNESCO, Center for Migration and Development Study, Paris, pp. 191–203. Cohen, J.H. (2005), Remittances outcomes and migration: theoretical contests, real opportunities. Studies in Comparative International Development 40, 88–112. Cohen, J.H., Rodriguez, L. (2004), Remittance Outcomes in Rural Oaxaca, Mexico: Challenges, Options and Opportunities for Migrant Households. Available at http://www.ccis-ucsd.org/publications/ wrkg102.pdf. Retrieved on April 2009. Dunn, K. (2004), Diaspora Giving and the Future of Philanthropy. Available at http://www.tpi.org/downloads/pdfs/whitepaper-diaspora_ giving.pdf. Retrieved on July 2009. Durand, J., Parrado, E., Massey, D.S. (1996), Migradollars and development: a reconsideration of the Mexican case. International Migration Review 30, 423–444. Epstein, G.S., Kahana, N. (2008), Child labor and temporary emigration. Economics Letters 99, 545–548. Findley, S.E. (1994), Does drought increase migration? A study of migration from rural Mali during the 1983–1985 drought. International Migration Review 28, 539–553. Global Development Finance: Harnessing Cyclical Gains for Development. (2004), The World Bank. Available at http://siteresources.worldbank. org/GDFINT2004/Home/20177154/GDF_2004%20pdf.pdf. Retrieved on July 2009. Hansen, K.T. (2000), Salaula: The World of Secondhand Clothing and Zambia. University of Chicago Press, Chicago, IL. Helweg, A.W. (1983), Emigrant remittances: their nature and impact on a Punjabi village. New Community 10, 67–84. Itzigsohn, J. (1995), Migrant remittances, labor markets, and household strategies: a comparative analysis of low-income household strategies in the Caribbean Basin. Social Forces 74, 633–655. Koc, I., Onan, I. (2004), International migrants’ remittances and welfare status of the left-behind families in Turkey. International Migration Review 38, 78–112.

A Study of Remittances in Rajasthan, India

515

Lu, Y., Treiman, D.J. (2006), The effect of labor migration and remittances on children’s education in South Africa. Paper presented at the annual meeting of the Population Association of America, Los Angeles. Madhavan, M.C. (1985), Indian emigrants: numbers, characteristics, and economic impact. Population and Development Review 11, 457–481. Massey, D.S. (1990), Social structure, household strategies, and cumulative causation of migration. Population Index 56, 3–26. Massey, D.S. (1994), An evaluation of international migration theory. Population and Development Review 20, 699–751. Massey, D.S., Arango, J., Hugo, G., Kouaouci, A., Pellegrino, A., Taylor (1993), Theories of international migration: a review and appraisal. Population and Development Review 19, 417–446. Massey, D.S., Arango, J., Hugo, G., Kouaouci, A., Pellegrino, A., Taylor, J.E. (1998), Worlds in motion: understanding international migration at the end of millennium. Oxford University Press, New York. Massey, D.S., Parrado, E. (1994), Migradollars: the remittances and ravings of Mexican migrants to the United States. Population Research and Policy Review 13, 3–30. Orozco, M., Lowell, B.L., Bump, M., Fedewa, R. (2005), Transitional engagement, remittances and their relationship to development in Latin America and the Caribbean. Final report submitted to the Rockefeller Foundation for Grant 2003 GI 050. Washington, DC: Institute for the study of international migration, Georgetown University. Available at http://www.thedialogue.org/publications/2005/summer/trans_ engagement.pdf. Retrieved on May 2007. Russel, S.S. (1986), Remittances from international migration: a review in perspective. World Development 14, 677–696. Seddon, D. (2004), South Asian remittances: implications for development. Contemporary South Asian 13, 403–420. Semyonov, M., Gorodzeisky, A. (2004), Occupational destinations and economic mobility of Filipino overseas workers. International Migration Review 38, 5–25. Semyonov, M., Gorodzeisky, A. (2005), Labor migration, remittances and household income: a comparison between Filipino and Filipina overseas workers. International Migration Review 39, 5–25. Semyonov, M., Gorodzeisky, A. (2008), Labor migration, remittances and economic well-being of households in the Philippines. Population Research Policy Review 27, 619–637. Semyonov, M., Lewin-Epstein, N. (2000), The impact of parental transfers on living standards of married children. Social Indicators Research 54, 115–137. Singh, S. (2007), Sending money home – maintaining family and community. The International Journal of Asia Paciﬁc Studies 3, 93–109.

516

Yan Xing et al.

Stark, O. (1984), Migration decision making: a review essay. Journal of Development Economics 14, 251–259. Taylor, J.E. (1987), Undocumented Mexico – U.S. migration and the returns to households in rural Mexico. American Journal of Agricultural Economics 69, 619–638. Zlotnik, H. (1990), International migration policies and the status of female migrants. International Migration Review 24, 372–381.

CHAPTER 22

Promoting the Educational Success of Latin American Immigrant Children Separated from Parents during Migration Sara Z. Poggioa and T.H. Gindlingb a

Department of Modern Languages and Linguistics, University of Maryland Baltimore County, 1000 Hilltop Circle, Baltimore, MD 21250, USA E-mail address: [email protected] b Department of Economics, University of Maryland Baltimore County, 1000 Hilltop Circle, Baltimore, MD 21250, USA E-mail address: [email protected]

Abstract For many immigrants, especially those from Central America and Mexico, it is common for a mother or father (or both) to migrate to the United States and leave their children behind. Then, after the parent(s) have achieved some degree of stability in the United States, the children follow. In our previous research, we found that children separated from parents during migration are more likely to be behind others their age in school, are more likely to repeat a grade, and are more likely to drop out of high school. The negative impact of separation during migration on educational success is largest for children separated from their mothers (in contrast to fathers), for those whose parents have lived in the United States illegally, and for those who reunited with parents as teenagers (rather than at younger ages). In this chapter, we suggest public policies to help immigrant children separated from parents during migration to succeed in U.S. schools. The policies that we discuss are based on focus group discussions with parents separated from their children during migration, interviews with psychologists and school administrators, and an online survey of elementary and high school teachers. Keywords: immigrant children, education, family separation JEL classiﬁcations: I2, J13, J61

Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008028

r 2010 by Emerald Group Publishing Limited. All rights reserved

518

Sara Z. Poggio and T.H. Gindling

1. Introduction Because of the recent surge in immigration to the United States, immigrant children are one of the fastest growing segments of the U.S. school-age population. Special challenges and opportunities face immigrant children in school. On the positive side, immigrant children recognize the sacriﬁces they and their parents make for their beneﬁt, and many are therefore highly motivated to succeed in school (Rumbaut, 2005a). On the other side, challenges that immigrant children face include lack of English proﬁciency, culture shock, and the low socioeconomic status of many immigrant parents (Suarez-Orozco et al., 2008). In a companion paper, we identify another factor common to the migration experiences of many recent immigrant children that also contributes to the difﬁculties some immigrant children face in school – separation from parents during migration (Gindling and Poggio, 2009, 2010). For many immigrants, especially those from Central America and Mexico, it is common for a mother or father to migrate to the United States and leave their children behind, in the care of relatives or family friends. Then, after the parent(s) have achieved some degree of stability in the United States, the children follow (Suarez-Orozoco et al., 2002). Using both qualitative and quantitative methodological techniques, Gindling and Poggio (2009, 2010) show that, compared to Latin American immigrant children who remain with their parents during the migration process, children separated from parents during migration are more likely to be behind others their age in school, are more likely to repeat a grade, and are more likely to drop out of high school. In the present chapter, we suggest public policies to help immigrant children separated from parents during migration to overcome these obstacles and succeed in U.S. schools. The policies that we discuss are based on focus group discussions with parents separated from their children during migration, interviews with psychologists and school administrators, and an online survey of elementary and high school teachers. The rest of this chapter is organized as follows. In Section 2, we review literature related to the issue of family separation during migration and the educational success of immigrant children. In Section 3, we present the results of the policy suggestions made by parents and teachers to promote the educational success of children separated from parents during migration and then reunited with parents in the United States. Section 4 is a concluding section.

2. Literature review The existing literature on immigrant children rarely distinguishes between immigrant children who migrate with parents and those who are separated from parents during the migration process. Yet, studies that have made

Promoting the Educational Success of Latin American Immigrant Children

519

this distinction indicate that it is common for child migrants to the United States to be separated from parents during migration. Suarez-Orozco et al. (2002, 2008) report results from the Longitudinal Immigrant Student Adaptation (LISA) study, a survey of young recent immigrants from Central America, China, the Dominican Republic, Haiti, and Mexico recruited from 51 schools in 7 school districts in the Boston and San Francisco greater metropolitan area. Eight-ﬁve percent of the youth in this sample were separated from one or both parents during the migration process. Separation was most likely for immigrants from Central America (96% of children in the sample). Gindling and Poggio (2009, 2010), in a study of children of immigrants who received green cards in 2003, report that 31% of immigrant children (and 45% of immigrant children from Latin America) were separated from at least one parent for at least two years because of migration. Separation of young children from parents during migration can have profound negative psychological effects on children and their parents (Schen, 2005; Smith et al., 2004). Attachment theory, taken from the psychology literature, provides one framework to explain why. Attachment theory argues that disruptions in ‘‘affection bonds’’ with parental ﬁgures (especially mothers) can have profound negative psychological and developmental implications later in life. Young children can interpret separation from parents as a complete loss of their love and protection. Attachment theory focuses on the effect of the bond that children develop in their relationship with parents and in the meaning of the interruption of the relationship reﬂected in the child’s behavior. The loss of this bond with parents triggers grief responses that affect behavior. Separation from parents during migration, in particular, can lead to emotional distress and have an impact on later relationships and behavior. Immigrants in general experience ‘‘ambiguous loss’’ in relation to friends and family members in the country of origin (Boss, 1991). Ambiguous loss is deﬁned as the impossibility to mourn and heal after losing a loved one in the case of someone who is physically absent but psychologically present – friends and relatives who are alive but do not physically interact with the immigrant anymore. Immigrant children have to deal with ambiguous loss after their mother or father leaves them, when they have to leave their caregiver in the country of origin, and when they leave the rest of their family and friends. This burden that immigrant children bring to their new country and new school can become a signiﬁcant constraint for them to succeed at school in America. The emotional impacts of separation and reuniﬁcation are further complicated by pre- and postarrival events and conditions that the child experiences in relation with his/her particular family situation. It is reasonable to expect that school performance in the country of origin will also be affected by the sense of ambiguous loss that children have to endure. In some cases when the child is expecting to be reunited

520

Sara Z. Poggio and T.H. Gindling

with his/her parents in United States, he/she will be not concentrate enough on learning in their local school.1 Children with a family member in the United States may also be more likely to see migration, rather than education, as the route to higher earnings, and therefore less likely to ﬁnd schooling in the home country to be worthwhile, and are therefore likely to get less educational attainment while separated from their parents (McKenzie and Rapoport, 2006; Miranda, 2007). In a study of children of immigrants from Oaxaca, Sawyer et al. (forthcoming) ﬁnd that while remittances from extended family members abroad contribute to increased education levels of children, having a close family member in the United States actually reduces the education of children left behind. Similarly, Amuedo-Dorantes (2008) ﬁnds evidence that in some communities in Haiti, remittances raise school attendance only for children from households who do not experience any family out-migration. Suarez-Orozoco et al. (2002) collected longitudinal (1999–2002) data on 407 recently arrived immigrants, ages 9–14 in 1997, in San Francisco and Boston (the LISA study). They found that the following factors affect academic achievement: English language proﬁciency, parental education, income, gender, behavioral engagement, school characteristics, peers, and family structure. In their quantitative analysis, Suarez-Orozco et al. (2008) did not explicitly test for an impact of separation during migration on academic performance. However, in the qualitative portraits of high and low achievers at school, they write that among ‘‘protracted decliners,’’ ‘‘many families had been strained by protracted separations and complicated reuniﬁcations’’ (p. 170), whereas the ‘‘high achieving students y were also much less likely to report long separations from their parents’’ (p. 296). Gindling and Poggio (2009, 2010) report both qualitative and quantitative evidence that children separated during migration have less educational success compared to children who remain with their parents during the process of migration or children of immigrants born in the United States. Gindling and Poggio (2009, 2010) analyze data from the New Immigrant Survey (NIS), a representative sample of all of those new legal immigrants (those who received ‘‘green cards’’) in 2003. They report that over 40% of Latin American immigrant children who were separated from parents during migration drop out of high school, compared to less than 20% of those who migrated with their parents or are U.S.-born children of Latin American immigrant parents. Similarly, 16.7% of Latin American immigrant children in the NIS sample are at least one grade behind others their age, compared to only 6.4% of Latin American 1

Children left behind are, in effect, living in two worlds. ‘‘Piedras Blancas, El Salvador fourth grade teacher Roney Ramirez on Josselin Mendez, whose parents are both in the United States: ‘I try to tell her that what she learns here can serve her over there y But she really doesn’t take it in. Her mind is so focused on over there that it’s as though she’s left already’ ’’ (Aizenman, 2006).

Promoting the Educational Success of Latin American Immigrant Children

521

children who migrated with their parents and 5.3% of U.S.-born children of Latin American immigrants. These results continue to be true when other factors that affect drop-out rates and education gaps are controlled for using regression analysis (Gindling and Poggio, 2009, 2010). The negative impact of separation during migration on the schooling success of immigrant children is larger if parents were undocumented before receiving their green cards and larger for children separated from mothers or both parents compared to separation from fathers. This last result is consistent with other ﬁndings in the literature that separation from mothers during migration has more impact on children than does separation from fathers (Menjivar and Abrebo, 2009; Debry, 2009). This is an important issue because, in contrast to past waves of migration when fathers tended to leave their families behind, it is more likely today that it is women who leave their families to work in low-wage jobs in the service and domestic sectors of the United States and other developed countries, leaving their children behind with their husbands or other relatives (Salazar-Parren˜as, 2006; Debry, 2009). A striking ﬁnding of Gindling and Poggio (2009, 2010) relates to the ages at which children are separated from, and reunited with, their parents. The impact of separation is largest for children who were separated from their parents at older ages and who migrated as teenagers. Gindling and Poggio (2009) ﬁnd no empirical evidence of a negative impact of separation on children who migrated at younger ages. These age-related results are consistent with those reported by Gonzalez (2003) for immigrants to the United States and by Van Ours and Veenman (2006) for the Netherlands. Gonzalez (2003) reports that age of arrival has a signiﬁcant negative impact on years of schooling completed, but only for children who arrive as teenagers and only for Latin American, Mexican, and European immigrant children. On the contrary, immigrants who arrive as preteens complete more years of education compared with immigrants who arrived as infants. Van Ours and Veenman (2006) ﬁnd that a later age at migration lowers the level of educational attainment, but that again the impact varies by region of origin of the migrant. Gonzalez (2003) conducts a cost-beneﬁt analysis and concludes that a policy of allowing Latin American children to enter the United States before ﬁrst grade is cost-effective because the higher wages brought about by more schooling in the United States results in increased tax revenues that more than offsets the cost of education for these children in elementary, middle, and high school. There are several reasons why we might expect separation to have a bigger impact on children who migrated as teenagers. One set of explanations has to do with the children and the special challenges faced by teenagers. In our focus groups, for example, parents suggested that younger children are more responsive to parental expectations regarding success in school compared with teenagers. School counselors made

522

Sara Z. Poggio and T.H. Gindling

similar comments. The teenage years are already a period when the child– parent relationship experiences signiﬁcant strain. It is likely that adding family separation and reuniﬁcation to the mix adds an additional level of stress to this process. The evidence is also clear that language acquisition is easier for younger children, and some researchers have argued that the best time to learn a language is before puberty. According to this view, at puberty, there is a signiﬁcant and dramatic decline in the ability of children to learn languages (Scovel, 2000). As adaptation is already harder for those separated from their parents, this suggests that teenagers with poor English skills may have an even harder time adapting to U.S. schools than younger children with poor English skills. Because language acquisition becomes more difﬁcult as children age, adaptation to a new language, culture, and educational system will be more difﬁcult for children who migrate when they are older (Scovel, 2000; Chiswick and Miller, 2008). Another set of explanations for why the impact of separation is likely to be greater for teenagers focuses on the different nature of elementary, middle, and high school in the United States. Elementary schools are generally more supportive of students personally than are middle and high schools; teachers are with the same children all day and get to know them and their individual problems. Elementary school teachers also may, in general, be better able to help children separated during the migration experience adapt to U.S. schools because elementary school teachers have training in teaching English reading and writing, whereas it is generally assumed by middle and high school teachers that students already know how to read and write English. School counselors suggested to us that these differences could explain why those who were separated and reunited with their families as teenagers have a more difﬁcult problem adapting to U.S. schools. Another possible reason could be that the education of these children was interrupted in their home country or during the migration process. Many Latin American immigrants are from rural areas, where interrupted education for children is common, especially for older children who can work productively on farms. This could also explain why children who migrated when they were older are more likely to have an education gap or drop out. Because older children are more likely to be taken out of school to work on the farm than younger children, it is reasonable to suppose that only older immigrants will be more likely to have an education gap. It may also be that immigrant children are older than others in their grade in the United States because the process of migration may take time away from school or that, when they enter school in the United States, immigrant children are assigned to a lower grade than the grade that they completed in their home country. Finally, a growing literature suggests that older children who migrate never really ‘‘drop in’’ to school and are more interested in entering the labor market (and not education) as a means of economic advancement (Vernez et al., 1996).

Promoting the Educational Success of Latin American Immigrant Children

523

3. Policy recommendations from parents and teachers 3.1. Parents In November of 2008, we conducted a focus group with eight parents (all of them mothers) of children who had been separated from their parents for at least two years because of migration and who are currently attending schools in Baltimore, Maryland. This focus group also included one young adult who had experienced separation and reuniﬁcation from her parents during migration. The Hispanic Apostolate of Baltimore recruited group participants and provided the space for the meeting. Focus group participants were women from Honduras, El Salvador, and Mexico. They were all housewives. The age of participants was 25–39 years. The main themes discussed in this focus group were centered on the educational experiences of separated and reuniﬁed children and what parents believed schools and local governments could do to help their children succeed at school. Mothers veriﬁed that immigrant students who enter school in the United States are often assigned to a lower grade than they have completed in their home country. Mothers identiﬁed this as an important factor that negatively affected the academic success of immigrant (separated and reuniﬁed) children. Being assigned to a grade lower than they completed in their home country makes the student older than the rest of the students in his/her class and, as one group participant stated, made her child vulnerable to teasing and disrespect from other kids in the family and at school. Lack of knowledge of English language is considered as one of the main constraints for the academic achievement of the students at school, followed by the inability of parents to effectively help their children to overcome the problem. Not knowing the language or the school system, parents are poorly equipped to help their children to succeed in American schools. In this context, the inability of parents to help their children with homework (because of the parents’ lack of English skill) was identiﬁed as a particular problem. Asked about ways in which social policies at the county, state, or school level can help and support children separated and reunited with their parents, our group participants mentioned that (a) school counselors ﬂuent in Spanish; (b) teachers calling parents whenever children have problems (using parent’s cell phones if necessary); (c) help with homework in after-school programs; (d) and an increase school discipline, more rules, dress code or uniform and punishment. Regarding the last point, the Latin American mothers in our focus group all agreed that corporal punishment in school is necessary. In general, mothers were convinced that physical punishment is a convenient tool for education.

524

Sara Z. Poggio and T.H. Gindling

3.2. Teacher survey We designed an online, anonymous survey for teachers that complements and expands our understanding of the parent focus group results. The Appendix in this chapter contains a copy of the questionnaire for the teacher survey.2 The questions on the teacher survey are centered around the following themes: (a) what are the biggest challenges facing Latin American immigrant children in school; (b) are teachers aware of the prevalence and potential problems of family separation during migration, and if so what are the problems they see most often; (c) which immigrant children have an easier time adjusting to the U.S. school system – teenagers or younger children, boys or girls; (d) which existing programs or potential programs would most help in improving the performance of Latin American immigrants in school? To obtain teacher responses we e-mailed a letter to all teachers of English for Speakers of Other Languages (ESOL) in the northern region of the Prince George’s County Maryland Public School System. In addition, we e-mailed the letter to all teachers (not only ESOL teachers) at the two high schools with the largest proportion of immigrant students in the Prince George’s County Public School system (Northwestern and High Point high schools). The e-mail provided teachers with the URL and a link to the online survey and requested that they visit the site and complete the questionnaire. We were clear that the survey was voluntary and completely anonymous. The Appendix in this chapter contains a copy of this e-mail. We focused on high school because, according to our quantitative analysis, it is in high school where separation during migration has the biggest negative effect on success at school. We focused on Latin American students because they are by far the largest group of immigrant students in Prince George’s County schools as well as other Maryland and U.S. schools. 3.2.1. What are the biggest challenges facing Latin American immigrant children in school? The most frequently noted challenges facing Latin American immigrant students were lack of knowledge of English and lack of academic preparation (90% of teachers who responded cited both of these as challenges facing Latin American immigrant students). When asked why Latin American immigrant children arrived in the United States with a lack of academic preparation, the most common response was that the education of the children had been interrupted at some point before or during migration. The majority of teachers also identiﬁed emotional problems because of family separation and reuniﬁcation as a challenge facing Latin American immigrant students. However, relatively few teachers identiﬁed behavior 2

The questionnaire can be viewed at http://www.umbc.edu/mll/teachers/.

Promoting the Educational Success of Latin American Immigrant Children

525

or discipline problems as an issue, especially for newly arrived immigrants. Some did note, however, that behavior and discipline problems tended to become more of an issue the longer the immigrant student has been in the country.3 The most common discipline problem mentioned was the risk that boys who have been in the country for a few years would join gangs (the Salvadoran gang MS13 was speciﬁcally mentioned). The most common health problems mentioned were lack of adequate dental care and immunizations. Many teachers also identiﬁed low levels of involvement among the parents of Latin American immigrant children in their child’s education as a problem. One particularly common comment was that parents provided very little help to their children with homework. Teachers saw this problem arising from cultural differences, the long hours worked by immigrant parents, lack of material resources (computers, a private room in which to study at home, etc.), and the lack of English skill among immigrant parents (where students often have to translate for their parents in interactions with teachers and other school ofﬁcials). One ESOL teacher wrote ‘‘Many classroom teachers complain that Hispanic parents do not make sure that their children do homework (as compared to Asian parents). As an ESOL teacher, I see that many parents are working long hours; however, even those students with mothers at home, often do not do homework. Sometimes this is corrected after conferencing with the teacher and our Spanish-speaking parent liaison, but usually homework remains erratic.’’ Another a problem that makes it difﬁcult for parents to fully participate in the schooling of their children is the undocumented status of many Latin American immigrant parents. In response to an open-ended question about possible challenges facing immigrant students in schools, the most common response we received was legal issues surrounding undocumented immigrants (32% of teachers mentioned this). Several teachers wrote that parents were reluctant to come to school meetings, volunteer in school, or provide any personal information to teachers because they were afraid their undocumented status would come to the attention of authorities.4 The impressions of teachers are consistent with the conclusion from the quantitative analysis that children separated during migration are more likely to have less educational success if their parents are undocumented.

3

One teacher writes that behavioral problems are ‘‘very minimal – most are respectful until they become fully acculturated, then some problems crop up.’’ Another writes ‘‘They are eager and wonderful when they arrive. They quickly learn American behavior or get caught up by MS13.’’ 4 An example of how school rules can discourage undocumented parents from participating fully in their children’s schooling comes from Prince George’s County, were any parent who volunteers to work with students or chaperone ﬁeld trips is required to be ﬁngerprinted by county authorities.

526

Sara Z. Poggio and T.H. Gindling

Several teachers also expressed the view that high-stakes standardized testing mandated by the No Child Left Behind Act discouraged immigrant students from ﬁnishing high school (in Maryland, high school students are required to pass a set of subject-speciﬁc standardized tests to graduate). One teacher writes, ‘‘High stakes state exams become the focus of instruction and classroom teachers become frustrated with what they feel is the slow progress of English Language Learner (ELL) students.’’ 3.2.2. Are Latin American immigrant children older than others in their grades, and if so why? Most teachers agreed that immigrant children are often older than others in their grade (70% of those who responded to this question reported that Latin American immigrant students are at least sometimes older than others in their grade). When asked why they believe this has happened, the results are consistent with our expectations. The most common response (from 70% of respondents) was that immigrant children are assigned to a grade below the grade level in their home county. The next most popular choices were that they had repeated a grade in the United States (55%) or had missed a year of schooling during the migration process or interrupted schooling in their home country (50%). One teacher wrote ‘‘many students are 16, 17 or 18 in 9th grade – if they don’t have transcripts available from the home country, or if they were out of school working for a year or two.’’ Teachers also noticed that students who had interrupted schooling in their home countries tended to come from rural areas.5 3.2.3. Are teachers aware of the prevalence and potential problems of family separation during migration, and if so what are the problems they see most often? Yes, the teacher survey results suggest that teachers understand that separation during migration is a common phenomenon for immigrant students and that it may lead to problems in school. Most of those who responded to this question said that family separation during migration was common among the Latin American immigrant children that they teach. Teachers reported that, on average, almost 40% of the immigrant students in their classes had been separated during migration. The teachers 5

Counselors in Prince George’s County Public Schools (who are responsible for assigning immigrant students to a grade) told us that immigrants who enter in elementary schools are assigned to a grade based on their age, regardless of academic skills, whereas immigrants who enter in high school are assigned to a grade based on their academic skills (primarily English language proﬁciency), regardless of their age. The result is that most immigrants who enter the school system in high school, whatever their age, end up being assigned to 9th grade. Once students are assigned to a lower grade, they remain behind – students generally are not allowed to skip a grade to catch up to their peers.

Promoting the Educational Success of Latin American Immigrant Children

527

recognized that separation has negative effects on the relationship between parents and children, with 45% of respondents noticing problems in the relationships between children separated during migration and their parents. Several teachers identiﬁed resentment toward parents because of perceived abandonment and difﬁculties with new siblings or step parents as particular problems. One teacher wrote, ‘‘Yes, they sometimes have trouble ﬁtting into a family that has grown while they were still in their home country, i.e., new brother or sister born to parents in this country, or they have a new step parent they don’t know. Parents in this situation often don’t have a lot of time to spend with their children even after they are reunited because they are working to make ends meet.’’ ‘‘One student came in the 4th grade with no English and joined his father and the new wife and his new U.S.-born little brother. The little brother was ashamed of him and made fun of him in front of the other children at the bus stop for not speaking English. That same boy grew up to become a gang leader.’’ Several teachers also pointed to another perceived problem with children from transnational families (who have family members in both the United States and the home country); a tendency for children to miss part of the school year because they travel to their home country to be with their extended family. Several teachers also noted that many such students return to their home country and family members during the summer break and argued that this travel increased the loss of knowledge most students experience during summer break, making it more difﬁcult for immigrants from transnational families to catch up to fellow students when they return to school in the fall. Teachers also noted that children who immigrate when they were older have a harder time adjusting to the U.S. school system and that these problems are particularly noticeable among those who have been separated from their families. As one teacher wrote, ‘‘It seems to me that middle school is tough enough without having to adjust to ‘new’ parents, new country, new language, and new friends. Younger kids have the advantage of more ﬂexible brains and if they are in the primary grades, having to make less of a cognitive leap.’’ ‘‘The parents are often new at raising teenagers, which is never easy, anyway. They left cute little kids and spent years talking on the phone, but it’s hard to build a new relationship.’’ Another teacher wrote, ‘‘Middle schools are not the friendliest environment. There is no easing into the system. In many middle schools newcomers are easy prey for gang recruiters who offer protection from bullies and a sense of family and belonging.’’ 3.2.4. What existing programs or potential programs would most help in improving the performance of Latin American immigrants in school? Five types of programs were mentioned frequently by teachers as being most helpful in improving the performance of Latin American immigrants

528

Sara Z. Poggio and T.H. Gindling

in school: (1) ESOL programs for students; (2) Spanish-speaking counselors who are familiar with the culture of the Latin American immigrants and are aware of the prevalence of family separation during migration; (3) after-school help with homework for Latin American immigrant children; (4) support for parents, including Spanish-speaking parent liaisons, English classes, and workshops to help parents understand the structure of U.S. school systems; and (5) an ESOL summer school that would work as a ‘‘bridge program’’ to ease the transition of immigrant children into U.S. schools. Most teachers identiﬁed ESOL classes for students as the most helpful existing program for improving the performance of Latin American immigrant students in school. This is consistent with the literature. Many teachers felt that Latin American immigrant students are more likely to conﬁde in and seek the help of counselors if those counselors speak Spanish and are comfortable with the culture from which they came. The best situation is where the counselors themselves are immigrants from the same countries as their students. Interestingly, teachers did not think it is important that the classroom teachers speak Spanish, only that counselors and some staff speak Spanish. Almost all teachers identiﬁed difﬁculties that immigrant children have with homework as a signiﬁcant barrier to the success of these children in school. Teachers also recognized that the parents of these children are unable (or unwilling) to provide their children with much help with homework. As a result, the most common suggestions that we received from teachers designed to improve Latin American student performance were for programs where regular classroom teachers provided homework help after school (in ‘‘extended day’’ programs). Once again, none mentioned that it was important for these teachers to be able to speak Spanish, although several mentioned that it was important for the homework help to be provided by classroom teachers who know what is expected of the students in completing their homework. Several teachers also suggested after-school social activities for teenagers that would insulate middle and high school students from gangs and allow them to socialize in a safe environment. The second most common type of program suggested by teachers was programs ‘‘to help immigrant parents learn English and to navigate the American school system and the communities in which they live.’’ Lack of English language skills on the part of parents contribute to the problems children have completing homework, to the lack of involvement of parents in their children’s school, to solving discipline problems at school, and in general to a lack of communication between teachers, administrators, and parents. Interestingly, teachers often saw the lack of English language skill of parents to be a bigger problem than the lack of English language skill on the part of the children. Perhaps, this is because the school system already has programs in place to help immigrant students learn English, but few programs to help parents. Several teachers suggested that these language

Promoting the Educational Success of Latin American Immigrant Children

529

classes be held at the child’s school. In this way, the classes serve both to increase the language skills of parents and to make the parents more comfortable with increased involvement at school and increases communication with teachers and school administrators. Several teachers also suggested that English language programs for parents would be most successful if they got ‘‘the entire family involved in learning English together.’’ Some suggested ‘‘mommy and me’’ classes for mothers and younger children. To facilitate parents’ understanding of their children’s schools, it is seen as important to have school staff who know the language and culture of the parents of their students. One existing Prince George’s County Public School program that was praised by many of the teachers was the ‘‘parent liaison’’ program. Parent liaisons are Spanish-speaking staff hired by schools to act as an intermediary between parents and teachers and school administrators. They are often parents of former or current students who themselves emigrated from the same countries as the parents of other students at the school and so are not only ﬂuent in the language but also comfortable with the culture of parents. Teachers use liaisons when they need to contact parents; liaisons can be present to translate during parent– teacher conferences; and parents use teacher liaisons to communicate concerns to teachers. One teacher wrote, ‘‘for our school, parent liaisons are most important. We are very lucky to have such a person who is fantastic. She is bilingual and calls the parents whenever it is needed. We have a huge ESOL population here and she is very friendly with the students, they trust her and she is also friendly with the parents. We have systems in place for those students who start to slip out of line so that with parent help, they stay in line.’’ As many parents do not speak English, it is important that liaisons to Spanish-speaking parents also speak Spanish. A similar cultural background is also important. As one teacher wrote, ‘‘if (the liaison) is a well-educated, high status person from South America y there is a distinct disjunction between that person and the kids and parents,’’ who are often from rural areas in Central America and Mexico. Many teachers also mentioned the importance of ‘‘parent workshops’’ to help parents of new immigrant children understand the structure of American school systems (which is often very different from the structure of the school systems in the country of origin). Such workshops should be held at the beginning of the school year and in the evening so that working parents can attend. Both parents and teachers also stressed that such workshops should inform parents that most American school systems expect a high level of parental involvement in their children’s education. Several parents in our focus groups told us that in their country of origin, both teachers and parents view schools and home life as separate, where parents are expected to leave the education of their children to the professional teachers. As U.S. teachers in our survey made clear, however, teachers in the United States view parents as irresponsible if they do not

530

Sara Z. Poggio and T.H. Gindling

help their children complete homework and other projects and do not have an active presence in the school. Several teachers proposed a summer bridge program, limited to new immigrants, to assist in the adjustment from the culture and school system in the home country and the one in the United States. One teacher wrote that ‘‘the International Student Counseling Ofﬁce tries to run acculturation sessions for ESOL newcomers but it is not enough – three one hour sessions.’’ Another called for ‘‘a six week summer bridge program that provides the student with an opportunity to know that he/she is not alone in his/her plight, an opportunity to make friends or acquaintances with peers and instructors prior to attending school, an opportunity to be in the school in which they are registering.’’

4. Conclusions Family separation during migration has a negative impact on the educational success of children that goes beyond the problems experienced by all migrants. Family separation during migration matters and should be taken into account in schools. School counselors in particular should be aware that, compared with nonimmigrants and immigrants who migrate with their parents, children separated during migration are more likely to be depressed, to have difﬁculty adapting to the popular and school culture in the United States (and therefore may be more likely to be attracted into gangs), to have had traumatic experiences during the process of migration, and to have strained relationships with parents and siblings from whom they have been separated. The teachers and parents we surveyed argued that is also important that school counselors (although not necessarily classroom teachers) speak the language and are comfortable with the culture of the immigrant child. If not, students are less likely to trust or accept help from counselors. Useful programs would provide immigrant students with help adjusting to American schools and teen culture, as well as foster a feeling of belonging through connections to peers who model positive behavior. Such programs are not common. ‘‘We have no national policies for helping young immigrants who arrive during the middle and high school years’’ (Suarez-Orozco et al., 2008, p. 360). One successful program directed toward teenage immigrant students who have been separated from their parents during migration can be found in Northwestern High School, in the school district where we conducted our teacher survey, Prince George’s County, Maryland. This program was developed jointly by the Northwestern High School ESOL Intervention Specialist and the Prince George’s County Immigrant School Counseling Ofﬁce speciﬁcally to ease the transition to U.S. schools for immigrants separated during migration. The program includes individual counseling, group counseling sessions, and support groups that include

Promoting the Educational Success of Latin American Immigrant Children

531

peers who have also experienced family separation but have been in the United States for several years. Participants in these support groups, or ‘‘reuniﬁcation groups,’’ compare personal stories and discuss the differences between U.S. school culture and that of the immigrant’s home country, difﬁculties of acculturation, and strategies for success in high school (Bock and Chiancone, 2006). Children separated during migration are more likely to be older than others in their grade. Children separated during migration are also more likely to drop out of high school. These two results are probably related; children who are older than others in their grade are often less motivated to succeed at school, more likely to face pressure to enter the work force, and less likely to complete high school before they reach the maximum age at which they are eligible for free public education (they ‘‘age out’’ of the public school system before they graduate from high school). Immigrant children separated from their parents during migration are more likely to be older than others in their grade for various reasons: they may have repeated a grade either before or after migration, they may have interrupted schooling in their home country to work or take care of family members, they may have lost a year or more of schooling because of the trauma of migrating or because of inconsistencies in the timing of the school year between their home country and in the United States, or they may have been assigned to a lower grade than other children their age when they entered school in the United States. When asked which of these was most important, the most common response we received from parents and teachers was that students were assigned to a lower grade in U.S. schools than they had completed in their home country. Most often, this was due to a low level of English proﬁciency. Our results suggest that separation during migration has a negative effect on the ability of immigrant children who immigrate as teenagers to stay on track to graduate from high school. This implies that it is important to help children who immigrate when they are middle school or high school age to stay in school. Older immigrant students face strong pressure to work to help out their extended families (family members in the United States and by sending remittances to family members abroad). This suggests that one important set of programs to lower high school dropout rates would allow high school students to take classes at night or on the weekend (so as to not interfere with work), to attend high school part time (around work schedules), and to receive free public school education at older ages (it can take immigrant students longer to ﬁnish high school both because they may be working and because they lost years of schooling when they migrated to the United States). Several school systems in the immigrant-rich Washington, D.C., suburbs have programs in high schools that are focused on educating older students during nontraditional school hours. Younger students (up to 21 or 22 years old) who attend regular high schools can generally

532

Sara Z. Poggio and T.H. Gindling

transfer and attend these night high schools tuition free, whereas older students are required to pay tuition. In Prince George’s County, there are four evening high schools, including one at Northwestern High School. In Fairfax County (Virginia), the High School Continuation Program at Arlington Mill High School, although begun in 1929, currently tailors its teaching style to the needs of its primarily immigrant student body. Students at Arlington Mill High School may take special classes for English language learners (HILT), making for an easier transition for recent teenage immigrants into regular high school classes taught in English. Fairfax County (Virginia) public schools offer evening programs at four ‘‘transitional ESOL high schools’’ to provide instruction to older ESOL students (18 and older) who want to earn a high school diploma (the schools are Bryant Alternative High School, Mountain View Alternative High School, Summit Hills Alternative High School, and Woodson Adult High School). The transitional high school programs do not offer a diploma but offer a way to transition into English language high school classes at these alternative high schools. The transitional high school program appears to have a good reputation among teachers in the area. Several teachers from Prince George’s County in Maryland pointed to the transitional high school program in Northern Virginia as effective and a program that should be replicated in their school district. In the focus groups and teacher surveys, the most frequent policy interventions mentioned were programs to help the parents of students understand the structure and expectations of the schools system in the United States (which can be very different from the school systems in their home countries). One frequent problem that arises in this context is that parents may have very poor English proﬁciency. Poor English proﬁciency of immigrant parents was identiﬁed as a problem more often in the focus groups and teacher surveys than poor English proﬁciency of immigrant children. In fact, we were often told that it is the children who translate for the parents in interactions with the school system, not vice versa. Helpful programs mentioned by parents and teachers included ‘‘mommy and me’’ English classes at local schools and after-school English classes for parents at schools (taught by teachers at those schools so that parents become comfortable with their children’s teachers). Teachers and parents also agreed that, to facilitate the participation of parents in the education of their children, it is important to have at least some school staff who speak Spanish and are comfortable with the culture of the immigrant parents. One speciﬁc program identiﬁed as helpful and successful in our qualitative analysis of the Prince George’s County public school system was the parent liaison program, where schools hire a Spanish-speaking staff member (often a current or past parent of a student in that school) to be someone that parents of immigrant children can consult when they have questions about the school and whom teachers can use as go-between to communicate with parents who have limited English skills.

Promoting the Educational Success of Latin American Immigrant Children

533

Another area of concern that was mentioned consistently in the focus groups of parents and teacher surveys was homework. Teachers believe that Hispanic immigrant parents are not involved enough in making sure that their children successfully complete homework assignments. Teachers in the survey often attributed this to a cultural difference between Hispanic immigrants and those born in the United States. The parents in our focus groups also identiﬁed completing homework successfully as a problem, but pointed to a lack of English proﬁciency on the part of parents as the primary reason for this. Parents found it difﬁcult to know how to help and to understand what was required when the homework, instructions, textbooks, and related materials are all in English only. A common request from parents with limited English skills was to have the homework, instructions, and related materials translated into Spanish; without this translation, many immigrant parents ﬁnd it impossible to help students. Many parents and teachers also suggested that immigrant students beneﬁtted greatly from an after-school program that provided help to students on homework – where regular classroom teachers provided homework help in ‘‘extended day’’ programs.

Acknowledgments We are grateful for ﬁnancial support received for this research project from the Spencer Foundation through grant number 200800052; the UMBC Graduate School through a Special Research Initiative Support grant; and the Maryland Institute for Policy Analysis and Research (MIPAR) through a MIPAR Fellowship. Helpful comments were received from Ramona Bock, Patricia Chiancone, Dennis Coates, Lisa Dickson, Dave Marcotte, the participants at the conference on ‘‘Emerging Perspectives on Children in Migratory Circumstances,’’ Drexel University, June 20–22, 2008, and the participants of a UMBC Public PolicyEconomics seminar. Leif Huber, Luis Peralta, Elizabeth Arevalo, Lisa Fink, and Claudia Rybero provided valuable research assistance. Focus group participants were recruited by Evelyn Rosario of the Hispanic Apostolate of Baltimore.

Teacher Questionnaire

Principle Researchers: Sara Poggio, Associate Professor, Department of Modern Languages and Linguistics, UMBC and Tim Gindling, Professor, Department of Economics, UMBC.

1. Approximatively what proportion of your students are immigrants from Latin America?

How many years have you been teaching?

Are you an ESOL teacher?

In what school district?

Grade(s) that you teach:

First, please tell us about yourself.

Appendix

534 Sara Z. Poggio and T.H. Gindling

d. Health problems

c. Behavior or discipline problems

b. Lack of academic preparation

a. Lack of knowledge of English

3. In your opinion, when Latin American immigrant children enter American schools, do they experience more problems than other students because of the following (if so, please provide a short description of the problem):

2. From which Latin American countries do your students come?

Promoting the Educational Success of Latin American Immigrant Children 535

a. From what you know, approximately what percent of your latin American immigrant students have spent time separated from their parents for one year or more because of migration?

5. We are particularly interested to know if the separation of families during the migration process has an impact on later school performance. That is, the parents of many Latin American immigrant children migrate to the U.S. first, leaving their children in the care of other family members or friends. Later, the children also migrate and are reunited with their parents.

b. If so, why? (Please write your answer in the box below.) Possible answers: i. They were assigned to a lower grade than others of thier age group when they arrived in the United States. ii. They repeated a grade while going to school in the United States. iii. They missed a year of school because of the move from their home country. iv. Other (explain)

4. a. Are Latin American immigrant children older than others in their grade?

e. Emotional problems

536 Sara Z. Poggio and T.H. Gindling

7. Who has an easier time in school, children who immigrated as pre-teens (11 years old and younger) or children who immigrated as teenagers (age 12 and above)? Why do you think this is?

6. Do you think that there is a difference between Latin American immigrant girls and boys in the way they adjust to school?

c. Do you notice any special problems in the relationships between such children and their parents that affect school performance?

b. Do you notice any special difficulties such children have in school? If so, can you briefly describe these special difficulties?

Promoting the Educational Success of Latin American Immigrant Children 537

Thank You

10. From your experience, what resources or programs (that are not currently available) would help most in improving the performance of Latin American immigrant students?

9. From your experience, what existing resources or programs help most in improving the performance of Latin American immigrant students?

8. Please, feel free to share with us any thought about factors that in your opinion could affect the school performance of Latin American immigrant students.

538 Sara Z. Poggio and T.H. Gindling

Promoting the Educational Success of Latin American Immigrant Children

539

On subject line of email: Quick survey from UMBC regarding immigrant students. Dear Teacher, We are professors Sara Poggio and Tim Gindling of UMBC and are conducting research about the factors that affect the performance in Maryland schools of children who have migrated from Latin America. We would like to know what you, as teachers who work directly with immigrants, believe are the important factors that affect the success or failure in school of immigrant children from Latin America. Your knowledge and opinions are extremely important to us and will guide us as we decide which factors should be the focus of our research. For these reasons, we would appreciate if you would take a few minutes to answer the questions in a short, anonymous, on-line questionnaire. To access this questionnaire, please click on the following link: http://www.umbc.edu/mll/teachers/ After you have answered the questions to your satisfaction, please click the ‘‘submit’’ button at the bottom of the web page. Be assured that your responses to this questionnaire are completely anonymous: the on-line questionnaire has been designed so that we will not know your name, school or email address. If anyone else at your school would be interested in this survey, please give them the above web address. For any additional questions, you may contact us at the email addresses below. Thank you for your help. Sincerely yours, Tim Gindling ([email protected]), Professor of Economics, University of Maryland Baltimore County (UMBC), and Sara Poggio (poggio@umbc. edu), Associate Professor of Modern Languages and Linguistics, UMBC. References Aizenman, N.C., (2006), Emigration empties Salvadoran Village: left behind are students and the elderly – and silence. Houston Chronicle, May 9, 6. Amuedo-Dorantes, C. (2008), Migration, remittances and children’s schooling in Haiti. IZA Discussion Paper No. 3657, Institute for the Study of Labor (IZA), Bonn, Germany, August. Bock, R., Chiancone, P. (2006), Counseling immigrant adolescents after long term separation from their families. Presentation to the Annual Convention of the American Counseling Association and the Canadian Counseling Association, Montreal, Quebec, Canada, April 1. Boss, P. (1991), Ambiguous loss. In: Walsh, M.M.F. (Ed.), Living Beyond Loss: Death in the Family. W. W. Norton, New York, pp. 164–175.

540

Sara Z. Poggio and T.H. Gindling

Debry, J. (2009), Negotiating work and parenting over the life course: Mexican family dynamic in a binational context. In: Foner, N. (Ed.), Across Generation: Immigrant Families in America. New York University Press, New York, pp. 191–206. Chiswick, B.R., Miller, P.W. (2008), A test of the critical period hypothesis for language learning. Journal of Multilingual and Multicultural Development 9 (2), 16–29. Gindling, T.H., Poggio, S. (2009), Family separation and reuniﬁcation as a factor in the educational success of immigrant children. UMBC/MIPAR Working Paper. Available at http://www.umbc.edu/blogs/mipar/ Final%20Report-Family%20Separation%20and%20Reuniﬁcation.pdf Gindling, T.H., Poggio, S. (2010), The effect of family separation and reuniﬁcation on the educational success of immigrant children in the United States. IZA Discussion Paper No. 4887, Institute for the Study of Labor (IZA), Bonn, Germany, April. Gonzalez, A. (2003), The education and wages of immigrant children: the Impact of age at arrival. Economics of Education Review 22 (2), 203–212. McKenzie, D., Rapoport, H. (2006), Can migration reduce educational attainment? World Bank Policy Research Working Paper No. 3952, June, Washington. Menjivar, C., Abrebo, L. (2009), Parents and children across borders: legal instability and intergenerational relations in Guatemala and Salvadorian families. In: Fonner, N. (Ed.), Across Generation: Immigrant Families in America. New York University Press, New York, pp. 207–226. Miranda, A. (2007), Migrant networks, migrant selection, and high school graduation in Mexico. IZA Discussion Paper No. 3204, Institute for the Study of Labor (IZA), Bonn, Germany, December. Rumbaut, R. (2005a), Children of immigrants and their achievement. In: Taylor, R., Walberg, H.J. (Eds.), Addressing the Achievement Gap: Findings and Applications. Information Age Publishing, Charlotte, NC, pp. 23–59, Chapter 2. Salazar-Parren˜as, R. (2006), Understanding the backlash: why transnational migrant families are considered the ‘Wrong Kind Of Family’ in the Philippines. Working group in childhood and migration. Available at http://globalchild.rutgers.edu/about.htm. Accessed on July 2010. Sawyer, A.D. Keyes, Valasquez, C., Bautista, M. (forthcoming), Going to school, going to El Norte: the impact of migration on education. In: Cornelius, W., Fitzgerald, D., Hernandez-Diaz, J., Borger, S. (Eds.), Between Two Worlds: Oaxacan Migrants and the Mexican Mixteca and California. Center for Comparative Immigration Studies, University of California, San Diego. Schen, C. (2005), When mothers leave their children behind. Harvard Review of Psychiatry 13 (4), 233–243. Scovel, T. (2000), A critical review of the critical period research. Annual Review of Applied Linguistics 20, 213–223.

Promoting the Educational Success of Latin American Immigrant Children

541

Smith, A., Lalonde, R.N., Johnson, S. (2004), Serial migration and its implications for the parent-child relationship: a retrospective analysis of the experiences of the children of Caribbean immigrants. Cultural Diversity and Mental Health 10 (2), 107–122. Suarez-Orozco, C., Suarez-Orozco, M., Todorova, I. (2008), Learning in a New Land: Immigrant Students in American Society. The Belknap Press of Harvard University Press, Cambridge. Suarez-Orozoco, C., Todorava, I., Louie, J. (2002), Making up for lost time: the experience of separation and reuniﬁcation among immigrant families. Family Process 41 (Winter), 625–643. Van Ours, J., Veenman, J. (2006), Age at immigration and educational attainment of young immigrants. Economic Letters 90 (3), 310–316. Vernez, G., Abrahamse, A., Quigley, D. (1996), How Immigrants Fare in U.S. Education, Rand Institute on Education and Training. Rand Monograph Report 718. Available at http://www.rand.org/pubs/ monograph_reports/MR718

CHAPTER 23

Cultural Differences in the Remittance Behaviour of Households: Evidence from Canadian Micro Data Don DeVoretza and Florin Vadeanb,c a

Department of Economics, Simon Fraser University, Burnaby, BC V5A 1S6, Canada E-mail address: [email protected] b Centre for Economic and International Studies, University of Rome Tor Vergata, Rome, Italy c School of Economics, University of Kent, Canterbury Kent CT2 7NP, United Kingdom E-mail address: [email protected]

Abstract This chapter analyses the effect of cultural differences among ethnic groups on the remittance behaviour of native and immigrant households in Canada. In contrast to the New Economic of Labour Migration (NELM) literature that examines remittance motivation in the framework of extended family agreements, we embed remittances in a formal demand system, suggesting that they represent expenditures on social relations with relatives and/or friends and contribute to membership in social/religious organisations respectively. The results indicate strong ethnic group cultural differences in the remittance behaviour of recent Asian immigrant households and highlight the importance of differentiating with respect to cultural background when analysing the determinants of remittances. Keywords: International migration, Household behaviour, Remittances JEL classiﬁcations: C31, D12, F22, F24

1. Introduction The literature on household money transfers to persons outside the household is substantial, analysing the remittance behaviour mainly in terms of motivation. These are categorised to be either altruism, selfinterest, exchange, co-insurance or loan-agreements between extended family members (Lucas and Stark, 1985). The motivation models make different predictions about the effect of speciﬁc determinants on remittances. For example, under the altruistic Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008029

r 2010 by Emerald Group Publishing Limited. All rights reserved

544

Don DeVoretz and Florin Vadean

hypothesis (i.e. the migrant cares about the relatives left behind), once the migrant’s income is taken into account, the education level should not have an effect on remittances (Rapoport and Docquier, 2006). Under the exchange hypothesis, i.e. remittances ‘buy’ various types of services like taking care of the migrant’s relatives or his assets in the home country, the education level is expected to have a negative impact on remittances as educated migrants have a lower propensity to return and thus reunite with their families in the host country and invest less in home country assets (Cox, 1987). Although under the informal family loan hypothesis, more educated migrants are expected to remit more in order to repay for the initial investment made by the family in their education (Poirine, 1997; Cox et al., 1998; Ilahi and Jafarey, 1999). To date, most studies focus on analysing remittance receiving households in a particular migrant sending country and the empirical results are quite diverging: conﬁrming or contradicting the predictions of one or the other remittance motivation models. For example, Lucas and Stark (1985) established that remittances received by households in Botswana rose signiﬁcantly with the migrant’s years of schooling. The effect is even stronger among the recipient household head’s own young (i.e. children, grandchildren, nephews and nieces), giving support to the notion that remittances are partially a result of an understanding to repay initial educational investment. Similarly, Cox et al. (1998) found evidence for the loan repayment hypothesis by analysing remittance receiving households in Peru and distinguishing between parents-to-children and children-toparents transfers. On the contrary, Brown’s (1997) estimation results illustrate that Western Samoan migrants to Australia (conversely to migrants from Tonga) remitted more if they received ﬁnancial assistance from their relatives at home for migration proposes. However, he found no evidence that the level of education attainment could be associated with any difference in remittance behaviour. Research to date suggests that these observed differences in remittance behaviour might be caused by cultural differences in social and/or family norms. Nevertheless, little systematic research has been done so far on the effect of ethnic group cultural differences on the remittance behaviour of households.1 As reﬂected by the Canadian Ethnic Diversity Survey (see Statistics Canada, 2003), Canadian ethnic groups exhibit differential contact with their relatives in their country of origin. For example, 62% of those of Filipino ancestry reported monthly or more frequent contact with their relatives compared to 46% of those of Chinese, 31% of Italian and 20% of German origin.2 And such differences are determined, at least partially, by

1 2

One exception would be Wolff et al. (2007). See Statistics Canada (2003); these numbers are in part reﬂecting time of arrival in Canada.

Cultural Differences in the Remittance Behaviour of Households

545

cultural differences in social/family norms. For example, Elliott and Gray (2000) explain in a report for the New Zealand Immigration Service that the responsibility to care for parents and grandparents is a key component of the family systems in South and South East Asia. Similarly, in Oceania young adults are expected to contribute to both nuclear and extended family commitments. Conversely, in Western societies such family obligations are less important because they have been replaced by welldeveloped social security and ﬁnancial systems. This chapter builds on this literature by assessing the effect of ethnic group cultural differences on the remittances behaviour of native and immigrant households in Canada. For several reasons, we consider the Canadian context appropriate for this exercise. Canada’s foreign-born resident population is large: 5.6 million or approximately 19% of the total population in the 2001 census; the vast majority of these foreign-born residents are admitted into Canada on a permanent basis (96%) and due to quick accession to citizenship over 75% of Canada’s foreign-born population is naturalised. Canada’s immigrant population is thus quite homogeneous in terms of legal resident and citizenship status. However, it is quite diverse in terms of ethnicity. Traditional migration sources are countries from Western and Southern Europe (i.e. UK, Italy, Germany and Portugal) which in 2001 still made up approximately 30% of the stock of foreign-born population. Nevertheless, in the 1980s and 1990s immigration dramatically shifted towards Asian and Central and Eastern European sources, which now represent approximately 38% and 11% of the immigrant population respectively (Citizenship and Immigration Canada, 2002).

2. Theoretical considerations In contrast to extant micro-analysis that models remittance behaviour in the framework of informal agreements between extended family members, we embed remittances in a formal demand system and suggest that they represent expenditures on social relations with relatives and/or friends and contribute to membership in social/religious organisations respectively. This modelling is, nevertheless, consistent with remittance motivation theory, in which remittances are expected to inﬂuence the social relations between family members too. For example in the altruistic model, the degree of altruism of the relatives towards the migrant may inﬂuence and be inﬂuenced by remittances. In the inheritance model, the migrant is assumed to send remittances to maintain a good relationship with the parents and, thus, help insure a possible inheritance. In the exchange model, remittances ‘buy’ various types of services such as taking care of the migrant’s relatives or his assets in the home country. In the co-insurance model, the ﬁnancial support provided by the migrant ensures that relatives will support him in the future in case of need. Similarly, loan repayments

546

Don DeVoretz and Florin Vadean

for the investment of the family in the initial education and/or migration cost could be regarded as assuring his further membership in the family. In line with Deaton and Muelbauer (1993), we allow for a two-stage budgeting process, in order to characterise the household’s remittance decisions with respect to consumption. Thus, in the ﬁrst stage, the household may allocate total expenditures on consumption and on the composite good ‘social relations outside the household’. In the second stage, the expenditures on ‘social relations outside the household’ determined in the ﬁrst stage are distributed between expenditures on social relations with relatives and/or friends and contributions to group membership (i.e. membership in a religious, charitable, professional group, etc.). The differentiation between the expenditures on the two types of social relationships is not only of sociological relevance. The costs involved are also different: whereas contributions to group membership are in the majority of cases tax deductible, transfers to relatives are not. 2.1. The demand system It is a basic premise of this chapter that the act of private remittances is embedded in the household’s utility maximisation framework and is, thus, a part of the household’s allocation process across a general expenditure system. The chosen demand system estimated is the Linear Approximate/ Almost Ideal Demand System (LA/AIDS) proposed by Deaton and Muelbauer (1980), because it satisﬁes the micro-economic theory of demand (i.e. allows for exact aggregation and the imposition of homogeneity and symmetry restrictions) and permits a two-stage budgeting procedure. For the ith commodity, the model can be speciﬁed as follows: X gij ln pj þ bi lnðy=p Þ þ ei (1) w i ¼ ai þ j

where wi ¼ pi qi =y is the budget share of the ith good, pj the price of the jth good, y P represents total expenditures and p* a Stone price index wi ln pi ). To ensure that this demand system conforms to (i.e. ln p ¼ the utility maximisation properties, Equation (1) must satisfy the adding up, homogeneity and symmetry conditions: n X

Adding up :

ai ¼ 1;

i¼1

Homogeneity :

n X i¼1

n X

gij ¼ 0

bi ¼ 0;

n X

gij ¼ 0

(2)

i¼1

(3)

j¼1

Symmetry :

gij ¼ gji

(4)

The adding up conditions are ensured by the P fact that the budget shares of the goods in the system add up to one: wi ¼ 1. Homogeneity and

Cultural Differences in the Remittance Behaviour of Households

547

symmetry have to be tested and they can be parametrically imposed. The LA/AIDS is simple to interpret. In the case of constant relative prices and ‘real’ expenditure (y/p*), the budget shares are constant. This is the natural starting point for the predictions using the model. Changes in real expenditures operate through bi; these add to zero and are positive for luxuries and negative for necessities. Using the estimate bi, Engel elasticities can be calculated as follows: ei ¼ 1 þ ðbi =w i Þ

(5)

where ei is the Engel elasticity and w i the mean share of expenditures on the ith good for the entire sample. The Engel elasticity is greater than unity for luxuries, less than unity for necessities and equal to one for normal goods. 2.2. Demographic controls, immigration entry and assimilation In the demand analysis for various commodities, the LA/AIDS is often supplemented with demographic variables in order to reduce the bias due to unobserved household characteristics (see Teklu, 1996; Adrangi and Rafﬁee, 1997; Meenakshi and Ray, 1999). Following this approach, we additionally estimate a demographically enhanced demand system: wi ¼ ai þ

n X

gij ln pj þ bi lnðy=p Þ þ dik X k þ ei

(6)

j¼1

where Xk represents a set of demographic control variables that depict the life cycle stage of the immigrant and Canadian households, i.e. gender, age, education, marital status, household size, home ownership and net change in assets and liabilities (i.e. savings). Finally, based on the model of Carroll et al. (1994), the demand system is further augmented in order to capture eventual immigration entry and assimilation effects with respect to the remittance behaviour of households: w i ¼ ai þ

n X j¼1

gij ln pj þ bi lnðy=p Þ þ dik X k þ

X ðjis þ yis DÞ IGs þ ei s

(7) where IGs is a dummy variable that is equal to one if the household belongs to immigrant group s and zero otherwise. D denotes the duration of the foreign-born household residence in Canada. Immigrants are assumed to arrive with a set of cultural values and tastes which are different from those of the natives; this is reﬂected by possible non-zero values for jis. Thus, the set of parameters jis can be interpreted ﬁrst as a general immigration entry effect. If jis differs signiﬁcantly across

548

Don DeVoretz and Florin Vadean

immigrant groups, we consider this an evidence for country-/regionspeciﬁc cultural effects as well. Over time, via assimilation, the behaviour of immigrants may become more similar to that of the host group. In our model, this would be the case when the sign of yis is opposite to the sign of jis. In this case, the immigration entry and/or cultural effects would vanish after jis/yis years of residence in the host country. 2.3. Weak separability According to Deaton and Muelbauer (1993), a necessary and sufﬁcient condition for the second stage of the two-stage budgeting process is weak separability. Weak separability of a utility function over a given set of commodities implies that the marginal rate of substitution between any two goods within one group of goods is independent of the level of consumption of any other group of goods. If this condition holds, then it is correct to specify the demand for these product groups separately. The sole connection between the commodity groups is via the income or expenditure effect. Following Hansen (1993), Allen’s partial elasticities of substitution allow us to test for the existence of weak separability. The utility function is weakly separable into the commodity groups A and B if two conditions are satisﬁed: (a) the partial substitution elasticities between different commodities of the groups A and B are identical, i.e. slm ¼ s for all lAA and mAB and (b) the utility sub-functions are homothetic: X X bl ¼ 0 and bm ¼ 0. (8) l

m

From the relation between substitution elasticities and compensated price elasticities we have slm ¼ 1=w m Ylm . The compensated price elasticities are calculated as Yij ¼ w j þ gij =w i for i 6¼ j. Thus, from the condition (a) we obtain the testable restriction: 1 þ glm =w l w m ¼ s.

(9)

To test if the conditions (8) and (9) are satisﬁed, we apply a likelihood ratio (LR) test comparing the system of equations with and without the restrictions imposed. 3. Data and descriptive statistics 3.1. Family expenditure survey (FAMEX) The data sets used for this analysis are taken from the waves 1986 and 1992 of the Family Expenditure Survey (FAMEX), Statistics Canada. Data were collected in the form of a detailed questionnaire during one or

Cultural Differences in the Remittance Behaviour of Households

549

several interviews. Thus, income, expenditure and remittance data in the surveys are self-reported. The focus of the empirical part of this study is to investigate the possible differential patterns of private remittances by Canadian- and foreign-born households. The Canadian-born population is used as a reference group since presumably its members have no immediate attachments abroad. The survey years 1986 and 1992 are of interest because they encompass a dynamic period of expanding Canadian immigration inﬂows which dramatically shifted to Asian source countries.3 The wave 1990 was not included because in comparison to 1986 and 1992 it has observations only from households in urban areas.4 Data from the year 1996 were omitted as well because they do not include information on the immigrant’s year of arrival, which is assumed to signiﬁcantly affect remittances. Only observations with positive and non-zero income and total expenditures were kept in the regressions. Observations with negative expenditures for the different expenditure groups and with ‘masked’ or ‘non-stated’ responses for the variables of interest were excluded as well.5 In addition, the household head is considered to be the member of the household mainly responsible for its ﬁnancial maintenance (i.e. pays the rent, mortgage, property taxes, etc.).6 This deﬁnition of the household head will enable us to categorise a foreign-born household as one in which the ﬁnancial maintenance responsibility is borne by a foreign-born person. The data from the pooled 1986 and 1992 surveys, given the above screening, yield 18,995 surveyed households. Data used in this study do not allow us to differentiate between transfers sent inside or outside Canada. However, we can distinguish between a transfer to a person and to a charity. An inspection of the actual remittance data indicates that some households specialise in the type of transferred funds. Speciﬁcally, 8.5% of the households remit money exclusively to charitable organisations and approximately 17% remit money only to individuals, whereas 66% remit to both individuals and charitable groups.7 We hypothesise that charitable remittances should respond differently to household income since these donations are tax deductible in Canada and do not imply a contractual motive to extended family members. Table 1 reports some descriptive statistics by birth status and for the two survey years included in the study: 1986 and 1992. We are able to

3

In 1968, 75% of Canadian immigrants came from Western Europe and North America and by 1992, 25% came from these regions. 4 A further reason for the omission of the survey year 1990 was the rejection of the test for non-occurrence of structural breaks when pooling the 1986, 1990 and 1992 surveys. 5 Following this screening, a total of 853 observations were excluded. 6 We assume that this person determines also the household’s expenditure patterns. 7 The remaining 8.5% did not make any remittances.

36,189 13,503 1,563 988 395 8,530

HH income after taxes (CA$) Income per HH member (CA$) Net change in assets (CA$)

Remittances to persons (CA$) Remittances to charities (CA$)

Number of observations

7,898

1,033 322

36,404 14,165 1,737

0.43 46.69 2.69 0.63 0.14 0.23 2.57 0.60 N.A.

1992

780

1,173 557

39,012 15,605 3,634

0.33 53.93 2.75 0.62 0.09 0.29 2.50 0.66 27.58

1986

594

1,711 588

37,807 15,819 1,865

0.42 54.44 3.04 0.63 0.08 0.29 2.39 0.66 31.16

1992

North American and West European

405

1,500 225

39,966 13,190 2,432

0.23 51.13 1.94 0.76 0.04 0.20 3.03 0.71 24.72

1986

317

1,322 309

35,784 13,012 1,365

0.32 52.97 2.36 0.72 0.05 0.23 2.75 0.74 28.30

1992

South and East European

233

1,227 292

43,063 10,956 282

0.21 42.00 3.22 0.85 0.09 0.06 3.93 0.55 11.19

1986

238

1,173 316

38,213 11,867 1,222

0.32 45.18 3.18 0.75 0.13 0.12 3.22 0.55 13.36

1992

Chinese, Asian and Oceania

Notes: Education levels are 1 ¼ less than 9 years, 2 ¼ some or completed secondary, 3 ¼ some post-secondary, 4 ¼ Post-secondary degree and 5 ¼ University degree. Monetary values in 1992 Canadian dollars. N.A., not applicable.

0.31 45.60 2.49 0.63 0.15 0.22 2.68 0.56 N.A.

1986

Canadian

Population group

Descriptive statistics by population group (1986/1992; mean values)

Female as HH head (proportion) Age of HH head Education Married with HH member (proportion) Single – never married (proportion) Separated/divorced/widowed (proportion) HH size Home ownership (proportion) Years since immigration

Variable

Table 1.

550 Don DeVoretz and Florin Vadean

Cultural Differences in the Remittance Behaviour of Households

551

differentiate between ﬁve pre-deﬁned population groups: Canadian-born; immigrants from North America and Western Europe; from Southern and Eastern Europe; from China, Asia and Oceania; and others and nonstated.8 The last group was excluded from the analysis since it was deemed too heterogeneous. Group mean values show that the Asian immigrant population contains more males as household heads, is younger and more educated, includes a lower portion of separated/divorced household heads, has households with the largest average size and has a signiﬁcantly shorter immigration history in Canada than the remaining foreign-born groups. Also, Asian immigrant households earn the highest average incomes but save least. However, the greatest average remittances, both to persons and to charities, are made by immigrant households from North America and West Europe. They remitted approximately 35% more than Asian immigrant households in 1992. The North American and West European group have the greatest share of household heads separated or divorced (which we assume to positively affect remittances to persons) and the greatest income per household member (which we assume to positively affect remittances to both persons and charities). Age of the household head seems to signiﬁcantly inﬂuence the remittance activity of the household as well (Figure 1), however, with differences among the population groups. On average Canadian-born and South and East European immigrants make the greatest remittances to persons after age 65 (CA$ 1,375/year and CA$ 1,944/year respectively). Although, North American and West European households remit the greatest average amounts between age 35 and 64 (CA$ 1,678/year). Only Asian immigrant households keep their average remittances to persons quite stable over the whole lifetime. As a share of total expenditure, all population groups remit most after age 65. The share is the biggest for South and East European (9.5%) and the smallest for Asian immigrants (4.7%). Finally, the largest average remittances to charities are made by households in all population groups after age 65 (CA$ 400–600 or 2–3% of the total expenditures).

3.2. Prices The prices used for 8 (of the 10) commodity groups (i.e. Food, Shelter, Household Operations and Furnishing, Clothing, Transportation, Personal and Health Care, Recreation, Education, and Tobacco and Alcoholic 8

The geographical origin of the individual is deﬁned in these broad regional groups by the Canadian Family Expenditure Survey and there is no additional information about the country of birth. Therefore, a disaggregation of the population by smaller regional groups or by country of birth was not possible.

Fig. 1.

0

2000

4000

6000

8000

Age 35-64

Food

Age 34 and less

Age 35-64

Shelter

Age 65 and over

Age 35-64

Age 34 and less HH Op. & Furnishing

Age 65 and over

Age 34 and less

Age 35-64

Clothing

Age 65 and over

Age 35-64

Age 34 and less Transportation

Age 65 and over

Age 34 and less

Age 35-64

Age 65 and over

Age 35-64

Age 34 and less

Heath & Pers. Recreation & Care Educ.

Canadian-born North American and West European South and East European Chinese, Asian and Oceanian

Age 65 and over

Age 35-64 Age 34 and less

Tobacco & Alcohol

Age 35-64 Age 34 and less

Remit. to persons

Age 65 and over

Remit. to charities

Age 34 and less

10000

Age 35-64

12000

Age 65 and over

Age 65 and over

Age 65 and over

Age 34 and less

Mean expenditures by age and population groups. Notes: Values in 1992 Canadian dollars. Source: Own calculations; Family Expenditures Survey (FAMEX) 1986/1992, Statistics Canada.

CA$

552 Don DeVoretz and Florin Vadean

Cultural Differences in the Remittance Behaviour of Households

553

Beverages) included in this study are consumer price indices (CPI) that vary over time and across ﬁve regions (Atlantic Provinces, Quebec, Ontario, Prairies and British Columbia) and are assumed to be ﬁxed within the regions (Table 2). For Remittances to Persons Outside the Household and Remittances to Charities we computed price indices based on the CPIs of the above eight commodity groups. We argue that the value of one remitted dollar to a person outside the household equals to one dollar of forgone consumption. Thus, we calculated for each household in our sample the price index of Remittances to Persons as the sum of the CPIs of the eight expenditure groups presented earlier, weighted by the respective share of the expenditure group in total expenditures. Charitable donations are tax deductible. Thus, the price for one dollar donated to charities equals the value of forgone consumption minus the tax deduction received for the donation of that one dollar. The CPIs for Remittances to Charities are computed as follows: CPIchaor;i ¼ 100þ ðCPIpoh;i 100Þ ð1 Taxri Þ, where: CPIchaor,i is the CPI of Remittances to Charities for the ith household; CPIpoh,i the CPI of Remittances to Persons for the ith household and Taxri stands for the tax rate applicable for the ith household. The tax rates are uniquely computed for each household through a combination of the federal and provincial tax rates. 4. Empirical results LA/AIDS is a system of seemingly unrelated equations with identical regressors and cross-equation restrictions, e.g. gij ¼ gji. For estimating the system, therefore, we use Zellner’s Seemingly Unrelated Regression (SUR). For the dependent variable the following must hold: Swi ¼ 1. This restriction implies further restrictions on the right-hand side, in particular Sei ¼ 0. The residuals are linearly dependent and their covariance matrix is singular (see Hansen, 1993). Green (2002) shows that the solution to the singularity problem is to arbitrarily drop one of the equations and estimate the remainder. The residuals covariance matrix of the system with n1 equation is non-singular. The coefﬁcients of the nth equation result from the ‘adding-up’ restriction. Furthermore, in the SUR model, when all equations have the same regressors, the efﬁcient estimator is single-equation ordinary least squares; i.e. GLS is the same as OLS. Thus, we use in this analysis SUR and OLS alternatively: SUR in most cases, in particular when we impose cross-equation restrictions and OLS for single equation estimation. 4.1. Homogeneity and symmetry One of the tasks of this empirical analysis is to test if the restrictions implied by utility theory hold for the demand equations when including

Atlantic Quebec Ontario Prairies BC

Atlantic Quebec Ontario Prairies BC

1986

1992

98.2 97.8 100.0 98.6 104.7

82.9 87.6 85.7 84.0 88.4

Food

80.4 72.0 100.0 75.1 102.0

68.2 58.4 78.0 62.2 80.3

Shelter

98.1 96.7 100.0 92.1 99.2

85.2 81.2 83.5 77.1 84.5

HH operations and furnishing

96.5 99.7 100.0 102.8 99.8

75.9 74.3 78.3 80.5 81.7

Clothing

75.9 90.1 100.0 77.5 97.9

60.3 79.1 77.3 57.3 63.5

Transportation

Expenditure group

88.7 90.7 100.0 92.2 88.0

71.0 67.9 74.2 68.5 71.0

Personal and health care

Prices indices across Canadian regions: 1986 and 1992

Notes: The base used for the price indices is Ontario 1992. Source: Pendakur (2001), Didukh (2001) and Browning and Thomas (1998a, 1998b).

Region

Year

Table 2.

101.3 100.1 100.0 94.6 97.1

77.7 71.7 75.5 71.5 77.6

Recreation and Education

104.5 101.1 100.0 95.1 104.4

58.3 58.3 54.1 50.7 55.3

Tobacco and alcohol

554 Don DeVoretz and Florin Vadean

Cultural Differences in the Remittance Behaviour of Households

555

the unique expenditure items relating to remittances. The homogeneity restriction is ﬁrst tested by using a LR test comparing the separate OLS regressions for each commodity group in the study, with and without the restriction imposed. Then, we test for homogeneity, symmetry and both homogeneity and symmetry by comparing the SUR estimates for the whole system, with and without the restrictions imposed. The test is undertaken for both the uncontrolled for demographics LA/AIDS model (Equation (1)) and the demographics augmented model (Equation (6)). The test results for the homogeneity and symmetry restrictions are presented in Table 3. Since we assumed different expenditure patterns for the four population groups in the study, we conducted the tests for each group separately. In fact, different results are generated by the restriction tests. In the uncontrolled for demographics setting, when running separate OLS regressions, the hypothesis of homogeneity cannot be rejected at the 95% level in 6 of the 10 equations in the system for the Canadian-born population, all 10 equations for the North American and West European and the South and East European immigrant population, and 9 of the 10 equations for the Asian immigrant population. When running the entire system, the homogeneity restriction cannot be rejected in the case of the North American and West European and the South and East European immigrant groups. Finally, the symmetry restriction is rejected at the 99% level for all population groups. In the controlled for demographic characteristics setting, the tests for homogeneity and symmetry performed similarly. The weak performance of the homogeneity and symmetry tests is not necessarily proof of irrational behaviour. In fact, it might have been caused by the lack of sufﬁcient cross-variation for the price variables (i.e. the prices vary only between 2 years and 5 Canadian provinces/regions) and, thus, leads to large standard errors on the price parameters. Nevertheless, when estimating the expenditure elasticities we will impose the homogeneity and symmetry restrictions (Equations (3) and (4)) parametrically in the SUR model.

4.2. Weak separability The LR-test results show that weak separability cannot be rejected only in the case of Asian immigrant households. The w2 statistic is 10.84 in an unrestricted setting and 8.93 when restricting for homogeneity and symmetry, with both values lower than 95% level critical value. For all other population groups weak separability is rejected by the LR test. This implies that in the case of Asian households remittances to charities are a substitute for remittances to persons. This behaviour actually resembles Muslim charity traditions. According to the Koran, 2–5% of the income should be donated to the poor (including extended

System Homogeneity Symmetry Homogeneity and symmetry

Uncontrolled setting Food Shelter HH Operation and furnishing Clothing Transportation Health and personal care Recreation and education Tobacco and alcohol Remittances to persons Remittances to charities

Commodity group

32.43 7,523.39 10,358.00

5.62 5.22 2.29 2.11 0.29 0.86 2.18 1.04 7.08 8.35 0.000 0.000 0.000

0.018 0.022 0.130a 0.147a 0.591a 0.355a 0.140a 0.307a 0.008 0.004 5.37 531.94 795.41

1.36 1.05 1.20 0.38 0.54 0.92 0.14 0.01 0.01 0.29 0.801a 0.000 0.000

0.244a 0.307a 0.274a 0.536a 0.462a 0.337a 0.713a 0.940a 0.941a 0.591a

p-value

w2

w2 p-value

North American and West European

Population

3.28 2.14 0.86 2.37 0.04 1.12 0.09 0.02 1.62 0.79 9.56 190.65 321.71

w2

0.387a 0.000 0.000

0.070a 0.143a 0.353a 0.124a 0.849a 0.291a 0.770a 0.881a 0.203a 0.373a

p-value

South and East European

Test for homogeneity and symmetry restrictions

Canadian

Table 3.

24.83 329.48 388.63

1.35 0.87 12.10 0.04 2.85 0.02 1.68 0.31 0.53 2.64

w2

0.003 0.000 0.000

0.246a 0.350a 0.001 0.841a 0.091a 0.902a 0.195a 0.578a 0.466a 0.104a

p-value

Chinese, Asian and Oceania

556 Don DeVoretz and Florin Vadean

a

0.000 0.000 0.000

System Homogeneity Symmetry Homogeneity and symmetry

w-value smaller than the 95% critical level.

72.98 6,289.78 7,426.94

0.184a 0.000 0.008 0.004 0.427a 0.071a 0.574a 0.229a 0.001 0.000

Controlled for demographic characteristics Food 1.76 Shelter 26.89 HH operation and furnishing 6.95 Clothing 8.37 Transportation 0.63 Health and personal care 3.27 Recreation and education 0.32 Tobacco and alcohol 1.45 Remittances to persons 10.94 Remittances to charities 20.51 10.00 446.69 506.40

2.02 4.04 1.79 1.36 0.76 0.32 0.10 0.00 0.51 0.79 0.351a 0.000 0.000

0.155a 0.044 0.181a 0.244a 0.382a 0.572a 0.757a 0.976a 0.477a 0.374a 11.65 158.00 179.64

1.74 3.36 0.61 6.39 0.23 1.60 0.43 0.12 0.84 0.05 0.234a 0.000 0.000

0.187a 0.067a 0.435a 0.012 0.629a 0.207a 0.511a 0.726a 0.360a 0.819a 26.47 291.03 337.09

0.82 0.05 13.91 0.04 2.77 1.89 1.83 1.18 0.35 2.02

0.002 0.000 0.000

0.366a 0.817a 0.000 0.849a 0.096a 0.169a 0.176a 0.277a 0.556a 0.155a

Cultural Differences in the Remittance Behaviour of Households 557

558

Don DeVoretz and Florin Vadean

family members) or an Islamic cause; this Islamic tax is known as Zaqqat (ECORYS, 2005). However, Asian immigrants in Canada are predominately from China, Hong Kong, India and the Philippines and belong to other religious groups. Therefore, it is not straightforward how the above explanation applies to non-Muslim Asian immigrants. Another explanation for the weak separability in the case of Asian households is owing to the quite low variance in the transfers share to relatives and friends (see Sections 4.3 and 4.4). Because transfers to charities represent only approximately 20% of the total transfers outside the household, their variance has a less signiﬁcant impact on the weak separablility test results. In the remainder of the chapter, the demand sub-system for relations with relatives and/or friends and group membership is speciﬁed for Asian households separately. The LA/AIDS system thus contains only two equations (one for the share of remittances to persons and the other for the share of remittances to charities) and has total remittances as an independent argument (instead of total expenditures).

4.3. Expenditure elasticities Engel elasticities for Canadian- and foreign-born residents across income groups are estimated in an LA/AIDS system, under an uncontrolled as well as a controlled setting. Table 4 reports the estimated expenditure elasticities for the pooled 1986 and 1992 surveys without imposing restrictions for homogeneity and symmetry.9 The estimated expenditure elasticities with restrictions imposed (Table 5) mimic those of the unrestricted estimates. If the model is correct and demographic arguments condition remittances then signiﬁcant differences should arise between the controlled and uncontrolled elasticity measures. And indeed, expenditure elasticities for remittances to persons and remittances to charity/religious groups are in a controlled setting up to two times greater than estimates derived in an uncontrolled one. In the remainder of this section, we would like to focus on the ﬁrst set of estimates because they reﬂect the net income effect on the remittance activity more accurately. The results are differentiated by foreign-born status and income group to capture any effects owing to the immigrant origins or their position in Canada’s income distribution. Given these categories, the range of calculated values for the expenditure elasticities for remittances to persons greatly exceed unity for the Canadian-born households and the 9

Elasticity estimates for the traditional goods on the basis of FAMEX as reported by Didukh (2001, 2002) and Geiger (2002) over a wide variety of commodities are within the range reported here.

Food Shelter HH operation and furnishing Clothing Transport Health and personal care Recreation Tobacco and alcohol Remittances to persons Remittances to charities

Food Shelter HH operation and furnishing Clothing Transport Health and personal care Recreation Tobacco and alcohol Remittances to persons Remittances to charities

North American and West European

Expenditure group

0.78 0.63 1.14 1.24 1.51 0.89 1.46 1.04 1.11 0.72

0.74 0.60 1.06 1.26 1.68 0.92 1.40 0.93 1.13 0.60 0.72 0.77 1.15 1.12 1.41 0.68 1.40 0.72 0.98 0.31

0.69 0.67 1.08 1.18 1.57 0.75 1.33 0.87 1.09 0.40 0.74 0.62 1.23 1.33 1.63 0.92 1.59 1.15 1.24 0.81

0.74 0.58 1.07 1.31 1.91 0.98 1.46 1.00 1.32 0.60 0.63 0.65 1.14 1.18 1.48 0.85 1.39 1.01 1.91 1.17

0.63 0.61 1.03 1.27 1.65 0.90 1.36 1.00 1.85 1.02

0.62 0.68 1.20 1.08 1.43 0.59 1.34 0.91 1.83 0.60

0.62 0.66 1.11 1.22 1.48 0.73 1.33 1.04 1.45 0.84

Top Y/2

All

Bottom Y/2

All

Top Y/2

Controlled

Uncontrolled

Income group

Expenditure elasticities calculated from LA/AIDS, unrestricted (1986/1992)

Canadian

Population group

Table 4.

0.59 0.69 1.23 1.41 1.47 1.00 1.49 1.07 1.82 1.20

0.64 0.63 1.06 1.35 1.81 1.00 1.38 0.98 1.95 0.98

Bottom Y/2

Cultural Differences in the Remittance Behaviour of Households 559

Remittances to persons Remittances to charities

Chinese, Asian and Oceania 1.08 0.77

0.78 0.50 1.14 1.36 1.71 1.00 1.53 1.10 0.93 0.38 1.08 0.80

0.68 0.52 1.16 1.27 1.65 0.86 1.41 1.17 0.60 0.37 1.10 0.71

0.79 0.52 1.20 1.33 1.99 1.03 1.43 1.08 1.20 0.47 1.09 0.75

0.67 0.49 1.11 1.34 1.59 0.96 1.41 1.10 2.03 0.86

1.08 0.79

0.63 0.55 1.10 1.29 1.48 0.83 1.29 1.60 1.55 0.30

Top Y/2

All

Bottom Y/2

All

Top Y/2

Controlled

Uncontrolled

Income group

1.10 0.69

0.71 0.53 1.23 1.47 1.67 1.01 1.22 0.70 2.31 0.95

Bottom Y/2

Notes: Elasticities are computed using the formula ei ¼ 1 þ ðbi =w i Þ, where w i is the mean share of expenditures on the ith good for the entire sample and bi the estimated household total expenditures coefﬁcient.

Food Shelter HH operation and furnishing Clothing Transport Health and personal care Recreation Tobacco and alcohol Remittances to persons Remittances to charities

Expenditure group

South and East European

Population group

Table 4. (Continued )

560 Don DeVoretz and Florin Vadean

Expenditure group

Food Shelter HH operation and furnishing Clothing Transport Health and personal care Recreation Tobacco and alcohol Remittances to persons Remittances to charities

Food Shelter HH operation and furnishing Clothing Transport Health and personal care Recreation Tobacco and alcohol Remittances to persons Remittances to charities

Canadian

North American and West European

0.78 0.68 1.08 1.24 1.54 0.85 1.40 1.07 1.06 0.56

0.75 0.65 1.07 1.30 1.58 0.91 1.40 0.89 1.17 0.62 0.68 0.81 1.09 1.12 1.49 0.64 1.36 0.80 1.04 0.15

0.65 0.76 1.03 1.17 1.60 0.73 1.23 0.84 1.13 0.31 0.69 0.68 1.14 1.24 1.82 0.88 1.47 1.37 1.13 0.50

0.71 0.62 1.03 1.29 1.95 0.96 1.42 1.03 1.28 0.50 0.61 0.72 1.11 1.11 1.62 0.76 1.27 0.88 1.77 1.03

0.61 0.71 1.03 1.22 1.66 0.84 1.28 0.91 1.78 0.97

0.60 0.72 1.16 1.05 1.60 0.51 1.25 0.80 1.64 0.42

0.58 0.77 1.09 1.14 1.58 0.67 1.18 0.90 1.43 0.83

Top Y/2

All

Bottom Y/2

All

Top Y/2

Controlled

Uncontrolled

Income group

0.54 0.76 1.22 1.24 1.74 0.92 1.32 0.90 1.66 1.06

0.59 0.70 1.05 1.25 1.91 0.95 1.29 0.89 1.81 0.90

Bottom Y/2

Expenditure elasticities calculated from LA/AIDS, restricted for homogeneity and symmetry (1986/1992)

Population group

Table 5.

Cultural Differences in the Remittance Behaviour of Households 561

Food Shelter HH operation and furnishing Clothing Transport Health and personal care Recreation Tobacco and alcohol Remittances to persons Remittances to charities

Remittances to persons Remittances to charities

South and East European

Chinese, Asian and Oceania

1.07 0.80

0.79 0.56 1.12 1.35 1.66 0.98 1.51 1.12 0.85 0.25 1.07 0.82

0.66 0.55 1.08 1.20 1.74 0.81 1.33 1.24 0.65 0.35 1.10 0.71

0.74 0.57 1.15 1.29 2.04 1.01 1.39 1.20 1.19 0.34 1.09 0.76

0.65 0.57 1.08 1.29 1.66 0.90 1.35 1.03 1.89 0.70

1.08 0.79

0.62 0.55 1.06 1.20 1.69 0.69 1.16 1.47 1.52 0.39

Top Y/2

All

Bottom Y/2

All

Top Y/2

Controlled

Uncontrolled

Income group

1.10 0.69

0.65 0.62 1.20 1.37 1.81 0.97 1.14 0.54 2.15 0.81

Bottom Y/2

Notes: Elasticities are computed using the formula ei ¼ 1 þ ðbi =w i Þ, where w i is the mean share of expenditures on the ith good for the entire sample and bi the estimated household total expenditures coefﬁcient.

Expenditure group

Population group

Table 5. (Continued )

562 Don DeVoretz and Florin Vadean

Cultural Differences in the Remittance Behaviour of Households

563

North American and all European immigrant households, which seem to treat social relations to persons outside the household as a luxury. At the same time, elasticity estimates for remittances to persons in the Asian case are close to unity, meaning that they consider expenditures in these social ties as a normal good. The results indicate signiﬁcant differences in the remittance activity of the population groups across the cited income classes and imply that households value differently the relationships with relatives and/or friends outside the household, dependent on their cultural background. On the one hand, for the North American and all European immigrant households, the relationship among the household members (i.e. the nuclear family) appears to be of primary importance and only when total household consumption is large enough do these households become more generous towards other relatives and friends. On the other hand, for Asian households, the remitted share to persons outside the household is more stable with changes in total expenditure, which could be evidence of stronger ties with their extended family. The estimated expenditure elasticities to charities of all households in the top income half are below unity, implying that they consider group membership a necessity. This is actually in line with the general experience, that religious participation weakens (or at least it does not strengthen) as a person/household becomes wealthier. However, for households in the bottom income half, the elasticity is around unity for the Canadian-born and the South and East European (for North American and West European even exceeding unity), meaning that these households increase charitable spending probably as a means to improve their status in their social group as their income rises. Asians are again an exception with households in both income halves treating remittances to charities as a necessity.

4.4. Demographic controls We now turn to the effects of household demographic characteristics on remittance behaviour. We argue that remittances are embedded in the household’s life cycle experience and illustrate it with a series of simulations. These simulations are depicted in Figures 2 and 3 and are constructed from the reported estimates for remittances to persons and to charities in Tables 6 and 7. In short, for each representative household we place the mean values for all the model’s variables (except age and agesquared) and cross multiply by the relevant coefﬁcients. This produces the household’s estimated remittances share by age for its constituent parts. Figure 2 reveals several important features of the remittance experience over time and across various population groups. We note that the share of remittances to persons of Asian households has the lowest variance over

564

Don DeVoretz and Florin Vadean

Fig. 2. Expenditure share of remittances to persons by population group over the life cycle. Source: Own calculations; Family Expenditures Survey (FAMEX) 1986/1992, Statistics Canada.

Fig. 3. Expenditure share of remittances to charities by population group over the life cycle. Source: Own calculations; Family Expenditures Survey (FAMEX) 1986/1992, Statistics Canada.

Female

Log of price for remittances to charities

Log of price for remittances to persons

Log of price for tobacco and alcohol

Log of price for recreation

Log of price for health and personal care

Log of price for transportation

Log of price for HH operation and furnishing Log of price for clothing

Log of price for shelter

Log of price for food

Log of total remittances

0.459 [0.183]** 0.236 [0.080]*** 1.946 [0.804]** 0.302 [0.135]** 0.049 [0.040] 0.122 [0.053]** 0.532 [0.293]* 0.900 [0.367]** 0.096 [0.066] 0.109 [0.096]

0.004 [0.001]***

Uncontrolled

0.590 [0.175]*** 0.260 [0.076]*** 2.614 [0.764]*** 0.329 [0.128]** 0.051 [0.038] 0.170 [0.049]*** 0.803 [0.278]*** 1.165 [0.347]*** 0.145 [0.062]** 0.026 [0.096] 0.002 [0.001]*

0.027 [0.002]***

Controlled

Canadian

0.832 [0.654] 0.264 [0.270] 2.027 [2.689] 0.768 [0.492] 0.062 [0.135] 0.175 [0.176] 0.023 [0.983] 1.448 [1.238] 0.319 [0.196] 0.323 [0.290]

0.004 [0.005]

Uncontrolled

0.516 [0.582] 0.107 [0.241] 0.660 [2.402] 0.392 [0.440] 0.021 [0.121] 0.079 [0.166] 0.367 [0.888] 0.725 [1.101] 0.426 [0.210]** 0.355 [0.316] 0.004 [0.005]

0.038 [0.007]***

Controlled

North American and West European

0.129 [1.199] 0.051 [0.488] 0.181 [4.843] 0.046 [0.936] 0.042 [0.234] 0.241 [0.423] 0.163 [1.768] 0.100 [2.261] 0.432 [0.438] 0.633 [0.640]

0.003 [0.007]

Uncontrolled

0.488 [1.241] 0.094 [0.485] 1.348 [4.749] 0.267 [0.959] 0.083 [0.228] 0.073 [0.371] 0.243 [1.739] 0.641 [2.226] 0.614 [0.367]* 0.655 [0.533] 0.013 [0.009]

0.048 [0.011]***

Controlled

South and East European

2.938 [1.057]*** 4.653 [1.574]***

0.062 [0.014]***

Uncontrolled

0.859 [1.168] 1.593 [1.726] 0.010 [0.039]

0.067 [0.013]***

Controlled

Chinese, Asian and Oceania

Regression equation coefﬁcients (OLS) predicting the expenditure share of remittances to persons (1986/1992)

Log of total expenditures

Table 6.

Cultural Differences in the Remittance Behaviour of Households 565

16,428 0.02

Observations R2

Note: Robust standard errors in brackets * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

1.753 [0.465]***

Constant

Years since immigration 100

Log of net change in A&L

House ownership

Number of persons a member

Separated/divorced/widowed

Married (with HH member) 100

Education 10

Age2 1,000

Age 100

Uncontrolled

16,428 0.14

2.324 [0.584]***

0.168 [0.023]*** 0.025 [0.003]*** 0.002 [0.004] 0.002 [0.142] 0.013 [0.002]*** 0.010 [0.001]*** 0.001 [0.001] 0.049 [0.028]*

Controlled

Canadian

1,374 0.02

1.322 [1.513]

Uncontrolled

1,374 0.11

0.068 [0.105] 0.014 [0.011] 0.027 [0.017] 0.411 [0.584] 0.016 [0.006]** 0.014 [0.002]*** 0.003 [0.004] 0.046 [0.069] 0.023 [0.020] 0.486 [1.542]

Controlled

North American and West European

Table 6. (Continued )

722 0.03

3.891 [2.471]

Uncontrolled

722 0.17

0.411 [0.164]** 0.053 [0.018]*** 0.047 [0.022]** 0.093 [1.261] 0.014 [0.015] 0.014 [0.003]*** 0.009 [0.009] 0.135 [0.091] 0.006 [0.030] 1.325 [3.167]

Controlled

South and East European

417 0.07

8.450 [2.722]***

Uncontrolled

417 0.15

0.501 [0.895] 0.030 [0.092] 0.542 [0.123]*** 4.130 [5.898] 0.005 [0.077] 0.024 [0.013]* 0.053 [0.039] 0.013 [0.225] 0.405 [0.219]* 4.551 [3.793]

Controlled

Chinese, Asian and Oceania

566 Don DeVoretz and Florin Vadean

0.504 [0.100]*** Log of price for shelter 0.261 [0.042]*** Log of price for HH operation and 2.410 furnishing [0.423]*** Log of price for clothing 0.250 [0.071]*** Log of price for transportation 0.078 [0.021]*** Log of price for health and personal care 0.155 [0.027]*** Log of price for recreation 0.842 [0.154]*** Log of price for tobacco and alcohol 0.985 [0.193]*** Log of price for remittances to persons 0.204 [0.041]*** Log of price for remittances to charities 0.348 [0.059]*** Female

Log of price for food

Log of total remittances

0.005 [0.001]***

Uncontrolled

0.296 [0.096]*** 0.155 [0.040]*** 1.460 [0.407]*** 0.119 [0.068]* 0.034 [0.020]* 0.120 [0.026]*** 0.487 [0.149]*** 0.601 [0.185]*** 0.033 [0.042] 0.089 [0.062] 0.001 [0.001]

0.001 [0.001]

Controlled

Canadian

0.541 [0.400] 0.156 [0.166] 1.208 [1.702] 0.360 [0.313] 0.025 [0.083] 0.109 [0.114] 0.172 [0.609] 0.719 [0.800] 0.046 [0.111] 0.027 [0.159]

0.005 [0.002]**

Uncontrolled

0.299 [0.385] 0.041 [0.163] 0.192 [1.677] 0.167 [0.300] 0.020 [0.081] 0.073 [0.114] 0.162 [0.601] 0.258 [0.782] 0.211 [0.121]* 0.223 [0.174] 0.002 [0.003]

0.003 [0.004]

Controlled

North American and West European

0.085 [0.473] 0.058 [0.178] 1.191 [1.921] 0.059 [0.375] 0.070 [0.091] 0.131 [0.145] 0.709 [0.676] 0.426 [0.939] 0.046 [0.114] 0.091 [0.143]

0.007 [0.003]***

Uncontrolled

0.042 [0.473] 0.067 [0.182] 1.033 [1.913] 0.120 [0.359] 0.043 [0.082] 0.105 [0.141] 0.615 [0.708] 0.370 [0.907] 0.011 [0.110] 0.042 [0.162] 0.004 [0.004]

0.002 [0.003]

Controlled

South and East European

2.938 [1.057]*** 4.653 [1.574]***

0.062 [0.014]***

Uncontrolled

0.859 [1.168] 1.593 [1.726] 0.010 [0.039]

0.067 [0.013]***

Controlled

Chinese, Asian and Oceania

Regression equation coefﬁcients (OLS) predicting the expenditure share of remittances to charities (1986/1992)

Log of total expenditures

Table 7.

Cultural Differences in the Remittance Behaviour of Households 567

16,428 0.02

Observations R2

Note: Robust standard errors in brackets * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

1.495 [0.270]***

Constant

Years since immigration 100

Log of net change in A&L

House ownership

Number of persons a member 100

Separated/divorced/widowed

Married (with HH member)

Education 10

Age 1,000

2

Age 100

Uncontrolled

16,428 0.11

0.765 [0.251]***

0.056 [0.013]*** 0.011 [0.001]*** 0.030 [0.002]*** 0.005 [0.001]*** 0.007 [0.001]*** 0.004 [0.021] 0.003 [0.001]*** 0.033 [0.007]***

Controlled

Canadian

1,374 0.02

1.040 [0.881]

Uncontrolled

1,374 0.08

0.151 [0.059]*** 0.020 [0.006]*** 0.034 [0.013]*** 0.008 [0.007] 0.009 [0.007] 0.076 [0.097] 0.005 [0.003] 0.015 [0.018] 0.030 [0.010]*** 0.596 [0.924]

Controlled

North American and West European

Table 7. (Continued )

722 0.06

0.591 [0.934]

Uncontrolled

722 0.16

0.118 [0.046]*** 0.018 [0.006]*** 0.026 [0.014]* 0.014 [0.014] 0.024 [0.017] 0.025 [0.088] 0.002 [0.003] 0.013 [0.017] 0.009 [0.009] 0.102 [1.053]

Controlled

South and East European

417 0.07

7.450 [2.722]***

Uncontrolled

417 0.15

0.501 [0.895] 0.030 [0.092] 0.542 [0.123]*** 0.041 [0.059] 0.005 [0.077] 2.383 [1.294]* 0.053 [0.039] 0.013 [0.225] 0.405 [0.219]* 3.551 [3.793]

Controlled

Chinese, Asian and Oceania

568 Don DeVoretz and Florin Vadean

Cultural Differences in the Remittance Behaviour of Households

569

lifetime. Moreover, from all population groups, Asian households remit to persons the greatest share of expenditures over the active lifetime of the household head (i.e. until age 60).10 Both these could be a sign of contributions to the extended family, whose size is more stable over their lifetime. Non-Asian households’ transfers to persons increase dramatically as the age of the household head exceeds 50. This result may arise as members of the nuclear family (i.e. own children) leave the household. The largest transfers are, however, made after retirement age, perhaps as inter vivo transfers to heirs. These simulated patterns conform to our earlier reported stylised facts (Figure 1). To wit, the Canadian-born increase their remittances to persons from an average of CA$ 700/year under the age of 34 to around CA$ 1,050 between age 35 and 64 and further to about CA$ 1,375/year after age 65. Similarly, South and East Europeans increase their remittances to persons from an average of about CA$ 900/year under the age of 34 to about CA$ 1,300/year between age 35 and 64 and almost CA$ 2,000/year over age 65. The remittances to persons sent by North American and West European immigrant households reach a maximum at midlife (ca. CA$ 1,700/year) and fall again after age 65 to about CA$ 1,250/year. Although those of Asian households being quite stable among age groups, at values between CA$ 1,100 and 1,250/year. From Figure 1 we should further note that the substantial increase of expenditure shares remitted to persons after age 65 observed from the simulation is partly due to the signiﬁcant decrease in all expenditures (except remittances to persons and to charities). The possible explanation that the share of expenditures remitted to person increases with the number of the close family members living outside the household is also conﬁrmed for the Canadian-born and North American and West European households by the positive sign of the coefﬁcient of the separated/divorced dummy. This implies that if the spouse lives outside the household or the household head is divorced,11 the household remits a signiﬁcantly higher share of its expenditures to persons outside the household. Another important result is that education negatively affects the budget share remitted to persons in the case of South and East European and 10

The F-tests employed conﬁrm the existence of signiﬁcant differences in means between the predicted values. 11 The FAMEX marital status group includes widowed persons as well. However, we expect that this will not bias our results. Both separated, divorced and widowed household heads might have a higher propensity to remit. Separated and divorced household heads might remit more because they have a greater number of close relatives (i.e. [ex]spouse and children) living outside the household. Similarly, widowed household heads might invest more in relations to persons outside the household (i.e. remit more) in order to substitute for their loss of social relations within the household.

570

Don DeVoretz and Florin Vadean

Asian immigrant households, conﬁrming the prediction of the exchange hypothesis (see Cox, 1987). Under the exchange hypothesis, because more educated migrants have a lower propensity to return, they are less likely to invest in home country assets and likely to reunite with their close family members in the host country, both negatively affecting remittances. Figure 3 depicts the simulated charitable remittances for various households. In general, all population groups increase their minuscule charitable donations from approximately 0.5% at age 25 to around 2–3% at age 75. Furthermore, all immigrant groups remit slightly less to charities compared to the Canadian-born, with no sign of convergence over time.

4.5. Immigration entry and assimilation effects We ﬁnally estimate the augmented share equation (Equation (7)) with the immigration entry and assimilation effects. Table 8 reports the estimation results for the expenditure share of the remittances to persons, the expenditure share of the remittances to charities, and the related F-test comparing the immigrant group coefﬁcients (i.e. the entry effects) and the interactions of the immigrant group coefﬁcients with the variable for the time spent in Canada (i.e. the assimilation effects). The immigrant group coefﬁcient for remittances to persons is signiﬁcant only for the Asian households. This indicates that at the time of entry, their expenditure share remitted to persons is 1.7% higher compared to that of Canadian-born households (and implicitly also 1.7% higher compared to other immigrant households). The coefﬁcients are signiﬁcantly different between immigrant groups. The w2 statistic of the F-test being 3.54 and, thus, greater than the 95% critical value. This implies the existence of ethnic group cultural differences in the remittance behaviour of households at time of entry. There is no evidence of assimilation between the foreign- and the Canadian-born remittance behaviour over time. In the case of immigrant households from Southern and Eastern Europe, the remittance behaviour difference grows over time. Each additional year spent in Canada increases their expenditure share remitted to persons by 6.2%. The w2 statistic derived from comparing the convergence patterns is 3.66 and thus greater than the 95% critical value. This implies that there exist ethnic group cultural differences with respect to the speed of assimilation to the Canadian-born remittance norm as well. Regarding remittances to charities, all foreign-born households remit a slightly smaller share of expenditures (0.5–0.7%) compared to Canadianborn households. However, the w2 statistic of both F-tests is lower than 95% critical value, showing that there is no evidence for ethnic group cultural differences in the remittance behaviour of households to charities.

Cultural Differences in the Remittance Behaviour of Households

Table 8.

Entry and assimilation effects (1986/1992) Share of remittances to persons OLS

Log of total expenditures

571

0.028 [0.002]*** Log of price for food 0.688 [0.163]*** Log of price for shelter 0.276 [0.070]*** Log of price for HH operation and furnishing 2.686 [0.708]*** Log of price for clothing 0.384 [0.121]*** Log of price for transportation 0.065 [0.035]* Log of price for health and personal care 0.149 [0.047]*** Log of price for recreation 0.782 [0.257]*** Log of price for tobacco and alcohol 1.242 [0.323]*** Log of price for remittances to persons 0.171 [0.057]*** Log of price for remittances to charities 0.018 [0.088] Female 0.002 [0.001]* Age 100 0.171 [0.022]*** 2 Age 1,000 0.025 [0.002]*** Education 10 0.024 [0.040] Married (with HH member) 0.027 [0.134] Separated/divorced/widowed 0.013 [0.002]*** Number of persons a member 100 0.103 [0.004]*** House ownership 0.001 [0.001] Log of net change in A&L 0.029 [0.024]

Share of remittances to charities

F-test OLS (p-value) 0.001 [0.001] 0.261 [0.087]*** 0.132 [0.037]*** 1.218 [0.373]*** 0.100 [0.064] 0.023 [0.018] 0.104 [0.024]*** 0.385 [0.135]*** 0.508 [0.170]*** 0.012 [0.037] 0.060 [0.055] 0.001 [0.001] 0.064 [0.012]*** 0.012 [0.001]*** 0.296 [0.024]*** 0.498 [0.113]*** 0.007 [0.001]*** 0.001 [0.002] 0.003 [0.001]*** 0.030 [0.006]***

F-test (p-value)

572

Don DeVoretz and Florin Vadean

Table 8. (Continued ) Share of remittances to persons OLS

North American and West European (NAWE) South and East European (SEE) Chinese, Asian and Oceania (CAO)

NAWE years since immigration SEE years since immigration CAO years since immigration

Constant

Observations R2

F-test OLS (p-value)

0.004 [0.005] 3.54 0.006 [0.006] 0.017 (0.029) [0.006]*** 0.015 [0.018] 0.062 [0.027]** 0.052 [0.048] 2.420 [0.522]*** 18,995 0.13

Share of remittances to charities

3.66

(0.026)

F-test (p-value)

0.007 [0.003]** 0.15 0.005 [0.002]*** 0.005 (0.863) [0.002]*** 0.024 [0.011]** 0.011 [0.009] 0.009 [0.016]

0.49

(0.611)

0.726 [0.224]*** 18,995 0.11

Note: Robust standard errors in brackets. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

5. Conclusions Most studies on remittance behaviour have used household data in a particular migrant sending country, thus analysing the behaviour of a culturally homogenous population. The empirical results vary signiﬁcantly between the studied populations, leading to suggestions that the disparities might be attributable to cultural differences. However, little systematic research, if any, was done to directly test the hypothesis that cultural differences play a discernable role in the remittance behaviour of migrants. We have tested for cultural effects by comparing the remittance behaviour of immigrants to Canada who come from different world regions, and therefore represent different cultures. The empirical results suggest signiﬁcant differences in the remittance behaviour among the population groups. Expenditure elasticities computed separately for each immigrant group reveal that Asian households consider remittances to relatives and/or friends a normal good, while all other immigrants and the Canadian-born regard them a luxury good. Moreover, running

Cultural Differences in the Remittance Behaviour of Households

573

estimations with pooling all population groups and controlling for immigrant groups and time spent in Canada shows that Asian households remit to persons a greater share of their expenditures at the time of arrival, with no evidence of convergence to the Canadian-born norm over time. We must note that in the Canadian context remittances represent only a small share of the immigrant households’ budget until retirement years, most probably because the vast majority of immigration to Canada is permanent, with a generous family reuniﬁcation policy. Nevertheless, our ﬁndings give additional insights into the transfer behaviour of permanent migrants in general and have important policy implications. The differential response with respect to changes in total expenditures (or income) suggests that during periods of economic downturn in migrant host countries – like the one we are currently passing through – migrants originating from countries with a nuclear family tradition (and/or with more developed social systems) would probably decrease their private monetary transfers more dramatically. These differences in transfer behaviour will certainly change the geography of international remittance ﬂows. Recent World Bank estimations of regionally aggregated remittance ﬂows to developing countries conﬁrm these expectations. Remittance ﬂows to developing countries in Europe and Central Asia are estimated to have fallen in 2009 by about 14.7%, while migrants’ remittances to South and East Asia and Paciﬁc by only from 1.5% to 1.8% (Ratha et al. 2009).

Acknowledgments Support from Friedrich Naumann Foundation, IZA, Bonn, HWWA/ HWWI, Hamburg and RIIM/Simon Fraser University, Vancouver are noted with appreciation. The authors wish to thank Jagjit Chadha, Amanda Gosling, Matloob Piracha, Moshe Semyonov, anonymous referees, and participants at the XXII Annual Conference of the European Society for Population Economics for valuable comments.

References Adrangi, B., Rafﬁee, K. (1997), An econometric analysis of health care reform in the US. International Advances in Economics Research 3 (2), 181–192. Brown, R.P.C. (1997), Estimating remittance functions for Paciﬁc Island migrants. World Development 25 (4), 613–626. Browning, M., Thomas, I. (1998a), Prices for the FAMEX: methods and sources. Department of Economics Working Paper. McMaster University, Hamilton.

574

Don DeVoretz and Florin Vadean

Browning, M., Thomas, I. (1998b), Reconciled FAMEX codebook. Department of Economics Working Paper. McMaster University, Hamilton. Carroll, C., Rhee, B.K., Rhee, C. (1994), Are there cultural effects on saving? Some cross-sectional evidence. Quarterly Journal of Economics 35 (3), 685–699. Citizenship and Immigration Canada. (2002), Facts and ﬁgs. 2002: Statistical overview of the temporary resident and refugee claimant population, Online resources: Citizenship and Immigration Canada. Cox, D. (1987), Motives for private income transfers. Journal of Political Economy 95 (3), 508–546. Cox, D., Eser, Z., Jimenez, E. (1998), Motives for private transfers over the life cycle: an analytical framework and evidence from Peru. Journal of Development Economics 55 (1), 57–80. Deaton, A., Muelbauer, J. (1980), An almost ideal demand system. American Economic Review 70 (3), 312–324. Deaton, A., Muelbauer, J. (1993), Economics and Consumer Behavior. Cambridge University Press, Cambridge. Didukh, G. (2001), Health and personal care consumption patterns of foreign-born and Canadian-born consumers: 1984–1996, RIIM Working Paper No. 01-13, Research on Immigration and Integration in the Metropolis, Vancouver. Didukh, G. (2002), Immigrants and the demand for shelter, RIIM Working Paper No. 02-01, Research on Immigration and Integration in the Metropolis, Vancouver. ECORYS (Ed.). (2005), Study on Improving the Efﬁciency of Workers’ Remittances in Mediterranean Countries. ECORYS, Rotterdam. Elliott, S., Gray, A. (2000), Family structures: a report for the New Zealand immigration service, Department of Labour, Immigration Service, New Zealand. Geiger, B. (2002), Clothing demand for Canadian-born and Foreign-born households, RIIM Working Paper No. 02-04, Research on Immigration and Integration in the Metropolis, Vancouver. Green, W.H. (2002), Econometric Analysis, 5th ed. Prentice-Hall, NJ. Hansen, G. (1993), Quantitative Wirtschaftsforschung. Mu¨nchen, Vahlen. Ilahi, N., Jafarey, S. (1999), Guestworker migration, remittances and the extended family: evidence from Pakistan. Journal of Development Economics 58 (2), 485–512. Lucas, R., Stark, O. (1985), Motivations to remit: evidence from Botswana. Journal of Political Economy 93 (5), 901–918. Meenakshi, T., Ray, R. (1999), Regional differences in India’s food expenditure pattern: a complete demand systems approach. Journal of International Development 11 (1), 47–74. Pendakur, K. (2001), Consumption poverty in Canada 1969 to 1998. Canadian Public Policy 27 (2), 125–149.

Cultural Differences in the Remittance Behaviour of Households

575

Poirine, B. (1997), A theory of remittances as an implicit family loan arrangement. World Development 25 (4), 589–611. Rapoport, H., Docquier, F. (2006), The economics of migrants’ remittances. In: Kolm, S., Ythier, J.M. (Eds.), Handbook of the Economics of Giving, Altruism and Reciprocity (vol. 2). Elsevier. Ratha, D., Mohapatra, S., Silwal, A. (2009), Migration and remittance trends 2009: a better-than-expected outcome so far but signiﬁcant risks ahead, Migration and Development Brief 11. World Bank, Washington DC. Statistics Canada (2003), Ethnic diversity survey catalogue 89-593-XIE, Online resources: Statistics Canada. Teklu, T. (1996), Food demand studies in sub-Saharan Africa: a survey of empirical evidence. Food Policy 21 (6), 479–496. Wolff, F., Spilerman, S., Attias-Donfut, C. (2007), Transfers from migrants to their children: evidence that Altruism and cultural factors matter. Review of Income and Wealth 53 (4), 619–644.

PART V

Selection, Attitudes and Public Policy

CHAPTER 24

FSU Immigrants in Canada: A Case of Positive Triple Selection? Don DeVoretza and Michele Battistia a

Department of Economics, Simon Fraser University, Burnaby, BC, Canada V5A 1S6 E-mail address: [email protected]; [email protected]

Abstract This chapter investigates the economic performance of immigrants from the Former Soviet Union (FSU) countries in Canada. The contribution of this chapter lies in its use of the fall of the Soviet Union as a natural experiment to detect possible differential labour market performances of immigrants undergoing different screening systems and affected by different push and pull factors. In short, the collapse of the former Soviet Union allows an exogenous supply change in the number and type of FSU immigrants potentially destined to enter Canada. For this purpose, Census micro-level data from the 1986, 1991, 1996 and 2001 Canadian Population Census are utilised to estimate earnings and employment outcomes for immigrants arriving from the Soviet Union and from FSU countries before and after the collapse. Keywords: Immigration, integration, Canadian immigration policies JEL classiﬁcations: J61, F22

1. Introduction The post-1990 rise in immigration in general to Canada and from two disparate formerly closed systems – the Soviet Union and China – may have led to profound changes in the paradigm of economic integration into Canada’s labour force. Before the collapse of the Soviet Union, potential Soviet e´migre´s could not decide to move to Canada based on an open and easily transparent exit system. Thus, immigrants to Canada from the Former Soviet Union (hereafter FSU) were largely designated by Canada as refugees and therefore they were not screened in terms of their credential and readiness for integration into Canada’s labour market. Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008030

r 2010 by Emerald Group Publishing Limited. All rights reserved

580

Don DeVoretz and Michele Battisti

Fig. 1. Distribution of immigrants to Canada by entry gate: FSU immigrants. Source: LIDS (Landed Immigrants Data System) from IMDB Immigration Database. Figure 1 illustrates this point graphically. From 1980 through 1991 the distribution of immigrants from the USSR across entry gates was as follows: 58% refugees, 14.1% family class and 27.5% skilled class.1 By the year 2000, FSU refugees made up only 13.9% of the entrants with 63.9% of FSU immigrants now appearing in the skilled group. In short, before 1992 the majority of Soviet e´migre´s to Canada were only self-selected from the refugee portion of the potential pool of all USSR e´migre´s, whereas after 1992 most FSU immigrants entered under a double selection system. Figure 2 illustrates the uniqueness of the exogenous shock to FSU immigrant ﬂow ca. 1992–2001 when all immigrants to Canada had only a 5.3% drop in the proportion of refugees whereas the FSU immigrant share of refugees fell by more than 40%. Thus, after the fall of the Soviet Union (as with China ca. 1995), immigrants who left that region were drawn from a larger pool of potential movers with a different set of observable and unobservable human capital attributes. This important change should ultimately reveal itself in differential labour force outcomes of FSU immigrants in Canada after 1991 if our thesis of positive selection holds. After 1991, for most immigrants the initial positive self-selection was combined with a second level of selection as these FSU e´migre´s were subjected to a ‘points assessment’ system which favoured the admission of FSU immigrants with higher level of schooling and more skills that are of use in the Canadian labour market and society.2 Thus, looking at the labour market

1

The skilled class potential entrant is assessed under a ‘points system’ which yielded points for human capital attributes. 2 Pivnenko and DeVoretz (2003) note that a majority of Ukrainian immigrants to Canada came through non-economic entry gates before 1991.

FSU Immigrants in Canada

581

Fig. 2. Distribution of immigrants to Canada by entry gate: all other immigrants. Source: LIDS (Landed Immigrants Data System) from IMDB Immigration Database.

performance of FSU immigrants entering before and after 1991 has the potential of shedding light on the effectiveness of Canada’s selection process. The ﬁnal or third selection process arises when the immigrant decides to ascend to Canadian citizenship or to remain a non-citizen.3 It should be noted that traditionally only a portion of ‘points-assessed’ immigrants self-select into citizenship, yet almost all refugees naturalise given their inability to return home. Given that refugees often feel compelled to naturalise, some of the economic premium owing to citizenship is often lost due to adverse selection; those FSU e´migre´s who arrived after 1991, however, were largely not refugees and should reveal a positive selection into citizenship since they were not compelled to naturalise.4 In fact, we argue that only those post-1991 FSU e´migre´s who acquired additional Canadian-speciﬁc human capital will tend to naturalise and reap the labour market rewards from acquiring this human capital. Further empirical research on the role of citizenship acquisition for this speciﬁc group is left for future research. In sum, the following thesis is offered in terms of the labour market integration of e´migre´s from the USSR/FSU into the Canadian context: before 1991, most e´migre´s from the Soviet Union to Canada were singly

3

DeVoretz and Pivnenko (2006) document the positive effect of citizenship status on the labour market outcomes of Ukrainian immigrants. 4 DeVoretz and Pivnenko (2006) verify this empirically for all Canadian refugees ca. 2006 in Canada.

582

Don DeVoretz and Michele Battisti

selected by themselves and/or along unobservable characteristics that may not have been related to those that are beneﬁcial in the Canadian context; after 1991 the new cohort of FSU emigrants to Canada were often selected three times. This triple selection procedure in turn implies that a greater human capital stock will be embodied in this post-1991 cohort and would lead to more rapid integration into Canada’s labour market in the absence of discrimination or other forms of labour market failure. It is the purpose of this study to test this thesis in the context of a ‘gap analysis’ in terms of income and employment. The traditional immigrant earnings literature owing to Chiswick (1978) argues that upon entry, immigrants suffer an earnings deﬁcit due to the absence of speciﬁc and general (language and knowledge of institutions) human capital attributes. It was inferred by Chiswick from census data that over time – generally 8–12 years – immigrants overcame these human capital deﬁcits by investing in themselves and their earnings subsequently ‘caught-up’ to and then perhaps surpassed their Canadian-born colleagues. Figure 3 depicts the ‘gap’ hypothesis from both optimistic and pessimistic viewpoints. Given our thesis of ‘triple selection’ we would expect that Figure 1 would apply to highly skilled FSU e´migre´s to Canada since increased observable human capital attributes owing to triple selection should hasten the diminution in the earnings gap and may lead to its complete evaporation at X. Beyond X, in the optimistic case the immigrant now can overachieve with respect to their Canadianborn cohort’s earnings performance. However, if there exist ‘unobservable’

Fig. 3.

Idealised age-earnings proﬁle.

FSU Immigrants in Canada

583

factors that intervene in this process of labour market integration, the case of underachieving may arise. These ‘unobservables’ include inhibitions on the immigrant’s desire to self-select into the labour market, employer discrimination of the immigrant’s human capital characteristics (i.e. foreign education) and discounting their foreign labour market experience, as well as unobservables that may be beneﬁcial in the labour market and that also may have helped immigrants being accepted as refugees before the collapse of the Soviet Union. It is this ‘gap’ framework as depicted in Figure 1 which will inform our labour market integration analysis given the triple selection thesis outlined earlier.

2. Literature review Canadian literature on the economics of immigration provides an extensive empirical immigrant labour market integration (Reitz, 2001). The literature has largely focused on Canadian immigrant earnings’ performance in general but a series of in-depth studies based on the immigrants’ country of origin have recently appeared. Two major ﬁndings from the general Canadian immigrant earnings experience appear to date. First, an age earnings proﬁle analysis based on a human capital model forms the underlying analytical framework for immigrant labour market performance across entry groups and over time in Canada. Second, and this is more germane to this study, recent immigrant entry cohorts have failed to ‘catch-up’ to their Canadian-born cohorts while older vintages of immigrants have overachieved. Figure 4 presents two interesting empirically based variants of the gap model illustrated in Figure 3 in the Canadian context.5 First, British immigrants ca. 1996 could be termed ‘overachievers’ since they do not suffered an earnings penalty at entry and upon gaining citizenship outperformed their Canadian-born cohort every year over their life cycle. This overachieving phenomenon is repeated by immigrants from the United States and several other western European immigrant groups in Canada (Pivnenko and DeVoretz, 2003). However, the literature also argues that many groups of immigrants entering Canada after 1990 have not performed as described earlier. These observers report that each successive wave of post-1990 immigrants had a larger earnings entry penalty and rarely overcame this increased penalty with time in Canada (Li, 2003). In addition, work on discriminatory behaviour in the Canadian immigrant labour market argued that institutional barriers prevented credential recognition 5

Since these ﬁndings were derived from a series of pooled Canadian Censuses care in this interpretation must be made since ageing, cohort and time in Canada effects are difﬁcult to disentangle.

584

Don DeVoretz and Michele Battisti 35000

Wage earnings, $

30000 25000 20000 15000 10000 5000 0 25

35

45

55

65

Age CB

BritIm_NC

BritIm_C

ChinIm_C

ChinIm_NC

Fig. 4. Age-earning proﬁles. CB, Canadian Born; Britlm_C, British Immigrants Canadian citizens; Britlm_NC, British Immigrants non-citizens of Canada; Chimlm_C, Chinese Immigrants Canadian citizens and Chinlm_NC, Chinese Immigrants non-citizens of Canada (Chinlm_NC). Source: Census of Canada, 1996. (Ferrer and Riddell, 2002) and when coupled with overt discrimination (Pendakur and Pendakur, 1998) prevented the post-1990 wave of Canadian immigrants from successfully integrating into the Canadian labour market. In Figure 4 these ‘underachievers’ appear in the form of the most recent wave of highly educated Chinese immigrants in Canada. In fact, as reported in Figure 4, the average Chinese immigrant with or without citizenship status does not ‘catch-up’. We currently have two econometric studies which address the economic performances of select groups of Canadian immigrants from part of the FSU. Pivnenko and DeVoretz (2003) investigated the economic performance of recent Ukrainian immigrants to Canada and the United States with available census data. Their underlying approach was to test for ethnicity, foreign birth status and destination effects on the economic performance of a pooled set of pre- and post-1990 FSU Ukrainian immigrants in Canada and the United States. In particular, they tested for the existence of earnings overachieving in the context of Ukrainian immigrants in North America. An important sub-hypothesis is also addressed when they speculate that Ukrainian immigrants overachieve because they enjoy a ‘sheepskin effect’ which raises Ukrainian immigrant

FSU Immigrants in Canada

585

Fig. 5. Age-earnings proﬁle for Ukrainian Immigrants to Canada (UI), nonUkrainian Immigrants to Canadian Born (UCB) and non-Ukrainian Canadianborn (NUCB). Source: 1996 Census of Canada authors’ calculations. earnings relative to other immigrant graduates because employers may value Ukrainian degrees more. Finally, Pivnenko and DeVoretz conducted a comparative analysis of Ukrainian immigrant earnings in the United States versus Canada to detect if Canada’s highly selective immigration policy encouraged more productive immigrants to enter Canada from the Ukraine. Their reported results indicate that recent Ukrainian immigrants to Canada are indeed a select group. For the period 1991–2001, Ukrainian immigrants to Canada arrived with higher educational attainment, a greater propensity to speak English at home and contained the largest percentage of professionals for any immigrant cohort over the 1991–2001 period. These human capital attributes led to above-average earnings performance for Ukrainian immigrants which in turn was explained by their occupational distribution (largely professionals), numbers of weeks worked, and a substantial ‘sheepskin effect’.6 The result of this robust earnings function is that Ukrainian immigrants in Canada outperformed the earnings of all other Canadian immigrants and ‘caught-up’ and surpassed their Canadian-born cohort at the age of 36 as depicted in Figure 5. Finally, Pivnenko and DeVoretz (2003, p. 13) conclude from their study that: For Ukrainian immigrants, the assimilation process starts at a higher income level that exceeds the income earned by non-Ukrainians with 6

In fact, Pivnenko and DeVoretz (2003) report that this earnings effect derived from completing a university degree was the greatest for Ukrainian immigrants relative to all other Canadian immigrants.

586

Don DeVoretz and Michele Battisti

the difference growing over time. The greater intercept reﬂects the more favorable entry effect for the Ukrainian immigrants. This positive earnings premium implies that y the quality of the earnings enhancing characteristics Ukrainians have acquired y is relatively higher than for the rest of the immigrant population. In other words, Ukrainian immigrants in general were found to be overachievers. They further report that Ukrainian immigrants to the United States do even better because they were endowed with greater human capital than Ukrainian immigrants resident in Canada. However, we must be cautious not to draw hasty conclusions from this Ukrainian study for the FSU immigrant experience in general. First, the Pivnenko–DeVoretz sample is restricted to Ukrainians only and includes both pre- and post-FSU populations of all skill types. It is possible that the subject of this study, namely post-FSU arrivals, will exhibit a pattern of underachieving that appears in Figure 3. Dean and DeVoretz (2000) conduct a similar analysis exploring the ‘gap’ thesis for all Jews living in Canada ca. 1996. Again, this population does not match the former FSU immigrant stock which is the focus of this study but does include many former FSU immigrant arrivals. Dean and DeVoretz (2000) ask whether ethnicity (i.e. Jewish or non-Jewish in this case) is related to the economic performance of immigrants in Canada. Their underlying argument is that income-enhancing non-cultural characteristics (e.g. education) are correlated with cultural characteristics. Using Canadian census data, their study group overlaps with the immigrant sample considered in this study, namely those immigrants from the FSU.7 They isolated two Jewish sub-groups: Jewish Canadianborn and Jewish immigrants and estimated earnings functions for these two groups as well as their non-Jewish counterparts. They reported that the stock of human capital characteristics which were normally correlated with higher income (age, education and English language skills) exceeded all other immigrant groups to such an extent that any earnings entry penalty owing to immigrant status was overcome by Jewish immigrants by virtue of other income correlates. For example, almost 100% of Jewish immigrants reported speaking English at home while only 69% of all immigrants reported a similar capability. In addition, Jewish immigrants are older and more likely to be married than non-Jewish immigrants. However, one glaring inconsistency occurs when they observe that Jewish immigrants have less education than their Canadian-born Jewish counterparts. Nonetheless, Jewish immigrants are highly concentrated in the professions in Canada. In terms of gender, it is reported that ca. 1990 Jewish immigrant women opted out of the Canadian labour market 7

In fact, in the 1991 Canadian 2% PUST a cross tabulation of the Jewish sample indicates that over 60% of recent Jewish immigrants originated from the FSU.

FSU Immigrants in Canada

587

and, when in the labour force, were more likely to work for wages and salaries. Their regression analysis of the earnings model allowed Dean and DeVoretz (2000) to conclude that both the substantial human capital endowments of Jews born in Canada and the differential rewards paid to these educational endowments allowed Jewish–Canadian immigrants resident in Canada ca. 1990 to outperform other immigrants and avoid an earnings penalty upon entry into Canada. Therefore, two econometric studies that partially cover our immigrant group of interest (FSU) indicate that before 1995 Jewish and Ukrainian immigrants to Canada were exceptional groups. They earned more than their other foreign-born cohorts in Canada due to either greater human capital endowments or a better recognition of their credentials, or both.

3. Data 3.1. Data source The data we use in this chapter are drawn from the individual Public Use Micro Files (PUMF) of the Canadian Census of Population for the years 1986, 1991, 1996 and 2001. These datasets contain information on a representative sample of people living in Canada in the years 1985, 1990, 1995 and 2000 respectively. The total sample sizes of the PUMFs are 500,434 for the 1986 Census, 809,654 for the 1991 Census, 792,448 for the 1996 Census and 801,055 for the 2001 Census. The choice to work with census data has a number of inherent disadvantages. Census data do not have a panel structure that would allow us to follow the same individuals over time, and thus in our econometric estimation we cannot control for unobservables that happen to be correlated with the variable that identiﬁes FSU as the origin region. In addition, stacking four different Censuses together introduces a number of possible sources of bias which makes the data cleaning process for this paper particularly complex. In spite of these drawbacks, the individual-level census data seem to be the best choice to analyse our research question. The most important reason for this conclusion is that it is the only dataset that leaves us with a sufﬁciently large number of observations for FSU immigrants that entered Canada before and after 1991.8

8

In particular, the publicly available version of the IMDB dataset does not offer a sufﬁciently large sample size (because FSU immigrants are a relatively small population) and, even more crucially, does not include immigrants who arrived in Canada before 1981.

588

Don DeVoretz and Michele Battisti

3.2. Construction of our dataset A few variables of interest for our study are coded differently across the different censuses. In other words, the coding system adopted for the construction of most of the categorical variables vary across different censuses, so that it is not possible to simply stack the data. The information contained in our dataset is ultimately equivalent since we will only include variables for which this is the case. In the very few cases in which the recoding procedure does have an impact on the informational content. Variables identifying birthplace and those identifying year of immigration do not have a perfectly overlapping coding across census years, however we did recode to avoid any bias.9 PUMF ﬁles for each census report variables are expressed in Canadian Dollars (CAD) as reported by the respondents for one year before the release of the relevant census, therefore all monetary variables employed in our estimates (wages and salaries, self-employment income, total income and government transfers) need to be adjusted for inﬂation. For the reported statistics in our summary tables and our regressions, all monetised variables are expressed in dollars for the year 2000 leaving the monetised values for 2001 unchanged.10 3.3. Data selection For the summary statistics and for our wage equations, we restricted our sample to individuals of working age (i.e. aged 20–65) and excluded individuals for whom the primary source of reported income is selfemployment income. It is apparent from even a cursory reading of Table 1 that substantial differences emerge between the Canadian and FSU or USSR-born populations in Canada for these selected socio-economic indicators. First, FSU immigrants are much older (47) than the Canadian-born stock (39). Next, the USSR/FSU-born immigrant group is more likely to be married (57.3%) and more highly educated (42.6%) than their Canadian-born cohort. These strong human capital characteristics, however, did not translate into greater incomes or wages and salaries for the USSR/FSU group. In fact, USSR/FSU personal income was $1,096 less than their Canadian-born cohort. Nonetheless, the consumption of total government transfers is nearly equal between the two groups. In sum, all the standard variables contained in a typical human capital model of earnings are strong for the USSR/FSU-born group in Canada across the survey period. 9

Codes for the construction of the merged dataset are available from the authors. For this task, we use a conversion utility offered by Statistics Canada (2006). According to this conversion tables and in comparison with 2000 CAD, values from the 1996 Census ought to be multiplied by 1.10, values from the 1991 Census by 1.21, values from the 1986 Census by 1.51. 10

589

FSU Immigrants in Canada

Table 1.

Relevant sample statistics for the period 1986–2001

Variables

Canadian-born

USSR/FSU-born

English speaking (%) English and French speaking (%) Percentage married Age Number of years in Canada Percentage male Bachelor degree or higher (%) Value dwelling Total personal income Wages and salaries Self-employment income Child beneﬁts received Total government transfers Sample size Percentage of living in GTA Percentage from Census 1986 Percentage from Census 1991 Percentage from Census 1996 Percentage from Census 2001

63 21 44.9 39.4 Not Applicable 49.4 20.4 $133,524a $29,245 $23,386 $1,697 $267 $2,227 1,068,272 10.5 17.6 28.0 27.0 27.5

85 9.2 57.3 47.3 23.0 48.1 42.6 $179,292 $27,149 $19,552 $2,109 $257 $2,332 3,278 49 20.6 21.9 24.1 33.4

Source: Authors’ calculations based on census data. a All dollar values in constant in year 2000.

In spite of this, however, the labour market outcomes in terms of earnings contradict these strong human capital characteristics, suggesting that the returns to those strong characteristics are lower for the USSR/FSU group than for their Canadian counterparts. We next turn to a regression analysis to explain this anomaly. 4. Regressions results 4.1. OLS results Following the naı¨ ve Mincer earnings equation framework we report the results derived from a simple ordinary least squares (OLS) wage regression in log form for people aged 20–65 in Table 2. We isolate two areas of birth in this equation: immigrants from the USSR/FSU, and we use the Canadian-born respondents as our control group.11 It should be noted that this formulation does not take into account the fact that the labour market participation decision is endogenous to other labour market conditions. It is nevertheless an interesting ﬁrst step to evaluate the overall discrepancy between the 11

We exclude non-wage earners and earners who derived their incomes from selfemployment.

590

Don DeVoretz and Michele Battisti

Table 2. Dependent variable: log(wage) in 2000 CAD

(1)

Born in USSR or FSU countries Age Age2 Male Married and not separated Dummy for children in the household Dummy for Toronto CMA Census year ﬁxed effects Educational attainment dummies Observations R2

0.03

OLS earnings equations (2)

(3)

(4)

(0.02) 0.09*** (0.02) 0.23*** (0.02) 0.45*** (0.02) 0.16*** (0.00) 0.15*** 0.001*** (0.000) 0.002*** 0.56*** 0.18*** (0.00) 0.19***

(0.00) 0.17*** (0.000) 0.002*** (0.00) 0.51*** (0.00) 0.22***

(0.00) (0.000) (0.00) (0.00)

0.12*** (0.00) 0.19*** No

No

Yes

Yes

No

No

Yes

Yes

1,047,041 0.000

1,047,041 0.114

1,047,041 0.217

385,104 0.227

(0.00)

Note: Standard errors in parentheses. * po0.05, **po0.01 and ***po0.001.

wages of Canadian-born and USSR/FSU-born workers, controlling for a few covariates including gender, age, education and marital status. Our preferred speciﬁcation is presented in column 4 (which includes dummies for each census year) to control for the fact that the proportion of USSR/FSU immigrants varies across census periods. Table 2 reports several interesting results. First, the estimated wage differential is 45% in favour of the Canadian-born control group and is signiﬁcant at the 1% signiﬁcance level. Moreover, the basic human capital model we hypothesised to explain earnings are conﬁrmed by these results. Age and age2 obtain positive and negative signs respectively, conﬁrming the argument that earnings increase at a decreasing rate over their lifetimes. Moreover, the effects derived from an earner being male (0.506), and married (0.220), or living in Toronto (0.185) are all positive and these variables signiﬁcantly raised earnings. However, if the wage earner was born in the USSR/FSU, then as earlier noted their earnings signiﬁcantly declined (0.450) relative to the control group of Canadian-born earners. Table 3 reports equivalent earning equations using other immigrants as a control group. Using immigrants from other countries as controls allows us to include a dummy for having entered Canada after 1991, as well as an interaction effect for allowing the effect of having entered after 1991 to be different for FSU immigrants and for other immigrants, which is consistent with a change in selection mechanisms for FSU immigrants.

Note: Standard errors in parentheses. * po0.05. ** po0.01. *** po0.001.

No 248,974 0.021

0.11 0.48*** 0.06

Born in USSR or FSU countries Arrived in Canada after 1991 FSU*post 1991 arrival Age Age2 Married and not separated Male Dummy for children in the household Dummy for Toronto CMA Educational attainment dummies Observations R2

**

(1) (0.03) (0.001) (0.04)

No 248,974 0.095

0.04 0.38*** 0.02 0.13*** 0.001*** 0.14***

(2) (0.03) (0.01) (0.04) (0.00) (0.0000) (0.01)

**

Yes 248,974 0.181

0.09 0.42*** 0.07 0.13*** 0.001*** 0.13*** 0.50***

(3) (0.03) (0.01) (0.04) (0.00) (0.000) (0.01) (0.003)

USSR/FSU earnings equations with all other immigrants as control group

Dependent variable: log(wage) in 2000 CAD

Table 3.

0.10** 0.47*** 0.05 0.13*** 0.001*** 0.15*** 0.47*** 0.10*** 0.16*** Yes 147,952 0.173

(4) (0.03) (0.01) (0.05) (0.00) (0.000) (0.01) (0.001) (0.01) (0.01)

FSU Immigrants in Canada 591

592

Don DeVoretz and Michele Battisti

Table 4.

Earnings equation, USSR/FSU-born only

Dependent variable: (1) log(wage) in 2000 CAD Entering after 1991 Age Age2 Male Married and not separated Dummy for children in the household Dummy for Toronto CMA Census year Fixed effects Years since immigration dummies Educational attainments dummies Observations R2

0.50***

(2)

(0.04)

0.22**

(3)

(0.07)

0.02 0.15*** 0.002*** 0.52*** 0.09

(4)

(0.08) 0.06 (0.01) 0.12*** (0.000) 0.001*** (0.04) 0.49*** (0.05) 0.11 0.01

(0.10) (0.02) (0.000) (0.05) (0.06) (0.06)

0.22*** (0.05) No

No

Yes

Yes

No

Yes

Yes

Yes

No

No

No

Yes

3,180 0.041

3,180 0.100

3,180 0.209

1,999 0.227

Note: Standard errors in parentheses. * po0.05, **po0.01 and ***po0.001.

The coefﬁcient on the interaction term ‘FSU immigrant * post 1991 arrival’ in columns 3 and 4 of the table show that, conditional on all our controls, FSU immigrants arriving after 1991 do marginally better than those arriving earlier. The effect is however not statistically signiﬁcant. There seems to be a positive effect for FSU immigrants. Immigrants from the FSU coming after 1991 experience less of an earnings disadvantage when compared to all other immigrants. We can run equivalent regressions comparing the earnings equations for USSR/FSU-born immigrants who arrived in Canada before 1991 and after 1991. This division is motivated by the supply side change in the composition of FSU immigrants to Canada. These estimations are presented in Table 4, which further investigates some of the ﬁndings brieﬂy discussed in Table 3. The ﬁrst column suggests that post-1991 immigrants have lower unconditional average earnings than those who came before that date. This should not be surprising given that at any point in time immigrants that arrived in Canada earlier have more experience in the Canadian labour market. In column 2 we control for years in Canada by including a series of dummies,12 and as a result the estimated sign obtained for our main parameter of interest changes and it 12

Unfortunately, years since immigration are a categorical variable in the Canadian Censuses of the years we use.

593

FSU Immigrants in Canada

Table 5.

Labour force activity aged 20–65 by gender and birth place

Labour force status

All Canadians

All FSU/ Soviet Union

Pre-1992 FSU

Post-1992 FSU

Males and Females Percentage employed Percentage unemployed Percentage not in labour force

61 7 24

61 6.5 33.5

61.5 9.2 28.3

60.7 5 35

Males Percentage employed Percentage unemployed Percentage not in labour force

79.5 7 24

59 10 31

72.5 9.4 18.1

70.9 5.6 23.5

Females Percentage employed Percentage unemployed Percentage not in labour force

65 7 28

51 6 43

52.0 10.5 37.6

51.3 4.5 44.2

Source: Authors’ calculations based on census data.

is now positive and signiﬁcant: after the same number of years in Canada FSU immigrants arriving after 1991 earn more than their predecessors. Including census ﬁxed effects, age and educational attainment (column 4), leaves the sign of the estimated coefﬁcient on the dummies for arrival after 1991 unchanged, but renders the estimate insigniﬁcant. It is arguable that the estimates of column 2 are actually more informative, since shutting down all variation obtained from comparing observations across different censuses13 effectively eliminates most of the overall variation because recent FSU immigrant are necessarily observed in later censuses only. We do ﬁnd it comforting, however, that the point estimates are not qualitatively different from those in column 2. The sign obtained by the human capital arguments of interest (age, age2, marital status, years in Canada and gender) all conformed to our theoretical predictions. Again, FSU immigrants living in the Toronto Metropolitan area have conditionally higher earnings.

4.2. Labour force activity The reported regression results in Tables 3 and 4 only concerns workers. Presumably, working is a result of a decision made by people that may depend on unobservable factors that may also affect wages. Table 5 reports the labour force activity of all Canadians and those who arrived before and after 1992 from the FSU which portrays their respective commitments to the Canadian labour force. For males and females 13

Including census ﬁxed effects does exactly that.

594

Don DeVoretz and Michele Battisti

combined, the greatest difference in labour force status appears in the percentage not in the labour force with those from the FSU/Soviet Union reporting an average 33.5% not in the labour force. This percentage absent from the labour market grows to 43% for females from FSU/Soviet Union. In short, whether we cross tabulate the percentage not in the labour force by gender, birth place or year of entry, FSU immigrants to Canada are signiﬁcantly less likely to report themselves not in the labour force than the rest of the Canadian population.

5. Two-stage models 5.1. FSU vs. Canadian born As we brieﬂy discussed earlier, our previously estimated models do not incorporate the decision to enter the labour force which now appears to be an important omission given the reported labour force activity differentials as reported in Table 5. In the following regressions, we estimate a probit model to detect the probability of participating in the Canadian labour force. Our binary dependent variable (1 ¼ employed and 0 ¼ otherwise) is constructed from the variable reporting full-time and part-time participation. Therefore, at this stage we do not distinguish between part-time and full-time employment, and investigate the choice of entering the labour market only for males and females separately. The main parameter of interest corresponds to the effect of being born in the USSR/FSU on the probability of working. We include ﬁxed effects for the census reporting year, control for age (and age2), educational attainment and presence of children in the households. We run separate regressions for males and females.14 Table 6 reports the results of our selection and earnings equations for the pooled sample of all Canadians and USSR/FSU immigrants residing in Canada for the period 1986–2001. This is our base speciﬁcation, which argues that participation in the labour force is endogenous and depends on prospective earnings. Given that males and females have very different labour market participation rates, we run these regressions separately. In the ﬁrst stage (or in the selection equation) in addition to the basic human capital variables (age, age2, gender and education levels) we further isolate residence in Toronto and the individual’s foreign birth status (USSR/FSU). In our preferred results (reported in columns 1 and 2) the variables that are excluded from the wage equation are marital status, the presence of children, and the presence of employment beneﬁts in the individual’s reported 14

Owing to large differences in male–female labour market participation and labour supply elasticities, it is not sensible to impose a restriction that labour force participation is the same for men and women.

0.116*** 0.00173*** 0.632*** 0.335*** 0.124*** 0.184*** 0.526*** 0.00471 0.0862*** 0.699*** 0.710*** 2.026*** 152,574

Selection (ﬁrst stage) Age Age2 Grade 9 or less High school dropouts Bachelor Graduate studies certiﬁcate Married and not separated Children in the household Toronto CMA Born in USSR or FSU countries Constant Mills Lambda Observations (0.00261) (0.0000307) (0.0181) (0.0116) (0.0125) (0.0204) (0.0136) (0.0121) (0.0120) (0.0448) (0.0481) (0.0888)

(0.00604) (0.0000802) (0.0396) (0.0186) (0.0139) (0.0227) (0.0141) (0.0715) (0.119) 0.108*** 0.00167*** 0.887*** 0.418*** 0.264*** 0.334*** 0.0571*** 0.611*** 0.132*** 0.410*** 0.456*** 0.722*** 335,899

0.129*** 0.00131*** 0.187*** 0.108*** 0.197*** 0.347*** 0.160*** 0.190*** 7.165*** (0.00153) (0.0000185) (0.0106) (0.00651) (0.00744) (0.0125) (0.00604) (0.00610) (0.00677) (0.0290) (0.0286) (0.0184)

(0.00135) (0.0000179) (0.0156) (0.00719) (0.00564) (0.00880) (0.00535) (0.0291) (0.0267)

(4) Females

(0.00111) (0.0000129) (0.00644) (0.00512) (0.00688) (0.0106) (0.00464)

(0.0298) (0.0213) (0.0275)

0.299*** 0.144*** 1.698*** 650,741

(0.0380) (0.0359)

0.321*** 0.215*** 0.684*** 696,069

0.0752*** 0.00116*** 0.920*** 0.452*** 0.326*** 0.412*** 0.0878***

0.298*** 6.138***

(0.00184) 0.166*** (0.0000245) 0.00193*** (0.0117) 1.038*** (0.00645) 0.465*** (0.00650) 0.418*** (0.00959) 0.641***

0.0867*** 0.00138*** 0.629*** 0.308*** 0.145*** 0.257*** 0.504***

0.0440 7.881***

0.107*** 0.000870*** 0.0364** 0.0102 0.131*** 0.260***

(3) Males

Notes: Standard errors in parentheses. Marital status and presence of children (or the former only) used as instruments in the participation equation. * po0.05, **po0.01 and ***po0.001.

0.110*** 0.000847*** 0.181*** 0.0221 0.151*** 0.277*** 0.117*** 0.0924 7.767***

Age Age2 Grade 9 or less High school dropouts Bachelor Graduate studies certiﬁcate Toronto CMA Born in USSR or FSU countries Constant

(2) Females

Heckman’s procedure-endogenous participation for FSU immigrants and Canadian-born

Dependent variable: log(wages) in 2000 CAD (1) Males

Table 6.

(0.0260) (0.0185) (0.0479)

(0.000965) (0.0000114) (0.00624) (0.00422) (0.00534) (0.00946) (0.00378)

(0.0289) (0.0475)

(0.00208) (0.0000313) (0.0279) (0.0119) (0.00725) (0.0101)

FSU Immigrants in Canada 595

596

Don DeVoretz and Michele Battisti

earnings.15 All the variables obtain the expected sign and are signiﬁcant at the 5% level or better for both the male and female speciﬁcations. Lambda, or the coefﬁcient of the inverse of the Mill’s ratio, is negative as expected with the exception of the female speciﬁcation in column 4.16 In the case of males (columns 1 and 3), the FSU origin dummy has a negative effect on the probability of participating in the labour market but, for people who work, it does not impact wages signiﬁcantly (there is a negative effect, but it is not signiﬁcant). Turning to the results reported in column 3 (which are more reliable given the many more observations) male FSU immigrants are 30% less likely to participate in the labour force, controlling for age, education and marital status. Female FSU immigrants, however, exhibit both a lower probability of being in the labour market (32% using column 4) and lower predicted wages than their Canadian-born counterparts (30% wage earnings) after controlling for the other covariates and for the endogeneity of labour market participation. In sum, FSU foreign birth status impacts female and male labour market outcomes relative to the entire Canadian population except for FSU male earnings. 5.2. USSR/FSU immigrants versus all immigrants Our preferred speciﬁcations in Table 7 are found under columns 3 and 4 since the under-reported or missing values for the presence of children does not appear in these speciﬁcations. The selection equation which is embedded in the lower half of Table 7 reports the usual feature that labour force participation is increasing in age at a decreasing rate for all foreign-born residents in Canada. In addition, married foreign-born males are more likely to participate in the labour force (holding age and education constant) than unmarried males and married foreign-born females are less likely to work if married. More striking is the fact that FSU-born males have an average expected participation rate that is 15.8 points lower than that of other immigrants. The equivalent ﬁgure for females is 20 points lower. Finally, the coefﬁcient obtained on the IMR (Inverse Mills Ratio) is negative for males and females, although for females it is not signiﬁcant. Thus, women who work are positively self-selected, but not as strongly as males are. In the wage equation (upper half of Table 7), all variables obtain the expected sign. Age and education variables both have a much greater impact on wages for females when compared to males. The age2 variable 15

The inclusion of Toronto CMA and presence of children while our preferred equation greatly reduces the number of observations. 16 Under this speciﬁcation without the presence of children or location in Toronto a positive lambda infers negative selection of females into the labour market.

(0.0159) (0.000201) (0.0514) (0.0320) (0.0217) (0.0299) (0.0846) (0.354) (0.00315) (0.0000359) (0.0160) (0.0141) (0.0130) (0.0184) (0.0169) (0.0124) (0.0425) (0.0598) (0.234)

0.0109 0.000295 0.0737 0.0132 0.179*** 0.298*** 0.0883 9.845***

0.132*** 0.00173*** 0.392*** 0.208*** 0.0309* 0.0445* 0.263*** 0.0651*** 0.304*** 1.418*** 2.429*** 95,179

0.147*** 0.00196*** 0.499*** 0.275*** 0.119*** 0.186*** 0.0692*** 0.394*** 0.189*** 1.683*** 0.561*** 186,053

0.0921 0.000860*** 0.172*** 0.107*** 0.203*** 0.361*** 0.0729* 7.691*** (0.00197) (0.0000233) (0.00945) (0.00878) (0.00903) (0.0139) (0.00810) (0.00766) (0.0268) (0.0378) (0.0334)

(0.00202) (0.0000230) (0.0110) (0.0100) (0.00975) (0.0129) (0.00862) (0.0292) (0.0395) (0.100)

0.158*** 1.360*** 2.115*** 187,707

(0.00715) (0.0000885) (0.0224) (0.0168) (0.0141) (0.0177) (0.0477) (0.159)

0.125*** 0.00164*** 0.329*** 0.174*** 0.0460*** 0.0711*** 0.315***

(0.00274) 0.0297 (0.0000350) 0.0000200 (0.0146) 0.00280 (0.0106) 0.00965 (0.00843) 0.171*** (0.0125) 0.284*** (0.0294) 0.0136 (0.0619) 9.504***

***

(3) Males

Notes: (1) Reference group: high school graduate, not from the FSU countries, single, no children. (2) Standard errors in parentheses. (3) Marital status and presence of children (or the former only) used as instruments in the participation equation. (4) po0.05, **po0.01 and ***po0.001.

Age Age2 Grade 9 or less High school dropouts Bachelor Graduate studies certiﬁcate Born in USSR or FSU countries Constant Selection (ﬁrst stage) Age Age2 Grade 9 or less High school dropouts Bachelor Graduate studies certiﬁcate Married and not separated Children in the household Born in USSR or FSU countries Constant Mills Lambda Observations

***

(2) Females

0.200*** 1.477*** 0.0469 203,146

0.124*** 0.00161*** 0.512*** 0.285*** 0.129*** 0.184*** 0.146***

0.123*** 0.00128*** 0.330*** 0.188*** 0.235*** 0.400*** 0.117*** 6.917***

(4) Females

(0.0256) (0.0351) (0.0782)

(0.00178) (0.0000206) (0.00910) (0.00843) (0.00851) (0.0131) (0.00694)

(0.00520) (0.0000685) (0.0258) (0.0150) (0.00884) (0.0129) (0.0287) (0.125)

Heckman’s procedure-endogenous participation for FSU immigrants and all immigrants: 1986–2001

Dependent variable: log(wages) in 2000 CAD (1) Males

Table 7.

FSU Immigrants in Canada 597

598

Don DeVoretz and Michele Battisti

has no signiﬁcant effect for males. For males, having completed high school also has no signiﬁcant wage returns.17 Once we control for the fact that participation in the labour force is endogenous (and it is so especially for males), FSU-born males who work have basically the same average wage as all other immigrants. In the case of females, there is a gap against the FSU-born. Controlling for age and education and for the endogeneity of labour market participation, FSU-born females yields 11.7% lower average wage when compared to all other female immigrants. An additional selection device beyond immigrants self-selecting into (or out of) the labour market is their citizenship status. As we argued earlier, naturalising to Canadian citizenship can affect labour force dimensions in terms of job selection and provide a signal to potential employers about a naturalised citizen’s integration. Thus, we add a citizenship variable to our reported preferred speciﬁcations (reported in Table A1). In the foreign-born population, a citizenship premium appears and, more importantly, when we interact citizenship status with FSU origin the interaction term is strongly positive yielding our predicted ‘citizenship’ premium.

6. Simulations Our earnings gap thesis outlined in Figure 3 can be reproduced empirically from our reported results and we do so for two control groups and the FSU-born group in Canada. Figure 6 compares Canadians to the FSU group with older and pre-1992 immigrants from the FSU now doing very poorly when we control for years in Canada. Figure 7 below compares the earnings proﬁles of FSU immigrants relative to all other immigrants in Canada. The results are revealing since regardless of age FSU immigrants earn less than all other immigrant population.

6.1. Decomposition analysis The simulated wage gap between the two immigrant groups may be owing to differences in individual characteristics and differences in returns to these characteristics. The Oaxaca–Blinder decomposition method (Neuman and Oaxaca, 2003) has become a routine method in labour market discrimination studies to explain a segmented group’s wage difference vis-a-vis a control group. In our case, we argue that the foreign birth status of an immigrant segments the labour market, and thus we 17

Coefﬁcients on ‘grade 9 or less’ and on ‘High School dropouts’ are negative but small and insigniﬁcant.

FSU Immigrants in Canada

599

Fig. 6.

Age-earning proﬁles for FSU- and Canadian-born, based on the second stage of our Heckman model.

Fig. 7.

Age-earning proﬁles for FSU-born and other immigrants to Canada, based on the second stage of a Heckman model.

employ a decomposition analysis. In its simplest version, the idea is to isolate the fraction of the wage differential unexplained by human capital endowments which is usually attributed to labour market discrimination. Accordingly, we have to adopt one of the estimated wage structures as the non-discriminatory norm for the group believed to be dominant in

600

Don DeVoretz and Michele Battisti

the labour market (Canadian-born) relative to the comparison group (FSU immigrants). The human capital portion of the overall wage differential is obtained as a sum of the differences in the mean characteristics of the two groups weighted by the estimated coefﬁcients for the non-discriminatory wage standard. The portion of the overall wage differential owing to discrimination will then be the residual left over after netting out for the human capital portion.18 In our study we adapt this decomposition methodology to explain Canadian/FSU immigrant wage differentials. Further, we treat FSU immigrants as a disadvantaged group since as (initially) non-citizens they are discriminated against in the public sector which blocks job access of non-citizens. Moreover, in the private sector we argue that non-citizenship status serves as a signal: ﬁrst, it indicates a weak attachment to Canada, and, secondly, it may indicate need for greater cultural integration. Neuman and Oaxaca (2003) acknowledged that when you introduce a correction for selectivity bias this in turn introduces some fundamental ambiguities in the context of wage decompositions. Thus we follow one of their suggested decomposition modiﬁcations below. Based on our earlier estimates, we deﬁne the wage differential between FSU and non-FSU (reference group [REF]) as ¼ lnðW REF Þ lnðW FSU Þ ¼ X REF bREF X FSU bFSU D lnðWÞ ¼ ðX REF X FSU ÞbFSU þ X FSU ðbREF bFSU Þ þ ðX REF X FSU ÞðbREF bFSU Þ In the second line of equation the overall wage effect is decomposed into the effect due to different endowments, the effect due to different returns to these characteristics, and the effect due to the interaction of the difference in endowments and the difference in their returns. Table 8 reports several decompositions which use alternative reference groups when analysing the wage gap with respect to the FSU population. In the ﬁrst case (Row 1), the wage gap in total or for males between Canadians and all FSU immigrants regardless of entry date is in favour of the FSU population with 1% premium for males and a trivial premium accruing to FSU females. However, this small wage gap is a product of two offsetting forces: the FSU population in general has a greater human capital endowment (by 19%) than Canadians, but they receive a lower rate of return (14%) on these assets. The second decomposition experiment is conducted now for FSU immigrants and all other immigrants. Again, only a minor wage gap 18

This could also be directly calculated as a sum of the difference in estimated coefﬁcients between the two groups weighted by the mean characteristics of the discriminated group.

601

FSU Immigrants in Canada

Table 8.

Sources of wage differentials

Decomposition Total difference

Canadian born Total vs. FSU Males immigrants Females Total All other Males immigrants Females vs. FSU immigrants FSU before vs. Total Males FSU after Females 1991

Difference explained by endowments

Interaction Number of Difference observations unexplained by endowments (returns to endowments)

0.01307 0.01630 0.00148 0.01294 0.01534 0.00474

0.1939 0.2177 0.1582 0.0550 0.0664 0.0389

0.1431 0.1527 0.1237 0.0388 0.0329 0.0447

0.4948 0.4744 0.5180

0.5606 0.7758 0.2893

0.0866 0.05812 0.1353

0.0377 0.0486 0.0330 0.0032 0.0182 0.0104

961,509 522,165 439,344 262,222 143,113 119,109

1.1421 1.3083 0.9426

3,180 1,711 1,469

appears between these two groups whether in total (1%) or males (1%) or females (0.4%). This slight wage gain in favour of the FSU population is explained as in the earlier case. The FSU population in general has a slightly greater human capital endowment (5%) than all Canadian immigrants but they receive a lower rate of return (3.8%) on these assets which in turn leads to the small positive wage gap in favour of the FSU population. A substantial wage gap arises when the control group is deﬁned as the pre-1991 FSU immigrants and the excluded group is post-1991 FSU immigrants. Now a nearly 50% wage gap emerges in favour of pre-1991 FSU immigrants to Canada. FSU immigrants entering Canada after 1991 have a much ‘better’ endowment in human capital (higher level of education, younger) and receive a slightly greater return to those endowments. However, given that the interaction term is large and positive (1.1), the effects derived from a greater human capital endowment and a higher reward to this human capital for the post-1991 FSU cohort are outweighed.19 7. Conclusions This chapter argued that the post-1991 changes in the exit rules for potential immigrants from the FSU amounted to a natural experiment with predicable supply side changes. The evidence presented supports this interpretation. First, before 1991 FSU immigrants to Canada were largely (76%) unscreened by Canada’s points system. By 2000, 65% of FSU immigrants to Canada were screened. This shift to entry under the 19

Note the two terms in the interaction term are negative and thus when cross multiplied becomes a positive term.

602

Don DeVoretz and Michele Battisti

screened portal, namely as economic immigrants, led to enhanced human capital attributes for post-1991 FSU immigrants to Canada but sample statistics provided by census data did not support the anticipated strong labour market outcomes. These observed weak outcomes are in sharp contrast to the two reported case studies for Jews and Ukrainians in the Canadian contexts which characterised these members of the FSU e´migre´ groups as groups of ‘overachievers’. With the aid of a human capital model, we estimate a naı¨ ve (OLS) Mincer earnings equation which identiﬁes a 45% negative wage gap relative to the Canadian-born population arising from FSU foreign birth status. A crucial modelling correction is made once it is noted that, for post-1991 FSU immigrants, more than 35% of these immigrants ca. 1991–2000 were not in the labour force. We now estimate a two-stage model to correct for the fact that labour force participation and earnings are endogenous, foreign birth status (FSU-born) impacts labour force participation but not earnings for FSU males. However, female FSU immigrants still experience a 30% wage gap vis-a-vis Canadian females. These gap results are identical when we use all foreign-born as the reference group. In sum, we conclude that FSU immigrants are earnings ‘underachievers’ relative to either all Canadians or just all immigrants in Canada. When we decompose the sources of the simulated wage gap, we ﬁnd that although the FSU population has in general a greater human capital endowment, they receive a lower rate of return than either the Canadian- or all foreign-born reference groups, which suggest discrimination against FSU immigrants in Canada’s labour market. In sum, the ‘natural experiment’ of free movement after 1991 did lead to a shift in entry gates to economically assessed immigrants who possessed greater human capital attributes than resident Canadians. However, this ‘double selection’ process did not lead to economic ‘overachieving’ since FSU immigrants opted out of the labour market and, when in the labour market, did not receive the same returns as other Canadians. Acknowledgments We note with appreciation the copyediting services M. Hayden of [email protected] and the ﬁnancial support provided by the Research Authority, Ruppin Academic Centre, Israel. We also thank Professor Douglas Baer for the help with the IMDB dataset. The comments of M. Justus at the 14th International Metropolis Conference helped clarify some data issues.

603

FSU Immigrants in Canada

Appendix A

Table A1.

USSR/FSU immigrants by entry gate 1980–2005

Year of landing 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002

Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage Count Percentage

within year within year within year within year within year within year within year within year within year within year within year within year within year within year within year within year within year within year within year within year within year within year within year

Family class

Skilled immigrants

Refugees

Total

56 2.7 98 11.2 131 34.7 74 34.9 73 52.1 51 45.9 42 38.5 60 26.4 79 12.8 174 10.8 255 10.3 451 21.2 769 27.0 1,065 31.6 999 23.7 973 18.7 1,186 17.6 1,392 14.6 1,437 13.0 1,638 18.7 2,006 22.0 2,381 23.0 2,343 23.3

108 5.2 59 6.8 66 17.5 35 16.5 39 27.9 o20

1,914 92.1 715 82.0 174 46.2 98 46.2 26 18.6 39 35.1 40 36.7 101 44.5 345 56.0 971 60.0 1,200 48.5 735 34.6 967 34.0 731 21.7 900 21.3 1,223 23.5 1,147 17.0 1,283 13.5 1,098 10.0 646 7.4 852 9.3 1,076 10.4 921 9.1

2,079 100.0 872 100.0 377 100.0 212 100.0 140 100.0 111 100.0 109 100.0 227 100.0 616 100.0 1,618 100.0 2,472 100.0 2,126 100.0 2,847 100.0 3,372 100.0 4,221 100.0 5,207 100.0 6,742 100.0 9,538 100.0 11,012 100.0 8,740 100.0 9,132 100.0 10,364 100.0 10,075 100.0

23 21.1 65 28.6 191 31.0 471 29.1 1,015 41.1 928 43.7 1,056 37.1 1,455 43.1 2,173 51.5 2,883 55.4 4,212 62.5 6,596 69.2 8,149 74.0 6,255 71.6 6,020 65.9 6,668 64.3 6,425 63.8

604

Don DeVoretz and Michele Battisti

Table A1. (Continued ) Year of landing 2003 2004 2005

Count Percentage within year Count Percentage within year Count Percentage within year

Family class

Skilled immigrants

Refugees

2,283 24.9 2,046 22.1 1,510 15.8

5,264 57.4 5,491 59.2 5,912 61.8

1,045 11.4 1,148 12.4 1,598 16.7

Total

9,165 100.0 9,270 100.0 9,560 100.0

Note: The total by year is higher than the sum of the three columns because the categories entrepreneur and the category other are excluded due to very low numbers, which cannot be released by IMDB. Source: Special tabulations of LIDS (Landed Immigrants Data System) from IMDB Immigration Database.

References Chiswick, B. (1978), The effect of Americanization on the earnings of foreign-born men. Journal of Political Economy 86 (5), 897–922. Dean, J., DeVoretz, D. (2000), The economic performance of Jewish immigrants to Canada: a case of double jeopardy? In: Elazar, D., Weinfeld, M. (Eds.), Still Moving. Transaction Publishers, London. DeVoretz, D., Pivnenko, S. (2006), The economics of Canadian citizenship. Journal of International Migration and Integration 6 (3/4), 435–468. Ferrer, A., Riddell, M. (2002), The role of credentials in the Canadian labour market. Canadian Journal of Economics 35 (4), 87–95. Li, P.S. (2003), Initial earnings and catch-up capacity of immigrants. Those FSU e´migre´s who arrived after 1991. Canadian Public Policy 29 (3), 319–337. Neuman, S., Oaxaca, R. (2003), Estimating labor market discrimination with selectivity corrected wage equations: methodological considerations and an illustration from Israel. The Pinhas Sapir Center for Development Paper No. 2-2003. Pendakur, K., Pendakur, R. (1998), The colour of money: earnings differentials among ethnic groups in Canada. Canadian Journal of Economics 31 (2), 347–378. Pivnenko, S., DeVoretz, D. (2003). The recent economic performance of Ukrainian immigrants in Canada and the U.S. IZA DP No. 913. Reitz, J. (2001), Immigrant skill utilization in the Canadian labour market: implications of human capital research. Journal of International Migration and Integration 2 (3), 347–378.

CHAPTER 25

What Drives Immigration Policy? Evidence Based on a Survey of Governments’ Ofﬁcials Giovanni Facchinia,b and Anna Maria Maydac a

Dipartimento di Scienze Economiche, Aziendali e Statistiche, Universita’ degli Studi di Milano, Milano, Italy b Department of Economics, Erasmus University Rotterdam, Rotterdam, The Netherlands E-mail address: [email protected] c Department of Economics and School of Foreign Service, Georgetown University, Washington, DC, USA E-mail address: [email protected]

Abstract We analyze a newly available dataset of migration policy decisions reported by governments to the United Nations Department of Economic and Social Affairs between 1976 and 2007. We ﬁnd evidence indicating that most governments have policies aimed at either maintaining the status quo or at lowering the level of migration. We also document variation in migration policy over time and across countries of different regions and income levels. Finally, we examine patterns in various aspects of destination countries’ migration policies (policies toward family reuniﬁcation, temporary vs. permanent migration, high-skilled migration). This analysis leads us to empirically investigate the determinants of destination countries’ migration policies. In particular, we examine the link between public opinion toward immigrants and governments’ policy decisions. While we ﬁnd evidence broadly consistent with the median voter model, we conclude that this framework is not sufﬁcient to understand governments’ migration policies. We discuss evidence that suggests that interest-groups dynamics may play a very important role. Keywords: Immigration, immigration policy, median voter, interest groups, political economy JEL classiﬁcations: F22, J61

Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008031

r 2010 by Emerald Group Publishing Limited. All rights reserved

606

Giovanni Facchini and Anna Maria Mayda

1. Introduction As has been pointed out by many observers (Freeman, 2006; Pritchett, 2006), we are experiencing a wave of globalization that involves everything but labor. Capital ﬂows have increased dramatically in the past decades, trade is becoming more and more important as a share of world GDP but, as of today, only slightly less than three per cent of the world population lives in a country different from the one in which it was born. Every year, international migration ﬂows involve on average only one in 600 individuals (United Nations). What explains the modest size of observed ﬂows of people across borders? The income gap between sending and receiving countries is still substantial, whereas transportation and communication costs have declined rapidly in the past few decades. Thus, supply side considerations would seem to imply – if anything – an increase in the ﬂow of migrants. It follows that the most likely driver of the limited ﬂows is to be found on the demand side and is represented by the migration policies implemented by the receiving countries. The ﬁrst goal of this chapter is to develop a political economy model of receiving countries’ migration policies based on the insights of the existing literature. In particular, we focus on the role played by voters’ attitudes toward migration as well as on political mechanisms – the median voter and the interest groups models – through which voters’ attitudes translate into migration policy decisions. The second goal is to provide information on the restrictiveness of policies toward migration as reported by governments to the United Nations Department of Economic and Social Affairs between 1976 and 2007. We ﬁnd evidence that most governments have policies aimed at either maintaining the status quo or at lowering the level of migration. We also document variation in migration policy over time and across countries of different regions and income levels. Finally, we examine patterns in different aspects of destination countries’ migration policies, such as policies toward family reuniﬁcation, temporary vs. permanent migration, high-skilled migration. Next, we merge the information contained in this dataset with crosscountry data on individual attitudes toward immigrants. We use data on public opinion from the International Social Survey Programme (ISSP), National Identity Module, for the years 1995 and 2003, and from the World Value Survey, for the years 1995–1997. The merged datasets allow us to investigate whether – within a median voter framework (Benhabib, 1996; Ortega, 2005; Facchini and Testa, 2009) – voters’ migration attitudes are consistent with governments’ migration policy decisions. Our answer is yes, but only in part. Given the very low fractions of voters in favor of increasing the number of immigrants in the majority of destination countries, restrictive migration policies are broadly consistent

What Drives Immigration Policy?

607

with a median voter framework. We also ﬁnd that in countries where the median voter and, in general, public opinion are more favorable to migration, governments’ policies tend to be more open. In other words, there exists a positive correlation between actual migration policy and public opinion across countries. This evidence suggests that policy makers take public opinion into account as they formulate policy decisions, which is in line with the predictions of the median voter model. However, given the extent of opposition to migration revealed by voters’ attitudes in the majority of destination countries, it is a puzzle that migration is allowed to take place at all. We document a ‘‘public opinion gap,’’ that is a gap between very restrictionist public opinion on one side and more open stated policy goals on the other. We conclude the chapter by discussing the alternative political mechanism – in particular, the interest groups model (Facchini and Willmann, 2005) – and its role in reconciling the evidence on attitudes with the patterns of migration policy decisions. Work in the literature suggests that many interest groups are pro-migration – for example, native workers who are complemented by foreign immigrants and capital owners (Facchini and Mayda, 2008; Facchini et al., 2008). The activities of these groups can explain the public opinion gap. The remainder of the chapter is organized as follows. Section 2 develops the political economy model of migration policy, while Sections 3 and 4, respectively, describe both the country level and the individual level data we are using. Section 5 presents evidence on the median voter model while Section 6 concludes. 2. Political economy model of migration policy How do migration policies come about? In developing our analysis, we review the explanations that have been proposed in the economics literature, while referring the interested reader to Joppke (1998), Money (1997), Freeman (1992), and Freeman (1995) for important contributions in the political science ﬁeld. A useful conceptual scheme to analyze the migration policy formation process – which is based on Rodrik (1995)1 – is illustrated in Figure 1. The basic idea is that the formulation of migration policies involves at least four elements. First, policy making necessarily needs to take into account voters’ individual preferences, and how these preferences are shaped by the inﬂows of foreign workers. Both economic and noneconomic factors are likely to play a role in shaping public opinion. The second step is to map these preferences into a policy demand. Various channels have been suggested in the literature, ranging from pressure groups to grass-roots 1

Rodrik (1995) uses the conceptual scheme in Figure 1 to analyze trade policy outcomes.

608

Giovanni Facchini and Anna Maria Mayda

individual preferences on immigration policy (A)

pressure groups, political parties, grass-roots movements (B)

“demand side” of immigration policy

Immigration policy outcomes

policymaker preferences (C)

Fig. 1.

institutional structure of government (D)

“supply side” of immigration policy

Determination of immigration policy.

movements. On the supply side of migration policies, we need to identify the policy maker preferences and to understand the details of the institutional setting in which migration policy is formulated. Building upon this framework, our starting point is the analysis of drivers of individual attitudes toward immigration.

2.1. What drives individual attitudes toward immigration? A substantial body of literature has studied the effect of both economic and noneconomic factors on attitudes toward immigration. The overall message from these studies is that, whereas noneconomic drivers have an important and independent effect on individual preferences, economic characteristics of the respondents systematically shape attitudes toward international labor mobility. The early contributions have mainly focused on individual countries like the United States (see, e.g., Espenshade and Hempstead, 1996; Citrin et al., 1997; Kessler, 2001; Scheve and Slaughter, 2001) and the United Kingdom (Dustmann and Preston, 2001, 2007). More recently, cross-country studies have taken advantage of newly available surveys, which cover large samples of both advanced and developing countries (Chiswick and Hatton, 2003; O’Rourke and Sinnott, 2005; Mayda, 2006; Facchini and Mayda, 2008, 2009b). The analysis of the economic determinants of attitudes toward immigration focuses on the income distribution effects of the inﬂow of foreign workers. Most of the literature considers a highly stylized economy

What Drives Immigration Policy?

609

that is usually described by a simple factor proportions analysis model or a two-sector Heckscher–Ohlin trade model. In both these frameworks, ignoring cases in which wages are not affected, the labor market impact of immigration depends on the skill composition of the migrants relative to the natives in the destination country. If immigrants are on average less skilled than the native population, their presence will hurt unskilled natives and beneﬁt skilled ones. On the other hand, if immigrants are on average more skilled than natives, they will beneﬁt the domestic unskilled, while hurting the skilled ones. In an early inﬂuential study using the 1992 US National Election Study, Scheve and Slaughter (2001) analyze the labor-market drivers of attitudes toward immigration and ﬁnd support for the theoretical predictions we have just discussed. In particular, in the United States, where immigrants are on average less skilled than natives, unskilled workers are more likely to oppose immigrants than skilled ones. Using the 1995 round of the ISSP and the 1995–1997 round of the World Value Survey, Mayda (2006) fully exploits the predictions of the model by taking advantage of the different skill compositions of migrants (relative to natives) across countries. This chapter ﬁnds robust evidence suggesting that individual skill is positively correlated with pro-immigration attitudes in countries where immigrants are on average unskilled, while it is negatively correlated with pro-immigration attitudes in countries where immigrants are on average skilled (relative to the native population). The main OECD destination countries of immigrant ﬂows are often characterized by large welfare states (Boeri et al., 2002), in which the public sector redistributes a substantial fraction of national income across individuals. In these contexts, immigration has a non-negligible impact on public ﬁnances, since foreign workers both contribute to and beneﬁt from the welfare state. The aggregate net effect of immigration on the welfare state can be either positive or negative, depending on the socioeconomic characteristics of immigrants relative to natives. To understand how the welfare state shapes attitudes toward immigration, Facchini and Mayda (2009b) consider a simple redistributive welfare state. In their model, an inﬂow of unskilled migrants (relative to natives) will make all natives worse off, by causing a given level of redistribution to become more costly. More speciﬁcally, assuming the welfare state adjusts through changes in taxes to maintain the same level of per capita beneﬁts (tax adjustment model), taxes will have to increase. As a consequence, everybody will be negatively affected but higher income individuals to a greater extent, as they are on the ‘‘contributing’’ end of the system. On the contrary, if the adjustment takes place through changes in per capita beneﬁts to maintain the same level of taxation (beneﬁt adjustment model), per capita beneﬁts will have to decrease. As a consequence, everybody will be adversely affected but lower income individuals to a greater extent,

610

Giovanni Facchini and Anna Maria Mayda

as they are on the ‘‘receiving’’ end of the system. Finally, if an inﬂow of skilled migrants takes place, all the above effects are reversed. All natives will gain with migration through the welfare-state channel. Under the tax adjustment model, higher-income individuals will be more positively affected than poor ones by the decrease in tax rates. Under the beneﬁt adjustment model, lower-income individuals will be more positively affected than rich ones, given that the increase in per capita beneﬁts is mostly relevant for this income category. Figures 2 and 3 (adapted from

Fig. 2.

Fig. 3.

The tax adjustment model.

The beneﬁt adjustment model.

What Drives Immigration Policy?

611

Facchini and Mayda, 2008) represent the correlations between individual income and pro-migration attitudes implied by the tax adjustment and beneﬁt adjustment models when, respectively, migration is unskilled (right panel) and skilled (left panel). In their analysis of the variation in attitudes across US states, Hanson et al. (2007) ﬁnd evidence that the positive correlation between proimmigration attitudes and education – driven by the labor market – becomes smaller in absolute value and even negative in US states where the ﬁscal exposure to immigration is high. This evidence provides empirical support for the tax adjustment model, since education and income tend to be positively correlated at the individual level. Using two surveys covering a large sample of advanced countries, and information on both the characteristics of the immigrant population and of the destination country’s welfare state, Facchini and Mayda (2009b) also ﬁnd evidence consistent with the tax adjustment model (as well as with the labor market channel). In countries where natives are on average more skilled than immigrants, individual income is negatively correlated with pro-immigration preferences, whereas individual skill is positively correlated with them. These relationships have the opposite signs in economies characterized by skilled migration (relative to the native population). Thus, their results suggest that the very same skilled and high-income German businessman may feel ambivalent regarding the arrival of unskilled immigrants since he might beneﬁt from hiring them (labor market complementarity) but be hurt by paying their way through the welfare state. The authors conﬁrm these results when they exploit international differences in the characteristics of the destination countries’ welfare state. Which economic channel matters most in shaping attitudes? Focusing on a group of advanced countries, Dustmann and Preston (2004) ﬁnd that welfare state determinants are more important than the two other economic channels (labor-market competition and efﬁciency considerations) in shaping immigration preferences. Besides economic drivers, Scheve and Slaughter (2001), Mayda (2006), and O’Rourke and Sinnott (2005) also consider noneconomic factors such as the perceived crime and cultural impact of immigration, racism, sciovinism, etc., although these factors are not the main focus of their analysis. In addition, a recent paper investigates the role of voters’ media exposure in shaping individual migration preferences (Facchini et al., 2009).

2.2. From individual preferences to migration policy Individual preferences are aggregated and become political demands thanks to the working of grass-root movements, political parties and/or

612

Giovanni Facchini and Anna Maria Mayda

interest groups (box B in Figure 1). This process of aggregation is clearly affected by how severe the collective action problem is for certain groups, which in turn is driven by several factors, for example the geographic concentration of members of a group. On the supply side of migration policy, government preferences play an important role (box C). Do ofﬁcials care only about aggregate welfare, that is, do they just wish to maximize society’s well being? Do they care only about being re-elected, that is, do they try to please the majority? Are their choices driven by ideological considerations? Do policymakers care more about the demands of speciﬁc groups within society, that is, do they use migration policy as a tool to transfer resources to a speciﬁc group? Finally, the institutional structure of the government, that is, for instance which body is in charge of setting migration policy, plays an important role (box D). These three dimensions of the policy making process are modeled together by the existing literature and the detail to which they are analyzed varies substantially. Whereas quite a bit of attention has been dedicated to the process through which individual preferences are aggregated, the policy makers preferences are modeled in a very reduced form fashion, and almost no attention is paid to the details of the institutional setting in which migration policy is set. This is an important shortcoming, as the destination countries vary substantially in their political institutions. In the remainder of this section, we describe the two main frameworks that have been proposed by the literature, the median voter framework and the pressure group model, and assess their empirical performance. 2.2.1. The median voter model What is the migration policy chosen by a stylized democracy? In a very elegant paper, Benhabib (1996) considers the human (physical) capital requirements that would be imposed on potential immigrants by an income-maximizing polity under majority voting. Output is modeled using a constant returns to scale production function combining labor with human (or physical) capital. Each individual is endowed with labor and capital, and the distribution of the latter in both the native and (potential) immigrant populations are known. Benhabib shows that the policy that will defeat any other in a binary contest is the one in which the median voter chooses to admit individuals who supply a set of factors that are complementary to her own endowment. This implies that, if the median voter is unskilled, she will choose the policy that guarantees the highest possible ex-post average level of human capital. On the contrary, if the median voter is highly educated, she will choose the policy that minimizes the ex-post average human capital level in society. In a dynamic extension of Benhabib’s model, Ortega (2005) explores the trade off between the short run economic impact of immigration and its

What Drives Immigration Policy?

613

medium to long run political effect. In particular, while immigration affects only the labor market in the current period, in the future it also inﬂuences the political balance of the destination country, as the descendants of migrants gain the right to vote. As a result, on the one hand, skilled natives prefer an immigration policy that admits unskilled foreign workers since, due to complementarities in production, this policy will increase the skilled wage. On the other, the arrival of unskilled immigrants and the persistence of skill levels across generations can give rise to a situation in which unskilled workers gain the political majority and, therefore, vote for policies that beneﬁt them as a group. Thus, through the political channel, skilled natives prefer an immigration policy that admits skilled foreign workers. The interplay between these two forces allows Ortega to characterize under which conditions an equilibrium migration quota might arise, that is, to derive a prediction in terms of the size of the migration inﬂows. 2.2.2. The pressure group model The median voter model is a useful framework to understand the process of aggregation of individual preferences into migration policy but is hardly able to capture the complexity of the political process in modern democratic societies. In particular, there is substantial anecdotal evidence suggesting that interest groups have been very actively involved in shaping policy toward immigration. For instance, in the United States – at least until very recently – labor unions have been active in limiting the inﬂows of foreign workers (Briggs, 2001; Watts, 2002). At the same time, there is ample evidence on the role played by pro-immigrant lobbies, representing the business sector, in shaping migration policy (Facchini et al., 2008). To formally study the role played by pressure groups in shaping policy toward international factor mobility, Facchini and Willmann (2005) develop a simple theoretical model, which is based on the menu auction framework pioneered by Bernheim and Whinston (1986). In their setting, policy is determined as the result of the interaction between organized groups – representing production factors – who maximize the net welfare of their members, and an elected politician who trades off aggregate welfare vis-a-vis political contribution. Using a one-good multiple factors framework, Facchini and Willmann (2005) ﬁnd that policies depend on both whether a production factor is represented or not by a lobby and on the degree of substitutability/complementarity between domestic and imported factors. In particular, ﬁrst they show that a nonorganized factor will not be able to inﬂuence the policy determination process. Second, an organized factor will instead be effective in reducing the inﬂow of a substitute, while it will increase the inﬂow of a complement. A lobbying model has also been studied by Epstein and Nitzan (2006), in a setting where two groups compete against each other to determine the policy and the role played by the existing status quo migration policy is highlighted.

614

Giovanni Facchini and Anna Maria Mayda

These models are able to rationalize both the intense lobbying activities recently carried out, for example, by healthcare providers in the United States – which resulted in the introduction of the new H1C visa category for nurses in 1999 – and the ﬁerce opposition of the union representing local nurses (Facchini et al., 2008).

3. Governments’ views and policies toward immigration In this section, we explore governments’ views and policies toward immigration using the information gathered by the United Nations Department of Economic and Social Affairs, between 1976 and 2007.

3.1. Governments’ views toward immigration The 1974 World Population Conference held in Bucharest developed a World Population Plan of Action and called for a systematic monitoring of population policies across member countries. Data have been collected since the mid seventies and provide information on a broad range of issues. Concerning immigration, two sets of questions have been asked. First, an effort has been carried out to elicit governments’ views on the overall level of immigration. Second, information has been collected on government policies toward immigration, both at the aggregate level, as well as with respect to speciﬁc issues. We start by considering government views. View on immigration is the government’s view on the level of documented immigration into the country, including immigration for permanent settlement, temporary settlement, high-skilled work, and family reuniﬁcation. Government views toward asylum seekers, refugees and undocumented migrants are not reﬂected in this variable.2 The variable can take three possible values: ‘‘too high,’’ ‘‘satisfactory,’’ and ‘‘too low.’’ The summary statistics are reported in Table 1. As we can see, on average for the period considered (1976–2007) about 79% of the ofﬁcials who have been interviewed have claimed to be satisﬁed with the current levels of immigration. In addition, 17% deem the immigration level too high, while only 5% share the view that immigration is too low (these percentages are broadly similar when we focus on Western European and North American countries, which represent the main destinations of migration ﬂows – see Table A1). 2

Notice that this question focuses on the current, status quo immigration level. While the wording of the question is not ideal for the purpose of eliciting government preferences on the overall level of immigration, it is very close to the one used in the individual level surveys that we discuss in Section 4.

615

What Drives Immigration Policy?

Table 1.

Governments’ view on the level of immigration, by year and income Governments’ view on immigration Too high

Satisfactory

Too low

Total

Total # observations

By year 1976 1986 1996 2007

6.67 20.12 21.35 17.44

86 76.22 76.56 76.92

7.33 3.66 2.08 5.64

100 100 100 100

150 164 192 195

Total

16.83

78.6

4.56

100

701

24.72 18.8 12.97 13.16

68.54 75.19 82.16 85.79

6.74 6.02 4.86 1.05

100 100 100 100

178 133 185 190

17.2

78.28

4.52

100

686

By income High income Upper middle income Lower middle income Low income Total

Notes: The table presents row percentages by year and by income. View on immigration is the government’s view on the level of documented immigration into the country, including immigration for permanent settlement, temporary and high-skilled work, and family reuniﬁcation. Governments’ views toward asylum seekers, refugees, and undocumented migrants are not reﬂected in this variable. The possible values of View on immigration are: the government has indicated that immigration is too high, satisfactory, too low.

These aggregate data hide a substantial degree of heterogeneity both over time and across country groups (see top and bottom panel of Table 1). Over time, between 1976 and 2007, the number of countries where immigration levels are perceived to be ‘‘too high’’ becomes almost three times as large. This is true even though the pattern is nonlinear: between 1976 and 1996 we observe a monotonic increase in the share of governments opposed to immigration while, between 1996 and 2007, a slight decline (from 21% to 17%). Consistent with that, the share of governments which perceive immigration levels as too low has declined from 7.3% to 5.6% between 1976 and 2007, but again the pattern is nonlinear. In general, governments’ views toward migration worsen between 1976 and 1996 and improve in the last decade. It is also interesting to explore how views are affected by different levels of development. The bottom panel of Table 1 reports summary statistics according to income levels (we use the World Bank classiﬁcation of high income, upper middle income, lower middle income and low income countries, as contained in the 2009 World Development Report). It is immediately evident that governments are more likely to perceive migration levels as too high, the higher is the per capita income level in the receiving country. Per capita GDP is thought to be a good proxy for the

616

Giovanni Facchini and Anna Maria Mayda

size of migration ﬂows and the relative skill composition of the native vs. immigrant populations – i.e., richer countries receive on average larger ﬂows and more unskilled immigrants relative to natives (Mayda, 2006; Mayda, 2009). The summary statistics for governments’ view are then consistent with evidence that larger ﬂows of migrants and unskilled migrants are perceived less favorably by individual voters as well (Mayda, 2006; Facchini and Mayda, 2008). Considering different regions of the World (Table A1 in the Appendix), it is interesting to note that among the most important destinations, ofﬁcials in the Gulf countries3 are the most concerned with the current levels of immigration. On average, 46% of the respondents in Gulf countries believe immigration levels to be too high. By comparison, in Western Europe and North America, 29% and 27% of the ofﬁcials respectively share the same view.

3.2. Governments’ policies toward immigration Although a systematic, objective measure of the restrictiveness of immigration policies does not exist on a cross-country comparable scale, the United Nations collect a wide range of information on government policies toward immigration through a survey of each country’s government ofﬁcials.4 The survey questions cover not only the general policy, but also individual policy aspects. Table 2 reports summary statistics on policies regarding the ‘‘overall level of immigration’’.5 While 11% of the ofﬁcials report that their country did not pursue an active migration policy, the majority of government ofﬁcials appears to exhibit a strong status quo bias. On average, between 1976 and 2007, over 60% of the government respondents reports that their country has policies in place to keep immigration levels unchanged. On the contrary, 23% reports that their countries have actively pursued a reduction in immigration ﬂows, while only 5% reports that their countries have tried to increase immigration levels. Once again, aggregate data hide substantial heterogeneity. Over time, the number of countries which have implemented policies geared toward lowering immigration ﬂows has ﬂuctuated widely. In 1976 less than 7% of the ofﬁcials interviewed report policies aimed at reducing inﬂows, while 3

Gulf countries include Bahrain, Kuwait, Oman, Qatar, Saudi Arabia, United Arab Emirates. 4 See UN ‘‘World Population Policies,’’ various issues. 5 Note that the question on governments’ policy does not explicitly distinguish between stocks and ﬂows of immigrants. Also, the contents of the question on governments’ views and governments’ policy are slightly different. In the former, issues related to asylum seekers, refugees and undocumented migrants are speciﬁcally excluded, while in the latter no such distinction is explicitly made.

617

What Drives Immigration Policy?

Table 2.

Governments’ policy on the level of immigration, by year and income Policy on immigration Lower Maintain Raise No intervention Total Total # observations

Year 1976 1986 1996 2007 Total Income High income Upper middle income Lower middle income Lower income Total

6.67 20.12 40.63 19.49

86 76.22 30.21 58.46

7.33 3.66 4.17 5.64

0 0 25 16.41

100 100 100 100

150 164 192 195

22.68

60.77

5.14

11.41

100

701

37.08 21.8 16.76 16.84

53.93 67.67 63.78 58.42

7.87 6.02 5.95 1.05

1.12 4.51 13.51 23.68

100 100 100 100

178 133 185 190

23.03

60.5

5.1

11.37

100

686

Notes: The table presents row percentages by year and income. Policy on immigration is the government’s policy regarding the overall level of immigration. The possible values of Policy on immigration are: The government has policies in place to lower, maintain, raise the overall level of immigration; the government does not intervene with regard to the overall level of immigration.

in 1996 this number grows to over 40%. In 2007, this number shrinks to 19%, suggesting a growing acceptance of migration. Across country groups, destinations with higher GDP per capita are more likely to report restrictive immigration policies in place (bottom panel of Table 2). This mirrors closely the ﬁndings on government views. Looking at geographic aggregates (Table A1 in the Appendix), the Gulf countries stand out as the group most actively pursuing policies aimed at reducing immigrant ﬂows, with 50% of the ofﬁcials interviewed characterizing policy in this way. Similarly, 46% of Western European ofﬁcials also report policies in place to reduce foreign workers arrivals, while the same is true for only 20% of their North American counterparts. Migration policy involves the use of a complex array of measures, and several important dimensions are considered in the data published by the UN. The ﬁrst important distinction involves permanent vs. temporary migration, and both in 1996 and 2007 ofﬁcials have been asked to state whether policy on permanent settlement and on temporary workers has been aimed at lowering, maintaining, increasing the current level or has involved no intervention at all (Tables 3 and 4). Overall, slightly over 20% of the respondents report no speciﬁc intervention on either dimension. Over 40% aim instead at maintaining

618

Table 3.

Giovanni Facchini and Anna Maria Mayda

Governments’ policy on permanent settlement, by year and income Policy on permanent settlement Lower Maintain Raise No intervention Total Total # observations

Year 1996 2007 Total Income High income Upper middle income Lower middle income Lower income Total

10 21.66

20.44 58.6

4.38 6.37

35.77 13.38

100 100

137 157

29.93

40.82

5.44

23.81

100

294

42.5 30.16 26.92 18.84

42.5 55.56 42.31 23.19

11.25 3.17 5.13 0

3.75 11.11 25.64 57.97

100 100 100 100

80 63 78 69

30

40.69

5.17

24.14

100

290

Notes: The table presents row percentages by year and income. Policy on permanent settlement is the government’s policy on migration for the purpose of permanent settlement. The possible values of Policy on permanent settlement are: The government has policies in place to lower, maintain, raise migration for permanent settlement; the government does not intervene with regard to migration for permanent settlement (or it is not known whether the government intervenesy ).

Table 4.

Governments’ policy on temporary workers, by year and income Policy on temporary workers Lower Maintain Raise No intervention Total Total # observations

Year 1996 2007 Total Income High income Upper middle income Lower middle income Lower income Total

34.48 24.85

25 60

1.72 5.45

38.79 9.7

100 100

116 165

28.83

45.55

3.91

21.71

100

281

37.18 25 24.32 25.81

50 60.94 44.59 27.42

6.41 3.13 2.7 1.61

6.41 10.94 28.38 45.16

100 100 100 100

78 64 74 62

28.42

46.04

3.6

21.94

100

278

Notes: The table presents row percentages by year and income. Policy on temporary workers is the government’s policy on the migration of temporary workers. The possible values of Policy on temporary workers are: The government has policies in place to lower, maintain, raise the migration for temporary workers; the government does not intervene with regard to the migration of temporary workers (or it is not known whether the government intervenesy ).

What Drives Immigration Policy?

619

the current levels of permanent/temporary settlements, while 28–30% try to lower these levels. Only 4–5% of the respondents report that their policies have the objective of increasing the number of migrants. Interestingly, comparing the two years for which we have data, we can see that the fraction of ofﬁcials – reporting policies in place to maintain the current levels of permanent and temporary migrants – has substantially increased between 1996 and 2007, from 20–25% to 58–60%. This suggest once again growing acceptance of the phenomenon. Looking across income groups (bottom panel of Tables 3 and 4), we can note how high GDP countries have a more interventionist stance (only 4–6% of the respondents report no intervention), and are on average more likely to have policies that are both aimed at reducing permanent settlements (43% vs. a sample average of 30%) and the number of temporary workers (37% vs. a sample average of 28%). Among the immigrant destinations, Gulf countries stand out once again as the most likely to have active policies in place to limit both temporary and permanent settlement (88% and 67%, respectively) (Table A2 in the Appendix). North America is instead the region in which ofﬁcials are more likely to be trying to raise the inﬂow of both temporary and permanent immigrants (17% and 14% respectively). The main channels of entry of immigrants into a destination country are family reuniﬁcation, work, and asylum seeker/political refugee. As selective policies have become more widespread, it is interesting to investigate the role played by the channel of entry – as it has been shown to have an important role in shaping the skill composition of the foreign population arriving in the country.6 Ofﬁcials have been asked whether governments have policies in place to raise/maintain/lower migration for family reuniﬁcation or if they do not actively intervene in this policy area. Summary statistics are reported in Table 5. Interestingly, they suggests that less than 4% of the countries considered have policies in place to increase the number of immigrants arriving for the purpose of family reuniﬁcation, while 16% are actively trying to make it harder to migrate taking advantage of this channel. Interestingly, it is important to note that over time the acceptance of family reuniﬁcation as a channel of entry has vastly increased. As reported by the OECD (2008), 44% of the new immigrants arrived in OECD countries in 2006 were admitted as family members and, in the face of this development, government policies are becoming more accommodating. In fact, in 1996 only 27% of the ofﬁcials reported efforts to maintain the current immigration levels through this channel, while 25% indicated that policy was trying to limit the inﬂows for the purpose of family reuniﬁcation. In 2007, these ﬁgures had changed to 68% and 9% respectively. As both theoretical models and the empirical evidence suggest, the skill composition of migrants compared to the native population is likely to be a

6

See for instance Boeri et al. (2002), Borjas (1999).

620

Table 5.

Giovanni Facchini and Anna Maria Mayda

Governments’ policy on family reuniﬁcation, by year and income Policy on family reuniﬁcation

Year 1996 2007 Total Income High income Upper middle income Lower middle income Lower income Total

Lower Maintain Raise No intervention Total

Total # observation

24.78 8.57

27.43 67.86

1.77 5.71

46.02 17.86

100 100

113 140

15.81

49.8

3.95

30.43

100

253

23.68 14.04 10.77 12.96

61.84 57.89 56.92 16.67

5.26 3.51 3.08 3.7

9.21 24.56 29.23 66.67

100 100 100 100

76 57 65 54

15.87

50

3.97

30.16

100

252

Notes: The table presents row percentages by year and income. Policy on family reuniﬁcation is the government’s policy concerning migration for the purpose of family reuniﬁcation. The possible values of Policy on family reuniﬁcation are: The government has policies in place to lower, maintain, raise migration for the purpose of family reuniﬁcation; the government does not intervene with regard to migration for family reuniﬁcation (or it is not known whether the government intervenesy ).

key factor to understand the labor market impact of immigration. In the 2007 survey, a question has been introduced to assess the policies implemented toward highly skilled workers (Table 6). Ofﬁcials in only ﬁve countries (Bhutan, Botswana, Jordan, Saudi Arabia and the United Arab Emirates) have reported that policies are in place to reduce the arrivals of this type of workers. Over 80% of the countries in the sample have instead policies in place to maintain or increase the number of skilled migrants arriving. This ﬁnding is particularly interesting. From our previous discussion of the labor market effects of immigration, we expect that skilled migrants will be welcome in countries where skilled labor is the scarce factor (and therefore the median voter is likely to be unskilled), but they will not be viewed favorably in countries that are skilled labor abundant (i.e. countries where the median voter might be skilled). The results from the UN survey suggest instead that government policies are favorable to skilled migrants also in countries where skilled labor is ‘‘abundant’’. There are at least three possible explanations for this result. First, through the welfare state channel, every individual – both skilled and unskilled – prefers skilled migration to unskilled migration, since skilled (unskilled) migrants are likely to represent a net contribution (burden) for the destination country’s welfare state. Secondly, high income countries might prefer highly skilled migrant workers for simple but intuitive political reasons. Natives might realize that immigrants or their children will

621

What Drives Immigration Policy?

Table 6.

Governments’ policy on highly skilled workers, by year and income Policy on highly skilled workers Lower Maintain Raise No intervention Total Total # observations

Year 2007

3.47

58.33

25

13.19

100

144

Income High income Upper middle income Lower middle income Lower income

4.44 2.78 5.41 0

40 66.67 78.38 52

44.44 25 10.81 12

11.11 5.56 5.41 36

100 100 100 100

45 36 37 25

3.5

58.74

25.17

12.59

100

143

Total

Notes: The table presents frequencies and row percentages by year and by income. Policy on highly skilled workers is the government’s policy on the migration of highly skilled workers. The possible values of Policy on highly skilled workers are: The government has policies in place to lower, maintain, raise the migration of highly skilled workers; the government does not intervene with regard to the migration of highly skilled workers (or it is not known whether the government intervenesy ).

eventually become citizens and thus will be allowed to vote. If skill levels are persistent across generations – as the existing evidence suggests (Ortega, 2005) – skilled native voters will favor (oppose) skilled (unskilled) migrants because the latter ones will tend to vote for policies that favor skilled (unskilled) individuals (Ortega, 2005). Third, cultural assimilation is easier for highly skilled migrants than for low-skilled migrants, and therefore countries of different income groups in general tend to prefer highly skilled migrants (Chiswick and Miller, 2006). Finally, as the number of individuals living in foreign countries has rapidly grown in the past decades, it is important to assess whether destination countries are actively promoting the integration of immigrants. A question on this issue has been asked both in 1996 and 2007 (Tables 7–9). On average, 54% of the ofﬁcials interviewed report active integration policies to be in place, and this number has grown over time (from 44% in 1996 to 64% in 2007). Furthermore, the overwhelming majority of high income countries (82%) have policies in place for the integration of foreigners, while poorer countries appear to be less active in this policy area. This result is conﬁrmed in Table A4, where we investigate the determinants of policies on the integration of noncitizens. While it appears at ﬁrst that countries receiving higher immigrant inﬂows are the ones promoting integration policies [see columns (1) and (4)], it is the per capita GDP of the destination country which seems to drive this result. In other words, higher income countries (which are also the ones receiving

622

Giovanni Facchini and Anna Maria Mayda

Table 7.

Policies on integration of noncitizens, by year

Year

Integration of noncitizens No

Yes

Total

Total # observations

1996 2007

56.3 36.29

43.7 63.71

100 100

119 124

Total

46.09

53.91

100

243

Notes: The tables present row percentages. Integration of noncitizens is government’s policies or programs to foster the integration of noncitizens into society. The possible values are: Yes, the government has policies or programs to foster the integration of noncitizens (e.g., language classes, provision of social services); No, the government has no policies or programs to foster the integration of noncitizens (or it is not known whether the government has a policy or program to foster the integration of noncitizens into society).

Table 8.

Policies on integration of noncitizens, by income

Income

Integration of noncitizens No

Yes

Total

Total # observations

High income Upper middle income Lower middle income Lower income

17.57 36 64.41 72.88

82.43 64 35.59 27.12

100 100 100 100

74 50 59 59

Total

46.28

53.72

100

242

Notes: Please refer Notes in Table 7.

larger immigrant inﬂows) are more likely to develop integration policies [see columns (2) and (5)]. Finally, we ﬁnd some evidence that countries characterized by less skilled immigrants are more likely to promote integration policies [see columns (3) and (6)].

4. Individual attitudes toward immigrants Are natives in favor of or against an increase in migration to their countries? Are there differences in public opinion toward immigration across destination countries? We consider evidence from two sets of individual level surveys. The ﬁrst is the National Identity module of the ISSP (see also Mayda, 2006; Facchini and Mayda, 2008), which has been carried out in 1995 and 2003 and covers a large sample of respondents from mainly advanced OECD and middle income countries (see Tables 10 and 11 – based on Facchini and Mayda (2008) – for

623

What Drives Immigration Policy?

Table 9.

Policies on integration of noncitizens, by region

Region

Integration of noncitizens No

Yes

Total

Total # observations

Central Asia East Asia and Paciﬁc Eastern Europe Gulf countries Latin America and Caribbean Middle East North America North Africa South Asia Sub-Saharan Africa Western Europe

33.33 53.85 31.25 50 53.33 62.5 42.86 50 83.33 64.29 7.69

66.67 46.15 68.75 50 46.67 37.5 57.14 50 16.67 35.71 92.31

100 100 100 100 100 100 100 100 100 100 100

9 26 32 4 45 8 7 4 12 56 39

Total

46.28

53.72

100

242

Notes: Please refer Notes in Table 7.

summary statistics on these surveys). The second is represented by the third wave of the World Value Survey (WVS), which was carried out in 1995–1997. The WVS data set includes more than 50,000 respondents based in 44 mostly developing countries (see Table 12 for summary statistics based on this survey). To construct measures of attitudes toward immigration from the ISSP survey, we use respondents’ answers to the following question: ‘‘There are different opinions about immigrants from other countries living in (respondent’s country). By ‘immigrants’ we mean people who come to settle in (respondent’s country). Do you think the number of immigrants to (respondent’s country) nowadays should be: (a) reduced a lot, (b) reduced a little, (c) remain the same as it is, (d) increased a little, or (e) increased a lot’’? The survey format also allows for ‘‘can’t choose’’ and ‘‘not available’’ responses (which we treat as missing values and thus exclude from the sample in our speciﬁcations). In 1995, in the sample of countries considered (see list in Table 10), individuals are on average very opposed to immigration: only 7.39% of individuals – who give an opinion on this issue – agree with the statement that the number of immigrants to their countries should be increased either a little or a lot. The average of the variable Pro Immig Opinion in the overall sample equals 2.13.7 Finally, the median value of the same variable in the overall sample is equal to 2.

7

Pro Immig Opinion uses answers to the immigration question and ranges from 1 (reduced a lot) to 5 (increased a lot).

29.60

Overall

24.59

24.72 17.19 20.58 25.75 22.37 23.77 24.19 13.56 30.31 21.82 20.05 30.99 31.65 29.26 27.14 17.53 22.15 24.51 29.92 26.64 29.25 25.19

26.27

37.74 9.77 32.89 21.14 17.37 25.89 13.51 55.35 19.87 35.03 17.19 30.79 24.06 27.32 25.63 19.91 22.28 24.3 31.76 45.49 21.88 21.83

4.82

2.93 2.17 12.17 1.9 1.74 2.8 0.71 15.6 2.56 10.11 0.26 4.42 8.59 5.7 7.2 4.13 3.99 1.81 1.35 6.39 4.13 4.58 1.60

0.81 1.54 5.99 0.27 0.54 1.06 0.71 2.24 0.82 2.95 0.13 0.68 2.22 1.21 3.77 1.82 1.46 0.65 0.39 1.07 2.11 2.14

Reduced Remain the Increased Increased a little (2) same as it is (3) a little (4) a lot (5)

Pro Immig Opinion

13.12

5.45 36.74 11.9 11.2 9.9 6.38 4.94 6.62 4.67 16.72 12.63 6.75 6.68 6.98 4.36 30.68 34.05 18.51 6.66 11.64 6.97 16.57 2.13

2.19 1.78 2.67 1.84 1.72 1.94 1.59 2.93 1.85 2.61 1.64 2.16 2.23 2.14 2.20 2.11 2.28 2.00 2.06 2.60 2.01 2.09 2

2 1 3 2 1 2 1 3 2 3 1 2 2 2 2 2 2 2 2 3 2 2 0.07

0.04 0.06 0.21 0.02 0.03 0.04 0.01 0.19 0.04 0.16 0.00 0.05 0.12 0.07 0.11 0.09 0.08 0.03 0.02 0.08 0.07 0.08 12

15 12 10 11 11 12 12 12 11 12 14 12 10 10 11 12 11 9 11 13

14.76 12.91 10.92 11.32 10.49 12.25 11.03 11.87 11.61 12.69 14.31 12.66 9.39 10.29 11.19 11.84 10.68 10.13 11.41 13.43 11.68

9

10.36

20171

29102 6312 27861 16144 27148 25268 11216 22908 25324 28235 6487 29039 20948 39455 2502 9398 7591 11485 16227 21343 25142 34619 2.12

0.26 1.32 5.57

2.52

2.74 0.96 2.83

4.27 1.38 1.84 0.41 0.76 1.43

2.74

2.73

Per capita Relative skill Median Average Average Median Average mix (natives vs. Pro Immig Pro Immig Pro Immig education education GDP, Missing Opinion (7) Opinion (8) Dummy (9) years (10) years (11) PPP (12) immigration) values (6) (13)

0.0015

0.0013 0.0025 0.0049 0.0002 0.0028 0.0017 0.0016 0.0050 0.0021 0.0004 0.0032 0.0023 0.0021 0.0024 0.0026 0.0016 0.0030 0.0000 0.0024 0.0040 0.0011 0.0046

1.3636

1 2 1 2 1 1 1 2 1 2 1 1 3 1 1 2 1 1 1 1 1 2

Policy on Net migration immigration 1996 (14) 1996 (15)

Summary Statistics of Individual Attitudes towards Immigration (ISSP 1995) and country-level variables

Sources: United Nations. Data for columns (1)–(11) is from the 1995 ISSP National Identity Module. Data for per capita GDP, PPP is for 1996 (World Development Indicators). Data for the relative skill mix is for 1990/1991/1996 (Docquier, 2007; Barro and Lee, 2000). Data on net migration 1996 is from the United Nations. Data on policy on immigration 1996 is from the United Nations. Notes: The survey sample excludes non-citizens. Pro Immig Opinion uses answers to the immigration question (‘‘Do you think the number of immigrants to (R’s country) nowadays should be y’’: reduced a lot, reduced a little, remain the same as it is, increased a little, increased a lot) and ranges from 1 (reduced a lot) to 5 (increased a lot). Pro Immig Dummy equals one if Pro Immig Opinion is equal to 4 or 5, zero if Pro Immig Opinion is equal to 1, 2 or 3. Both variables exclude missing values. Net migration is equal to the net migration inﬂow, divided by the destination country’s population, in 1996.

28.36 32.58 16.48 39.75 48.07 40.1 55.95 6.63 41.76 13.38 49.74 26.37 26.79 29.53 31.91 25.92 16.08 30.22 29.92 8.77 35.66 29.69

Reduced a lot (1)

Austria Bulgaria Canada Czech Republic Germany Great Britain Hungary Ireland Italy Japan Latvia Netherlands New Zealand Norway Philippines Poland Russia Slovak Republic Slovenia Spain Sweden USA

Country

Table 10.

Australia Austria Bulgaria Canada Chile Czech Republic Denmark Finland France Germany Great Britain Hungary Ireland Israel Japan Latvia Netherlands New Zealand Norway Philippines Poland

Country

Pro Immig Opinion

16.79 32.72 16.17 10.21 22.78 26.19 25.87 15.83 35.37 44.29 50.88 34.38 27.65 26.68 20.15 26.36 37.84 26.81 36.37 17.92 19.42

19.65 26.75 18.89 18.65 37.23 30.95 21.63 15.61 21.38 23.66 22.68 30.56 28.81 16.49 22.32 24.09 26.95 27.62 29.80 19.58 20.67

34.71 29.94 20.11 34.51 29.23 4.76 35.93 36.97 22.30 19.39 14.81 27.23 30.73 26.68 28.58 30.01 23.86 25.28 19.28 37.67 28.97

15.81 5.25 2.26 19.92 4.84 2.38 7.87 18.70 4.09 2.79 3.41 1.67 7.32 12.10 8.44 1.51 2.47 10.70 5.01 11.50 3.52

5.72 1.03 0.85 5.99 1.61 2.38 1.21 3.02 2.20 0.90 1.76 0.39 1.06 13.92 2.36 0.63 0.95 3.06 1.13 5.58 1.72

7.32 4.31 41.72 10.72 4.31 33.34 7.49 9.87 14.66 8.97 6.46 5.77 4.43 4.13 18.15 17.40 7.93 6.53 8.41 7.75 25.70

2.72 2.11 2.19 2.92 2.22 1.86 2.32 2.75 2.02 1.82 1.74 1.97 2.22 2.69 2.40 2.10 1.93 2.31 1.96 2.64 2.29

3 2 2 3 2 2 2 3 2 2 1 2 2 3 2 2 2 2 2 3 2

0.23 0.07 0.05 0.29 0.07 0.07 0.10 0.24 0.07 0.04 0.06 0.02 0.09 0.27 0.13 0.03 0.04 0.15 0.07 0.19 0.07

13.06 11.08 11.11 13.46 10.71 13.15 13.18 11.98 13.68 10.68 11.78 10.74 12.92 13.41 12.03 12.69 13.59 13.28 13.45 9.66 10.82

Average Average Median Average Pro Immig Pro Immig Pro Immig education Opinion (7) Opinion (8) Dummy (9) years (10)

13 10 11 13 12 12 13 12 13 11 11 11 13 12 12 12 13 13 13 10 10

Median education years (11)

33011 35974 10665 36178 13108 22505 35062 33269 31455 32149 32766 17960 40168 24466 31607 16536 36580 25306 49707 3219 15401

2.38

2.08 1.02 1.93

1.34

1.73 2.26 3.69 4.46 0.72 1.87 0.28

1.65

1.50 2.13

3 2 2 3 2 2 1 3 1 2 2 2 2 3 2 2 1 3 2 2 2

Policy on Per capita Relative skill GDP, PPP mix (natives vs. immigration 2007 (14) immigration) (12) (13)

Summary Statistics of Individual Attitudes toward Immigration (ISSP 2003) and country-level variables

Reduced Reduced Remain Increased Increased Missing a lot (1) a little (2) the same a little (4) a lot (5) values (6) as it is (3)

Table 11.

Pro Immig Opinion

23.88

Overall

23.78

35.01 25.14 15.58 32.05 23.35 35.16 27.30 27.02 31.76 20.35 28.74 28.38

27.32

39.10 10.26 25.15 43.34 34.52 35.66 26.95 45.64 18.01 46.41 28.66 42.18

6.63

2.38 1.64 7.14 2.48 17.57 5.80 8.05 5.11 3.33 12.80 5.47 3.95 2.48

0.59 1.68 2.09 0.37 5.32 2.44 2.27 0.32 1.09 5.89 3.34 2.81 15.91

3.83 22.27 23.67 5.05 10.11 7.74 9.88 5.00 11.47 8.38 10.09 2.64 2.29

2.28 1.74 2.25 2.34 2.85 2.45 2.27 2.42 1.93 2.91 2.29 2.40 2

2 1 2 2 3 2 2 3 2 3 2 3 0.11

0.03 0.04 0.12 0.03 0.25 0.09 0.11 0.06 0.05 0.20 0.10 0.07 11.89

8.12 11.59 13.51 11.20 12.30 10.00 12.10 11.36 11.30 9.12 13.88

Average Average Median Average Pro Immig Pro Immig Pro Immig education Opinion (7) Opinion (8) Dummy (9) years (10)

12

6 12 13 11 12 10 12 10 12 9 14

Median education years (11)

26374

10609 43227 11487

20488 13918 19241 25576 23363 28333 33760 36873

1.96

5.32

0.70 0.36 2.11 2.98

0.56

2.1212

2 3 2 2 3 2 2 2 2 2 2 2

Policy on Per capita Relative skill GDP, PPP mix (natives vs. immigration 2007 (14) immigration) (12) (13)

Sources: Data for columns (1)–(11) is from the 2003 ISSP National Identity Module. Data for per capita GDP, PPP is for 2007 (World Development Indicators). Data for the relative skill mix is for 1999/2000/2001/2002 (Docquier, 2007; Barro and Lee, 2000). Data on policy on immigration 2007 is from the United Nations. Notes: The survey sample excludes non-citizens. Pro Immig Opinion uses answers to the immigration question (‘‘Do you think the number of immigrants to (R’s country) should be y’’: reduced a lot, reduced a little, remain the same as it is, increased a little, increased a lot) and ranges from 1 (reduced a lot) to 5 (increased a lot). Pro-Immig Dummy equals one if Pro Immig Opinion is equal to 4 or 5, zero if Pro Immig Opinion is equal to 1, 2 or 3. Both variables exclude missing values.

19.09 39.01 26.37 16.71 9.13 13.20 25.55 16.91 34.34 6.17 23.70 20.04

Reduced Reduced Remain Increased Increased Missing a lot (1) a little (2) the same a little (4) a lot (5) values (6) as it is (3)

Portugal Russia Slovak Republic Slovenia South Korea Spain Sweden Switzerland Taiwan Uruguay USA Venezuela

Country

Table 11. (Continued )

Argentina Armenia Australia Azerbaijan Belarus Bosnia Brazil Bulgaria Chile China Croatia Dominican Republic Estonia Finland Georgia Germany India Japan Latvia Lithuania Macedonia Mexico

Country

31.96 17.05 43.37 16.37 19.54 18.16 24.85 25.12 31.31 45.18 30.49 46.35

40.74 50.94 24.26 37.41 26.98 43.80 38.88 33.01 43.17 30.15

12.07 7.48 8.09 6.32 23.73 6.51 13.51 21.30 19.29 7.99

Limits ¼ 2 (2)

39.37 30.04 46.95 43.78 17.04 45.48 41.89 38.34 15.85 43.38

51.44 47.74 49.69 56.74 50.87 32.16 36.05 44.68 49.95 37.27 49.96 36.52

4.12 7.69 17.47 10.57 10.64 4.20 3.74 3.94 13.87 14.30

7.22 20.79 3.63 17.95 15.19 39.11 24.50 5.74 10.23 6.70 11.48 11.34

If jobsy Anyone ¼ 3 (3) ¼ 4 (4)

Pro Immig Opinion

3.70 3.85 3.23 1.92 21.61 0.00 1.98 3.41 7.82 4.16

0.00 5.20 0.00 4.32 6.25 6.12 1.57 11.85 1.72 0.00 0.00 2.77 2.37 2.39 2.76 2.60 2.19 2.47 2.37 2.26 2.26 2.67

2.56 2.84 2.54 2.92 2.78 3.13 2.73 2.49 2.65 2.40 2.65 2.58 2 2 3 3 2 2 2 2 2 3

3 3 3 3 3 3 3 3 3 2 3 2 0.45 0.39 0.67 0.55 0.35 0.50 0.47 0.44 0.32 0.60

0.59 0.72 0.53 0.78 0.70 0.76 0.62 0.57 0.61 0.44 0.61 0.49 5 3 7 5 4 6 5 5 5

6.38 5.85 5.21 5.41

4 7 6 7 5 5 6 5 5 5 2 8 8492 22598 1870 27148 1489 28235 6487 7806 6209 9356

10006 1831 25476 1859 4331 3024 7794 6312 9678 2018 9284 3909

Per capita Median education GDP, PPP attainment (11) (10)

5.93 2.75 6.35 5.67 4.53

4.78 6.59 6.09 6.67 5.69 5.32 5.99 5.07 5.32 4.46 2.03 7.49

Average Median Average Average Pro Immig Pro Immig Pro Immig education Don’t Opinion (6) Opinion (7) Dummy (8) attainment know (5) (9)

0.4791

1.4260

4.2747

3.3226

1.2302

0.0052 0.0008 0.0141 0.0028 0.0003 0.0004 0.0032 0.0061 0.0005 0.0026

0.0006 0.0142 0.0051 0.0033 0.0000 0.0173 0.0003 0.0025 0.0008 0.0002 0.0067 0.0034

Net Relative skill mix (natives vs. migration 1996 (13) immigration) (12)

1 1 No intervention 1 2 2 1 1 1 1

2 No intervention 2 2 2 No intervention 2 2 2 2 No intervention 1

Policy on immigration 1996 (14)

Summary Statistics of Individual Attitudes towards Immigration (WVS 1995) and country-level variables

9.38 9.22 3.32 4.63 8.15 4.45 13.04 12.61 6.79 10.84 8.07 3.02

Prohibit ¼ 1 (1)

Table 12.

20.90 10.49

Venezuela Overall

36.60 34.32

29.01 29.83 39.95 52.06 37.51 63.00 26.59 48.38 33.41 26.54 30.22 22.90 56.34 36.39 32.36 17.40 23.51 55.98

Limits ¼ 2 (2)

39.15 40.40

41.32 31.51 36.61 41.09 36.51 15.67 45.82 27.53 45.19 37.01 58.00 55.47 30.46 50.00 37.19 47.83 55.03 30.03

3.35 10.73

10.99 14.29 17.51 4.32 7.30 9.00 5.96 5.59 4.01 19.21 2.44 13.74 8.73 3.72 3.20 17.65 11.40 4.59

If jobsy Anyone ¼ 3 (3) ¼ 4 (4)

Pro Immig Opinion

0.00 4.05

3.08 10.92 0.00 0.84 7.72 0.75 3.97 2.29 8.73 6.22 0.00 3.39 3.51 5.26 0.87 10.81 3.29 0.00 2.25 2.53

2.49 2.52 2.66 2.48 2.44 2.22 2.42 2.23 2.49 2.69 2.54 2.81 2.49 2.56 2.17 2.86 2.73 2.30 2 3

3 3 3 2 2 2 3 2 3 3 3 3 2 3 2 3 3 2 0.43 0.53

0.54 0.51 0.54 0.46 0.47 0.25 0.54 0.34 0.54 0.60 0.60 0.72 0.41 0.57 0.41 0.73 0.69 0.35 5 5

4 5 5 3 4 5 4 5 4 6

4.25 4.91 4.98 4.16 4.41 5.23 4.59 5.84 4.72 6.24 5.02 5.39

5 5 6 5 5 7 5

9935 11431

16227 21343 25142 31884 8378 3540 8598 34619

1454 39455 5364 2502 7591 7466 15597

1481

Per capita Median education GDP, PPP attainment (11) (10)

5.43 5.28 5.59 5.85 5.61 6.07 5.80

Average Average Median Average Pro Immig Pro Immig Pro Immig education Don’t Opinion (6) Opinion (7) Dummy (8) attainment know (5) (9)

2.0188

5.5674

0.2591 1.3209 2.0809 0.7004

0.7301

2.8345

0.0004 0.0008

0.0115 0.0079 0.0002 0.0024 0.0045 0.0026 0.0030 0.0019 0.0004 0.0038 0.0024 0.0040 0.0011 0.0008 0.0003 0.0021 0.0016 0.0046

Net Relative skill mix (natives vs. migration 1996 (13) immigration) (12)

No intervention

1 1 1 1 1 No intervention 3 2

2 1 No intervention 1 1 2 1

No intervention

Policy on immigration 1996 (14)

Sources: United Nations. Data for columns (1)–(11) is from the 1995 WVS. Data for per capita GDP, PPP is for 1996 (World Development Indicators). Data for the relative skill mix is for 1990/1991/1996 (Docquier, 2007; Barro and Lee, 2000). Data on net migration 1996 and on policy on immigration 1996 is from the United Nations. Notes: The survey sample excludes foreign-born. net migration is equal to the net migration inﬂow, divided by the destination country’s population, in 1996.

15.60 13.45 5.93 1.69 10.95 11.58 17.66 16.21 8.65 11.02 9.33 4.50 0.96 4.63 26.38 6.31 6.78 9.40

Prohibit ¼ 1 (1)

Moldova Montenegro Nigeria Norway Peru Philippines Russia South Africa South Korea Serbia Slovenia Spain Sweden Switzerland Turkey Ukraine Uruguay USA

Country

Table 12. (Continued )

What Drives Immigration Policy?

629

In addition, Column 9 in Table 10 clearly shows that there exists substantial variation across countries in terms of individual attitudes toward immigrants. In 1995, Canada and Ireland are the countries most favorable to migration (with, respectively, 20.61% and 19.10% of their population favoring an increase in the number of immigrants) while Latvia and Hungary are the most opposed (with, respectively, 0.45% and 1.48% of their population supporting higher migration). In general, most Central and Eastern European countries are characterized by very low percentages of voters favoring migration (Latvia, Hungary, Slovenia, Czech Republic, Slovak Republic). Among Western European countries, Italy (3.55%) and Germany (2.54%) have the most hostile public opinion to immigration. Besides Ireland, Spain is the Western European country whose citizenry is most receptive toward migrants (8.44%). Finally, in the United States, 8.05% of the population welcomes increases in migration. The 2003 data set, based on a larger sample of countries (see list of countries in Table 11), conﬁrms that voters are indeed hostile to immigration on average: only 10.84% of individuals – who give an opinion about migration – in the overall sample of countries agrees that the number of immigrants should be increased either a little or a lot. The average of the variable Pro Immig Opinion in the overall sample equals 2.29. Finally, the median value of the same variable is, in the overall sample, again equal to 2. Like in 1995, there are substantial differences in attitudes toward immigrants across countries in 2003. In particular, Column 9 in Table 11 shows that in Canada and Israel, respectively, 29.02% and 27.14% of the population favors an increase in the number of immigrants, while in Hungary and Latvia these percentages are, respectively, equal to 2.18% and 2.60%. Among Western European countries, Portugal (3.09%), the Netherlands (3.72%) and Germany (4.06%) show the public opinion that is most hostile to immigration. Finland (24.10%) is the only Western European country among the top ﬁve most open countries toward migration. In the United States, 9.8% of individuals favors larger numbers of immigrants, which is a higher percentage than in 1995 (8.05%): this is remarkable given that the September 11 terrorist attacks took place in 2001, that is, between the two surveys. In France, 7.37% of voters welcomes increases in migration. The immigration question in the 1995 round of the WVS asks the following: ‘‘How about people from other countries coming here to work. Which one of the following do you think the government should do? (a) Let anyone come who wants to? (b) Let people come as long as there are jobs available? (c) Place strict limits on the number of foreigners who can come here? (d) Prohibit people coming here from other countries? (e) Don’t know.’’ Summary statistics are reported in Table 12. To simplify the exposition, we have also constructed a Pro Immig Opinion variable that uses answers to the immigration question and ranges from 1 (prohibit) to 4 (let anyone).

630

Giovanni Facchini and Anna Maria Mayda

As can be seen from Column 9 in Table 12, 53% of individuals who gave an opinion are in favor of ‘‘letting anyone come who wants to’’ or of ‘‘letting people in as long as there are jobs available.’’ The average of the variable Pro Immig Opinion in the overall sample equals 2.53. Finally, the median value of the same variable in the overall sample is equal to 3. Note that the values of immigration attitudes in the WVS display much more favorable opinions toward migration than the ISSP dataset. This can be due in part to the different wording of the question. However, the most important reason for this difference is likely to be the different coverage of countries in the two samples. While the ISSP dataset mostly covers middle and high income countries – and therefore is representative of the most important destinations of migration ﬂows in the world – the WVS mostly covers low-income countries, which may be more favorable to migration because they are at the same time immigration and emigration countries. As in the case of the ISSP surveys, average data hide substantial differences in attitudes toward immigrants across countries. In particular, Column 8 in Table 12 shows that in Azerbaijan and Bosnia, respectively, 78% and 76% of the population favors an increase in the number of immigrants, while in the Philippines and Macedonia these percentages are, respectively, equal to 25% and 32%. Finally, comparing the summary statistics in this section and the previous one, it appears that overall governments’ views (Table 1) and policies (Table 2) are more favorable to immigration than individual voters’ attitudes. While this ‘‘public opinion gap’’ has been pointed out before in previous works (see, e.g., Freeman, 1992; Joppke, 1998; Facchini and Mayda, 2008), the UN dataset merged with individual survey data allows us to document it quantitatively for the ﬁrst time in the literature. 5. Individual opinions and immigration policy In a democratic society, voters’ attitudes should be the basis of policy making. This idea is at the core of the median voter model according to which migration policy should be correlated with the opinion of the median voter and, more in general, with public opinion.8 We next evaluate whether these predictions are consistent with the data. Our ﬁrst piece of evidence is the summary statistics in Table 10 (ISSP 1995) and Table 11 (ISSP 2003). As we have seen, voters across countries are, on average, very much opposed to immigration. Given restrictive migration policies observed across destination countries, this evidence is indeed consistent with the median-voter framework. 8

From the literature on the determinants of individual attitudes, we know that in a country that receives unskilled migrants relative to natives, a voter (and therefore the median voter) will be more pro migration the more skilled he is. On the contrary, if immigration is skilled, a voter (and therefore the median voter) will be more pro migration, the more unskilled he is.

What Drives Immigration Policy?

631

Figures 4–6 and Tables 13 and 14 provide additional evidence that is consistent with the median-voter model. Figure 4 and the left hand panel of Table 13 use data on attitudes from the 1995 ISSP dataset. Figure 5 and Table 14 use data on attitudes from the 2003 ISSP dataset. Finally, Figure 6 and the right hand panel of Table 13 use data on attitudes from the 1995–1997 WVS. In these ﬁgures and tables, we show that migration policy across countries is positively correlated with the opinion of the median voter and, in general, public opinion across countries.9 In particular, we start by relating the opinion on immigration of the median voter in each country to the migration policy of that country, as reported by its government to the United Nations. We identify the median voter using the Pro Immig Opinion variable: we rank individuals in each country according to their Pro Immig Opinion value and we next select the individual who corresponds to the 50th percentile [the opinion of this individual – median Pro Immig Opinion – appears in column (8), Table 10, in column (8), Table 11 and in column (7), Table 12]. We ﬁnd that the two variables – the opinion on immigration of the median voter and the migration policy of each country – are positively correlated with each other, even though this result is not always statistically signiﬁcant [the regression results appear in columns (1) and (4), Table 13 and column (1), Table 14]. Next, we carry out a set of robustness checks, considering the impact on the migration policy of each country of average attitudes toward immigrants [the average of the Pro Immig Opinion variable, which appears in column (7), Table 10, in column (7), Table 11, and in column (6), Table 12]. Once again, the correlation is positive and almost always signiﬁcant [the regression results appear in columns (2) and (5), Table 13 and column (2), Table 14]. Finally, we look at the impact on migration policy of the fraction of voters, in each country, favorable to an increase in the number of immigrants [the average of the Pro-Immig Dummy variable, which appears in column (9), Table 10, in column (9), Table 11, and in column (8), Table 12]. We ﬁnd a positive and this time always signiﬁcant correlation between the two variables [the regression results appear in columns (3) and (6), Table 13 and column (3), Table 14]. The higher signiﬁcance of the correlation between actual policy and voters preferences obtained using the opinion of the average rather than the median voter seems to suggest that the views of ‘‘extreme’’ groups play an important role in shaping policy measures. This suggests that the direct democracy model might be too simple to capture all the complexities of the political process.

9 See also Figures A1 and A2 and Table A5, based on Facchini and Mayda (2008), which relate migration ﬂows (divided by population) to attitudes and to migration policy, using the 1995 ISSP dataset.

Fig. 4.

The impact of individual attitudes toward immigrants on migration policy (ISSP 1995, United Nations 2007).

632 Giovanni Facchini and Anna Maria Mayda

Fig. 5.

The impact of individual attitudes toward immigrants on migration policy (ISSP 2003, United Nations 2007).

What Drives Immigration Policy? 633

Fig. 6.

The impact of individual attitudes toward immigrants on migration policy (WVS 1995, United Nations 2007).

634 Giovanni Facchini and Anna Maria Mayda

635

What Drives Immigration Policy?

Table 13. The impact of individual attitudes towards immigrants (ISSP 1995 and WVS 1995) on migration policy, 1996 OLS

1

2

3

Dependent variable

5

6

Migration policy, 1996 Using attitudes from ISSP 1995

Median Immig Opinion Average Immig Opinion Average Pro Immig Dummy Constant Observations R-squared

4

0.125 0.2087

Using attitudes from WVS 1995 0.2917 0.2011

0.3482 0.3706

1.1136 0.4359** 22 0.02

0.8566 0.5204 3.8068 2.1883* 1.0856 0.1989***

0.6257 0.7952 22 0.04

22 0.13

0.75 0.516

0.6526 1.3018

31 0.07

31 0.09

1.3401 0.7986* 0.8003 0.4193* 31 0.09

Sources: 1995 ISSP National Identity Module, WVS 1995, and United Nations. Note: Standard errors under each coefﬁcient. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

Table 14.

The impact of individual attitudes towards immigrants (ISSP 2003) on migration policy, 2007

OLS Dependent variable Median Immig Opinion

1

2

Migration policy, 2007 0.3352 0.1759*

Average Immig Opinion

0.6934 0.2650**

Average Pro Immig Dummy Constant Observations R-squared

3

1.3743 0.3955*** 37 0.09

0.5523 0.6008 37 0.16

Sources: 2003 ISSP National Identity Module and United Nations. Note: Standard errors under each coefﬁcient. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

4.2208 1.0454*** 1.6865 0.1304*** 37 0.32

636

Giovanni Facchini and Anna Maria Mayda

Finally, note also that these ﬁgures and tables treat the independent variable (attitudes) as given and exogenous. This assumption might be problematic. In particular, our estimates might be biased because of reverse causality: that is, migration policy will impact migration inﬂows, which in turn may themselves affect attitudes. As a matter of fact, Mayda (2006) ﬁnds that, in countries with higher immigrant inﬂows, voters tend to be on average more opposed to immigration. Notice, however, that this reverse causality biases the coefﬁcients in our tables and ﬁgures toward zero, thus it is not problematic for our results. Hence, Figures 4–6 and Tables 13–14 provide evidence which is broadly consistent with the median-voter framework. This evidence is the ﬁrst of its kind and complements our previous work (Facchini and Mayda, 2008). In that paper, we have used an indirect measure of migration policy, that is, net migration ﬂows to each destination country. In this chapter, on the contrary, we consider a direct measure of migration policy decisions, that is, reports to the UN by government ofﬁcials in each destination country. This represents a substantial improvement relative to our previous work since net migration ﬂows are an equilibrium outcome, that is, the result of the interaction between demand and supply factors. Instead, government ofﬁcials’ reports on immigration policy are a pure ‘‘demand’’ indicator. In addition, the UN dataset we use in this chapter allows us to document differences in migration policies across a much larger set of countries and over a longer period of time, spanning over three decades. While the empirical evidence in this section is consistent with the median-voter model, it is clear that this framework is not sufﬁcient to explain migration policy decisions. In the summary statistics section, we have documented a systematic gap between very restrictionist public opinion and more open government policies. What are the other factors that are relevant in shaping governments’ migration policies and help explain the public opinion gap? In Facchini and Mayda (2008) we investigate the impact of interest groups. We focus on the United States and use a panel covering the period 1995–2005. Differentiating labor according to both skill levels and occupation, we ﬁnd systematic evidence suggesting that the lobbying activities of organized labor lead to an increase in the inﬂow of foreign workers in different occupation/education cells. This effect is likely to be driven by complementarity.10 In addition, Facchini et al. (2008) ﬁnd even stronger evidence that pro-migration interest groups make policy more open to migration. The number of temporary work visas in a given sector, in the United States in the period 2001–2005, is positively affected by the 10

We also ﬁnd evidence of an effect driven by substitutability. The lobbying activity of organized labor leads to a reduction in the inﬂow of foreign workers in the same occupation/ education cell.

What Drives Immigration Policy?

637

lobbying expenditures for migration of ﬁrms in that sector. In particular, a 10% increase in lobbying expenditures by business groups in a given sector leads on average to a 2.3–7.4% increase in the number of work visas allocated to that sector. This is consistent with the interest groups model, given that capital and labor are complements. As mentioned in the introduction, interest groups are likely to be a very important factor explaining the gap between public opinion and government policies.

6. Conclusions In this chapter, we have used a newly available dataset on migration policy decisions reported by governments, which has been constructed by the United Nations Department of Economic and Social Affairs. We have found evidence suggesting that most governments have policies aimed at either maintaining the status quo or at lowering the level of migration. For example, between 1976 and 2007, approximately 61% of the government ofﬁcials interviewed reported policies to maintain the status quo, while 23% were trying to reduce the number of immigrants. We have then merged the UN dataset with two large individual level surveys, the ISSP and the WVS. We have found evidence suggesting that government policies are correlated with individual opinions, as is consistent with a simple median voter model. Still, individual opinions appear to be substantially more restrictionist that the actual policies implemented by governments. Thus, we have documented – for the ﬁrst time in a quantitative assessment – the existence of a public opinion gap. We have argued that the activities of pro-immigration interest groups, which heavily lobby governments in destination countries, are a primary candidate to explain the public opinion gap we have documented. Of course, alternative factors can also be at work. For example, another reason why migration ﬂows continue to take place – notwithstanding the great opposition of voters in destination countries – is that policymakers may not have full control on migration inﬂows through their policies. In other words, migration pressure on the supply side might give rise to increasing inﬂows through illegal migration. We tend to believe that this is an unplausible explanation, that is, we think that governments are not willing – rather than able – to block migration inﬂows. For instance, it is well known that most destination countries manage migration through border enforcement rather than interior enforcement, although the latter is much more effective than the former (Hanson and Spilimbergo, 2001). Thus, allowing large ﬂows of illegal immigrants like those which have been estimated for the United States between 1995 and 2005 (Passel, 2005) might well represent a government attempt to reconcile the restrictionist views of the broader public and the pro-migration requests of domestic pressure groups. Analogously, the (re)introduction of guest worker

638

Giovanni Facchini and Anna Maria Mayda

programs – like the one which was part of the failed Kennedy-McCain proposal in 2005 – suggests that increasing temporary migration might be another possible politically viable way of allowing the needed pool of talents into the country. Acknowledgments We would like to thank Cecilia Testa for useful comments on an earlier draft. This paper builds on research we carried out for the United Nations for the 2009 UN Human Development Report (see Facchini and Mayda, 2009a).

Fig. A1.

Appendix

The impact of individual attitudes toward immigrants on migration inﬂows (ISSP 1995, United Nations)

What Drives Immigration Policy? 639

640

Giovanni Facchini and Anna Maria Mayda

fitted net migration/pop, 1996

net migration/pop, 1996 .005

IRL USA

CAN

net migration, 1996 (divided by population)

ESP RUS DEU SVN NOR NLD ITA GBR HUN AUT SWE

0

NZL

JPN CZE

SVK

POL BGR

PHL LVA

-.005 Lower

Fig. A2.

1.5

Maintain policyimmREC

2.5

Raise

The impact of migration policy on migration inﬂows (United Nations 2007, United Nations)

Government’s view on immigration

Policy on immigration

Government’s view and policy on the level of immigration, by region

17.2

Total

78.28

93.75 83.7 87.5 50 82.54 50 66.67 56.25 76.92 84.07 69.57 4.52

6.25 5.43 1.79 4.17 6.35 21.43 6.67 6.25 2.56 2.75 1.09 100

100 100 100 100 100 100 100 100 100 100 100 686

16 92 56 24 126 28 15 16 39 182 92 23.03

6.25 13.04 21.43 50 14.29 32.14 20 37.5 28.21 17.58 45.65

Notes: The table presents row percentages by region. Gulf countries exclude Iran and Iraq.

0 10.87 10.71 45.83 11.11 28.57 26.67 37.5 20.51 13.19 29.35 60.5

50 77.17 69.64 41.67 70.63 46.43 66.67 50 64.1 51.65 52.17 5.1

11.37

6.25 37.5 5.43 4.35 1.79 7.14 4.17 4.17 7.14 7.94 21.43 0 6.67 6.67 6.25 6.25 5.13 2.56 3.85 26.92 1.09 1.09

100

100 100 100 100 100 100 100 100 100 100 100

686

16 92 56 24 126 28 15 16 39 182 92

Too high Satisfactory Too low Total Total # Lower Maintain Raise No intervention Total Total # observations observations

Central Asia East Asia and Paciﬁc Eastern Europe Gulf countries Latin America and Caribbean Middle East North America North Africa South Asia Sub-Saharan Africa Western Europe

Region

Table A1.

What Drives Immigration Policy? 641

Policy on temporary workers

30

Total

33.33 8.57 5.56 33.33 25 0 14.29 25 20 58.21 2.38

5.17 24.14

8.33 11.43 2.78 0 3.57 20 14.29 0 13.33 2.99 0 100

100 100 100 100 100 100 100 100 100 100 100 290

12 35 36 6 56 10 7 4 15 67 42

Notes: The table presents row percentages by region. Gulf countries exclude Iran and Iraq.

40.69

8.33 50 28.57 51.43 33.33 58.33 66.67 0 19.64 51.79 50 30 42.86 28.57 50 25 40 26.67 19.4 19.4 47.62 50 28.42

46.04

9.09 63.64 28.95 55.26 27.78 55.56 87.5 0 15.38 55.77 50 30 0 66.67 50 33.33 13.33 60 29.82 21.05 38.46 53.85 3.6

0 10.53 5.56 0 1.92 0 16.67 0 0 1.75 2.56

21.94

27.27 5.26 11.11 12.5 26.92 20 16.67 16.67 26.67 47.37 5.13

100

100 100 100 100 100 100 100 100 100 100 100

278

11 38 36 8 52 10 6 6 15 57 39

Lower Maintain Raise No intervention Total Total # Lower Maintain Raise No intervention Total Total # observations observations

Policy on permanent settlement

Governments’ policy on permanent settlement and on temporary workers, by region

Central Asia East Asia and Paciﬁc Eastern Europe Gulf countries Latin America and Carribean Middle East North America North Africa South Asia Sub-Saharan Africa Western Europe

Region

Table A2.

642 Giovanni Facchini and Anna Maria Mayda

Policy on family reuniﬁcation

58.74

Total

0 0 11.11 0 0 14.29 33.33 0 0 55.56 18.18

25.17 12.59

16.67 36.36 33.33 0 12.5 0 66.67 0 28.57 16.67 45.45 100

100 100 100 100 100 100 100 100 100 100 100 143

6 22 18 5 32 7 3 3 7 18 22

Notes: The table presents row percentages by region. Gulf countries exclude Iran and Iraq.

3.5

83.33 63.64 55.56 60 87.5 71.43 0 100 57.14 22.22 36.36 15.87

0 9.38 14.29 37.5 14.29 14.29 16.67 20 23.08 13.46 23.68 50

42.86 62.5 62.86 50 57.14 57.14 50 60 38.46 17.31 65.79

57.14 15.63 20 12.5 26.53 28.57 16.67 20 38.46 65.38 7.89 3.97 30.16

0 12.5 2.86 0 2.04 0 16.67 0 0 3.85 2.63

100

100 100 100 100 100 100 100 100 100 100 100

252

7 32 35 8 49 7 6 5 13 52 38

Lower Maintain Raise No Total Total # Lower Maintain Raise No Total Total # intervention observations intervention observations

Policy on highly skilled workers

Governments’ policy on highly skilled workers and on family reuniﬁcation, by region

Central Asia 0 East Asia and Paciﬁc 0 Eastern Europe 0 Gulf countries 40 Latin America and Carribean 0 Middle East 14.29 North America 0 North Africa 0 South Asia 14.29 Sub-Saharan Africa 5.56 Western Europe 0

Region

Table A3.

What Drives Immigration Policy? 643

118

42.55180421 13.10843051***

2007

108

0.12215755 9.48110047 0.00002572 0.00000684***

5

27

0.00002797 0.00056058

6

Notes: Standard errors under each marginal effect. The table reports marginal effects. The dependent variable policy on the integration of non-citizens indicates whether or not (1 and 0, respectively) there exists a policy to foster the integration of non-citizens. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

24

113

4

Policy on the integration of noncitizens

3

Observations

2.57638449 5.20404834 0.00001723 0.00000470***

2

0.1061808 0.05510909*

117

2.34041133 4.68707925

1996

1

The determinants of the policy on the integration of noncitizens

Relative skill mix (natives vs. immigration)

Per capita GDP (PPP-adjusted)

Net migration, divided by population

Year

Dependent variable

Probit

Table A4.

644 Giovanni Facchini and Anna Maria Mayda

118

42.55180421 13.10843051***

2007

117

27.04520477 13.74180074** 1.18374051 0.30364651***

5

27

0.00002797 0.00056058

6

Notes: Standard errors under each marginal effect. The table reports marginal effects. The dependent variable policy on the integration of non-citizens indicates whether or not (1 and 0, respectively) there exists a policy to foster the integration of non-citizens. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

24

116

4

Policy on the integration of noncitizens

3

Observations

3.50589187 4.75068949 1.08835177 0.27183735***

2

0.1061808 0.05510909*

117

2.34041133 4.68707925

1996

1

The determinants of the policy on the integration of noncitizens

Relative skill mix (natives vs. immigration)

Human development index

Net migration, divided by population

Year

Dependent variable

Probit

Table A4 (cont.).

What Drives Immigration Policy? 645

646

Table A5.

Giovanni Facchini and Anna Maria Mayda

The impact of individual attitudes towards immigrants and migration policy on migration inﬂows

OLS Dependent variable Median Immig Opinion

1

2

3

4

Net migration, 1996 (divided by population) 0.002 0.0007***

Average Immig Opinion

0.0035 0.0013**

Average Pro Immig Dummy

0.0159 0.0088*

Migration policy, 1996 Constant

0.0025 0.0015

0.006 0.0028**

Observations R-squared

22 0.26

22 0.26

0.0003 0.0008 22 0.14

0.0002 0.0009 0.0017 0.0013 22 0

Sources: 1995 ISSP National Identity Module and United Nations. Note: Standard errors under each coefﬁcient. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

References Benhabib, J. (1996), On the political economy of immigration. Economic European Review 40, 1737–1743. Bernheim, B.D., Whinston, M.D. (1986), Menu auctions, resource allocation, and economic inﬂuence. Quarterly Journal of Economics 101, 1–31. Boeri, T., Hanson, G., McCormick, B. (2002), Immigration Policy and the Welfare State. Oxford University Press, Oxford. Borjas, G.J. (1999), Heaven’s Door. Princeton University Press, Princeton, NJ. Briggs, V.M. (2001), Immigration and American Unionism. Cornell University Press, Ithaca, NY. Chiswick, B.R., Hatton, T.J. (2003), International migration and the integration of labor markets. In: Bordo, M.D., Taylor, A.M., Williamson, J.G. (Eds.), Globalization in Historical Perspective. University of Chicago Press, Chicago, pp. 65–119, Chapter 3. Chiswick, B.R., Miller, P.W. (2006), Language skills and immigrant adjustment: what immigration policy can do!. In: Cobb-Clark, D., Khoo, S.-E. (Eds.), Public Policy and Immigrant Settlement. Edward Elgar Publishing, Celthenham, pp. 121–148.

What Drives Immigration Policy?

647

Citrin, J., Green, D., Muste, C., Wong, C. (1997), Public opinion toward immigration reform: the role of economic motivation. The Journal of Politics 59, 858–881. Dustmann, C., Preston, I. (2001), Attitudes to ethnic minorities, ethnic context, and location decisions. Economic Journal 111 (470), 353–373. Dustmann, C., Preston, I. (2004), Is Immigration Good or Bad for the Economy? Analysis of Attitudinal Responses. CReAM Discussion Paper No. 06/04. London: Centre for Research and Analysis of Migration, University College London. Dustmann, C., Preston, I. (2007), Racial and economic factors in attitudes to immigration. The B.E. Journal of Economic Analysis & Policy 7 (1), (Advances), Article 62. Epstein, G.S., Nitzan, S. (2006), The struggle over migration policy. Journal of Population Economics 19, 703–723. Espenshade, T.J., Hempstead, K. (1996), Contemporary American attitudes toward U.S. Immigration. International Migration Review 30, 535–570. Facchini, G., Mayda, A.M. (2008), From individual attitudes towards migrants to migration policy outcomes: theory and evidence. Economic Policy 56, 651–713. Facchini, G., Mayda, A.M. (2009a), Does the welfare state affect individual attitudes towards immigrants? Evidence across countries. Review of Economics and Statistics 91, 291–314. Facchini, G., Mayda, A.M. (2009b), The Political Economy of Immigration Policy, Background Paper for the 2009 United Nations Human Development Report, Human Development Research Paper 2009/03. Facchini, G., Testa, C. (2009), Who is against a common market? Journal of the European Economic Association 7, 1068–1100. Facchini, G., Willmann, G. (2005), The political economy of international factor mobility. Journal of International Economics 67, 201–219. Facchini, G., Mayda, A.M., Mishra, P. (2008), Do interest groups affect US immigration policy? CEPR Working Paper No. 6898. Facchini, G., Mayda, A.M., Puglisi, R. (2009), Media exposure and illegal immigration: evidence on attitudes from the United States. Mimeo. Freeman, G. (1992), Migration policy and politics in the receiving states. International Migration Review 26, 1144–1167. Freeman, G. (1995), Modes of immigration politics in liberal democratic states. International Migration Review 29, 881–902. Freeman, R. (2006), People ﬂows in globalization. Journal of Economic Perspectives 20, 145–170. Hanson, G., Scheve, K., Slaughter, M. (2007), Public ﬁnance and individual preferences over globalization strategies. Economics and Politics 19, 1–33. Hanson, G., Spilimbergo, A. (2001), Political economy, sectoral shocks, and border enforcement. Canadian Journal of Economics 34, 612–638.

648

Giovanni Facchini and Anna Maria Mayda

Joppke, L. (1998), Why liberal states accept unwanted immigration. World Politics 50, 266–293. Kessler, A. (2001), Immigration, economic insecurity, and the ‘‘ambivalent’’ American public. CCIS Working Paper No. 41, The Center for Comparative Immigration Studies, University of California, San Diego. Mayda, A.M. (2006), Who is against immigration? A cross country investigation of individual attitudes towards immigrants. Review of Economics and Statistics 88, 510–530. Mayda, A.M. (2009), International migration: A panel data analysis of the determinants of bilateral ﬂows, May 2008. Journal of Population Economics, October 2010 23 (4), 1249–1274. Money, J. (1997), No vacancy: the political geography of immigration control in advanced industrial countries. International Organization 51, 685–720. O’Rourke, K.H., Sinnott, R. (2005), The determinants of individual attitudes towards immigration. European Journal of Political Economy 22, 838–861. Ortega, F. (2005), Immigration quotas and skill upgrading. Journal of Public Economics 89, 1841–1863. Passel, J.S. (2005), Unauthorized migrants: Numbers and characteristics. Mimeo, Pew Hispanic Center. Pritchett, L. (2006), The future of migration: accommodating irresistible forces and immovable ideas. Mimeo, JFK School of Government, Harvard University. Rodrik, D. (1995), Political economy of trade policy. In: Grossman, G., Rogoff, K. (Eds.), The Handbook of International Economics, Vol. 3. North-Holland, Amsterdam, pp. 1457–1494, Chapter 28. Scheve, K., Slaughter, M. (2001), Labor market competition and individual preferences over immigration policy. Review of Economics and Statistics 83, 133–145. United Nations, World Population Policies, United Nations, New York, various issues. Watts, J.R. (2002), Immigration Policy and the Challenge of Globalization. Cornell University Press, Ithaca and London.

CHAPTER 26

Changes in Attitudes toward Immigrants in Europe: Before and After the Fall of the Berlin Wall Ira N. Ganga,b,c, Francisco L. Rivera-Batizd and Myeong-Su Yunb,e a

Department of Economics, Rutgers University, New Brunswick, New Jersey, 08901-1248, USA Institute for the Study of Labor (IZA), Bonn, Germany c CReAM-Center for Research and Analysis of Migration, London, UK E-mail address: [email protected] d Economics and Education, Teachers College and International and Public Affairs (Afﬁliate), Macy Hall 350, Teachers College, Columbia University, 525 West 120th Street, New York, NY 10027 E-mail address: ﬂ[email protected] e Department of Economics, Tulane University, New Orleans, LA, 70118, USA E-mail address: [email protected] b

Abstract This chapter provides a statistical analysis of the determinants of attitudes toward foreigners displayed by Europeans sampled in Eurobarometer surveys in 1988 and 1997. Those who compete with immigrants in the labor market are more negative toward foreigners. An increased concentration of immigrants in neighborhoods increases the likelihood of negative attitudes. Racial prejudice exerts a strong inﬂuence on anti-foreigner sentiment. Greater racial prejudices, and the decline in the strength of educational attainment in reducing negative attitudes toward foreigners, contribute to the increased anti-foreigner attitudes between 1988 and 1997. Keywords: Anti-foreigner, labor force status, ethnic concentration, probit decomposition, racial prejudice, attitudes, sentiments toward migrants JEL classiﬁcations: J15, J61, F22

1. Introduction On November 9, 1989, the Berlin Wall fell. Rapidly the barriers separating East and West disintegrated, and the population of foreigners (from all over) in Western Europe rose. The 1990s saw the share of the population change in the European Union accounted for by net immigration exceed that of natural population growth for the ﬁrst time in many decades (OECD, 2001, Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008032

r 2010 by Emerald Group Publishing Limited. All rights reserved

650

Ira N. Gang et al.

p. 33). For some countries, migration has played a dominant role in population growth. In Germany, the natural increase of the population since the 1970s has been negative, net immigration totally accounting for population growth. The same phenomenon has emerged in Greece, Italy and Sweden, where natural population growth turned negative in the late 1990s and their populations rose only as a result of net immigration. This demographic phenomenon has led to a substantial increase in the portion of the population accounted for by foreign nationals. Table 1 shows that, in 1998, foreigners constituted 36 percent of the population in Luxembourg, 9.1 percent in Austria, 8.7 percent in Belgium, and 8.9 percent in Germany. For some countries, these ﬁgures underestimate the signiﬁcance of immigration, as they include only foreign-born individuals who are not citizens or have not been naturalized. In Sweden, for example, foreign nationals constituted 4.2 percent of the population in 1998 but the total foreign-born population, including foreign nationals as well as naturalized foreign immigrants and Swedish citizens born abroad, was 10.8 percent of the population. By comparison, the proportion of the foreign-born in the population of the United States, a country famed for its open immigration policy, was 9.8 percent in 1998.1 The immigration ﬂows in the European Union were magniﬁed by the rise of refugees and asylum-seekers in the late 1980s and throughout the 1990s. Although laws to curb refugees and asylum-seekers were passed in some countries, the ripples of the massive immigration ﬂows associated with civil war and socioeconomic strife in Africa, Eastern Europe, and Central Asia remained. In 1983, approximately 30,000 people asked for asylum in the European Community countries. This number rose quickly in the late 1980s and early 1990s and it peaked at 680,000 in 1992. After, the number of asylum seekers has declined but still remains at high levels compared to the situation in earlier decades. In 1999, a total of 390,000 asylum-seekers entered the European Community countries. In Germany, the European country receiving the largest number of refugees, the number of asylum-seekers rose from 121,000 in 1989 to a peak of 438,000 in 1992, gradually declining to 95,100 in 1999. The rise of immigration in the 1990s was associated with increased manifestations of anti-foreigner attitudes in some countries. In Great Britain, the number of racially motivated incidents reported to the police grew from 4,383 in 1988 to 7,793 in 1992 and 13,878 in 1998.2 This trend

1

Note that international migration data have not been standardized and hence, when one refers to the fraction of ‘‘foreigners’’ in the population or to the ‘‘foreign-born,’’ one may be talking about different demographic groups in different countries (see OECD, 2001, pp. 295–301). 2 The number of incidents reported to the police grossly underestimates the actual number of such incidents since most remain unreported. In 1996, the British Crime Survey estimated that 143,000 offences against ethnic minorities (transgressions considered by the victim to be racially-motivated) had been committed the year before (Channel 4, 2000).

Changes in Attitudes toward Immigrants in Europe

651

Table 1. The fraction of foreigners in the population (% of the population, by country) Country

1985

1998

Austria Belgium Denmark Finland France Germany Ireland Italy Luxembourg The Netherlands Portugal Spain Sweden United Kingdom

4.0 8.6 2.3 0.3 – 7.2 2.3 0.7 26.7 3.8 1.0 0.6 4.6 3.1

9.1 8.7 4.8 1.6 6.4a 8.9 3.0 2.1 35.6 4.2 1.8 1.8 5.6 3.8

Source(s): OECD (1998, 2001). a Data is for 1990.

exploded in the summer of 2001 when South Asian immigrants in Britain rioted in the cities of Bradford, Oldham, Leeds, and Burnley, in large part to protest growing violence and anti-immigrant attitudes. In Germany, the number of criminal offences with racist/xenophobic motives was 10,037 in 1999, of which there were 746 racially motivated acts of violence reported to the police. A number of these attacks resulted in death, as in the case of an Algerian man who died on February 13, 1999 as a result of injuries he suffered as he was ﬂeeing from his attackers, and a man from Mozambique who died in Bavaria as a result of injuries received in an attack on August 15, 1999. In France, the killing of a 17-year-old African immigrant in Marseilles in February 1995 led to a wide debate over the foreign-born population in the country, a controversy that spilled-over into the French presidential campaign at the time. Anti-foreigner violence has also been on the rise in other European Union countries (see EUMC, 2000). Attitudes toward foreigners often depend on where the foreigners come from. Dustmann and Preston (2007) show that in the United Kingdom, attitudes toward foreigners from other European countries are more favorable than those toward Asians or West Indians. Table 2 shows the decomposition of the population of foreigners in some European Union countries in 1998 on the basis of country of origin. The proportion of non-European Economic Community (EEC) countries in the contingent of foreigners vary from 37 percent in Belgium to 86 percent in Italy. In the United Kingdom, 60 percent of the foreign-born population comes from non-EU countries and 14 percent from Asian countries. What explains the rise in negative sentiments toward non-European immigrants among some segments of the European population after the

652

Table 2.

Ira N. Gang et al.

Stocks of foreign population in selected European countries, by nationality, 1998 Recipient Country Belgium Francea Germany Italy

Total foreigner population (thousands)

The Netherlands

United Kingdomb

892

3,607

7,320

1,250

662

2,208

Foreigners from EEC countries Italy Spain Portugal Greece France The Netherlands Germany United Kingdomc Belgium Ireland

63%

36%

25%

14%

29%

40%

23% 5% 3% 2% 12% 9% 4% 3%

7% 6% 18%

8% 2% 2% 5% 1% 2%

3% 3% 1% 1%

4% 2% 2%

Foreigners from non-EEC countries Morocco Algeria Turkey Tunisia Poland Yugoslavia USA Caribbean/Guyana Asia SS Africad

37%

64%

14% 1% 8% 1% 1% 1% 1%

16% 17% 6% 6% 1% 1%

2%

2% 3% 2%

3% 8% 6% 4%

4%

20%

3% 1%

1%

75%

86%

71%

12%

19%

29% 4% 15% 1%

60%

15% 4% 2% 3% 4% 11% 3%

3% 2%

6% 3% 14% 6%

Source: OECD (2001). Note: Individual country data are presented only for those foreign groups with the largest populations in the host country. a 1990. b 1999. c Includes Hong Kong. d Selected sub-Saharan African countries in France and Italy (Senegal) and Belgium (Congo).

fall of the Berlin Wall? Is economic strain in host countries, in the form of stagnant earnings and rising unemployment, the key to understanding anti-immigrant activities or are non-economic factors, such as prejudice and racism, more inﬂuential in determining such behavior? This chapter explores the determinants of the attitudes of European citizens toward non-European Union foreigners using samples from the 1988 and 1997 Eurobarometer surveys. The Eurobarometer survey is carried out every

Changes in Attitudes toward Immigrants in Europe

653

year and samples European attitudes toward a wide array of subjects. Both in 1988 and in 1997, the surveys included speciﬁc questions measuring attitudes toward immigrants and immigration. We utilize the answers to these questions to carry out an analysis of some of the key factors inﬂuencing the attitudes of European Union citizens toward foreigners and their changes over time. In the next section we discuss the various forces that have been presented in the literature as possible factors generating anti-immigrant sentiments in host nations. Section 3 offers some background on the Eurobarometer survey data sets utilized in this chapter, and presents mean characteristics of the sampled populations. Section 4 examines the determinants of attitudes toward foreigners by utilizing a probit analysis of the relative inﬂuence of various economic and non-economic variables on such attitudes. Section 5 employs a probit decomposition analysis to examine and explain the changes in attitudes between 1988 and 1997. Finally, Section 6 provides a summary of our results. 2. The determinants of anti-immigrant attitudes Over the years, conjectures on the determinants of anti-immigrant sentiments have been based more on heavy theorizing and casual evidence (see, for instance, Alber, 1994). Recently, however, a number of empirical studies have emerged utilizing comprehensive survey data. The most popular explanation for the emergence of negative sentiments toward immigrants is ethnic or racial prejudice, whose strength is often related to the presence and concentration of immigrants within particular communities. In the United States, historically, there is substantial evidence that racial prejudice was a major factor behind restrictionist movements that reduced immigration ﬂows from particular countries or regions, such as China and Mexico (see, for example, Gutierrez, 1995).3 Racial prejudice has also been found in many of the anti-immigrant activities documented in the past few years in European countries. This was evident in the 2001 British immigrant riots. In the city of Oldham, England, where immigrant unrest occurred in May 2001, immigrants complained of racial prejudice as a source of their frustration. As the New York Times (Lyall, 2001, p. A4) reported: ‘‘[There is] a sense among 3

The extensive literature on racial segregation in the United States also serves as a background for this perspective. This literature suggests that the rise of racial segregation and housing discrimination in certain cities of the United States was linked to the increased visibility and growth in the concentration of black Americans in those cities. As Massey and Denton (1993, p. 10) point out: ‘‘the black ghetto was constructed through a series of welldeﬁned institutional practices, private behaviors, and public policies by which whites sought to contain growing urban black populations.’’ Growing immigrant communities may have faced similar behavior.

654

Ira N. Gang et al.

nonwhites that Britain, and Oldham, are racist places. Leaving aside obviously provocative language, many nonwhites say that low-grade racism is an underlying fact of life here.’’ In fact, the growing visibility of the immigrants appears to have intensiﬁed anti-immigrant sentiments. In Bradford, England, where immigrant riots erupted in July 2001, the city of half a million residents includes a visible population of about 100,000 Asian immigrants. This population is itself highly segregated within Bradford, further contributing to tensions between the immigrants and the rest of the population. A second force which is frequently postulated as an explanation for anti-immigrant attitudes is economic in nature. It is hypothesized that in countries where economic strain is present, with stagnant or collapsing income and/or employment opportunities, immigrants will be partly blamed for the economic stress thus generating the resentment of the native-born population. Whether immigration does in fact act to lower wages or reduce unemployment opportunities is a matter of debate. For instance, evidence on the impact of immigrants on European labor markets is inconclusive, often ﬁnding small effects of immigration on employment.4 Studies on the United States also ﬁnd small wage and unemployment effects of immigration.5 Indeed, economic theory warns us against hastily assuming that a ﬂow of immigrants into an economy will raise the unemployment of natives or reduce their wages. If native-born are complements to immigrants, the foreign labor inﬂow increases the demand for natives, thus raising – rather than lowering– their employment.6 However, this debate on the economic effects of immigration may not be directly relevant to the formation of attitudes of natives toward the immigrants, which are based on perceptions about how immigrants affect the economy, perceptions that are not necessarily based on reality.7 For example, even if the measured employment or wage effects of immigrants are very small, people may be inﬂuenced by rumors and stories reported in the media or heard in the streets about the ‘‘immigrant invasion’’ which is taking jobs away from them. Those who are directly competing with immigrants for jobs and who may be seeking a factor to blame for job losses, may be more responsive to these rumors and biased stories, developing strong negative attitudes toward foreigners – particularly if the press and politicians make the topic a big issue. 4

See the survey paper by Zimmermann (1995), as well as Hunt (1992), DeNew and Zimmermann (1994), Muhleisen and Zimmermann (1994), Winkelmann and Zimmermann (1993), and Winter-Ebmer and Zweimueller (1999). 5 See Card (1990) for empirical evidence on this issue relating to the Mariel immigrant ﬂow. See also Card (2001) for further evidence. 6 For a discussion of the issue of complementarity between immigrant and native-born workers, see Gang and Rivera-Batiz (1994a). 7 The question of perception versus reality in viewing migrants is taken up by Fertig and Schmidt (2002).

Changes in Attitudes toward Immigrants in Europe

655

The relative roles of the various factors inﬂuencing attitudes toward foreigners have been examined. Krueger and Pischke (1997) provide a statistical analysis of the various forces inﬂuencing crimes against foreigners in Germany. They ﬁnd signiﬁcant variation in the incidence and pattern of violence against foreigners on the basis of location. However, they also conclude that ‘‘economic strain,’’ as measured by high unemployment or low wages, seem to contribute little to the incidence of violence once location is taken into account. Although Krueger and Pischke (1997) study the causes of crime against foreigners in Germany, Gang and Rivera-Batiz (1994b), Dustmann and Preston (2001, 2007), Bauer et al. (2000), and Mayda (2006) study attitudes toward foreigners. One ﬁnding is that high concentrations of ethnic minorities are associated with more hostile attitudes toward immigrants in Germany (Gang and Rivera-Batiz, 1994b) and in the United Kingdom (Dustmann and Preston, 2001). On the other hand, Dustmann and Preston (2007) ﬁnd evidence that both welfare and labor market concerns also matter for the formation of attitudes toward future immigrants, but that the most important factor is non-economic: racial bias. Bauer et al. (2000) study the effect of different immigration policies in OECD countries on attitudes toward immigrants. Scheve and Slaughter (2001) ﬁnd that an individual’s attitude toward immigration policy is related to labor market skills; that is, less skilled workers are restrictionist so as to avoid labor market competition. In this chapter we study the determinants of attitudes toward foreigners among European Union citizens, and the changes in these attitudes between 1988 and 1997. We examine the relative role played by economic strain, racial prejudice and ethnic concentration in determining negative attitudes toward immigrants.

3. The eurobarometer survey and the empirical model The analysis in this chapter uses the 1988 and 1997 Eurobarometer surveys. The Eurobarometer surveys are carried out every year in European countries in order to examine attitudes toward a variety of issues. The surveys give rise to unique datasets consisting of cross-sections of a geographically distributed random sample of households across Europe (see Reif and Melich, 1992; Melich, 1999 for detailed descriptions of the procedures followed in each country). In addition to information on household economic and demographic behavior, the 1988 and 1997 Eurobarometer surveys contained detailed questions on attitudes toward immigrants and foreigners. We ﬁrst measure attitudes toward foreigners on the basis of the responses of residents of European Union countries, as sampled by the 1988 Eurobarometer survey, to the question: ‘‘Is the presence of (non-European Community) foreigners good or bad for the

656

Ira N. Gang et al.

future of our country?’’ We construct the variable ANTI-FOREIGNERS, equal to one if the respondent declares that foreigners are ‘‘bad’’ or ‘‘a little bad’’ for the future of their country, and zero otherwise. This qualitative variable will be used later as the dependent variable in a probit analysis establishing the impact of various explanatory variables on the probability of a person displaying negative attitudes toward foreigners. The variable ANTI-FOREIGNERS captures in a very straightforward and obvious manner a dislike of foreigners. Unfortunately, the question was not asked of survey respondents in the 1997 Eurobarometer survey, and hence its use does not allow us to make inter-temporal comparisons. We construct another dependent variable for use in our analysis, based on responses to another question which appears in both the 1988 and 1997 Eurobarometer surveys. This question asked respondents whether there were ‘‘too many,’’ ‘‘a lot, but not too many,’’ or ‘‘not many’’ foreigners living in their country. The binary variable TOO-MANY-FOREIGNERS is equal to one if the respondent answered that there were too many foreigners in their country and zero otherwise. We utilize TOO-MANYFOREIGNERS as our main dependent variable. In addition, at several places in the chapter we will make comparisons of our results for 1988 using the ANTI-FOREIGNERS and TOO-MANY-FOREIGNERS variables. We will show that these two variables reﬂect similar behavioral patterns. Table 3 presents the attitudes of European residents toward foreigners, on the basis of the two variables, ANTI-FOREIGNERS and TOO-MANY-FOREIGNERS. The ﬁrst row shows the results for the overall Eurobarometer samples in 1988 and 1997. As can be seen, in 1988, 31.4 percent of all respondents answered that foreigners were ‘‘bad’’ or ‘‘a little bad’’ for the future of their country. In addition, in that same year, 29.5 percent indicated that there were ‘‘too many foreigners’’ in their country. By 1997, the percentage of people answering that there were too many foreigners in their country had risen to 42.1 percent. This substantial increase in anti-foreigner attitudes is consistent with the observations noted in the introduction, indicating greater anti-foreigner activity in some European countries. What explains negative attitudes toward foreigners and why has there been such a marked increase in anti-foreigner sentiment among native European residents? The previous section points out that one answer may lie in the economic strain encountered by people. In a large part of Europe, a slowdown in growth and rising unemployment characterized the 1980s and 1990s. In Germany, for instance, unemployment rose from 2.9 percent in 1979 to 6.3 percent in 1988 and 9.9 percent in 1997. In France, the unemployment rate jumped from 6.1 percent in 1979 to 10.3 percent in 1988 and 12.4 percent in 1997. And in Italy, unemployment rose from 4.9 percent in 1979 to 7.9 percent in 1988 and 12.3 percent in 1997. The rising unemployment could lead to anti-foreigner attitudes as those

657

Changes in Attitudes toward Immigrants in Europe

Table 3.

Attitudes of European residents toward foreigners 1988

Overall Employed-wage/salary jobs Unemployed Retired Self-employed Students Nonretirees out of the labor force Many foreigners in the neighborhood Few foreigners in the neighborhood No foreigners in the neighborhood Disturbed by the existence of other race Not disturbed by the existence of other race

1997

Antiforeigners (%)

Too many Sample foreigners size (%)

Too many Sample foreigners size (%) (%)

31.40 31.84 31.37 38.05 32.29 26.15 28.95 41.98

29.50 29.86 33.50 39.13 27.69 19.82 28.14 46.20

9,775 3,794 612 1,201 1,087 1,105 1,976 892

42.10 40.36 42.30 52.25 44.48 26.71 46.51

11,868 5,154 896 1,776 1,223 1,243 1,576

31.48

28.13

4,209

29.47 65.24

27.80 63.18

4,737 1,165

79.85

1,841

26.82

24.95

8,610

35.17

10,027

Source(s): Eurobarometer survey, 1988, 1997; authors’ calculations. Notes: ANTI-FOREIGNERS has a value of one when the respondent says that the presence of foreigners are bad for the future of the country. TOO-MANY-FOREIGNERS has a value of one when the respondent feels that there are too many foreigners in the country.

who are ﬁred and laid-off seek a target to blame their ills on. Has that been the case in Europe? On the top half of Table 3 we decompose the 1988 and 1997 Eurobarometer samples into various labor force categories which show that anti-foreigner attitudes vary by labor force group and by year. For instance, in 1988, almost one third, or 31.37 percent, of the people who were unemployed declared that foreigners were ‘‘bad’’ or ‘‘a little bad’’ for their economy, but only 26.15 percent of students declared the same. In 1997, 42.30 percent of the people who were unemployed declared that there were ‘‘too many’’ foreigners, but only 26.71 percent of students declared the same that year. The retired have the highest rates of antiforeigner behavior in the sample. In 1988, 38.05 percent of the retired declared that there were too many foreigners in their country; in 1997, 52.25 percent of the retired declared the same. Note that negative attitudes toward foreigners rise between 1988 and 1997 for every group in the sample, especially for the nonretirees out of the labor force. Often unemployment is used in economic studies to capture economic strain. But Table 3 shows that the unemployed do not demonstrate a markedly more negative view toward foreigners than the employed or

658

Ira N. Gang et al.

retirees. The fact is that it is not just the unemployed but various groups in the economy who face, or perceive that they face, economic strain brought on by foreigners, as discussed in Section 2. The employed may be threatened by the possibility of lower wages resulting from immigration; retirees may be encouraged to retire by the presence of younger immigrants ready to take their place. Another major explanation for anti-foreigner attitudes is the greater racial or ethnic bias that may be generated by the visible presence of immigrants in certain communities. The rise in European immigration was previously noted in the introduction. Is this factor behind the rise in antiforeigner attitudes? To test this hypothesis, we use two different questions as explanatory variables. First, in 1988, the survey asked respondents to characterize the relative size of the foreign-born population in their neighborhood, with answers including: ‘‘many foreigners,’’ ‘‘few foreigners,’’ and ‘‘no foreigners.’’ On this basis, we construct two dummy variables, MANY and FEW, equal to one if a person answered that there were ‘‘many,’’ or ‘‘few’’ foreigners in their neighborhood, respectively, and zero otherwise. Although this measure is subjective, this is ﬁne since we are dealing with subjective attitude. We do not need an objective cut-off line for ‘‘too many.’’ We utilize the MANY/FEW variables in our probit analysis later to examine the role played by larger concentrations or greater visibility of foreigners on European attitudes toward them. If we ﬁnd that greater visibility breeds anti-foreigner attitudes, we presume that an element of racial/ethnic prejudice is involved in those sentiments. The simple correlation between increased concentrations of immigrants and negative attitudes toward foreigners is clearly shown among those sampled by the Eurobarometer survey. Of those who thought that their neighborhood had ‘‘many’’ foreigners, 42 percent answered that foreigners ‘‘were not good for the future of the country’’ and 46 percent thought that there were ‘‘too many’’ foreigners in their country. This is substantially higher than the comparable ﬁgures for those who thought that their neighborhood had ‘‘few’’ foreigners: for this group, 31 percent thought foreigners ‘‘were not good for the future of the country,’’ and 28 percent thought that there were ‘‘too many’’ foreigners in their country. For those who responded that their neighborhood had no foreigners, the corresponding percentages are 29 and 28 percent, respectively. Of course, our analysis below will examine whether this simple correlation holds in a multivariate analysis of anti-foreigner attitudes. A second and different question concerning racial bias was asked both in the 1988 and 1997 Eurobarometer surveys, inquiring whether respondents felt that the presence of people from another race was disturbing or not. We construct a dummy variable DISTURBING equal to one if the person answered yes to this question and zero otherwise. This variable allows us to examine the role racial bias plays in the manifestation of attitudes toward foreigners (see Dustmann and Preston, 2007).

Changes in Attitudes toward Immigrants in Europe

659

Racial bias and anti-foreigner attitudes are, though overlapping, two different negative potential biases native-born Europeans may have against different nationalities. Table 3 shows that 65 percent of those disturbed by the existence of other races also held anti-foreigner attitudes in 1988 (that the presence of foreigners was bad for the future of the country), and 63 percent thought that there were too many foreigners in their country. With respect to the latter, by 1997 the percentage of people who felt that there were too many foreigners in their country climbs to 80 percent among those disturbed by the existence of other races. However, Table 3 also shows that people who were not disturbed by the presence of other races also displayed sharply lower anti-foreigner attitudes, although even among these groups antiforeigner attitudes also rose between 1988 and 1997. Taken together with the MANY/FEW variables we can assess racial and ethnic considerations from native-born anti-foreigner perspectives. Although interesting, the simple correlations obtained from Table 3 can only be suggestive. In order to systematically identify effects of various economic and ethnic/racial factors on the formation of anti-foreigner attitude, we carry out a probit analysis of the two attitude variables deﬁned earlier: ANTI-FOREIGNERS and TOO-MANY-FOREIGNERS. In the probit analysis, the probability of observing negative attitude toward foreigners is deﬁned as Prob(TOO-MANY-FOREIGNERS ¼ 1) ¼ F(Xb), where F is a standard normal cumulative distribution function, b a set of estimated coefﬁcients and X includes various explanatory variables to be speciﬁed below. We carry out a similar analysis using ANTI-FOREIGNER as the dependent variable in the probit equation. In addition to explanatory variables capturing labor force status, racial/ethnic bias, and the relative visibility of foreigners in the neighborhood, we include in our analysis a number of background and demographic variables. First, we include the generational impact reﬂected by the age of the person. We use two variables: AGE (number of years) and AGESQUARE (number of years of age2 divided by 100), to reﬂect decreasing or increasing effects of age on attitudes. In addition, we deﬁne the variable EDUCATION (years of schooling), which we expect to be inversely associated with negative attitudes toward immigrants, partly because most educational systems willfully act to reduce prejudice and bias, and partly because more educated European residents are less likely to be negatively affected by the less-skilled foreigners; in fact, they may have beneﬁted from low-wage foreigners. We also examine differences in attitudes based on gender, including a dummy variable MALE, which is equal to one if the person was male and zero otherwise. The dummy variable HEAD-of-HOUSEHOLD is equal to one if the person is the head of household and zero otherwise. We expect HEAD-ofHOUSEHOLD to be associated with more negative attitudes toward

660

Ira N. Gang et al.

foreigners because any perceived negative economic effects of the immigrants will be magniﬁed for heads of household, who are in charge of the economic affairs of their families. We also include the variable CHILDREN15, which equals the number of children less than 15 years living in the household where the respondent resides. Since households with a greater number of children may be subject to deeper economic strains, we anticipate CHILDREN15 to be associated with anti-immigrant attitudes. As users of public services, immigrants proﬁt greatly from public education. Being comparatively young, and with family sizes that exceed the average, immigrant families tend to have on average more children in public schools than the average. The impact of this on the budget of the public sector has not gone unnoticed. Both in Europe and in the United States, a controversy has raged in recent years about the impact of immigrants on social spending, including public education spending.8 4. Results This study focuses on the attitudes of the citizens of European Union countries aged 16–70 years, not of foreign origin and not in the military. Respondents who did not answer questions as to their nationality, occupation, age, or gender were removed from the sample. The remaining sample of citizens of the European Union was equal to 9,775 in 1988 and 11,868 in 1997.9 Table 4 shows the mean characteristics of the samples. The average age is approximately 40–41 years for both 1988 and 1997. The educational attainment of the sample rises from 11.2–11.8 years. About half of the two samples (49 percent) consist of men; close to one-half are household heads; and the average number of children less than 15 years of age per household is 0.65 in 1988 and 0.59 in 1997. In 1988, 39 percent of the sample was employed in wage/salary jobs, 6 percent were unemployed, 12 percent were retired, 11 were percent self-employed, 11 percent were students, and 20 percent were nonretirees out of the labor force. By 1997, employment increased to 43 percent, unemployment to 8 percent, retirees to 15 percent, whereas the self-employed, students, and nonretirees out of the labor force fell to 10 percent, 10 percent, and 13 percent, respectively. 8

This concern has been especially sharp in relation to the children of illegal immigrants. Indeed, legislation has been debated about whether the children of illegal immigrants can, or should, be excluded from access to public education. Given the publicity accorded to these issues, parents of young children, concerned with the impact of foreigners on social spending, may have more negative attitudes toward foreigners. 9 We restrict the sample to respondents in countries that were members of the European Union in 1988: Belgium, Denmark, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Portugal, Spain, and the United Kingdom.

661

Changes in Attitudes toward Immigrants in Europe

Table 4.

Sample means

Variable

1988

Age Years of education Children15 (number of children less than 15) Head of household Male

40.01 11.23 0.65 0.49 0.49

(15.67) (3.04) (1.02) (0.50) (0.50)

40.85 11.81 0.59 0.53 0.49

(15.32) (4.39) (0.95) (0.50) (0.50)

0.39 0.06 0.12 0.11 0.11 0.20 0.12

(0.49) (0.24) (0.33) (0.31) (0.32) (0.40) (0.32)

0.43 0.08 0.15 0.10 0.10 0.13 0.16

(0.50) (0.26) (0.36) (0.30) (0.31) (0.34) (0.36)

0.08 0.43 0.48 0.30

(0.28) (0.50) (0.50) (0.46)

0.42

(0.49)

0.31

(0.46)

Labor force status Employed: wage/salary jobs Unemployed Retired Self-employed Students Nonretirees out of the labor force Disturbing (disturbed by the existence of other race) Foreigners in the neighborhood Many foreigners Few foreigners No foreigners Too many foreigners (feel that there are too many foreigners in their country) Anti-foreigners (the presence of foreigners is ‘‘Bad’’ or a ‘‘Little Bad’’ for the future of the country) Number of observations

1997

9,775

11,868

Source(s): Eurobarometer Survey, 1988, 1997; authors’ calculations. Note: Standard deviations are reported in parentheses.

The visibility of foreigners in a neighborhood is measured by the dummy variables MANY and FEW. As noted earlier, these variables are based on responses to the 1988 Eurobarometer survey question as to whether there were ‘‘many,’’ ‘‘few,’’ or ‘‘no’’ foreigners residing in the neighborhood of the respondents. Almost half of the sample (48 percent) declared that there were no foreigners residing in their neighborhood, whereas 43 percent stated there were a few foreigners in their neighborhood, and 8 percent said that there were many foreigners. Finally, the variable DISTURBING measures the percentage of the sample stating that they found the presence of people of other races disturbing. The proportion of the sample saying yes to this question was 12 percent in 1988 and 16 percent in 1997. Table 4 also shows the mean values of the dependent variables in the probit analysis. In terms of the equation for the dependent variable ANTI-FOREIGNERS, which is available only for 1988, about 31 percent of the sample had a value equal to one, that is, 31 percent declared that the presence of immigrants is ‘‘bad’’ or a ‘‘little bad’’ for the future of their

662

Ira N. Gang et al.

country. For the dependent variable TOO-MANY-FOREIGNERS, which is available both in the 1988 and 1997 samples, 30 percent of the sample in 1988, and 42 percent of the 1997 sample had a value of one, that is, answered that there were too many foreigners in their country. The estimated probit coefﬁcients are reported in Tables 5 thru 7. In Tables 5 and 6 we present the results using TOO-MANY-FOREIGNERS as the dependent variable. Table 7 repeats the analysis presented in Table 6, this time with ANTI-FOREIGNERS as the dependent variable. The main sets of probit results are displayed in Table 5. First, the estimated coefﬁcients show that greater educational attainment is associated with a statistically signiﬁcant reduction in the probability of displaying negative attitudes toward foreigners, with everything else held constant. The negative coefﬁcient on CHILDREN15 suggests that people residing in households that have a larger number of children under the age of 15 have more positive attitudes toward foreigners. This is contrary to the expectations we had on the sign of this coefﬁcient. This may be Table 5.

Probit analysis of attitudes toward foreigners-dependent variable: too many foreigners

Variable

1988

1988

1997

With neighborhood variables

No neighborhood variables

No neighborhood variables

Estimate **

(S.E.)

Estimate **

(S.E.)

Estimate **

(S.E.)

Constant Age Age2/100 Years of education Children15 Head of household Male Self-employed Employed Unemployed Retired Nonretirees out of the labor force Disturbing Many foreigners Few foreigners

0.650 0.012 0.011 0.039** 0.059** 0.013 0.000 0.039 0.160** 0.253** 0.244** 0.041

(0.130) (0.007) (0.008) (0.005) (0.015) (0.038) (0.036) (0.073) (0.060) (0.076) (0.080) (0.073)

0.622 0.012 0.010 0.035** 0.064** 0.003 0.015 0.025 0.155* 0.257** 0.230** 0.029

(0.130) (0.007) (0.008) (0.005) (0.015) (0.038) (0.036) (0.073) (0.060) (0.075) (0.080) (0.072)

0.410 0.001 0.003 0.022** 0.050** 0.034 0.070* 0.337** 0.283** 0.274** 0.371** 0.301**

(0.105) (0.006) (0.007) (0.003) (0.014) (0.030) (0.028) (0.066) (0.054) (0.066) (0.069) (0.065)

0.960** 0.445** 0.011

(0.041) (0.051) (0.030)

0.990**

(0.041)

1.183**

(0.036)

Log-likelihood

5,479.998

5,521.064

7,294.283

Source(s): Eurobarometer Survey, 1988, 1997; authors’ calculations. Notes: Dependent variable is a binary variable, Too Many Foreigners, and ** and * mean statistically signiﬁcant at 1% and 5% levels, respectively. The reference group of occupation is Student.

663

Changes in Attitudes toward Immigrants in Europe

Table 6.

Probit analysis of attitudes toward foreigners dependent variable: too many foreigners

Variable

1988

1988

1997

With neighborhood variables

No neighborhood variables

No neighborhood variables

Estimate

(S.E.)

Estimate

(S.E.)

Estimate

(S.E.)

Constant Age Age2/100 Years of education Children15 Head of household Male Labor market competitors Disturbing Many foreigners Few foreigners

0.591** 0.010 0.008 0.040** 0.057** 0.012 0.004 0.157**

(0.126) (0.006) (0.007) (0.005) (0.015) (0.037) (0.034) (0.030)

0.567** 0.009 0.007 0.036** 0.062** 0.002 0.018 0.161**

(0.125) (0.006) (0.007) (0.005) (0.015) (0.037) (0.034) (0.030)

0.449** 0.014** 0.008 0.025** 0.043** 0.024 0.077** 0.054*

(0.100) (0.005) (0.006) (0.003) (0.014) (0.030) (0.028) (0.027)

0.961** 0.444** 0.009

(0.041) (0.051) (0.029)

0.992**

(0.041)

1.185**

(0.036)

Log-likelihood

5,482.095

5,523.149

7,309.620

Source(s): Eurobarometer Survey, 1988, 1997; authors ¼ calculations. Notes: Dependent variable is a binary variable, Too Many Foreigners, and ** and * mean statistically signiﬁcant at 1% and 5% levels, respectively. The variable ‘‘attached to the formal labor force’’ has a value of one if employed in wage/salary jobs or unemployed or retired.

explained by the increased likelihood that the children of natives will mix with the children of immigrants, increasing their contact and diffusing the tensions between the adults in the two groups. The probit results in Table 5 show the partial correlation between attitudes toward foreigners and various, disaggregated labor market groups, including the employed, unemployed, retired, self-employed, and nonretirees out of the labor force (note that students are the reference group). The statistical signiﬁcance of the coefﬁcients on the dummy variables varies substantially across these groups and across years. The main focus of the existing literature is ‘‘Do the unemployed have stronger negative attitudes toward foreigners than other groups?’’ A null hypothesis, b(employed) ¼ b(unemployed), was tested using a likelihood ratio test (see Amemiya, 1985, section 4.5) in order to determine whether or not unemployment increases the likelihood of forming more negative attitudes toward foreigners relative to the employed. This null hypothesis cannot be rejected at the 5 percent level in 1988 or 1997. This means that the likelihood of having negative attitudes toward foreigners is not different between the employed and unemployed in either year.

664

Table 7.

Ira N. Gang et al.

Probit analysis of anti-foreign attitudes dependent variable: anti-foreigners

Variable

1988

1988

With neighborhood variables Without neighborhood variables Estimate

(S.E.)

Estimate

(S.E.)

Constant Age Age2/100 Years of education Children15 Head of household Male Labor market competitors Disturbing Many foreigners Few foreigners

0.855** 0.009 0.006 0.004 0.041** 0.005 0.007 0.071* 0.978** 0.220** 0.029

(0.122) (0.006) (0.007) (0.005) (0.015) (0.036) (0.033) (0.030) (0.041) (0.050) (0.029)

0.839** 0.008 0.006 0.002 0.044** 0.013 0.000 0.074* 0.996**

(0.122) (0.006) (0.007) (0.005) (0.015) (0.036) (0.033) (0.030) (0.041)

Log-likelihood

5,727.636

5,737.157

Source(s): Eurobarometer Survey, 1988, 1997; authors’ calculations. Notes: Dependent variable is a binary variable, Anti-Foreigners, and ** and * mean statistically signiﬁcant at 1% and 5% levels, respectively. The variable ‘‘attached to the formal labor force’’ has a value of one if employed in wage/salary jobs or unemployed or retired.

Does this result imply that economic strain is unrelated to negative attitudes toward foreigners? It may imply that economic stress in the form of unemployment may not be the only cause of negative attitudes toward foreigners. Both the employed and unemployed may perceive economic strain caused by immigrants, albeit in different ways. The important commonality is that their well-being is affected by changing wages and by employment rates, both of which popular lore feels are negatively affected by immigrants. Another ﬁnding common to both years is that being retired is also associated with negative attitudes toward foreigners. The retired were workers in the past, subject to the vagaries of wages and employment rates, and attitudes toward foreigners formed in earlier years do not necessarily disappear over time, after retirement. From the results, one may conjecture that the people in these three categories (the employed, the unemployed, and the retired) may have similar attitudes toward foreigners, as compared to people in other categories, holding other things constant. One might expect that individuals, who are in these three categories, would perceive (currently or in the past) that their earnings and employment opportunities could be negatively affected by immigrants, and thus would have stronger anti-immigrant attitudes. In order to formally test this hypothesis, we substituted the various labor force status categories shown in Table 5 with

Changes in Attitudes toward Immigrants in Europe

665

the variable LABOR MARKET COMPETITORS, which is equal to one if the person is employed, unemployed or retired, and zero otherwise. Table 6 presents our probit analysis for negative attitudes toward foreigners, using the variable LABOR MARKET COMPETITORS as an explanatory variable instead of the disaggregated categories that appear in Table 5. Note that labor market competitors made up 57 percent of the sample in 1988 and 66 percent in 1997. For both 1988 and 1997, we ﬁnd that being a labor market competitor (currently or in the past) has a statistically signiﬁcant positive impact (at least at a 5 percent signiﬁcance level) on the likelihood that a person has negative attitudes toward foreigners, holding other things constant. This suggests that economic strain magniﬁes negative attitudes toward foreigners. Some comments are warranted here. First, it is well documented that immigrants often choose to be self-employed (e.g., Yuengert, 1995). Also, since the self-employed are labor market participants, it could be argued that they should be considered in the category LABOR MARKET COMPETITORS. Being self-employed in 1988, however, does not signiﬁcantly increase one’s negative attitudes toward foreigners. It is possible that the self-employed did not face intensive competition with immigrants in 1988. Second, it is interesting to note the changes in the signiﬁcance of the coefﬁcients of the different labor force categories between 1988 and 1997. Indeed, every category signiﬁcantly demonstrated increased negative attitudes toward foreigners relative to the students in 1997. By 1997, it appears that only the students are different from other labor force categories. This may be because students perceive the least economic strain caused by the increasing immigration since their current interests are the most remotely related to the labor market. Their future competition with foreigners when they enter the labor market may not resonate with them.10 We now turn our attention to racial/ethnic bias and its role in fomenting anti-foreigners attitude. The results on this issue reported in Tables 5 and 6 are consistent and robust. These tables include two measures that are linked to racial/ethnic prejudice. First, the results for 1988 presented in column 1 of Table 5 show that a greater concentration of foreigners in a particular location increases negative attitudes toward foreigners. Recall that the variable MANY is equal to one when individuals declared that their neighborhood had many foreigners, and zero otherwise. The estimated coefﬁcient on this variable is positive and statistically signiﬁcant at 1 percent signiﬁcance level. Similarly, the variable FEW, which is equal 10

A series of joint hypothesis tests are performed, though not reported, using the likelihood ratio test in order to ﬁnd a fault line between distinctive groups which show different attitudes between them in 1988 and 1997. The tests conﬁrm what we describe in the text; the distinction can be found between groups we called labor market competitors and others in 1988 and between students and other categories in 1997.

666

Ira N. Gang et al.

to one when people stated that their neighborhood had a few foreigners, displays a positive coefﬁcient, but it is not statistically signiﬁcant. In fact, additional analysis indicates that the hypothesis b(Many Foreigners) ¼ b(Few Foreigners) can be rejected (the likelihood ratio test statistic was 73.437 with one degree of freedom). There is deﬁnitely a stronger anti-foreigner sentiment among people whose neighborhoods have many foreigners when compared to those who have few or no foreigners in their neighborhood. Tables 5 and 6 show the estimated coefﬁcients of the variable DISTURBING, which is equal to one if the person ﬁnds people of a different race disturbing and zero otherwise. The estimated coefﬁcient on this variable is positive and statistically signiﬁcant (at the 1 percent signiﬁcance level) in both 1988 and 1997. This indicates that racial prejudice is associated with stronger negative sentiments toward foreigners. Tables 5 and 6 present our results using the variable ‘‘too many foreigners in our country’’ as the dependent variable. Table 7 carries out a probit analysis adopting the dependent variable ANTI-FOREIGNER, which is equal to one if the respondent declares that foreigners are ‘‘bad’’ or ‘‘a little bad’’ for the future of their country, and zero otherwise. This dependent variable is only available in 1988 and it does not allow an analysis of changes in attitudes over time. However, we report our results for 1988 because it represents a dependent variable which more directly reﬂects anti-foreign sentiment. The results, though, are remarkably consistent with those presented in Table 6 using the alternative measure of anti-foreign attitudes. For instance, the estimated coefﬁcient on the variable LABOR MARKET COMPETITORS is again positive and statistically signiﬁcant in determining anti-foreigner attitudes. This means that the economic strain felt by people who compete or have competed with foreigners in the labor market strengthen negative attitudes toward foreigners. Also, the neighborhood variable, reﬂecting the existence of ‘‘many foreigners’’ in the neighborhood, is also positively associated to negative attitudes toward foreigners. Finally, the coefﬁcient on the variable DISTURBING, reﬂecting racial bias, is also positive and statistically signiﬁcant, showing that those who ﬁnd the presence of another race disturbing also tend to feel foreigners are bad for the country. The results of probit models and hypothesis tests imply that people who are – or were – competitors of immigrants in the labor market (those who are employed at wage/salary jobs, unemployed or retired) have more strongly negative attitudes toward immigrants, when compared with people who do not compete with immigrants in the labor market (students, nonretired people out of the labor force and the self-employed). This is especially so in 1988. This may be because there is a perception – whether correctly or incorrectly – that the economic strain facing them through labor market unemployment and sluggish wages is due to the presence of

Changes in Attitudes toward Immigrants in Europe

667

immigrants in the labor market. We do not ﬁnd evidence, however, that the unemployed and the employed have different attitudes toward immigrants, a dichotomy emphasized (or perhaps overemphasized) by the previous literature. In 1997, we ﬁnd that only students differ in their attitude toward immigrants, compared to other groups. Our results also show that people who live in neighborhoods with a greater concentration of foreigners tend to have stronger negative attitudes toward foreigners. The increasing contact with foreigners might ignite racial bias and discrimination against foreigners. Those with racial prejudice are also prejudiced against foreigners.

5. Changes in attitudes: a decomposition analysis In this section, we seek to explain the jump in hostile attitudes toward foreigners in Europe between 1988 and 1997. There are two broad approaches to explain the changes in attitudes over time. One relies on the possibility that the characteristics of individuals that give rise to negative attitudes toward immigrants have changed over time, increasing the dislike of foreigners. For instance, those who feel that they are being hurt by immigrants via current competition in the labor market (such as the employed, whose wages may decline as a result of immigration, or the unemployed, whose employment opportunities may shrink), as well as those who feel they were hurt by the labor market competition suffered from immigrants in the past (the retired population), can have strong negative feelings toward foreigners. If the number of these labor market competitors of immigrants rises in the population, one expects that the society will suffer from more negative attitudes toward foreigners. Indeed, in the Eurobarometer surveys, the proportion of the population that competes (or has competed) with immigrants in the labor market increased from 57 percent in 1988 to 66 percent in 1997. More generally, we can describe this type of explanation as a characteristics effect, because it reﬂects how changes in the characteristics of individuals over time affect the likelihood that someone has negative attitudes toward foreigners. Suppose, however, that individual characteristics were not different in 1988 and 1997. A second approach to explain the rising negative attitudes toward foreigners relies on the possibility that the effects of the given individual characteristics on attitudes have changed over time. For example, suppose that those individuals who are competing – or have competed – with foreigners in the labor market become more frustrated over time as a result of the lasting economic strain they suffer. This may spillover into stronger negative attitudes toward foreigners. This suggests that the increased bias against foreigners is not due to rising unemployment or a more sluggish economy, but rather to the fact that the unemployment and recession have lasted for so long, causing people to

668

Ira N. Gang et al.

develop more strongly anti-foreigner attitudes. In terms of our earlier analysis, this type of explanation is reﬂected in the higher coefﬁcients of the variables EMPLOYED, UNEMPLOYED, and RETIRED in the probit equations explaining negative attitudes toward foreigners. As an example, the probit coefﬁcient on the variable UNEMPLOYED reported in Table 5 rises from 0.257 in 1988 to 0.274 in 1997, suggesting that the unemployed had more strongly negative attitudes toward foreigners in 1997 than in 1988. More generally, this type of effect is associated with changes in the coefﬁcients of the probit equations between 1988 and 1997, and we may refer to it as a coefﬁcients effect. Studying characteristics and coefﬁcients effects was formally introduced by Oaxaca (1973) and Blinder (1973). Their works have led to a long literature on decomposing differences over time (or between groups). The implementation and extensions of their analyses have generally been in the context of wage differentials (in general, any continuous variable), although recent extensions allow for discrete dependant variables (e.g., Even and Macpherson, 1990, 1993; Nielsen, 1998; Yun, 2004). Until these recent innovations, decomposing when one had discrete dependant variables (e.g., employment status) was generally accomplished by so-called simulation (see Abowd and Killingsworth, 1984). In these analyses, logits or probits would be estimated for each period, and the coefﬁcients for later period would be replaced with those of the other period in order to calculate a counter-factual predicted probability. Subtracting this counterfactual prediction from the observed probability for the later time period, one sees the effects of the differences in coefﬁcients between the two time periods, holding the characteristics constant. The coefﬁcients can be switched one-by-one to see contribution of each variable, or completely to see the effect of overall change. However, this simulation method is not only tedious but also problematic since it may be sensitive to the order of switching (see Ham et al., 1998, p. 1137, for a discussion of pathdependency). Also, the simulation method usually only partially explains the changes since it looks only at the effect of switching coefﬁcients, that is, the coefﬁcients effect explained earlier. The decomposition method in this chapter can provide a complete picture of the changes. We follow a decomposition method proposed by Yun (2004).11 The likelihood of having anti-foreigner attitude for individual i is estimated by F(Xib), where Xi is a 1 K vector of independent variables, b a K 1 vector of coefﬁcients, and F the standard normal cumulative distribution function. The observed rate to have anti-foreigner attitudes is asymptotically equal to the sample average of individual’s likelihood 11

What Yun (2004) proposed is a general method to decompose differences in the ﬁrst moment for nonlinear models which have already been applied to count-data model (Park and Lohr, 2010) and hazard rate model (Powers and Yun, 2009) in addition to probit/logit models.

669

Changes in Attitudes toward Immigrants in Europe

P to have anti-foreigner attitudes, or P ¼ FðXbÞ ¼ ð1=NÞ N i¼1 FðX i bÞ. Algebraically, the changes between years A and B in the average probability of having anti-foreigner attitudes (PA PB ), where A ¼ 1997 and B ¼ 1988, may be decomposed into two components which represent the characteristics effect and coefﬁcients effect as following, PA PB ¼ ½FðX A bB Þ FðX B bB Þ þ ½FðX A bA Þ FðX A bB Þ, where the ﬁrst and the second components in the right-hand side represent the characteristics effect and coefﬁcients effect; ‘‘over bar’’ represents the value of the sample’s average. The above decomposition provides us with the overall coefﬁcient and characteristics effects. In order to ﬁnd the relative contribution of each variable to changes in negative attitudes toward foreigners between 1988 and 1997, in terms of characteristics and coefﬁcients effects, we employ a decomposition equation for the probit model (as proposed by Yun, 2004);12 PA PB ¼

K X

W kDX ½FðX A bB Þ FðX B bB Þ

k¼1

þ

K X

W kDb ½FðX A bA Þ FðX A bB Þ,

k¼1 k

k

k

k where W kDX ¼ ðX W kDb ¼ ðX A ðbkA bkB Þ= PAK X BkÞbB =ðX PAK X Bk ÞbB Þ, k k X A ðbA bB Þ, and k¼1 W DX ¼ k¼1 W Db ¼ 1, where X A and X B are average values of explanatory variables k for groups A and B, respectively.13

12

In order to obtain a proper weight, the following approximations are used; ﬁrst, an approximation of the value of the average of the function, FðXbÞ, with that of the function evaluated at the average value of exogenous variables, FðXbÞ; second, a ﬁrst order Taylor expansion to linearize the characteristics and coefﬁcients effects around X B bB and X A bA . See Yun (2004) for details. 13 For computing asymptotic standard errors of the characteristics and coefﬁcients effects, see Yun (2005a). We deal with robustness issues, known as the index or parameterization problem and the identiﬁcation problem in detailed decompositions. A decomposition equation with a different parameterization, that is, ½FðX A bA Þ FðX B bA Þ þ ½FðX B bA Þ FðX B bB Þ, is possible; our results with it are not substantially different from those presented here and are available from the authors upon request. Another issue when interpreting the decomposition results is that the coefﬁcients effect in the detailed decomposition is not invariant to the choice of omitted groups when dummy variables are used (see Oaxaca and Ransom 1999, for details of this issue). We follow a solution suggested by Yun (2005b, 2008) that, if alternative reference groups yield different estimates of the coefﬁcients effects for each individual variable, it is natural to obtain estimates of the coefﬁcients effects for every possible speciﬁcation of the reference groups and take the average of the estimates of the coefﬁcients effects with various reference groups as the ‘‘true’’ contributions of individual variables to differentials. Although appearing cumbersome, this can be accomplished with a single estimation. We can transform our probit estimates into a normalized equation and use the normalized equation for our decomposition.

670

Table 8.

Ira N. Gang et al.

Decomposition analysis of increase in anti-foreigner attitudes Characteristics effect Estimate

Aggregate effect Sub-aggregate effects Age and its square Years of education Children15 Head of household Gender Labor market statues Disturbing/nondisturbing Intercept

***

0.0015** 0.0083*** 0.0017*** 0.00004 0.00001 0.0058*** 0.0143*** 0

0.0151

Share (%)

Coefﬁcients effect Estimate

Share (%)

12.04

***

0.1105

87.96

1.21 6.61 1.36 0.04 0.01 4.62 11.41 0

0.0572 0.0506** 0.0027 0.0004 0.0003 0.0004 0.0223*** 0.1364**

45.51 40.26 2.17 0.33 0.21 0.32 17.73 108.58

Source(s): Eurobarometer Survey, 1988, 1997; authors’ calculations. Notes: Percentage share of differences in predicted probabilities of anti-foreigner attitude measured by the variable ‘‘Too many’’ between 1997 and 1988 (0.1256 ¼ 0.4207 0.2951) are reported. ***, **, and * denote signiﬁcance at the 1%, 5%, and 10% levels, respectively.

Table 8 reports the results of this decomposition analysis. The probit equations upon which we base the decompositions were presented at the second and third columns in Table 5 and represent the results of our analysis for 1988 and 1997 (no neighborhood variables). In essence, we decompose the changes in the probability of having negative attitudes toward foreigners between 1988 and 1997, as measured by the variable TOO-MANY-FOREIGNERS in the Eurobarometer surveys. This probability, which is the percentage of the sample who believed that there were ‘‘too many’’ foreigners in their country, rises sharply by 12.6 percent from 29.5 percent to 42.1 percent between 1988 and 1997. The top row in Table 8 shows the overall effects of characteristics versus coefﬁcients in explaining the increased negative attitudes toward foreigners, whereas the sub-aggregate effects depict the role of various groups of variables. About 12 percent of the increased anti-foreigner attitudes (of 12.6 percent) are explained by differences in people’s characteristics between the two years. This means that if people in the sample have had the same characteristics in 1988 and 1997, then the increased probability of having anti-foreigner sentiment would have been 12 percent less. The variable with the largest effect among the various individual characteristics affecting attitudes is DISTURBING, which rises sharply between 1988 and 1997. This means that rising prejudice accounts for 12 percent of the rise in anti-foreigner attitudes. The compositional changes related to labor market statues such as the employed, the unemployed, and so on, also explain some of the rising negative attitudes toward foreigners, but this is less than 6 percent. On the contrary, the increased educational attainment

Changes in Attitudes toward Immigrants in Europe

671

of the population acted to lower the anti-foreigner sentiment between 1988 and 1997. The majority (88 percent) of the increased anti-foreigner sentiment (of 12.6 percent) is explained by the differences in probit coefﬁcients between 1988 and 1997. One could refer to these as behavioral changes since they represent the changes in the strength of the various individual characteristics inﬂuencing attitudes toward foreigners.14 If in both years the various variables inﬂuencing attitudes toward foreigners had the same strength (their coefﬁcients in the probit equation had been equal), then about 88 percent of the increased probability of having negative attitudes toward foreigners would disappear. First, among the various coefﬁcient effects, it is remarkable that the coefﬁcient of educational attainment became less negative between 1988 and 1997. This means that the strength of the ameliorating impact of education on anti-foreigner attitudes diminished over time, explaining about 40 percent of the increased anti-foreigner attitudes between 1988 and 1997. In other words, the educated are increasingly displaying anti-foreigner attitudes in Europe, accounting for a substantial portion of the overall increased anti-foreigner attitudes in the European Union. Second, the coefﬁcient on the DISTURBING variable rises between 1988 and 1997, meaning that racial bias appears to be reﬂected in anti-foreigner behavior more strongly in 1997 than in 1988. However, this also implies that, if the individual does not show racial bias (DISTURBING ¼ 0), the coefﬁcients effect on lowering negative attitudes toward foreigners would be strengthened between 1988 and 1997. When both effects of increased negative attitudes among those with racial bias and of lowered negative attitudes among those without racial bias are taken together, the combined effect turns out negative. Third, the strength of being (or having been) a labor market competitor with immigrants increases the negative attitudes toward immigrants over time. However, the decomposition analysis shows that the effects of changes in coefﬁcients of labor market statuses combined together are not large. There are other interesting ﬁndings. The coefﬁcient effect of age structure (age and age squared taken together) is negative, implying that the negative attitudes of older people, in general, toward immigrants declined in strength between 1988 and 1997. This helped to reduce anti-immigrant sentiments, but clearly not enough to compensate for the other coefﬁcients – or behavioral – changes over time. The changes in the constant term also contribute signiﬁcantly to increasing negative attitudes toward foreigners. The constant term may reﬂect underlying changes in 14

In the well-known Blinder–Oaxaca decomposition analysis for wage differentials, the part explained by the differences in coefﬁcients is usually called ‘‘discrimination.’’ In the decomposition analysis utilized in this chapter, ‘‘behavioral differences’’ or ‘‘behavioral changes’’ are better descriptors.

672

Ira N. Gang et al.

attitudes toward foreigners between 1988 and 1997 which are not captured by the other explanatory variables.

6. Summary and conclusions This chapter has examined the relative signiﬁcance of some of the key forces that inﬂuence the attitudes of European Union citizens toward foreigners (non-European Union people). Using attitudinal survey data from the 1988 to 1997 Eurobarometer surveys, we analyze the role of labor market competition, immigrant concentration, racial/ethnic bias, educational attainment, and a set of other variables that potentially determine attitudes toward immigrants. Estimating probit equations of the likelihood that people in the sample had negative attitudes toward foreigners, the chapter provides an analysis of the connections between an array of explanatory variables and negative attitudes toward (non-European Union) foreigners. The Eurobarometer survey ﬁnds a sharp increase in anti-foreigner attitudes in Europe between 1988 and 1997. For instance, in 1988, a total of 29.5 percent of the sample felt that there were ‘‘too many foreigners’’ in their country, but by 1997 this percentage had risen to 42.1 percent. What are the factors that explain negative attitudes toward foreigners? We ﬁnd that people who directly compete (or have competed) in the labor market with immigrants have stronger negative attitudes toward foreigners, ceteris paribus. This includes not only the unemployed but also employed, salaried workers (who may perceive that their wages are negatively affected by immigrants), and the retired (who may have developed anti-foreigner attitudes in the past, when they were employees in the labor market). This is especially true in 1988. By 1997 students stand out as the one group with minimal anti-foreigner attitudes. We also ﬁnd strong evidence that a greater concentration of foreigners in the neighborhoods where citizens reside also raises the probability of a person displaying negative attitudes toward foreigners, holding other things constant. This may suggest that ethnic bias and discrimination are key forces generating negative attitudes toward foreigners. Larger concentrations of immigrants, being more visible, can set aﬁre the ethnocentric sentiments of prejudiced individuals. Communities with larger concentrations of immigrants may give rise to greater antiimmigrant sentiment. The signiﬁcance of racial/ethnic prejudice is conﬁrmed by our ﬁnding that people who ‘‘feel disturbed by people of a different race’’ also have stronger negative attitudes toward foreigners, holding other things constant. Educational attainment is found to be a strong antidote to antiforeigner sentiments. Older people, however, generally have stronger negative attitudes toward foreigners. And, contrary to our expectations,

Changes in Attitudes toward Immigrants in Europe

673

people who have children less than 15 years of age tend to have more positive attitudes toward foreigners, holding other things constant. This may be explained by the increased likelihood that the children of natives will mix with the children of immigrants, increasing the contact and diffusing the tensions between the adults in the two groups. Using our probit decomposition analysis of the factors determining negative attitudes toward foreigners, we were also able to provide some explanations for the jump in hostile attitudes toward foreigners in Europe between 1988 and 1997. There are two broad approaches to explain the changes in attitudes over time. One relies on the possibility that the characteristics of individuals that give rise to negative attitudes toward immigrants have changed over time, incrementing the dislike of foreigners. We describe this type of explanation as a characteristics effect. A second approach relies on the possibility that the effects of the given individual characteristics on attitudes have changed over time. This type of effect is associated with changes in the coefﬁcients of the probit equations between 1988 and 1997, and we may refer to it as a coefﬁcients effect. The decomposition analysis indicates that about 12 percent of the increased anti-foreigner attitudes displayed by the people sampled in the Eurobarometer survey are explained by differences in people’s characteristics between the two years. The variable with the largest effect among the various individual characteristics affecting attitudes is racial prejudice. The increased proportion of people who compete –or have competed– with immigrants in the labor market explains some of the rising negative attitudes toward foreigners, but only less than 6 percent. However, the increased educational attainment of the population acted to lower the antiforeigner sentiment between 1988 and 1997. We also ﬁnd that 88 percent of the rising anti-foreigner sentiment is explained by coefﬁcient effects. This means that most of the increased antiforeigner sentiment is related to behavioral changes among the population that has strengthened the impact of various individual characteristics on negative attitudes toward foreigners. Key among these behavioral changes is the fact that the strength of the ameliorating impact of education on anti-foreigner attitudes diminished over time. In other words, the highly skilled are increasingly displaying anti-foreigner attitudes in Europe and this accounts for close to 42 percent of the overall increased anti-foreigner attitudes in the Union. In addition, racial bias appears to be reﬂected in stronger anti-foreigner behavior in 1997 than in 1988. However, the negative attitudes of older people in general toward immigrants declined in strength between 1988 and 1997. This helped to reduce anti-immigrant sentiments, but clearly not enough to compensate for the other behavioral changes over time. The rising anti-foreigner trend documented by the Eurobarometer surveys is alarming and requires serious discussion and policy responses. Particularly sobering is the ﬁnding that educational attainment, one of the

674

Ira N. Gang et al.

most signiﬁcant variables acting to reduce anti-foreigner sentiment, diminished its role between 1988 and 1997, with a growing number of skilled workers displaying anti-foreigner sentiment. Our analysis strongly suggests that European countries face a major challenge in battling the ignorance and the social environment that give rise to prejudice and discrimination.

Acknowledgments This manuscript is beneﬁted from comments on an earlier version by Christian Dustmann, Christoph M. Schmidt, and by seminars at Lafayette College and the European Society for Population Economics. Ira Gang thanks the Humboldt Fellowship and Rutgers Research Council for their partial support.

References Abowd, J.M., Killingsworth, M.R. (1984), Do minority/white unemployment differences really exist? Journal of Business and Economic Statistics 2, 64–72. Alber, J., (1994), Toward explaining anti-foreigner violence in Germany. Center for European Studies, Harvard University, Working Paper No. 53. Amemiya, T. (1985), Advanced Econometrics. Harvard University Press, Cambridge. Bauer, T.K., Lofstrom, M., Zimmermann, K.F. (2000), Immigration policy, assimilation of immigrants and natives’ sentiments toward immigrants: evidence from 12 OECD-countries. Swedish Economic Policy Review 7, 11–53. Blinder, A.S. (1973), Wage discrimination: reduced form and structural estimates. Journal of Human Resources 8, 436–455. Card, D. (1990), The impact of the Mariel Boatlift on the Miami labor market. Industrial and Labor Relations Review 43, 245–257. Card, D. (2001), Immigrant inﬂows, native outﬂows, and the local labor market impacts of higher immigration. Journal of Labor Economics 19, 22–64. Channel 4 Hate Commission. (2000), Hate Crime Report. London. DeNew, J.P., Zimmermann, K.F. (1994), Native wage impacts of foreign labor: a random effects panel analysis. Journal of Population Economics 7, 177–192. Dustmann, C., Preston, I. (2001), Attitudes to ethnic minorities, ethnic context and location decisions. Economic Journal 111, 353–373.

Changes in Attitudes toward Immigrants in Europe

675

Dustmann, C., Preston, I.P. (2007), Racial and economic factors in attitudes to immigration. The B.E. Journal of Economic Analysis & Policy 7 (1), Article 62, http://www.bepress.com/bejeap/vol7/iss1/art62 European Monitoring Centre on Racism and Xenophobia (EUMC). (2000), Annual Report 1999. European Commission, Brussels. Even, W.E., Macpherson, D.A. (1990), Plant size and the decline of unionism. Economics Letters 32, 393–398. Even, W.E., Macpherson, D.A. (1993), The decline of private-sector unionism and the gender wage gap. Journal of Human Resources 28, 279–296. Fertig, M., Schmidt, C.M., (2002), First- and second-generation immigrants: what do we know and what do people think? In: Rotte, R., Stein, P. (Eds.), Migration Policy and the Economy: International Experiences, Studies and Comments 1 of the Hans-Seidel-Stiftung Munich, pp. 179–218. Gang, I.N., Rivera-Batiz, F.L. (1994a), Labor market effects of immigration in the United States and Europe: substitution vs. complementarity. Journal of Population Economics 7, 157–175. Gang, I.N., Rivera-Batiz, F.L. (1994b), Unemployment and attitudes toward foreigners in Germany. In: Steinmann, G., Urich, R. (Eds.), Economic Consequences of Immigration in Germany. Physica-Verlag, Germany, pp. 121–154. Gutierrez, D.G. (1995), Walls and Mirrors: Mexican Americans, Mexican Immigrants and the Politics of Ethnicity. University of California Press, Berkeley, CA. Ham, J.C., Svejnar, J., Terrell, K. (1998), Unemployment and the social safety net during transitions to a market economy: evidence from the Czech and Slovak Republics. American Economic Review 88, 1117–1142. Hunt, J. (1992), The impact of the 1962 repatriates from Algeria on the French labor market. Industrial and Labor relations Review 45, 556–572. Krueger, A.B., Pischke, J.-S. (1997), A statistical analysis of crime against foreigners in uniﬁed Germany. Journal of Human Resources 32, 182–209. Lyall, S., (2001), In Ravaged English City, Racial Mix was Volatile. The New York Times, May 31, p. A4. Massey, D.S., Denton, N.A. (1993), American Apartheid: Segregation and the Making of the Underclass. Harvard University Press, Cambridge. Mayda, A.M. (2006), Who is against immigration? A cross-country investigation of individual attitudes toward immigrants. Review of Economics and Statistics 88 (3), 510–530. Melich, A., (1999), Eurobarometer 48: Holiday Travel, October-November 1997 [Computer ﬁle]. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributors].

676

Ira N. Gang et al.

Muhleisen, M., Zimmermann, K.F. (1994), A panel analysis of job changes and unemployment. European Economic Review 38, 793–801. Nielsen, H.S. (1998), Discrimination and detailed decomposition in a logit model. Economics Letters 61, 115–120. Oaxaca, R. (1973), Male-female wage differentials in Urban labor markets. International Economic Review 14, 693–709. Oaxaca, R., Ransom, M.R. (1999), Identiﬁcation in detailed wage decompositions. Review of Economics and Statistics 81, 154–157. OECD (1998, 2001), Trends in International Migration (SOPEMI). Paris, 1998. Park, T.A., Lohr, L. (2010), A oaxaca-blinder decomposition for count data models. Applied Economics Letters 17, 451–455. Powers, D.A., Yun, M.-S. (2009), Multivariate decomposition for hazard rate models. Sociological Methodology 39, 233–263. Reif, K., Melich, A., (1992), Euro-Barometer 30: Immigrants and OutGroups in Western Europe, October-November 1988 File. Ann Arbor MI: Inter-University Consortium for Political and Social Research Producer and Distributor. Scheve, K.F., Slaughter, M.J. (2001), Labor market competition and individual preferences over immigration policy. Review of Economics and Statistics 83, 133–145. Winkelmann, R., Zimmermann, K.F. (1993), Ageing, migration and labor mobility. In: Paul, J., Zimmermann, K.F. (Eds.), Labor Markets in an Aging Europe. Cambridge University Press, Cambridge, pp. 255–282. Winter-Ebmer, R., Zweimueller, J. (1999), Do immigrants displace young native workers? The Austrian experience. Journal of Population Economics 12, 327–340. Yuengert, A.M. (1995), Testing hypotheses of immigrant self-employment. Journal of Human Resources 30, 194–204. Yun, M.-S. (2004), Decomposition differences in the ﬁrst moment. Economics Letters 82, 273–278. Yun, M.-S. (2005a), Hypothesis tests when decomposing differences in the ﬁrst moment. Journal of Economic and Social Measurement 30, 295–304. Yun, M.-S. (2005b), A simple solution to the identiﬁcation problem in detailed wage decomposition. Economic Inquiry 43, 766–772. Yun, M.-S. (2008), Identiﬁcation problem and detailed Oaxaca decomposition: a general solution and inference. Journal of Economic and Social Measurement 33, 27–38. Zimmermann, K.F. (1995), Tackling the European migration problem. Journal of Economic Perspectives 9, 45–62.

CHAPTER 27

The Implications of Social Norms on Immigration Policy Shirit Katav-Herz School of Management and Economics, Tel-Aviv-Yafo Academic College, PO Box 8401, Tel-Aviv-Yafo 61083, Israel E-mail address: [email protected]

Abstract This chapter investigates the relationship between social norms and the local population’s attitude toward immigration. Although there are beneﬁts from immigration in terms of greater consumption opportunities, disutility from changes in social norms due to immigration may vary across different segments of the local population. This social disutility leads to opposition to foreigners even through anti-immigrants actions. The chapter shows how the disutility from changes in social norms will affect behavior toward immigrants and the formulation of immigration policy. Keywords: Immigration, social norms, social capital JEL classiﬁcations: J15, O16, Z10

1. Introduction Afﬂuent countries often encourage immigration from other countries to augment their labor force. At the same time, it is often the case that immigrants are not welcomed by some segments of the local population, particularly in European countries. The opposition to immigration is often unrelated to its economic consequences and has more to do with the fear of losing national and cultural identity. Attitudes toward immigrants appear to be simply an emotional response. The massive inﬂux of immigrants into Western countries has raised concern about loss of national identity and has led to anti-immigrant feelings and even violent acts against immigrants. The interplay between national identity and attitudes toward immigrants is likely to remain a charged issue for a long time to come. Attitudes toward immigration differ even among host countries that have large immigrant populations. Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008033

r 2010 by Emerald Group Publishing Limited. All rights reserved

678

Shirit Katav-Herz

Esses et al. (2006) compare the attitudes toward immigration in Canada and Germany, which have both accepted large numbers of immigrants over the past 50 years. Canada has historically considered immigration to be an integral part of its national development, whereas Germany perceives the large immigrant population as an unintended result of the entry of guest workers and an inﬂux of refugees and asylum seekers. These differing approaches are reﬂected not only in naturalization laws and government policies but also in the attitudes of the host population with regard to accepting and promoting cultural diversity. The chapter examines the implications for immigration policy when immigration brings about change in social norms and when the attitude toward this process differs among various segments of the population. As immigrants leave their country of origin to improve their standard of living, the direction of immigration is usually from impoverished countries to afﬂuent ones. Immigration is in some cases encouraged by the host country; in other cases, it is illegal. Highly skilled immigrants are often invited to a country because they bring needed human capital with them. However, for the most part, immigrants are usually low-skilled and low-paid workers who are on the margins of society. In many developed countries, low-paid immigrants are concentrated in sectors that are more exposed to import competition as a result of the globalization process or work in low-wage sectors producing nontraded goods. Nonetheless, low-skilled and low-paid immigrants produce beneﬁts for the local population through complementarily with other factors of production of the local population, by reducing production costs and by enabling the demand for low-wage services to be satisﬁed.1 The immigration surplus might be even larger when the immigration ﬂow is composed exclusively of skilled workers (Borjas, 1995). Despite the economic beneﬁts from immigration, certain segments of host societies have indicated – through election outcomes and survey results – that they oppose the presence of immigrants. There are several possible explanations. For example, low-skilled immigrants may to some extent displace local workers because they are often willing to work for lower wages.2 In this case, anti-immigrant attitudes are based on threats to personal income. This is supported by evidence linking the number of anti-immigrant incidents to the unemployment rate or the wage rate (see Alber, 1994). However, Krueger and Pischke (1996) found that the threat to personal income explains little of the tendency to participate in violent acts against foreigners in Germany. The threat to social norms is another possible explanation for the opposition to immigration. Immigrants bring with them customs and 1 See Zimmermann (2005) who concludes that immigration is usually beneﬁcial in Western host countries. 2 Overall, there is no evidence that immigration has an adverse effect on local wages or employment levels (Kahanec and Zimmerman, 2008; D’Amuri et al., in press).

The Implications of Social Norms on Immigration Policy

679

values that are foreign to the host society and that affect local social norms. This raises the question of why a change in social norms would be a source of disutility among certain segments of the local population. A form of behavior becomes a social norm when it is adopted by a majority of the population.3 Deviating from a social norm results in disutility for the individual because he is censured by other members of society; he becomes an outsider and all that that implies.4 It is often the case that immigrants continue to behave according to the social norms of their country of origin and thus remain outsiders. As long as they are few in number relative to the local population, there is no perception of a threat to local social norms and it is presumed that they will be assimilated within a short period or at least will not have a major inﬂuence on local social norms. Immigrants can, however, form a subsociety, thus preferring to associate with other immigrants. In this case, immigrants are likely to maintain the modes of behavior from their country of origin and will deviate as a group from local social norms.5 As a result, the local population may come to fear that foreign norms will supplant local ones.6 Examples include lower labor standards and tolerance for child labor; opposition to education for girls and participation of women in the labor force; conservative attitudes toward how women dress (veils and the like); and the perception of animals such as cats and dogs as food rather than pets. Customs can also be sources of conﬂict between local residents and immigrants. For example, female circumcision may be an accepted rite among the immigrants but totally unacceptable to the native population. Social norms are also considered part of social (collective) capital,7 which includes relationships between members of the society that have developed over time. For instance, social capital may include the norm that in certain circumstances individuals will act in the interest of society as a whole, even at the expense of their own self-interest. Foreigners who are not familiar with these norms (because they carry with them different social capital) not only reduce local social capital but may also take advantage of the local norms in ‘‘prisoner’s dilemma’’-type encounters. For an individual considering immigration, opposition (for whatever reason) to immigration in the host country leads to a trade-off between improving his economic situation and the probability of being physically harmed or being treated in a demeaning or insulting way. At the same

3

Social norms can, for example, be instruments for solving the voluntary supply problem of public goods (Coleman, 1988). 4 See, for example, Akerlof (1980), Bernheim (1994), and Sugden (1998). 5 Ko’nya (2007) demonstrates a model where immigrant’s decision composed also from capability and willingness to assimilate with the destination population. 6 See Hillman (2002). 7 See Coleman (1988), Lindbeck (1995), and Paldam and Svendsen (2000).

680

Shirit Katav-Herz

time, local opposition to immigration affects immigrants who are already living in the host countries. Their reaction may be to further cohere and to more strongly adhere to their social norms. For local residents, there is a trade-off between the economic beneﬁts they derive from immigrants and the disutility from the change in social norms. There is also a potential for conﬂict between segments of the local population who differ in the degree they beneﬁt from immigration or in the importance they attribute to preserving social norms. The model presented here incorporates the differences between the various segments of the local population and time dimension. According to the model, opposition to foreigners increases with the number of immigrants arriving in a particular period, and as a result, fewer immigrants will arrive in the following period. A far-sighted immigration policy will take this into account to reduce opposition to immigration. The model is also used to investigate the ﬂow of immigration over time and the conditions under which gradual ﬂow of immigrants is preferable on the entry of the same number of immigrants all at once. As the local population responds to the effect of immigration on local social norms, the equilibrium ﬂow of immigration over time depends on the degree of deviation from local social norms and their immunity to the inﬂuence of foreign norms. The chapter is organized as follows. Section 2 describes the model for the one-period and two-period cases and for myopic and far-sighted immigration policies. Section 3 discusses the distribution of immigration over time. Section 4 concludes. 2. The model We begin by showing that the difference in social norms between the host country and the immigrant population should be a factor in the formulation of immigration policy. Consider a local population composed of t individuals. A local individual i has the following utility function: ui ¼ ui ðci ; Di Þ @ui 40 @ci

and

(1) @ui 40 @Di

where ci is individual i’s level of consumption, which is deﬁned as a proportion of the surplus produced by the employment of immigrants reﬂecting the beneﬁts to the local economy (as detailed in Section 1), and Di is the subjective beneﬁt from the local social norms. Local individuals are satisﬁed with prevailing social norms and their utility declines when those norms change, which occurs with the arrival of

The Implications of Social Norms on Immigration Policy

681

immigrants from a different culture. To incorporate this effect into the model, we specify that beneﬁt Di depends on the ratio between the size of the local population L and the number of immigrants F. The presence of immigrants causes local social norms to change, and therefore, a larger number of immigrants entering the host country leads to greater opposition to immigration. In addition, local individuals differ from each other in the weight they assign to the preservation of local norms, which we denote by Ai. Thus, D i ¼ Ai

L F

(2)

In the labor market, the demand for immigrant workers, denoted by FD, is a negative function of the real wage w: F D ¼ F D ðwÞ

(3)

The supply of immigrants, on the contrary, is a positive function of w and a negative function of anti-immigrant incidents I, which create an uncomfortable environment for immigrants in the host country. Thus, the supply of immigrants is given by: F s ¼ F s ðw; IÞ

(4)

At ﬁrst, we ignore the social norms’ effect on the utility of the local population. We denote the beneﬁt or surplus from the employ^ which determines the individual’s level of ment of immigrants by c, consumption. ^ The utility F** denotes the number of immigrants that maximizes c. from consumption therefore depends on the demand for immigrant workers, which is obtained from: ^ max u0 ðcðF; wðFÞÞÞ

(5)

where F is the number of immigrants. Therefore, F** satisﬁes, @u0 @u0 @c^ @u0 @c^ @w þ ¼0 ¼ @F @c^ @F @c^ @w @F

(6)

The utility of a local individual depends not only on c~ but also on the beneﬁt from the social norms. Thus, the utility function of a local individual is given by: L ^ max u cðF; wðFÞÞ; Di Ai (7) F With social norms now taken into account, we can deﬁne the number of immigrants that maximizes utility, denoted as F*. Clearly, F oF .

682

Shirit Katav-Herz

The consumption surplus c^ increases with the number of immigrants in equilibrium F*. In other words, @c^ 40 @F

(8)

~ The surplus c~ is therefore a function of F through cðFÞ. Thus, when the number of immigrants entering the host country increases, the surplus c^ increases as well, and in equilibrium, the net effect of the number of immigrants F on a local individual i’s surplus ci is positive ð@ci =@FÞ40. As individuals differ from each other in their degree of disutility from changes in social norms, each individual will prefer a different number of immigrants to maximize his own utility. We complete the model by introducing majority voting as the method for deciding on immigration policy. 2.1. The choice of the median voter in a one-period model Members of the local population have single-peaked preferences regarding the number of immigrants to be allowed into the country. A majority vote will therefore result in a stable equilibrium immigration policy according to the preference of the median voter. The number of immigrants Fm that maximizes the utility of the median voter (F m ¼ F ) is determined by: Am

@um @um @cm ¼ @F m @cm @F m

(9)

where Am is the weight attributed by the median voter to the preservation of local norms. The median voter’s preferred number of immigrants F* is a positive function of the size of the local population L because a larger local population reduces the effect of immigrants on local norms and thus reduces the social cost resulting from the presence of immigrants. Therefore, a larger local population will lead to a larger number of immigrants in the median-voter equilibrium, that is, ð@F =@LÞ40. As mentioned, the optimal immigration policy F* is determined by the choice of the median voter Fm, and therefore, all other voters prefer more or less immigrants because they attribute a different weight Ai to the preservation of local norms. A local individual for whom Ai oAm prefers more immigrants, whereas an individual for whom Ai 4Am prefers fewer immigrants. Voters who favor fewer immigrants may take action to discourage immigration, which will provide a public-good beneﬁt to all individuals who would prefer less immigration. Taking action to discourage immigration has the characteristics of a volunteer-type public good. When the beneﬁt differs among individuals, such public goods are

The Implications of Social Norms on Immigration Policy

683

provided by the highest-beneﬁt individual. In our case, this is the individual with the highest disutility from changes in local social norms or in other words the person with the largest Ai.8 2.2. Anti-immigrant actions Consider a two-period model in which immigration policy is determined in each period for that period. A volunteer, who we denote by v, takes action I, which maximizes his lifetime utility uv. Taking such action in the ﬁrst period increases his utility in the second period, denoted as uv2, by reducing the number of immigrants in that period.9 The volunteer’s utility in the second period is an increasing function of the severity of the actions he took to deter immigration (i.e. the level of I). Taking action I involves a cost for the volunteer which is incurred only in the ﬁrst period, i.e. when the action takes place. To economize on notation, the cost is deﬁned as being equal to the level of the action taken against immigrants, i.e. as being equal to I. Moreover, in taking action I, there is a probability p of being caught by the authorities during that period. If caught, a punishment (utility loss) denoted by g is imposed on the individual. This punishment is a positive function of the severity of I, i.e. g ¼ gðIÞ and ð@g=@IÞ40. The volunteer’s expected utility is therefore given by: Eðuv Þ ¼ Eðuv1 Þ pgðIÞ þ uv2 L L ¼ puv1 cðF 1 Þ I; Av pgðIÞ þ ð1 pÞ uv1 cðF 1 Þ I; Av F1 F1 L þ uv2 cðF 2 ðIÞ; Av ð10Þ F 2 ðIÞ where F j is the number of immigrants in period j, which is determined by the median voter’s preferences in each period and by the action I taken in the previous period (only relevant in period 2). The volunteer maximizes utility with respect to the action I and therefore chooses I to satisfy, @Eðuv Þ @uv1 @g @uv2 @c @F 2 @uv2 @F þp þ þ Av 2 ¼ 0 ¼ @I @c @F 2 @I @F 2 @I @I @I

(11)

We denote I* as the solution to (11). The severity of the action I increases with the disutility of the volunteer as a result of the deviation from local social norms, that is, ð@I =@Ai Þ40. As a result, the severity of 8

Higher social dominance-oriented individuals generally hold more negative attitudes toward immigrants (Esses et al., 2006). 9 See Appendix for proof of this claim.

684

Shirit Katav-Herz

the action against immigrants increases with the number of immigrants in the ﬁrst period, that is, ð@I =@F Þ40.

2.3. The number of immigrants in the second period The utility of the median voter in the second period is given by: L um2 ¼ um2 cðF 2 ðIÞ; Am2 F 2 ðI Þ

(12)

The only change in period 2 is as a result of the action I in period 1. Given this action, the number of immigrants in period 2, denoted by F 2 , is smaller than the number of immigrants in the ﬁrst period, denoted by F 1 . A larger number of immigrants in the ﬁrst period encourages stronger action I, which leads to fewer immigrants in the second period. This is summarized in the following proposition: PROPOSITION 1. The number of immigrants in the second period is a decreasing function of the number of immigrants in the ﬁrst period. The severity of actions against immigrants is a decreasing function of the probability of being caught and the severity of the resulting punishment. The disutility resulting from the deviation from local norms is an increasing function of the number of immigrants in the ﬁrst period and therefore the severity of the action against immigrants is as well. A government that understands the behavior of individuals taking action against immigrants should take it into consideration when determining the number of immigrants allowed into the country in the ﬁrst period.

2.4. A far-sighted median voter The median voter, and all other voters for that matter, has been assumed to be myopic up to this point. In other words, she only considers the current period’s costs and beneﬁts when voting on immigration policy. A far-sighted voter on the contrary takes into account the effect of F 1 on F 2 (recall that F 2 is a function of I which is determined by the volunteer’s anti-immigrant action, given the level of F 1 ). A far-sighted median voter therefore maximizes his lifetime utility in period 1:10 uv ¼ u1 ðF 1 Þ þ u2 ðF 2 ðF 1 ÞÞ

10

For simplicity, utilities are presented as depending only on F.

(13)

685

The Implications of Social Norms on Immigration Policy

The number of immigrants, which we now denote as determined by: @um1 @um2 @F 2 ¼ @F 1 @F 2 @F 1

F~ 1 ,

is

(14)

We therefore are able to formulate the following proposition: PROPOSITION 2. When voters are far-sighted and therefore internalize the effects of their actions on second-period outcomes, the number of immigrants in the ﬁrst (second) period is smaller (larger) than in the case of myopic voters, that is, F~ 1 oF 1 and F~ 2 4F 2 . PROOF. When voters do not take into consideration the effect of antiimmigrant acts (I) (the myopic case), F is chosen to maximize u1 ðF 1 Þ. When this effect is internalized (the far-sighted case), F is chosen to maximize u1 ðF 1 Þ þ u2 ðF 2 ðF 1 ÞÞ. For values of F2 smaller than the optimal value in the myopic case, we have ð@u2 =@F 2 Þ40. Moreover, ð@F 2 =@F 1 Þo0 (because ð@I =@F 1 Þ40;) and ð@F 2 =@IÞo0. As a result, the number of immigrants in the ﬁrst period is greater in the myopic case than in the far-sighted case, that is, F 1 ð1Þ4F~ 1 ð2Þ. We also conclude that the number of immigrants in the second period is larger in the myopic case than in the far-sighted case, that is, F 2 ð1ÞoF~ 2 ð2Þ. 3. The distribution of immigration over time As discussed earlier, there are economic advantages to the local population in allowing immigrants to enter the country; however, at the same time, there are costs in the form of changes in local social norms. This section deals with the consequences of immigration policy on the distribution of immigration over time and its effect in turn on the preservation of local social norms. We ﬁrst examine how the timing of immigration affects local social norms. It is assumed that a large number of immigrants arriving in a short period will have a larger effect on local social norms than the same number of immigrants arriving over a longer period. Thus, local social norms will be less threatened by foreign norms brought over by immigrants if the immigrants arrive at a more gradual ﬂow and are thus assimilated in the local society to a greater extent. The social norms in the host country can be speciﬁed as a weighted average of the original local social norms (before the arrival of immigrants) and the foreign norm, which depends on the number of individuals that adhere to it. ‘‘New’’ immigrants always act according to the foreign norm during their initial period in the country and the local norm is altered by exposure to the foreign norm. The social norms

686

Shirit Katav-Herz

therefore depend on the composition of the population and on its immunity to the foreign norm. The local social norms SN before any immigration thus depend on the number of local residents L: (15)

SN0 ¼ f ðLÞ

In the ﬁrst period of immigration, the new local social norms are a weighted average of the pure local norms SN0 and the foreign norms G(F1) where F1 is the number of immigrants who enter the country during this period. The foreign norms depends on the number of immigrants arriving in each time period F and the weight l attached to the local norms represents its immunity to the foreign norm (l40:5). Thus, SN1 ¼ lSN0 þ ð1 lÞGðF 1 Þ.

(16)

The function G will be negative because the foreign norms always affect the local ones. In the second period, the new local norms are now a blend of the local norms from the previous period and the pure foreign norms brought in by the new immigrants, as given by: SN2 ¼ lSN1 þ ð1 lÞ GðF 2 Þ

(17)

For the general case of n periods: " # n X n ni l GðF i Þ SNn ¼ l f ðLÞ þ ð1 lÞ

(18)

i¼1

Due to the gradual entry of immigrants, the original social norms will be diluted. An alternative immigration policy would be to permit the entry of an identical number of immigrants, but only in the second period. In this case, the new local norms will be, SN2 ¼ lSN0 þ ð1 lÞ GðF 1 þ F 2 Þ and for the general case of n periods: ! n X SNn ¼ lf ðLÞ þ ð1 lÞ G Fi .

(19)

(20)

i¼1

Although gradual immigration has a weaker effect on local social norms in the period when the immigrants arrive (the direct effect), it dilutes the original local norms (the indirect effect) for more periods. A necessary but not sufﬁcient condition for ensuring that the direct effect on the local norms is stronger when all immigrants arrive together is for G to be

The Implications of Social Norms on Immigration Policy

687

characterized by increasing returns to scale, that is,. GðF 1 þ F 2 ÞoGðF 1 Þ þ GðF 2 Þ

(21)

We now derive the condition for determining which policy will be more effective in preserving the original local norm. Gradual immigration is preferred whenever the difference between the direct effect of all immigrants entering the country together and the direct effect of gradual entry dominates the erosion of the original local norms through gradual entry, that is, when, lf ðLÞolGðF 1 Þ þ GðF 2 Þ GðF 1 þ F 2 Þ

(22)

This condition11 depends on the function G and also on the immunity of the local norm to the foreign norm brought over by the immigrants, which is represented by l.

4. Conclusions The chapter has attempted to show how social norms inﬂuence immigration policy. Assuming that social norms reﬂect individual preferences, a change in those norms will reduce the utility of local individuals. Immigration threatens the local social norms, thus creating local opposition to immigration, in spite of its economic beneﬁts. The model presented here explains the actions of a local minority in response to the policy chosen by the median voter in a system of majority voting. The reaction to immigration, although it involves illegal actions, can be modeled as a volunteer public good. The model also shows how immigration policy can vary across countries according to the preferences of the local population for preserving social norms and according to differences in income and wealth. For example, in the long run, there will be a greater degree of change in social norms in more afﬂuent countries. The effect of the timing of immigration on social norms, and thereby on the opposition to immigration, was also examined, and the conditions under which gradual immigration is preferred over immigration all at once were examined. The most signiﬁcant conclusion of the chapter is that policy makers should take into account besides the economic beneﬁts of immigration also the local population’s sensitivity toward the assimilation of foreigners into the society.

11

The general condition for the superiority of gradual immigration is: P P ð1 lÞ½ ni¼1 lni GðF i Þ Gð ni¼1 F i Þ4lf ðLÞ½1 ln1 , where n equals the number of periods.

688

Shirit Katav-Herz

Acknowledgment I thank Arye L. Hillman, an anonymous referee, and the editors for helpful comments. Appendix This purpose of this appendix is to explain the positive relationship between the number of immigrants permitted into the country, denoted by F*, and the severity of anti-immigrant actions, denoted by I. Recall that utility increases monotonically with the level of consumption, denoted by c. Therefore, for simplicity, we relate F** to the number of immigrants that maximizes c: max cðF; wðFÞÞ; IÞ @F ð@2 c=@F@IÞ o0 ¼ @I ð@2 c=@F 2 Þ because ð@2 c=@F 2 Þo0 and obviously ð@2 c=@F@IÞo0. As F oF , we have, @F o0 @I References Akerlof, G.A. (1980), A theory of social custom, of which unemployment may be one consequence. The Quarterly Journal of Economics 94, 749–775. Alber, J. (1994), Toward explaining anti-foreigner violence in Germany. Working Paper No. 53, Center for European Studies, Harvard University. Bernheim, D.B. (1994), A theory of conformity. Journal of Political Economy 102 (5), 841–877. Borjas, G.J. (1995), The economic beneﬁts from immigration. Journal of Economic Perspectives 9 (2), 3–22. Coleman, J.S. (1988), Social capital in the creation of human capital. American Journal of Sociology 94, S95–S120. D’Amuri, F., Ottaviano, G.I.P., Peri, G. (2010), The labor market impact of immigration in Western Germany in the 1990s. European Economic Review 54 (4), 550–570. Esses, V.M., Ulrich, W., Carina, W., Mattias, P., Wilbur, C.J. (2006), Perception of national identity and attitudes toward immigrants and immigration in Canada and Germany. International Journal of Intercultural Relations 30, 653–669.

The Implications of Social Norms on Immigration Policy

689

Hillman, A.L. (2002), Immigration and intergenerational transfers. In: Siebert, H. (Ed.), Economic Policy for Aging Societies. Kluwer Academic Publisher, Dordrecht, pp. 204–217. Kahanec, M., Zimmermann, K.F. (2008), International migration, ethnicity and economic inequality. Discussion Paper No. 3450, IZA. K’onya, I. (2007), Optimal immigration and cultural assimilation. Journal of Labor Economics 25 (2), 367–391. Krueger, A.B., Pischke, J.-S. (1996), A statistical analysis of crime against foreigners in uniﬁed Germany. National Bureau of Economic Research, Working Paper No. 5485. Lindbeck, A. (1995), Welfare state disincentives with endogenous habits and norms. Scandinavian Journal of Economics 97 (4), 477–494. Paldam, M., Svendsen, G.T. (2000), An essay on social capital: looking for the ﬁre behind the smoke. European Journal of Political Economy 16 (2), 339–366. Sugden, R. (1998), Normative expectations: the simultaneous evolution of institutions and norms. In: Ben-Ner, A., Putterman, L. (Eds.), Economics, Values and Organization. Cambridge University Press, Cambridge, pp. 73–100. Zimmermann, K.F. (2005), European Migration: What Do We Know? Oxford University Press, Oxford.

CHAPTER 28

Ethnic Fragmentation, Conﬂict, Displaced Persons and Human Trafﬁcking: An Empirical Analysis Randall K.Q. Akeea, Arnab K. Basub, Nancy H. Chauc and Melanie Khamisd a

Department of Economics, Tufts University, Medford, MA 02155, USA and Institute for the Study of Labor (IZA), Bonn, Germany E-mail address: [email protected] b Department of Economics, College of William and Mary, Williamsburg, VA 23187, USA, Center for Development Research (ZEF), University of Bonn and Institute for the Study of Labor (IZA), Bonn, Germany E-mail address: [email protected] c Department of Applied Economics and Management, Cornell University, Ithaca, NY 14853, USA, Center for Development Research (ZEF), University of Bonn and Institute for the Study of Labor (IZA), Bonn, Germany E-mail address: [email protected] d Institute for the Study of Labor (IZA), Schaumburg-Lippe-Str. 5-9 D-53113, Bonn, Germany E-mail address: [email protected]

Abstract Ethnic conﬂicts and their links to international human trafﬁcking have recently received a surge in international attention. It appears that ethnic conﬂicts exacerbate the internal displacement of individuals from networks of family and community, and their access to economic and social safety nets. These same individuals are then vulnerable to being trafﬁcked by the hopes of better economic prospects elsewhere. In this chapter, we empirically examine this link between ethnic fragmentation, conﬂicts, internally displaced persons, refugees, and international trafﬁcking, making use of a novel dataset of international trafﬁcking. We conduct a direct estimation, which highlights the ultimate impact of ethnic fragmentation and conﬂict on international trafﬁcking through internal and international displacements. Keywords: ethnic fragmentation, conﬂict, displaced persons, human trafﬁcking JEL classiﬁcations: R23, D74, O11, Z12

Frontiers of Economics and Globalization Volume 8 ISSN: 1574-8715 DOI: 10.1108/S1574-8715(2010)0000008034

r 2010 by Emerald Group Publishing Limited. All rights reserved

692

Randall K. Q. Akee et al.

1. Introduction The link between ethnic conﬂicts and international trafﬁcking is an issue that has recently received a surge in international attention. The main argument is that internal conﬂicts encourage the internal displacement of individuals from networks of family and community, and their access to economic and social safety nets. These same individuals are particularly vulnerable to being trafﬁcked, by the hopes of better economic prospects elsewhere. In this chapter, we take this link between ethnic fragmentation, conﬂicts, internal and internally displaced people from a country, and international trafﬁcking to the data for the ﬁrst time, making use of a novel dataset of international trafﬁcking. While there is an extensive empirical literature linking ethnic fragmentation to conﬂict, and an equally extensive literature linking conﬂict to internally displaced persons (IDPs) and international refugees, scant empirical attention has been accorded either to the link between IDPs/refugees and international human trafﬁcking or to the link between ethnic fragmentation, conﬂict, IDPs/refugees, and trafﬁcking. Given that the factors that force migration within and across international borders are similar to the ones that are at play in the trafﬁcking context, it would seem natural to explore the link. However, before we look into the existing literature in this area, it is necessary to unravel the causal link from ethnic fragmentation to conﬂict since the issue of IDPs and refugees is crucially linked to the incidence and intensity of conﬂict. Is ethnic fragmentation a cause for conﬂict within nations? The answer is not clear. Before 1980, only 15 countries could be identiﬁed as homogenous with the two Koreas, Portugal, and Japan leading this select group (Connor, 1983; Lee et al., 2004). Correspondingly over the 1950–1989 period, Gurr (1993) ﬁnds that nonviolent protests by ethnic minority groups increased by 230%, violent protests rose by 430%, and rebellions increased by 360%. Since the end of the Cold War the fraction of countries that could be characterized as ethnically homogenous has fallen further with the break-up of the ex-Soviet Union, and as Reilly (2000) reports, of the 110 major conﬂicts globally over this time period, 103 were intra-state in nature. A primary cause of ethnic conﬂict across nations arises out of interactions between ethnic groups and government, through (i) government coercion in response to the threat of dissent and rebellion from ethnic groups, (ii) ethnic groups having a mobilization advantage and are therefore more likely to engage in protest, or (iii) democratization that ties a government’s hands with respect to coercion thereby allowing ethnic groups the leeway to mobilize and protest (Lee et al., 2004). Lindstro¨m and Moore (1995) in fact ﬁnds support for the hypothesis that ethnic fragmentation is positively correlated with conﬂict, in consonance with the theory that ethnic fragmentation is closely related with mobilization for dissent. The literature on the link between ethnic fragmentation and conﬂict can be further classiﬁed into two branches – democratic and nondemocratic

Ethnic Conﬂicts and Their Links to Human Trafﬁcking

693

societies. In democratic societies, the main cause of conﬂict is increasing inter ethnic-group economic inequality caused by (i) economic policies instituted by the government, (ii) the classic tragedy of the commons case wherein one ethnic group fails to internalize the costs that their choices impart on other ethnic groups, (iii) bureaucratic corruption, (iv) the ‘‘resource curse’’ where the rents from natural resource extraction accrue only to a minority group within a country, and ﬁnally (v) political transitions. In so far as issues (i), (ii), and (iii) above are concerned, Horowitz (1985) notes that individuals often derive enjoyment from seeing beneﬁts accrue to members of their group (be it class- or ethnic-based) even when they themselves do not directly share in those beneﬁts. Empirically, Alesina et al. (1999) ﬁnd that U.S. cities characterized by higher levels of ethnic fragmentation and economic inequality exhibit higher overall levels of both government spending and debt while at the same time devote lower shares of total spending toward investment in public goods. This suggests that cities with higher ethnic fragmentation and inequality spend more on patronage for conﬂicting special interest groups. Thus, any economic or social policy that is deemed to confer additional beneﬁts purely to a particular ethnic group can be a cause for dissent and conﬂict. An additional cause of ethnic conﬂict stems from the inability of international, national and regional powers to adequately provide security for minority groups. Finally, for nondemocratic countries, ethnic conﬂict is more often than not linked to repression of ethnic groups by military dictatorships, and frequent struggles for power through coups and rebellions. However, it can be argued that ethnic fragmentation (or the share of an ethnic group in the total population) within a country does not necessarily imply dissent even in the face of perceived unjust economic and social policies – what matters probably more is the relative strength of an ethnic group vis-a`-vis others within a country. While the theories linking ethnic fragmentation and conﬂict are wide-ranging and still debated, the link between conﬂict (ethnic, religious, ideological, or otherwise) and the problem of IDPs and refugees is far more clear-cut. IDPs and international refugees constitute the spectrum of people ﬂeeing conﬂict, post-conﬂict returnees, people displaced by environmental and natural disasters, people displaced by development projects like large dams. Spiegel (2004) notes that by the end of 2002, there were approximately 40 million displaced people globally with 15 million refugees (UNHCR, 2003), and 25 million IDPs (Global IDP Project, 2003). Moreover, IDPs and refugees are either the poorest or those stripped of resources by stronger groups (Mani, 2005). Displacement leads to the breakdown of social structures and informal and formal insurance mechanisms along with a disruption of employment, healthcare, education and ﬁnancial services making IDPs and refugees a vulnerable group. Specially affected amongst the IDPs and refugees are women and children who suffer the most in terms of food insecurity, hunger and unequal distribution of material goods. As a result, women and children are the most at risk of exploitation and abuse, including

694

Randall K. Q. Akee et al.

coercion into transactional sex for survival. Indeed, Mani (2005) points to the case of Sierra Leone where provisions were allocated to women refugees in exchange of sexual favors with the threat of violence. While studies do document that women and children are the most vulnerable amongst IDPs and refugees, it does not necessarily imply that this group are also victims of trafﬁcking across international borders. Our hypothesis that conﬂict-prone countries with a high number of IDPs and source countries for refugees may also turn out to be the source country for trafﬁcked victims is based on two sets of information delineated below. First, data on the countries of origin for asylum seekers into North America, Western Europe and Australia closely mirror the country of origin of trafﬁcked victims into these regions. Second, case studies based on two ethnically fragmented and conﬂict prone countries – Nigeria and Nepal – show that the IDPs and refugees, specially, women and children, are indeed victims of trafﬁcking. Data on international refugees and asylum seekers from the United Nations High Commissioner for Refugees (UNHCR) show that the global refugee population grew from 2.4 million in 1975 to 14.9 million in 1990. After reaching a peak at the end of the Cold War, the global refugee population had declined to 12.1 million in 2000 (UNHCR, 1995; UNHCR, 2000; Castles and Loughna, 2003; Zlotnik, 1999). Refugees came mainly from countries affected by civil conﬂict with the top ten countries of origin in 1999 being Afghanistan (2.6 million), Iraq (572,000), Burundi (524,000), Sierra Leone (487,000), Sudan (468,000), Somalia (452,000), Bosnia (383,000), Angola (351,000), Eritrea (346,000), and Croatia (340,000). Next we look at the top 10 countries of origin for asylum seekers entering USA, Canada, Western Europe, and Australia from 1990 to 2001 According to Castles and Loughna (2003), the countries of origin of the top 10 asylum seekers into the USA over the 1990–2001 period were El Salvador (223,887), Guatemala (178,047), Mexico (66,338), China (60,926), Haiti (51,308), Nicaragua (34,411), India (30,985), Russia (20,913), Pakistan (16,700), and Cuba (16,600) with the share of the top 10 being 70% of the total number of asylum seekers over this time period. Notable here is the fact that the top three countries in question (El Salvador, Guatemala, and Mexico) are also ethnically fragmented and conﬂict ridden countries with a signiﬁcant number of IDPs and refugees from neighboring conﬂict prone countries like Nicaragua and Honduras. A similar pattern for asylum seekers can be also observed for Canada with the top 10 countries of origin for asylum seekers being Sri Lanka (40,009), Somalia (21,120), Pakistan (18,680), China (17,651), Iran (15,590), India (14,106), Mexico (8,940), Hungary (8,915), Israel (8,527), DR Congo (8,229) with the share of the top ten being 46% of the total number of asylum seekers over the 1990–2001 time period. Once again, the top three countries of origin of asylum seekers into Canada (Sri Lanka, Somalia, and Pakistan) are ethnically fragmented and conﬂict ridden countries. Moreover, Somalia is also host to refugees ﬂeeing conﬂict and food insecurity from Ethiopia and

Ethnic Conﬂicts and Their Links to Human Trafﬁcking

695

Sudan, Pakistan being host to Afghan refugees while DR Congo hosts refugees from Rwanda. Coincidentally, data from 2002 also show that the USA and Canada are also host countries of trafﬁcked victims from Mexico, Guatemala, and El Salvador (Akee et al., 2009). For Western Europe the top 10 countries of origin of asylum seekers over the 1990–2001 time period were FR Yugoslavia (935,973), Romania (412,326), Turkey (392,867), Iraq (272,918), Afghanistan (192,581), Bosnia & Herzegovina (184,005), Sri Lanka (169,666), Iran (143,651), Somalia (142,148), DR Congo (123,441) with the top 10 share of the total number of asylum seekers over the 1990–2001 time period being 59%. All countries in the top 10 list are ethnically fragmented with either prolonged periods of civil conﬂict or in political transition. Similar to the USA and Canada, Western European countries are also host to trafﬁcked victims from the conﬂict prone countries of Eastern Europe (Bosnia & Herzegovina, Georgia, and Kyrgyzstan, among others) and Africa (Somalia, Nigeria, Sierra Leone, and Mozambique, among others). Finally, for Australia the top 10 countries of origin of asylum seekers between 1996–2001 were Indonesia (7,529), China (6,649), Iraq (5,378), Philippines (4,665), Afghanistan (4,241), Sri Lanka (4,025), India (2,873), Fiji (2,134), Iran (1,910), Thailand (1,263) with the top 10 share of the total number of asylum seekers over this time period being 65%. Needless to say, Indonesia and Sri Lanka have been plagued by prolonged ethnic conﬂicts, and Australia is also a host country of trafﬁcked victims from Indonesia, Thailand and Cambodia, among others. However, one cannot draw a deﬁnite conclusion that the pattern of asylum seekers and trafﬁcked victims are positively correlated from a casual look at the data. After all, asylum migration (which is legal) and trafﬁcking (illegal) are distinctly different channels through which people move or are moved across borders. While it might well be either that trafﬁcked victims could be misled with the hope of gaining asylum in host countries by middlemen or that trafﬁcking becomes the next best alternative in response to stringent immigration laws in host countries, the dynamics of legal migration is distinct from that of illicit migration like trafﬁcking where migrants face different levels of risk, vulnerability, and employment outcomes in the host countries. Sufﬁce to say, that some of the push factors behind asylum migration, namely (i) repression of minorities or ethnic conﬂict, (ii) civil war, (iii) high numbers of IDPs relative to total population, (iv) poverty as reﬂected in low per capita income, (v) low position on the Human Development Index (HDI), (vi) low life expectancy, (vii) high population density, and (viii) high adult literacy rate; mirror closely the push factors that induce trafﬁcking, namely (i) poverty, (ii) lack of educational opportunities, and (iii) armed conﬂict (Castles and Loughna, 2003; Akee et al., 2009). To get a better sense of what drives the positive correlation between conﬂict, IDPs/refugees, and trafﬁcked victims, we look at two case studies, from Nigeria (ethnic, religious, and resource conﬂict) and Nepal (ideological conﬂict).

696

Randall K. Q. Akee et al.

Our data on trafﬁcking suggests that Nigeria is hub country of trafﬁcked victims (women and children). In other words, it is both a source as well as a host country for trafﬁcking. Victims are trafﬁcked out of Nigeria to Western Europe, South Africa and Gulf countries while trafﬁcked victims enter Nigeria from West African countries like Liberia, Sierra Leone, Mali, Cameroon, and Coˆte d’Ivoire (Akee et al., 2009). Recent evidence suggests that trafﬁckers in Kano state exploited the annual pilgrimage to Mecca to trafﬁc children, men, and women for different exploitative purposes like prostitution, begging, and domestic work. Moreover, and historically, women are recruited in the Edo state and are trafﬁcked into Italy, Netherlands, Spain and other countries to work as prostitutes (Carling, 2005; de Haas, 2006; Ehindero et al., 2006; Ibeanu, 1999). Concurrently, Nigerians are the ﬁfth largest group of asylum seekers in Western European countries while at the same time host a large number of refugees and asylum seekers – majority of whom are from Sierra Leone, Chad, Liberia, DR Congo, Sudan (Darfur), Somalia, Coˆte d’Ivoire, Niger, and Cameroon (de Haas, 2006). There are also two major issues that lead to a high degree of labor mobility within Nigeria: conﬂict and the institution of ‘foster’ children. Religious and ethnic conﬂicts (in Plateau and Kano states in 2004 and in Benue state in 2001 respectively) as well as conﬂicts over crude oil mining and reﬁning in the Delta area has led to Nigeria having the highest number of IDPs in West Africa – estimated to be as high as 1.2 million at the end of the 1990s (de Haas, 2006). Furthermore, child fostering is a well-established practice in Nigeria in which poor rural families send their children to family members in urban areas so that they can get better education and employment opportunities. However, many of these children end up working as child laborers. Thus, IDPs, refugees into Nigeria, and foster children combine to constitute a vulnerable group susceptible to trafﬁckers (de Haas, 2006). Nepal, unlike Nigeria that is a hub country of trafﬁcking, is a source country for trafﬁcked victims primarily into India. Armed conﬂict between Maoist guerillas and the ex-Royal Nepalese Army has led to an estimated displacement of 200,000 people both internally and across the border (Internal Displacement Monitoring Centre, 2006; Bharadwaj et al., 2007; UNOCHA 2008). However, a high percentage of the displaced population in Nepal constitute women and young girls, who in addition to lacking access to basic needs are vulnerable to trafﬁcking (Tamang, 2009). With the above background in place, we empirically test whether there is indeed a correlation between ethnic fragmentation, conﬂict, IDPs/ refugees, and trafﬁcking. In what follows, we use religious and linguistic fragmentation, in addition to ethnic fragmentation and different measures of conﬂict to understand which type(s) of fragmentation and conﬂict(s) are most signiﬁcant in explaining the presence of IDPs within and refugees from a country, as well as the factors that increase the incidence of trafﬁcking between countries.

Ethnic Conﬂicts and Their Links to Human Trafﬁcking

697

2. Data For this research on human trafﬁcking and the relationship between IDPs and refugees, ethnic fragmentation and conﬂict, data were collected from various sources described later in text. A description of these speciﬁc variables can be found in Table 1, which contains all the variables employed in our estimations. Table 2 shows the basic descriptive statistics of the dataset at the country level for the main explanatory variables of interest: fragmentation, IDPs/refugees and conﬂict measures. Finally, Table 3 provides the descriptive statistics of the estimation sample. Data on the reported incidence of trafﬁcking were compiled from the country-by-country descriptive accounts of the Trafﬁcking in Persons Table 1.

Variable description

Human trafﬁcking Trafﬁcking

Incidence of trafﬁcking (host–source country), (0/1)

Fragmentation Ethnic Religion Language

Ethnic fractionalization index Religious fractionalization index Language fractionalization index

IDPs and refugees Refugees/IDPs

Refugee and internally displaced persons, (0/1)

Conﬂict Cumulative intensity Cumulative intensity level of conﬂict, taking into consideration the conﬂict history – 0: conﬂict has not over time resulted in more than 1,000 battle-related deaths, 1: if conﬂict reaches threshold Intensity Intensity level of conﬂict – 1: minor, 2: war Count Number of conﬂicts within a country Type1 No extra-state conﬂict (0), extra-state minor armed conﬂict (1), extrastate intermediate armed conﬂict (2), extra-state war (3) Type2 No interstate conﬂict (0), interstate minor armed conﬂict (1), interstate intermediate armed conﬂict (2), interstate armed war (3) Type3 No internal conﬂict (0), internal minor armed conﬂict (1), internal intermediate armed conﬂict (2), internal war (3) Type4 No internationalized internal conﬂict (0), internationalized internal minor armed conﬂict (1), internationalized internal intermediate armed conﬂict (2), internationalized internal war (3) Other controls GDP Prostitution Rule of Law Landlocked Common border Common region

Log of GDP per capita constant 1995 US dollars Prohibition of prostitution (0/1) Rule of law index Landlocked country (0/1) Countries share common border (0/1) Countries are in common region (0/1)

Sources: US Department of State (2002); Protection Project (2002); Alesina et al. (2003); UNHCR; UCDP/PRIO Armed Conﬂict Dataset; Gleditsch et al. (2002); World Bank (2004).

698

Table 2.

Randall K. Q. Akee et al.

Descriptive statistics: fragmentation, IDPs/refugees and armed conﬂict, country level

Variable

Observations Mean

Standard Minimum Maximum Deviation

Fragmentation data Ethnic fractionalization Religious fractionalization Language fractionalization

180 177 184

0.440 0.391 0.437

0.255 0.277 0.235

0 0.002 0.002

0.930 0.923 0.860

187

25676.09

101896.5

0

875402.9

187

63.490

259.093

0

3134.588

187

87613.63

312417.1

0

3378246

187

0.390

0.489

0

1

108 108

1.272 0.452

0.296 0.375

1 0

2 1

187

0.257

0.471

0

4.148

187 187 187 187

0.025 0.093 0.206 0.144

0.119 0.217 0.415 0.353

0 0 0 0

0.893 1.104 2.463 2.259

IDPs and refugees data IDPs and IDP-like situations (1993–2001 average) Refugee population by origin (1992–2001 average) IDPs/IDP-like situation and refugees IDPs/IDP-like situation and refugees dummy Armed conﬂict Conﬂict 1 Intensity level of conﬂict Cumulative intensity level of conﬂict Conﬂict 2 (1946–2001 average) Number of conﬂicts within a country Type 1 (extra-state) Type 2 (interstate) Type 3 (internal) Type 4 (internalized internal)

Sources: Authors’ calculations based on Alesina et al. (2003); UNHCR; UCDP/PRIO Armed Conﬂict Dataset; Gleditsch et al. (2002).

(TIP) report (United States Department of State, 2002). A country is designated as a ‘‘Host’’ country for trafﬁcked victims only if 100 cases were reported in the past year. Country host–source pairs of trafﬁcking are coded from these reports for the year 2002. The number of IDPs and IDPlike situations are data collected by UNHCR, the United Nations Refugee agency and we use their data for the period 1993–2001. We use data on the number of refugees from origin from UNHCR (2002) to generate an average for the years 1993–2001. The incidence of refugees from a source country along with IDPs within a source country is then combined in a single IDPs/refugees dummy variable for a source country. Fragmentation measures were taken from Alesina et al. (2003), where fragmentation P 2 (ethnic, religious, or linguistic) is deﬁned as FRACTj ¼ 1 N i¼1 S ij , where Sij is the share of group i, (i ¼ 1, y, N) in country j. Ethnic,

699

Ethnic Conﬂicts and Their Links to Human Trafﬁcking

Table 3.

Descriptive statistics of the estimation sample

Variable

Observations

Mean

Standard Deviation

Minimum

Maximum

Trafﬁcking Ethnic Religion Language Refugees/IDPs Cumulative intensity Intensity Count Type1 Type2 Type3 Type4 GDP source GDP host Prostitution source Prostitution host Rule of Law source Rule of law host Landlocked source Landlocked host Common Border Common Region

34,969 33,660 34,408 33,099 34,969 20,196 20,196 34,969 34,969 34,969 34,969 34,969 31,416 31,416 34,221 34,221 29,546 29,546 33,660 33,660 34,969 34,969

0.017 0.440 0.437 0.391 0.390 0.452 1.272 0.257 0.025 0.093 0.206 0.144 7.608 7.608 0.383 0.383 0.006 0.006 0.217 0.217 0.015 0.154

0.130 0.254 0.234 0.276 0.488 0.373 0.294 0.470 0.119 0.216 0.413 0.352 1.597 1.597 0.486 0.486 0.912 0.912 0.412 0.412 0.120 0.361

0.000 0.000 0.002 0.002 0.000 0.000 1.000 0.000 0.000 0.000 0.000 0.000 4.553 4.553 0.000 0.000 2.153 2.153 0.000 0.000 0.000 0.000

1.000 0.930 0.860 0.923 1.000 1.000 2.000 4.148 0.893 1.104 2.463 2.259 10.976 10.976 1.000 1.000 1.996 1.996 1.000 1.000 1.000 1.000

Sources: Authors’ calculations based on US Department of State (2003); Protection Project (2002); Alesina et al. (2003); UNHCR; UCDP/PRIO Armed Conﬂict Dataset; Gleditsch et al. (2002); World Bank (2004).

religious, and language fractionalization cover a larger range of countries and various aspects of fragmentation, which we employ here to understand the impact on international trafﬁcking. As Alesina et al. (2003) discuss, these three indices are correlated, and we will employ them separately in our estimations. The ethnic fragmentation measures were available for various years until 2001, while language and religious fragmentation were measured for the year 2001. Conﬂict measures were collected from the Uppsala Conﬂict Data Program (UCDP)/International Peace Research Institute, Oslo (PRIO) Armed Conﬂict Dataset (2009), and Gleditsch et al. (2002) for the period 1946 to 2001. We employ an average of these measures over this time period in our estimations. We account for various measures of conﬂict in our estimations. First, we use two measures that capture the intensity of conﬂict: (i) the cumulative intensity dummy takes into account the history of the conﬂict. It takes the value 0 if the conﬂict has resulted in less than 1,000 battle-related deaths and 1 otherwise and (ii) the level intensity of conﬂict is measured by distinguishing between either a minor conﬂict or a

700

Randall K. Q. Akee et al.

war (where a minor conﬂict has more than 25 battle-related deaths per year for every year in the period while a war is deﬁned as one where are more than 1,000 battle-related deaths per year for every year in the period). Second, we use a count measure of the number of conﬂicts within a country. Third and ﬁnally, we use a more complex measure that differentiates between the types of conﬂict into four categories. The type 1 variable takes the value 0, 1, 2, and 3 respectively and distinguishes between no conﬂict, minor armed conﬂict, intermediate armed conﬂict, and war at the extra-state level [extra-state conﬂict is a conﬂict over a territory between a government and one or more opposition groups, where the territory is a colony of the government. Also, a minor conﬂict is one where more than 25 battle-related deaths per year for every year in the period has been reported, an intermediate conﬂict is one where more than 25 battle-related deaths per year (but less than 1,000) with a total conﬂict history of more than 1000 battle-related reported deaths while a war is where more than 1000 battle-related deaths has been reported per year for every year in the period]. The type 2 variable takes the value 0, 1, 2 and 3 respectively for no conﬂict, minor armed conﬂict, intermediate armed conﬂict and war at the interstate level (interstate conﬂict is a conﬂict between two or more countries and governments). The type 3 variable takes the value 0, 1, 2, and 3 respectively and separates the conﬂict levels at the internal level (internal conﬂict is conﬂict within a country between a government and one or more opposition groups, with no interference from other countries). The type 4 variable distinguishes whether the conﬂict/war was an internationalized internal one or not, and then at what level (minor armed conﬂict, intermediate armed conﬂict or war. An Internationalized internal conﬂict is similar to internal conﬂict, but where the government, the opposition or both sides receive support from other governments). In addition to the data on trafﬁcking, ethnic/religious/linguistic fragmentation, IDPs/refugees, and conﬂict, we use the following as control variables. First, we use Gross Domestic Product (GDP) for the host and source countries as the refugees and trafﬁcking literature ﬁnds that relatively poorer countries are source countries while richer countries are the host for international refugees and trafﬁcked victims. Second, we use an indicator variable – landlocked – that takes on a value of 1 if a country (trafﬁcking host or source) is landlocked and 0 otherwise. Third, we use a common border and common region dummies for host and source countries of trafﬁcking. These last three variables (landlocked, common border, and common region) are meant to account for the ease with which both refugees and trafﬁcked victims can move between source and host countries, and whether the incidence of refugee migration and trafﬁcking are more likely to be observed within a common region and

Ethnic Conﬂicts and Their Links to Human Trafﬁcking

701

between neighboring countries. The GDP and the landlocked indicator were obtained from the World Bank Development Indicators for the year 2000. The common border and common region measures were self-collected and coded from the Central Intelligence Agency’s World Fact book (2001). Frequently cited in the trafﬁcking literature is the role of (or the lack thereof) legislation surrounding prostitution activities and the enforcement of these laws since most trafﬁcked victims are women and children who are forced into exploitative sexual activities. We use data on legislation surrounding prostitution activities (pimping, pandering, brothels) from the Protection Project Country Report (2002) for the year 2001 and the ‘‘rule of law’’ index from Kaufmann, Kraay, and Zoido-Lobaton (1999a, 1999b) to proxy for corruption and governance for all countries in our dataset. The ‘‘rule of law’’ indicator is a composite index of (i) voice (e.g., freedom of press and the freedom to associate) and accountability, (ii) political stability/lack of violence, (iii) government effectiveness, (iv) regulatory framework, (v) rule of law, and (vi) control of corruption. The index ranges from 3 (worst) to þ3 (best). The reason for using legislation surrounding prostitution as a variable across host and source countries of trafﬁcking stems from the following argument. First, stricter laws surrounding prostitution in source countries makes it an illegal activity in these countries, and one would expect that prostitution commands a higher return in an illegal market rather than a legal one to compensate for the risk of getting caught and punished. However, a higher return from illegal prostitution in the source country allows trafﬁckers and middlemen to extract a higher price from buyers of trafﬁcked victims in the host countries if these earnings from illegal prostitution in the source country constitute the ‘‘reservation’’ price at which trafﬁckers sell their victims in the host country. Similarly, a ban on prostitution in the host countries of trafﬁcking would imply that prostitutes in this illegal activity would earn a higher return than otherwise. Again, this allows for a higher price to be extracted for trafﬁcked victims by the middlemen who sell these victims to the host country buyer. As we show in Akee et al. (2010), a ban on prostitution in either a source or a host country can tilt the incentive in favor of trafﬁckers to move victims across borders for the purpose of prostitution.

3. Empirical methodology and results To determine the link between ethnic conﬂicts and international trafﬁcking, we estimate the direct effect of ethnic fragmentation, various types of external and internal conﬂicts, presence of IDPs/refugees in a source country on the incidence of trafﬁcking between countries.

702

Randall K. Q. Akee et al.

This model is presented below: traffickingij ¼ a1 þ b1 fragi þ b2 frag2i þ b3 refidpi þ b4 conflicti þ b5 Controlsi þ b6 Controlsj þ ij where trafﬁcking is the binary dependent variable for the incidence of trafﬁcking from country i to country j (source i to host j). This variable takes the value 1 if an incidence of trafﬁcking from country i to country j is reported and 0 otherwise. The variable frag measures fragmentation in the source country of trafﬁcking. It is measured continuously from 0 to 1 while frag2 is the squared value of the fragmentation variable. Three measures, ethnic, religious, and language fractionalization, are included in turn in the different regression speciﬁcations. The dummy variable reﬁdp indicates the presence of refugees as well as IDPs in the source country. The variable conﬂict captures the various measures of conﬂict in a source country that we have discussed earlier, and we include these various measures in separate regression speciﬁcations for each fragmentation measure (ethnic, religious, and linguistic). The control variables for the host and source countries include GDP, prohibition of prostitution, rule of law index, common border, common region, and the landlocked dummy. We report the results related to the direct impact of ethnic/religious/ linguistic fragmentation, conﬂict, and IDPs/refugees on the dummy variable capturing the incidence of trafﬁcking between two countries – host and source. Tables 4–6 reports the results, with Table 4 being speciﬁc to the ethnic fragmentation variable while Tables 5 and 6 speciﬁc to the religious and linguistic fragmentation variables, respectively. From Table 4, we start with those results that are robust to all the conﬂict measures used in our regression (cumulative intensity, level intensity, count, types 1–4 and location). We ﬁnd that a lower GDP for source countries and a higher GDP for host countries increases the likelihood of a source–host match for trafﬁcking. This is in consonance with both earlier empirical studies (Akee et al., 2010) and the literature based on government reports and victim surveys that the countries of origin for trafﬁcked victims are relatively poorer as compared to the destination countries. The coefﬁcient on common border and common region are positive and signiﬁcant for both the host and the source countries of trafﬁcking implying that the likelihood of a host–source match for trafﬁcking increases for countries that share a common border or are in a common geographical region. This underscores the fact that lower transportation costs play a signiﬁcant role in explaining the incidence of trafﬁcking between countries. The likelihood to a host–source match for trafﬁcking also increases if a host country is not landlocked – possibly since countries with sea ports allows trafﬁckers to import victims from a wider spectrum of countries.

703

Ethnic Conﬂicts and Their Links to Human Trafﬁcking

Table 4. Direct impact, marginal effects, ethnic fractionalization Probit regressions: direct impact Trafﬁcking incidence (host–source) dependent variable

Fragmentation Ethnic Ethnic squared Refugees/IDPs Refugees/IDPs dummy Conﬂict measures Cumulative intensity Intensity Count Type1 Type2 Type3 Type4 Other controls GDP Source GDP host Prostitution source Prostitution host Rule of law source Rule of law host Landlocked source Landlocked host Common region

(1)

(2)

(3)

(4)

0.043*** [0.016] 0.053*** [0.016]

0.040*** [0.015] 0.051*** [0.016]

0.049*** [0.010] 0.060*** [0.010]

0.043*** [0.009] 0.050*** [0.010]

0.007*** [0.002]

0.007*** [0.002]

0.005*** [0.001]

0.006*** [0.001]

0 [0.002] – – – – – – – – – – – – 0.004*** [0.001] 0.007*** [0.001] 0.003* [0.002] 0.002 [0.002] 0.002 [0.001] 0.003* [0.002] 0.006*** [0.002] 0.008*** [0.002] 0.037*** [0.005]

– – 0.003 [0.003] – – – – – – – – – –

– – – – 0.001 [0.001] – – – – – – – –

– – – – – – 0.015** [0.007] 0.015*** [0.002] 0.004*** [0.001] 0.006*** [0.001]

0.004*** [0.001] 0.007*** [0.001] 0.003* [0.002] 0.002 [0.002] 0.002 [0.001] 0.003* [0.002] 0.006*** [0.002] 0.008*** [0.002] 0.037*** [0.005]

0.003*** [0.001] 0.006*** [0.001] 0.002* [0.001] 0.001 [0.001] 0.003*** [0.001] 0.001 [0.001] 0.002** [0.001] 0.005*** [0.001] 0.024*** [0.003]

0.002*** [0.001] 0.006*** [0.001] 0.001 [0.001] 0.001 [0.001] 0.005*** [0.001] 0.001 [0.001] 0 [0.001] 0.005*** [0.001] 0.024*** [0.003]

704

Randall K. Q. Akee et al.

Table 4. (Continued ) Probit regressions: direct impact Trafﬁcking incidence (host–source) dependent variable (1) Common border Observations

(2) ***

0.111 [0.020] 13912

(3) ***

0.111 [0.020] 13912

(4) ***

0.084 [0.014] 21904

0.079*** [0.013] 21904

Notes: Marginal effects reported. Robust standard errors in brackets. Mean dependent variables, for the different columns: 0.027 for (1) and (2), 0.025 for (3) and (4). Fragmentation, refugees/IDPs and conﬂict measures only at source level while other controls at source and host level. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

As expected, the presence of IDPs/refugees in the source countries increases the likelihood of trafﬁcking in support of the literature that IDPs and refugees are particularly vulnerable to being lured into relocating for employment opportunities and can subsequently be coerced into illicit activities abroad. Table 4 also shows the impact of ethnic fragmentation on the incidence of trafﬁcking under the various conﬂict measures. Higher ethnic fragmentation increases the likelihood of trafﬁcking from a country while the coefﬁcient on the squared term on ethnic fragmentation is negative and signiﬁcant under all the conﬂict measures. This implies that ethnic fragmentation increases the likelihood of trafﬁcking but at a decreasing rate. A possible explanation of this result might be that higher ethnic fragmentation allows middlemen or trafﬁckers to easily target members of different ethnic groups and take advantage of the limited information that potential job seekers have of their credibility. However, as the number of ethnic groups becomes too large, or crosses a critical threshold, middlemen and trafﬁckers may ﬁnd difﬁcult to operate across different groups if the level of mistrust between members of different ethnic groups rises correspondingly. As far as our legislative variable is concerned, a ban on prostitution in the source countries of trafﬁcking is associated with a higher likelihood of trafﬁcking. However, it is weakly signiﬁcant and only for the cumulative intensity, level intensity and count measures of conﬂict. A possible explanation (and as we discussed earlier) might have to do with a higher reservation price that an illegal market for prostitution bestows on a middleman in the source country in bargaining with a potential buyer in the host country of a trafﬁcked victim. A higher likelihood of trafﬁcking is also associated with a source country not being landlocked under our cumulative intensity, level intensity and count measures of conﬂict.

705

Ethnic Conﬂicts and Their Links to Human Trafﬁcking

Table 5.

Direct impact, marginal effects, religious fractionalization Probit regressions: direct impact Trafﬁcking incidence (host–source) dependent variable (1)

Fragmentation Religion Religion squared Refugees/IDPs Refugees/IDPs dummy Conﬂict measures Cumulative intensity Intensity Count Type1 Type2 Type3 Type4 Other controls GDP source GDP host Prostitution source Prostitution host Rule of law source Rule of law host Landlocked source Landlocked host Common region

(2)

(3)

(4)

0.059*** [0.013]

0.060*** [0.013]

0.055*** [0.010]

0.045*** [0.010]

0.073*** [0.016]

0.074*** [0.016]

0.062*** [0.011]

0.048*** [0.011]

0.008*** [0.002]

0.008*** [0.002]

0.005*** [0.001]

0.005*** [0.001]

0.002 [0.002] – – – – – – – – – – – –

– – 0.005* [0.003] – – – – – – – – – –

– – – – 0.001 [0.002] – – – – – – – –

– – – – – – 0.020*** [0.007] 0.016*** [0.002] 0.003** [0.001] 0.007*** [0.001]

0.003*** [0.001] 0.007*** [0.001] 0.005*** [0.002] 0.002 [0.002] 0.001 [0.001] 0.003* [0.002] 0.005*** [0.002] 0.008*** [0.002] 0.036*** [0.005]

0.003*** [0.001] 0.007*** [0.001] 0.005*** [0.002] 0.002 [0.002] 0.001 [0.001] 0.003* [0.002] 0.005*** [0.002] 0.008*** [0.002] 0.036*** [0.005]

0.003*** [0.001] 0.007*** [0.001] 0.004*** [0.001] 0.001 [0.001] 0.003*** [0.001] 0 [0.001] 0.002 [0.001] 0.006*** [0.001] 0.024*** [0.003]

0.002*** [0.001] 0.006*** [0.001] 0.002** [0.001] 0.001 [0.001] 0.006*** [0.001] 0 [0.001] 0.001 [0.001] 0.005*** [0.001] 0.024*** [0.003]

706

Randall K. Q. Akee et al.

Table 5. (Continued ) Probit regressions: direct impact Trafﬁcking incidence (host–source) dependent variable (1) Common border Observations

(2) ***

0.111 [0.020] 14060

(3) ***

0.111 [0.020] 14060

(4) ***

0.086 [0.014] 22052

0.080*** [0.013] 22052

Notes: Marginal effects reported. Robust standard errors in brackets.Mean dependent variables, for the different columns: 0.027 for (1) and (2), 0.024 for (3) and (4). Fragmentation, refugees/idps and conﬂict measures only at source level while other controls at source and host level. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

Again existence of sea ports facilitates easier export of trafﬁcked victims from a source country. The rule of law index is negative and signiﬁcant for source countries but again only for the regression speciﬁcations associated with the count and types 1–4 measures of conﬂict. As one might expect, higher corruption levels and weaker governance structures in poorer countries are likely to lead to these countries becoming origins for trafﬁcked victims. Turning to the various conﬂict measures, we ﬁnd that the cumulative intensity, level intensity, and count measure of conﬂict are insigniﬁcant predictors of trafﬁcking incidence. However, types 1–4 measures of conﬂict are signiﬁcant, with types 1, 3, 4 reducing while type 2 increasing the incidence of trafﬁcking. Recall that type 1 is extra-state conﬂict (conﬂict within a territory with a colonial power and opposition within the colony as adversaries), type 3 is intra-state (between a government and opposition forces with no help from outside governments), type 4 is internationalized intra-state conﬂict (between a government and opposition forces with help from outside governments for either parties), and type 2 inter-state conﬂict (between two countries). Thus, types 1, 3, and 4 can be categorized as internal conﬂicts while type 2 is an external conﬂict. Our results therefore indicate that internal conﬂicts reduce the likelihood of trafﬁcking while external conﬂicts exacerbate the problem. This result, while paradoxical, may be because internal conﬂicts are more likely to disrupt internal transportation networks in a country that impedes the movement of trafﬁckers and potential victims alike. Table 5 reports the results when we use religious fragmentation along with the different measures of conﬂict to predict the incidence of trafﬁcking. Once again, irrespective of the measure of conﬂict used,

707

Ethnic Conﬂicts and Their Links to Human Trafﬁcking

Table 6.

Direct impact, marginal effects, language fractionalization Trafﬁcking incidence (host–source) dependent variable

Fragmentation Language Language squared Refugees/IDPs Refugees/IDPs dummy Conﬂict measures Cumulative intensity Intensity Count Type1 Type2 Type3 Type4 Other controls GDP source GDP host Prostitution source Prostitution host Rule of law source Rule of law host Landlocked source Landlocked host Common region

(1)

(2)

(3)

(4)

0.005 [0.012] 0.009 [0.013]

0.003 [0.012] 0.008 [0.013]

0.019** [0.008] 0.024*** [0.009]

0.014* [0.007] 0.017** [0.008]

0.010*** [0.002]

0.010*** [0.002]

0.007*** [0.002]

0.007*** [0.001]

0.001 [0.002] – – – – – – – – – – – –

– – 0.005 [0.003] – – – – – – – – – –

– – – – 0.001 [0.002] – – – – – – – –

– – – – – – 0.021*** [0.007] 0.016*** [0.002] 0.003** [0.001] 0.007*** [0.001]

0.003*** [0.001] 0.007*** [0.001] 0.004** [0.002] 0.002 [0.002] 0.002 [0.002] 0.003* [0.002] 0.004** [0.002] 0.008*** [0.002] 0.038*** [0.005]

0.003*** [0.001] 0.007*** [0.001] 0.004** [0.002] 0.002 [0.002] 0.002 [0.002] 0.003* [0.002] 0.004* [0.002] 0.008*** [0.002] 0.038*** [0.005]

0.003*** [0.001] 0.007*** [0.001] 0.003** [0.001] 0.001 [0.001] 0.004*** [0.001] 0 [0.001] 0.001 [0.001] 0.006*** [0.001] 0.025*** [0.003]

0.002*** [0.001] 0.006*** [0.001] 0.001 [0.001] 0.001 [0.001] 0.006*** [0.001] 0 [0.001] 0.001 [0.001] 0.005*** [0.001] 0.024*** [0.003]

708

Randall K. Q. Akee et al.

Table 6. (Continued ) Trafﬁcking incidence (host–source) dependent variable (1) Common border Observations

0.107*** [0.020] 13616

(2) 0.107*** [0.020] 13616

(3) 0.083*** [0.014] 21460

(4) 0.077*** [0.013] 21460

Notes: Marginal effects reported. Robust standard errors in brackets. Mean dependent variables, for the different columns: 0.028 for (1) and (2), 0.025 for (3) and (4). Fragmentation, refugees/IDPs and conﬂict measures only at source level while other controls at source and host level. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

a lower GDP for a source and a higher GDP for the host country leads to higher likelihood of a host–source trafﬁcking match. The presence of IDPs/refugees in the source country, common border, common region, and the host country not being landlocked all positively affect the likelihood of trafﬁcking. Akin to the case of ethnic fragmentation, religious fragmentation in the source country has a positive but decreasing impact on the likelihood of trafﬁcking under all measures of conﬂict. The rule of law index for source countries is negative and signiﬁcant for source countries but only for the count and types 1–4 measures of conﬂict while a ban on prostitution in the source country is positively and signiﬁcantly related to the likelihood of trafﬁcking for the cumulative intensity, level intensity and count measures of conﬂict. Finally, internal conﬂicts (types 1, 3, and 4) decreases, while external conﬂict (type 2) in a source country increases the likelihood of trafﬁcking. In this respect, the direct impacts of religious fragmentation are very similar to those for ethnic fragmentation. Lastly, we use the linguistic fragmentation along with the various measures of conﬂict to see if linguistic fragmentation plays any role in predicting the likelihood of trafﬁcking between countries. As Table 6 reports, similar to the case for ethnic and religious fragmentation (and irrespective of the measure of conﬂict used), a lower GDP for a source and a higher GDP for the host country, presence of IDPs/refugees in the source country, common border, common region and a non-landlocked host all increase the likelihood of a source–host trafﬁcking match. A lower rule of law in the source country has a positive and signiﬁcant effect on trafﬁcking but only under the count and types 1–4 measure of conﬂict while a ban on prostitution in the source country increases the likelihood of trafﬁcking under the cumulative intensity, level intensity and count measures of conﬂict. However, the case of ethnic and religious fragmentation, linguistic fragmentation has a positive and signiﬁcant impact on the likelihood of

Ethnic Conﬂicts and Their Links to Human Trafﬁcking

709

trafﬁcking only under the count and types 1–4 measures of conﬂict. Nevertheless, once again internal conﬂicts (types 1, 3 and 4) decreases, while external conﬂict (type 2) in a source country increases the likelihood of trafﬁcking. To summarize, the direct impacts on the incidence of trafﬁcking are (i) lower GDP in source countries, (ii) higher GDP in host countries, (iii) presence of IDPs/refugees in the source country, (iv) common border, (v) common region, (vi) a non-landlocked host country, (vii) a lower rule of law for source countries, and (viii) external conﬂicts in source countries are all positive and signiﬁcant predictors of a source–host trafﬁcking match irrespective of whether we use ethnic, religious or linguistic measures of fragmentation. Meanwhile, internal conﬂicts are negative but signiﬁcant predictors of trafﬁcking under for all three fragmentation indices (ethnic, religious and linguistic). While linguistic fragmentation is positive and signiﬁcant predictor of trafﬁcking likelihood only under the count and types 1–4 measures of conﬂict, ethnic and religious fragmentation both increases the likelihood of trafﬁcking under all measures of conﬂict. 4. Conclusion This chapter is a ﬁrst to explore the nexus between ethnic/religious/ fragmentation, different types of conﬂict, the presence of IDPs and refugees, and the incidence of trafﬁcking. Speciﬁcally, we relate international trafﬁcking with two separate literatures that studies the link between ethnic/religious fragmentation and conﬂict on one hand, and the link between conﬂict and IDPs/refugee on the other to show that ethnic and religious fragmentation, along with various measures of conﬂict and the presence of IDPs and refugees in a country, is signiﬁcant predictors of the likelihood of trafﬁcking amongst countries. However, our results in the chapter, specially, those that relate to the various measures of conﬂict are sometimes paradoxical and require further analysis by either regrouping or recoding the various types of conﬂict that might be overlap with each other, or employing a different estimation strategy to eliminate possible collinearity between the fragmentation, conﬂict, and the IDPs/refugees variables. To partially account for this latter problem we run probit regressions with the IDPs/refugees as a dependant variable and fragmentation, conﬂict measures and controls for the source countries. The results are reported in the Appendix. Appendix The results related to the measures of conﬂict (specially, internal ones) are paradoxical at ﬁrst glance. More often than not, ethnic fragmentation, conﬂict and IDPs/refugees are highly correlated and while estimating the

710

Randall K. Q. Akee et al.

direct effects, multicollinearity among these three variables might well be at play. To partially account for the collinearity problem, we turn to the results from a probit regression below. refidpi ¼ a1 þ b1 fragi þ b2 frag2i þ b3 conflicti þ b4 controlsi þ i In the above equation, the dependent binary dummy variable is now the incidence of refugees and internally displaced at source country level while the explanatory variables are the measures of fragmentation, conﬂict, and controls at the source country level. We report the regression results for the ethnic, religious and linguistic fragmentation below. Overall our results suggest that there is a relationship between the different fragmentation variables and the presence of refugees and IDPs in our data. We also ﬁnd evidence that there is some relationship between the presence of refugees or IDPs and other variables of interest. This high level of correlation may explain the relatively large standard errors for certain coefﬁcients in Tables 4–6. Table 7 reports the results for ethnic fragmentation. A lower GDP and a lower rule of law index are associated with the presence of IDPs and refugees under all the conﬂict measures. A ban on prostitution has a positive impact on IDPs and refugees under the count and types 1–4 measures of conﬂict while a non-landlocked country has an increased likelihood of the presence of IDPs and refugees under the cumulative intensity and level intensity measures of conﬂict. Ethnic fragmentation plays a positive but decreasing role under the cumulative and level intensity of conﬂict but a negative and increasing role under the count and types 1–4 measures of conﬂict in explaining the presence of IDPs and refugees. Cumulative intensity, level intensity, count and types 3 and 4 (internal conﬂicts) are positive and signiﬁcant while types 1 (internal conﬂict within a colony) and 2 (inter-state conﬂicts) are negative and signiﬁcant predictors of the presence of IDPs and refugees in a country. Next we use religious fragmentation as a predictor of IDPs/refugees along with the different conﬂict measures and controls for the source countries of trafﬁcking. As Table 8 shows religious fragmentation is a positive and signiﬁcant predictor of the presence of IDPs/refugees in a source country of trafﬁcking for all measures of conﬂict. A lower GDP, a lower rule of law index as well as legislations banning prostitution positively increase the likelihood of the presence of IDPs and refugees in source countries of trafﬁcking. A nonlandlocked country has an increased likelihood of the presence of IDPs and refugees under the cumulative intensity and level intensity measures of conﬂict but a decreased likelihood under the count measure of conﬂict. Similar to the case of ethnic fragmentation, cumulative intensity, level intensity, count and types 3 and 4 (internal conﬂicts) are positive and signiﬁcant while types 1 (internal

711

Ethnic Conﬂicts and Their Links to Human Trafﬁcking

Table 7.

Probit regressions, marginal effects, ethnic fractionalization Probit regressions Refugees/IDPs dummy dependent variable

Fragmentation Ethnic Ethnic squared Conﬂict measures Cumulative intensity Intensity Count Type1 Type2 Type3 Type4 Other controls GDP Prostitution Rule of law Landlocked Observations

(1)

(2)

(3)

(4)

0.660*** [0.075] 1.077*** [0.078]

0.701*** [0.076] 1.092*** [0.079]

0.477*** [0.056] 0.032 [0.059]

0.891*** [0.058] 0.464*** [0.061]

0.368*** [0.011] – – – – – – – – – – – –

– – 0.378*** [0.016] – – – – – – – – – –

– – – – 0.601*** [0.010] – – – – – – – –

– – – – – – 0.401*** [0.028] 0.138*** [0.016] 0.510*** [0.013] 0.446*** [0.012]

0.172*** [0.005] 0.016* [0.008] 0.170*** [0.008] 0.114*** [0.012]

0.170*** [0.005] 0.007 [0.008] 0.194*** [0.008] 0.097*** [0.012]

0.162*** [0.004] 0.111*** [0.007] 0.196*** [0.007] 0.004 [0.010]

0.153*** [0.004] 0.137*** [0.007] 0.155*** [0.007] 0.011 [0.010]

17484

17484

27528

27528

Notes: Marginal effects reported. Robust standard errors in brackets. Mean dependent variables, for the different columns: 0.585 for (1) and (2), 0.426 for (3) and (4). Only source country side. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

conﬂict within a colony) and 2 (inter-state conﬂicts) are negative and signiﬁcant predictors of the presence of IDPs and refugees in a country. Finally, we ﬁnd that the results for linguistic fragmentation (Table 9) are very similar to those for religious fragmentation above. Linguistic

712

Table 8.

Randall K. Q. Akee et al.

Probit regressions, marginal effects, religious fractionalization Refugees/IDPs dummy dependent variable

Fragmentation Religion Religion squared Conﬂict measures Cumulative intensity Intensity Count Type1 Type2 Type3 Type4 Other controls GDP Prostitution Rule of law Landlocked Observations

(1)

(2)

(3)

(4)

1.873*** [0.075] 1.994*** [0.088]

1.878*** [0.077] 2.067*** [0.089]

1.972*** [0.062] 2.449*** [0.072]

2.028*** [0.063] 2.457*** [0.073]

0.362*** [0.011] – – – – – – – – – – – –

– – 0.326*** [0.014] – – – – – – – – – –

– – – – 0.618*** [0.011] – – – – – – – –

– – – – – – 0.129*** [0.027] 0.075*** [0.017] 0.491*** [0.015] 0.434*** [0.013]

0.126*** [0.004] 0.067*** [0.008] 0.162*** [0.007] 0.085*** [0.012]

0.123*** [0.004] 0.056*** [0.009] 0.185*** [0.007] 0.068*** [0.012]

0.134*** [0.004] 0.146*** [0.007] 0.162*** [0.007] 0.021** [0.009]

0.130*** [0.004] 0.162*** [0.007] 0.127*** [0.007] 0.001 [0.010]

17670

17670

27714

27714

Notes: Marginal effects reported. Robust standard errors in brackets. Mean dependent variables, for the different columns: 0.579 for (1) and (2), 0.423 for (3) and (4). Only source country side. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

fragmentation is a positive and signiﬁcant predictor of the presence of IDPs/refugees in a source country of trafﬁcking for all measures of conﬂict. Similar to the case of ethnic and religious fragmentation, cumulative intensity, level intensity, count and types 3 and 4 (internal conﬂicts) are positive and signiﬁcant while types 1 (internal conﬂict within a colony) and 2 (inter-state conﬂicts) are negative and signiﬁcant

713

Ethnic Conﬂicts and Their Links to Human Trafﬁcking

Table 9.

Probit regressions, marginal effects, language fractionalization Probit regressions Refugees/IDPs dummy dependent variable

Fragmentation Language Language squared Conﬂict measures Cumulative intensity Intensity Count Type1 Type2 Type3 Type4 Other controls GDP Prostitution Rule of law Landlocked Observations

(1)

(2)

(3)

(4)

1.361*** [0.058] 1.258*** [0.064]

1.641*** [0.060] 1.468*** [0.065]

0.457*** [0.051] 0.525*** [0.055]

0.211*** [0.052] 0.238*** [0.057]

0.400*** [0.012] – – –– – – – – – – – – –

– – 0.503*** [0.016] – – – – – – – – – –

– – – – 0.587*** [0.010] – – – – – – – –

– – – – – – 0.364*** [0.026] 0.063*** [0.016] 0.482*** [0.013] 0.389*** [0.012]

0.121*** [0.005] 0.014 [0.009] 0.149*** [0.008] 0.120*** [0.013]

0.117*** [0.005] 0.01 [0.009] 0.179*** [0.008] 0.131*** [0.012]

0.125*** [0.004] 0.116*** [0.007] 0.150*** [0.007] 0.039*** [0.010]

0.116*** [0.004] 0.133*** [0.007] 0.105*** [0.007] 0.033*** [0.010]

17112

17112

26970

26970

Notes: Marginal effects reported. Robust standard errors in brackets. Mean dependent variables, for the different columns: 0.565 for (1) and (2), 0.414 for (3) and (4). Only source country side. * Signiﬁcant at 10%. ** Signiﬁcant at 5%. *** Signiﬁcant at 1%.

predictors of the presence of IDPs and refugees in a country. A lower GDP and a lower rule of law index positively increase the likelihood of the presence of IDPs and refugees in source countries of trafﬁcking. A nonlandlocked country has an increased likelihood of the presence of

714

Randall K. Q. Akee et al.

IDPs and refugees under the cumulative intensity and level intensity measures of conﬂict but a decreased likelihood under the count and types 1–4 measures of conﬂict. Finally a ban on prostitution has a positive inﬂuence on the presence of IDPs and refugees only under the count and types 1–4 measures of conﬂict. In effect, our results show that ethnic, religious and linguistic fragmentation, along with our various measures of internal conﬂicts turn out to be strong predictors of the presence of IDPs and refugees. We acknowledge that this may have increased the standard errors in our main regressions of interest which may explain why some coefﬁcients do not achieve statistical signiﬁcance. Acknowledgment Financial support from the Alexander von Humboldt Foundation (Transcoop Grant) is gratefully acknowledged. The usual disclaimer applies. References Akee, R., Basu, A.K., Bedi, A., Chau, N.H. (2009), Combating trafﬁcking in women and children: a review of international and national legislation, co-ordination failures and perverse economic incentives. The Protection Project Journal of Human Rights and Civil Society 2, 1–24. Akee, R., Basu, A.K., Bedi, A., Chau, N. H. (2010), Transnational trafﬁcking, law enforcement and victim protection: a middleman’s perspective. Mimeo, Department of Applied Economics and Management, Cornell University. Alesina, A., Baqir, R., Easterly, W. (1999), Public goods and ethnic divisions. Quarterly Journal of Economics 114, 1243–1284. Alesina, A., Devleeschauwer, A., Easterly, W., Kurlat, S., Wacziarg, R. (2003), Fractionalization. Journal of Economics Growth 8, 155–194. Bharadwaj, N., Dhungana, S.K., Hicks, N., Crozier, R., Watson, C. (2007), Nepal at a Crossroads: The Nexus between Human Security and Renewed Conﬂict in Rural Nepal. Friends for Peace and International Alert, Kathmandu, Nepal. Carling, J. (2005), Trafﬁcking in women from Nigeria to Europe. Migration Information Source. Available at http://www.migrationinformation.org/ Feature/display.cfm?id=318 Castles, S., Loughna, S. (2003), Trends in asylum migration to industrialized countries: 1990–2001. Discussion Paper No. 2003/31, World Institute of Development Research (WIDER), Helsinki. Central Intelligence Agency (CIA) (2001), The world factbook. Available at http://www.faqs.org/docs/factbook/index.html

Ethnic Conﬂicts and Their Links to Human Trafﬁcking

715

Connor, W. (1983), Beyond reason: the nature of the ethnonational bond. Ethnic and Racial Studies 16, 373–389. Ehindero, S., Awolowo, O., Idemudia, P. (2006), Baseline Study on Forced Labour and Human Trafﬁcking in Kwara, Kano, Cross Rivers and Lagos State in Nigeria. ILO/PATWA Ofﬁces in Nigeria, Ghana, Sierra Leone and Liberia, Abuja. Gleditsch, N.P., Wallensteen, P., Eriksson, M., Sollenberg, M., Strand, H. (2002), Armed conﬂict 1946–2001: a new dataset. Journal of Peace Research 39, 615–637. Global IDP Project. (2003), A global overview of internal displacement. Available at http://www.idpproject.org/global_overview.htm Gurr, T. (1993), Minorities at Risk: A Global View of Ethnopolitical Conﬂicts. United States Institute for Peace, Washington, DC. de Haas, H. (2006), International migration and national development: viewpoints and policy initiatives in countries of origin. The case of Nigeria. Working Papers Migration and Development Series, Report 6, International Migration Institute, University of Oxford, UK. Horowitz, D.L. (1985), Ethnic Groups in Conﬂict. University of California Press, Berkeley, CA. Ibeanu, O. (1999), Exiles in their own home: conﬂicts and internal population displacement in Nigeria. Journal of Refugees Studies 12 (2). Internal Displacement Monitoring Centre. (2006), Nepal: IDP return still a trickle despite ceaseﬁre. Available at http://www.internal-displacement. org/8025708F004BE3B1/(httpInfoFiles)/DDDDE17440C134D7C12572 0900397F78/$ﬁle/Nepal%20overview%2016oct%202006.pdf. Retrieved on March 1, 2009. Kaufmann, D., Kraay, A., Zoido-Lobaton, P. (1999a), Aggregating governance indicators. World Bank Policy Research Department Working Paper No. 2195. Kaufmann, D., Kraay, A., Zoido-Lobaton, P. (1999b), Governance matters. World Bank Policy Research Department Working Paper No. 2196. Lee, C., Lindstro¨m, R., Moore, W.H., Turan, K. (2004), Ethnicity and repression: the ethnic composition of countries and human rights violations. In: Carey, S.C., Poe, S.C., (Eds.), The Systematic Study of Human Rights. Ashgate, Burlington, VT. Lindstro¨m, R., Moore, W.H. (1995), Deprived, rational or both? ‘‘Why minorities rebel’’ revisited. Journal of Political and Military Sociology 23, 167–190. Mani, D. (2005), Strengthening Decentralized Governance for Human Security. United Nations Centre for Regional Development (UNCRD), Nagoya, Japan, http://unpan1.un.org/intradoc/groups/public/documents/ UN/UNPAN020337.pdf Protection Project (2002), Human Rights Report on Trafﬁcking in Persons, Especially Women and Children. School of Advanced International Studies (SAIS), The Johns Hopkins University, Washington, DC.

716

Randall K. Q. Akee et al.

Reilly, B. (2000), Democracy, ethnic fragmentation and internal conﬂict: confused theories, faulty data and the ‘‘crucial case’’ of Papua New Guinea. International Security 25, 162–185. Spiegel, P.B. (2004), HIV/AIDS among conﬂict-affected and displaced populations: dispelling myths and taking action. Disasters 28, 322–339. Tamang, R. (2009), Internally displaced persons in Nepal: neglected and vulnerable. Asian Social Science 5, 3–11, http://www.ccsenet.org/ journal.html UCDP/PRIO (2009), UCDP/PRIO armed conﬂict dataset. Version 4-2009. Available at www.ucdp.uu.se/database. Retrieved on February 1, 2010. UNHCR (1995), The State of the World’s Refugees: In Search of Solutions. Oxford University Press, Oxford. UNHCR (2000), The State of the World’s Refugees: Fifty Years of Humanitarian Action. Oxford University Press, Oxford. UNHCR. (2002), UNHCR statistical yearbook 2001. Available at http:// www.unhcr.org/4a02e3406.html UNHCR (2003), 2002 statistics on asylum-seekers, refugees and others of concern to UNHCR. Geneva. United Nations Ofﬁce for the Coordination of Humanitarian Affairs (UNOCHA) (2008), Nepal situation overview. Available at http:// ochaonline.un.org/OchaLinkClick.aspx?link=ocha&docId=1061738 United States Department of State (2002), Trafﬁcking in persons report. Ofﬁce of the Under-Secretary for Global Affairs. United States Department of State Publication. World Bank. Various Issues. World Development Indicators. CD-Rom. Washington DC: The World Bank. World Bank (2004), World Development Indicators CD-ROM. The World Bank, Washington, DC. Zlotnik, H. (1999), Trends of international migration since 1965: what existing data reveal. International Migration 37 (1), 21–62.

AUTHOR INDEX

Abowd, 171, 668 Abraham, 181 Abrebo, 521 Abuja, 715 Adams, 12, 16 Adler, 499, 514 Adrangi, 547 Ahearne, 107 Aigner, 206 Aizenman, 520 Akee, 15–16, 691–692, 694–696, 698, 700–702, 704, 706, 708, 710, 712, 714, 716 Akerlof, 208, 679, 688 Alba, 138–139, 171–172 Alber, 653, 678, 688 Alesina, 9, 16, 194, 202, 326, 337, 693, 697–699, 714 Alexander, 304, 714 Allensteen, 715 Altonji, 8, 16, 181, 193, 202, 206, 234, 326, 337, 342, 355 Amemiya, 663 Amuedo-Dorantes, 520 Anas, 9, 16, 194, 202, 326, 337 Angel, 451 Antecol, 83 Arango, 515 Arellano, 126–127 Arnold, 360, 372 Aronson, 46, 66 Arrow, 206, 233 Ashenfelter, 16, 202, 337, 355 Aslund, 138, 140, 171 Attanasio, 130, 134 Attias-Donfut, 575 Azariadis, 182 Azzarri, 477 Baldwin-Grossman, 6, 16 Banerjee, 28, 43, 110, 113, 126 Baqir, 714 Barjaba, 472–473, 485 Bartel, 2, 16, 27, 43, 138–140, 159, 167, 171

Basu, 16, 714 Battisti, 14, 16, 579–580, 582, 584, 586, 588, 590, 592, 594, 596, 598, 600 Bauer, 2, 4, 8, 16–17, 27, 43, 138–141, 171, 178, 180–181, 193, 202, 326, 337, 342, 355, 376, 655 Bautista, 540 Baye, 330–331, 336–337 Beach, 237 Beaudry, 182 Becker, 180, 206, 231, 233, 241, 261 Bedi, 714 Behringer, 414 Benhabib, 15, 17, 606, 612 Berger, 297 Bernheim, 613, 679, 688 Betts, 68–69, 78 Bhagwati, 341, 355 Bharadwaj, 696, 714 Bhattacharya, 46, 65 Bhaumik, 8, 17, 193, 202, 326, 337, 342, 355 Bikhchandani, 28, 32, 43 Bisin, 9, 17, 326, 337 Blalock, 206 Blanchard, 135 Blank, 8, 16, 193, 202, 206, 234, 326, 337, 342, 355 Blau, 8, 17, 193, 202, 326, 337–338, 342, 355 Blinder, 236, 257–258, 598, 668, 671 Bock, 531 Bodenhorn, 9, 17, 295–296, 298, 300, 302, 304, 306, 308, 310, 312, 314, 316, 318 Bodvarsson, 6–7, 17, 232–234, 236–238, 240, 242, 244, 246, 248, 250, 252, 254, 256, 258, 260, 262 Boeri, 609, 619 Bond, 126–127, 519, 715 Borjas, 6, 17, 138, 140, 146, 171, 178, 180–181, 206, 232, 272, 290, 364, 371, 379, 468, 470, 477, 488–489, 493, 498, 514, 619, 678, 688 Boss, 519 Bourguignon, 46, 65

718

Author Index

Bratsberg, 68–69, 78, 85, 468, 470, 477, 488–489, 493 Brezis, 35, 43 Briggs, 613 Brown, 206, 544 Browning, 554 Bucci, 235, 237 Buhai, 208 Bump, 515 Burkhauser, 414 Burr, 448, 451 Bursik, 299 Cain, 206, 234 Calomiris, 305 Capella, 360, 372 Cappellari, 480 Card, 6, 16–17, 71, 79, 138–139, 146, 171, 202, 285, 290, 337, 355, 654 Carletto, 477 Carling, 696, 714 Carolyn, 295 Carrington, 2, 17, 33, 43 Carroll, 547 Carter, 297 Casella, 358, 373 Castano, 499, 514 Castles, 694–695, 714 Chakravarty, 49, 65 Charvet, 171 Chau, 16, 691, 714 Chavez, 173 Chiancone, 531 Chiappori, 182 Chiswick, 2, 4, 8, 17, 25, 43, 68–72, 74, 76, 78–80, 82, 84, 86–88, 90–93, 139–140, 171, 178, 180, 186, 193, 202, 206, 232, 235–236, 271–272, 276, 290–291, 326, 338, 342, 355, 364, 372, 379, 402, 413, 522, 582, 608, 621 Choi, 33, 43 Church, 2, 18, 25, 44, 384 Citrin, 608 Clark, 106 Cline, 120 Coate, 206 Cobb-Clark, 646, 102 Cohen Goldner, 13 Cohen, 13, 18, 27, 44, 271, 274, 276–278, 291, 330, 338, 447–448, 450, 452, 454, 456, 458, 460, 462, 464, 499, 514 Conley, 207 Connor, 692, 715

Constant, 8, 12, 18, 21, 80, 84, 88–89, 112, 115, 119, 126, 130, 139, 161, 179, 184, 193, 203, 209–210, 239, 254, 256, 277, 281, 285–286, 288, 326, 338, 342, 348, 352–353, 355, 361, 365, 416, 434, 442, 452, 457, 459, 462, 468, 479, 482, 484–485, 490–493, 507–509, 547, 566, 568, 572, 589, 595–597, 612, 635, 662–665, 668, 671–673, 697 Cornell, 691, 714 Cotton, 259–260 Courant, 111 Cowell, 46, 65 Cox, 12, 18, 544, 570 Crowder, 173 Crozier, 714 Dagum, 46, 51, 65 Dalipaj, 495 D’Amuri, 678, 688 Daneshvary, 237 Davis, 129, 368, 417, 442 Dayton-Johnson, 472 Dean, 586–587 Deardorff, 111, 365, 372 Deaton, 546, 548 Debry, 521 Decressin, 130, 134 Delgado, 108 DeNew, 654 Denny, 264, 266 Detragiache, 17, 43 Deutsch, 4, 8, 18, 45–46, 48, 50, 52–54, 56, 58, 60, 62, 64–66, 193, 203, 326, 338, 402, 413 Devereux, 402, 413 Devleeschauwer, 714 DeVoretz, 14, 16, 18, 543–544, 546, 548, 550, 552, 554, 556, 558, 560, 562, 564, 566, 568, 570, 572, 579–588, 590, 592, 594, 596, 598, 600 Dhungana, 714 Didukh, 554, 558 Dietz, 379, 387 Diewert, 238 Dimova, 13, 18 DiNardo, 182 Disdier, 366, 372 Djajic, 111, 468, 470, 487 Dobson, 404, 413 Docquier, 544, 624, 626, 628 Dornbusch, 135 Douglas, 602, 226 Du Bois, 311

Author Index Duleep, 180 Duncan, 414 Dunlevy, 358, 364, 372 Dunn, 499, 514 Durand, 138, 171, 499, 514 Dustmann, 9, 18, 179–181, 194, 203, 272, 291, 326, 338, 379–380, 391, 468, 470, 487, 608, 611, 651, 655, 658 Easterlin, 111 Easterly, 714 Edmonston, 138, 172 Ehindero, 696, 715 Eliott, 298 Ellingsen, 336–338 Elliott, 545 Epstein, 2–4, 6–14, 16–21, 26–30, 32, 34, 36, 38, 40, 42–44, 171, 193–194, 196–198, 200, 202–203, 326–328, 330, 332, 334, 336–338, 342, 355, 413, 459, 471, 481, 486, 499, 501, 514–515, 613 Eriksson, 715 Eser, 574 Espenshade, 271, 291, 608 Espinosa, 468, 486 Esses, 678, 683, 688 Esteban, 45, 49, 66 Even, 5, 34, 36, 38, 40, 79, 105–109, 111–112, 114, 119–121, 126–127, 129–130, 138, 152, 161, 166, 178–179, 181, 183, 185–186, 197, 206, 217, 223, 238, 257–258, 300–303, 308, 311, 315, 320, 326, 330, 353, 377, 381, 385, 389, 391–392, 405, 411–412, 418, 421, 423, 425, 428, 432, 434, 438, 447, 450, 468, 486, 488–489, 522, 525, 527, 544, 563, 586–588, 611, 615, 631, 636, 654, 659, 668, 677–679, 693 Fabbri, 18, 203, 338 Facchini, 15, 19, 605–614, 616, 618, 620, 622, 630–632, 634, 636 Faini, 5, 19, 106, 108, 110–112, 114, 116, 118, 120–122, 124, 126, 128, 487 Fan, 237, 342–343, 356 Farber, 181–182 Farrell, 317 Fatas, 130, 134 Fedewa, 515 Feenstra, 367, 372 Feick, 207 Ferenczi, 106 Ferrer, 584

719

Fertig, 10, 19, 178, 180, 375–376, 378, 380, 382, 384, 386, 388, 390, 654 Fertig, 10, 19, 178, 180, 375–376, 378, 380, 382, 384, 386, 388, 390, 654 Fijalkowski, 399 Fischler, 360, 372 Fishlow, 305 Fitch, 322 Florin, 467–468, 470, 472, 474, 476, 478, 480, 482, 484, 486, 488, 490, 492, 543–544, 546, 548, 550, 552, 554, 556, 558, 560, 562, 564, 566, 568, 570, 572 Foster, 106, 207, 530, 622, 696 Freeman, 171, 606–607, 630 French, 111, 126, 235, 287, 589, 651 Friedberg, 6, 19, 342, 356 Friedman, 302 Frisbie, 206 Fudenberg, 207 Fuess, 233, 237–238 Fuguitt, 206 Fujii, 235–236 Furstenberg, 303 Gabriel, 236–237 Gallman, 304 Gang, 1–2, 4, 6, 8–12, 14–21, 40, 43–44, 171, 193–194, 197, 202–203, 325–328, 330, 332, 334, 336–338, 342, 355–356, 401–404, 406, 408, 410, 412–414, 525, 527, 649–650, 652, 654–656, 658, 660, 662, 664, 666, 668, 670, 672 Gartner, 321 Gataullina, 18, 203, 338, 355 Gaytan-Fregoso, 345, 356 Geiger, 558 Gertler, 402, 414 Gibbons, 178, 182 Gindling, 14, 21, 517–522, 524, 526, 528, 530, 532 Gladwell, 207 Glaeser, 207, 300, 314 Gleditsch, 697–699, 715 Glewwe, 402, 414 Glick, 448 Goeken, 322 Gonzalez, 521 Gorodzeisky, 498–499, 501, 515 Gottlieb, 2, 19, 25, 44 Gould, 122, 291, 358–359, 372, 480 Goyal, 207 Goyette, 237 Gradstein, 197, 203 Grant, 182, 206, 326–336, 515, 714

720

Author Index

Gray, 545 Green, 20, 519–521, 553 Greenwood, 139–140, 151, 166, 171–172, 364, 372 Griever, 133 Grossman, 2, 6, 16, 20, 25, 44, 206 Guiso, 108 Gurak, 140–141, 172 Gurr, 296, 692, 715 Gutierrez, 653 Haig, 236 Haines, 321 Haisken-DeNew, 189 Haiyan, 372 Hall, 649 Halter, 358, 373 Ham, 668 Hamermesh, 206 Handlin, 298–299, 314, 317 Hanoka, 65 Hansen, 124, 499, 514, 548, 553 Hanushek, 69–72, 75–80, 84–85, 87, 89–90, 92 Harmon, 266 Harris, 110, 179, 182–183, 185, 187 Hartog, 68, 70 Hatt, 322 Hatton, 106, 110, 122, 342, 356, 608 Hatzipanayotou, 346, 356 Haveman, 402, 414 Hayfron, 235, 237 Head, 4, 6, 259, 297, 358–359, 366, 372–373, 408, 431, 439–440, 450, 453–458, 460–462, 472, 501–502, 504, 507–510, 544, 549–551, 569, 659, 661–664, 670 Heckman, 208, 274, 278–280, 595, 597, 599 Heer, 206 Helpman, 365, 372 Helweg, 499–500, 514 Hempstead, 608 Herander, 358, 373 Hicks, 210, 714 Hildebrand, 12, 20 Hill, 471, 487–488 Hillman, 40, 44, 336–338, 679, 688–689 Hirschman, 206 Hirshleifer, 43 Hofstede, 206 Holmstro¨m, 190, 182 Horowitz, 693, 715 Hufbauer, 373 Hugo, 515 Hunt, 6, 19, 342, 356, 654

Husted, 267 Hutchinson, 358, 364, 372 Ilahi, 544 ILO-UNCHR, 135 Im, 123 Itzigsohn, 499, 514 Jaeger, 2, 20, 140, 172 Jafarey, 345, 356, 544 Jenkins, 480 Jones, 298–299 Joppke, 607, 630 June, 298, 306–307 Kaestner, 5, 20, 137–138, 140, 142, 144–146, 148, 150, 152, 154, 156, 160, 164, 166, 168, 172 Kahana, 13, 19, 499, 514 Kahanec, 7, 9–10, 20, 194, 203, 206, 208, 210, 212, 214–216, 218, 220, 222, 326, 339, 415–416, 418, 420, 422, 424, 426, 428, 430, 432, 434, 436, 438, 440, 442, 678, 689 Kahn, 8, 17, 193, 202, 239, 245, 326, 337–338, 342, 355 Kanbur, 54, 66, 110, 113, 126 Katav-Herz, 16, 20, 677–678, 680, 682, 684, 686, 688 Katseli, 494, 130 Katz, 451 Kaufmann, 701, 715 Kaushal, 5, 20, 137–138, 140, 142, 144, 146, 148, 150, 152, 154, 156, 160, 164, 166, 168, 172 Kaushal, 5, 20, 137–138, 140, 142, 144, 146, 148, 150, 152, 154, 156, 160, 164, 166, 168, 172 Kee, 237 Kelkar, 46, 66 Kendall, 47, 66 Kessler, 608 Keyes, 540 Khamis, 16, 691 Khan, 124 Killingsworth, 668 Kim, 373 Kimko, 69–72, 75–80, 84–85, 87, 89–90, 92 King, 2, 18, 25, 44, 119, 487 King, 2, 18, 25, 44, 119, 487 Kiviet, 124 Koc, 499, 514 Koch, 138, 171 K’onya, 689

Author Index Kossoudji, 271, 291, 403, 414 Kouaouci, 515 Kovenock, 337 Kritz, 138, 140–141, 151, 159, 166–167, 172 Krueger, 71, 79, 655, 678, 689 Krugman, 35, 43 Kurlat, 714 Kurthen, 379 Kwuon, 356 La Ferrara, 9, 16, 194, 202 Lahiri, 11, 20, 341–342, 344–346, 348, 350, 352, 354, 356, 479 Laibson, 228 LaLonde, 180 Lambert, 46, 66 Lane, 296, 300, 320 Lang, 290 Langbein, 305 Layard, 106 Lazear, 9, 20, 194, 203, 207, 287, 291, 326, 339 Lecker, 18, 203, 338, 413 Lee, 143, 173, 315–316, 624, 626, 628, 692, 715 Leigh, 342, 356 Lerman, 46, 66 Lewin, 459, 501, 515 Lewis, 108, 139, 171 Li, 239–244, 361, 583 Light, 5, 105, 107, 129, 188, 211, 222, 306, 332, 343, 356, 358, 373, 382, 413, 581 Lii, 206 Lindbeck, 679, 689 Lindstro¨m, 692, 715 Lipsey, 372 List, 120, 367, 503, 623, 629, 695 Little, 5–6, 14, 68–69, 77, 86, 90, 109, 127, 129, 137, 149, 164, 231, 233–234, 236, 271, 306, 320, 345, 351, 377, 388–389, 391, 418, 525, 527, 544, 572, 623–626, 629, 655–657, 661, 666, 678 Locatelli, 109 Lockard, 331, 339 Lofstrom, 68–69, 78 Log, 110, 121, 123, 125–126, 247, 249, 253, 255, 276–277, 280–281, 284–285, 287–288, 366, 368–369, 457, 462, 474, 480, 482–485, 490–493, 565–568, 571, 589–592, 595, 597, 662–664, 697 Logan, 138–139, 171–173 Long, 12–13, 35, 40, 106–107, 109, 122, 130, 142, 145–146, 155, 181–182, 184, 187, 199, 201, 206, 233, 245–246, 287, 305,

721

312, 316, 341, 343, 364, 417–418, 438, 441–442, 467–469, 471, 497, 507, 513, 520, 525, 613, 629–630, 667–668, 677, 679, 687 Lopez, 106 Loughna, 694–695, 714 Lowell, 515 Lowenstein, 451 Lu, 499, 515 Lucas, 106, 207, 543–544 Lundberg, 206 Macpherson, 668 Madhavan, 499–500, 515 Mahalanobis, 46, 65 Mai, 495 Majumder, 49, 65 Mak, 235–236 Malcomson, 178 Mani, 693–694, 715 Maniatis, 494 Marks, 2, 20, 25, 44, 209 Martin, 138, 172, 205–206, 208, 210, 212, 214, 216, 218, 220, 222, 359, 373, 415–416, 418, 420, 422, 424, 426, 428, 430, 432, 434, 436, 438, 440, 442–443 Martinez, 299, 315–316 Marzan, 171 Massey, 139, 166, 171–172, 468, 486, 498–499, 514–515, 653 Mattias, 688 Mayda, 15, 19, 605–612, 614, 616, 618, 620, 622, 630–632, 634, 636, 655 McCaffrey, 295, 297–298, 317, 319 McCarthy, 138, 172 McCormick, 646 McDowell, 364, 372 McFadden, 478 McKenzie, 12, 20, 520 Mealem, 7, 19, 193–194, 196, 198, 200, 202 Meenakshi, 547 Meng, 236 Menjivar, 521 Mesnard, 12, 20, 468, 470, 489 Metropolis, 574, 602 Michael, 356, 375–376, 378, 380, 382, 384, 386, 388, 390 Milbourne, 111, 468, 470, 487 Millar, 404, 410, 414 Miller, 2, 4, 8, 17, 25, 43, 67–72, 74, 76, 78–80, 82, 84, 86–88, 90–93, 139–140, 171, 193, 202, 271–272, 290–291, 326, 338, 342, 355, 364, 372, 522, 621 Mincer, 140, 172, 180, 589, 602

722

Author Index

Miranda, 520 Mishra, 647 Moehling, 17, 299–300, 314 Mohapatra, 575 Molle, 106 Money, 120, 244, 273, 305, 356, 468, 499–500, 515, 543, 549, 607 Monkkonen, 296, 300, 304, 320 Mookherjee, 46, 66 Moore, 138, 172, 692, 715 Morrison, 295 Moscato, 135 Muelbauer, 546, 548 Muhleisen, 654 Mundra, 12, 20, 357–358, 360, 362, 364, 366, 368, 370, 372 Munshi, 2, 20, 207 Murphy, 178 Muste, 647 Mutchler, 448, 451 Mutlu, 415–416, 418, 420, 422, 424, 426, 428, 430, 432, 434, 436, 438, 440, 442 Nagase, 479 Naon, 451 Narayanan, 182 Nawata, 479 Naylor, 298 Neidert, 206 Neuman, 598, 600 Neumark, 259 Neville, 237 Niedert, 263, 268 Nielsen, 12, 21, 232–233, 237, 261, 668 Nitzan, 329, 334, 338–339, 613 North, 111, 115, 310, 356, 420, 457, 462, 472, 476, 485, 549–552, 555–556, 559, 561, 563, 565–569, 572, 584, 614, 616–617, 619, 623, 694 O’Rourke, 608, 611 Oaxaca, 15, 144, 236, 257–260, 499, 514, 520, 598, 600, 668–669, 671 Ofer, 404, 414 oido-Lobaton, 701, 715 Olmstead, 321 Orozco, 499, 515, 518–520, 530 Ortega, 15, 20, 606, 612–613, 621 Ottaviano, 6, 21, 238, 688 Padoa-Schioppa, 130, 134 Paldam, 679, 689 Papademetriou, 494 Park, 668

Parrado, 498–499, 514–515 Parsons, 182 Pasqua, 135 Pedace, 237–238 Pellegrino, 515 Pendakur, 554, 584 Peri, 6, 21, 238, 688 Pesaran, 123 Phelps, 206 Piehl, 17, 299–300, 314 Piracha, 13, 21, 467–468, 470, 472, 474, 476–478, 480, 482, 484, 486–488, 490, 492–493 Piracha, 13, 21, 467–468, 470, 472, 474, 476–478, 480, 482, 484, 486–488, 490, 492–493 Pitblado, 495 Pivnenko, 580–581, 583–586 Poggio, 14, 21, 517–522, 524, 526, 528, 530, 532 Poirine, 544 Polachek, 3–4, 21, 355 Portes, 108, 358, 373 Poterba, 111 Preston, 18, 203, 338, 608, 611, 651, 655, 658 Pritchett, 606 Puglisi, 647 Pyatt, 46, 66 Quigley, 541 Radu, 471 Rafﬁee, 547, 573 Ransom, 257, 259, 669 Rapoport, 3, 21, 197, 203, 520, 544 Ratha, 573 Rauch, 358, 366, 373 Ray, 45, 49, 66, 547 Razin, 129 Regets, 180 Reilly, 692, 716 Reimers, 236, 259 Reiss, 306 Reitz, 583 Relles, 479 Rey, 108, 402, 414 Rhee, 574 Rice, 305 Richman, 206 Riddell, 584 Riley, 336–338 Riphahn, 380, 402, 414 Rivera-Batiz, 6, 8, 19, 44, 194, 203, 327, 338, 342, 356, 649, 654–655

Author Index Roberson, 124 Roche, 266 Rodriguez, 499, 514 Rodrik, 356, 607 Ronnander, 322 Rosholm, 267 Ruggles, 303 Rumbaut, 358, 373, 518 Russel, 499, 515 Rybaczuk, 119 Ryzin, 172 Saavedra, 358, 373 Sabol, 310 Sacerdote, 300, 314 Sadka, 129 Salanie´, 190 Salazar-Parren˜as, 521, 540 Salt, 115 Salvanes, 402, 413 Sapienza, 135 Sastry, 46, 66 Sawyer, 520 Scharfstein, 28, 44 Scheinkman, 228 Schen, 519 Scheve, 608–609, 611, 655 Schiff, 106, 113, 197, 203 Schmidt, 6, 21, 177–178, 180, 182, 184, 186, 188, 376, 379–380, 654 Schmitz, 236–237 Schott, 373 Schultz, 12, 21, 402–403, 411, 414 Schweikart, 305 Scovel, 522 Seddon, 499, 515 Sela, 330, 338 Sellin, 299 Semyonov, 21, 459, 498–499, 501, 515 Senhadji, 124 Sessions, 7, 17, 231–232, 234, 236, 238, 240, 242, 244, 246, 248, 250, 252, 254, 256, 258, 260, 262, 530 Shakotko, 181 Shamsuddin, 237 Sharpless, 303 Shin, 123 Shorrocks, 46, 48, 50, 66 Shortridge, 303 Sik, 403, 414 Sims, 124 Singh, 499–500, 515 Siniver, 7, 21, 269–270, 272, 274, 276, 278, 280, 282, 284, 286–288, 290–292

723

Sinning, 181 Sinnott, 608, 611 Sjaastad, 139, 172 Slaughter, 608–609, 611, 655 Smith, 8, 21, 111, 138, 172, 326, 339, 519 Sobek, 322 Sollenberg, 715 Song, 479 South, 58, 76, 111, 114, 140, 172–173, 237, 371, 448, 457, 462, 472, 515, 529, 545, 550–552, 555–556, 560, 562–563, 565–569, 572–573, 623, 626, 628, 651, 696 Spence, 533 Spiegel, 693, 716 Spilimbergo, 106, 120, 637 Spoerer, 297 Sribney, 495 Stah, 514 Stampini, 494 Stark, 110, 113, 172, 179, 261, 342–343, 356, 389, 392, 428, 468, 470, 477, 498–499, 516, 543–544 Startz, 206 Steckel, 303 Stein, 28, 44 Stock, 3, 12, 27, 38, 41, 119, 123, 126–127, 306, 357–358, 364–371, 503, 545, 582, 586, 588 Stolzenberg, 479 Strand, 417, 715 Strosberg, 451 Stuart, 47, 66, 404, 413 Stults, 171 Suarez-Orozco, 518–520, 530 Sugden, 679, 689 Sutch, 297 Svejnar, 675 Svendsen, 679, 689 Sweetman, 69, 79, 85 Symons, 124 Taft, 299 Tamang, 696, 716 Tandon, 235 Tansel, 402, 414 Taylor, 13, 21, 477, 499, 515–516, 669 Tenorio, 235, 237 Terrell, 68–69, 78, 85 Testa, 15, 19, 606 Thomas, 182, 297, 402, 414, 554 Tienda, 139, 173, 206, 451 Tilghman, 342, 356 Todorova, 541

724

Author Index

Topel, 180–181 Tranaes, 12, 21 Tran-Nam, 237 Trejo, 102 Tullock, 331, 339 Turan, 715 Udry, 207 Ulrich, 688 Vadean, 13–14, 18, 21, 468, 470, 472, 474, 476–478, 480, 482, 484, 486, 488, 490, 492–493, 544, 546, 548, 550, 552, 554, 556, 558, 560, 562, 564, 566, 568, 570, 572 Valente, 207 Valentin, 190 Van den Berg, 6, 17 van der Leij, 208 Van Hook, 448 Van Ours, 521 van Soest Language, 391, 399 Veenman, 521 Velez-Ibanez, 343, 356 Venables, 111 Venturini, 5, 19, 105–106, 108, 110, 112, 114, 116, 118–120, 122, 124, 126–128 Vernez, 138, 172, 522 Vinokur, 404, 414 Vishwanath, 17, 43 Vogler, 106 Vullnetari, 472–473 Wacziarg, 714 Wagner, 359, 373, 403, 407, 414 Wang, 49, 66 Warnock, 133 Watchman, 297–298 Watson, 714 Watts, 613 Weber, 237 Weiss, 3, 21, 44, 197, 203, 271, 291–292, 450 Welch, 8, 21, 43, 206, 291, 326, 339 Whinston, 613

Widgren, 138, 172 Wilbur, 688 Willcox, 134 Williams, 138, 173 Williamson, 106, 110, 122 Willmann, 607, 613 Wilson, 139, 173, 237, 342, 355 Winkelmann, 654 Winter-Ebmer, 418, 442, 654 Wittke, 317 Wolf, 108 Wolfe, 402, 414 Wolff, 13, 18, 544 Wolfson, 49, 66 Wong, 206 Worrall, 182 Worswick, 237 Wright, 321 Wyatt, 13, 21 Wyplosz, 110–111 Xie, 237 Ying, 237 Yitzhaki, 46, 66 Yuengert, 665 Yuksel, 10, 20, 415–416, 418, 420, 422, 424, 426, 428, 430, 432, 434, 436, 438, 440, 442 Yun, 17, 19, 44, 202–203, 337–338, 355–356, 668–669 Zavodny, 138–140, 173 Zellner, 143, 173, 553 Zezza, 494 Zhang, 54, 66, 171–172 Zhong, 356 Zhou, 139, 173, 373 Zimmermann, 12, 18–19, 21, 127, 203, 326, 338, 355, 376, 403–404, 414, 416, 442, 468, 654, 678, 689 Zingales, 135 Zlotnik, 499, 516, 694, 716 Zweimueller, 654

SUBJECT INDEX

adjustment costs, 8, 193, 326, 342 ageing population, 16 assimilation, 1–3, 6–12, 149, 167, 178, 193–201, 231–233, 235–236, 270, 287–289, 325–327, 329, 331, 333, 335, 341–346, 348, 351–355, 357, 359, 364–365, 367, 369–371, 448, 487–488, 547–548, 570–571, 585, 621, 687 attitudes, 1–2, 10, 14–16, 40, 234, 375, 377–380, 382, 385–389, 391, 451, 606–609, 611, 622–625, 627, 629–636, 649–673, 677–679, 683

crime, 5, 9, 137, 142, 146–147, 150, 152–155, 158, 162–164, 166, 168–170, 295–302, 304–306, 308, 310–312, 314–320, 611, 650, 655 cultural heritage, 8, 205, 326 cultural identities, 3 cultural identity, 9, 326–327, 677 culture, 1–3, 5, 7, 9, 11–13, 15, 40, 108, 238, 269–271, 273, 275, 277, 279, 281, 283–285, 287, 289–290, 299, 341, 477, 481, 485, 518, 522, 528–532, 681 customs, 1, 205, 678–679

borrowing, 6, 11, 184, 341, 344, 347–348, 350, 352–355

demographic factors, 5, 128–129, 137 destination countries, 15, 115, 119–120, 122, 127, 129, 177–178, 188, 378, 472, 476, 488, 493, 605–607, 609, 611–612, 621–622, 630, 637, 702 differentiated products, 12, 365 discrimination, 1, 6–8, 85, 107, 194, 206, 223, 231–239, 241–245, 247, 249, 251–253, 255, 257–259, 261–262, 327, 378, 415, 419, 421–423, 438, 582–584, 598–600, 602, 653, 667, 671–672, 674

children, 12–14, 86, 146–147, 158, 170, 308, 311, 376, 401–413, 415, 417, 421–422, 424–428, 432–441, 448, 451–453, 455, 457–464, 472–473, 475–476, 484–487, 489–493, 512–513, 517–533, 544, 569, 590–592, 594–597, 620, 660–663, 673, 693–694, 696, 701 circular migrants, 13, 467, 469–470, 473–474, 476–478, 485, 488–489, 491, 493 competition, 2, 6–7, 10, 26, 188, 205–207, 209–211, 213, 215, 217, 219, 221–223, 325–328, 335, 611, 655, 665, 667, 672, 678 complementarity, 6, 210, 220, 222, 239–241, 344, 352, 611, 613, 636, 654 complements, 6, 8, 39, 179, 220, 239–242, 327, 359, 524, 636–637, 654 conﬂict, 1, 8–10, 15, 193, 299, 325–326, 416–418, 420–421, 441, 679, 680, 691, 692–700, 702–714 consumption, 13–14, 27, 35, 111–112, 184, 209, 213, 216, 344–346, 350–351, 353, 417, 470, 497, 499, 501, 503, 511, 513, 546, 548, 553, 563, 588, 677, 680–682, 688 country-of-origin, 2 credit, 1–2, 11, 139, 329, 341, 343, 345, 347, 349, 351–353

earnings, 2, 4, 6, 7, 15, 27, 34, 182–183, 185–187, 231–232, 235–238, 241, 251–252, 257–258, 261, 263, 269–276, 278, 280, 282, 284, 289, 306, 342–343, 379, 381, 418, 423, 428, 434, 488, 498–499, 520, 582–598, 701 education, 4–6, 13–15, 67–70, 72, 75, 78, 80, 84–91, 109–110, 137–138, 140–142, 144, 146–150, 154, 156–158, 160, 162–163, 165–166, 168–170, 181, 188, 235–236, 238, 269–272, 274–286, 288, 364, 376, 379, 381–382, 385, 402–406, 408–412, 415, 418–419, 421–423, 425, 428, 432, 434, 438–441, 450, 458, 461, 467, 469, 473–476, 481–485, 487–488, 490–493, 501–504, 506–512, 517, 520–525, 529, 531–532, 544, 546–547, 550–551, 554, 556–557, 566, 568–569, 571, 583, 586, 590, 594, 596, 598, 601,

726

Subject Index

611, 624–628, 636, 649, 659–664, 670–671, 673, 679, 693, 696 educational attainment, 11, 16, 68–70, 78, 139, 142, 306, 380, 402–403, 411–412, 415, 417–419, 421, 423, 425, 428, 432, 434, 436, 438–439, 441, 520–521, 585, 590–594, 649, 660, 662, 670–673 educational credentials, 6, 177, 179 effort, 7–8, 10, 182, 188, 193, 195–197, 199–200, 325–335, 342, 344, 352, 354, 614 emigration, 2, 28–30, 32–36, 40, 107, 120–123, 126–127, 129, 342, 404, 491, 630 enclave, 1, 3, 141, 289, 357, 364, 367, 369 ethnic conﬂicts, 15, 691–693, 695, 696, 701 ethnic goods, 3, 27, 366–367 ethnic group, 7–8, 10, 14, 45, 47, 49, 51, 53, 55, 57, 59, 61, 63, 152, 205–213, 215–223, 235, 401, 412, 416, 543–545, 570, 692–693, 704 ethnic minorities, 7, 205, 223, 650, 655 ethnic networks, 2, 11, 27, 358, 423 externalities, 2, 25–28, 33–41, 207 family issues, 2, 12 family reuniﬁcation, 12, 15, 573, 605–606, 614–615, 619–620, 643 ﬂow, 3 foreign culture, 2 foreign-born, 4–5, 71, 85, 90, 137–138, 140, 142, 146–148, 150, 152, 154, 157–161, 166, 168–170, 234–239, 296, 298, 300, 305–308, 310–314, 316, 320, 342, 545, 547, 549, 551, 558, 570, 587, 596, 598, 602, 628, 650–651, 658 free-rider, 9, 194, 327 friction, 6 full employment, 361 generations of migrants, 182 geographical location, 13, 467, 469 german socio-economic panel (gsoep), 10, 287, 375, 377, 381, 391, 401, 403–405, 407 gini index, 4, 45–49, 54–56, 58–60, 488 harassment, 9, 193, 327, 342 highly skilled migrants, 2, 621 hindering assimilation, 3 history, 3, 26–27, 178, 182, 186, 205, 231, 233, 245, 270, 288, 320, 343, 469, 472–474, 477, 485, 551, 697, 699–700

home country, 2, 5, 8, 11–13, 28, 33–34, 38, 105–106, 111–114, 117, 119, 123, 126–127, 129, 181, 343, 357–361, 364–367, 369–370, 407, 467–471, 477, 487, 493, 499, 520, 522–523, 526–527, 530–531, 544–545, 570 homogenous goods, 12 host country, 2–3, 7–8, 11–12, 26, 28, 33–35, 37–38, 40, 106, 110–112, 120, 123, 125–127, 139–141, 179–180, 182, 184, 193, 232, 234, 244, 269–271, 278, 284–285, 289–290, 326, 342–344, 358, 360–362, 370–371, 468–471, 483, 487, 544, 548, 570, 652, 678–682, 685, 695–696, 701–702, 704, 708–709 hostile, 3, 629, 655, 667, 673 human capital, 4, 6–8, 10–11, 13, 68–69, 71–72, 75–76, 78, 87, 92, 177–181, 186–188, 205, 207, 209, 212–213, 215–216, 222–223, 232–233, 236, 238, 240, 244, 252, 289, 342, 378–379, 402, 404, 411, 415, 417–421, 423, 425, 427, 429, 431–435, 437–441, 470, 488, 498, 503, 513, 580–583, 585–590, 593–594, 599–602, 612, 678 human capital theory, 6, 177, 179–180, 186 identity, 1, 8–10, 12, 15, 28–29, 326–327, 380, 606, 622, 624, 626, 635, 677 immigrant characteristics, 5, 137, 139, 141, 143, 146, 166, 287 immigrant composition, 5, 40, 137, 164 immigrant earnings, 6, 69, 180, 231–232, 235–238, 261, 582–583, 585 immigrant stock, 357–358, 364–365, 367–371, 586 incarceration patterns, 9, 311 inﬂuential factor, 4, 67 informal barriers, 11, 108 information costs, 11 intergenerational familial conﬂicts, 8, 193, 326 intergenerational transfers, 16 interlinkages, 342 internal labor mobility, 5, 105, 129 international trade, 11, 358–359 international trafﬁcking, 15, 691–692, 699, 701 isolation, 8, 194, 327, 376 labor market, 4, 6–8, 12–15, 40, 67–68, 70, 77, 79, 84, 87, 92–93, 108, 115, 119–120, 125, 140–141, 177–189, 193–194, 205–206, 208–212, 215–217,

Subject Index 220–223, 231, 233–235, 237–239, 241, 243, 245, 247, 249, 251, 253, 255, 257, 259, 261, 271, 274, 278, 289, 306, 326–327, 342, 376, 379–380, 392, 402, 415, 418–420, 422–423, 425, 428, 432, 434, 436–438, 447, 450–451, 455–456, 458, 460–461, 464, 498, 501–502, 504, 506–512, 522, 609, 611, 613, 620, 649, 654–655, 663–667, 670–673, 681 labor market performance, 6, 13–14, 177, 180, 206 lack of knowledge, 8, 193, 326, 523–524 language skills, 8, 70, 79, 193, 270–271, 275, 287, 289, 326, 342, 528–529, 586 language, 2–3, 7–8, 25–27, 68, 70, 79, 179, 193, 205, 238, 261, 269–275, 277–281, 283–285, 287–290, 303, 326, 342–343, 358, 366, 368, 370, 378–380, 383, 390–391, 476, 483, 488, 520, 522–523, 526–530, 532, 582, 586, 622, 654, 697–699, 702, 707, 713 legal status, 4, 27, 85 leontief production function, 7, 238, 244 local population, 1–3, 6–7, 11, 14, 16, 26, 34–35, 38, 40, 194–195, 197–201, 327, 417, 677–682, 685, 687 location attributes, 5, 137, 142, 144, 146, 150, 153–155, 157–158, 161, 164, 168–170 location choice, 1–2, 4–5, 27, 39, 111, 137–138, 140, 142–147, 149–155, 159–160, 166–169 locational choice, 5, 39 low-skilled, 6, 127, 621, 678 migrants, 1–3, 6–8, 11–13, 26–27, 33, 35, 38–41, 106–107, 109–110, 113–115, 119–120, 122–123, 125–130, 177–180, 182–189, 193–201, 235, 298, 300, 310, 325–327, 335, 345, 359, 376–381, 385, 391, 403–404, 449, 467–470, 472–474, 476–479, 481–493, 497–501, 503–506, 510–513, 519, 530, 544, 570, 572–573, 606, 609–610, 613–616, 619–621, 629–630, 649, 654, 695 migration, 1–5, 7–9, 11–15, 25, 27–28, 33, 40, 67–68, 70, 85–86, 92, 105–117, 119–123, 125–130, 137–143, 145–147, 149, 151, 153, 155, 157–161, 163, 165–167, 169–170, 177, 179, 181–182, 184, 187–188, 193, 235, 273, 276, 297, 325, 341–342, 354, 359, 361, 363, 376–379, 401, 418, 448–449, 454, 467–479, 481–493, 497–501, 506–507, 511–513, 517–528, 530–531, 543–546,

727

605–608, 610–622, 624, 627–638, 649–650, 692, 695, 700 migration choices, 5, 137, 139, 141–142, 151, 166 migration destination, 1 migration enclaves, 3 migration policies, 14–15, 120, 245, 579, 605–608, 616–617, 630, 636, 655, 680 minority, 3, 7–9, 40, 193–194, 205–206, 209–210, 212–216, 218–223, 233, 241, 261–262, 327, 416–417, 687, 692–693 mobility, 1, 5, 89, 105–106, 108–110, 129, 137, 140, 155, 157, 159–160, 289, 359, 365, 386–387, 608, 613, 696 national identity, 15, 380, 606, 622, 624, 626, 635, 677 nationality discrimination, 7, 231–239, 241–245, 247, 249, 251, 253, 255, 257, 259, 261–262 native-born, 4, 6–7, 71, 90, 147, 231–232, 234–235, 237–240, 244, 251–252, 257–258, 298, 300, 303, 305–306, 310–311, 314, 319, 326, 403, 411, 654, 659 natives, 3, 11, 108–109, 138, 140, 145–147, 157–158, 160, 166, 170, 178–180, 182–183, 186, 232–233, 235–236, 238, 240–241, 244, 251–252, 261, 269–271, 295, 299, 301, 303, 305–306, 311–312, 314–316, 320, 341–343, 345–346, 348–349, 354, 357, 359–360, 362, 367, 369–370, 377–380, 383, 385, 391, 448–456, 459–461, 472, 547, 609–611, 613, 616, 620, 622, 624–628, 630, 654, 663, 673 negative externalities, 2, 26, 37–38 network, 2, 12, 25–28, 33–41, 207–208, 212, 216–221, 223, 357, 359, 364, 368–370, 464, 485–486 network size, 12 oaxaca decomposition, 144, 257–258, 671 origin, 2–4, 27, 34, 45–46, 60–61, 67–69, 71, 73, 75, 77–79, 81, 83–87, 89, 91–93, 108–111, 121–122, 130, 142, 145–147, 150, 152, 155, 157–160, 164, 170, 178, 182, 184, 187, 236, 341–343, 345–346, 354, 358, 360, 364, 379, 383, 390, 403, 405, 416, 418, 447–449, 451, 464, 468, 471, 489, 491, 493, 497–498, 519, 521, 529, 544, 551, 583, 587, 596, 598, 651, 660, 678–679, 694–695, 698, 702 outperforming, 7, 205

728

Subject Index

payoff to schooling, 4–5, 67–71, 78–82, 84–87, 90–93 permanent migrants, 6, 177, 184, 187, 467, 473–474, 476–477, 486–487, 573 permanent migration, 15, 181, 467, 469, 471–473, 478, 481, 483, 485–486, 491, 605–606 polarization, 4, 45–51, 53–61, 63 political culture, 2 political institutions, 9, 194, 327, 612 political parties, 9, 40, 194, 327, 382, 608, 611 population subgroups, 4, 45–46, 51, 53, 62 poverty trap, 419, 432 production activities, 7, 193, 200 production, 2, 6–7, 69, 75, 106, 182, 193–197, 200, 209–211, 213–218, 220, 223, 233, 238–239, 241, 244–245, 262, 346, 360–363, 365, 370–371, 612–613, 678 productivity, 6–7, 11, 69, 178–181, 185–186, 188, 195–196, 198, 231–232, 238, 240–246, 252–258, 261–262, 289, 341, 361 propensity to migrate, 5, 105–106, 112–113, 126, 128, 130, 167 public good provision, 11, 346 public policies, 14, 342, 517–518, 653 public policy, 2, 9, 14, 194, 201, 327, 415 quality of schooling, 4, 67, 71, 89–90, 92–93 quality school, 4, 67, 90, 92 recipient countries, 652 remittances, 2, 8, 12–14, 341–345, 353–354, 428, 493, 497–507, 509–513, 520, 531, 543–546, 549–551, 553, 555–565, 567, 569–573 repeat migrant, 8, 13, 467, 469, 491 risk, 11, 28, 110–111, 179, 182, 184, 236, 308, 328, 332, 343, 376, 380, 392, 441, 468, 486, 488, 525, 693, 695, 701 schooling, 4–5, 7, 10, 67–71, 78–87, 89–93, 148–149, 155–157, 232, 238, 269, 271–273, 275, 278, 280, 282–286, 288–290, 379–380, 401–413, 423–426, 428–429, 433, 436–437, 439–440, 455–459, 461–464, 501, 520–521, 525–526, 531, 544, 580, 659 selection, 2, 4, 13–14, 67–68, 78, 90, 92–93, 274, 278–279, 286, 305, 418, 467–470, 479–481, 483, 485–489, 493, 506, 579–583, 588, 590, 594–598, 602

separate identity, 8, 326 separation for children, 12 separation, 1–2, 12, 14, 232, 476, 499, 517–524, 526–528, 530–531 signals, 3, 30, 32–33, 35, 38, 246 skill acquisition, 7, 205, 207, 209, 212, 216–217, 219–222 skill-biased technological, 491 social distances, 7, 205, 208–209, 217, 222 social norms, 16, 677–683, 685–687 socio-economic characteristics, 467, 469, 479 specialization, 7, 205–209, 211, 213, 215, 217–223 spillover effects, 7, 11, 205, 207–208, 212–213, 215, 217–218, 222–223, 415 spot markets, 6, 177–178 standard of living, 13–14, 28, 121, 188, 476, 486, 497–498, 500–507, 509–513, 678 statistical discrimination, 206, 234 status quo, 14–15, 605–606, 613–614, 616, 637 stock, 3, 12, 27, 38, 41, 119, 123, 126–127, 306, 357–358, 364–371, 503, 545, 582, 586, 588 stress of relocating, 2 structure of the family, 447–448 substitutability, 6, 206, 209, 215–216, 220–222, 238–241, 613, 636 substitutes, 6–8, 112, 194, 216–217, 220–221, 233, 238–242, 244, 261–262, 327, 359, 364–365 taste-based discrimination, 423 temporary migration, 8, 13, 467–473, 477, 479, 484–488, 490–493, 617, 638 timing of immigration, 16, 685, 687 trade, 9, 11–12, 16, 34, 107–108, 194, 217, 219–220, 222, 327, 347, 350, 357–371, 606–607, 609, 612, 679–680 trade liberalization, 9 trade organizations, 9, 194, 327 transactions, 3, 11, 358 transferable human capital, 7 uncertainty, 11, 26, 31, 33, 40, 111, 182–183, 185, 188, 449, 471 underperforming, 7, 205 unemployment, 8, 108, 115, 118–120, 122–123, 125–127, 139–140, 179, 194, 326, 419, 421, 428, 434, 437–438, 470, 652, 654–657, 660, 663–664, 666–667, 678 unionization, 5, 137, 150–151, 153–154, 158, 162–163, 166, 168–170 unions, 9, 194, 327, 613

Subject Index values, 1, 32, 48, 51–52, 54–55, 58, 60, 62, 73, 78–80, 84, 86, 88–91, 125, 147, 184, 205, 219, 319, 334, 346, 352, 377, 379–381, 387, 448, 473, 475, 547, 550–552, 555, 558, 563, 569, 588–589, 596, 614–615, 617–618, 620–626, 630, 661, 669, 679, 685 violent crimes, 9, 295, 300, 315–316, 319–320 wage, 33–36, 38, 89–90, 110–112, 114–115, 119, 121–122, 125, 127, 129–130, 139, 142, 146–147, 150–151, 154, 158, 161–163, 168–170, 177–181, 183–187, 209–211, 213–215, 218, 232–233, 235, 239–241, 243–244, 257–261, 269, 271,

729

273, 278–279, 286, 305, 344, 346–347, 354, 360, 362, 371, 379, 419, 424–426, 432, 434–435, 470, 474–475, 481–482, 484, 490, 492, 498, 521, 584, 588–592, 594, 596, 598–602, 613, 654, 657, 659–661, 663–664, 666, 668, 671, 678, 681 wealth, 113, 420, 687 welfare, 16, 139–140, 179–180, 376, 380, 428, 432, 436, 472, 609–613, 620, 655 willingness to assimilate, 12, 679 workplace interaction, 1 xenophobia, 3, 25, 40