This Page Intentionally Left Blank
Immigration, Trade, and the Labor Market
A National Bureau of Economic Research Project Report
Immigration, Trade, and the Labor Market
Edited by
John M. Abowd and Richard B. Freeman
The University of Chicago Press
Chicago and London
JOHNM. ABOWDis professor of labor economics and management, Cornell University, and a research associate of the NBER. RICHARD B. FREEMAN is professor of economics at Harvard University and director of the Labor Studies program at NBER.
The University of Chicago Press, Chicago 60637 The University of Chicago Press, Ltd., London 0 1991 by the National Bureau of Economic Research All rights reserved. Published 1991 Printed in the United States of America 00999897969594939291 5 4 3 2 1
Library of Congress Cataloging-in-Publication Data Immigration, trade, and the labor market / edited by John M. Abowd and Richard B. Freeman. cm. - (National Bureau of Economic Research Project p. paper) “Papers presented at a conference held in Cambridge, Massachusetts, 11-12 September 198T-Pref. Includes bibliographical references and indexes. ISBN 0-226-00095-8 (acid-free paper) 1. Alien labor-United States-Congresses. 2. Alien labor-Canada-Congresses. 3. Alien labor-Australia-Congresses. 5 . Labor market-Can4. Labor market-United States-Congresses. ada-Congresses. 6. Labor market-Australia-Congresses. 7. Foreign trade and employment-United States-Congresses. 9. Foreign 8. Foreign trade and employment-Canada-Congresses. I. Abowd, John M. trade and employment-Australia-Congresses. 11. Freeman, Richard B. (Richard Barry). 111. Series. HD8081.A5153 1991 331.6’2-dc20 90-24954 CIP
8 The paper used in this publication meets the minimum requirements of the American National Standard for Information Sciences-Permanence of Paper for Printed Library Materials, ANSI 239.48-1984.
National Bureau of Economic Research Officers George T.Conklin, Jr., chairman Paul W. McCracken, vice chairman Martin Feldstein, president and chief executive oficer
Geoffrey Carliner, executive director Charles A. Walworth, treasurer Sam Parker, director offnance and administration
Directors at Large John H. Biggs Andrew Brimmer Carl F. Christ George T. Conklin, Jr. Kathleen B. Cooper Jean A. Crockett George C. Eads Morton Erhlich
Martin Feldstein George Hatsopoulos Lawrence R. Klein Franklin A. Lindsay Paul W. McCracken Leo Melamed Michael H. Moskow James J. O’Leary
Robert T. Parry Peter G. Peterson Robert V. Roosa Richard N. Rosett Bert Seidman Eli Shapiro Donald S. Wasserman
Directors by University Appointment Jagdish Bhagwati, Columbia William C. Brainard, Yale Franklin Fisher, Massachusetts Institute of Technology Jonathan Hughes, Northwestern Saul H. Hymans, Michigan Marjorie B. McElroy, Duke James L. Pierce, California, Berkeley
Andrew Postlewaite, Pennsylvania Nathan Rosenberg, Stanford Harold T. Shapiro, Princeton Craig Swan, Minnesota Burton A. Weisbrod, Wisconsin Michael Yoshino, Harvard Arnold Zellner, Chicago
Directors by Appointment of Other Organizations Richard A. Easterlin, Economic History Association Gail Fosler, The Conference Board A. Ronald Gallant, American Statistical Association Bruce Gardner, American Agricultural Economics Association Robert S . Hamada, American Finance Association Robert C. Holland, Committeefor Economic Development
David Kendrick, American Economic Association Ben E. Laden, National Association of Business Economists Rudolph A. Oswald, American Federation of Labor and Congress of Industrial Organizations Douglas D. Purvis, Canadian Economics Association Charles A. Walworth, American Institute of Certified Public Accountants
Directors Emeriti Moses Abramovitz Emilio G. Collado Frank W. Fetter
Thomas D. Flynn Gottfried Haberler Geoffrey H. Moore
George B. Roberts Willard L. Thorp William S. Vickrey
Relation of the Directors to the Work and Publications of the National Bureau of Economic Research 1. The object of the National Bureau of Economic Research is to ascertain and to present to the public important economic facts and their interpretation in a scientific and impartial manner. The Board of Directors is charged with the responsibility of ensuring that the work of the National Bureau is carried on in strict conformity with this object. 2. The President of the National Bureau shall submit to the Board of Directors, or to its Executive Committee, for their formal adoption all specific proposals for research to be instituted. 3. No research report shall be published by the National Bureau until the President has sent each member of the Board a notice that a manuscript is recommended for publication and that in the President’s opinion it is suitable for publication in accordance with the principles of the National Bureau. Such notification will include an abstract or summary of the manuscript’s content and a response form for use by those Directors who desire a copy of the manuscript for review. Each manuscript shall contain a summary drawing attention to the nature and treatment of the problem studied, the character of the data and their utilization in the report, and the main conclusions reached. 4. For each manuscript so submitted, a special committee of the Directors (including Directors Emeriti) shall be appointed by majority agreement of the President and Vice Presidents (or by the Executive Committee in case of inability to decide on the part of the President and Vice Presidents), consisting of three Directors selected as nearly as may be one from each general division of the Board. The names of the special manuscript committee shall be stated to each Director when notice of the proposed publication is submitted to him. It shall be the duty of each member of the special committee to read the manuscript. If each member of the manuscript committee signifies his approval within thirty days of the transmittal of the manuscript, the report may be published. If at the end of that period any member of the manuscript committee withholds his approval, the President shall then notify each member of the Board, requesting approval or disapproval of publication, and thirty days additional shall be granted for this purpose. The manuscript shall then not be published unless at least a majority of the entire Board who shall have voted on the proposal within the time fixed for the receipt of votes shall have approved. 5. No manuscript may be published, though approved by each member of the special manuscript committee, until forty-five days have elapsed from the transmittal of the report in manuscript form. The interval is allowed for the receipt of any memorandum of dissent or reservation, together with a brief statement of his reasons, that any member may wish to express; and such memorandum of dissent or reservation shall be published with the manuscript if he so desires. Publication does not, however, imply that each member of the Board has read the manuscript, or that either members of the Board in generahor the special committee have passed on its validity in every detail. 6. Publications of the National Bureau issued for informational purposes concerning the work of the Bureau and its staff, or issued to inform the public of activities of Bureau staff, and volumes issued as a result of various conferences involving the National Bureau shall contain a specific disclaimer noting that such publication has not passed through the normal review procedures required in this resolution. The Executive Committee of the Board is charged with review of all such publications from time to time to ensure that they do not take on the character of formal research reports of the National Bureau, requiring formal Board approval. 7. Unless otherwise determined by the Board or exempted by the terms of paragraph 6, a copy of this resolution shall be printed in each National Bureau publication.
(Resolution adopted October 25, 1926, as revised through Seprember 30, 1974)
Contents
ix
Preface
1
Introduction and Summary John M. Abowd and Richard B. Freeman I. IMMIGRANT FLOWSAND PERFORMANCE IN THE UNITEDSTATES
29
1. Immigration and Self-Selection George J. Borjas 2. Undocumented Mexican-born Workers in the United States: How Many, How Permanent? George J. Borjas, Richard B. Freeman, and Kevin Lang
3. The Effect of Policy Restrictions on Capital and Labor Flows in Mexico Juan Diez-Canedo R.
5. Migration, Ethnicity, and Labor Force Activity Marta Tienda and Franklin D. Wilson IMPACT OF IMMIGRATION, TRADE,AND CAPITALFLOWSON U.S. LABORMARKET
101 121
4. Internal Migration of U.S. Immigrants Ann P. Bartel and Marianne J. Koch
11.
77
135
THE
6. Labor Market Adjustments to Increased Immigration Robert J. LaLonde and Robert H. Tope1
167
viii
Contents
7. The Effects of Immigration on the Labor
Market Outcomes of Less-skilled Natives Joseph G. Altonji and David Card
20 1
8. Industrial Wage and Employment Determination in an Open Economy Richard B. Freeman and Lawrence F. Katz
235
9. Foreign-Owned Businesses in the United States Jonathan S. Leonard and Rachel McCulloch
26 1
10. Immigration, International Ikade, and the Wages of Native Workers Peter Kuhn and Ian Wooton
285
11. Immigrants, Labor Market Pressures, and the Composition of the Aggregate Demand Susan M. Collins
305
EXPERIENCES: CANADA AND AUSTRALIA 111. COMPARATIVE 12. An Analysis of the Earnings of Canadian Immigrants David E. Bloom and Morley Gunderson
13. The Effects of International Competition on Collective Bargaining Outcomes: A Comparison of the United States and Canada John M. Abowd and Thomas Lemieux 14. Male Immigrant Wage and Unemployment Experience in Australia John J. Beggs and Bruce J. Chapman 15. Why are Low-skilled Immigrants in the United States Poorly Paid Relative to Their Australian Counterparts? Some of the Issues Illustrated in the Context of the Footwear, Clothing, and Textile Industries R. G. Gregory, R. Anstie, and E. Klug
321
343
369
385
Appendix: The NBER Immigration, Ikade, and Labor Markets Data Files John M. Abowd
407
List of Contributors
423
Author Index
425
Subject Index
429
Preface
This volume consists of papers presented at a conference held in Cambridge, Massachusetts, 11-12 September 1987, and is part of the National Bureau of Economic Research Labor Studies program. Support for the project came from the Ford Foundation. We are grateful to Jennifer Amadeo-Holl, Jane Konkel, and Jean Brown for help in getting these papers ready for publication. Any opinions expressed in this volume are those of the respective authors and do not necessarily reflect the views of the National Bureau of Economic Research or the sponsoring organization.
This Page Intentionally Left Blank
Introduction and Summary John M. Abowd and Richard B. Freeman
During the 1970s and 1980s, immigration, trade, and foreign investment in the United States became increasingly important in the U.S. labor market. The number of legal and illegal immigrants to the country increased, altering the size and composition of the work force and substantially raising the immigrant share of labor in “gateway” cities such as Miami, Los Angeles, and New York. The national origins of immigrants changed from primarily European to Mexican, Latin American, and Asian. Foreign trade rose relative to gross national product, and a massive trade deficit developed in the 1980s, turning the United States into a substantial debtor nation. Because the composition of employment shifted from manufacturing to nontraded services, the immediate burden of adjusting to trade-induced changes fell on a decreasing segment of the work force. As the flip side of the trade deficit, foreign investment in the United States grew rapidly, with foreign direct investment increasing until 3% of American workers were employed in foreign-owned firms. While at one time labor market analysts could look on the United States as a largely closed economy, the changes of the 1970s and 1980s brought about the internationalization of the U.S. labor market. What are the interrelations among the flows of foreign outputs and inputs that have caused such a change in the way we look at the U.S. labor market? How have the flows changed over time? Which industries or areas are most John M. Abowd is professor of labor economics and management, Comell University, and a research associate of the National Bureau of Economic Research. Richard B. Freeman is professor of economics, Harvard University, and director of the Labor Studies Program of the National Bureau of Economic Research. The authors acknowledge financial support from the Ford Foundation and the National Science Foundation (grant 88-13847 to Abowd). The authors thank George Bojas, Lawrence Katz, and Robert Topel for their help in preparing this paper. Daniel Kessler, Laura Leete, and Ana Revenga served as research assistants.
2
John M. Abowd and Richard B. Freeman
heavily affected by the movements in goods, the influx of immigrants, or foreign direct investment? What have we learned from studying the effects of these increased flows in the U.S. labor market? This paper presents background information about the growing internationalization of the U.S. labor market and summarizes the results of the NBER studies contained in this volume. The paper highlights several aspects of the internationalization of the U.S. work force. 1 . Although the number of immigrants relative to the population increased from the 1950s to the 1980s, the immigrant share of the growth of the work force was relatively moderate. The rapid growth of the labor force due to increased female participation and entry of the baby boom generation to the labor market kept pace with the influx of immigrant workers. 2 . The trade content of the U.S. economy, measured by exports plus imports relative to sales or GNP, has increased markedly, but the share of labor in traded sectors, notably manufacturing, has fallen, so that a smaller fraction of workers are directly affected by trade than in the past. Those workers are, however, more closely tied to world markets than in the past. 3. Direct foreign investment was substantial in the 1980s, reaching 34% of gross U.S. investment in 1988. Three percent of the private U.S. work force was employed in foreign-owned enterprises by the mid- 1980s. 4. The immigrant share of the labor force differs largely across geographic areas, whereas the trade share of product markets differs largely across industries. This motivates the research strategy for the studies of the effects of immigration and trade on the U.S. labor force: studies concerned with how immigration affects labor market outcomes contrast wages and employment in local areas with different immigrant shares of the work force; studies concerned with how trade affects labor market outcomes contrast wages and employment in industries with different trade shares of output. 5 . Industries are related to the open economy in a variety of ways. Industries in which there are considerable imports employ a disproportionate share of immigrants, whereas high export industries employ relatively few immigrants. Direct foreign investment is concentrated in manufacturing. Overall, the first-order effects of the internationalization of the labor market fall on manufacturing. 6. There are significant differences in the characteristics of workers between export-intensive, import-intensive, and immigrant-intensive sectors. Women workers and lower-paid, less-skilled workers are highly concentrated in sectors where imports are significant and where relatively many immigrants work. Perhaps surprisingly, foreign-owned enterprises have a comparable unionization rate to domestically owned enterprises: they are concentrated in traded goods sectors and have higher wages than domestic producers. The paper is divided into four parts. Section 1 deals with the aggregate
3
Introduction and Summary
flows of people, goods, and capital from overseas to the United States. Section 2 describes the industrial pattern and regional dimensions of the labor, goods, and capital flows as shown in the NBER data files developed for this project.' Section 3 turns to the characteristics of workers in sectors most affected by trade, immigration, and foreign direct investment. In sections 1-3, we have compiled statistics from a wide variety of sources in order to present comparable figures for 1960, 1970, 1980, and the most recent year available. Section 4 summarizes the findings of the papers included in this volume. 1. The Aggregate Flows of People, Goods, and Capital In this section, we review the basic data on each of the three flows under study: labor, goods, and capital. In contrast to demographic studies that focus on the immigrant share of the population and the increase in population, we focus on the immigrant share of the labor force and the increase in the labor force. In contrast to trade studies that focus on balance of payments issues, we focus on the proportion of workers in traded sectors and the ratio of exports plus imports to output in those sectors. In contrast to financial studies that consider international capital mobility broadly defined (and equal to imports minus exports by definition), we focus on direct foreign investment in plant and equipment. Flows of Labor Table 1 presents the basic data on the flows of immigrants entering the country (pt. A) and the stock of immigrants in the United States (pt. B) from the 1940s through the 1980s. The table provides figures for legal immigrant flows and legal plus estimated illegal immigrant flows in absolute numbers and relative to the population, labor force, and change in the labor force. The data in the first two columns of the table show that the number of immigrants coming into the United States and the number per one thousand inhabitants rose in the 1970s and 1980s, consistent with the increased public concern about immigration. When we consider the immigrant share of changes in the population and labor force, however, a different story emerges. Because the baby boom increase in the U.S. population occurs in the early postwar years (through 1960), the legal immigrant share of population growth is relatively small during this period. Because of the increased participation of women, the influx of baby boomers into the labor markets, and the fact that many legal immigrants enter for family unification reasons rather than for labor market reasons, the estimated immigrant flow share of the growth of the labor force
1. For a description of these data, see Abowd (in this volume).
4
John M. Abowd and Richard B. Freeman
Table 1
Flows and Stocks of Immigrants Relative to the Population
and Labor Force
A. Flows of Immimants
Period
DeCadal Flow of Immigrants (thousands)
Legal flows only:" 1941-50 1951-60 1961-70 1971-80 1981-90 Legal and illegal flowsb 1971-80 1981-90
Inflow Per 1,Ooo U.S. inhabitants
Immigrant Flow Share of Change: In Population (8) In Labor Force
1,035 2,515 3,322 4,493 5,900
.7 1.5 1.7 2.1 2.5
5.2 8.9 13.6 19.8 26.8
7.3 14.5 11.1 9.3 16.2
5,800 8,400
2.7 3.6
25.6 38.2
12.0 23.1
B . Stocks of Immigrants
Census of Population
Number of Foreign Born Counted (thousands)
As reported: 1940 1950 1960 1970 1980 Adjusted for undercount:c 1980
Population
Number of Foreign Born in Civilian Labor Force (thousands)
11,657 10,431 9,738 9,619 14,080
8.8 6.9 5.4 4.7 6.2
4,838 4,134 4,223 7,001
8.2 6.1 5.2 6.7
15,380
6.8
7,647
7.3
% of
Immigrant % of Labor
Force
Sources: Part A Flow of immigrants from U.S. Bureau of the Census, Statistical Abstract of the United States, 1989, table 7 (from the Statistical Yearbookof the Immigration and Naturalization Service), with the 1981-90 flow estimated by extrapolating the 1981-87 flows. Immigrant flow shares of changes were obtained by dividing flows by changes in decadal population from the relevant decades (tables in Council of Economic Advisers, Economic Report of the President, 1990). To obtain immigrants in the labor force, we assumed that the labor force participation rate of the decadal flow of immigrants was the same as the ratio of foreign-born workers in the civilian labor force to the foreign-born population (see pt. B). Part B: Foreign-born count and percentage of population from Sruristical Abstracr of the UnitedStates. 1988, table 44 (from the U.S. Census of Population). Foreign-born in the civilian labor force from various Censuses of Population. Qfficial counts from the Immigration and Naturalization Service summed over the indicated years. bAdjustedfor illegal flows using estimates from Borjas, Freeman, and Lang (in this volume) and Warren and Passel (1987), as described in the text. cAdjustedby adding the 1.3 million estimated uncounted illegal immigrants to the 1980 Census counts.
5
Introduction and Summary
actually falls from the 1950s to the 1970s, raising serious doubts about the labor market basis for concern over immigration until the 1980s. Then the number of immigrants rises substantially, and the contribution to both population and labor force growth reaches a postwar high. The figures in part B for actual counts of the stock of immigrants (which depend not only on inflows of immigrants but also on emigration and the death or retirement of persons who immigrated decades earlier and which include some illegal immigrants) tell a generally similar story. While declines in the immigrant share of the population and labor force are reversed for the decade 1971-80, the immigrant proportion of the population or labor force in 1980 remains below the 1950 proportion. What happens to this picture when adjustments are made for the widely publicized illegal immigration into the United States? We have made adjustments in the table based on the methods of Borjas, Freeman, and Lang (in this volume) and earlier research on illegal immigration (Warren and Passel 1987). The bases for our adjustments are Warren and Passel’s estimate that the 1980 Census included about two million illegal immigrants and Borjas, Freeman, and Lang’s estimate that approximately 6 1% of illegal (Mexican) immigrants were counted in the Census. Taken together, these estimates suggest that there were on the order of 3.3 million illegal aliens in the United States in 1980. Warren and Passel estimate that 75% of the illegals counted in the Census came in the 1970s. Assuming, conservatively, that 75% of the uncounted illegal immigrants also came in the 1970s, we get 2.5 million as the estimated flow of illegal immigrants in the 1970s. Adding this number to the number of legal immigrants reported by the Immigration and Naturalization Service (INS) in the rows giving “legal and illegal flows” changes greatly the picture of immigrant flows given in part A. Immigrant flows now rise sharply in the 1970s compared to the 1960s. Similarly, adding 1.3 million uncounted immigrants to the 1980 Census count raises the immigrant’s share of population and labor force in part B of the table to levels close to those of 1950. If illegal flows proceeded in the 1980s at the same rate as in the 1970s, then, given the INS estimates of legal immigrant flows, we estimate that some 8,400,000 immigrants came to the United States in the 1980s. This raises the immigrant inflow per one thousand United States inhabitants and the immigrant share of the change in population and labor force above the levels of the 1970s. The 1970s and 1980s were periods of marked acceleration in immigration, in large part because of illegal flows. Another aspect of the flow of immigrants to the United States deserves attention. The change in the geographic origins of immigrants following the 1965 Immigration Act has produced a dramatic shift in immigrant origins from Europe and Canada to Asia. Figure 1 illustrates this change. If we adjusted the proportions in the figure for illegal immigrants (largely Mexican), the share from Latin America would also rise.
6
John M. Abowd and Richard B. Freeman
Fig. 1 Distribution of immigrant origins Source: Immigration and Naturalization Service Statistical Yearbook, 1987, table 2,
“Immigration by Region and Selected Country of Last Residence.”
Flows of Goods Figure 2 shows the widely heralded increase in the role of trade in the U.S. economy in terms of two related measures-exports ( X ) plus imports (M) relative to GNP, which we will call the trade content of the economy and the ratio of the trade balance (exports minus imports) to GNP. In the 1950s and 1960s, the overall trade content of GNP was roughly 10%-11%, with U.S. exports exceeding imports. In the 1970s, the trade content jumped, particularly after 1978, reaching a peak in 1981, then hovered around this level for the rest of the decade. The balance of trade diverged modestly from year to year until 1983, when it became negative. Large negative trade balances characterize the rest of the decade and are unlike any other postwar period. While the trade content of the U. S. economy has risen sharply, the proportion of workers employed in the traded goods sectors-manufacturing , mining (including crude oil), and agriculture-has fallen, so that relatively fewer workers are directly imported by foreign competition. Table 2 shows the ratio of exports plus imports to sectorul GNP for traded goods (agriculture, mining, and manufacturing), all other sectors, and the entire U.S. economy for the
Introduction and Summary
7
a,
Y
8 -
a
6 4 -
-4
-
,
-6
+
I
I
,
,
, ,
1
,
1
1
1
1
1
1
1
,
1
,
1
1
1
,
I
,
I
,
,
r , , , ,I
,
(Exports - Irnports)/GNP
Fig. 2 Openness of the U.S. economy Source: U.S. National Income and Product Accounts, July 1989.
years 1960, 1970, 1980, and 1987. The table also shows the percentage of GNP originating in the sector and the percentage of full-time equivalent employment in the sector. Exports plus imports as a percentage of sectoral GNP rise sharply in the traded goods sectors, but the share of GNP and the share of employment in the traded goods sectors fall. Whereas in 1960 33% of the work force and 35% of GNP were in the traded goods sectors, by 1987 only 21% of employment and 23% of GNP were in those sectors. A smaller fraction of the labor force is directly affected by foreign competition by 1987 than in the earlier decades. The table also shows the employment-weighted exports plus imports as a percentage of sectoral GNP (last row) and the comparable ratio for the overall economy (“total all sectors”). The traded portion of the entire U.S. economy (goods and services) rose from 10% in 1960 to 22% in 1987 by either overall measure (also shown in fig. 2). The economy-wide trade ratios rise by much less than the ratios in the traded goods sectors. In terms of direct competition from foreign-produced goods, a decreasing proportion of the labor force faces the consequences of increased traded goods flOWS.2
2. This assumes that exports plus imports is a good measure of trade dependence. Under some circumstances it will be. Under others it may understate trade dependence: e.g., when prices are determined by the world market but there are no trade flows.
8
John M. Abowd and Richard B. Freeman The Changing lkade Content of the U.S. Labor Market (%)
Table 2
1960
1970
1980
1987
28.6 22.6 17.5 19.2 5.8 10.5
27.8 25.0 26.6 26.6 6.9 12.7
53.5 76.6 56.8 59.3 11.0 24.5
33.8 45.3 64.4 60.0 11.2 22.3
4.2 2.5 28.0 34.7 65.3
2.9 1.8 24.8 29.6 70.4
2.8 3.9 21.3 28.0 72.0
2.1 1.9 18.9 22.8 77.2
3.1 1.2 28.6 33.0 67.0
1.8 0.9 26.6 29.2 70.8
1.8 1.2 22.8 25.7 74.3
1.6 0.7 19.0 21.3 78.7
10.1
12.7
22.9
21.8
+
(Exports Imports)/GNPin sector? Agricultureb Mining' Manufacturing Total traded goods' All other sectorsf Total all sectors' Percentage of GNP in sector: Agriculture Mining Manufacturing Total traded goods All other sectors Percentage of employment in sector:h Agriculture Mining Manufacturing Total Traded Goods All other sectors (Exports Imports)/GNP in sector (employment weighted)'
+
Sources: Exports and imports 1960, 1970, and 1980 from Bureau of the Census, U.S. Commodiiy Exports and Imports as Related to Output 1981/80 (1983), table A. Exports and imports 1987 from U.S. Department of Commerce online data base of official statistics. GNP in sector from National Income and Product Accounts, table 6.1 (extracted from CITIBASE). Full-time employment in sector from National Income and Product Accounts, table 6.7B (extracted from CITIBASE).
*Exportsplus imports as a percentage of GNP originating in the industry group. bAgricultureis SIC industry groups 01-09. cMiningis SIC industry groups 10-14. dManufactureis SIC industry groups 20-39. Traded goods are agriculture, mining, and manufactures. 'All other sectors include SIC industry groups 15-17 and 40-99. Exports (imports) in all other sectors are defined as the difference between total exports (imports) and traded goods exports (imports). SExports and imports from the National Income and Product Accounts. Traded goods sectors consist of manufacturing (SIC 20-39), mining (SIC 10-14), and agriculture (SIC 01-09). hFull-timeequivalent employees from the National Income and Product Accounts. 'Exports plus imports as a percentage of GNP originating in the industry group weighted by employment in the industry group.
Capital Flows The flow of capital across international borders is the most difficult flow to measure and analyze. Net capital flows should equal the balance on current accounts (plus allocations of special drawing rights), but, in fact, the two differ significantly, requiring a statistical discrepancy line to produce the definitional equality. In terms of the effects on labor markets, we want to distin-
9
Introduction and Summary
guish a foreign capital investment that is a long-term job creating flow from a short-run financial flow. If all net capital flows were of the former kind, public focus on the disemployment effects of an imbalance on the current account would be erroneous. If all the net capital flows were of the latter kind, by contrast, such concerns might be valid, although the imbalance would eventually alter the real exchange rate and, in principle, correct itself. It is not easy, however, to determine the degree to which capital flows fall along a spectrum from long-term job-creating to short-term financial flows. Presumably, direct foreign investment is job creating, while currency transactions are likely to be short run, though we still need to know the “motive” and likely holding period of these intermediate investments. A foreigner who buys stocks, corporate bonds, or U.S. Treasury obligations or even leaves money in a U.S. bank account for a long time can, through the flow of funds, produce as much longterm investment in the United States as a foreigner who builds a plant. We distinguish in table 3 between direct foreign investments in plant and equipment, likely to be long run, and other forms of capital flows. As can be seen in the table, both direct and indirect capital flows increased dramatically in recent years. Net U.S. investment abroad (the change in U.S. assets abroad from the international transactions accounts) increased from $4,099 million in 1960 (shown as a negative number in the table to reflect a capital outflow) to $82,110 million in 1988. Net foreign investment in the United States (the change in foreign-owned assets from the international transactions accounts) increased from $2,294 million in 1960 to $219,299 million in 1988. Direct U.S. investment abroad and direct foreign investment in the United States also increased dramatically since 1960. By 1988, over a quarter of foreign investment in the United States consisted of direct foreign investment. Are the international capital flows sizable or negligible in the context of the U.S. economy? Table 3 also compares net foreign investment in the United States and direct foreign investment in the United States to GNP and U.S. gross investment. Direct foreign investment in the United States rises from . l % of GNP and .4% of gross investment in 1960 to 1.2% of GNP and 9.2% of gross investment in 1988. While Japanese investment in the United States has received the most public attention, the percentage of direct foreign investment by country of ultimate beneficial ownership in table 3 shows that European direct investment is quantitatively much larger, although Japan increased its share dramatically in the late 1980s.
2. Industrial and Geographic Patterns Flows of goods, people, and capital occur differently by sector and area of the economy. Some industries produced traded goods, while others do not. Immigrants are overrepresented in some sectors and underrepresented in others, and immigrants go to some areas of the country, and not to others. For some long-term general equilibrium purposes, the sectoral division of the
10 Table 3
John M. Abowd and Richard B. Freeman Capital Market Flows between the United States and the Rest of the World Investments (millions of dollars)
Net U.S. investment abroad' Direct investment abroad Net foreign investment in the United Statesb Direct foreign investment Investment outlaysc
1960
1970
1980
1988
- 4,099
- 9,337 - 7,590
- 86,118
-82,110
- 2,940
- 19,222
- 17,533
2,294 315 NA
6,359 1,464 NA
58,112 16,918 12,172
219,299 58,436 65,019
Relative Figures (% of base) 1960
Net foreign investment in the United StatesiGNP Net foreign investment in the United States/gross investmentd Direct foreign investment'GNP Direct foreign investment'gross investment Percentage of direct foreign investment by country' Canada Japan Europe West Germany The Netherlands United Kingdom Rest of the World
1970
1980
1988
.4
.6
2.1
4.5
2.8 .1
4.1
.I
12.9 .6
34.7 1.2
.4
I .o
3.8
9.2
100.0 NA NA NA NA NA NA NA
100.0 NA NA NA NA NA NA NA
100.0 16.1 4.9 62.9 11.7 13.6 25.2 16.1
100.0 16.0 21.8 52.5 2.1 3.0 33.1 9.7
Sources: U.S. International Transactions accounts from the Survey of Current Business (June 1989). Percentage distribution by country of ownership from Survey of Current Business, U.S. Business Enterprises Acquired or Established by Foreign Direct Investors, 1980 and 1988. National Income and Product Account data extracted from CITIBASE. Note: NA = not available on a comparable basis. *From U.S. assets abroad, net (increaseicapital outflow [ -1). in the U.S. International Transactions accounts. Negative numbers indicate a net outflow. Direct investments abroad is a subaccount of U.S. private assets, net. bFrom foreign assets in the United States, net (increaselcapital inflow [+I), in the U.S. International Transactions accounts. Positive numbers indicate a net inflow. Direct foreign investments is a subaccount of other foreign assets in the United States, net. CInvestmentoutlays from the U.S. Department of Commerce, Bureau of Economic Analysis, Survey of New Foreign Direct Investment in the United States (1983). Cross investment series from the U.S. National Income and Product Accounts, annual data. 'Percentage of Bureau of Economic Analysis survey investment outlays by country of ultimate beneficial owner. Figures for 1988 are preliminary.
11
Introduction and Summary
flows is unimportant. For many short- and intermediate-term questions, however, sectoral flows are critical. To deal with this issue, the NBER developed the Immigration, Trade, and Labor Markets Data Files (see Abowd, in this volume). These data allow us to examine the pattern of trade across industry lines over time, to contrast the industrial distribution of trade and the employment of immigrants, to determine the characteristics of workers in industries with more or less trade and with sizable or limited employment of immigrants, and to compare the geographic and industrial patterns of trade and immigration effects. Table 4 uses the NBER immigration and trade data files to assess the variability of trade ratios across manufacturing industries. The table records the mean, standard deviation, and coefficient of variation of trade ratios among the 450 four-digit SIC manufacturing industries and of immigration ratios Table 4
Variation in Trade and Immigratino Ratios for Manufacturing Industries (employment weighted)
+
(Exports (Exports Import/ Imports)/ Imports)/ Shipments New Supply Shipments Shipments
Export/
1960: Mean Standard deviation Coefficient of variation 1970: Mean Standard deviation Coefficient of variation 1980: Mean Standard deviation Coefficient of variation 1985: Mean Standard deviation Coefficient of variation Change, 196W30: Mean Standard deviation
(%)
(%'.)"
4.27 5.91
2.30 4.61
7.00 10.17
1.38
2.01
1.45
5.62 7.07
4.59 6.11
11.08 12.91
1.26
1.33
1.17
10.31 11.38
7.41 8.51
19.94 23.59
1.10
1.14
1.18
8.48 11.09
10.94 11.38
24.95 52.17
1.31
1.04
2.11
5.50 8.37
5.38 6.42
12.62 17.34
Immigrants/ Labor Force (%)
1.56 9.87
8.48 3.66 .43
.20 11.40
7.13 3.37 .47
.75 18.56
1.96 4.34 .55
- 7.95 50.78
- 1.58 14.03
- .40 2.82
Source: NBER Immigration, Trade, and Labor Markets Data Files (see Abowd, in this volume). Nofe: All ratios are stated as percentages of the relevant base. The statistics are averages over four-digit SIC industries using the annual employment in the industry as the weight. There are 450 SICs with valid immigrant ratio data and 430 SICs with valid import and export data. *New supply is the sum of shipments and imports.
12
John M. Abowd and Richard B. Freeman
among those industries. The table shows considerable variation in ratios of trade (functions of exports X and imports M) to shipments (S) across industries. The relatively stable coefficient of variation in the exports-to-shipments ratio contrasts with a declining coefficient of variation in the imports-tc+new supply ratio, implying an unchanged concentration of the former compared to an increasing concentration of the latter. Among the sectors with the largest increase in trade content are footwear except rubber, electrical equipment, and electronic resistors. Imports grew especially rapidly in footwear, and exports grew especially rapidly in electrical equipment. Both exports and imports increased in electronic resistors. Comparing the trade and immigration ratios across industries, one striking fact emerges: trade ratios are much more variable among sectors than are immigration ratios. This has important consequences for the way in which NBER and other researchers study the effects of trade on the labor market: focusing on differences across industries. A very different pattern emerges when we consider regional differences in trade ratios, immigrant flows, and foreign direct investment in the United States. Here we find exactly the opposite: immigration ratios vary much more across regions than across industries. This fact leads NBER and other researchers to study the effect of immigration on the labor market by focusing on differences across areas. The geographic concentration of immigration is documented in table 5. The table shows INS figures on the number of immigrants declaring selected standard metropolitan statistical areas (SMSAs) as their intended residence from 1976 to 1979 and the contribution of that flow to the growth of the labor force. Table 5 displays the ten SMSAs with the largest percentage of foreign born in the area. These gateway cities absorbed a very substantial fraction of all immigrants who entered the United States in the four-year period illustrated (comparable data were not collected for 1980). Table 6 shows Census of Population figures on the percentage of the labor force that are immigrants by SMSAs in 1970 and 1980. What stands out in these tables is the substantial concentration of immigrant flows by SMSA. How might the flow of illegal immigrants into the United States change the picture of geographic concentration shown in tables 5 and 6? Given the concentration of illegal aliens in California, where Warren and Passel estimate that 50% of illegals counted in the Census are located, the concentration of immigrants would become even more dramatic. Table 7 presents data on all our flows by state. Columns 1-3 give data from the Census of Population for immigrants as a percentage of the population in 1970 and 1980 and in 1980 adjusted for the likely undercount of illegals in the Census on a state-by-state basis using the estimates in Passel and Woodrow (1984). While the geographic diffusion of the stock of immigrants is lower than the diffusion of the flow of new immigrants among SMSAs, there is still considerable variation across areas. Columns 4-6 of the table turn from immigrant to trade figures. As data on
13
Introduction and Summary
Table 5
SMSA Miami Los Angeles New York City El Paso Newark Washington, D.C. Houston Cleveland Philadelphia Dallas
Flows of Immigrants into Selected Standard Metropolitan Statistical Areas (SMSAs) Immigrants Declaring SMSA as Intended Place of Residence’
Change in Labor Force from 1976 to 1979b
Estimated Immigrant Contribution to Labor Force Growth (%)
79,099 74,515 247,052 13,053 8,879 8,359 23,868 3,800 10,571 10,735
54,233 254,000 38,000 8,836 40,738 166,193 255,367 38,108 85,016 220,331
73.3 14.7 326.8 74.3 11.0 2.5 4.7 5.0 6.3 2.4
Sources: Number of immigrants from the Statistical Yearbook of the Immigration and Naturalization Service, 1976-79: table number varies; table title “Immigrants Admitted by Specified Countries of Birth and Rural and Urban Area and City.” Change in the labor force from the Bureau of the Labor Statistics, Employment and Earnings, various issues. The immigrant labor force-tc+immigrant population ratio was estimated from the 1980 Census of Population Detailed Population Characteristics U.S. Summary, sec. A-U.S. PCSO-1-D1-A. Total immigrants is from table 254, “Citizenship and Year of Immigration for Foreign Born Persons by Country of Birth.” Immigrants in the labor force is from table 255, “Selected Economic and Social Characteristics by Nativity.” ‘Number of immigrants who declared the SMSA as the intended place of permanent residence during the period from 1 October 1975 to 30 September 1979. SMSAs are listed in descending order of percentage foreign born in the area. Thange in the size of the labor force from 1976 to 1979, inclusive. CEstimated as 50.3% of col. 1 divided by col. 2.
exports or imports by geographic location are unavailable, our estimates of the trade content of a state’s industry mix are obtained by weighting industry trade ratios according to the industrial distribution of state labor forces as follows: T, = C W i j T i , i
where T = relevant state trade ratio, W , = proportion of workers in state j who work in industry i, and Ti = national trade ratio in industry i. In contrast to the wide variation in immigration ratios across states, the trade ratios differ relatively moderately, except for the net export ratio ([X - M ] / S ) . For example, the five states whose industry structures have the highest import ratios ( M / [ S M I ) have an average value of 8.8, compared to an average figure of 3.8 for the five states with the lowest import ratios. While there are surely individual localities that are greatly sensitive to trade, the implication of the table is that trade flows are unlikely to have great effects on local labor markets, except, possibly, where there is a substantial net export ratio ([X - M ] / S ) , as in Alaska.
+
14
Table 6
John M. Abowd and Richard B. Freeman Immigrants as a Percentage of the Labor Force Selected SMSAs ~
Atlanta Baltimore Boston Chicago Dallas-Fort Worth Detroit Houston Los Angeles Miami New York City Philadelphia Pittsburgh St. Louis San Francisco Washington, D.C.
~~
~
I970
1980
1.3 3.9 10.3 10.2 2.4 8.4 3.3 13.6 27.9 18.0 6.1 4.3 2.3 13.7 6.0
2.9 3.5 10.6 11.6 4.8 6.3 8.3 24.2 41.2 24.0 5.0 2.7 2.3 16.2 9.0
~~~
Sources: Based on individual data from the 1970 Census of Population and Housing 1/100 Public Use County Group Sample and the 1980 Census of Population and Housing Public Use Microdata A Sample. Note: The numerator is the number of immigrants in the labor force in the SMSA indicated. The denominator is the number of individuals in the labor force in the SMSA. SMSA definitions in the 1970 and 1980 Censuses of Population were made comparable by selecting the appropriate area and subarea codes (1970) and SMSA codes (1980).
The popular and business press are filled with stories about the decision of Japanese and other foreign investors to locate plants in certain regions of the country as opposed to others. Column 7 of table 7 presents data from the 1980 Benchmark Survey of Direct Foreign Investment in the United States (U.S. Department of Commerce 1983) on the proportion of the private work force employed in foreign-owned enterprises among the states. It shows considerable variation in employment in foreign-owned affiliates, with a range far exceeding that for trade shares, and a regional pattern differing greatly from that for immigrant employment.
3. Characteristics of Workers in Sectors Affected by Internationalization To evaluate the type of workers most likely to be affected by trade or immigration, we have performed a two-part analysis. First, we tabulated the average characteristics of workers by employment. Second, we calculated correlation coefficients between worker characteristics by industry and the relevant trade or immigrant worker ratio. Table 8 presents the results of the first analysis with sectors divided between traded goods and nontraded goods, between export- and import-intensive
15
Introduction and Summary
Table 7
Geographic Distributionof Immigration, Bade, and Direct Foreign Investment Immigrant % of Labor Force
Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware District of Columbia Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee (conrinued)
% of U.S. % Net
Ratio
Export Ratio X - M/S
Affiliate Employment in Total Private Employmentb
%
1970
1980
Adjusted 1988
Export Ratio XIS
.7 5.9 5.4 .5 11.4 3.8 10.6 2.7
1.1 4.7 6.1 1.1 16.3 4.0 8.9 3.1
1.2 4.9 6.9 1.2 19.3 4.5 9.0 3.2
8.6 14.9 13.8 8.4 11.8 10.4 14.8 12.3
7.1 2.9 6.6 7.5 6.4 6.7 7.2 6.6
.8 11.7 6.5 -.2 4.8 2.9 6.8 4.9
2.0 6.9 1.7 2.2 2.5 1.9 2.7 3.9
8.0 10.0 1.2 11.4 1.8 7.3 2.2 1.8 1.3 .2 1.4 5.8 4.4 10.6 6.7 3.1 .2 1.8 1.6 1.4 6.2 5.1 10.6 3.3 13.9 .8 2.7 3.5 1.1 4.4 4.1 8.8 .7 1.2 1.1
6.9 11.5 1.9 15.3 3.0 8.0 1.9 1.5 2.1 .9 2.3 3.5 5.0 8.5 4.3 2.3 1.0 1.8 2.2 1.9 7.9 4.3 11.2 4.1 14.8 1.4 2.0 2.7 2.0 4.1 3.1 8.8 1.6 1.3 1.2
8.7 12.2 2.1 15.2 3.5 8.9 2.0 1.6 2.3 1.1 2.4 3.4 5.7 8.8 4.4 2.4 1.1 1.9 2.2 2.1 8.6 4.3 11.6 4.9 15.8 1.5 2.1 2.8 2.3 4.5 3.1 9.0 1.7 1.3 1.3
2.5 10.2 9.2 5.5 9.4 11.5 10.3 13.1 14.6 9.9 9.1 8.8 8.2 12.1 11.3 12.5 8.1 10.1 11.4 9.8 8.6 13.4 9.6 7.5 10.7 7.7 8.2 10.9 11.6 11.0 9.1 11.1 9.5 9.4 9.0
1.6 6.2 7.2 7.1 5.4 6.8 8.2 6.6 6.0 7.2 6.1 9.4 6.6 7.6 10.4 6.0 7.6 7.5 4.8 6.6 7.6 7.8 6.4 7.6 1.3 6.2 5.5 7.5 6.7 6.4 7.3 8.1 6.8 4.5 7.7
.8 3.2 1.2 - 2.6 3.3 4.0
.6 2.1 3.6 4.6 1.3 2.8 2.6 2.0 1.7 2.3 2.9 2.7 3.1 4.0 2.3 2.0 1.4 1.8 .8 1.o 1.3 3.6 4.6 2.0 2.9 3.2 1.2 2.2 2.0 1.2 2.8 2.0 5.3 .6 2.9
% Import
MIS
+M
1.1
5.8 7.9 1.9 2.4 -2.3 .8 3.5 - .9 5.9 - .4 1.5 6.2 2.5 - .2 4.6 2.4 - 1.2 2.5 .9 2.1 2.4 4.2 3.9 1.o 2.0 2.0 4.3 .3
16
John M. Abowd and Richard B. Freeman
Table 7
(continued) Immigrant % of Labor Force
Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming Summary: Mean of top 5 Mean of bottom 5 Difference
% Net
%
Adjusted 198W
Export Ratio XIS
% Import
Ratio MIS + M
Export Ratio X - MIS
% o f U.S. Affiliate Employment in Total Private Employmentb
1970
1980
3.5 2.9 7.4 2.4 5.9 1.2 3.3 1.6
6.5 4.0 4.5 3.7 6.1 1.1 2.5 1.9
7.6 4.5 4.5 4.2 6.5 1.1 2.6 2.1
12.0 11.8 14.2 8.5 15.2 10.5 11.5 9.6
6.6 6.6 6.5 6.6 6.1 7.1 7.1 6.0
4.7 4.5 7.1 1.3 8.5 2.5 3.5 3.0
2.7 2.2 3.6 2.2 1.5 3.6 3.4 1.7
13.5
13.8
14.8
14.7
8.8
8.4
5.1
.5 13.0
1.0 12.8
1.1 13.7
6.3 8.4
3.8 5.0
1.5 9.9
1.o 4.2
-
Sources: Immigrant, export, and import ratios are from the NBER Immigration, Trade, and Labor Markets Data Files (see Abowd, in this volume). Employment in U.S. affiliates is from U.S. Department of Commerce (1983), Foreign Direct Investment in the United States, 1980. T h e adjustment is based on Passel and Woodrow’s (1984) table 1, “Estimates of Undocumented Aliens Counted in the 1980 Census and Legally Resident Aliens by State of Residence and Period of Entry.” bFrom the 1980 Benchmark Survey of Foreign Direct Investment in the United States, “U.S. Affiliate Employment by State.”
manufacturing (reported separately for high-import and high-export industries), and by high and low immigrant-worker ratios. The differences between traded and nontraded sectors reflect basic economic differences between characteristics of workers in goods and services industries: workers in nontraded goods are better educated, younger, more likely to be female, and less likely to be union than workers in traded goods. In addition, wages and GNP per worker in this sector are lower than in traded goods. While immigrant ratios are nonnegligible in nontraded sectors, they are lower than in the traded goods sector, indicating that the traded goods sector is more directly tied to the international economy by flows of people as well as by flows of goods. Decomposing manufacturing into high (top quintile) export and import to shipments sectors, we find striking differences in the characteristics of the work forces. These differences indicate which workers are more or less likely to be directly positively or negatively affected by trade. The principal differences among workers revealed by the table are that export sectors have proportionately more educated workers, fewer blacks, and strikingly fewer female workers than import-intensive industries; that high-export manufactur-
17
Introduction and Summary
Table 8
Average Characteristics of the Labor Force in 1980 (industry averages weighted by employment) Production Workers Earnings With W o Years of Black 16-24 Female Immigrant Who Are per College (8)(%) (%) (%) (%) Union(%) Worker
Traded goods Manufacturing High exports (top 20) High imports (top 20) Nontraded goods and services All industries Low immigrant /labor force High immigrant /labor force
18.7 19.3
9.3 21.4 10.2 20.2
32.4 35.9
7.8 8.2
36.8 43.7
25.5
1.5
18.3
29.9
7.5
41.6
15.4
10.9
20.2
44.8
10.4
40.6
31.2 28.6
10.2 24.9 10.0 24.1
50.6 46.8
6.1 6.5
21.8 25.0
36.2
9.4
17.4
41.2
3.4
33.0
20.9
14.1
22.6
52.4
12.0
24.1
Value Added per Worker
16.9 17.4
32.6 28.6
13.0
21.8
~
Source: Calculated from the NBER Immigration, Trade, and Labor Markets Data Files (see Abowd, in this volume). Thousands of dollars per worker.
ing sectors have lower proportions of immigrants than high-import manufacturing (though their ratio still exceeds the economy-wide average). The fraction of blue-collar workers unionized in the sectors does not, by contrast, show any noticeable differences. Turning to the characteristics of workers by immigrant ratios, we find that high-immigrant-ratio sectors tend to have less educated workers, relatively more blacks, relatively more women, relatively more young workers, and relatively fewer union workers. With the exception of the unionization pattern, these differences mirror those between export- and import-intensive industries in manufacturing. Table 9 records the correlation coefficients between mean characteristics of workers and export and import ratios and the net export ratio in manufacturing and between the mean characteristics of workers and immigrant ratios in manufacturing, nonmanufacturing, and all industries. The correlations confirm the evidence given in table 8, revealing a strikingly high positive correlation between the percentage of workers who are women and the percentage of workers who are immigrants in manufacturing industries. The correlations also show that female, black, and immigrant workers tend to be concentrated in industries with negative net exports and that educated workers tended to be in industries with positive net exports. These calculations suggest that both trade and immigrant flows may have an especially large effect on the female work force, especially in manufacturing.
John M. Abowd and Richard B. Freeman
18
Table 9
Correlation Coefficients for Immigration and lkade Ratios with Various Labor Force Characteristics, 1980 (employment weighted) ~~~~
Nonmanufacturing
Manufacturing Only
XIS 2 Years of college % Black % 16-24 % Female % Immigrant % Unionized Eamingsiworker Shipmentsiworker
.31 - .25 -.29 - .22 -.16 .05 .32 .I3
Mi(S
+ M)
All Industries
Immigrant/ Immigrant! Immigrant/ Labor Force (X - M ) / S Labor Force Labor Force
-.I3 .08
-.14 .I1 .22
.oo
-.06 -.09
- .21
.38
- .01
.06 .I6
- .29
.I2 .29 .20
.I1
1 .oo - .36 - .53 - .29
- .18 - .32 - .30 .05 .35
- .35
- .21 .15 .18 .25 -.18
.I9
Source: NBER Immigration, Trade, and Labor Markets Data Files (see Abowd, in this volume). Note: The statistics are painvise correlation coefficients computed using the percentage of total employment in the industry as the weight.
Finally, we consider the characteristics of workers in foreign- and U.S.: owned businesses operating in the United States. Table 10 displays a collection of comparisons from the 1974, 1980, and 1987 Benchmark Surveys of Foreign Direct Investment. The 1980 survey is the most detailed, and it reveals that employees of nonbank U.S. affiliates of foreign companies are about as likely as employees of U.S.-owned companies to be unionized. Further, 1980 hourly earnings levels are somewhat higher, though sales per employee levels are similar in affiliates. Because the benchmark surveys are not comparable in the universe (banking affiliates are included in 1974 but excluded in 1980 and 1984) and in the summary data tables (employment is not reported by industry in 1974), it is difficult to discern trends in the comparisons of foreign-owned to U.S.-owned businesses. It seems likely that the differences are not substantial, and this conclusion is supported by other research (Leonard and McCulloch, in this volume). 4.
Findings of the NBER Project
Motivated by the internationalization of the American labor market described in sections 1-3, the NBER undertook the research project whose results are given in this volume. The first part of the project studied the factors that influence the number and characteristics of immigrants and their location in the country, including the undocumented, largely Mexican aliens who have aroused so much public concern. The second part of the project examined how immigration and trade affect the wages and employment of American workers. The third part of the project added an international comparative dimen-
19
Introduction and Summary
Table 10
Selected Business and Employment Characteristicsof U.S. Affiliates of Foreign Companies, by Industry
Number of employees (thousands)’ Percentage of civilian labor force Percentage union in U.S. affiliates Percentage union in all U.S. businesses Average hourly earnings for production work: Workers in U.S. affiliates (manufacturing) Workers in all U.S. businesses (manufacturing) Sales per employee (thousands) of dollars):b U.S. manufacturing affiliates U.S. manufacturing businesses Percentage of total employment: Traded goodsc manufacturing Selected nontraded goodsd
1974
1980
1987
1,083 1.2
2,034 1.9 29.3 25.2
3,160 2.6
7.85 7.27
3.3 2.8
9.9
88.86 91.27
137.65 125.63
5.4 5.0 3.0
1.7 8.4 3.6
Sources: Survey of Current Business, “Benchmark Survey of Foreign Direct Investment in the United States, 1974” (May 1976). U.S. Department of Commerce (1983), Foreign Direct Investment in the United States, 1980. Survey of Current Business, “U.S. Affiliates of Foreign Companies: 1987 Benchmark Survey Results” (July 1989). U.S. shipments and employment data from Survey of Current Business, various issues. ”11 U.S. affiliates in 1974; nonbank U.S. affiliates in 1980, 1987. bManufacturingindustries only, thousands of dollars per employee. ‘Agriculture, mining, and manufacturing only. dWholesale trade, retail trade, finance (except banks), insurance, and real estate.
sion by studying immigration, trade, and the labor market in two other countries that are major immigrant recipients, Canada and Australia. Canada is of particular interest for several reasons: Canadian immigration policies historically have stressed job skills as a condition for entry to a greater extent than the United States does; Canada had a significant balance of trade surplus with the United States in the 1980s; and Canada has long depended on foreign (largely U.S.) capital to employ a large share of its work force. Australia is of interest because immigrants constitute an exceptionally large proportion of its work force, raising issues about how immigration affects the macroeconomy, and because its protectionist trade policies contrast with the free trade policies of the United States. The differing experiences among the United States, Canada, and Australia indicate the degree to which different labor market institutions and economic policies can condition the effect of immigration and trade on economic outcomes. The project researchers used different strategies to study the flow of immigrants, the effects of immigration on labor market outcomes, and the effects of trade on those outcomes. The studies that focus on the flow of immigrants
20
John M. Abowd and Richard B. Freeman
compare the labor market and migration behavior of individual immigrants since it is the immigrant decisions and performance in the labor market that are at issue. The studies of the effects of immigration on wages and employment compare local labor markets that have different immigrant shares in the work force. The principal reason for focusing on local markets is that immigrants are concentrated by geographic area, constituting large and increasing proportions of the work force in gateway cities but negligible proportions elsewhere, as indicated in tables 5 and 6. By contrast, the studies focusing on trade examined the effect of trade on industry labor markets. This is because the export and import components of economic activity vary and change greatly among industries, suggesting that the first-order labor market effects of trade occur at the industry level. Most of the findings are based on data from government surveys such as the Census of Population and the Census of Manufactures. To answer certain questions, however, researchers developed new data sets, ranging from one that links import prices to collective bargaining contracts in Canada (Abowd and Lemieux) to a survey of illegal Mexican immigrants in the San Diego area (Borjas, Freeman, and Lang). Because trade, immigration, and labor market data are collected using different standards by diverse government surveys, researchers developed the industry-based trade and labor markets data file for U.S. manufacturing industries from the 1950s through the 1980s and the areabased immigration and labor market data file for the 1960s, 1970s and 1980s (Abowd; Altonji and Card; and LaLonde and Topel). Studying immigration and trade by comparing outcomes across individuals, areas, or industries differs from most studies in international trade, where researchers use general equilibrium models to make inferences about the economic effects of immigration, trade, and capital flows. While there is no inherent conflict between these two types of research approaches (some of the studies use input-output and trade models; (e.g., Kuhn and Wooton; Collins), our decision to concentrate on individuals and markets was a conscious one that conditions the issues we address and our major findings. Our approach pins down the first-order effects of trade and immigration on the economic well-being of the groups most affected by the internationalization of the U.S. labor market but does not yield estimates of the broader benefits of trade or immigration to the overall society. The approach has the advantage of basing inferences on the great variation in the experiences of individuals, areas, or industries and of requiring less formal structure than general equilibrium analyses, at the cost of being unable to answer questions about how things may work out for the society as a whole in the long run. As a broad generalization, the American labor market adjusted well to immigrant flows, absorbing immigrants into local area work forces with little redistributive losses to natives, but it had greater difficulty adjusting to the surge of imports, which produced some noticeable losses to natives in affected industries. Still, industry wages were as flexible to changes caused by trade as
21
Introduction and Summary
to changes caused by domestic factors, falling where imports reduced domestic production and thus buffering employment to some extent. By contrast, in Australia, industry wage responsiveness to imports was limited, and the government sought to protect labor through import restrictions. The research highlights the supply responsiveness of immigrants to economic and political conditions and to foreign as well as to American immigration policies in the context of a “world market for immigrants.” Immigrant Flows 1. The flow of illegal immigrants to the United States, while sizable, falls far short of the huge numbers often reported in the media. NBER estimates of Mexican-born illegal immigrants based on the number of deaths and births of Mexican-born persons in the United States, Mexican surveys of returned migrants, and analyses of apprehension statistics that take account of the fact that apprehensions are determined by Border Patrol activity as well as by immigrant flows support the claim of Census Bureau demographers that the 1980 Census enumerated over half the illegal immigrants. The number of illegal Mexican immigrants in 1980 was on the order of two million rather than ten to twelve million. Moreover, most of the likely illegal Mexican immigrants counted in the Census have a family composition and type. of employment similar to those of legal immigrants (Borjas, Freeman, and Lang; DiezCanedo). Consistent with a factor endowment explanation of immigration, most illegal aliens are unskilled. 2. The characteristics of immigrants are influenced significantly by the economic and political situation in the home countries and by the attractiveness of the United States in the “world market for immigrants,” where the United States competes with other immigrant-recipient countries such as Canada and Australia. All else the same, workers with a high earnings potential are especially likely to migrate to the United States from a country with an egalitarian wage structure (where they cannot easily make high earnings), while workers with a low earnings potential are especially likely to migrate from a country with great wage inequality. The 1965 changes in U.S. immigration policy produced a wave of immigrants whose labor market skills were lower relative to those of native Americans than was true of earlier waves of immigrants, who did especially well in the labor market relative to natives (Borjas). Changes in Canadian immigration laws produced a similar pattern of declining skills in the late 1970s. Australia, by contrast, attracted immigrants who did well compared to natives through 1980 (Borjas; Beggs and Chapman; Bloom and Gunderson). 3. New immigrants to the United States are as mobile across geographic areas as natives, on average, but their mobility has not led them to spread out across the country. Instead, they move to cities where their fellow countrymen reside in large numbers. The tendency of immigrants to cluster dominates such economic incentives as differences in unemployment rates or welfare
22
John M. Abowd and Richard B. Freeman
benefits across areas in determining immigrant migration flows (Bartel and Koch). Cuban, Mexican, and Puerto Rican immigrants and natives who move from cities with a high proportion of persons of their ethnic background to cities with a low proportion of persons with their ethnic background have roughly similar earnings and employment experiences as their peers who move from cities with a low proportion of persons of their ethnic background to cities with a high proportion of persons of that ethnic background (Tienda and Wilson). The direct advantages and costs of immigration thus continue to be borne by gateway cities, while the persistent geographic concentration of immigrants may reduce their economic progress and rate of long-run assimilation into the broader society. The Effects of Trade and Immigration on Labor Markets 4. Increased immigration has a modest adverse effect on the wages of the immigrants themselves and on the wages of earlier waves of immigrants, but it has only a modest effect on the wages of the young black and Hispanic Americans who are likely to be the next closest substitutes (LaLonde and Topel). Neither the employment nor the wages of less educated black and white natives worsened noticeably in cities where immigrant shares of the population rose in the 1970s. On the positive side, there is some evidence that, in cities with more immigrants, employment grew more rapidly or declined more slowly in low-wage industries where immigrants tended to find jobs and that less-skilled natives moved into better jobs (Altonji and Card). The broad implication is that immigrants have been absorbed into the American labor market with little adverse effect on natives. 5. “General equilibrium analysis” of the potential effects of immigration on the labor market through changes in sectoral outputs and prices further supports the claim that immigration has not harmed American labor. Indeed, the concentration of immigrants in import-intensive, traded goods manufacturing industries and the distribution of capital and native labor among export, import, and nontraded goods sectors suggests that increased immigration may actually benefit native labor, at least in the short run (Collins; Kuhn and Wooton). 6 . Wages in industries where sales are adversely affected by trade tend to decline relative to wages in other industries, just as do wages in industries where sales are adversely affected by domestic market developments, buffering to some extent the loss of jobs in industries facing large increases in imports. Unionized sectors make greater wage adjustments than nonunion sectors, apparently because workers in those industries earn above-market wages that can be reduced to save jobs whereas nonunion wages are closer to competitive levels (Abowd and Lemieux; Freeman and Katz). Once workers are dislocated by trade, however, they appear to have greater difficulty finding work than workers displaced for other reasons (Kruse 1988). 7. Foreign-owned firms employ nearly 3% of American workers. Despite
23
Introduction and Summary
the concern about foreign ownership, wages of production workers appear to be higher in foreign-owned enterprises, and rates of unionization are not different from domestic-owned companies. Moreover, notwithstanding all the attention given to Japanese firms, the bulk of direct foreign-owned enterprises in the United States are European. Foreign-owned firms use substantially more highly educated research-and-development employees (Leonard and McCulloch). Comparative Experiences: Canada and Australia 8. Canadian and Australian immigration policies, traditionally based on labor market skill considerations, have moved toward admitting immigrants for reasons of family unification, as in the United States. Since 1974, Canada has given preferential treatment to persons with close relatives in the country as well as to those who fulfill certain labor-market criteria. As a consequence, immigrants coming to Canada after the mid-1970s apparently do worse in the labor market relative to natives than earlier immigrant cohorts (Bloom and Gunderson). Australia admitted immigrants on the basis of a labor market point system from the 1970s through the early 1980s, with the result that the labor market skills of Australian immigrants did not drop in the 1970s relative to those of native workers, as in the United States and Canada (Beggs and Chapman). The implication is that immigration policies significantly affect the type of immigrants and their labor market performance. 9. Low-skill immigrants are relatively more highly paid in Australia than in the United States. There are three reasons for this: (1) wage differentials by occupation are smaller in Australia than in the United States; (2) immigrants are more highly unionized in Australia than in the United States; and (3) Australia has enacted trade policies that protect industries employing low-skill immigrants. Australian protection of immigrant-intensive industries produces relatively higher prices for the outputs of those sectors and extracts a sizable social cost on the order of 50% to 100% of the wage bill in footwear, clothing, and textiles (Gregory, Anstie, and Klug). 10. In Canada, changes in import and export prices, which reflect the pressure of the international economy on producers, have significant effects on the employment of workers covered by collective bargaining agreements. Increases in import prices, which shift demand to domestic producers, and in export prices, which reflect greater returns from increasing sales overseas, are associated with increases in employment of sizable magnitudes. In both the United States and Canada, unionized employment is more sensitive to import competition than to a comparable reduction in domestic production (Abowd and Lemieux) .
Concluding Remarks Perhaps the most intriguing finding of the Immigration, Trade, and Labor Markets studies is the apparently different direct effect of immigration and
24
John M. Abowd and Richard B. Freeman
trade on workers in the affected labor markets. Whereas immigration does not discernibly reduce the wages and employment of less-skilled native workers in immigrant-intensive localities, imports reduce the pay as well as employment of workers in heavily affected industries. Why? What might account for this differential effect? While we cannot give a conclusive and quantifiable answer, the general factors likely to underlie the differences do seem clear. First, differences in the concentration and magnitude of imports and immigration in affected areas certainly have an influence. In the ten industries with the largest growth of import shares of sales from the 1960s to the mid-l980s, import shares rose by 14% of domestic sales to 73% of domestic sales on average. By contrast, in the ten standard metropolitan areas with the greatest 1970-1980 growth of immigrants relative to the work force, new immigrants averaged 20% of the 1970 work force. Employment fell by 56% in the tradeaffected industries, while employment of natives increased in all the immigrant-affected localities save for New York City. Second, immigration has potential offsetting effects on the demand for labor in affected areas, while trade has no such effects on demand for labor in affected industries. Immigrants purchase goods and services in the area in which they work, raising demand for labor. Immigrant skills are also likely to complement the skills of some native workers, raising demand for them. By contrast, even with balanced trade, workers in an industry facing a surge of imports are unlikely to benefit directly from offsetting export-created demand for labor or from complementary demands for native labor in retail and wholesale trade. Third, it is possible that the concentration of immigrants in gateway cities did not increase the labor supply in those areas by as much as the immigration numbers would suggest. This would be the case if natives adjusted their choice of location of residence to take account of the immigrant flows. The flow of immigrants to, say, Los Angeles could have deterred midwesterners or southerners from migrating there or impelled natives to move elsewhere, so that the labor force in the city was not all that different from what it would be absent immigration. No such mitigating response exists for trade-affected industries. All these considerations suggest that the 1980s import surge caused a greater "shock" in affected labor markets than did the influx of immigrants and, thus, created greater difficulties of labor market adjustment. Trade upset the demand-supply balance in industry labor markets more than immigration upset the demand-supply balance in local labor markets. One additional factor may also contribute to the greater effect of imports than immigration on affected workers. In some industries, worker skills and earnings are industry specific, so that shocks cause greater economic losses to the affected employees. Consequently, labor mobility may be easier for work-
25
Introduction and Summary
ers facing immigrant competition in a local labor market than for workers facing import competition in a trade-affected industry. In summary, while trade and immigration may have the same long-run economic effects on an economy, there are good reasons (and, more compelling, empirical evidence) that they have different transitional costs for affected workers.
References 1978. Citibank Economic Database [machine-readable magnetic data file, 1946 to present]. New York: Citibank. Council of Economic Advisers. 1990. Economic Report of the President. Washington, D.C.: U.S. Government Printing Office, February. Kruse, Douglas. 1988. International Trade and the Labor Market Experience of Displaced Workers: Evidence from the Displaced Worker Survey. Industrial and Labor Relations Review 41, no. 3 (April): 402-17. Passel, Jeffrey S., and Karen A. Woodrow. 1984. Geographic Distribution of Undocumented Immigrants: Estimates of Undocumented Aliens Counted in the 1980 Census by State. Washington, D.C.: U S . Bureau of the Census, Population Division. U.S. Department of Commerce. Bureau of the Census. 1983. US.Commodify Exports and Imports as Related to Output 1981180; Washington, D.C.: U.S. Government Printing Office, November. . Annual. Statistical Abstract of the United States. Washington, D.C.: U S . Government Printing Office. . Decennial. Census of Population Detailed Population Characteristics U.S. Summary. Washington, D.C.: U.S. Government Printing Office. U.S. Department of Commmerce. Bureau of Economic Analysis. 1983. Foreign Direct Investment in the United States, 1980. Washington, D.C.: U.S. Government Printing Office. . Monthly. Survey of Current Business. Washington, D.C.: U.S. Government Printing Office. U.S. Department of Labor. Bureau of Labor Statistics. Monthly. Employment and Earnings. Washington, D.C.: U.S. Government Printing Office. U.S. Department of Justice. Immigration and Naturalization Service. Annual. Statistical Yearbook of the Immigration and Naturalization Service. Washington, D.C.: U S . Government Printing Office. Warren, Robert, and Jeffrey S. Passel. 1987. A Count of the Uncountable: Estimates of the Undocumented Aliens Counted in the 1980 Census. Demography 24 (August): 375-93. CITIBASE.
This Page Intentionally Left Blank
Immigrant Flows and Performance in the United States
This Page Intentionally Left Blank
1
Immigration and Self-selection George J. Borjas
The insight that migrants may be systematically different from persons who do not choose to migrate has long played an important role in sociological and historical studies of the immigration phenomenon (see, e.g., the studies contained in Jackson 1969). The selectivity hypothesis has also played a major role in the modem economic literature that analyzes how immigrants do in the U.S. labor market. For example, the early studies of Chiswick (1978) and Carliner (1980) invoke the assumption that immigrants are positively selected from the population of the countries of origin to explain the cross-sectional empirical finding that immigrant earnings (after a short time period) “overtake” the earnings of natives with the same observed socioeconomic characteristics, such as age and education.’ My recent work in this area (Borjas 1985, 1987) has addressed two related questions raised by the early studies. Since most of the literature analyzing immigrant earnings focuses on the study of single cross-sectional data sets, my 1985 paper raised the possibility that the overtaking findings could be due to the fact that cross-sectional regressions confound aging and cohort effects.2 The positive correlation between immigrant earnings and years of residence in the United States observed in the cross section could arise because immigrants “adapt” rapidly to the U.S. labor market or because earlier waves of immigrants differ in substantial ways (labor market productivities, unobserved George J. Bojas is professor of economics at the University of California, San Diego, and a research associate of the National Bureau of Economic Research. The author is grateful to Richard Freeman for many helpful discussions of the ideas presented in this paper, to Charles Brown for insightful suggestions and comments, and to Bemt Bratsberg for excellent research assistance. He is also grateful to the National Science Foundation (grants SES-8604973and SES-8809281) and to the National Institute of Child Health and Human Development (grant R01-HD22344) for financial support.
29
30
George J. Borjas
abilities or skills) from more recent waves. Borjas (1985) adapted well-known techniques (see, e.g., Heckman and Robb 1983) to separately identify aging and cohort effects using the 1970 and 1980 U.S. Censuses. This methodology, which “tracks” synthetic cohorts of immigrants over time, showed that ( a ) immigrant assimilation was not as fast as the cross-sectional studies indicate, (b) the more recent immigrant waves performed substantially worse in the labor market than the early postwar waves, and ( c ) there was little likelihood that the most recent immigrant waves would ever earn substantially more than natives of comparable age and education. An important insight provided by the study of synthetic cohorts is that invoking the assumption of positive selection, though it may be correct for some cohorts of immigrants, may be completely wrong for other cohorts of immigrants. This raises the important question of exactly which factors determine whether immigrants are positively or negatively selected from the population in the countries of origin. Borjas (1987) presents an initial attempt to address this problem and derives a simple economic model of selection on the basis of unobserved characteristics (which, after all, form the focus of much of the literature on immigrant earnings). This model, which will be discussed in detail below, shows that there is no general law stating that immigrants must be positively selected. In fact, under a reasonable set of conditions, it is likely that immigrants are negatively selected (i.e., persons who have below-average earnings and productivities are the most likely to migrate to the United States). My empirical analysis revealed that positive selection was more likely to characterize immigrants from the advanced industrial countries and negative selection was more likely to characterize immigrants from the Third World countries, who form the bulk of migration to the United States in the post-1965 period. This paper expands my earlier work in a number of significant ways. The theoretical analysis below will argue that, although most of the literature has focused on the role that selection in unobserved Characteristics plays in determining immigrant earnings, there is also selection in observed characteristics such as education. The theoretical framework clearly shows that it is completely possible for the most educated persons to migrate to the United States (i.e., positive selection in education) but for these persons to be the least productive in the population of highly educated persons (i.e., negative selection in unobserved characteristics). The analysis below presents a number of propositions that yield insights into the process that determines the selection of immigrants in these separate dimensions of “quality.” The empirical analysis in this paper expands my previous work in two ways. First, it presents a detailed analysis of the U.S. earnings of immigrants by focusing on the roles played by selection in both observed and unobserved characteristics. It will be seen that a number of the theoretical predictions are confirmed by the data. Second, it is clear that potential migrants can choose from a number of potential host countries. The empirical analysis below will
31
Immigration and Self-Selection
present a systematic study of the selection biases generated by the sorting of migrants among three potential countries of destination: Australia, Canada, and the United States. The evidence indicates that both country-of-origin and country-of-destination characteristics play an important role in determining the performance of immigrants in any labor market.
1.1 Theory of Immigration 1.1.1 The Roy (1951) Model Migration is assumed to flow from country 0, the country of origin or the “home” country, to country 1, the country of destination or, for concreteness, the United States. This simple framework ignores three potential complications. First, it is likely that persons born in the United States also consider the possibility of migrating to other countries, and perhaps many of them do so. Second, even persons choosing the United States as a country of destination may find that things do not work out (or perhaps work out much better than expected), and some return migration is thereby generated. Third, individuals contemplating migration in a particular country of origin enter the “immigration market” in which a number of other host countries (such as Australia and Canada) compete for the immigrant’s human and physical capital. Little is known about the size and composition of the migrant flows from the United States to other countries; hence, these possibilities are ignored in what follows. Much more, however, is known about the size and composition of the flows from any given home country to each of three potential host countries (Australia, Canada, and the United States), and the implications of the simpler two-country model will be applied below to the more general framework where potential migrants not only decide whether to migrate but also choose a country of destination. Residents of the home country face an earnings (w)distribution given by (1)
In w,
=
XS,
+
E ~ ,
where X is a vector of socioeconomic characteristics with value 6, in country 0, and the disturbance E~ is independent of X and is normally distributed with mean zero and variance @. The earnings distribution facing individuals in the United States is given by (2)
In w,= (1 - M ) X S ,
+ MX6, +
E,,
where M is a dummy variable indicating whether the individual is foreign born or native. The vector 6, gives the value that the U.S. labor market attaches to the socioeconomic characteristics X for natives. This valuation may differ because of discrimination or other unobserved factors from the value 6 , that the labor market attaches to the characteristics brought in by potential migrants. The disturbance E, is again independent of X (and M )and is normally distrib-
32
George J. Borjas
uted with mean zero and variance 0:.Finally, the random variables E~ and E , have correlation coefficient p. Equations (1) and (2) completely describe the earnings opportunities facing a potential migrant (as well as U.S. natives). Three questions are raised by this simple framework. First, what factors determine the size of the migration flow generated by the income-maximization hypothesis? Second, what types of selection in the unobserved characteristics E are created by the endogenous migration decision? Third, what types of selection in the observed characteristics X are created by the endogenous migration decision? The migration decision is determined by the sign of the index function:
where C gives the level of mobility costs, and 7~ gives a “time-equivalent” measure (T = C/wo)of the costs of migrating to the United States. The level of migration costs C is likely to vary among individuals for two reasons. First, there are time costs associated with migration, and these time costs are likely to be higher for persons with higher opportunity costs. Second, there are transportation costs associated with migration. These direct costs include not only the air fare (which is likely to be constant across individuals) but also moving expenses of family and household goods, and it is reasonable to suppose that these expenses may also be a positive function of w,. These assumptions give little hint as to how the time-equivalent measure of mobility costs, T ,varies across individuals. It is instructive to assume first that T is constant across individuals since the main implications of the Roy model are clearest in this special case. The analysis below will show that the treatment of T as a random variable in the population does not substantially alter the analysis and will, in some instances, reinforce the conclusions of the simpler model. Since migration to the United States occurs when I > 0, the emigration rate from the country of origin for persons of given characteristicsX is given by
(4)
P ( X ) = pr{v > - [ X ( S , - So) -
T]}=
1
-
@(z),
where v = E , - E ~ z, = - [ X ( S , - So) - T]/u,, and 0 is the standard normal distribution function. If the characteristicsX have a joint density function given byf(x), then the emigration rate from country 0 is given by
P
P(x)f(x)dx.
= X C i l
Equations (4) and (5) summarize the (rather obvious) economic content of the theory of migration proposed by Hicks (1932) and further developed in Sjaastad (1962). In particular, the emigration rate is a negative function of mean income in the home country (po = XS,), a positive function of mean
33
Immigration and Self-Selection
income in the United States (p, = XS,), and a negative function of migration costs. Much of the literature on the internal migration of persons in the United States is devoted to testing these theoretical predictions (see the survey in Greenwood 1975). The immigration literature, on the other hand, has historically focused on explaining not the size of migration flows but their composition or labor market quality. As far back as 1919, for example, Douglas was asking whether the skill composition of immigrant cohorts was constant across successive immigrant waves. The theory of migration contained in equations (l)-(5)has important implications about the selection biases that characterize the pool of migrants in terms of both unobserved and observed characteristics. Consider initially the selection mechanism in the unobserved characteristics E. In particular, consider the conditional expectations E(1n w, I X, I > 0) and E(1n w ,I X, I > 0). Note that these means condition on two dimensions: the observed characteristics X and the decision to migrate. Under the normality assumptions, these conditional means are given by
(7)
+
where A = + (z) /P (X ), and is the density of the standard normal. The variable A is inversely related to the emigration rate and will be positive as long as some persons find it profitable to remain in the country of origin (i.e., P [ X ] < 1). Let Q, = E ( E , I X, I > 0), Q, = E ( E , I X , I > 0), and k = U,/U,. The variables Q, and Q,measure the “quality” (in terms of unobserved characteristics) of the migrant pool. The Roy model identifies three cases of substantive interest. Positive Selection, Q,
> 0 and Q,> 0.
This type of selection exists when migrants have above-average earnings in the country of origin (for given characteristicsx) and also have U.S. earnings that exceed the earnings of comparable U.S. natives (ignoring the possibility that immigrant earnings may be reduced because of their ethnic or racial background). Inspection of equations (6) and (7)shows that the necessary and sufficientconditions for this type of selection to occur are
If p is sufficiently high, and if income is more dispersed in the United States than in the country of origin, immigrants arriving in the United States will be selected from the upper tail of the home country’s income distribution and
34
George J. Borjas
will outperform comparable natives on arrival in the United States. Intuitively, this occurs because the home country, in a sense, is “taxing” high-ability workers and “insuring” low-ability workers against poor labor market outcomes. Since high-income workers benefit relatively more than low-income workers from migration to the United States (regardless of how much higher mean incomes in the United States may be relative to the country of origin), a brain drain is generated, and the United States, with its greater opportunities, becomes a magnet for persons who are likely to do well in the labor market. Negative Selection, Q, < 0 and Q , < 0 This type of selection is defined to exist when the United States draws persons who have below-average incomes in the country of origin and who, holding characteristics constant, do poorly in the U.S. labor market. The necessary and sufficient conditions for negative selection to occur are (9)
p>k,
k < 1.
Negative selection also requires that p be “sufficiently” positive but that the income distribution in the country of origin be more unequal than that in the United States. Intuitively, negative selection is generated when the United States “taxes” high-income workers relatively more than the country of origin and provides better insurance for low-income workers against poor labor market outcomes. This opportunity set leads to large incentives for low-ability persons to migrate, since they can improve their situation in the United States, and to decreased incentives for high-ability persons to migrate, since income opportunities in the home country are more profitable. Refugee Sorting, Q, < 0 and Q , > 0 This kind of selection occurs when the United States draws below-average immigrants (in terms of the country of origin) but migrants have aboveaverage earnings in the U.S. labor market. The necessary and sufficient condition is p
< min
(k,
k).
In other words, if p is negative or “small,” the composition of the migrant pool is likely to resemble a refugee population. For instance, it is likely that p is negative for countries that have recently experienced a Communist takeover. After all, the change from a market economy to a Communist system is often accompanied by structural changes in the income distribution and by confiscation of entrepreneurial assets and redistribution to other persons. The Roy model suggests that immigrants from such systems will be in the lower tail of the “revolutionary” income distribution but will outperform the average U. S. native worker. The basic Roy model thus provides a useful categorization of the factors
35
Immigration and Self-Selection
that determine the quality or composition (in terms of unobserved characteristics) of the migrant pool. Even at this level, several important implications are generated that give some insight into a number of empirical findings in the literature. For example, many studies have documented the fact that refugee populations perform quite well in the U.S. labor market when compared to native workers of similar socioeconomic characteristics. These empirical results are explained by the income-maximization hypothesis and by the fact that these refugee populations, prior to the political changes that led to a worsening of their economic status, were relatively well off in the country of origin. It is, therefore, unnecessary to resort to the arbitrary distinctions between “economic” and “noneconomic” migrants to explain the refugee experience. The Roy model also provides an interesting explanation for the empirical finding that the quality of migrants to the United States has declined in the postwar period (where quality is defined by the wage differential between migrants and natives of the same measured skills). Prior to the 1965 amendments to the Immigration and Nationality Act, immigration to the United States was regulated by numerical quotas. The distribution of the fixed number of quotas across countries was based on the ethnic population of the United States in 1920 and thus encouraged migration from (some) Western European countries and strongly discouraged immigration from other continents, particularly Asia. The favored countries have one important characteristic: their income distributions are probably much less dispersed than those of countries in Latin America or Asia. The 1965 amendments abolished the discriminatory restrictions against immigration from non-European countries, established a twenty thousand numerical limit for legal migration from any single country (subject to both hemispheric and worldwide numerical limitations), and led to a substantial increase in the number of migrants from Asia and Latin America. The new flow of migrants thus originates in countries that are much more likely to have greater income inequality than the United state^.^ It would not be surprising, therefore, if the quality of immigrants declined as a result of the 1965 amendments. The theoretical analysis yields two equations that can guide empirical analysis. These equations are given by
(1 1)
Q,
= g b 0 , CL,,n, u,,, u,, P),
Equation (1 1 ) gives a “reduced-form’’ equation, where immigrant quality in the United States (i.e., the wage differential between migrants and natives of equal measured skills) is a function of all the primitive parameters of the model (i.e., the parameters of the two income distributions and migration costs). My earlier paper (Borjas 1987) provides a detailed analysis of the theoretical restrictions implied by the income-maximization hypothesis on the direction of the effects of the various variables in the model. These effects are
36
George J. Borjas
usually ambiguous and can be categorized in terms of “composition effects” and “scale effects.” In particular, a change in any variable OL will create incentives for a different type of person to migrate (the composition effect) and for a different number of persons to migrate (the scale effect). Equation (12) is a “structural” equation and states that, if knowledge of A is available, a subset of the parameters of the model enters multiplicatively through the h function (see eq. [7]). By holding A constant, the structural equation essentially nets out the scale effect and leads to more unambiguous predictions of the effect of the exogenous variables on the quality of immigrants. It is important to note that the h function in (12) does not depend on mean income levels in the countries of origin and the country of destination or on the level of migration costs since these factors play a role only through the selectivity variable A. Three comparative statics results are implied by analysis of the A-constant structural quality equation. 1. An increase in the variance of the income distribution in the home country leads to a decrease in the quality of migrants in the United States. 2. An increase in the variance of the income distribution in the United States leads to an increase in the quality of migrants in the United States.’ 3. An increase in the correlation coefficient between earnings in the home country and earnings in the United States increases immigrant quality if there is positive selection and decreases immigrant quality if there is negative selection. The ambiguity arises because, the larger the correlation coefficient, the better the “match” between the two countries. The improvement in the match increases the quality of the immigrant flow if there is positive selection and decreases it if there is negative selection. 1.1.2 Random Mobility Costs These insights have been derived from the simplest version of the Roy model, which treats mobility costs (defined as a fraction of potential income in the country of origin) as a constant in the population. This assumption may be restrictive, and it is important to ascertain how its relaxation affects the results of the model. Suppose that mobility costs are normally distributed in the population and can be written as
(13)
Tr= P,
+
E,,
where p, is the mean level of mobility costs in the population, and E, is a normally distributed random variable with mean zero and variance u;. The random variable E, may be correlated with E, and E, and the correlation coefficients are given by pno and p,, , respectively. The conditional expectations of migrant incomes in the home and destination countries are now given by
(14)
E(ln w, I X, I > 0) = X 6 ,
+u
u
Ud
[(P
):
-
- Pd
3,
37
(15)
Immigration and Self-Selection E(ln w , I X, I
> 0) = X 6 ,
+
where v’ = E , - E, - E,. Equations (14) and (15) show that the addition of migration costs does not affect any of the substantive results of the simplest version of the Roy model if migration costs are uncorrelated with earnings opportunities. However, if migration costs are correlated with earnings opportunities, the type of selection that is generated may change in either direction. Suppose, for example, that migration costs are positively correlated with earnings opportunities. For instance, high-ability persons may take longer to find appropriate jobs. This positive correlation makes both Q, and Q, more negative and hence increases the likelihood of negative selection. Conversely, if migration costs (measured in time units) and earnings opportunities are negatively correlated, the likelihood of positive selection is increased. Two additional points about this more general model are worth stressing. First, the importance of variable migration costs in the analysis will diminish greatly if the variance in migration costs is relatively small compared to the variance in the income distributions. Second, regardless of how important migration costs are, the key result that negative selection is more likely from countries with high levels of income inequality and positive selection is more likely from countries with more equal income distributions is unaffected. 1.1.3 Selection in Observed Characteristics Equation (4), the probit equation determining the migration rate, contains an additional insight: the migration rate is a function of X through the parameter (6, - 8,). Hence, the migration of persons with larger levels ofX is more likely if X has a higher return in the United States than in the country of origin, and the migration of persons with lower levels of X is more likely if the country of origin values the characteristicX more than the United States. A complementary analysis to the Roy model can be derived if it is assumed that the vector X consists of only one variable, say education (s), that this variable is uncorrelated with the disturbances in the earnings functions, and that this variable, too, is normally distributed in the population. The assumption of only one variable in the vector X is irrelevant since the results can be easily generalized to any number of variables. The assumption of normality, though unrealistic for some socioeconomic characteristics, does simplify the mathematics substantially and allows a useful extension of the Roy approach to the study of selection in observed skills and the analysis of the actual wage differential between immigrants and natives (as opposed to the standardized wage differential). Suppose the earnings functions in the two countries are given by (16)
In w, = po + 6,s
+ E,.
38
George J. Borjas
In w , = p,
(17)
+ 6,s + E,,
and that the education distribution in the population of the country of origin can be written as (18)
s =
+
Er,
where E, is normally distributed with mean zero and variance US. Assuming that mobility costs are constant, the emigration rate for the population in the country of origin is given by
(19)
p
=
pr{(El -
Eo)
+ (4
+ (6, - 6,)p$
- $)E, > - [(PI - Po) - TI} = 1 - @(z*),
+
where t = (E, - E ~ ) (6, - SO)&,, and z* = -[(p, - po) + (6, - 6JPs - TI/U,. Two interesting questions can be addressed within this framework. First, consider the conditional expectation of schooling of persons who do migrate. It is easy to show that
Hence, the mean schooling of migrants will be less than or greater than the mean schooling of the population depending on which of the two countries values schooling more. Positive selection in schooling will be observed when (6, - 6,) > 0 so that the U.S. labor market attaches a higher value to schooling, while negative selection in schooling will be observed when (6, - 6,) < 0 so that highly educated individuals have little incentive to leave the country of origin. It is important to stress that these selection conditions seem to have little to do with the conditions determining selection in unobserved characteristics. Any permutation of selection mechanisms in unobserved and observed characteristics is theoretically possible. Hence, negative selection in unobserved characteristics (or ability) may be jointly occurring with positive selection in education, or vice versa. Simply because the United States attracts highly educated persons from some countries does not imply that these highly educated persons are the most productive highly educated persons in that particular country of origin. At a more fundamental level, however, the determinants of the two types of selection are not all that different. The sorting in observed characteristics is guided by international differences in the prices So and 6,. In the case of unmeasured skills, the sorting is guided by the variances ui and a:. In a sense, these variances measure the “prices” of unmeasured skills in the respective countries since these abilities are more highly rewarded in countries with higher levels of income inequality. The sorting in all the dimensions of skills,
39
Immigration and Self-Selection
therefore, is guided by the same basic process: skills flow to whichever country offers the highest price for them. The actual mean earnings of the migrant pool in each of the two countries are given by (21)
E m w, I I > 0 ) =
Il.0
+
6oPs
(6, - 6,)6,
+ -(p *OU1
u,
-
)]A,
Mean earnings of migrants depend on the mean education of migrants, as given by (20), and on the mean level of their unobserved characteristics. Since the two kinds of selections are independent, nothing can be said about how the average migrant performs in the host country unless the kinds of selections that occurred in each of the two dimensions of quality are known. Nevertheless, it is of interest to document the net effect of the selection in all the various dimensions of skills on immigrant earnings, and the empirical analysis below presents a detailed study of the unstandardized earnings differential between immigrants and natives in the host country. Equations (21) and (22) show that generalizations about the quality of immigrants based solely on observed education levels (or other measures of X) are extremely misleading. In addition, it is well known that observed characteristics such as education, age, marital status, health, etc. explain a relatively small fraction of earnings variation across individuals. It is not uncommon, for example, to find that the observed characteristics explain much less than a third of the variance in wage rates or weekly earnings. The selection in unobserved characteristics, therefore, is likely to be much more important empirically than the selection in observed characteristics. A number of comparative statics results can be generated by analysis of equation (20). Perhaps the most interesting of these results is
That is, a one-year increase in the mean education level of the country of origin will increase the mean education level of persons who actually migrate to the United States, but this increase will be by less than one year. The intuition for this result follows from the fact that an increase in kSwill change the size of the immigrant flow. Suppose, for concreteness, that (6, - 6,) > 0 so that there is positive selection in education. The increase in p, makes it worthwhile for more persons to migrate and thus dilutes the mean education level of the population of migrants. Hence, the increase in the conditional expectation is less than the increase in the population mean. An important implication
40
George J. Borjas
of this theoretical prediction is that the variance in education levels across immigrants (from different countries) in the United States will be smaller than the variance in education levels of the actual populations across countries in the world. In other words, the population of migrants in the United States is more homogeneous (in terms of education) than the populations of the different sending countries. In general, equation (20) implies the existence of observable quality equations analogous to (1 l ) and (12): (24) (25)
QT
= g * b 0 , CL.,,m, uo, ul, P, F,, us,(6, - 6,,)1,
QT
= h*[uo, ul, P, IJ-~, us,(8, -
6,)lL
where QT gives the mean level of the observed characteristics of immigrants in the United States. The estimation of (24) and (25), of course, is likely to be extremely difficult in practice since they introduce a number of primitive parameters (e.g., 6, - 6,) that are unobservable and likely to remain so.
1.2 Empirical Framework Recent empirical research on the earnings of immigrants stresses the importance of disentangling the cohort and aging effects that are confounded by a single cross section of data. In the analysis presented in this paper, two Censuses in the country of destination will be pooled (e.g., the 1970 and 1980 U.S.Censuses), and the following regression model will be estimated: (26) (27)
In wo = Xjai
+ a l y j + azy,’ + 2 p,C, + yimj + E ~ , In wnl = X,S, + y,ml + t
E,~,
where wii is the wage rate of immigrantj, wnris the wage rate of native person 1, X is a vector of socioeconomiccharacteristics (e.g., education, age, etc.), y is a variable measuring the number of years that the immigrant has resided in the country of destination, C is a vector of dummy variables indicating the year in which migration occurred, and IT is a dummy variable set to unity if the observation is drawn from the 1980 Census and zero otherwise.6The vector of parameters (a,, az),along with the age coefficients in the vector X, provides a measure of the assimilation effect (i.e., the rate at which the ageearnings profile of migrants is converging to the age-earnings profile of natives), while the vector of parameters p estimates the cohort effects. The period effects are given by yi for immigrants and by y, for natives. The model in equations (26) and (27) is underidentified. In particular, some of the right-hand-side variables in the immigrant earnings function are perfectly collinear. Suppose, for example, that the immigrant arrived in calendar year 8 so that C, = 1. Then
41
Immigration and Self-Selection y = (T
(28)
- k - 8)
+ nk,
where T is the calendar year in which the latest cross section is observed, and k is the number of years separating the two cross sections. The variable capturing the period effect, therefore, is a linear combination of the cohort variable and of the years-since-migration variable. Obviously, two cross sections cannot be used to identify three separate effects: period, cohort, and aging effects. In order to estimate the structural parameters describing the extent of immigrant assimilation and cohort quality change, a restriction must be imposed on the size of the period effect in the migrant population. A reasonable, though unverifiable, assumption is that the period effect experienced by immigrants (y,) is identical to the period effect experience by natives (y,). In other words, changes in the wage rate due to shifts in aggregate economic conditions affect the immigrant and native wage levels by the same relative magnitude. It is easy to show that this restriction is sufficient to identify all the structural parameters in equations (26) and (27) exactly. This theoretical restriction leaves some amplitude for its empirical implementation since the choice of the native base is essentially arbitrary. The choice of a native base for the various immigrant groups under study will be discussed in detail below. There are two dimensions of migrant quality that can be calculated from the estimated regressions in (26) and (27): (a) the entry wage of immigrants when they arrive into the United States and (b) the rate at which this wage changes over time. To simplify the empirical analysis, the two measures will be combined into a single measure of immigrant quality. In particular, let G,(8) be the entry wage of an immigrant cohort that arrives in the United States at age twenty in calendar year 8, and let Gnbe the entry wage of a comparable (in terms of all observable economic variables) native person who enters the labor market at age twenty. Similarly, let gi be the rate at which the earnings of immigrants grow over their lifetime, and let g, be the growth rate for natives. Finally, let r be the rate of discount (assumed to be the same for migrants and natives). If persons are infinitely lived, the present values associated with the earnings streams of migrants and natives are given by
(29)
vj(e) =
I
Gi(8)e-@-gJ'df =
wi(8)/ (r - g i ) ,
0
(30)
=
@
(r- g).
The percentage difference in present values between immigrants of cohort 8 and natives is defined by
42
George J. Borjas
and a first-order approximation (using the assumption that earnings growth rates are small relative to the discount rate) yields In [Vi(e) / V,] = [In ai(8)- In
an]+ .-gi
-g,
r
Hence, the percentage difference in the present value of the earnings streams faced by immigrants and natives is an additive function of the wage differential at the time of entry and of the difference in earnings growth rates over the life cycle.7 The present value differential in (32) can be easily evaluated from the estimates of equation (26) and (27) if two assumptions are made. First, the rate of discount is assumed to be 5 percent. Clearly, the assumption of any higher discount rate would lead to a worsening of relative immigrant earnings since the latter part of the working life cycle (where immigrants tend to do better) would be more heavily discounted. Second, the growth rates g, and g, must be evaluated from the age and years-since-migration coefficients in the earnings functions in (26) and (27). The quadratic specification for age and years since migration in the earnings functions implies that the growth rate is not constant over time. The empirical analysis below will define the growth rate g, and g, by (33) (34)
2,
= ( Y , [ X ,50, 30, 81- Y , [X,20, 0, 8])/30,
2,
=
( Y , [ X , 501 - Y,[ X , 20])/30,
where Y , [ X ,A , y, 81 is the predicted (In) earnings for an immigrant with characteristics X , at age A, with y years of residence in the United States, and who migrated in cohort 8 . Similarly, Y J X , A] gives the predicted earnings for a native with characteristics X at age A . In other words, the average growth rate experienced by immigrants and natives between ages twenty and fifty (evaluated at the mean characteristics of the migrant population, X) is used for estimation of the growth rate in the present value expressions. This approach has the useful property that the growth rates (for both immigrants and natives) are a linear function of regression coefficients, and, since the entry wages are given by Y,[X, 20,0, 01 for immigrants and Y J X , 201 for natives, the present value expressions in (33) and (34) are also linear functions of regressions coefficients; hence, a standard error can be easily evaluated. This approach makes a departure from the tradition in the empirical literature that analyzes immigrant earnings. The entire literature essentially focuses on the estimation of entry wage levels and on the calculation of “overtaking” points (if they exist). This type of analysis is not useful if overtaking points occur rather late in the life cycle (or if they do not occur at all), as some recent evidence suggests. The empirical use of the present value of earnings is much more consistent with the theoretical content of the theory of migration and
43
Immigration and Self-Selection
deemphasizes the somewhat misleading concept of overtaking points. The analysis of the success of migrant groups in the United States, to borrow from the human capital theory that guided much early research on immigrant earnings, should be based not on the calculation of wage differentials at given ages but on the life-cycle wealth accumulated by migrants and natives. Hence, the present value approach used in the empirical sections of this paper is much more in the tradition of the human capital literature and of the Roy model of immigration developed in the previous section.
1.3 Earnings of Immigrants in the United States 1.3.1 Data and Descriptive Statistics This section analyzes the relative earnings of immigrants in the U.S. labor market. The data are drawn from the 1970 2/100 U.S. Census (obtained by pooling the 5% SMSA and County Group Sample and the 5% State Sample) and the 1980 5/100 A sample. The complete samples are used in the creation of the immigrant extracts, but random samples are drawn for the native “baseline’’ populations.8 The analysis is restricted to men aged 25-64 who satisfied five sample selection rules: (1) the individual was employed in the calendar year prior to the Census; (2) the individual was not self-employed or working without pay; (3) the individual was not in the armed forces (as of the survey week); (4) the individual did not reside in group quarters; and (5) the individual reported annual earnings exceeding $1,000. Throughout this section, the dependent variable is the logarithm of the individual’s wage rate in the calendar year prior to the Census. The individual’s wage rate is defined as the ratio of annual earnings to annual hours worked. In the 1970 Census, annual hours worked is given by the product of weeks worked in 1969 and hours worked in the Census week, while, in the 1980 Census, annual hours is the product of weeks worked in 1979 and usual hours worked per week in that calendar year. Forty-one countries were chosen for analysis. These countries were selected on the basis that both the 1970 and the 1980 Censuses contained a substantial number of migrants from that country. In particular, it is necessary to have at least eighty observations of persons born in a particular foreign country in the pooled UlOO 1970 Census to enter the sample of forty-one countries. The countries thus chosen account for over 90 percent of all immigration to the United States between 1951 and 1980. It must be noted, however, that this restriction omits some countries that during the late 1970s became important source countries (e.g., Vietnam). Since two Censuses are required for the complete identification of the parameters of the model presented in section 1.2, however, a systematic analysis of the relative earnings of these migrants will have to await the 1990 Census. Table 1.1 begins the empirical analysis by presenting the unstandardized
44
George J. Borjas
differential between the log wage rate of the various migrant groups and ‘hatives.” In these statistics, the native population is defined as the group of U.S.born white, non-Hispanic, non-Asian men aged 25-64. Perhaps the most striking finding in the table is the fact that migrants from European countries tend to have wage rates that often exceed the wages of white natives, while migrants from Asian or Latin American countries tend to have wage rates that are substantially below those of white natives. Table 1.1 also presents the relative earnings of the 1965-69 cohort of migrants as of 1970, the relative earnings of the same cohort in 1980, and the relative earnings of the 1975-79 cohort as of 1980. These statistics yield important insights into the process of assimilation (the rate at which the earnings of migrants and natives are converging) and into the extent of productivity differences across successive cohorts. The “tracking” of the 1965-69 cohort across Censuses shows that the relative earnings of this cohort of migrants improved over time for most national groups. At the same time, the comparison of successive immigrant cohorts (i.e., the comparison of the 1965-69 cohort as of 1969 and the 1975-79 cohort as of 1979) shows that for some countries the relative earnings of migrants increased while for other countries the relative earnings of migrants decreased substantially. For example, the most recent migrant from France in 1970 was earning about 8 percent less than natives at the time of entry, while the most recent migrant from France in 1980 was earning about 22 percent more than natives at the time of entry. Conversely, the most recent migrant from India in 1970 earned about 4 percent more than white natives at the time of entry, but the most recent migrant from India in 1980 was earning 21 percent less than white natives at the time of entry. Table 1.2 continues the descriptive analysis by presenting the mean (completed) education level of four different cohorts of immigrants that arrived in the 1960-80 period. Since the education data available in the Census does not differentiate between education obtained prior to immigration and education obtained in the United States after immigration, the mean education levels for the 1970-74 and 1975-79 cohorts are obtained from the 1980 Census, and the mean education levels for the 1960-64 and 1965-69 cohorts are obtained from the 1970 Census. This use of the available data is designed to minimize the contamination of the education variable by postmigration schooling. The statistics in table 1.2 are consistent with the well-known secular increase in education levels over time for practically all migrant cohorts. It is worth noting, however, that for some countries the increase in education has been quite small (e.g., Portugal) while for others (e.g., Norway) it has been amazingly large. As the theoretical analysis in section 1.1 shows, these truncated education means can be understood only in terms of the population means of the education distribution in the countries of origin. To provide some insights into the extent of self-selection on the basis of education, table 1.2 also presents mean education levels calculated for the population in the countries of origin. The mean education level for the 1960s is calculated using
45
Immigration and Self-Selection
Table 1.1
Unstandardized U.S. Earnings of Immigrants Relative to White Natives
1970 Country Europe: Austria Czechoslovakia Denmark France Germany Greece HwwY Ireland Italy Netherlands Norway Poland Portugal Romania Soviet Union Spain Sweden Switzerland United Kingdom Yugoslavia Asia and Africa: China Egypt India Iran Israel Japan Korea Philippines Americas: Argentina Brazil Canada Colombia Cuba Dominican Republic Ecuador Guatemala Haiti Jamaica Mexico Panama Trinidad and Tobago
1980
All 1965-69 Immigrants Cohort
Sample All 1965-69 Size Immigrants Cohort
1975-79
Sample Size
,1969 ,1229 ,1208 ,1109 .I600 - ,1722 .1304 - .0369 - ,0150 .OW3 ,1653 .0392 - ,1913 .1153 ,0813 - ,1572 ,1485 ,2424 .I669 ,0353
,2182 ,0466 ,1803 - ,0766 ,1095 - .3704 - ,0631 - ,0260 - ,1707 .1412 ,2629 - ,0952 - ,2406 - .I915 - .I048 - ,3480 .2573 .0095 .1902 - .I382
380 398 141 317 2,399 634 650 754 3,068 430 243 1,629 349 259 907 210 221 177 2,231 646
,2108 ,1483 .2387 ,1071 ,1577 - .I874 ,1059 ,0688 - ,0124 ,1717 26.96 ,0165 - .2104
-.1258 .0273 ,4241 ,2237 ,2646 - ,3392 - ,1805 - ,1421 - ,1616 ,2824 ,2444 - .3698 - ,3240 - ,2913 - ,2856 - ,2143 ,1617 .4735 .1924 - ,1706
746 872 291 952 6,499 2,328 1,356 1,580 7,236
- .0533 - .0417 ,2392 ,3307 ,2111 ,0546
,3598 ,1141 ,3570 ,1158 ,2350 - ,2556 ,1027 ,0737 - ,0790 ,2179 .4183 .0207 - .1949 .0928 - .0578 - ,0184 .4570 ,2121 ,3188 - ,0191
408 3,278 2,213 614 2,104 730 335 397 5,475 1,967
- ,1543 - ,0073 ,1667 - ,0116 ,0707 ,0535 - ,0781 - ,1920
- .3459 - .2127 ,0413 - ,3556 -.I951 .0519 - .2183 - ,2389
880 136 363 121 141 228 142 816
- ,2212 ,0737 ,1221 - .0545 - ,0274 ,1362 - ,0881 - ,0707
- .1324 .3222 ,4050 ,1375 - ,0392 ,1492 ,2409 ,0694
- S372 - ,2892 - .2085 - ,2237 - .2483 .2020 - .3007 - ,3143
3,875 6% 3,629 1,027 789 1,634 2,013 4,955
,0319 .0212 ,1072 - .I452 - ,2822
- ,1644 - ,0993 ,1084 - ,2337 - .4461
218 - ,0096 101 ,0485 ,1258 3,430 254 -.2313 1,960 - ,1828
.0086 ,1407
- .2027 - .2698
,1440
- .1428 ,0481 ,1739 - ,4464 - .5392
834 345 7,083 1,760 6,837
- .3576 - ,2343 -.I940 - ,3041 - .I645 - ,4094 - ,0187
- ,5157 - ,4511 - ,5372 - ,3061 - .2462 - .6021 - .1899
210 174 82 130 263 3,122
- ,4768 - .2473 - .3425 - ,3726 - ,2132 - .3975 - ,0761
- ,4319 - ,2858 - ,2182 - ,2296 -.I245 - .3431 - ,1263
- ,6785 - ,5229 - s977 - ,6536 - .3604 - .6402 - ,3663
1,605 1,097 723 1,133 2,061 24,955 584
- ,1561
- .2909
86
- .1488
- ,0685
- ,4150
782
101
.0551
Cohort
1,161
46
George J. Borjas
Table 1.2
Completed Years of Schooling in Immigrant Cohorts Mean Education in Population
Year of Arrival Country Europe: Austria Czechoslovakia Denmark France Germany Greece Hungary Ireland Italy Netherlands Norway Poland Portugal Romania Soviet Union Spain Sweden Switzerland United Kingdom Yugoslavia Asia and Africa: China Egypt India Iran Israel Japan Korea Philippines Americas: Argentina Brazil Canada Colombia Cuba Dominican Republic Ecuador Guatemala Haiti Jamaica Mexico Panama Trinidad and Tobago
1975-79
1970-74
1965-69
1960-64
1970s
1960s
14.8 15.4 15.5 15.6 15.2 11.1 13.6 13.8 10.6 15.9 15.2 12.7 6.6 13.7 14.3 13.2 15.4 15.4 15.1 11.0
13.9 14.5 13.6 14.8 14.2 9.9 13.5 13.1 8.5 15.1 15.6 11.9 6.7 14.5 13.5 11.3 15.8 15.4 14.7 10.6
13.4 14.1 16.1 14.5 13.3 8.8 12.3 12.9 6.8 14.1 14.0 10.7 5.2 11.6 10.5 10.3 15.5 14.5 13.7 10.7
12.8 12.5 11.6 12.8 12.0 10.9 12.6 11.3 7.5 12.3 11.7 9.5 5.8 11.9 11.3 9.9 14.4 13.6 13.1 9.4
8.7 10.2 11.2 11.1 10.7 9.2 10.6 9.1 9.1 10.4 9.9 11.2 8.2 9.5 11.4 8.0 10.3 8.7 10.8 9.7
6.7 9.1 8.5 7.0 10.1 6.2 7.2 8.1 5.6 8.8 7.2 7.0 3.5 5.3 8.1 4.4 8.7 6.7 9.9 3.5
11.3 15.9 16.1 15.2 14.2 15.7 14.0 14.2
12.8 16.2 17.6 16.3 13.8 14.7 14.9 14.9
12.8 15.5 16.7 15.3 13.5 15.4 15.8 14.8
13.2 15.1 17.0 15.5 14.0 15.0 16.5 13.9
8.4 5.7 4.9 3.6 9.8 11.2 8.0 8.2
4.3 4.0 2.2 1.3 7.0 9.2 5.0 5.1
13.6 15.4 14.6 11.9 11.3 8.9 10.9 9.0 10.2 11.3 6.5 13.1 11.7
12.1 13.1 13.7 11.3 9.9 9.1 11.0 9.7 12. I 10.9 6.8 12.7 12.0
12.0 12.6 12.9 10.6 9.5 8.4 10.4 9.9 12.0 10.7 6.1 12.4 11.0
12.6 12.8 11.4 11.5 11.9 7.9 11.3 12.0 11.2 10.6 6.0 11.1 14.4
8.7 8.6 10.3 5.0 8.3 6.2 6.2 2.9 3.2 9.5 6.1 10.1 7.9
6.3 2.8 8.5 2.2 4.1 3.6 3.4 1.5 1.7 4.5 2.9 5.9 7.1
47
Immigration and Self-Selection
enrollment data in the various countries of origin during the 1950s, while the mean education level for the 1970s is calculated using enrollment data in the various countries of origin during the 1960s. The “lagged” construction of the variable giving mean education levels in the country of origin is designed to account for the fact that, in the samples used here, the average person migrated at about age 20. The relevant education distribution, therefore, is given by that of persons enrolled in school a few years earliec9 The means in table 1.2 present a remarkable picture. Even after allowing for the substantial errors involved in calculating the population means for each country of origin, the truncated means are almost always much greater than the population means. For example, the mean of education in Haiti is about three years, but the most recent Haitian immigrants report ten years of education in the 1980 Census. Surprisingly, the two statistics are most similar for Mexico, where both immigrants and the Mexican population have 6-7 years of education. Overall, table 1.2 suggests that immigrants are positively selected on the basis of education. The model presented earlier implies that this result is consistent with the hypothesis that the “rate of return” to education is greater in the United States than in most countries of origin. However, it is also consistent with the hypothesis that migration costs are lower for persons with higher education levels. This conjecture has received intensive study in the internal migration literature (Schwartz 1968). 1.3.2 Basic Regression Results The regression model in equations (26) and (27) was estimated on each of the 41 countries under analysis using the pooled 1970 and 1980 Census data. As noted earlier, the choice of the native baseline is an important step in the estimation procedure. In this section, the reference group is chosen according to the racial/ethnic background of the population of each country of origin. The estimation uses the white, non-Hispanic, non-Asian sample of native men as the reference group for migrants from Europe, Canada, and the Middle East. The group of Asian natives is the reference group for migrants from all other Asian countries. The group of Mexican natives is the reference group for Mexican migrants, and the group of “other Hispanic” men is the reference group for persons from all other Spanish-speaking countries in the American continent. Finally, the group of black natives is the reference group for migrants from countries with predominantly black populations (i.e., Haiti, Jamaica, and Trinidad and Tobago). The definition of the reference group in terms of the racial/ethnic background of the immigrant population is a simple way of specifying different period effects for the various immigrant groups. Presumably, the effect of changes in aggregate economic conditions on immigrant earnings is likely to be better approximated by the period effects experienced by populations that closely resemble the immigrant group. It is important to note, however, that, although the baseline populations differ across the 41 countries, the calcula-
48
George J. Borjas
tion of the present value differentials defined in equation (32) will always be relative to white, non-Hispanic, non-Asian natives (as in table 1 . 1 ) . In other words, the use of alternative reference groups is simply used to “net out” the period effect in the 1980 Census, and, after controlling for period effects, all comparisons between migrants and natives are conducted with respect to the “white” population. Initially, the regression model is estimated using a detailed list of demographic controls. These include education, marital status, health status, and metropolitan residence (as well as age and age squared). The calculated present value differentials estimated from the 41 runs of the model are presented in table 1.3 for each of the 6 cohorts identifiable in the Census data. It is worth stressing that these present value differentials measure the differences in earnings among migrants and white natives of equal measured skills and hence are empirical counterparts to the quality measure Q, defined in terms of unobserved characteristics. Table 1.3 shows that there are substantial differences in the “abilities” of migrant groups across the 41 countries of origin. Immigrants from European countries (particularly Western European countries) tend to do quite well relative to white natives of comparable socioeconomic characteristics. Recent immigrants from the United Kingdom, for example, can expected about 10 percent larger earnings over their lifetime than comparable white natives, recent immigrants from France will earn about 8-19 percent more than comparable white natives, and recent immigrants from Sweden will earn about 1020 percent more than white natives over their lifetime. On the other hand, immigrants from most Asian and Latin American countries do not perform well in comparison to white natives of equal observable skills. Recent immigrants from Taiwan, for example, will earn about 16-34 percent less over their lifetime than comparable white natives, immigrants from Israel will earn about 20-30 percent less, immigrants from Argentina about 20 percent less, and immigrants from Colombia about 24-38 percent less. An immigrant’s birthplace plays an important role in determining the type of selection that characterizes the migrant flow. In addition, table 1.3 shows that, even within a given country of origin, there are sizable differences in the unobserved quality of immigrants across the various cohorts. The quality of immigrants from some countries has been increasing rapidly, while the quality of immigrants from other countries has been declining rapidly. For instance, the most recent French immigrants have a higher earnings potential than earlier cohorts (particularly those arriving before 1970), while the most recent Polish migrants have much lower earnings potential than migrants of earlier cohorts. Similarly, the most recent Canadian immigrants earn about 8-15 percent more than most of the earlier cohorts, while the most recent Mexican immigrants earn about 9-13 percent less than the earlier Mexican cohorts.I0
49
Immigration and Self-Selection
lhble 1.3
Present Value Differentials between Immigrants and Natives Year of Arrival
Country
Europe: Austria Czechoslovakia Denmark France Germany Greece Hungary Ireland Italy Netherlands Norway Poland Portugal Romania Soviet Union Spain Sweden Switzerland United Kingdom Yugoslavia
(continued)
1975-79
1970-74
1965-69
- ,0841 ( - .74) - ,0141 (-.14) ,4432 (2.76) .I879 (2.29) ,0733 (1.69) -.lo60 (-2.00) ,1542 (- 1.94) ,1267 (1.58) .0498 (1.30) .2815 (3.66) ,1880 (1.45) - .1926 (-4.11) .0293 (.5u - ,2030 (-2.12) - ,2641 (-4.42) ,1047 (1.17) ,1141 (1.01) ,2395 (2.15) ,1052 (3.11) ,0602 (.93)
,1344 (1.25) - ,0546 ( - .73) ,0623 (.35) ,0829 (1.15) .0638 (1.50) - .I818 ( - 5.08) -.I132 (- 1.81) ,0817 (1.39) ,0424 (1.75) - .O917 (-1.11) .2468 (1.56) ,0727 (1.95) ,0348 (.82) .0911 (1.21) - .03W
,1945 (2.84) - .0036 (-.lo) ,1522 (1.45) - .0415 ( - .74) ,0385 (1 .a) -.I344 (-4.40) - ,0128 ( - .26) .1758 (3.75) .0693 (3.77) .0936 (1.69) .1757 (1.74) .0784 (2.65) ,0785 (2.31) - .0050 (-.lo) - .0332 ( - .57) .0518 (.96) ,2205 (1.97) .I407 (1.69) ,0948 (4.19) .0625 (1.94)
( - .55)
,1287 (2.01) ,1621 (1.15) .lo71 (.93) .0910 (2.88) .0746 (1.84)
1960-64 ,0707 (1.33)
.m
1950-59
< 1950
- .o004
- .0312 (-
( - .01)
,0182 ~42) - ,0434 ( - .65) - ,0626 (-1.56) ,0150 (.97) - .0402 - .0381 (-1.10) (-1.39) - ,0389 .0380 ( - .86) (1.45) - .0252 .0676 ( - .84) (2.14) .0839 .0695 (5.10) (5.04) ,0264 - ,0422 ~ 7 0 ) (-1.40) ,2017 ,1437 (2.55) (2.48) .0387 ,0526 (1.66) (2.44) .0954 ,0871 (2.18) (2.44) - ,0253 ,0534 ( - .39) (1 .04) - ,0456 ,0203 (-1.06) (.68) - .0022 - ,1186 ( - .01) (-2.22) ,0721 .o001 (.78) (.01) ,0967 ,0594 (1.41) (1.05) ,0449 ,0098 (2.32) (.%I ,1389 ,1089 (3.90) (4.15) (1.01) - ,0010 ( - .01) -.I179 (-2.57) ,0115
.05%
(.W . I 105 (.94) .0539 (.81) ,1174 (4.26) - ,1230 (-2.28) ,1441 (2.55) - ,2171 (-4.82) ,0627 (2.48) -.I736 ( - 2.77) - .0290 ( - .35) .0764 (2.31) .1746 (2.11) - ,0041 ( - .01) ,0322 (.67) - ,1001 ( - .94) ,0153 (.14) .0264 (.26) - ,0432 ( - 1.47) .0237 (.39)
50
George J. Borjas
Table 1.3
(continued) Year of Anival
Country Asia and Africa: China Egypt India Iran Israel Japan Korea Philippines Americas: Argentina
1975-79
1970-74
1965-69
- ,3662 - ,3362 (-7.58) ( - 10.26) -.I186 - ,1597 ( - 1.70) ( - 1.46) - .3365 - .1635 (-5.28) (-2.94) - .0084 .075 1 (-.lo) (.77) - ,3304 - ,2346 ( - 3.50) ( - 3.32) ,0741 - ,0145 ( - .22) (.95) - ,1840 -.I162 ( - 2.00) ( - 1.51) - ,0778 -.I884 ( - 2.53) (-4.01)
- ,2274 ( - 8.62) - ,0588 ( - .82) - ,0497 ( - .98) .0215 (.32) - ,2766 ( - 4.18) - ,0496 ( - .81) - ,0966 ( - 1.31) - ,0689 ( - 2.75)
- ,1842
- ,1908 ( - 3.10) .0623 (.75) ,1149 (5.27) -.I562 ( - 2.65) - .I366 ( - 6.08) - .0399 (-2.75) - .2348 (-3.07) - ,2551 (-3.32) - .1227 ( - 1.56) - ,1078 (-2.72) - .0399 ( - 2.75) - ,1267 ( - 1.80) - .0438 (-.W
- ,0822 (-1.47) .0782 (.88) ,0681 (4.24) - ,0614 ( - 1.11) - ,0687 (-3.52) - ,0338 (-2.39) - ,0657 ( - .89) - ,2085 (-2.33) - ,0189 ( - .22) - ,2182 (-4.51) - ,0338 (-2.39) - ,0972 (-1.47) - ,0002 (- .01)
- .2537 (-2.97) Brazil .0679 (.54) Canada .1440 (4.07) Colombia - ,3764 (-4.99) Cuba - ,2711 (-5.89) Dominican Republic - ,1566 ( - 5.73) Ecuador - ,2965 (-3.16) Guatemala - ,3163 ( - 2.68) Haiti - ,4721 (-4.81) Jamaica - ,2958 ( - 4.48) Mexico - ,1566 ( - 5.73) Panama - ,2717 ( - 2.20) Trinidad and - ,2433 Tobacco ( - 2.15)
- .2723 (-3.69) - ,0944 ( - .91) ,0497 (1.51) - .2372 ( - 3.62) - .0850 ( - 2.92) - ,0628 ( - 3.52) -.1742 ( - 2.08) - ,2695 (-3.17) - ,2447 ( - 2.85) - ,1505 ( - 3.24) - ,0628 ( - 3.52) - ,0221 ( - .24) - .0774 ( - .95)
Note: The t-ratios are presented in parentheses
1960-64
(-6.88) - ,0980 (-1.34) ,0391 (.78) - ,0470 ( - .71) - ,1978 (-3.10) - .0401 ( - .71) - ,0738 (-1.04) - ,1075 ( - 3.26)
1950-59
- .1228 (-4.20) - ,0900 ( - 1.33) ,0371 (.73) - ,0207 ( - .32) ,0060 (. 10) - ,0799 ( - 1.67) - .2214 (-3.55) -.I856 (-5.94)
- ,0944 ( - I .88) ,1511 (.84) ,1138 (.94) - .0143 (-.lo) .2476 (1.72) - .1887 (-1.39) - .0144
- ,0497 ( - .83) - .0006
,0221
- .01) ,0359 (2.45) .0920 (1.69) - .1752 (-7.87) - ,0311 ( - 1.94) - ,0810 ( - 1.20) - ,0959 (-1.13) - ,1056 (-1.29) - ,0780 ( - 1.67) -.0311 ( - 1.94) -.I131 (-1.74) .098 1 (1.04)
( - .01)
-.1108 ( - 2.15)
(1.44)
- ,3063 (-1.59) - ,0065 ( - .26) - ,2959 ( - 1.77) - ,0870 (-1.61) - ,0904 ( - 2.99) ,0581 (.28) ,3290 (1.85) - ,4107 (-2.05) - ,0451 ( - .51) - ,0904 ( - 2.99) ,0544 (.42) - ,1023 ( - .57)
51
Immigration and Self-Selection
1.3.3 Determinants of Selection in Unobserved Characteristics The Roy model suggests that the quality differentials documented in table 1.3 can be “explained” by economic and political characteristics of both the various countries of origin and the United States at the time of migration. Because it is easier to obtain such data for the post-1960 period, and also to maintain comparability with the analysis that will be conducted in the next section across host countries, the empirical study in this section focuses on explaining the variation in quality across the four cohorts that arrived in the post-1960 period. Hence, there are 164 observations (41 countries times 4 cohorts per country) in the data set analyzed here. The aggregate variables used in the analysis are, for the most part, obtained from my earlier study (Borjas 1987) and are described in table 1.4. They include measures of political conditions in the country of origin, mobility costs, and characteristics of the income distribution (the mean and the variance). The empirical analysis of the differences in the present value differentials between immigrants and natives in the 164-observation data set is presented in table 1.5.” The first column of the table presents estimates of the reducedform equation derived in (1 1). This regression reveals that a relatively small number of country-specificvariables explains a large fraction of the inter- and Table 1.4
Definition of Aggregate Variables
Variable FREE COMMUNIST LOSTFREE INEQUALITY
UNEMPLOYMENT USLAW
ENGLISH DISTANCE
AGNP
Definition
Mean (Standard Deviation)
.48 (.SO) .I7 (.38) .08 (.W 7.50 (6.08)
1 if the country had a competitive party system at the time of migration, 0 otherwise = 1 if the country had a Communist government at the time of migration, 0 otherwise = 1 if the country lost a competitive party system within the last 10 years, 0 otherwise = ratio of household income of the top 10 percent of the households to the income of the bottom 20 percent of the households ca. 1970 Unemployment rate in the United States at the time of migration = 1 if migration occurred after 1970,0 otherwise =
Fraction of 1975-80 cohort of immigrants who speak English well or very well Number of air miles (in thousands) between the country’s capital and the nearest U.S. gateway (Los Angeles, Miami, and New York) Difference in (In) GNP per capita between the country of origin and the United States at the time of migration
6.25
(1.64)
.5 ~50) .74 (.25) 3.73 (1.98)
- 1.39
Note: For additional details on the creation of these variables, see Borjas (1987).
(1.05)
52
George J. Borjas
Table 1.5
Determinants of Differences in Unobserved Characteristics Reduced-form Equation
Structural Equation'
Variable
Coefficient
CONSTANT
- ,1574
( - 1.61)
- ,0537
( - 1.70)
FREE
.0410 ,0113 - ,0333
(1.45) C.37) ( - .93) (-1.79) (1.81) (-2.61) (1.70) ~05) (4.02)
,0336 ,0072 - ,0106 - ,0029 ,0108 - ,0505
(4.31) (.69) ( - .86) (-4.56) (1.70) ( - 2.42)
COMMUNIST LOSTFREE INEQUALITY UNEMPLOYMENT USLAW ENGLISH DISTANCE
AGNP
R2 h
- ,0040
.0334
- ,1593 ,0797 ,0003 .0495
t
,402
... ...
...
t
... ...
,382 - .0167
~ ( U S L A W= 0 ) /@SLAW = 1)
Coefficient
...
.0085 - .O419
(-5.24) ( ,791 ( - 3.79)
'All the variables in the structural equation are interacted with A, the selection variable. For details, see eq. (12).
intracountry variance in the unobserved quality of immigrants. Many of the aggregate characteristics are statistically significant. Consider, for instance, the variable measuring the extent of income inequality in the country of origin. The coefficient of this variable is negative and marginally significant, as predicted by the theory. Similarly, the difference between mean GNP in the country of origin and mean GNP in the United States is positive and significant, indicating the fact that migrants from countries with advanced economies are characterized by larger levels of unobserved abilities or productivities. It is worth stressing that the measure of income inequality not only is statistically significant but also has a sizable numerical effect on the quality of the immigrant flow. This point is best illustrated by considering two countries: the United Kingdom and Mexico. The inequality measure takes on a value of 4.0 for the United Kingdom and of 12.3 for Mexico. The regression coefficient in table 1.5 suggests that, holding all other factors constant, Mexican immigrants earn 3-4 percent less than British immigrants simply because of the selectivity effects of higher levels of income inequality. Three other variables seem to be quite important in the regression. The first measures the English proficiency of the immigrant pool. Immigrants from countries where English is prevalent do much better in the United States than immigrants from non-English-speaking countries. Second, the unemployment rate in the United States is an important determinant of immigrant quality: the higher the unemployment rate at the time of migration, the better the quality of the migrant pool. This result is consistent with the Roy model if unemployment particularly effects the earnings opportunities of less-skilled workers.
53
Immigration and Self-selection
For instance, an increase in the unemployment rate will worsen the opportunities for persons in the lower end of the ability (i.e., income) distribution and hence will lead to reduced incentives for these persons to migrate. The quality of the self-selected immigrant pool increases as a result of the withdrawal of these persons from the immigration market. Finally, the reduced-form regression in table 1.5 introduces a dummy variable signaling whether the cohort arrived in the post-1970 period. Recall that U.S. immigration policy was changed drastically by the 1965 amendments (which became fully effective in 1968). Hence, post-1968 cohorts, holding constant characteristics of the country of origin, should differ from earlier cohorts. This is precisely what the results in table 1.5 indicate. In particular, post-1970 cohorts have nearly 16 percent lower (relative) earnings over the life cycle than immigrants who arrived prior to the change in U.S. policy. This result provides striking evidence of a significant structural shift that occurred in the unobserved quality of U.S. immigrants in the last two decades. As noted earlier, since data exist on the emigration rate of immigrants from any given country of origin (i.e., the number of immigrants in a particular cohort as a fraction of the population of the country of origin at the time of migration), the selectivity variable A can be calculated, and the structural equation in (12) can be estimated. The structural equation is written as Q, = hX, and the h-function can be approximated by h = PZ, where Z is the vector of variables proxying for the relevant primitive parameters. Hence, the empirical counterpart to (12) is Q, = P(Zh). This structural equation is presented in the second column of table 1.5. The selectivity variable directly controls for changes in mobility costs and means of income distributions, and these variables are omitted from the structural regression. Remarkably, the structural equation leads to estimates that are highly significant and very supportive of the Roy model. In particular, the inequality variable becomes negative and very significant, the unemployment variable remains positive and significant, and the dummy variable indexing post-1970 cohorts remains negative and strong. The estimated regression parameters can be used to calculate h = 6 Z . The estimates of h are presented at the bottom of column 2 in table 1.5. Three estimates are presented: one evaluated at the mean of all the variables, a second one evaluated at the same means but letting the dummy variable USLAW index pre-1970 cohorts, and a third evaluated at the same means but letting the dummy variable USLAW index post-1970 cohorts. These simulations show that there seemed to be weak positive selection prior to 1970 but very strong negative selection in the post-1970 period. 1.3.4 Determinants of Selection in Education
As noted earlier, self-selection occurs not only on the basis of unobserved ability but also on the basis of observed characteristics such as education. Table 1.2 documented the strong differences in educational attainment across immigrant groups from different countries. In addition, it was seen that the
54
George J. Borjas
observed educational attainment of immigrant groups differed from the mean educational attainment of the population in the country of origin. It is of interest, therefore, to analyze whether the same conceptual framework that explains the differences in unobserved characteristics also explains the differences in educational attainment. The key implication of the Roy model is that highly educated persons are most likely to originate in countries that have a low rate of return to education (relative to that found in the United States). Put another way, holding the mean educational attainment in the population of the source country constant, there should be a negative relation between the mean educational attainment of immigrant national origin groups and the rate of return to schooling in the source country. A detailed study by Psacharopoulos (1973) reports the private rate of return to higher education for a number of countries. Unfortunately, there are only 15 countries in common between his sample and the sample of countries that are important sources of immigration. Nevertheless, using the data presented in table 1.2 for each of the four post-1960 cohorts (so that there are 60 observations), the following regressions were estimated:
+
E(s 1 I > 0) = 13.02 .23 p, -9.76r, (9.78) (1.90) ( - 2.44)
R2 = .236,
E(s I I > 0) = 8.01 + .66p,, -9.82r - 1.44 AGNP, R2 = .421 (4.81) (4.51) (-2.79) (-4.24)
where the dependent variable is the mean educational attainment of the immigrant group in the United States, ps is the mean educational attainment in the source country, r is the rate of return to education in the source country, AGNP is the percentage difference in per capita GNP between the source country and the United States at the time of migration, and the t-statistics are presented in parentheses. The most important finding in these regressions is the significant negative effect of the rate of return to schooling in the source country on the educational attainment of the immigrant pool. A 10 percentage point increase in the rate of return to schooling decreases the mean educational attainment of the immigrant group by almost one year. The key implication of the Roy model, therefore, is confirmed by the data. The educational composition of the sample of immigrants is determined by the relative payoff to schooling in the source country. The regressions also indicate that the mean level of educational attainment in the country of origin has a positive effect on the mean educational attainment of immigrants and that the coefficient, as predicted by the theory, is between zero and one. I should add, however, that this confirmation of the theory-like the results regarding the rate of return to schooling-should be treated with some caution because the data on international differences in mean education levels and rates of return are measured with substantial error.
55
Immigration and Self-Selection
1.3.5 The Unstandardized Wage Differential between Immigrants and Natives Up to this point, the empirical analysis has focused on ascertaining the types of selection that occur on the basis of unmeasured skills and on the basis of educational attainment. As noted in the theoretical section, selection occurs along every single dimension of skills that is valued by the labor market, with particular skills flowing to countries that value those skills the most. As a result of these selection processes, the actual earnings of the typical immigrant in the United States are likely to differ substantially from the earnings of the typical white native. To determine the effect of self-selection in the various dimensions of skills on the earnings gap between immigrants and natives, the 41 regression models were reestimated without including any demographic characteristics in the vector X (except, of course, age and age squared so as to trace the ageearnings profile over the working life). The resulting present value differentials are presented in table 1.6. As can be seen by comparing tables 1.3 and 1.6, the relative earnings of some immigrant groups are lowered significantly when no standardization for demographic variables is conducted. For instance, the most recent immigrant cohort from Mexico earns about 47.8 percent less than white natives, but they earn only about 15.7 percent less than demographically comparable white natives. On the other hand, the most recent cohort of immigrants from the United Kingdom earns about 18.8 percent more than white natives, but only about 10.5 percent more than demographically comparable white natives. In other words, some immigrant groups have demographic characteristics that are much less valuable than those of natives, while other immigrant groups have demographic characteristics that are much more valuable than those of the typical native. It turns out that the same country-specific variables that explain the variation in standardized earnings differentials among source countries also explain the variation in the unstandardized differentials. Table 1.7 analyzes the determinants of the intercountry variation in unstandardized immigrant earnings (analogous to the second-stage regressions on standardized earnings presented in table 1.5). It is worth noting that, as before, immigrants originating in countries with high levels of income inequality have lower earnings than immigrants originating in other countries. This is not surprising since the level of income inequality can be interpreted as a summary index of skill prices. Thus, the key prediction of the Roy model is confirmed by the analysis of the actual earnings (as opposed to the standardized earnings) of immigrants.
1.4 Immigrant Sorting among Host Countries The last section showed that the labor market performance of immigrants currently living in the United States is strongly influenced by the economic and political characteristics of the country of origin at the time of migration.
56
George J. Borjas
Table 1.6
Unstandardized Present Value Differentials between Immigrants and Natives Year of Arrival
Country
Europe: Austria Czechoslovakia
1975-79
1970-74
1965-69
1960-64
1950-59
<1950
,0142 (. 10) ,0561
.1987 (1.72) ,0045 ~03) .0933 (.49) ,1464 (1.89) ,1054 (2.32) - ,2996 (-7.83) - ,0799 ( - 1.19) ,0306 ( .49) - .0947 ( - 3.72) ,0036 ~03) ,4233 (2.50) .0479 (1.21) - ,2518 ( - 5.66) ,2340 (2.94) - ,0061 (-.lo) ,0601 (.87) ,2013 (1.36) .1716 (1.39) .1705 (5.05) .w3 (.lo)
,3104 (4.20) ,0805 (1.54) .3028 (2.73) .0173 (.28) .w37 (3.27) - ,2326 (-7.13) .0442 (.82) .I284 (2.57) - ,0526 ( - 2.73) ,2121 (3.57) ,3319 (3.12) ,0736 (2.33) - ,1962 ( - 5.46) .1141 (1.61) - ,0293 ( - .47) ,0221 ( .39) ,3170 (2.64) ,2385 (2.67) ,1898 (7.88) ,0123 (.36)
,1157 (2.03) ,1289 (1.99) ,0409
,0535 (1.27) .I191 (2.58) .0312
,1135 (1.49) .1745 (2.46) ,1331 (1.07) ,1896 (2.68) .2464
(.50)
Denmark France Germany Greece HwzwY Ireland Italy Netherlands Norway Poland Portugal Romania Soviet Union Spain Sweden Switzerland United Kingdom Yugoslavia
,6004 (3.54) .2658 (3.05) ,1673 (3.64) - .I990 ( - 3.52) -.I137 (-1.34) ,0148 (. 17) - ,0645 (-1.61) .4170 (5.17) ,3004 (2.20) - ,2091 (-4.22) - .2826 (-4.65)
-.I151
(-1.13) - .2086 ( - 3.29) .I182 (1.24) .1826 (1.20) ,2868 (2.42) ,1875 (5.22) - ,0115 (-.17)
(.50) (.W - ,0847 ,0116 (-1.73) ( .26) .0526 ,0359 (3.29) (1.78) - ,0534 - ,0370 ( - 1.37) ( - 1.26) ,0412 ,0925 (3.36) (35) ,0676 ,0299 (2.00) (.94) - ,0132 .oO07 ( - .76) c.03) ,1008 ,0295 (2.50) ~90) ,2751 ,2018 (3.24) (3.26) ,0186 ,0563 (2.45) (.75) - ,1538 - ,1270 ( - 3.70) ( - 3.03) ,0724 ,0808 (1.15) (1.35) ,0429 - ,0118 ( - .26) (1.36) - ,0084 - ,1335 (-.14) ( - 2.35) .1426 ,0867 (1.18) (1.44) .I904 .I589 (2.66) (2.60) ,1198 ,0905 (5.80) (5.20) ,1105 .0938 (3.36) (2.90)
(8.44) - ,0627 ( - 1.09) .3162 (5.20) -.I139 ( - 2.40) ,0344 (1.29) - ,0980 (-1.48) ,0117 (. 14) ,1322 (3.75) ,0159 (. 17) ,0895 (33) .I402 (2.74) - ,0793 ( - .69) .1199 (.98) .2197 (1.99) .0490 (1.58) ,0549 (.83)
57
Immigration and Self-selection
Table 1.6
(continued) Year of Arrival
Country Asia and Africa: China Egypt India Iran Israel Japan Korea Philippines
1975-79
1970-74
- ,3703 (-7.24) ,0101
- ,2659 (-7.59) ,0583 ( .66) ,1339 (2.28) .lo77 (1.39) -.I891 ( - 2.50) ,0683 ( I .02) ,0047 ~03) ,0236 (.73)
(.lo) - ,0705 ( - 1.04) ,1011 (.94) - ,2692 (-2.66) ,1857 (2.24) - ,0846 ( - .87) - . I415 (-2.84)
Americas: Argentina
1965-69
1960-64
1950-59
-.I115
- .0379 (-1.33) ,0914
,0157
- ,0562 ( - 1.05) .3073 (1.59) ,2143 (1.68) ,2627 (1.68) ,3943 (2.56) - .0139 (-.lo) .3015 (1.30) - ,1758 (-3.31)
( - 3.97)
,1559 (2.03) (1.17) ,2617 ,3240 (4.87) (6.1 1) ,1501 ,1339 (2.09) (1.89) -.I761 - ,0764 (-1.12) ( - 2.49) .0927 ,0565 (1 S5) (37) ,1141 .0826 (1.06) (1.52) - ,0109 .0245 ~ 9 2 ) ( - .32)
(51)
.I147 (1.59) .2784 (5.24) ,2055 (2.90) ,1744 (2.99) ,0500 (1.00) - .0169 ( - .26) - ,1089 (-3.31)
- ,2239 - ,3123 -.I801 - ,0122 ,0736 ( - 3.98) (-2.76) (-2.47) (1.17) ( - .20) ,0717 Brazil .I575 - ,0816 .0615 .0807 (1.19) ( - .74) (.82) (.74) (.87) Canada .I862 ,0593 .0654 .I069 .0685 (4.95) (1.69) (4.58) (4.37) (3.80) Colombia - ,4451 - .3141 - ,1903 - .0522 ,1791 (-5.59) (-4.53) ( - .89) (-3.04) (3.10) - ,1693 Cuba - ,3450 - .0234 -.1877 - .I658 (-5.52) ( - 7.13) (-7.08) (-1.15) (-7.93) Dominican Republic - ,6794 - ,4560 - ,3963 - .3034 - ,1546 (-6.89) ( - 8.37) (-6.81) (-5.44) ( - 2.37) Ecuador - ,4016 - .2428 - ,2789 - ,0339 ,0186 ( - 2.75) (-4.06) (-3.44) (e.44) (.26) Guatemala - ,4927 - ,4041 - ,3604 - ,2532 - ,1017 (-4.41) ( - 3.94) ( - 4.48) ( - 2.68) (-1.14) Haiti - ,5907 - ,3037 ,0069 - .I307 - .0527 (-3.31) ( - 5.64) (-1.55) ( - .61) (. 10) Jamaica - .3938 - .2388 - ,1406 - ,2124 - .0471 (-4.85) ( - 5.62) (-4.12) (-3.35) ( - .94) Mexico - .4780 - ,3518 - .3074 - ,2798 - ,2350 ( - 17.14) ( - 19.94) ( - 21.97) ( - 20.59) ( - 15.06) Panama - ,3263 - ,0653 - ,1376 - .0905 - ,0433 (-.71) (- 2.49) (-1.85) (-1.29) ( - .63) Trinidad and - .3313 -.I167 ,0693 - ,0435 ,1741 Tobago ( - 2.73) ( - 1.35) (1.73) (.75) (e.60)
Note: The r-ratios are presented in parentheses.
,4027 (2.47) - .2328 (-1.14) ,0537 (2.06) -.I150 ( - .65) - ,0256 ( - .45) - ,0256 ( - .33) ,1277 (.58) ,4213 (2.23) - ,2614 (-1.22) ,0514 (.55) - ,2918 (-9.42) ,2219 (1.64)
.0340 (. 17)
58
George J. Borjas
Table 1.7
Determinants of Differences in UnstandardizedEarnings Differentials Reduced-form Equation
Variable
Coefficient
CONSTANT
- ,4069
FREE
,0372 ,1056 - ,0383 - ,0061
COMMUNIST LOSTFREE INEQUALITY UNEMPLOYMENT USLAW ENGLISH DISTANCE
AGNP R2 h h(USLAW = 0) h(USLAW = 1)
,0188
-.I322 ,3917 .0145 .0133 .754
t
(-4.39) (1.39) (3.71) ( - 1.34) ( - 2.30) (1
.w
(-2.13) (9.79) (3.38) (1.06)
Structural Equationa Coefficient - ,0505
,0692 ,0344 - ,0047 - ,0041
,0054 - .0354
t
(-1.24) (7.50) (2.73) ( - .34) (-4.00) ( .66) ( - I .27)
... 3 2 - ,0248 - ,0071 - ,0425
(-6.42) ( - .52) ( - 2.77)
"All the variables in the structural equation are interacted with A, the selection variable. For details, see eq. (12).
Potential emigrants in the source countries, however, chose to come to the United States instead of migrating to other potential countries of destination. In a sense, the observed pool of immigrants in the United States is the outcome of competition in the immigration market among various countries of destination. Different countries, by offering different immigration policies and different income distributions, will attract different kinds of immigrants. Three countries, Australia, Canada, and the United States, have been the main countries of destination for permanent migrants in recent years. U.N. statistics, for example, report that, in the period 1975-80, nearly five million persons migrated to a different country, with nearly two-thirds of these individuals migrating to one of these three countries. Each of these countries, of course, is characterized by a long history of immigration. The size of the recent flows generated by the self-selection of immigrants into each of the three potential countries of destination is illustrated in table 1.8. Over the period 1959-81, about 14.7 million persons legally left the various countries of origin and migrated to either Australia, Canada, or the United States. Sixty percent of these migrants chose the United States as their destination, and the remainder were evenly split between Australia and Canada. Table 1.8 also shows that these statistics vary significantly between the early part of the period (1959-70) and the later (197 1-8 1). Recent migrants are disproportionately more likely to select the United States as their destination (nearly two-thirds of migrants in the 1970s chose to do so) and disproportionately less likely to choose Australia as their destination (only 14 percent did so).
Migration Flows to the United States, Canada, and Australia
dom uding U.K.)
Period of Migration
1959-70
I97 1-8 1 Number
1959-81
Number (1,OOOs)
% to U.S.
% to Canada
% to Australia
(1,OOOs)
%to U.S.
%to Canada
% to Australia
115.1 2,111.6 708.3 1,322.9 2,583.4 123.7 6,965.0
37.5 84.9 69.5 20.3 47.5 18.9 55.2
29.6 13.4 19.2 28.8 28.9 32.5 23.3
32.8 1.7 11.3 50.9 23.6 48.6 21.5
220.5 2,687.7 2,580.8 751.1 1,309.2 176.9 7,726.2
48.3 81.0 73.5 18.4 55.7 23.5 65.9
32.4 15.9 17.7 31.7 26.0 19.4 20.3
19.3 3.1 8.7 49.9 18.3 57.2 13.8
~~
(1,OOOs)
9% to U.S.
%to Canad
335.5 4,799.3 3,289.0 2,074.0 3,892.6 300.5 14,690.9
44.6 82.7 72.7 19.6 50.3 21.6 60.8
31.5 14.8 18.0 29.8 27.9 24.8 21.7
Number ~~
S. Department of Commerce, Statistical Abstract of the United States; U.S. Immigration and Naturalization Service, Statistical Ye and Naturalization Service; Canada Statistics, Historical Statistics of Canada and Canada Yearbook; Australian Department of Im rs, Australian Immigration (1982).
60
George J. Borjas
These aggregate statistics mask important country-of-origin differences. During the period 1971-81, the United States was less likely to attract immigrants from Africa, the United Kingdom, and Europe and significantly more likely to attract immigrants from Asia or North and South America. Canada, on the other hand, seemed a relatively attractive destination for immigrants from Africa, the United Kingdom, and Europe, while Australia was the destination of choice for persons emigrating from the United Kingdom: nearly half the two million persons who left the United Kingdom in the period 19598 1 migrated to Australia. l 3 1.4.1 Migration Policies in Host CountriesI4 One important constraint on the size and the composition of the flow of migrants to potential host countries is the set of statutes and policies used by the various countries to screen the applicant pool. U.S. immigration policy, prior to the 1965 amendments to the Immigration and Nationality Act, was guided by the objective of restricting the migration of persons whose national origin did not resemble the national origin and ethnic composition of the United States population in 1920. The 1965 amendments abolished the “discriminatory” national origin quota system and instituted the goal of family reunification as the main objective of U.S. immigrant policy. These changes, as we saw above, may have been responsible for a very large decline in the unobserved skills of immigrants admitted by the United States. Canadian immigration policy, until 1961, also had a preferential treatment of immigrants originating in Western European countries. The 1962 Immigration Act (and further relatively minor changes in the statutes and regulations through the 1970s) removed the country-of-origin and racial restrictions and shifted emphasis toward skill requirements. Under the new regulations, potential migrants who were not relatives of Canadian citizens or residents could enter Canada if they passed a “test.” Applicants were graded and given up to 100 points according to a “point system,” and 50 points were necessary to obtain permission to migrate to Canada. These points were given according to the applicant’s education (a point per year of schooling, up to 20 points), occupational demand (10 points if the applicant’s occupation was in strong demand in Canada), age (up to 10 points for applicants under the age of thirtyfive, minus 1 point for each year over age thirty-five), a “personal assessment” by the immigration officer that was valued up to 15 points, etc. In 1976, the Canadian Immigration Act was amended to incorporate the goal of family reunification as an important policy objective. Australian immigration policy has a long history of restricting the migration of persons who are not of British origin. These restrictions, known as the “White Australia Policy,” operated both in terms of denying entry to persons of non-British or non-Northern European origin and in terms of denying financial assistance (to cover transportation and resettlement expenses) to undesirable migrants.
61
Immigration and Self-Selection
World War I1 raised doubts among Australian officials about the feasibility of defending a large continent with a small population, and a series of governments pursued a national policy of substantially increasing the number of immigrants who chose Australia as their destination. This objective, however, could not be achieved by allowing only British citizens to immigrate, and thus Australia began looking elsewhere for migrants (e.g., Germany, the Netherlands, Malta, Italy, and Greece all signed formal arrangements with the Australian government to recruit and assist persons from these countries in their migration to Australia). Further political changes in Australia led to the abolishment of the White Australia Policy in 1972. An immigration policy devoid of discrimination by national origin and race was announced, and a point system based on the Canadian system was instituted. During the early 1980s, Australia began to stress the concept of family reunification in its migration policy (see Birrell 1984). It is unlikely, however, that this shift in policy will have much effect on the 1981 Australian Census data that will be analyzed below. The effect of these changes in immigration policy on the national origin composition of the immigrant pool in each of the countries is documented in table 1.9. In all host countries, the national origin of the immigrant population has changed drastically over time. For example, in both Canada and the United States, the share of migrants originating in European countries declined drastically between the 1960s and the 1970s. During the 1960s, 23.5 percent of immigrants to Canada originated in the United Kingdom, and an additional 46.0 percent originated in other European countries. During the 197Os, the fractions had fallen to 15.2 and 21.7 percent, respectively. Conversely, the fraction of immigrants originating in Asia was only 8.4 percent during the 1960s, and this fraction had increased to 29.1 percent during the 1970s. Table 1.9 shows that the United Kingdom accounted for nearly half the migrants to Australia during the 1960s but for only a third of the migrants during the 1970s. A similar decline is observed in the fraction of Australian immigrants originating in other European countries: from 40.8 to 22.4 percent. On the other hand, the fraction of immigrants from Asia increased from 5.3 to 21.1 percent, a fourfold increase in a ten-year period. 1.4.2 Data and Descriptive Statistics The data are drawn from Public Use Samples of the Censuses available for each of the three destination countries. The U.S. data are identical to that used in the previous section and require no further description. The Canadian Censuses were conducted in 1971 and 1981. Both these Censuses have the important characteristic that they report the year in which foreign-born persons arrived in Canada. Hence, the agingkohort decomposition described in section 1.2 can be carried out. The 1971 data for both immigrants and natives residing in Canada are a 1/100 random sample of the
Table 1.9
Migration Flows into the United States, Canada, and Australia, 1959-81 Period of Migration United States 1959-70
Origin Africa America Asia United Kingdom Europe Oceania and other Total
1971-81
Canada 1959-81
1959-70
Australia
1971-81
1959-8 1
1959-70
197 1-81
1959-81
Number %of Number %of Number %of Number %of Number %of Number %of Number %of Number %of Number %of (1,OOOs) Total (1,OOOs) Total (1,OOOs) Total (1,OOOs) Total (1,OOOs) Total (1,ooOs) Total (1,OOOs) Total (1,OOOs) Total (1,OOOs) Total 149.7 1.7 106.5 2.0 43.2 1.1 1,792.0 46.6 2,175.7 42.7 3,967.7 44.3 492.2 12.8 1,898.1 37.2 2,390.3 26.7
34.1 2.1 283.5 17.5 136.3 8.4
71.5 4.6 427.9 27.3 457.3 29.1
268.8 7.0 1,228.2 31.9
381.2 23.5 745.4 46.0
237.8 15.2 619.0 340.1 21.7 1,085.5
23.4 3,847.8
.6
138.5 2.7 407.3 729.5 14.3 1,957.7 41.5 5,089.9
.8
64.9 8,937.7
4.6 21.9 .7
40.2 1,620.7
2.5
34.3 1,568.9
2.2
105.6 3.3 711.4 22.3 593.6 18.6
74.5 3,189.6
19.4 34.0 2.3
37.8 36.1 79.8
2.5 2.4 5.3
42.5 4.0 84.1 7.9 225.4 21.1
80.2 3.1 120.2 4.7 305.1 11.9
672.9 44.9 609.8 40.8
374.8 35.1 239.6 22.4
1,047.7 40.9 849.4 33.1
60.1 1,496.3
4.0
101.1 1,067.5
9.5
161.1 2,563.7
6.3
Sources: U.S. Department of Commerce, Statistical Abstract of the United States; U.S. Immigration and Naturalization Service, Statistical Yearbook of the Immigration and Naturalization Service; Canada Statistics, Historical Statistics of Canada and Canada Yearbook; Australian Department of Immigration and Ethnic Affairs, Australian Immigration (1982).
63
Immigration and Self-Selection
Canadian population, while the 1981 micro file is a 2/100 random sample of the Canadian population. All observations that satisfy the sample restriction of being prime-age men (aged 25-64), not self-employed, not residing in group quarters, and whose records report positive annual earnings in the year prior to the Census are used in the analysis. The Australian data used in this paper are drawn from the 1981 Census of Population and Housing, the only micro Australian Census file available at present. This Census file is a 1 / 1 0 random sample of the Australian population, and the entire sample (for both immigrants and natives) that satisfies the sample restrictions listed above is used. Three important problems are raised by the Australian data. First, only one Census is available; therefore, the aginglcohort decomposition cannot be conducted. The Australian results, therefore, are not directly comparable to those for the other two countries. Nevertheless, a simple solution that allows some rough comparisons will be proposed below. Second, the Australian Census does not report annual earnings but instead reports annual incomes (which include nonsalary receipts). This problem may not be very serious since the analysis focuses on nativehmmigrant earnings differences, and self-employed persons are omitted from the study. Finally, the Australian Census (unlike the U.S. or Canadian data) does not contain good measures of labor supply. Hence, a wage rate for the year prior to the Census cannot be calculated. The empirical analysis in this section, therefore, will be conducted on the logarithm of annual earnings. Table 1.10 presents summary statistics (mean earnings and education) as well as sample sizes for the various samples that will be used in the analysis.Is In addition, table 1.10 decomposes the immigrant population in each of the host countries according to the continent of origin. This decomposition by continent (rather than country) is mandated by the fact that, in both the Australian and the Canadian Censuses, the decomposition by country leads to a very small number of observations for most countries. In addition, the Canadian Censuses identify the country of origin only for a select group of Western European immigrants. The results for the United States, as expected, show a downward trend in the earnings of immigrants (relative to natives) over the decade. The average immigrant in 1970 earned, on average, about as much as the typical native worker. By 1980, however, immigrant earnings were about 15 percent below the native wage. The Canadian data show little change in the relative earnings of immigrants between 1971 and 1981. In both Censuses, the average immigrant had slightly higher earnings than the typical native worker. The exception seems to be immigrants originating in Latin America: their earnings are about 10 percent lower than those of Canadian natives in 1971 but 19 percent lower in 1981. The Australian Census shows that the typical immigrant in 1981 had roughly the same earnings as the typical native person and that the differential varied somewhat by country of origin.
64
George J. Borjas
Table 1.10
Summary Statistics Country of Destination: United States 1970
1980
Country of Origin
In ( w )
EDUC
N
In(w)
EDUC
N
Natives Asia Africa Europe Latin America All immigrants
8.99 8.88 8.88 9.06 8.67 8.95
11.5 13.3 13.9 10.8 9.2 10.8
28,978 3,495 172 16,922 7,507 32,491
9.61 9.47 9.40 9.69 9.23 9.46
12.7 14.6 15.3 12.1 9.4 11.7
15,071 25,288 2,622 42,734 48,929 134,252
Canada 1971
Natives Asia Africa Europe Latin America All immigrants
8.82 8.72 8.86 8.86 8.72 8.86
9.9 13.2 14.1 10.0 12.0 10.5
1981 28,049 409 119 6,633 223 8,018
9.79 9.66 9.74 9.86 9.60 9.81
11.3 13.6 14.0 10.9 12.1 11.7
61,205 2,372 504 12,193 1,229 17,417
Australia, 1981 Natives Asia Africa Europe Latin America All immigrants
9.39 9.34 9.45 9.34 9.35 9.36
11.6 12.9 13.1 11.4 12.1 11.7
23,086 1,074 267 7,799 102 9,936
It is instructive to compare the Australian statistics with the relevant numbers for Canada and the United States. For instance, European immigrants in Australia actually have the lowest education levels of any of the migrant groups in Australia and have a wage disadvantage of only 5 percent. In Canada, European immigrants also tend to have slightly lower educational levels but higher earnings than natives, while in the United States European immigrants outperform all other immigrant groups despite the fact that they have lower educational levels than the native population. This comparison (as well as similar comparisons for other regions of origin) reveals the nonrandom sorting of migrants across the various host countries. An important insight is provided by these statistics: generalizations about the productivity or earnings capacities of ethnic or national groups are misleading since they ignore the self-selectivity that generated the composition of the migrant pool in each of the host countries. In other words, there is no such thing as the effect of Asian ethnicity or race on immigrant earnings. The value
65
Immigration and Self-selection
attached by the host country’s labor market to ethnichacia1 characteristics depends greatly on the kinds of selections that generated the particular flow of migrants. 1.4.3 1980-81 Cross-sectional Regressions Since the agingkohort decomposition cannot be conducted for the Australian data, it is instructive to begin the empirical analysis by focusing on the 1980-81 cross section. Table 1.11 presents the cross-sectional earnings function estimated separately in the samples of immigrants and natives in each of the three countries of destination. The regressions in the native samples are of interest mainly because they are so similar across the destination countries. The coefficients of age, marital status, and urbanization status all have the expected signs and are of similar magnitudes whether the labor market is in Australia, Canada, or the United States. The only coefficient that seems to be an outlier in the native samples is that of education in Australia, where the coefficient is almost twice as large as that in the United States or Canada. The regressions in the immigrant samples are interesting because they illustrate the general result that practically all socioeconomic variables have a smaller effect among immigrants than among natives regardless of the country of destination. The earnings of immigrants are much less responsive to socioeconomic characteristicsthan the earnings of natives in these economies. The immigrant regressions in table 1.11 also include a vector of variables indicating the time of migration.I6 An important use of these coefficients (and of the socioeconomic variables) is to predict the size of the wage differentials between immigrants and natives for each of the cohorts. These predictions are calculated using the mean socioeconomic characteristics of the immigrant sample in each of the host countries. In addition, these predictions are obtained by holding the age of immigration constant at 20 for all cohorts. Hence, the typical immigrant in the 1975-1980 cohort is 23 years old when the prediction is calculated, the typical immigrant in 1970-74 is 28 years old, etc. The predicted age-earnings profile, therefore, incorporates both aging and cohort effects. These profiles are presented in table 1.12. The U.S. and Canadian profiles resemble the ones usually reported in the literature: the earlier cohorts, either because they are older and have been in the country longer or because there are vintage or cohort effects, do much better in the labor market than more recent cohorts. Table 1.12, however, shows that the Australian experience is very different. The Australian crosssectional age-earnings profile for immigrants is essentially flat! In fact, it is impossible to find any statistical difference in the relative earnings of immigrants among the cohorts that arrived in Australia after 1950. Their relative earnings hover around 7-8 percent less than natives, and there is no discemible trend over time. This result implies that, if there is any assimilation effect in Australia, the quality of immigrants to Australia must have increased during the period 1960-80. Hence, a simple comparison of the cross-sectional
66
George J. Borjas
Table 1.11
1980-81 Cross-sectional regressions
Country of Destination United States Sample
Canada
Australia
Coeff.
t
Coeff.
t
Coeff.
t
6.6488 .0587 ,0841 - ,0009 ,3151 - ,3337 ,1545 ,193
(76.33) (33.92) (20.17) ( - 18.00) (23.53) ( - 15.15) ( 12.07)
7.0465 .0510 ,0873 - .0009 ,2973
( 193.01)
6.3522 .0908 ,0886 - ,001 1 .2727
(104.68) (58.77) (32.01) ( - 34.61) (31.31)
...
...
. . .
...
,1036 1.71
(22.78)
,1605 ,245
(16.61)
6.6378 ,0497 ,0802 - ,0009 ,2325 - ,3502 ,0574 .2107 ,3141 ,3750 ,4436 ,4752 ,226
(223.77) (133.61) (55.39) ( - 51.35) (50.52) ( - 34.48) (9.43) (36.81) (51.89) (56.74) (74.88) (64.63)
7.3415 ,0415 ,0710 - .oO08 ,2190
(95.72) (40.97) (19.31)
6.7307 ,0748 .0779 - ,0010 ,2013
(66.17) (35.59) ( 16.86) ( - 18.70) (14.16)
Natives: CONSTANT EDUC AGE AGE' MAR HLTH URBAN
R'
(76.26) (49.42) (-45.21) (51.10)
All immigrants: C0N S TAN T EDUC AGE
AGE^ MAR HLTH URBAN
1970-74Wave 196549 Wave 1960-64 Wave 1950-59 Wave Pre-1950 Wave RZ
(- 18.44)
(18.42)
...
...
- ,0016 ,1609 ,2816 ,2825 ,3679 .4287 .I63
(-.16) (9.73) ( 18.03) (1 5.39) (25.59) (17.50)
... ,1079 .0444 ,0491 .0810 .0811
.1159
...
(5.41) (2.11) (2.36) (3.68) (4.18) (4.63)
1.88
regressions across the destination countries leads to an important finding about the trends that mark the self-selection of immigrant flows to the host countries over the last two decades. 1.4.4 Present Value Differentials
Since two Censuses are required to identify aging and cohort effects, the analysis of equations (26) and (27) is initially restricted to the U.S. and Canadian Censuses. Within each country of destination, five immigrant samples will be analyzed: the pooled sample and the subsamples of immigrants originating in Africa, Asia, Europe, and Latin America. The regression will contain a vector of demographic characteristics including education, marital status, health status (when available), and metropolitan residence. These regressions are used to calculate the present value differential between immigrants and demographically comparable natives for each of the cohorts. These present value differentials are presented in table 1.13. (The data presented in table 1.13 for Australia will be discussed in detail below.)
67
Immigration and Self-Selection
Table 1.12
Earnings Differentials between Immigrants and Comparable Natives in 1980-81 Cross Sections Immigrant Cohort
Origin and Destination All immigrants in: United States Canada Australia
1975-80
- .3460 (-14.48) -.2271 (-9.52) -.Of310 (-2.51)
1970-74
- ,1534 (-.10.42) -.1118 (-6.61) -.0642 (-2.87)
1965-69
- .0676 (-6.91) -.0286 (-2.35) -.0814 (-4.98)
1960-64
- ,0239 (-2.58) -.0571 (-3.99) -.0656 (-4.05)
1950-59
,0177 (1.79) -.0020 (-.22) -.0796 (-6.06)
< 1950 ,0045 (.39) ,0558 (2.78) -.0342 (-1.82)
Note: The r-ratios are presented in parentheses.
Consider initially the pooled sample of immigrants. Table 1.13 documents the systematic decline in the quality of immigrants arriving in the United States over the last two decades. For instance, the typical immigrant arriving in 1960-64 in the United States had only a slight wage disadvantage relative to a comparable native, while the typical immigrant arriving in the United States in 1975-79 has a wage disadvantage of nearly 27 percent over the life cycle as compared to the native baseline. Remarkably, the Canadian Censuses reveal very similar patterns: the 1960-64 migrant to Canada had a 6 percent wage disadvantage over the life cycle (relative to natives), while the disadvantage for the most recent migrants (1975-80) has increased to nearly 23 percent. The American and Canadian trends are less similar when the analysis is restricted to men from a specific country of origin. For example, among European immigrants, the U.S. Census reveals a substantial decline in quality (from a 4 percent advantage to an 11 percent disadvantage) over the last twenty years, while the Canadian Census reveals a roughly stable wage differential between immigrants and natives over the post- 1960 cohorts. Similarly, among Asian immigrants, the Canadian data reveal that the 1960-64 and the 1975-80 cohorts had essentially the same relative standing, while the U.S. data reveal a decline in quality from a 15 to a 27 percent disadvantage. These results, therefore, imply that at least part of the similarity between the United States and Canada at the aggregate level is due to the fact that the national origin composition of the cohorts shifted over time, away from European immigrants (who tend to do quite well in the labor market) to Asian and Latin American immigrants (who do much worse in the labor market). As noted earlier, the Australian Census is available only for 1981. Since cohort and aging effects cannot be identified, the present value differentials cannot be calculated directly. Recall, however, that the 1981 cross-sectional regressions estimated in the Australian data showed that immigrants in
68
George J. Borjas
n b l e 1.13
Present Value Differentials between Immigrants and Comparable Natives Year of Arrival
Group All immigrants in: United States Canada
1975-80
- ,2656 ( - 18.99) - .2297
( - 13.25)
Australia
.0149 (.46)
African immigrants in: - ,3779 United States (-5.11) Canada -.4092 (-3.00) Australia - ,1688 (- 1.01) Asian immigrants in: United States Canada Australia
- ,2692 ( - 11.47)
-.3930 (-6.88) - .0634 (- .84)
European immigrants in: - .lo68 United States ( - 6.06) Canada -.0516 ( - 2.22) Australia .0745 (1.68) Latin American immigrants in: United - ,2716 ( - 14.62) States Canada -.3312 ( - 3.77) Australia ,1671 (.61)
1970-74
1965-69
1960-64
1950-59
< 1950
-.1228 (-12.20) -.1306 (-8.57) ,0136 (.61)
-.0827 (-10.40) -.0449 (-3.75) -.0570 (-3.49)
-.0453 (-6.88) -.0632 (-4.63) -.0740 (-4.57)
-.0260 (-4.37) -.0344 (-3.57) -.1330 (-10.12 )
-.0451 (-4.38) ,0212 (1.10) -.IN14 (-4.86)
-.3097 (-6.08) -.4555 (-3.23) -.2197 (-1.90)
-.1425 (-3.21) -.2690 (-2.03) -.I191 (-1.42)
-.1577 (-3.62) -.3297 (-2.55) -.I317 (-1.26)
-.I997 (-4.28) -.2595 (-2.65) -.3413 (-.88)
-.I806 (-1.69) .2108 (.61)
-.2487 (-9.08) .0637 (34) ,0141 (.20)
-.1565 -.4117 ( - 10.53) ( - 8.33) - ,2534 - .3658 (-4.86) (-6.56) -.2348 .0022 (.04) (-4.75)
- ,0167 ( - 1.25) ,0113
-.I495
-.2551
( - 9.89)
(- 17.54)
- .3651 (-6.38) -.2367 (-3.63)
- .3868 (-10.19) -.3817 (-7.42)
(55) ,0350 (1.33)
,0218 (2.14) .0022 (.14) -.0524 (-2.87)
,0436 (5.07) -.0290 (-1.92) -.0732 (-4.26)
-.1273 (-9.53) - ,2820 (-3.25) - ,0677 (-.38)
-.1243 (-11.42) - ,1693 (-2.10) - .3991 (-2.45)
-.0841 (-8.91) - ,1230 (-1.46) - ,2721 (-1.15)
Note: The r-ratios are presented in parentheses.
,0307 (4.44) .0116 1. 04) -.1121 (-8.15)
-.1282 (-13.56) - ,1757 (-3.07) ,0827 (.15)
-.4481
(-3.26)
,0219 (1.79) .0423 (2.04) -.0833 (-4.18)
-.1629 (-8.18) ,1788 (.91) -.2868 (-.70)
69
Immigration and Self-Selection
Australia face significantly different age-earnings profiles than their counterparts in the United States and Canada. In particular, in the cross section, there seems to be little relation between the earnings of immigrants in Australia and the length of residence in Australia. If there is any assimilation effect in Australia, therefore, this result must imply that the quality of immigrants to Australia has increased over the last two or three decades. A rough estimate of this increase can be obtained if it is assumed that the unobserved assimilation effect experienced by immigrants in Australia resembles the assimilation effect of similar persons (i.e., persons from the same country of origin) in Canada or the United States. Given this approximation, the assimilation effects can then be subtracted from the Australian crosssectional coefficients (thus netting out the role played by pure aging in the generation of the cross-sectional results), and the present value differentials can be computed for each of the cohorts. Since there are two sets of assimilation parameters (one for Canada and one for the United States), a number of different approximations can be calculated. In general, these experiments led to similar qualitative findings. In this paper, therefore, the assimilation rate used is the average of the two assimilation rates (i.e., the U.S. and Canada aging effects) experienced by immigrants from the same continent of origin. Given these assimilation rates and the cross-sectional regressions estimated in the Australian Census for each region of origin, it is a simple matter to calculate the predicted present value differential between the various cohorts of immigrants and comparable natives in Australia. These predictions are also presented in table 1.13. Two substantive results are worth noting. As implied by the flat earnings profiles found in the (pooled) Australian cross section, the quality of immigrants to Australia increased slightly over the last twenty to thirty years. The typical immigrant in 1960-64 could expect a 7 percent wage disadvantage over his life cycle, while the typical immigrant in 1975-80 has no wage disadvantage relative to natives over his life cycle. Second, this increase in immigrant quality can essentially be found in every one of the national origin groups under analysis. For example, the typical European immigrant in the early 1960s had a 7 percent wage disadvantage, while the typical European immigrant in the late 1970s has a 7 percent wage advantage over natives. Similarly, the average Asian immigrant in the early 1960s had 24 percent lower earnings over his life cycle than natives, while the differential is only 6 percent (and insignificant) for the most recent migrants. The data presented in table 1.13 provide a unique descriptive analysis of an important question. Which host countries are the “winners” and which are the “losers” in the immigration market? This comparison, of course, depends on the assumption that the native base across countries has a similar level of productivity and skills. This assumption makes the relative wage of immigrants across host countries directly comparable as an index of immigrant quality. The assumption that natives among the three host countries are roughly simi-
70
George J. Borjas
lar is not empirically verifiable. However, it does not seem unreasonable since all three countries share a common language and culture, have similar political and economic systems, and are at similar stages of economic development. Given this assumption, the statistics presented in table 1.13 present an interesting story of the extent of self-selection in the generation of the foreignborn population in each of the countries. Consider the trends for the pooled sample. During the 1940s and 1950s, Australia was attracting immigrants who had lower productivities than the immigrants attracted by Canada and the United States. This type of selection, however, was drastically reversed during the 1960s, as both Canada and the United States began to attract persons who did not perform as well in the labor market and Australia began to attract immigrants with relatively high levels of unobserved skills. 1.4.5 Determinants of Immigrant Quality Consider the following regression model: (35) where Q,,(t) is the present value differential between immigrants and comparable natives of a cohort migrating from country i to country j at time t , X , ( t ) is a vector of variables describing conditions in the country of origin i at time t , and X J ( t ) is a vector of variables describing conditions in the country of destinationj at time t . The specification of (35) builds in a very important (and restrictive) assumption. In particular, the relative earnings of a person from country i in country j at time t are independent of events in other periods t’ ( t # t ’ ) , and, more important, they are also independent of conditions in other countries (particularly, they are independent of conditions in other potential countries of destination). This empirical framework, in a sense, introduces an “independence of irrelevant alternatives” assumption into the study. Although this assumption is not likely to be strictly satisfied, it does simplify the empirical analysis greatly. If the assumption was invalid, for instance, the right-hand side of (35) would have to be expanded to include the characteristics of all other potential countries of destination, and the increase in the number of variables would rapidly drive the number of degrees of freedom to zero. Table 1.14 presents the estimates of the reduced-form equation in (35). The sample consists of 48 observations (4 continents of origin times 4 post-1960 cohorts times 3 countries of destination). The regression in table 1.14 reveals that a small number of characteristics of the countries of origin and the countries of destination do “explain” a very large fraction of the variance in the unobserved quality of immigrants. The variables in the reduced-form equation, for example, explain over 80 percent of the variance in the quality measures presented in table l . 13. Despite this success, however, it must be noted that, because the countries of origin are defined in terms of continents, the
71
Immigration and Self-Selection
Table 1.14
Determinants of Immigrant Quality across Host Countries Variable
Coefficient
CONSTANT
.1252 -.0511 .0011
(-2.77) (- 1.79)
- ,0044
(- 1.89)
,0431 .0903 ,801
(4.35) (8.78)
USLAW
UNEMPLOYMENT
INEQUALITY(~) INEQUALITY(1) AGNP RZ
t
Note: Key to additional variables: UNEMPLOYMENT = unemployment rate in the host country at the time of migration; INEQUALITY(O) = average income inequality (as defined in table 1 . 4 ) in selected countries from continent of origin in decade of migration; INEQUALITY(~)= inequality measure for destination countries in decade of migration; AGNP = difference in (1n)GNP per capita between sending and host countries at time of migration.
two variables measuring country-of-origin characteristics (the relative GNP level and the extent of income inequality) are, in effect, averaged over a large and diverse number of countries.” It is unclear what biases are caused by this aggregation, but it is important to remember that the coefficients in table 1.14 are, at best, suggestive of the underlying economic behavior. Both the GNP of the continent of origin (relative to GNP per capita in the country of destination) and the inequality measure for the continent of origin affect the quality of migrants significantly. Migrants from wealthier regions do better no matter where they go, and migrants from regions with large levels of income inequality do worse than other migrants. Similarly, the inequality measure for the country of destination has a positive and significant effect on relative immigrant earnings, as predicted by the Roy model. Finally, the change in U.S. immigration policy (as measured by USLAW) has a negative and marginally significant effect and thus helps identify the effect of this major change in policy relative to other countries. The change in U.S. immigration policy lowered the earnings of migrants by 5 percent relative to the earnings of migrants who chose other countries of destination.
1.4.6 The Point System and Immigrant Quality It is somewhat surprising that the cohort quality trends in Canada and the United States are so similar despite the major differences in immigration policies between the two countries. Immigration policies, however, can screen applicants only on the basis of observed demographic characteristics such as education, occupation, and age. The results summarized in table 1.13 show that even a stringent point system, such as that used by Canada, was unable to prevent a decline in immigrant skills in that country similar to that experienced in the United States. On the other hand, however, the point system clearly affects the observable skills of the incoming immigrant flow. A clear way of ascertaining the impor-
72
George J. Borjas
tance of this direct effect is to reestimate the regression model on the pooled sample of immigrants without controlling for any demographic characteristics. The unstandardized differentials in the present value of lifetime earnings are presented in table 1.15. It is evident that, during the 1970s, immigrants admitted to the United States were substantially less skilled (relative to U.S. natives) than immigrants admitted to other host countries (relative to the natives of those countries). It is also evident that the point system generated an immigrant flow into Canada that had relatively favorable socioeconomic characteristics. For instance, even though the most recent immigrants into Canada (the post-1975 arrivals) earned 23 percent less than comparable Canadian natives, they earned only 12.2 percent less than the typical Canadian native. These immigrants, therefore, have more favorable demographic characteristics than Canadian natives. Although the point system was unable to prevent the decline in immigrant skills, it greatly tempered the extent of the drop. This fact has interesting implications. If a host country decides that it wishes to attract more skilled immigrants, a point system seems to be a very direct and simple way of doing so. At the same time, however, the point system has the major limitation that it cannot screen for unobservables, and these unobservables are major determinants of individual earnings. Hence, as long as economic and political conditions motivate relatively unskilled persons to emigrate, the point system may restrict entry to only those who pass the test, but the immigrant pool will be composed mostly of relatively unskilled persons with the acceptable demographic characteristics. Table 1.15
Standardized and UnstandardizedPresent Value Differentialsin Host Countries All Immigrants in:
Year of Arrival
United States
Canada
-
Australia
Standardized Unstandardized Standardized Unstandardized Standardized Unstandardized
- ,2656 18.99) - ,1228 1970-74 ( - 12.20) 1965-69 - ,0827 ( - 10.40) 1960-64 -.0453 (-6.88) 1950-59 - ,0260 (-4.37) < 1950 -.0451 (-4.38) 1975-79
(-
- ,3706 (-23.91) - ,2443 (-21.94) - .1710 ( - 19.52) - ,1086 (-15.18) - ,0712 (-11.12) - ,0562 (-5.05)
- ,2297
-.I222
( - 13.25)
( - 6.65)
- ,1306 (-8.57) - ,0449 (-3.75) - ,0632 (-4.63) - .0344 (-3.57) .0212 (1.10)
- .0271 (-1.67) ,0568 (4.47) - ,0079 (S4) ,0094 (.91) ,0286 (1.40)
Nore: The r-ratios are presented in parentheses.
,0149 (.46) ,0136 (.61) - ,0570 ( - 3.49) - ,0740 (-4.57) - ,1330 (-10.12) - ,0914 (-4.86)
- ,0020 ( - .23) - .0377 ( - .41) - ,0976 ( - 2.10) - ,0856 ( - 1.96) -.1803 (-5.56) - ,1997 ( - 7.21)
73
Immigration and Self-Selection
1.5.
Summary
Self-selection plays a dominant role in immigration (as it does in all other forms of turnover). There is selection in the determination of the composition of the persons who leave any given country, in terms of both observable characteristics (such as education) and unobservable characteristics (such as abilities and productivities). In addition, this nonrandom sample is then sorted across various possible host countries in a nonrandom way. Hence, the pool of immigrants in any host country is, in a sense, doubly self-selected: the pool of immigrants in the host country is composed of persons who found it profitable to leave the country of origin and who did not find it profitable to go anywhere else. This paper attempts to use the economic theory of self-selection as a guide to understanding how immigrants perform in the labor market. The assumption of wealth-maximizing behavior provides important insights into the mechanics that guide the selection process. The empirical analysis studied the role played by self-selection in the earnings of immigrants in the United States, and compared these migrants to the pool of migrants who chose to reside in other countries (Australia or Canada). The study of the various Censuses revealed that the United States, as a result of major changes in immigration policy, began to attract relatively less skilled persons in the 1970s. In a sense, the United States became less competitive in the international marketplace that determines the migration decision and the sorting of migrants across host countries.
Notes 1. A recent survey of this literature is given in Greenwood and McDowell(l986).
2. Jasso and Rosenzweig (1985) also stress this important technical point in their work. 3. A fourth case where Q, > 0 and Q, < 0 is theoretically impossible since it requires p > 1. 4. Data on international differences in income inequality are published by the World Bank (1986). These data, however, do not correspond directly to the variances that lie at the heart of the Roy model. In particular, a; and a: measure the dispersion in “opportunities” (for given X) rather than the variance in incomes across households in a given country. 5. There is a slight technical problem that must be taken into account in the derivation of this result. An increase in crl “stretches” the income distribution of the United States and will lead to a different mean wage level in the pool of migrants even if this pool is restricted to include the same persons. A simple solution to this problem is to define quality in terms of “standardized units,” or Q,/a,. The prediction in the text can then be easily derived. 6. Because the year of immigration is not precisely available in the U.S. Census, it is approximated as the midpoint of the available intervals. In the 1980 Census, y takes
74
George J. Borjas
on a value of 2.5 if the immigrant arrived in 1975-79, 8 if arrived in 1970-74, 13 if arrived in 1965-69, 18 if arrived in 1960-64, 25.5 if arrived in 1950-59, and 45 if arrived prior to 1950. In the 1970 Census, y is 2.5 if arrived in 1965-69, 8 if arrived in 1960-64, 15.3 if arrived in 1950-59, and 39.8 if arrived before 1950. The estimated y for these last two intervals are calculated using the more precise year of migration data available in the 1970 Census. 7. There is an implicit assumption in (25) that is directly responsible for this simple framework. In particular, growth rates for immigrants are independent of the year of migration 0. The model can be generalized to allow for these types of interactions. However, the estimating equations would include higher order polynomials, and the estimation of the underlying structural parameter may become quite sensitive to the very high correlation among the right-hand-side variables. 8. The construction of the data sets is described in detail in Bojas (1987). 9. The enrollment data are available in Unesco (1969, 1980). Enrollments are available for each “level” of education. The data sources also give the number of years of education associated with that “level” for each country. The means presented in table 1.2 are calculated using both these statistics. 10. It is important to note that many of these differences in quality across cohorts from a given country of origin are statistically significant at conventional levels. For some evidence on this point, see Borjas (1987). 11. Since the dependent variable in the “second-stage” regressions is a linear function of regression coefficients, the regressions are weighted to account for heteroskedasticity. For details, see Borjas (1987). 12. These statistics are available in the United Nations (1982, 44). The calculations ignore the large (and presumably) temporary flows from Ethiopia to Somalia in the late 1970s as well as the movement of guest workers to oil-producing countries in the Middle East. 13. A number of previous studies (e.g., Tandon 1978; Chiswick and Miller 1985; and Chiswick 1988) analyze the labor market performance of immigrants in Australia and Canada. These studies, however, do not study the nonrandom sorting of migrants across host countries. 14. This section is based on the excellent descriptions and summaries of immigration policies given by Boyd (1976), Keely (1979), and Keely and Elwell (1981), Kubat (1979), and Price (1979). 15. Throughout this section, the native base in each of the host countries is the entire population of native persons (regardless of ethnic or racial origin). This differs from the native baselines chosen in the previous section but makes the comparisons among host countries less arbitrary. 16. There are some differences in the calendar years bracketed by these dummy variables across the countries of destination. The brackets reported in the table are those that apply to U.S. data. The Canadian and Australian brackets are quite similar for post-1960 migrants but differ for pre- 1960 migrants. 17. The average was calculated over the two or three source countries that formed the bulk of immigration from that continent to the particular host country.
References Australian Department of Immigration and Ethnic Affairs. 1982. Australian Immigration, Consolidated Statistics no. 12. Canberra: Australian Government Publishing Service.
75
Immigration and Self-Selection
Birrell, R. 1984. A New Era in Australian Migration Policy. International Migration Review 18:65-84. Borjas, George J. 1985. Assimilation, Changes in Cohort Quality, and the Earnings of Immigrants. Journal of Labor Economics 3:463-89. . 1987. Self-Selection and the Earnings of Immigrants. American Economic Review 77531-53. Boyd, Monica. 1976. Immigration Policies and Trends: A Comparison of Canada and the United States. Demography 18533-104. Carliner, Geoffrey. 1980. Wages, Earnings, and Hours of First, Second, and Third Generation American Males. Economic Inquiry 18:87-102. Chiswick, Barry R. 1978. The Effect of Americanization on the Earnings of Foreignborn Men. Journal of Political Economy 86:897-921. . 1988. Immigration Policy, Source Countries, and Immigrant Skills: Australia, Canada, and the United States. In The Economics of Immigration, ed. Lyle Baker and Paul Miller, 163-206. Canberra: Australian Government Printing Service. Chiswick, Barry R., and Paul W. Miller. 1985. Immigrant Generation and Income in Australia. Economic Record 61 540-53. Douglas, Paul H. 1919. Is the New Immigration More Unskilled than the Old? Journal of the American Statistical Association 16:393-403. Greenwood, Michael J. 1975. Research on Internal Migration in the United States: A Survey. Journal of Economic Literature 13:397-433. Greenwood, Michael J., and John M. McDowell. 1986. The Factor Market Consequences of U.S. Immigration. Journal of Economic Literature 24: 1738-72. Heckman, James J., and Richard Robb. 1983. Using Longitudinal Data to Estimate Age, Period, and Cohort Effects in Earnings Equations. In Analyzing Longitudinal Data for Age, Period, and Cohort Effects, ed. H. Winsborough and 0. Duncan, 173-50. New York: Academic. Hicks, John R. 1932. The Theory of Wages. London: Macmillan. Jackson, J. A. 1969. Migration. Cambridge: Cambridge University Press. Jasso, Guillermina, and Mark R. Rosenzweig. 1985. How Well Do U.S. Immigrants Do? Vintage Effects, Emigration Selectivity, and the Occupational Mobility of Immigrants. University of Minnesota. Mimeo. Keely, Charles B. 1979. The United States of America. In The Politics of Migration Policies, ed. D. Kubat, 51-64. New York: Center for Migration Studies. Keely, Charles B., and Patricia J. Elwell. 1981. International Migration: Canada and the United States. In Global Trends in Migration, ed. M. Kritz, C. Keely, and S. Tomasi, 181-207. New York: Center for Migration Studies. Kubat, Daniel. 1979. Canada. In The Politics ofMigration Policies, ed. D. Kubat, 1936. New York: Center for Migration Studies. Price, Charles. 1979. Australia. In The Politics of Migration Policies, ed. D. Kubat, 3-18. New York: Center for Migration Studies. Psacharopoulos, George. 1973. Returns to Education: An International Comparison. Amsterdam: Elsevier. Roy, A. D. 1951. Some Thoughts on the Distribution of Earnings. Oxford Economic Papers 3:135-46. Schwartz, Aba. 1968. Migration and Lifespan Earnings in the U.S. Ph.D. diss., University of Chicago. Sjaastad, Larry A. 1962. The Costs and Returns of Human Migration. Journal of Political Economy 70:80-93. Tandon, B. B. 1978. Earnings Differentials among Native Born and Foreign Born Residents of Canada. International Migration Review 12906-10. Unesco. 1969. Statistical Yearbook. 1968. Paris: United Nations.
76
George J. Borjas
. 1980. Statistical Yearbook, 1978-1979. Paris: United Nations. United Nations. 1982. Demographic Indicators of Countries. New York: United Nations. U.S. Department of Commerce. Various Issues. Statistical Abstract of the United States. Washington, D.C.: U.S. Government Printing Office. U.S. Immigration and Naturalization Service. Various Issues. Statistical Yearbook of the Immigration and Naturalization Service. Washington, D.C.: U.S. Government Printing Office. World Bank. 1986. WorldDevelopment Report. New York: Oxford University Press.
2
Undocumented Mexican-born Workers in the United States: How Many, How Permanent? George J. Borjas, Richard B. Freeman, and Kevin Lang
Few issues in the area of immigration to the United States generate as much concern and confusion as the influx of illegal aliens. Estimates of the number of illegal immigrants vary widely. Some observers, noting the explosive growth of Border Patrol apprehensions of aliens to over a million a year, have suggested that the country has harbored five to ten million or more undocumented residents.’ Others, relying on 1980 Census of Population and related demographic data, put the numbers on the order of two to three million in that year2 Some think that illegal aliens are largely transient agricultural workers, slipping across the Mexican border for seasonal work. Others stress the permanence of many illegal aliens, who have sufficiently long stays in the United States to be eligible under the 1986 Immigration Reform and Control Act to attain legal resident status. In light of the difficulties in analyzing illegal immigration, one recent reviewer has written that the issue is “inaccessible to accurate measurement” with no “firm evidence” ever likely to become available (Teitelbaum 1986, 153). In this paper, we take a more positive approach, analyzing three government data sets and a small survey of illegal aliens in the San Diego area in an effort to evaluate conflicting claims about the illegal Mexican-born migrant population. Among the government data sets we analyze is, first, the 1980 U.S. Census George J. Borjas is professor of economics at the University of California, San Diego, and a research associate of the National Bureau of Economic Research. Richard B. Freeman is professor of economics at Harvard University and director of the Labor Studies program of the National Bureau of Economic Research. Kevin Lang is professor of economics at Boston University and a faculty research fellow of the National Bureau of Economic Research. The authors wish to thank Betty Smith of Statistical Resources at Public Health Service and Don Neeler, Rod Field, and Blanke Shanks of the Immigration and Naturalization Service for valuable assistance in providing data. For research assistance, they are grateful to Rachel Friedberg and Steven Rader.
77
78
G. J. BorjadR. B. FreemadK. Lang
of Population figures on the number of Mexican-born residents of the United States. As pointed out by Warren and Passel (1987), the Mexican-born population in the 1980 Census exceeds by over one million the number of legal immigrants expected in the Census on the basis of Immigration and Naturalization Service (INS) records. We build on the Census analysis by tabulating the family composition and economic characteristics of the nonnaturalized Mexican born most likely to be illegal. Second, we analyze vital statistics on the number of deaths of Mexican-born persons and of births to Mexican-born mothers in the United States. Under specified assumptions, these data provide us with independent estimates of the size and growth of the illegal Mexican population. Third, we analyze INS figures on apprehensions of illegal aliens. We relate these figures to Border Patrol expenditures and measures of economic incentives to migrate and examine the seasonality of apprehensions to make inferences about the growth and nature of border crossings. In addition, we examine the consistency between the level of apprehensions and Census/ Vital Statistics-based estimates of the size of the illegal Mexican population. Our original data consist of a survey of 289 illegal Mexican male aliens in fourteen locations around San Diego obtained during the summer of 1986. We use these data to estimate time spent in the United States and apprehensions by the Border Patrol per crossing as well as to determine the characteristics of this part of the illegal alien population. Our principal findings are as follows. 1. The number of illegal Mexican aliens in the United States in 1980 was on the order of 1.8 million. In addition to the approximately 1.1 million illegal Mexican immigrants counted in the Census, there were perhaps 600,000700,000 illegal Mexican immigrants not counted in the Census, giving a total illegal Mexican migrant population of less than two million. Births to Mexican-born women and deaths of Mexican-born persons suggest that the illegal population may have increased to 2.0-2.3 million by 1984. 2. The explosive growth of Border Patrol apprehensions is due in substantial part to increased Border Patrol activity, making the growth of apprehensions an upwardly biased estimate of the growth of illegal border crossings and a poor indicator of the growth of the illegal Mexican migrant population. Moreover, the seeming inconsistency between the 750,000-1,000,000 plus annual apprehensions and the estimate of fewer than two million illegals in 1980 appears to be due partly to short-term migrants who cross the border and generate apprehensions frequently but whose spells in the United States contribute less than a person-year to the stock of illegals. 3. The bulk of illegal Mexican aliens in the Census of Population live with their families, are engaged in work activities beyond agricultural labor, and have other characteristics suggesting that they are relatively permanent residents in the country. The earnings of these persons are considerably below those of other Mexican-born residents, but measures of work experience are similar.
79
Undocumented Mexican-born Workers in the U.S.
The difference between the relatively permanent illegal immigrants who seem to constitute the bulk of those counted in the Census and the more itinerant who cross the border frequently (and are covered in our survey in San Diego) may reflect a life-cycle change to illegal alien migration, as new young migrants come first without their families, go back often at holidays or at breaks in seasonal jobs, but later bring their families to the United States and move toward permanent residence. On the other hand, it may also reflect a substantial difference between permanent immigrant and sojourner populations.
2.1 Estimating the Size and Growth of the Illegal Mexican Population One of the more surprising facts about the 1980 Census of Population documented by Warren and Passel (1987) is that the Census counted a sizable number of illegal Mexican-born migrants (see table 2.1). In part, this results from the fact that, “for the 1980 Census, the Bureau of the Census made extra efforts to count difficult-to-enumerate groups such as undocumented aliens” and had sufficient success that the Bureau’s evaluation of population estimates turned up “a large national error of closure between the 1970 and 1980 Censuses” (U.S. Bureau of the Census, Population Growth and Distribution, 3 , 7). Our tabulations of the 1980 Census in table 2.1 show that 2.18 million Mexican-born persons (exclusive of native Americans born in Mexico) were counted, of whom 1.67 million reported that they were not citizens as of the date of the survey. According to Warren and Passel (1987), who used INS
Table 2.1
Size of the Mexican-born Population in the 1980 Census of Population Group Reporting as born in Mexico: 1. Total
2. Naturalized citizens 3. As percentage of total
4. Not a citizen 5. As percentage of total
N 2,182,900 509,400
1,673,500
%
100.0 23.3 16.1
Illegal immigrants from Mexico:
6.Estimated number 7.As percentage of total 8. As percentage of noncitizens 9.As percentage of noncitizens arriving after 1970
1,130,OOO
51.8 67.5 103.6
Sources: Lines 1-5 tabulated from the Public Use Sample of the U.S. Census of Population, the A Sample (5%). Line 6 from Warren and Passel (1987).
80
G . J. Borjas/R. B. FreemadK. Lang
figures on legal alien entry to the United States (modified in various ways) to estimate the number of legal migrants and, as a residual, the number of illegals, the 1980 Census counted 1.13 million illegal Mexican-born aliens. This is more than half the total reported (line 1 ) and two-thirds of those who were not citizens (line 4),implying that the Census data can be used to make inferences about the characteristics of a large number of illegal Mexican-born aliens. Going a step further, note that the figure of 1.09 million Mexican-born noncitizens in the Census who came after 1970 is approximately equal to Warren and Passel’s estimated number of illegal aliens. Given that relatively few illegal Mexican aliens are likely to have come to the United States prior to 1970 and that many legal immigrants are likely to have become citizens, one can reasonably treat Mexican-born noncitizens who came after 1970 as a population dominated by likely illegals, as we do in section 2.3. These conclusions are, of course, affected by the fact that the counts provided by the 1980 U.S. Census of Mexican nationals and of naturalized Mexicans are measured with error. It is well known, for instance, that naturalization rates calculated from individual responses to the Census questionnaire greatly overstate the naturalization rates recorded by the official INS documents. The careful study of the 1980 Census data by Warren and Passel, however, attempts to correct the counts for errors in misreporting both of country of birth and of citizenship status. Using these corrected population counts in our analysis does not alter the qualitative nature of any of our conclusions and does not greatly affect the order of magnitude of the statistics reported in table 2.1. How many illegal Mexican migrants might be missed in the Census count-one million, two million, ten million? To provide an answer to this important question, we make use of two pieces of data from the Vital Statistics of the United States: the number of deaths of persons of Mexican birth and numbers of births to Mexican-born women. Assuming, as seems reasonable, that mortality and birth data are more complete than the Census count of the population, we expect to find more deaths/ births for the Mexican born than the counted population could plausibly generate.3 Given assumptions about true death and birth rates, the “excess” deaths or births will yield estimates of Census undercounts and thus of the true population. Algebraically, the structure of our analysis can be most simply seen in the following accounting equation: (1)
R, = r,(POP,) +r; (HPOP,),
where R, = number of events, deaths or births, in Vital Statistics, r, = the true rate of occurrence of events to the measured population, I.: = the rate of Occurrence of events to the hidden population, POPt = measured population, HPOP, = hidden population, and i indexes an agehex group. Then HPOP, = (R, - r,POPt)/r: provides an estimate of the undocumented population of age i .
81
Undocumented Mexican-born Workers in the U. S .
The accuracy of estimates of the hidden or undocumented population based on (1) depends on two factors: the extent to which funeral directors and hospitals accurately record country of origin on death and birth forms and the accuracy of the postulated “true” death or birth rates for the measured population and the hidden population. While there may be some tendency for friends and relatives of illegal aliens to disguise country of origin, the head of Registration of Methods of the Public Health Service informed us that in his opinion these records are no less accurate for illegal aliens than for other groups, leading us to discount this potential source of The problem, then, is to estimate “true” birth or death rates for the undocumented population. Our approach is to assume that the rates for the undocumented are the same as those for the population as a whole (death rates) or for the documented migrant population (birth rates). This approach yields the following formula for estimating the hidden population:
(2)
HPOP; = R , / rj - POP,.
If, as seems reasonable, Mexican-born immigrants have higher death rates than native Americans (rr > r j ) ,estimates of the uncounted population based on national death rates will provide an upper bound to the population. Similarly, if undocumented Mexican-born women have higher birth rates than documented immigrants, estimates of the hidden population based on national birth rates will also provide an upper bound to the number of undocumented Mexican-born female migrants .5 2.1.1 Mortality of the Mexican Born Since 1979, Viral Statistics has published data on the number of deaths of the Mexican-born persons in the United States. Beginning in 1984, data are also available by age, allowing us to use equation (1) above to estimate the likely number of such persons by age. Columns 1-2 of table 2.2 record the basic data for this analysis: the number of deaths to Mexican-born persons in the United States by age (col. 1) and the 1984 mortality rates for all Americans by age (col. 2). Assuming that the U.S. mortality rate represents the true death rate for the Mexican born, we obtain the estimated 1984 population of Mexican-born persons by age in column 3. Summing down the column gives an estimated total population of 2.97 million persons. As we expect the mortality of the Mexican born to be higher than that of native Americans, this is likely to be an upper bound to the true number. To use our 1984 estimates to obtain estimates for 1980 when the Census was conducted, we must adjust them for potential changes in the Mexicanborn population from 1980 to 1984. According to Vital Statistics, the number of deaths of Mexican-born persons rose from 13,180 in 1980 to 14,050 in 1984, suggesting that the population may have been 6.6%higher in 1984 than in 1980. By this calculation, the Mexican-born population in 1980 was about 2.8 million persons, which implies that the Census missed 600,000-700,000
82
G . J. Borjask. B. Freeman/K. Lang
Table 2.2
Mortality and Undocumented Mexican-born Population Estimated from Deaths of the Mexican Born
Age Group
1984 Deaths (1)
1984 U.S. Mortality Rate/ 100,000 (2)
<1 1-4 5-14 15-24 25-34 3 5 4 45-54 5544 65-74 75-84 85
+
16 39 70 922 1,093 742 819 1,252 2,308 4,150 2,627
1,085.6 51.9 26.7 96.8 121.1 204.8 521.1 1,287.8 2,848.1 6,399.3 15,233.6
Total
14,038
1984 Mexican-born Population (3)
1980 Population (4)
1,500 75,100 262,000 952,000 902,600 362,000 157,000 97,200 8 1,000 64,900 17,300
7,800 52,300 271,000 516,000 551,800 310,000 186,000 126,000 94,900 53,600 13,900
2,972,600
2,183,000
“Uncounted” Mexican Born ( 5 ) - 6,300
22,800 -9,000 436,000 350,800 52,000 - 29,000 - 28,800 - 13,900 11,300 3,400 789,600 ~~
Sources: Columns I and 2 from Vital Statistics of the United States. Column 3 = (column 1)/ (column 2). Column 4 tabulated from 1980 U.S. Census of Population tapes. Column 5 = column 3 - column 4.
persons-roughly 25% of the Mexican-born population. Adding the 600,000-700,000 undocumented persons to Warren and Passel’s estimated 1.1 million in the Census yields a total illegal Mexican population in 1980 on the order of 1.8 million persons. As a check on the plausibility of our calculations, we record in column 4 of table 2.2 the number of Mexican-born persons by age in the 1980 Census and in column 5 the difference between these numbers and the 1984 numbers implied by the mortality data. The calculations show the biggest divergence to be among 15- to 24-year-olds and 25- to 34-year-olds, which seems plausible in terms of the likely age distribution of transient illegal aliens. The implication is that the mortality data are, indeed, giving us a reasonable handle on the order of magnitude of the missing population. 2.1.2 Births The number of births to Mexican-born women provides another source of information on the potential size of the population not counted in the Census. To estimate the size of the undocumented Mexican-born female population, we return to equation (1): as our measure of the number of events R, we take the number of births reported for Mexican-born women by age from Vital Statistics (col. 1 of table 2.3); as our measure of the true event rate, we take birth rates for Mexican-born women estimated by Bachu and O’Connell (1984) from the April 1983 Current Population Survey (col. 2 of table 2.3).6 We then divide the number of births by the birth rates to estimate the Mexican immigrant female population in 1980 (col. 3). For the group aged 18-39,
83
Undocumented Mexican-born Workers in the U.S. Births and Undocumented Mexican-bornFemale Population Estimated from Births to Mexican-born Women
Table 2.3
(1) 1980 Births to
Age 18-24 25-29 30-34 35-39
Total
Mexican-born Women 52,464 3 1,730 17,322 1,276
(2) 1983
Birth Rates (3) Estimated (4) 1980 Census (5) Hidden Female Population Population Population per 1,OOO 173 144 110 78
303,300 220,300 157,500 93,300
174,700 133,500 112,100 84,400
128,600 86,800 45,400 8,900
774,400
504,700
269,700
Sources: Column 1 from Viral Sratistics of the United Stares. Column 2 provided by Martin O’Connell from April 1983 CPS tapes. Column 3 = (column I)/(column 2). Column 4 tabulated from 1980 Census of Population tapes. Column 5 = column 3 - column 4.
which accounts for most births, this sums to 774,000. The difference between these figures and those in the Census given in column 4 is our estimate of the unrecorded Mexican-born female population in the child-bearing years: 269,000. If we take 700,000 as our estimate of uncounted Mexicans from mortality statistics and assume no illegal alien women outside the 18-to-39 age bracket, then women would constitute slightly less than 40% of the uncounted illegal Mexican migrant population. As the Census shows that 46% of Mexican-born nonnaturalized immigrants who came after 1970 were women, the implication is that the undocumented population contains more males than females, but not by as much as one would think on the basis of apprehensions of illegals, some 80%-85% of whom are males. One possible reason may be that Mexican women are more permanent migrants than men so that they contribute more to the total stock of persons in the country than to apprehensions at the border.’ All told, both the death and the birth figures support the growing consensus among demographers (see Passel 1986) that the number of illegal Mexican immigrants is on the order of two to three million. Finally, we also note that the estimated size of the illegal alien population revealed by the demographic data is roughly consistent with the number of persons who applied for legalization under the provisions of the Immigration Reform and Control Act of 1986 (IRCA). This legislation provides amnesty to illegal aliens under one of two provisions. First, amnesty is granted to aliens who have been present in the country illegally and continuously since before 1 January 1982 and who applied for amnesty in the year ending on 4 May 1988. Second, amnesty is granted to agricultural workers through the Special Agricultural Worker (SAW) program if the illegal alien worked in perishable crop agriculture in the United States for at least ninety days in the year ending on 1 May 1986 (Immigration and Naturalization Service 1989). Approximately 1.2 million Mexican-born persons applied for amnesty under the regular program, and an additional 1.1 million Mexicans applied under
84
G . J. Borjask. B. FreemadK. Lang
the SAW program. Because of widespread fraud in applications to the SAW program (“A Million Late Arrivals” 1988), the number of undocumented Mexicans who qualified for amnesty was probably under two million, a number roughly of the same order of magnitude as the estimates reported in this paper. 2.1.3 Trends in Mexican Immigration In addition to providing estimates of the numbers of illegal aliens in the United States, the Vital Statistics data can be used to estimate the growth rates of the underlying population. For the brief period in which we have deaths for the Mexican-born population, there is no dramatic trend upward, as the following numbers of deaths indicate: 1979: 12,288; 1980: 13,180; 1981: 13,135; 1982: 13,078; 1983: 13,066; 1984: 14,050. Taking the end periods, we have a rate of growth of 14% over the five years, or 2.7% per year compounded. For the longer period over which we have birth rates, the figures are more dramatic: in 1980, there were 117,126 births to Mexican-born women, compared to 48,796 in 1970. This implies a near two-and-a-half-fold increase in the Mexican-born female population, given constant birth rates. However, comparisons of the number of Mexican-born women counted in the 1970 and 1980 Censuses show that this is entirely consistent with the measured growth of the number of Mexican-born women-an approximate two-and-a-half-fold increase from 419,754 in 1970 to 1,038,700 in 1980 and a larger increase in the number in prime child-bearing years.8 Overall, there were 8 17,000 Mexican-born persons in the 1970 Census, compared to the 2,182,000 in the 1980 Census. If we take 700,000 as the number of Mexican immigrants missing from the Census, the true population of Mexican-born immigrants increased by some 260% over the period to nearly 2.9 million. This, in turn, implies a net immigration of 2.1 million persons to the United States in the 1970s-210,000 persons per year-of whom roughly three-quarters were illegal. This is an enormous increase in the Mexican-born population due to illegal immigration, but far below the growth in the number of apprehensions of Mexicans at the border that underlies much alarmist concern. We turn next to the apprehension data.
2.2 The Number and Growth of Apprehensions Apprehensions of illegal Mexican immigrants have increased at truly extraordinary rates since the late 1960s. In 1967, 100,OOO persons were apprehended for trying to cross the border; in 1986, nearly 1.7 million persons were apprehended-a seventeenfold increase that dwarfs our estimated growth of the Mexican-born pop~lation.~ The level of apprehensions as well as the growth also seems exceptionally large relative to the estimated size of the population: from 1970 to 1979 there were some 7.5 million apprehensions (three-quarters of a million per year), while from 1980 to 1986 there were 6.2
85
Undocumented Mexican-bom Workers in the U.S.
million apprehensions (nearly a million per year). If the number of persons who successfully crossed the border was, say, four times as large as the number of apprehensions, and if those who crossed successfully averaged a twoto three-year stay in the United States, these figures would indicate an illegal Mexican alien population of about six to nine million persons in 1980 and perhaps eight to twelve million in the mid- 1980s.I0 Do the apprehension data really imply such a large and rapidly growing illegal alien population? What caused the explosion in apprehensions? Is there a way to interpret apprehensions that would make this consistent with Censusand Vital Statistics-based estimates of the size and growth of the illegal immigrant population, or are the different data incommensurate? We suggest in this section that the growth of the number of illegal aliens trying to enter the United States is much less than indicated by the apprehension figures because a sizable part of the increase in apprehensions is due to the increased efficacy of the Border Patrol. We further suggest, on the basis of estimates of durations of time in the United States and of the ratio of apprehensions to successful crossings for a sample of relatively transient illegal migrants, that a population of illegals of the magnitude suggested by Census and Vital Statistics data together with a modest population of Mexicans who fail to cross the border could have generated the bulk of the apprehensions. 2.2.1 The Effect of Border Patrol Activity on Apprehensions The first factor that suggests that increased Border Patrol activity is a substantial determinant of the growth of apprehensions is the sharp increase in real expenditure on the Border Patrol in the 1970s and 1980s. In 1967, the Border Patrol spent twenty million dollars; in 1986, they spent forty-eight million in 1967 dollars-a 140% increase. If there were no change in the productivity of a dollar of resources in apprehending illegal border crossers, this increase in expenditures could by itself explain over half the growth of apprehensions. Indeed, as figure 2.1 shows, from the late 1970s to 1986, when so much concern was expressed about the explosion of apprehensions, apprehensions per dollar of Border Patrol expenditures rose only modestly, implying that the trend in apprehensions could have resulted largely from increased Border Patrol resources. From 1967 to 1976, on the other hand, the number of apprehensions per real dollar expenditure on the Border Patrol increased greatly. There are two possible explanations for the sharp pre-1976 increase in apprehensions shown in figure 2.1. The first is that the number of attempted illegal border crossings rose, presumably in response to economic incentives to migrate illegally to the United States in the wake of the termination of the Bracero Program. Increased real hourly earnings in the United States in the late 1960s and early 1970s may have made working in the United States more attractive, while the growth of real earnings in Mexico may have eased possible credit constraints in risking an illegal trip to the United States. On the
86
G. J. Borjash. B. FreernanlK. Lang
35 30 25 20 15 10 -
5
68
70
72
74
76
78
80
82
84
I 3
Fig. 2.1 Ratio of apprehensions to real Border Patrol expenditure (in constant dollars)
other hand, the number of legal Mexican immigrants exempt for family reasons from the INS quota limits fell in the period, which is the opposite of what one would expect given greater economic incentive and capital to enter the United States. Perhaps more important in terms of long-term immigration, the Census of Population data seem inconsistent with an explosion of permanent illegal migration in the period. According to the Census, the number of Mexican-born persons who immigrated in 1970-74 is just about twice the number who immigrated during 1965-69, whereas the number of apprehensions in 1970-74 is four times the number in 1965-69. Similarly, the number of Mexican-born persons who arrived during 1975-80 is about 40% higher than the number who arrived during 1970-74. The apprehension figures for 1975-80 are more than twice those for 1970-74.” Given that a higher proportion of individuals who immigrated in the earlier period are likely to have returned to Mexico, these figures raise serious doubts about interpreting the increased number of apprehensions as reflecting economically induced increases in the number of long-term illegal immigrants. The second possible interpretation of the 1967-76 spurt in apprehensions is that it represents a “learning curve” for the Border Patrol following the end of the Bracero Program. There is scattered evidence that the effectiveness of the Border Patrol increased over the period. Cornelius (1977) reports that the use of “coyotes” (smugglers of illegal aliens) increased over time in his sample, which suggests greater difficulty in crossing over time. And the Border Patrol introduced more capital intensive and modem technologies to detect illegal
87
Undocumented Mexican-born Workers in the U.S.
aliens, ranging from helicopters to sophisticated electronic detection devices planted along the border. As both increased border crossings and increased Border Patrol resources and effectiveness are likely to have contributed to the observed growth of apprehensions, a quantitative analysis is needed to evaluate the potential magnitude of each. Accordingly, we have regressed the log of the number of apprehensions on the log of the Border Patrol budget measured in real dollars; the log of average hourly earnings in the United States and the log of GDP per capita in Mexico, as indicators of the relative incentive to come to the United States; and a trend term. Our analysis covers the period 1967-84, when apprehensions skyrocketed following the end of the Bracero Program. Table 2.4 presents the regression results. As a base for judging the effect of the Border Patrol budget and other factors on apprehensions, column 1 records a regression estimate of the annual compound growth in apprehensions-approximately 12% per year. Column 2 includes the effect of real Border Patrol expenditure. The estimated elasticity of apprehensions with respect to Border Patrol expenditure exceeds two, and the annual growth rate in apprehensions falls to 5 % , indicating that over half the observed increase in apprehensions can be attributed to the growth of Border Patrol spending. Column 3 gives the regression coefficients and standard errors for the estimated effect of the log of real Border Patrol expenditures, average hourly earnings in the United States and GDP per capita in Mexico and the trend variable. Here, the estimated effect of Border Patrol resources on apprehensions has a near unit elasticity, and economic factors also appear to affect apprehensions, with U.S. Table 2.4
Determinants of Apprehensionsof Illegal Mexican Aliens (time-series estimates, 1967-84) (1)
Budget
(2)
(3)
2.22 (7.1)
.94 (2.3)
(4)
5.17 (3.5)
.99 (2.3) .29 (.4) 3.44 (1.3)
.66 (1.5) .08 (3.3) - 4.07 (1.3)
.48 (1.0) .07 (2.0) -5.22 (1.2)
Budget ( - 1) U.S. wage Mexican GDP per capita Time trend Constant
.12 (8.8) 12.05 (78.7)
.05
(4.1) - 10.14 (3.3)
(5)
.04 (4.3) -9.48 (4.0)
Sources: Apprehensions data from INS Budget, Central Border Patrol Office. U.S. nonagricultural average hourly wages from the ILO Yearbook (Geneva: ILO). Mexican GDP, population, and deflator from Intermtional Financial Statistics. U.S.CIP from Economic Report of the President (Washington, D.C.: U.S. Government Printing Office).
88
G. J. Borjas/R. B. FreemadK. Lang
earnings increasing apprehensions and Mexican GDP per capita also increasing them, perhaps as a result of the greater ability of poor Mexicans to raise capital for migration to the United States. As real wages in the United States and per capita income in Mexico declined in the late 1970s and early 1980s, we can treat the coefficient on trend as our estimate of the increase in illegal border crossings independent of the level of Border Patrol activity and with economic factors held fixed. It implies that one-quarter of the observed growth of apprehensions can be attributed to the increased border Patrol spending. Columns 4 and 5 include lagged Border Patrol expenditure in the regressions of columns 2 and 3 to capture potential learning effects. The results show a continued effect for Border Patrol spending, with residual trend effects ranging from one-third to over half the .12 coefficient in column 1. While we are leery of crude time-series regressions with just seventeen observations, it seems reasonable to conclude that Border Patrol activity has influenced the trend in apprehensions and thus that the trend uncorrected for Border Patrol activity exaggerates the growth of the flow of illegal Mexican immigrants. Quantitatively, the regressions in table 2.4 indicate that something on the order of half the increase in apprehensions is due to increased Border Patrol expenditure and thus that the growth of illegal crossings was perhaps half as great as the growth of apprehensions. 2.2.2 Relating Apprehensions to the Stock of Illegal Immigrants To analyze the relation between the level of apprehensions and the size of the stock of illegal immigrants, we decompose apprehensions into three categories: those that result from the apprehensions of new illegal aliens who eventually cross the border for a first trip to the United States ( P , ) ;those generated by experienced illegal aliens who make repeated crossings and are living in the United States (P,); and those generated by persons who fail to cross successfully (P,). The annual number of apprehensions per successful new crosser we denote as a , ; the number of apprehensions per successful repeat crosser we denote as a,; the number of apprehensions per failed crosser we denote as a,. Then total apprehensions ( A ) will be (3)
A = a , ( P , ) + a,(P,)
+ a,(P,).
This equation shows that apprehensions depend not only on the number of successful border crossers resident in the United States and the number of times they are apprehended in a year but also on the number of unsuccessful crossers and their rate of apprehension. To obtain information on the number of apprehensions per successful border crosser in the United States, we use the survey described in section 2.3. In that survey, we asked the illegal aliens in the United States the number of times they were apprehended and the number of visits they made to the United States.Iz The ninety-one persons in the sample who were on their first visit reported that they had been apprehended by the Border Patrol at least 95
89
Undocumented Mexican-born Workers in the U. S .
times, giving a ratio of apprehensions to successes of slightly more than one, which we will use to estimate a, in (3). The 132 individuals who were on a second or later trip had been apprehended at least 242 times in the course of 356 reported trips.I3Assuming that they were apprehended an average of once on their first trip, the implication is that these illegal migrants were apprehended about 110 times in the course of 224 second or later trips. The ratio of “successes” to apprehensions thus seems to be about two to one for “experienced” border crossers. While estimates of successful border crossings to apprehensions range all over the ballpark, the one-to-one and two-to-one ratios in our data are in line with the views of some informed observers. Alan Eliason, the chief Border Patrol agent in San Diego, estimates that “we’re locating, at best, about half the flow of illegal entrants” (Eliason 1986). “Official” estimates reported in the newspapers have been in the range of two to three to one.I4 The ratio of successes per apprehension for repeat crossers does not, however, give us the u2 parameter in equation (3). This is because repeat crossers may have made more than one trip to the United States in a year, generating more than one apprehension a year. To estimate the number of annual apprehensions generated by repeat crossers residing in the United States, we use the following steady-state condition:
(4)
u2 = (apprehensions per successful trip)/
(average length of trip measured as a fraction of year).
According to (4), shorter trips generate more apprehensions per person in the United States because they imply that each person makes multiple trips per year. Put differently, we must “blow up” apprehensions per trip to obtain u2because more than one successful trip is required to make up a full “personyear” in the United States. In our sample of illegal aliens, the average completed duration of the most recent trip of persons who were on at least their second trip was six months-.5 years-which, together with the estimated number of apprehensions per trip, yields a value of a, of about one. This estimate, however, may be biased. First, there is sample bias in our survey group that is due to the greater likelihood of our reaching those with longer spells. Put differently, failure to interview persons who made successful crossings but are now in Mexico biases our results. Data from CENIET’s Encuesta Nacional de Emigraci6n a la Frontera Norte del Pais y a 10s Estados Unidos suggests, however, that our estimates may not be that far off. The CENIET figures show that, of the nearly one million persons considered by their families to be living in Mexico who had migrated to the United States in the 1978-79 period, half were in Mexico at the time of the study (see DiezCanedo in this volume). This is consistent with the notion that these migrants average about half their time in each country. Another problem with our sample is that it may not be representativeof persons uncounted in the Census or of the sojourner population. Still, for want of better data, we shall use the
90
G . J. BorjadR. B. FreemadK. Lang
estimate that both new and repeated border crossers in the United States are likely to have generated one apprehension. Finally, while we have no information on how many times unsuccessful crossers were apprehended, they had to have at least one apprehension so that a, > 1. Given our estimate of one apprehension generated per successful crosser, what is the likely size of the relevant populations that made a successful crossing (PI P J in a year? As a crude estimate of the number who may cross frequently, we will take persons who were sufficiently transient to have been missed by the Census and those in the Census who have come after 1970 and are without their families. Estimates in the previous section suggest that there were 600,000-700,000 persons in the first category, while calculations given in the next section suggest that there were about 275,000 persons in the Census who came after 1970 and were living without their immediate families, giving a 1980 population of prospective new and repeated border crossers close to one million. Given our estimates of a, and a2,this is large enough to have generated apprehensions of the same magnitude. As additional apprehensions were undoubtedly generated by persons who failed to cross the border, we conclude that the observed number of apprehensions is more consistent with the estimated population of illegal Mexican migrants in the United States than first appears to be the case.
+
2.2.3 Seasonality of Apprehensions The notion that there is a significant population of Mexican migrants who cross the border often, generating apprehensions, can be checked further by examining the seasonal pattern of migration. While permanent illegal immigration to the United States might be seasonal, it seems more plausible that seasonality in apprehensions would reflect sojourner migration, with the same or different persons crossing regularly after short trips that do not add greatly to the stock of illegals in the United States at a moment in time.I5 To estimate the seasonality of apprehensions, we obtained monthly apprehension data from the INS. For each month, we took the ratio of apprehensions in that month to a twelve-month moving average centered on that month and averaged the fraction over the sample period. Figure 2.2 gives the monthly seasonality factors for the period 1957-64, when the Bracero Program was operative, and for 1967-85, when apprehensions skyrocketed.I6 The figure shows substantial seasonality in both periods, but of quite a different kind. In the post-Bracero period, apprehensions peak in March and bottom out in December, consistent with the view that many immigrants return to Mexico for extended Christmadwinter vacations. In contrast, in the earlier period, February has the lowest apprehension level, while there is a strong peak from July through October, which would appear to indicate workers heading to the U.S. harvests. While a more detailed analysis of the likely causes of seasonality in apprehensions is needed, the marked seasonality from 1967 to 1985 is consistent with the argument that illegal migrants with short
91
Undocumented Mexican-born Workers in the U.S.
-I
1.300
Jan
Peb
Mar
Apr
May
Jun
1-
Jul
Aug
1958-1964
Sep
Oct
Nov
Dec
_____ 1967-19851
Fig. 2.2 Seasonality of alien apprehensions
stays in the United States may have generated a large proportion of the apprehensions. Had we found no seasonal pattern, we would feel less comfortable with our argument that large numbers of apprehensions may be generated by short-term migrants.
2.3 Characteristics of Illegal Mexican Aliens In this section, we turn from counts of illegal Mexican aliens to the characteristics of those Mexican-born persons in the 1980 Census of Population who are likely to be illegal aliens and the characteristics of the illegal male Mexican-born migrants in our survey in the San Diego area. 2.3.1 Likely Illegal Immigrants in the Census As the Census does not contain direct information on whether a Mexicanborn migrant entered the country legally or illegally, we exploit the fact, noted earlier, that the number of Mexican-born noncitizens in the Census who came after 1970 was approximately equal in size to Warren and Passel’s (1987) estimate of illegal Mexican aliens in the Census. We define this group of persons as “likely illegal aliens” and compare them to naturalized Mexican-born persons for the purpose of making inferences about the correlates of illegal alien
92
G . J. Borjask. B. FreemadK. Lang
status.” Errors of classification in this (or any similar) scheme are likely to bias downward estimates differences between the groups. Tables 2.5-2.7 present our analysis of the likely illegal alien population in the Census. To begin with, table 2.5 records information on the distribution of the Mexican born by family status. Column 1 gives the percentage distribution of Mexican-born citizens; column 2 gives the distribution for noncitizens; column 3 focuses on noncitizens arriving after 1970 (our likely illegal migrant group); while column 4 records distributions for male noncitizens who arrived after 1970. Family status in these data is divided into householders (primarily adult males) and persons (primarily females and children), with subdivisions to reflect whether the individual is living with his immediate family (spouse, children, parents), other relatives (primarily siblings), or unrelated persons. Rows 8 and 9 at the bottom of the table show the proportion of the entire group living with close relatives and the proportion in all other categories. What stands out in the data is the large number of Mexican-born noncitizens who live with their families: 77% of all noncitizens, 73% of the recent immigrants, and even 65% of recently arrived male noncitizens live with their immediate families. Given that the bulk of these populations consist of illegals, the implication is that the majority of illegal Mexican-born residents in the 1980 Census reside here with their families. To the extent that immigrants living with their families should be viewed as relatively permanent migrants, table 2.5 suggests that most illegal Mexican migrants counted by the U.S. Census are permanent migrants. Table 2.5
Percentage Distribution of Mexican Born by Family Status Noncitizens Arrived
Family Status
All Citizens
Householders living with: 1. Closely related persons 36.2 2.0 2. “Other relatives” 1.1 3. Unrelated persons 8.0 4. Single householders Persons living with: 5. Closely related house45.1 holder 6. Other relative house4.9 holder 7. Unrelated householder 2.7 Total householders and persons living with: 8. Close relatives 81.3 9. All other 18.7 Number (OOOs)
509.0
Before 1970
After 1970
Male Noncitizens Arrived after 1970
26.6 2.1 1.2 4.1
20.3 2.1 1.3 3.1
32.9 3.1 2.2 4.9
49.9
52.2
31.6
10.5 5.6
13.7 7.3
15.6 9.6
76.5 23.5
72.5 27.5
64.5 35.5
1,674.0
1,090.0
593.0
Sources: Tabulated from the U.S.Bureau of the Census.
Undocumented Mexican-bom Workers in the U.S.
93
Table 2.6
Means of Socioeconomic Variables Noncitizens Citizens
Likely Legal
Likely Illegal
Variable
Male
Female
Male
Female
Male
Education Age LFP FXlIl Maid LWKS LWeekly Size
7.41 37.64 .77 .ll
7.02 40.62 .80 .15 3.12 5.35 4.83
6.88 42.68 .41 .07 .02 3.46 4.83 4.72
5.92 23.77 .88 .16
3.70 5.27 4.51
7.17 39.40 .43 .04 .01 3.41 4.88 4.51
.16 .18 .14 .ll .I7 .24
.I4 .18 .14 .10 .17 .27
.oo .oo
.oo .oo
.41 .22 .23 .22
.37 .23 .23 .17
.oo .oo .oo .oo
2,623
2,471
2,886
2,945
5,933
Year of migration: 1975-80 1910-74 1965-69 1960-64 1950-59 < 1950 Sample size
.oo
.oo
Female
.oo
3.65 5.04 5.33
5.51 23..96 .46 .06 .02 3.39 4.76 5.71
.59 .41
.56
.44
.oo .oo .oo .oo 4,971
Nore: LFP = 1 if participating in the labor force; 0 otherwise. Farm = 1 if employed in the agricultural sector, 0 otherwise. Maid = 1 if employed in the personal services industry, 0 otherwise. LWKS = log of weeks worked in 1979. LWeekly = log of weekly earnings in 1979.
Size
=
household size. The labor force variables are calculated among persons aged 16 or older.
Table 2.7
Determinants of Labor Market Outcomes Dependent Variable
Variable Likely illegal Citizen Education Age Age squared R2
LIT ,026 (3.09) - ,003 ( - .50) ,0003 (.38) ,047 (28.78) - ,0006 ( - 27.88)
.09
LWKS
LWeekly
- ,014
- ,179
( - .93)
- ,003 (-.16) .004
(3.09) ,049 (15.71) - ,0006 ( - 14.03)
.04
(7.90) - ,073 ( - 2.89) ,029 ( 13.40) .075 ( 15 SO) - .0008 (-13.29)
.09
Farm
- ,009 ( - .83) - ,005
( - .43)
- ,017 ( - 17.14)
- ,015 (-6.48) ,0002 (6.66) .05
Note: The r-ratios are presented in parenthesis. The regressions are restricted to men aged 16-64.
94
G. J. Borjask. B. FreemadK. Lang
Table 2.6gives the means of selected socioeconomic characteristics for the likely illegal alien group, for citizens, and for noncitizen Mexican-born persons who are likely to be legal. The differences between the groups shown in the table suggest that our classification succeeds in capturing important aspects of the likely illegal alien group. For example, the average age of a likely illegal alien man is 23.8 years, while the average age of other Mexican-born groups (citizens and noncitizens who arrived before 1970)is between 38 and 41 years. Similarly, the mean years of schooling of a likely illegal alien man is between 1 and 1.5 years below that of other Mexican-born men. The means in the table also reveal differences in labor force participation rates (higher for likely illegals) and in earnings (lower for likely illegals). These differences notwithstanding, table 2.6is also remarkable for what it does not show. It does not show the likely illegal population to be primarily male: only 54.4% of the group are men, compared to 50.4% of the remainder of the Mexican-born population. It does not show the likely illegal alien population to be heavily concentrated in agriculture; only 16% of men are employed in agriculture, as contrasted to 11%-15% of other Mexican-born men. What about the labor market experience of likely illegals? Table 2.7examines the effect of likely illegal status on four aspects of market performance for men aged 16-64: labor force participation (LFP); In weeks worked over the year (LWKS); In weekly earnings (LWKLY); and the probability of agricultural employment (FARM). It shows that, with other variables held fixed, the likely illegals have somewhat higher rates of participation in the work force than other Mexican-born persons, which makes sense if they migrate to obtain work and leave when they are out of work, and have much lower earnings than other Mexican-born persons, which also makes sense given their likely lower level of skill and lack of recourse to legal protections. The equations in table 2.7 constrain the coefficients of various socioeconomic characteristics to be the same for likely illegal aliens and other Mexican-born groups. This constraint is implausible given that illegals are likely to have less incentive to invest in human capital than legal migrants because of the likely shorter periods of time that they spend in the United States and are likely to have education that is less suitable to the job market. Accordingly, we estimated earnings equations separately for likely illegals and likely legal migrants, obtaining substantial differences in the effect of age (A)and education (E)and citizen status ( C ) on log weekly earnings, as the following regressions show for likely legals: LWKLY
=
.034*E + .105*A - .001*A2- .063*C, (11.6) (15.0) (13.3) (2.4)
R2 = .09
and for likely illegals: LWKLY = .023*E
(7.6)
+ .061*A - .001*A2, (8.1)
(7.0)
R2 = .04,
95
Undocumented Mexican-born Workers in the U.S.
where both regressions include a constant term, and the t-statistics are in parentheses. The finding that the earnings of legal migrants are more responsive to traditional human capital variables than the earnings of illegal migrants is consistent with evidence provided by Chiswick (1986a, 1986b) using a survey of undocumented workers in Chicago. The similarity between our results and those for the undocumented workers in Chicago suggests that problems of misclassification are not overly serious for our sample. Moreover, it suggests that Chiswick’s results for Chicago generalize to the broader population of illegal Mexican immigrants. 2.3.2
Results from Our Survey
As noted earlier, we conducted a small survey (289 observations) of illegal Mexican male migrants in the San Diego area; the participants were chosen to cast light on aliens unlikely to be counted in the Census.ls Interviews took place at downtown “shape-ups,” in agricultural “residences,” or wherever community contacts led us to illegal Mexican workers. The sample therefore captures the least stable and lowest-paid segment of the illegal immigrant community. While this sample of male illegal immigrants is by no means random, it is still instructive to look at their characteristics. As can be seen in table 2.8, the men in the sample are lower paid and more Sample Characteristics in Survey of Illegal Aliens
Table 2.8
Education Age Farm LWeekly Year of first trip (%): 1986 1985 1984 1983 1982 1981 1980 Before 1980
6.31 28.36 0.26 4.90 25
26 17
1
10 3 6 6
Family status: (number): With wives living in U.S. With children but without wives
39 3
Potential immigrant status: Intend to stay in U.S. permanently Will remain in U.S. indefinitely Will return to Mexico and not come back to U.S.
31 87 50
N Sources: 1986 Summer Survey of Illegal Aliens in the San Diego Area.
289
96
G . J. Borjask. B. FreemadK. Lang
Table 2.9
Determinants of Log-weeklyWages (survey sample) Education Age
- ,002 (1.2) ,045 (3.1)
Age squared
- ,005
R2
(1.3) .12
Note: The t-ratios are presented in parentheses.
heavily concentrated in agriculture than the likely illegal group in the Census (compare table 2.6). Adjusting for the fact that the survey was conducted six years after the Census, the survey sample arrived more recently than the illegal immigrants in the Census, implying that the sample does indeed reach the group we intended. Even so, a substantial number of persons in the sample appear to be on their way to becoming permanent immigrants: almost one in seven had wives living in the United States, a figure below that for likely illegals in the Census but still nonnegligible; thirty-seven said they intended to remain in the United States permanently, and eighty-seven intended to remain indefinitely, at least if they find employment. Only fifty said that they intended to return to Mexico and not come back to the United States. Finally, table 2.9 gives the results of the estimation of the weekly earnings equation for our sample. Relative to the estimates for the likely illegals in the Census, the effects of age on earnings are attenuated, and the effect of education is negative and statistically insignificant, further suggesting that human capital variables do not do much for the pay of illegal aliens. Whether these results reflect the differential experience of relatively recent and temporary immigrants or our sample design, which includes largely low-wage workers, is an open question. They do, however, confirm that we have indeed identified a very different set of illegal immigrants than those in the Census. 2.4
Conclusions
Given the difficulties in trying to measure any illegal activity, conclusions about the size and socioeconomic characteristics of the illegal Mexican population in the United States must inevitably be subject to numerous caveats. Our response to the problem of “inaccessibility” of the population of illegals has been to examine several different data sets, to search for consistencies among them that would allow for firm conclusions, and to make “strong” assumptions to obtain bounds on critical statistics. We found that the bulk of the data are consistent with the existence of an illegal Mexican-born population on the order of 1.8 million in the 1980s, that this population has grown rapidly over the decade but at a rate far below the growth of apprehensions, that a large portion of the illegal Mexican migrant population consists of “per-
97
Undocumented Mexican-born Workers in the U.S.
manent” migrants, but that border crossings by the transient part of the illegal migrant population may underlie a large portion of apprehensions. Despite the diverse data problems that we encountered, the consistency in our results across data sets lends some credence to our conclusions.Ig
Notes 1. The estimates from top officials in the 1970s included four to seven million (former Attorney General William Saxbe) and four to twelve million (Commissioner Chapman of the Immigration and Naturalization Service [INS]). Lesko and Associates estimated through a Delphic technique that there were 8.2 million illegal aliens in 1975, of which 5.2 million were Mexicans. The INS used a similar consensus method and came up with 5.5-6 million as of late 1975. 2. Census demographers have consistently come up with estimates below those in n. 1 above. See Lancaster and Sceuren (1977), Heer (1979), Robinson (1980), and Warren and Passel (1987). Other studies of the illegal alien population include Bean, King, and Passel (1983), Brown and Shue (1983), Corwin (1982), Cuthbert and Sterens (1981), Fogel (1978), Heer and Passel (1985), Jones (1984), North and Houstoun (1976), Passel and Woodrow (1984), Reichert and Massey (1979), and Siegal, Passel, and Robinson (1980). 3. The insight that death statistics can be used to measure the hidden population can be attributed to Robinson (1980). Death rate statistics by country of origin were not available at the time he conducted his research. Instead, he used changes in death rates in states such as California that were expected to have large illegal immigrant populations compared with changes for the United States as a whole. Since Robinson’s technique is not subject to bias if country of origin is misrecorded, we attempted to use it to get estimates of the “missing” illegal population. Unfortunately, our experience suggests that the technique is not robust. Robinson implicitly assumes that the death rate in California (or in other states with large illegal alien populations) changes by the same amount as death rates in other states unlikely to have many illegal aliens. The assumed unit coefficient linking death rates is, however, inconsistent with the actual pattern of death rates for the period 1960-70, when changes in the size of the illegal immigrant population are expected to be small: regression analysis shows very little connection between changes in state death rates and national totals. The problem is that state death rates are very “noisy.” When we “smoothed’ the data, we found strikingly different results depending on whether we started our analysis in 1969, 1970, or 1971 and whether we ended it in 1980 or 1984. Without smoothing, the results are even more sensitive to the choice of base year. Hence, we have eschewed use of Robinson’s technique in this paper. 4. Telephone interview, 10 August 1987. 5. We are less certain about the bias in the birth rate calculations since some of the Mexican-born women not counted in the Census may be temporary sojourners who are unlikely to have children. Others, however, may resemble the Mexican-born women in the Census, be living with their families, and have birth rates more like those in Mexico than like those of immigrants permanently established in the United States. 6. In their published article, Bachu and O’Connell (1984) do not report birth rates for Mexican-born women by detailed age. They kindly provided us with the relevant numbers from their computer printouts. 7. There is an important conceptual problem with the use of birth data to estimate
98
G . J. Borjas/R. B. FreemadK. Lang
the size of the illegal alien population. Because fertility decisions are endogenous, many Mexican women may temporarily migrate to the United States simply to bear their children. This ensures that their offspring are American citizens and thus have the option, on reaching adulthood, of migrating legally to the United States. To the extent that this type of migration is common among Mexican-born women, the illegal alien estimates provided by the birth data are biased upward. 8. The 1980 Census has 474,000 Mexican-born women in the age group 15-34, whereas the 1970 Census reports 156,000 in the age group 14-34. This is a threefold increase. 9. There was also a sizable number of apprehensions of illegal aliens in the 1950s. We have not contrasted the situation then to that in the 1970s and 1980s. 10. These estimates are obtained by multiplying the number of apprehensions per year by the postulated ratio of successful crossings to apprehensions by the postulated duration under steady-state assumptions. 11. Our tabulations of the 1980 Census show that 13% of migrants came in the 1965-69 period, compared to 25% in the 1970-74 period, while 33% of migrants came in 1975-80. Dividing these percentages gives the figures in the text. 12. There is some problem with interpretation of the questions. We have added together the total number of times individuals report having been refused entry to the United States and the number of people who report having been caught by the Border Patrol within the United States. The interviewer who conducted the survey believes that respondents interpreted the questions as meaning that they had been caught by the Border Patrol entering the United States (refused entry) or having made significant progress into the United States (apprehended by the Border Patrol within the United States) but before having reached their destination. Assuming that this interpretation is correct, we underestimate the number of apprehensions because we do not know how many times an individual was apprehended after having made significant progress into the United States. 13. Because not all illegal immigrants answered all questions, the number of responses we use to generate various statistics does not necessarily sum to the total in the survey. 14. New York Times, 21 February 1986; Newsweek, 17 March 1986. On the other hand, we note that our interviews with Border Patrol agents produced noticeably higher estimates, averaging around four to one. 15. In particular, there are no strong economic reasons for permanent migrants to move at one time in the year rather than another, given that any seasonal differences in returns to moving would be amortized over a long period. 16. We allowed a two-year gap for adjustment to the end of the Bracero Program. 17. For a comparable analysis, see Bean, Browning, and Frisbie (1984). 18. The survey was conducted by Eric Waggoner, a student at Harvard University. 19. Our conclusions are also consistent with the number of illegal Mexican immigrants who sought amnesty under the new immigration law and with the apparent reluctance of many of those eligible to apply for amnesty because of fears that other members of their families might not be eligible. Our analysis also suggests that the decline in apprehensions that began in about October 1986 may reflect in part not only fears by potential new illegal immigrants that employer sanctions will destroy their chances to obtain employment but also the fact that the new law made it potentially costly for current immigrants to return to Mexico for brief visits since being caught might jeopardize their claim to being continuously resident in the United States. Moreover, the “grandfather” clause exempting existing employees from the employer sanctions may also have the unintended consequence of turning sojourner laborers into permanent residents owing to the increased cost of giving up their jobs to return to Mexico.
99
Undocumented Mexican-born Workers in the U.S.
References Bachu, Amara, and Martin O’Connell. 1984. Developing Current Fertility Indicators for Foreign-born Women from the Current Population Survey. Review of Public Data Use 12:185-95. Bean, Frank D., Harley L. Browning, and W. Farker Frisbie. 1984. The Sociodemographic Characteristics of Mexican Immigrant Status Groups: Implications for Studying Undocumented Mexicans. International Migration Review 18(3):67291. Bean, Frank D., Allan G. King, and Jeffrey S. Passel. 1983. The Number of Illegal Migrants: Sex Ratio Based Estimates for 1980. Demography 20, no. l(Febmary):99-109. Brown, Peter G., and Henry Shue, eds. 1983. The Border That Joins: Mexican Migrants and U S . Responsibility. Totowa, N.J.: Rowman & Littlefield. Chiswick, Barry R. 1986a. Illegal Aliens: A Preliminary Report on an EmployeeEmployer Survey. American Economic Review 76, no. 2(May):253-57. . 1986b. Mexican Immigrants: The Economic Dimension. Annuls of the American Association of Political and Social Science 487(September):438-53. Cornelius, Wayne A. 1977. Illegal Mexican Migration to the United States: Recent Research Findings and Policy Implications. Congressional Record (July 13): 22726-32. Corwin, Arthur F. 1982. The Numbers Game: Estimates of Illegal Aliens in the United States, 1970-1981. Law and Contemporary Problems 45, no. 2(Spring):223-97. Cuthbert, Richard W., and Joe B. Sterens. 1981. The Net Economic Incentive for Illegal Mexican Migration: A Case Study. Znternational Migration Review 15, no. 3(Fall):54 1-49. Eliason, Alan. 1986. Letter to the Editor. New York Times, 29 April. Fogel, Walter. 1978. Mexican Illegal Alien Workers in the United States. Los Angeles: University of California Press. Heer, David M. 1979. What Is the Annual Net Flow of Undocumented Mexican Immigrants to the United States? Demography 16, no. 13(August):417-23. Heer, David M., and Jeffrey S. Passel. 1985. Comparison of Two Different Methods for Computing the Number of Undocumented Mexican Adults in the Los Angeles PMSA. Paper presented at the annual meeting of the Population Association of America, Boston. Immigration and Naturalization Service. 1989. Provisional Legalization Application Statistics. Statistical Branch Analysis. 12 May. Jones, Richard C., ed. 1984. Patterns of Undocumented Migration: Mexico and the United States. Totowa, N.J.: Rowman & Allanheld. Lancaster, Clarice, and Fredrick J. Sceuren. 1977. Counting the Uncountable Illegals: Some Initial Statistical Speculations Employing Capture-Recapture Techniques. Proceedings of the American Statistical Association, Social Statistics Section, pt. 1, pp. 530-35. “A Million Late Arrivals.” 1988. Time, 12 December. North, D. S., and M. F. Houstoun. 1976. The Characteristics and Role of Illegal Aliens in the U S . Labor Market: An Exploratory Study. Washington, D.C.: Linton. Passel, Jeffrey. 1986. Undocumented Immigration. Annals of the American Association of Political and Social Science 487(September): 181-200. Passel, Jeffrey S., and Karen A. Woodrow. 1984. Geographic Distribution of Undocumented Immigrants: Estimates of Undocumented Aliens Counted in the 1980 Census by State. International Migration Review 18(Fall):642-7 1. Reichert, J. S., and D. S. Massey. 1979. Patterns of U.S. Migration from a Mexican
100
G . J. BorjadR. B. FreemadK. Lang
Sending Community: A Comparison of Legal and Illegal Migrants. International Migration Review 13, no. 4(Winter):599-624. Robinson, J. G. 1980. Estimating the Approximate Size of the Illegal Alien Population in the United States by the Comparative Trend Analysis of Age-specific Death Rates. Demography 17, no. 2(May):159-76. Siegal, Jacob S., Jeffrey S. Passel, and J. Gregory Robinson. 1980. Preliminary Review of Existing Studies of the Number of Illegal Residents in the United States. In U.S. Immigration Policy and the National Interest. app. E, Papers on Illegal Immigration to the United States, 13-40. Staff Report of the Select Commission on Immigration and Refugee Policy. Washington, D.C.: U.S. Government Printing Office. Teitelbaum, Michael S. 1986. Immigration and Demographic Change. In World Population and U.S. Policy, ed. Jane Menken. New York: Norton. Warren, Robert, and Jeffrey S. Passel. 1987. A Count of the Uncountable: Estimates of Undocumented Aliens Counted in the 1980 Census. Demography 24 (August):375-93.
3
The Effect of Policy Restrictions on Capital and Labor Flows in Mexico Juan Diez-Canedo R.
This study analyzes the magnitude and characteristics of migratory flows to Mexican border states and to and from the United States using the Mexican Encuesta Nacional de Emigraci6n a la Frontera Norte del Pais y a 10s Estados Unidos, a special 1978 household survey designed to obtain data on such flows. The study focuses mainly on those aspects of the migratory flows that can be traced to the distinct free trade commercial regime in the border area. The findings of the study are as follows. (1) At the time of the 1978 survey, 1.5 percent of the Mexican working population-about 520,000 peoplewere working in the United States, a number far below the millions often cited in the press, while an additional 1.3 percent were return migrants. (2) The border states enjoyed an economic boom when Mexico was in a prolonged recession, with capita inflows and a migrant stream exceeding that to the United States. (3) Other characteristics held fixed, residence in the border area has become nearly as important a determinant of migration to the United States as residence in the historical source regions of the country. (4)Migrants to the border differed in their personal characteristics and regional source from migrants to the United States, for reasons seemingly related to the commercial policies in the border area and suggestive of a family migration pattern. Wives work at the border area, while husbands migrate back and forth to the United States.
3.1 Migration Patterns Because of the scarcity of migration data in general and the poor quality and inherent biases of the available information on illegal migration, in 1978 Juan Diez-Canedo R. is Deputy Director General, Investment Banking, Banco Intemacional, S.N.C. The author is grateful to Enriqueta Lopezlira, Fiorella Tapia, and Gisela Pineda for excellent research assistance and to Kenneth Flamm for comments made on a previous draft of this paper. Of course, he alone is responsible for the contents of this paper.
101
102
Juan Diez-Canedo R.
the Mexican Labor Department (Secretaria del Trabajo) conducted a national household survey in order to measure migratory flows to and from the United States (people who worked in the United States from January 1974 to November 1978 but were living in Mexico at the time of the survey), migratory flows to the Mexican border adjoining the United States, and internal migrati0n.l This national household survey was carried out from December 1978 to January 1979 and covered 115 locations and 62,500 homes selected on a probabilistic basis. A census was done for each household, and four different questionnaires were used depending on the different migratory characteristics. These questionnaires covered 300,000 individuals. The information was grouped according to region (see fig. 3.1) and according to variables such as age, sex, marital status, rural or urban origin, job characteristics, education, and whether respondents were employed, unemployed, or not in the labor force.* Only the most important results pertaining to migration to and return migration from the United States were analyzed in a study by Garcia y Griego and Zazueta (1982). The data on migration to the border and internal migration have not been analyzed. An overall view and the relative importance of the different migratory patterns can be seen in table 3.1. First, at the time of the interview, 1.5 percent of the working-age population were in the United States, 1.3 percent were return migrants, and 1.6 percent had migrated to a border county. In terms of absolute numbers, the figure of 520,000 migrants to the United States shown in table 3.1 is broadly consistent with the figures of Borjas, Freeman, and Lang (in this volume) and inconsistent with the millions of illegal immigrants bandied about in the popular American press. Using information from the
Agua Prieta
Fig. 3.1
The border area and other regions of Mexico
Table 3.1
Migratory Characteristic and Region (population fateen years and older) Region Total I
Total Migration to the United States Return migration Migration to the border Internal migration Nonmigrants Nonspecified
35,622,489 (100.0) 519,406 (100.0) 474,888 (100.0) 557,966 (100.0) 12,312,267 (100.0) 21,484,101 (100.0) 273,861 (100.0)
(100.0) (1.5) (1.3) (1.6) (34.6) (60.3) (.8)
1,688,098 (4.7)
50,044
(9.6) 58,194 (12.3) 317,972 (57.0) 637,723 (5.2) 614,551 (2.9) 9,614 (3.5)
I1 (100.0) 6,945,926 (19.5) (3.0) 205,740
(39.6) (3.4) 140,861 (29.7) (18.8) 75,929 (13.6) (37.8) 2,660,613 (21.6) (36.4) 3,797,756 (17.7) (.6) 65,027 (23.7)
I11 (100.0)
3,153,942 (8.9) (3.0) 140,010
(27.0) (2.0) 119,120 (25.1) (1.1) 37,838 (6.8) (38.3) 784,485 (6.4) (54.7) 2,051,931 (9.6) (.9) 20,558 (7.5)
IV (100.0)
5,152,855 (14.5) (4.4) 77,336
(14.9) (3.8) 96,491 (20.3) (1.2) 89,446 (16.0) (24.9) 1,868,745 (15.2) (65.1) 2,987,434 (13.9) (.7) 33,403 (12.2)
Source: Encuesta Nacional de Emigraci6n a la Frontera Norte del Phis y a 10s Estados Unidos, 1978-79 (CENIET 1984). Nore: For definitions of regions, see fig. 3.1. Numbers in parentheses are percentages.
V (100.0) (1.5) (1.9) (1.7) (36.3) (58.0) (.6)
18,681,668 (52.4) 46,276 (8.9) 60,222 (12.7) 36,781 (6.6) 6,360,701 (51.7) 12,032,429 (56.0) 145,259 (53.0)
104
Juan Diez-Canedo R.
survey, Garcia y Griego and Zazueta (1982, 50) estimated that 750,000 Mexican workers were working in the United States at some point during 1978. Second, most of those workers either migrating to or returning from the United States lived in regions I1 and I11 (66.6 and 54.8 percent of total migration from such regions, respectively), which together comprise the states of the center of Mexico (see the Appendix). Third, almost 60 percent of migrants living in the border area had come from another border county, while 16 percent came from an adjoining state (see fig. 3.1). Fourth, as could be expected, the bulk of the population does not migrate (60.3 percent), and the most important migratory flows are internal (34.6 percent). While there are several studies on Mexican internal migration (see DiezCanedo 1980; Isoard 1976; Greenwood and Ladman 1977; Greenwood 1978) and on Mexican migration to the United States, few analyze migration to the border areas, although the border cities were among the ones with the highest rates of growth in Mexico. From 1970 to 1980, these cities had annual average rates of growth as high as 7 percent for Tijuana or 6.5 percent for Matamoros (see table 3.2), placing them probably among the cities with the highest rates of growth by world standards. Such growth, however, may be explained in part by two important factors. First, the nearly two-thousand mile MexicoU.S. border is probably a unique case-within a few yards, the going minimum wage leaps from around $.40 (Mexico) to $3.35 (the United States) an Table 3.2
Population Change in Major Mexican Cities Bordering the United States and in U.S. SMSA’s Bordering Mexico Population (thousands)
1970 Tijuana Mexicali Ciudad Jufirez Nuevo Laredo Reynosa Matamoros United Mexican States
San Diego Tucson Las Cruces El Paso Laredo McAllen Brownsville South West United States Source: Hansen (1985).
1980
227 267 407 149 137 138
542 495 680 272 240 258
50,695
69,393
1,357 352 70 359 73 182 140 63,000 35,000 203,000
1,861 531 96 480 99 283 210 75 ,000 43 ,000 227,000
Percentage Change, 1970-80
96 85 67 83 75 87 37 51 38 34 36 56 49 19 23 11
105
The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
hour (see table 3.3). Second, also at the time of the study, the northern border of Mexico operated under a different trade regime (which basically allows free trade) from the one applied in the rest of the country. The wage differential, along with the U.S. labor structure, which conditions a high demand for migrant workers, has created a tremendous magnetic force that attracts millions of Mexican workers. Also, an interesting phenomenon of capital and labor attraction has been happening at the border. Through special trade and foreign-ownership laws and rapidly increasing employment creation, large migratory flows have been seeking permanent residence in the border counties. On the other side of the border, the U.S. Standard Metropolitan Statistical Areas (SMSAs) bordering Mexico have also experienced very rapid rates of growth (see table 3.2). El Paso, which had the lowest 1970-80 border SMSA population growth rate (34 percent), grew at over three times the corresponding U.S. national rate of 11 percent and well above the rates of the South (19 percent) and the West (23 percent). Also, the increase in personal income in those SMSAs was, except for the Las Cruces and El Paso, higher than the Minimum Hourly Wages (annual average in dollars)
Table 3.3
1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986
Mexico (1)
United States (2)
Mexican Minimum Wage as a Percentage of U.S. Minimum Wage (2)/(1)
.24 .24 .28 .28 .33 .33 .39 .41 .55 .65 .67 .59 .66 .77 .89 1.09 .78 .52 .58 .59 .42
1.oo 1.oo 1.14 1.29 1.44 1.59 1.60 1.60 1.80 2.00 2.20 2.30 2.65 2.90 3.10 3.35 3.35 3.35 3.35 3.35 3.35
24.0 24.0 24.5 21.7 22.9 20.7 24.3 25.6 30.5 32.5 30.4 25.6 24.9 26.5 28.7 32.5 23.2 15.5 17.3 17.6 12.6
Source: For Mexico, Salarios Minimos 1987, Comision Nacional de 10s Salarios Minimos (Salario Minimo General Promedio Nacional, in dollars using the average controlled rate). For the United States, Staristical Abstract of the United States (federal minimum hourly wage rate for nonfarm workers).
106
Juan Diez-Canedo R.
increases in the United States as a whole and in the South and the West (Hansen 1985).
3.2 The Border Commercial Zone Mexico’s Border Industrialization Program was created in the mid- 1960s and aimed in part to absorb what was perceived as growing unemployment in the border areas due to the termination of the Bracero Program in 1964. Its most important element was the creation of a different trade regime for the border areas through the Muquilu Program. Since 1965, duty-free imports of machinery, equipment, and components for processing and assembly within a twenty-kilometer strip along the border were allowed, provided that all imported products were reexported. The assembly plants are called maquiladoras; they allow for 100 percent foreign ownership as well as exemption of export Also, in some cases, firms may sell up to 20 percent of their production in Mexico. Correspondingly, U.S. tariff regulations 806.30 and 807.00 permit the return of the U.S. component portion duty free, taxing only the value added in Mexico. The development of the maquiladoras has been surprising. At the start of this program, 806.30/807.00 data show that Mexico was less than five times as important as Hong Kong and about as important as Taiwan in the process of industry exports to the United States (see Grunwald and Flamm 1985). In 1983, total 806.30/807.00 imports from Hong Kong and Taiwan were only 12 and 15 percent, respectively, of the Mexican 806.30/807.00 imports. In that year, the main imports under 806.301807.00 came from Japan (30.0 percent), Mexico (18 percent), and West Germany (13 percent). In the last few years, employment in the Mexican manufacturing sector has actually declined, and the gross capital formation in the economy has dropped by 28.5 percent in real terms from 1982 to 1986. While in that period total employment in the Mexican manufacturing industry diminished 6.7 percent, employment in the border assembly plants grew by 96.5 percent, and average work hours increased 88 percent. Thus, maquiladora employment as a percentage of total manufacturing employment increased from 3.4 percent in 1975 to almost 10.8 percent in 1986 (see table 3.4). Although offshore investment has been considered to be “footloose” (Piore 1979, 35-43), especially in the semiconductor assembly operations, and has been found to be highly dependent on the U.S. economic cycle (Bolin 1964), it has also been shown that the U.S. economic cycle has had greater effects on multinational corporations inside the United States than in their offshore operations, for they tend to cut the most costly operations first (Grunwald and Flamm) . The importance of labor to assembly costs, proximity, and the relatively unskilled but highly productive nature of the segment working in assembly plants which, as may be seen in table 3.4, is composed mostly (although in-
107
The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
Table 3.4
Total Employment and Participation of Women in the Muquila Industry, 1975-86 (in thousands) Total Employment
Maquiladoras Blue Collar
1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986
Female Participation
Manufacturing Industry (1)
Maquiladoras" (2)
Total (3)
Women (4)
Blue Collar (5) = (4)/(3)
Total (6) = (4)/(2)
2002 2046 205 1 2133 2291 2417 2543 2485 2340 2361 2415' 231gb
67.2 74.5 78.4 90.7 111.4 119.6 131.0 127.1 150.9 199.7 212.0 249.8
57.9 64.7 68.2 78.6 95.8 102.0 110.7 105.4 125.3 165.5 173.9 203.9
45.3 51.0 53.2 60.4 73.8 78.9 85.7 81.4 93.3 117.3 120.0 139.1
78.3 78.8 78.0 76.8 77.1 77.3 77.4 77.2 74.5 70.9 69.0 68.2
67.4 68.4 67.8 66.6 66.3 66.0 65.4 64.1 61.8 58.7 56.6 55.7
Sources: Estadfstica de la Industria Maquiladora de Exportacidn, Subdireccidn de Estadisticas Econbmicas, Secretm'a de Programacidn y Presupuesto, Mexico D.F. (1987). Sistema de Cuentas Nacionales de Mdxico, Secretm'a de Programacidn y Presupuesto, Mexico D.F. (1987). Indicadores Econdmicos, Banco de M&xico,Mexico D.F. (1987). 'Includes blue collar and white collar. bEstimation based on annual variations reported in the Encuesta Industrial Mensual, Secretaria de Programacidn y Presupuesto.
creasingly less so) of women, helps to explain the relatively steady growth of the maquiladoras. In the last ten years the number of assembly plans has practically doubled and, with the exception of 1982 (the year of the Mexican debt crisis), its growth has been steady. Except for 1982, during which the peso was clearly overvalued and there were signs of political instability and no economic growth, investment in the assembly plants has been not only stable but growing significantly, while as of the 1982 crisis and up to 1989 the opposite happened to investment in the manufacturing sector. Figure 3.2 helps to explain in part why such a phenomenon happened. After 1970 and up to 1986 the real exchange rate, estimated using the consumer price indexes, was considerably favorable for exports, with the exception of only three years. If the real exchange rate is calculated using, instead of consumer prices, wage indexes which are more relevant from the point of view of exporters, the Mexican real exchange rate has a considerable increase of 113.3 percent as of December 1985, from the base year of 1970. So the global competitiveness of manufacturing in the period under study was considerable, and even more so at the border where, on top of that, there have been no import restrictions whatsoever.
Juan Diez-Canedo R.
108 225
0.
175
t
. ’
. 0.
0
150
125 100
65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86
Fig. 3.2 Real exchange rate index (1970 = 100) Source: Gerencia del Sector Real, Banco de Mexico. 0Mean-Consumer Prices: Estimated in base to a divisia index of consumer price index of 133 countries, and a divisia index of the exchange rate of each country, weighted both by the share of each country’s GDP on total GDP. 0 Mean-Wages: Estimated in base to a divisia index of the wage indices of 19 countries, and a divisia index of the exchange rate of each country, weighted both by the share of each country’s GDP on total GDP.
The migratory flows both to the border and to the United States must be following better economic opportunities. However, it is interesting to see which factors separate migrants to the border from migrants to the United States or whether in fact migration to the border is just a step in a process of migration to the United States. After all, it would appear that factor complementarity should be relatively similar at the border, given its special trade regime, to that in the United States. Piore’s (1979, 35-43) dual labor market hypothesis for explaining the functioning of the labor market and the logical role of the migrants in the secondary sector serves to explain the role of migrants in the U.S. labor market and may also serve to explain the specific case of the Mexican assembly plants. These plants grew out of an external shock, which in this case was the end of the Bracero Program. This event may have been perceived by some U.S. plants as a new need for exporting at least part of the secondary labor requirements to Mexico and by Mexico as a need of capturing those same jobs. For these reasons, foreign technicians and managerial personnel (primary labor market types) are allowed to reside in Mexico. Also, foreign entrepreneurs, originally only from the United States, established twin plants, (Bolin 1964) with capital-intensive processes in the United States and labor-intensive ones in Mexico, thus minimizing costs and maximizing managerial functions (a sort of intraindustry Heckscher-Ohlin production scheme). Although there are a number of twin plants, in many cases the U.S. home office is actually a
109
The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
long distance away from the border (in 1978, forty-eight of the Fortune 500 companies had maquiladoras in Mexico; see Grunwald and Flamm 1985), but there is an actual trend, which includes growing numbers of Japanese companies in the United States (like Sony), toward the twin plant concept of production. Table 3.5 gives an idea of how labor markets could be complementary, both through Mexican migration to the United States and at the border, and tends to confirm a dual labor market hypothesis type of relation. Some differences, however, are apparent. While farm workers are the most important segment of migrants to the United States and white-collar workers represent only 4.5 percent of these workers, the most important segment of border migrants was blue collar, followed by service workers, and the percentage of white-collar workers was more than twice that of return migrants. Additional differences between flows to the United States and border areas are shown in table 3.6, which compares the sex and education composition of migrants to the United States, migrants returning from the United States, and migrants to Northern Mexico. While migrants going to and returning from the United States were men in 83 and 80 percent of the cases, respectively, migrants to the border were mainly women (5l .7 percent). The fact that migrants to the border are primarily females may reflect the particular demands of the maquiladora industry, which employs mostly women. The different patterns of migration also reveal different levels of education. Although the level of education is in general very poor in all cases-at least 50 percent of male and female migrants are virtually illiterate-a higher proportion of migrants to the border have junior high and high school educations. This fact could be conditioned by the relatively higher employment requirements that exist in the commercial and assembly plant sectors as well as by the presence of a higher proportion of white-collar workers (see table 3.5).
Table 3.5
Occupation Total White-collar workers Blue-collar workers Service workers Farm workers
Occupational Distribution of Border and Return Migrants Compared to U.S. Workers
United States 100.0 50.0 33.4 13.6 3.0
Population of Mexican Southwest Origin 100.0 53.8 30.6 13.2 2.4
100. 1 28.4 50.6 16.3 4.8
Return Migrants from the United States
Border Migrants
100.0 4.5 35.8 23.5 36.1
100.0 9.4 39.7 28.4 22.5
~
Sources: First four columns taken from Garcia y Griego and Zazueta (1982, 81). Column 1: U.S. Bureau of the Census, Statistical Abstract of the United States, 1979. Column 2: U.S. Bureau of Labor Statistics, Geographic Profile of Employment and Unemployment: States 1978, Metropolitan Areas 1977-78, September 1979. Column 3: U.S. Bureau of the Census, Current Population Reports, Series p. 20, no. 339, June 1979. Columns 4 and 5: Encuesta Nacional de Emigracdn a la Frontera Norte del Pais y a 10s Estados Unidos, 1978-79 (CENIET 1984).
110
Juan Diez-Canedo R.
Table 3.6
Migrants in the United States and Return Migrants from the United States, by Sex and Education
Migrants in the United States Percentage by sex Percentage male Percentage female Percentage by sex and education Percentage male Percentage female Return migrants Percentage by sex Percentage male Percentage female Percentage by sex and education Percentage male Percentage female Migrants to the northern border Percentage by sex Percentage male Percentage female Percentage by sex and education Percentage male Percentage female
Total
Without Formal Education
Elementary
Junior High
High School
College
503,803 100.0 83.0 17.0
302,936 100.0 85.5 14.5
148,862 100.0 79.2 20.8
41,770 100.0 17.2 22.8
9,434 100.0 87.4 12.6
80 1 100.0 95.5 4.5
100.0 100.0 100.0
60.1 61.9 51.4
29.5 28.2 36.1
8.3 7.7 11.1
1.9 2.0 1.4
.2 .2
465,766 100.0 80.0 20.0
308,672 100.0 83.3 16.7
107,415 100.0 75.8 24.2
33,598 100.0 67.1 32.9
13,907 100.0 76.4 23.6
2,174 100.0 32.8 67.2
100.0 100.0 100.0
66.3 69.0 55.3
23.1 21.9 27.8
1.2 6.1 11.8
3.0 2.9 3.5
.5 .2 1.6
532,802 100.0 48.3 51.7
266,623 100.0 48.7 51.3
138,749 100.0 46.4 53.6
91,203 100.0 47.5 52.5
30,833 100.0 52.0 48.0
5,394 100.0 61.2 32.8
100.0 100.0 100.0
50.0 50.5 49.6
26.0 25.0 27.0
17.1 16.8 17.4
5.8 6.2 5.4
1.o 1.4 .6
.o
Nore: Some migrants did not answer the schooling question.
All the figures given above seem to confirm the secondary labor market characteristics of the three patterns of migration.
3.3 The Determinants of Migration to the United States and the Northern Mexican Border In the Survey of Migration to the United States and to the Northern Border (Garcia y Griego and Zazueta 1982), data for the population aged fifteen years old and over and for the employed labor force are recorded according to different migratory patterns: migration to the United States (USM); return migration from the United States (RMUS); migration to the north border area (MNB); internal migration (IM); and nonmigration (NM). For each migratory pattern, the information was available aggregated in relation to socioeconomic
111
The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
and demographic factors such as region, age, sex, marital status, education, origin of the population, employment status, occupational status, and the economic sector in which the individual was occupied (see the Appendix). The available information did not make it possible to analyze all the population characteristics through one econometric model. Because information was available only through specific tabulations, the data had to be analyzed through six different models for each migratory pattern. Information was grouped in tables that contained three nonvarying characteristics and a fourth one that varied. For instance, one group included region, age, and sex as the nonvarying characteristics and marital status, schooling, or whether the origin was urban or rural, for example, as the fourth, variable, characteristic. Furthermore, the way in which the information was classified made it virtually impossible to include additional explanatory variables such as income differentials, differences in financial resources, investment, distance, etc. because the regions defined in the survey included groupings of states and, in some cases, one region included only one part of one state. Ideally, information at an individual level should have been used for the analysis. Unfortunately, however, such data disaggregation could not be obtained. As was mentioned before, available data were grouped and crossclassified in tables. In such cases, a generalized linear model can be defined for categorical data in which the observations consist of counts of frequencies in the cells of a contingency table formed by the cross-classificationof dependent (or response) and explanatory (or independent) variables. In this contingency table method, a log-linear model was specified and fitted to the different migratory patterns (USM, RMUS, MNB). For the purpose of estimation, a Poisson distribution was assumed. Since this involves unrestricted independent random variables with distributions from the exponential family, the Newton-Raphson estimation procedures implemented in GLIM (Numerical Algorithms Group, Oxford) were used to estimate the parameters (see Nelder 1974; and Dobson 1983, 99). (The GLIM estimates are maximum likelihood estimates.) The responsive variables of the log-linear models were standardized in the following form: each number in a contingency table cell was divided by the total number of related cells. The following examples illustrate this point. The number of 15- to 29-year-old single males who migrated to the United States from region I was divided by the total number of 15- to 29-year-old single males of region I, and the number of 30- to 44-year-old married women who migrated to the northern border from region I1 was also divided by the total number of 30- to 44-year-old married women in region 11. For each migratory characteristic, the following log-linear models were specified:
+ R, + A, + S, + CS,,
(1)
mrlasc= K ,
(2)
mzJasn= K , + R, + A a + S, + U,,
Juan Diez-Canedo R.
112
(3)
mrJaro= K ,
+ R, + A, + S, + OP,,
+ R, + A, + S, + Sch,, mlJOsq= K, + R, + A, + S, + ES,, mllasr = K , + R, + Aa + S, + OC,, m,,,sp = K,
(4) (5)
(6) where, for example,
h= I
where M = understandardized information; i = the different migratory patterns: USM, RMUS, MNB, and internal and nonmigration (which have been pooled); K = mean or constant; R = region; A = age; S = sex; cs = marital status; U = type or origin; OP = occupational status; Sch = schooling; ES = economic sector; oc = type of job; j = region I, 11, 111, IV, or V; a = age groups: 15-29, 30-44,45-49, 60 and over; s = male, female; c = single, married; n = rural, urban; o = employed, unemployed, not in the labor force; P = schooling: less than elementary, elementary, high school, college; 4 = economic sector: primary, secondary, tertiary; r = self employed, blue collar, day laborer, unpaid family member, ejidutario, landh~lder.~ 3.4
Results
For simplicity, tables 3.7 and 3.8 summarize the results of these estimates in terms of the estimated coefficients for analyses that treat age, region, sex, and either rural origin or occupational status, with appropriate interactions. The total population fifteen and over is the base for the calculations when rural origin is included, while the economically active population is the base when
113
The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
Table 3.7
Estimates of Log-Linear Main and Interaction Effect Parameters for Migration to the United States (Ml) Coefficient
SE
T
By Region, Age, Sex, and Origin Mean (K) Region (R): I I1 I11 IV V Age (A): 15-29 3044 45-59 Sex (S): Male Female Origin (0: Urban Rural Interaction, Region-Origin: Region I Urban Region I1 Urban Region 111Urban Region IV Urban
-1.145 2.804 2.888 3.161 2.064
,5918
-7.0*
,5563 ,5550
5.0*
.5739
5.2* 5.7* 3.6*
1.684 1.612 1.372
,2351 ,2364 ,2417
7.2* 6.8* 5.7*
1.619
,1487
10.9*
,675
.9
,5514
...
... .593
- ,6263 - 1.633 - 3.24 - ,9778
,7018 ,7208 ,8041 ,7413
- .9 -2.3* -4.0* - 1.3
By Region, Age, Sex, and Occupation Mean ( K ) Region (R): I I1 I11 IV V Age (A): 15-29 3044 45-59 Sex (S): Male Female Occupation status (OP): Self-employed Blue-collar Day laborer Unpaid family member Ejidatario Landholder (continued)
- 1.754
,6228
- 2.8*
2.797 1.859 2.531 1.438
.526 .549 ,5306 ,568
5.3* 3.4* 4.8* 2.5*
.3342 ,3327 ,3644
5.1* 5.3* 2.5*
... 1.695 1.755 ,8985
- 1.774
...
s605
-3.2*
- ,7971 - ,583 ,302 - 6.087
,3829 ,3566 .2815 4.166
-2.1* -1.6*
-1.334 .
,4668
-2.9*
1.1
- 1.5*
114
Juan Diez-Canedo R.
Table 3.7
(continued) Coefficient
Interaction, occupational status-sex: Self-employed, male Blue-collar, male Day laborer, male Unpaid family member, male Ejiduturio,male Note: Goodness-of-fit deviance = 29.1, mean deviance of the null model. *Significant at the 5 percent level.
2.572 2.4 1.542 5.313 ,7819
x2
=
SE
T
,6788 .6629 ,6247 4.267
3.8* 3.6* 2.5* 1.2
,9761
.8
90.53. The model explains 87.3 percent of the
occupation is included. Computations for other classifications gave similar results. Table 3.7 records estimates of the effect of the various factors on migration to the United States. It shows that region is an important determinant of migration to the United States, with residence in the border region having nearly as significant an effect in increasing the probability of migration to the United States as residence in region 111, which comprises the states that are commonly reported as the source of migration to the United States since the turn of the century. The implication is that the border has become an important stepping-stone for migration to the United States, controlling for other differences. Note further that, in the calculations with urban (as opposed to rural) origin included, that factor does not enter significantly for the border area but does enter for regions I1 and 111. With respect to occupation, the surprising result is the relative weakness of the occupational variables, which produced a poorer fit than did other classifications of the data. While male day laborers and blue-collar workers are especially likely to migrate to the United States, the stereotypes of the Mexican immigrant as an unskilled farm worker seems exaggerated on the basis of this calculation. The fact that male agricultural migrants work in areas close to the Mexican border, where about 90 percent of the Border Patrol is located, and that workers from regions I1 and I11 are more likely to be of rural origin makes it especially likely that they are captured by the Border Patrol, leading to the view that the vast bulk of migrants are agricultural laborers. In fact, while migrants to the United States are less educated than other Mexicans (see table 3.6), they are not overwhelmingly farm laborers (our results are consistent with Borjas, Freeman, and Lang’s [in this volume] findings from the U.S. Census). Finally, age coefficients show that the population group 30 to 44 years old was about as significant for explaining migration to the United States as the 15-29 age group. Table 3.8 presents my estimates of the effect of the factors on being a return migrant from the United States and on being a migrant to the Mexican border.
115
The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
Table 3.8
Estimates of Log-Linear Main and Interaction Effect Parameters for Migration Models Coefficient
T
SE
Return Migration from the United States Mean (K) Region (R):
I I1 I11 IV V Age (A): 15-29 30-44 45-59 Sex (S): Male Female Marital status (CS): Single Manied
- 2.736
,4286
- 6.4*
2.301 1.867 2.251 1.502
,3760 ,3850 .3770 ,3960
6.1* 4.8* 6.0* 3.8*
1.155 1.115 ,871
,2290 .23 .238
4.8*
1.424
,1630
8.7*
5.0* 3.7*
... - .419
,132
-3.2*
... Migration to the Border Areab
Mean (K) Region (R):
I I1 111
IV V Age (A): 15-29 30-44 45-59 Sex (S): Male Female Marital status (CS): Single Married
-1.162
,3633
- 3.2*
4.492 1.727 1.613 1.987
,359 ,387 ,391 .38
12.5* 4.5* 4.1* 5.2*
- ,471 -.186 - .01
,103 ,095 .09
-4.6* - 2.0* -.l
- ,087
.069
-
- ,223
,069
-3.2*
1.3
*Significant at the 5 percent level. Goodness-of-fit deviance = 48.6, x2 = 90.5. The model explains 76.3 percent of the mean deviance of the null model. b Goodness-of-fit deviance = 33.3, x2 = 90.5. The model explains 95.8 percent of the mean deviance of the null model. a
116
Juan Diez-Canedo R.
The return migrant calculations show that the probability of being a return migrant is highest for the border region, again indicating that residence in the border has become an important factor in explaining migration to the United States. In general, the coefficients for the determinants of return migrants are similar to those for migrants still in the United States, implying that I am identifying roughly comparable populations. The highest proportion of return migrants were aged 15 to 29 years, followed by the groups aged 30 to 44 and 45 to 49. Note also that being male raises the probability of having migrated to and returned from the United States. Models that included rural or urban origin, schooling level, occupational status, and economic sector of activity were not useful for explaining this migratory pattern. By contrast, the patterns of migration to the northern border differ considerably from those observed for migration to the United States. First, sex is not relevant for explaining this type of flow, therefore indicating that this migration is mostly of a permanent nature, as opposed to the temporary one shown in the results for migrants to the United States and return migrants. Second, the most important regions for this migratory flow are I (the border itself) and IV (the region adjoining the border). The increasing flow of migration from the adjoining counties happens first as a daily trip to the border (the transportation business is booming), which apparently becomes permanent after a while. The older age group (60and over) was the most important in relation to this migratory flow. Another interesting fact is that, although not significant, the proportion of women is more important than men in this migratory process, a fact that is explained by the existence of a higher proportion of women in the assembly plants. Examined together, tables 3.7 and 3.8 suggest an interesting pattern of interrelated migration to the border and to the United States. The different effect of sex on migration in the calculations can explain where males go when their wives work at the border. Since an important part of the work force, at least in maquiladoras, is female, the ideal overall migration strategy may be accomplished by a joint family decision where males migrate to the United States and females migrant to the border. This hypothesis is backed by the fact that married males who had migrated to the United States had the highest probability of being return migrants. In conclusion, it can be said that the 1964 termination of the Bracero Program conditioned a policy response that gave birth to a very successful border industrial venture. This venture has created an important number of jobs of a secondary market type, has attracted foreign and local capital, and has conditioned, through forward and backward linkages, growth in the border area while the rest of the country was, in the period under study, in the middle of a protracted recession. However, another important fact that should be noted is that migrants, whether to or from the United States or to the border, compose only a negligible portion of the total Mexican work force. Evidence found in this research also suggests that migration to the northern Mexican border may
117
The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
indeed be a step toward migration to the United States. Given that migration to the border seems to entail complete family units, that assembly plants employ mostly women, and that married males are an important element for explaining migration to the United States from the northern border, this pattern of migration may indeed be an optimal labor market decision for family units. From this perspective, one cannot understand Mexican migration to the United States separately from migration to the border.
Appendix The population aged 15 years old and over was recorded for each migration pattern, in various contingency tables formed by the cross-classification of four different independent variables each with two or more categories: Contingency Table 11.1. Region, age, sex, and education. Contingency Table 11.2. Region, age, sex, and origin. Contingency Table 11.3. Region, age, sex, and marital status. Contingency Table 11.4. Region, age, sex, and employment status.
Similarly, the employed labor force for each migratory pattern was crossclassified in accordance with four independent variables, each with two or more scales: Contingency Table 11.5. Region, age, sex, and occupation status. Contingency Table 11.6. Region, age, sex, and economic sector.
The survey divided the territory into five different zones according to the density flow of migrants (see fig. 3.1): Region I. The northern border area. Region 11. The states of Jalisco, Colima, Guanajuato, Michoach, part of Guerrero and the state of Mexico, and Ensenada in Baja California Norte. Region I l l . The states of Aguascalientes, Durango, Nayarit, Zacatecas, Querktaro, San Luis Potosi, and part of the state of Hidalgo. Region IR The states of Tamaulipas, Nuevo L e h , Coahuila, and Sonora (exempting the border area), plus the state of Sinaloa. Region K The states of Baja California Sur, Campeche, Chiapas, the Federal District, Oaxaca, Quintana Roo, Tabasco, Veracruz, Yucathn, Morelos, Puebla, Tlaxcala, and part of the states of Hidalgo, Mexico, and Guerrero.
The dichotomous variables available are a) Male or female. b ) Urban or rural population.
118
Juan Diez-Canedo R.
c) Marital status-single (includes single, widowed, divorced and separated) or married (including those who live in free union). The age variable is divided into four ranks: a ) fifteen to twenty-nine years old. b) thirty to forty-four years old. c ) forty-five to fifty-nine years old. d ) sixty years old and over.
The education variable is divided into four parts: Less than elementary. Those who are illiterate or did not finish elementary school. Elementary. Those who finished elementary school but not junior high school or the equivalent. Junior high school, Those who finished junior high school but not senior high school or the equivalent. Senior high school. Those who finished senior high school but not college. College.
The employment status variable included three categories: a ) Employed labor force. b) Unemployed labor force. c) Not in the labor force.
The occupational status is divided into six categories: a ) Self-employed. b) Blue collar. c) Day labor. d ) Unpaid family member. e ) Ejidatario. f ) Landholder.
The economic sector of occupation is divided into the following categories: a ) Primary sector. b) Secondary sector. c) Tertiary sector.
Notes 1. For a detailed explanation of available data on migration and biases in that data, see Diez-Canedo (1980). The Secretm’a del Trabajo conducted the household survey through the Centro Nacional de InformaciBn y Estudios del Trabajo (CENIET).
119
The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
2. See the Appendix. For a detailed explanation of the methodological aspects of this survey, see Garcia y Griego and Zazueta (1982). 3. At the time of the survey, except for the border areas, 100 percent ownership was allowed only a very few cases; normally, a 51 percent Mexican share was required. Recently, however, this restriction on foreign investment has been relaxed drastically. 4. An ejidatario is an ejido’s “landowner,” although the ejidos cannot be sold or mortgaged. The ejido is a plot of land owned by the nation through a community of “ejidatarios.” 5. Variables in contingency table analyses are described that are not reported in tables.
Bibliography Amemiya, Takeshi. 1981.Qualitative Response Models: A Survey. Journal of Economic Literature 19:1483-1536. Bolin, R. 1964.Industrial Opportunities for Ciudad Juhrez. Report prepared by Arthur D. Little Consulting Firm, 1 August. Diez-Canedo, J. 1980. A New View of Mexican Migration to the United States. Ph.D. diss., Massachusetts Institute of Technology. Dobson, Annette J. 1983.An Introduction to Statistical Modeling. London and New York: Chapman & Hall. Flamm, K. 1984.The Volatility of Offshore Investment. Journal of Development Economics, 16-34. Garcia y Griego, M., and C. Zazueta. 1982.Los Trabajadores Mexicanos en Estados Unidos. Mexico City: Secretaria del Trabajo, CENIET. Greenwood, M. J. 1978.An Econometric Model of Internal Migration and Regional Economic Growth in Mexico. Journal of Regional Science 18(1). Greenwood, M. J . , and J. Ladman. 1977.Economia de la Movilidad Geogr&ca de la Mano de Obra en MCxico. Demografia y Economi’a 1 l(2). Grunwald, J., and Kenneth Flamm. 1985. The Global Factory. Washington, D.C.: Brookings. Hansen, Niles. 1985.The Nature and Significance of Border Development Patterns. In The U S . and Mexico: Borderland Development and the National Economies, ed. L. Gibson and A. Corona, 3-17. Boulder, Colo.: Westview. Isoard, C. 1976.Inter-regional Migration in Mexico. University of Chicago, Department of Economics. Typescript. Nelder, J. A. 1974. Log-linear Models for Contingency Tables: A Generalization of Classical Least Squares. Applied Statistics 23:323-29. Piore, M. J. 1979.Birds of Passage. Cambridge: Cambridge University Press. Tamayo, J. 1981. Algunas Ideas Inexactas Acerca de la Frontera Norte de Mtxico. Mexico City: Centro de Investigacibn y Docencia Econhica. Mimeo. Tellez, L. 1986.Essays on the Financial and Real Aspects of an Open Economy: The Case of Mexico. Ph.D. diss., Massachusetts Institute of echnology.
This Page Intentionally Left Blank
4
Internal Migration of U. S . Immigrants Ann P. Bartel and Marianne J. Koch
In a recent study of Indochinese refugees who arrived in the United States in 1975, after the fall of Saigon, Baker and North (1984) observed that 45 percent of these individuals lived in a different state in 1980 than in 1975, compared to 9 percent of the U.S. population. This finding is interesting not only because of the sheer magnitude of the migration but also because of its effect; the result of the relocation was an increased concentration of the refugees in a small number of states. This particular group of immigrants constitutes a very special case in that they all came as refugees from the same part of the world and in that they did not choose their 1975 locations but were placed there under a program operated by the Inter-Agency Task Force for Indochinese Refugees. Whether these findings generalize to other groups of immigrants that have recently arrived in the United States is the subject of this paper. Specifically, we ask how dispersed U.S. immigrants are on their arrival to this country and whether they change locations as time in the United States elapses. Further, we explore the determinants of the internal migration of U.S. immigrants and consider the effect of mobility on their earnings. The study of the internal migration of immigrants is important in light of the expanding numbers of immigrants to this country. The ability to predict which areas of the country will receive immigrants will aid in planning for the provision of services to local populations. Further, if it is the case, as it was shown to be with the Indochinese refugees, that immigrants tend to cluster where their countrymen are located, planners in these cities may need to design bilingual or even multilingual programs to implement the delivery of services; this may, in turn, lead to a more permanently segregated U.S. society. Ann P. Bartel is professor of business at Columbia University’s Graduate School of Business and a research associate of the National Bureau of Economic Research. Marianne J. Koch is assistant professor of management in the College of Business Administration at the University of Oregon.
121
122
Ann P. Bartel and Marianne J. Koch
In this paper, we investigate the internal mobility patterns of Asian, Central and South American, and European immigrants who arrived in the United States after 1964. In section 4.1, we study the extent to which they disperse throughout the country as time in the United States elapses and compare their inter-SMSA (standard metropolitan statistical area) mobility rates to those of the native population. Section 4.2 presents the results of estimating an econometric model of the determinants of changing SMSAs. A brief analysis of the effect of internal migration on earnings is discussed in section 4.3, and conclusions and policy implications are presented in section 4.4
4.1 Mobility Patterns In this section, we present statistics that can be used to describe the internal mobility patterns of recent immigrants to the United States. We restrict our analysis to SMSAs that have at least ten male immigrants aged 22-64 in the Public Use B Sample of the 1980 Census of Population. This resulted in fiftythree SMSAs. Since 90% of all the immigrants in the Census files (and 43 percent of the U.S. population) live in one of these SMSAs, this is not a very restrictive selection rule.’ The “new immigrant” population-or immigrants arriving in the United States after 1964-is stratified into three cohorts based on year of arrival in the United States: arrivals between 1965 and 1969, 1970 and 1974, and 1975 and 1979. The individuals in each of these cohorts are restricted to be of working age at the time of their arrival in the United States; hence, the most recent arrivals, those who came between 1975 and 1979, are aged 22-54 in 1980, while those who came between 1970 and 1974 are aged 27-59 in 1980, and the 1965-69 arrivals are aged 32-64 in 1980. One way to describe the internal mobility patterns of recent immigrants to the United States is to calculate a Herfindahl index for the various samples of immigrants that we wish to study; this statistic provides a summary measure of the extent to which each of the groups is geographically dispersed throughout the United States.ZHigher values of the index represent greater geographic concentration. In table 4.1, Herfindahl indices are shown for three main ethnic groups-Asians, Central and South Americans, and Europeans-and for several subcategories in each of these groups. The Herfindahl indices for the most recent cohort (i.e., the 1975-79 arrivals) are shown in column 1, for the middle cohort (i.e., the 1970-74 arrivals) in column 2, and for the earliest group (i.e., the 1965-69 arrivals) in column 3. In order to appreciate the magnitude of these indices, we also calculated the Herfindahl index for the total population in the fifty-three SMSAs in 1980; as shown in the note to the table, this statistic is .04. Numbers in excess of .04 indicate that the group under study is more concentrated than the total population in this sample of cities.
123
Internal Migration of U.S. Immigrants
Table 4.1
Asians: China India Japan Korea Philippines Vietnam Central and South Americans: Cuba Mexico Other Europeans: England Italy Greece
Herlindahl Indices for 1980 Geographic Distributions of Various Immigrant Cohorts (1) Amved 1975-79 (aged 22-54 in 1980)
(2) Amved 1970-74 (aged 27-59 in 1980)
(3) Arrived 1965-69 (aged 32-64 in 1980)
.09 .17 .09 .15 .I8 .14 .08
.09 .I6 .08
.09
.I1 .I6
.07 .I9 .I1 .I2 .50
.13
.14 .34 .24 .24
.I4 .36 .26 .26
.40
.20 .17 .07 .05 .09 .14
.18 .I1
.10 .05
.19 .I2
.I5
.08 .06 .15 .14
Nore: The Herfindahl Index for the total U.S. population in the fifty-three cities in the sample is .04.
The data in table 4.1 can be used to answer the question whether immigrants become more geographically dispersed as they acquire experience in this country. The degree of geographical dispersion is one indicator of assimilation into the new country (Massey 1981). To the extent that immigrants are able to learn about opportunities in other parts of the country as time in the United States elapses, we would expect to observe greater dispersion over time or evidence of assimilation. In terms of table 4.1, we would expect to observer smaller values in column 3 as compared to column 2 and, certainly, as compared to column 1. A possible offsetting factor is that immigrants may move from their initial destinations in the United States only to discover that they are unable to live without the support of their ethnic enclaves and then return to their SMSAs of initial destination. If this happens, we would observe that the degree of dispersion of the three cohorts in the 1980 cross section would be very similar. We begin our discussion of table 4.1 by looking at the three main ethnic groups. Two important findings emerge. First, the immigrants, and especially the Central and South American immigrants, are more geographically concentrated than the U.S. population in the fifty-three cities. Second, for each of the main ethnic groups, there is no evidence of increased dispersion over time. Since the country composition of each of the main ethnic groups is likely to
124
Ann P. Bartel and Marianne J. Koch
have changed over time, the constancy of the Herfindahl indices across cohorts need not imply that there has been no dispersion for a given subcategory. For example, subgroups that are more dispersed on arrival to the United States may account for a larger proportion of the main ethnic group that has recently arrived; in table 4.1, this could mask the dispersion over time of other subgroups that represented a large share of the early cohorts. Hence, we also study the trend in the Herfindahl index for selected subgroups. There are only two cases, namely, the Koreans and the Cubans, in which we see greater geographic dispersion of the cohorts that arrived earlier.3In sum, the evidence of table 4.1 gives only limited support for the hypothesis that, as time elapses in the United States, the immigrants will become more dispersed throughout the country. Of course, one of the problems with table 4.1 is that it is calculated from cross-sectional data on immigrants who were living in the United States in 1980. We have attempted to interpret these data as a pseudo-panel in order to draw a conclusion about changes in geographic dispersion over time. Actually, this conclusion can be based only on data for a given sample of immigrants who are observed at more than one point in time. Since 50% of the individuals in the Public Use Sample of the 1980 Census of Population were asked where they lived in 1975, it is possible to create a panel for this group. We calculated the Herfindahl indices for these individuals first for their 1975 locations and then for their locations in 1980. If the dispersion-with-time hypothesis is correct, then we should observe a decrease in the Herfindahl index between 1975 and 1980. Table 4.2 reports these results for the 1965-69 and 1970-74 cohorts. When the main ethnic groups are not disaggregated, the Herfindahl indices hardly change between 1975 and 1980. For certain subgroups, however, we do find evidence of increased geographic dispersion between 1975 and 1980, and, to provide an arbitrary benchmark, we have indicated with an asterisk those groups for which there is at least a 20% fall in the Herfindahl index over the five-year period. These groups are entirely from the Asian category, namely, the immigrants from India, Japan, and Korea. We had inferred from table 4.1 that the Cubans dispersed over time, but, as table 4.2 clearly shows, this conclusion was erroneously based on our interpretation of a cross section of cohorts as a pseudo-panel. Relying on table 4.2 as the more correct picture of dispersion, we conclude that geographic dispersion over time is not a typical characteristic of the post- 1964 immigrants to the United States. Next we ask, although the new immigrants are not dispersing throughout the country, are they moving at all, or are they remaining in their original destination SMSAs? In other words, it is possible that these individuals are moving between SMSAs but that the degree of dispersion of the group is not changing; that is, person A is moving from city 1 to city 2 while person B is moving from city 2 to city 1. The percentages of various immigrant and ethnic
Internal Migration of U.S. Immigrants
125
Table 4.2
Herfindahl Indices in 1975 and 1980 Arrived 1970-74 (aged 27-59 in 1980)
Arrived 1964-69 (aged 32-64 in 1980)
1975
1980
1975
1980
Asians: China India Japan Korea Philippines Vietnam
.ll .22 .13 .14 .I4 .I2 .22
.10 .19 .lo* .11* .lo* .I1 .22
.09 .I6 .10 .25 .I5 .I4
.09 .I5
...
...
Central and South Americans: Cuba Mexico Other
.16 .32 .27 .30
.15 .36 .26 .28
.I3 .32 .27 .25
.13 .36 .27 .23
.I1
.I0 .07 .I3 .13
.10 .07 .22 .I4
Europeans: England Italy Greece
.08
.14 .15
.08*
.19* .12* .14
.G9 .08
.18 .14
Nore: Herfindahl indices are calculated for the sample of immigrants that resided in one of the fifty-three SMSAs in both 1975 and 1980. See text. *The Herfindahl index decreased by at least 20% between 1975 and 1980.
native groups who moved between one of the fifty-three SMSAs in our sample during the period 1975-80 are shown in table 4.34 As before, the results are specific to the particular ethnic group. We find that, although Asian immigrants are between two and three times as likely as natives of Asian ethnicity to change SMSAs, the Central and South American and European immigrants are less mobile than their native counterparts. Among the Asians, the Indians and Koreans stand out as the most mobile, and these were two of the three groups for which dispersion over time was observed. The major conclusion from our analysis of the data in tables 4.1-4.3 is that there is little systematic evidence of dispersion of the immigrants throughout our sample of cities. This is true even for the Asian ethnic group whose interSMSA mobility rate greatly exceeds that of any of the native groups. In other words, immigrants do move between SMSAs during their first five or ten years in this country, but this mobility does not substantially affect the index of dispersion, except in a few isolated cases. In the next section of the paper, we describe and estimate a logit model that can explain why the immigrants in our sample did change SMSAs between 1975 and 1980.
126
Ann P. Bartel and Marianne J. Koch
Table 4.3
Percentage of Immigrants and Natives Changing SMSAs between 1975 and 1980 (sample sizes in parentheses) (1) Arrived 1970-74 (aged 27-59 in 1980)
(2) Arrived 1965-69 (aged 32-64 in 1980)
(3) Natives (aged 27-64 in 1980)’
Asians: China India Japan Korea Philippines Vietnam Central and South Americans: Cuba Mexico Other Europeans: England Italy Greece
7.4 7.8 4.0 9.5
(838) (205) (253) (380)
10.6
(1,965)
8.4 12.1 7.8 5.1
(439) (33) (102) (59)
14.0
(1,275)
Nore: Percentages are. calculated for the sample of immigrants that resided in one of the fifty-three SMSAs in both 1975 and 1980. aThe natives are disaggregated into three categories: (1) Asian ethnicity; (2) Central and South American ethnicity; and (3) non-Hispanic whites or European ethnicity.
4.2 Determinants of Internal Migration In this section, we describe and estimate an econometric model of the determinants of the internal migration of recent immigrants to the United States. In section 4.2.1, the model is described and the variables we utilize defined. Section 4.2.2 presents the results for the immigrants and compares them to the results for natives of similar ethnicity as well as native whites. 4.2.1 Econometric Framework Beginning with Sjaastad (1962), economists have argued that an individual changes location within a country if the discounted net gain from moving is positive. In other words,
PM, = f ( G , ) ,
(1)
where PM,is the probability that the individual moves in time period t, and G, is the discounted net gain from moving. G, can be written as follows:
G,
=
Y,! - Y , - C,,
127
Internal Migration of U.S. Immigrants
where YT is the present value of the expected real income stream if the individual migrates in time period t, Y, is the present value of the expected real income stream in the current location calculated at time t, and C, are the costs of migration. In order to study the determinants of the probability of moving, those variables that measure the discounted net return from moving must be identified. Greenwood’s 1975 survey of the literature on geographic mobility and articles on this subject published after 1975 (e.g., Bartel 1979; and Fields 1979) show that economists have used information on the individual’s characteristics and the characteristics of the area in which he or she resides at the beginning of the period under study as proxies for the components of G , . We follow this procedure in specifying a model of the determinants of the 1975-80 inter-SMSA movement of immigrants who arrived in the United States between 1965 and 1974. Since our sample is obtained from the Public Use Sample of the 1980 Census of Population, information on personal characteristics as of 1975 is rather limited. We do have three variables in this category: (1) the individual’s age in 1975 (AGE) (2) his education in 1980 (EDUC); and (3) whether he reports the ability to speak English well or very well in 1980 (SPEAKS). Although education is measured in 1980, since the individuals in our sample were at least 22 years old at the time of immigration, it is very unlikely that educational attainment changed in any systematic way between 1975 and 1980. Similarly, SPEAKS is measured in 1980 and, if anything, is an overestimate of some individuals’ true ability to speak English in 1975, so that its estimated effect on migration is an understatement of the true effect. We expect age to have a negative effect on the probability of changing SMSAs since the time period over which to capture the discounted returns from migration shrinks as the individual ages. Education should have a positive effect since, as it has been argued in previous work, the more educated individual is better able to adapt to new locations and is more efficient in searching for jobs in other locations. SPEAKS should have a positive effect on internal migration since the immigrants who are more facile with the English language will have better information on opportunities throughout the country and will also be better able to adapt to a new location. Five variables are used to describe the SMSA in which the individual resided in 1975. They are (1) total population of the SMSA in 1975 (TPOP), (2) the 1975 unemployment rate in the SMSA (UNRATE), (3) the logarithm of the average wage in the city (LNWAGE), (4) the level of welfare benefits in the SMSA in 1975 as measured by the logarithm of the average monthly general assistance payment per recipient (LNGEN), and (5) the proportion of the SMSA’s population in 1975 that is foreign born and of the same ethnicity as the immigrant being studied (PFOR) (in other words, for an Asian immigrant, PFOR is coded as the percentage of the SMSA’s 1975 population that was born in Asia, for a Mexican immigrant, PFOR is the percentage of the SMSA’s 1975 population that was born in Mexico, etc.).
128
Ann P. Bartel and Marianne J. Koch
TPOP is obtained from the Statistical Abstract of the United States, UNRATE is from the State and Metropolitan Area Databook, LNWAGE is calculated from the data in the Public Use Sample, LNGEN is from the Public Assistance Statistics, and PFOR is calculated from the published volumes of the 1980 Census of Populati~n.~ TPOP is expected to have a negative sign since population acts as a measure of job opportunities and general economic activity. UNRATE should have a positive sign because, as the probability of finding a job in the SMSA decreases, out-migration should occur. LNWAGE and LNGEN are predicted to have negative signs since they measure attractive characteristics of the city. (LNGEN is likely to be correlated with the level of social services in the city.) Finally, the location of fellow countrymen has been shown to be a significant determinant of the settlement patterns of U.S. immigrants (see Bartel 1989; and Dunlevy 1980). We would then expect to see a negative effect of PFOR on the probability of leaving the 1975 SMSA; this effect should be smaller for immigrants who arrived between 1965 and 1969 than for those arriving later since time spent in the United States should weaken the attachment to fellow countrymen.
4.2.2 Logit Results Equation (1) is estimated on the sample of male immigrants who arrived in the United States between 1965 and 1974, who reported their 1975 and 1980 U.S. residences as one of the fifty-three SMSAs that were defined in section 4.1, and who were between the ages of 22 and 54 when they arrived in the United States. Columns 1-4 in table 4.4 report the results of estimating equation (1) on the two immigrant cohorts. The logit technique is used to estimate these equations. Among the personal characteristics, EDUC and AGE have the hypothesized signs, with the more educated and the younger immigrants being significantly more likely to change SMSAs between 1975 and 1980. Two ethnic dummy variables, ASIAN and CSA (Central and South Americans), are used to compare the migration rates of these groups to the excluded group, the Europeans. In columns 2 and 4, these dummy variables are interacted with PFOR to capture differences in responsiveness to the location of fellow countrymen. We find that, even when the other personal characteristics are included in the equation, the Asians who arrived in the 1970-74 wave are still significantly more mobile than the other immigrants who arrived during that time interval. The variables that measure relative economic opportunities in the cities (UNRATE, LNWAGE, and LNGEN) do not have the effects that we had predicted. In only one case, UNRATE for the 1970-74 cohort, do we find a significant effect. The result that stands out in table 4.4 is that for both cohorts PFOR is negative and significant. As predicted, the immigrants are most likely to move from those cities where their fellow countrymen constitute a very small percentage of the total population. This is precisely what Baker and North observed for the Indochinese refugees who did not choose their 1975 locations
129
Internal Migration of U.S. Immigrants Probability of Changing SMSAs between 1 9 5 and 1980 (&valuesin parentheses)
lsble 4.4 (1)
(2)
1965-69 Immigrants (aged 32-64 in 1980) EDUC AGE SPEAKS
.09 (4.35) - .03 (-2.57) - .01 ( - .05)
UNRATE LNWAGE LNGEN TPOP PFOR ASIAN CSA
.02 (.49) - 1.53 (-1.17) .21 (.62) - .37 (-1.06) - 6.45 ( - 2.27) .13 (.49) - .03 (-.14)
PFOR*ASIAN
PFOR*CSA
N
1,611
.10 (4.64) - .03 ( - 2.63) - .08 ( - .40) .02 (.45) - 1.67 (-1.26) .03
(.W - .37
(-1.04) 13.33 (1.43) .82 (1.54) .10 (1.95) - 10.71 (-.86) -23.38 ( - 2.38) 1.611
(3)
(4)
1970-74 Immigrants (aged 27-59 in 1980)
(5)
(6) White
(7) Ethnic
Natives Natives Natives (aged 27- (aged 27(aged 2764 in 1980) 64 in 1980) 64 in 1980)
.09 (5.07) - .02 (-2.34) - .ll ( - .65) .06 (1.62) - 1.03 ( - .94) .20 (.67) - .55 ( - 1.91) -6.94 ( - 2.41) .48 (2.10) .02 (. 12)
.09 (5.11) - .02 (-2.30) - .I2 ( - .71) .of5 (1.53) -1.14 (-1.00) .15 (.51) - .56 (-1.88) 4.52 (.50) .97 (1.95) .61 (1.25) - 9.42 ( - .80) - 13.25 ( - 1.40)
.15 (6.09) - .03 (-4.24) .17 (.98) .13 (2.95) 2.65 (2.68) - 1.12 ( - 3.65) .12 (4.34) - .45 ( - .23) - .57 ( - 2.52) - .30 (-1.30)
2,049
2,049
2,824
.21 (4.58) - .01 ( - .55) .97 (2.04) .10 (1.17) 2.72 (1.48) - 1.05 ( - 2.22) .02 033) 6.42 (2.11)
676
.ll (3.93) - .05 (-4.79) .07 (.39) .09 (1.51) 1.96 (1.59) - .86 ( - 1.90) .20 (4.52) -7.01 (-1.95) - .23 (-1.11)
2,148
on their own. We had expected to see a weaker effect of this variable for the immigrants who had spent more time in the United States as of 1975 (i.e., the 1965-69 arrivals), but their attachment to their fellow countrymen is just as strong. In columns 2 and 4, we interact PFOR with the ethnic dummy variables, ASIAN and CSA, in order to see if our ethnic groups are equally responsive to this variable. For both cohorts, the results show that only the Central and South Americans changed SMSAs in response to the presence of fellow countrymen. If the Central and South American immigrants are indeed moving from those cities where PFOR is low, then we would expect to observe an increase between 1975 and 1980 in the mean values of PFOR for those individuals who move. In rows 1, 3, and 5 of table 4.5, we show the ratio of mean PFOR in 1980 to its mean in 1975 separately for the movers and the stayers in our sample. The table also reports, for the movers only, the ratio of the 1980 mean
130
Ann P. Bartel and Marianne J. Koch
value of PFOR based on their 1975 locations, divided by the 1975 mean value. In other words, this ratio, shown in rows 2 , 4, and 6, describes how the 1975 locations changed between 1975 and 1980. Table 4.5 confirms that PFOR increased more between 1975 and 1980 for the Central and South American movers than the stayers; the immigrants in this ethnic group who changed locations in the United States experienced an increase in the concentration of fellow countrymen, and, as shown by comparing rows 3 and 4, this increase could only have occurred by changing SMSAs. Although these movers are initially in cities with lower values of PFOR compared to the stayers, by 1980 the gap has narrowed.6Consistent with the regressions in table 4.4, the results for the other ethnic groups in table 4.5 show that a change of SMSA does not increase the concentration of fellow countrymen; in fact, for the Europeans, there is a decrease for both cohorts. In columns 5-7 of table 4.4, we report the results of estimating equation (1) on a sample of natives aged 27-65 in 1980. Column 5 includes white natives and natives who report themselves of either Asian, Cuban, Mexican, or other Hispanic ancestry. The purpose of this analysis is to compare the behavior of the immigrants with that of individuals who were born and raised in the United States and presumably have better information about the country.’ Although EDUC and AGE have the same effects for the natives as they did for the immigrants, we do observe different effects for the city characteristics. LNGEN is now negative and significant, indicating a greater sensitivity (or awareness?) of the natives to welfare benefits and social services. But LNWAGE actually has the wrong sign! The most interesting finding is that the white natives (col. 6) are more likely to move from those cities with a high concentration of foreign born (in col. 6, PFOR refers to alf ethnic groups), while the ethnic natives, like the immigrants, prefer to stay in cities with a high concentration of individuals from their country of ancestry. Table 4.5
A Comparison of 1975 and 1980 Mean Values of PFOR for Movers and Stayers ~
1965-69 Cohort
1970-74 Cohort
Movers
Stayers
Movers
Stayers
Asians: 1. PFOR8O/PFOR75 2. P F O R ~ O(in 75 c i t y ) / ~ ~ 0 ~ 7 5
1.46 1.64
1.67
1.76 1.69
1.72
Central and South Americans: 3. PFOR80iPFOR75 4. PFOR80 (in 75 CitY)/PFOR75
1.63 1.28
1.19
1 .S3 1.27
1.32
Europeans: 5 . PFOR80/PFOR75 6. PFOR80 (in 75 CitY)/PFOR 75
.83 .97
1.10
.89 1.10
1.10
131
Internal Migration of U.S. Immigrants
4.3 The Effect of Internal Migration on Wages In this section of the paper, we study the effects of a change in SMSA between 1975 and 1980 on the 1980 wage rates of the immigrants who arrived between 1965 and 1974. We also compare the results to similar wage equations that are estimated on our sample of natives. Previous work on migration has not conclusively established that migrants necessarily experience an increase in wages following a change in location.8 We examine the effect of internal migration on wages by estimating a log(wage) equation on 1980 data that includes the standard variables such as education, experience, experience squared, marital status, health status, and a vector of ethnicity dummies. We then add a variable called MOVE, which equals one if the individual changed SMSAs between 1975 and 1980. Since the various ethnic groups are pooled in our sample, we also interact MOVE with our two ethnic dummy variables and create MOVEASN and MOVECSA; the comparison group is, therefore, the Europeans. We also distinguish two categories of moves, moving to an SMSA where PFOR is higher than it was in the origin SMSA (MOVEUP) and moving to an SMSA where PFOR is lower than it was in the origin SMSA (MOVEDOWN). We expect the coefficient on MOVEUP to be less positive than the coefficient on MOVEDOWN for two reasons. First, since immigrants prefer to live with fellow countrymen, compensating differentials will reduce the return to MOVEUP. Second, crowding effects should depress the wages of immigrants in cities with large foreign-born populations. The results of estimating the wage equation are shown in table 4.6. Only the coefficients on the mobility variables are shown; all the other variables had the usual signs. The equations were estimated separately for the two immigrant cohorts in order to see if time spent in the United States influences the returns to internal migration. In columns 1 and 4, we see that MOVE is insignificant for both cohorts but that it is positive fro the earlier immigrants. Distinguishing the returns to migration for the different ethnic groups is important, as shown in column 2. Among the 1965-69 arrivals, only the Europeans experienced a significant increase in wages when they changed SMSAs between 1975 and 1980. The hypothesis regarding the relative effects of MOVEUP and MOVEDOWN is confirmed for this cohort. MOVEDOWN has a positive and significant coefficient, while the coefficient on MOVEUP is insignificant. The fact that only 9% of the movers are in the MOVEDOWN category suggests that this small group is unique in its ability to leave ethnic networks and move to areas where their skills are more highly rewarded. In the case of the 1970-74 arrivals, however, ethnicity plays no role; MOVE is insignificant for all the groups. Finally, columns 7 and 8 show the effect of internal migration on the wages of natives. The results here conform to the findings of previous research, namely, an insignificant effect of geographic mobility on wage rates.
132
Ann P. Bartel and Marianne J. Koch Effects of Changing SMSA between 1975 and 1980 on Log (1980 wage rate) of Immigrants and Natives
Table 4.6
1965-69 Arrivals (1)
.06 (.94)
MOVE
MOVECSA MOVEASN
(2)
(3)
(4) - .01 (-.26)
.34 (2.52) - .38 ( - 2.30) - .34 ( - 1.92)
(5)
MOVEDOWN
1,400
1,400
1,400
Natives
(6)
.08 (.60) -.I6 ( - .99) - .07 ( - .46)
(7)
(8)
.03 (S9)
.03 (.27) .I1 (.87) -.I8 (-1.26)
2,387
2,387
- .01 ( - .24)
.02 (.31) .29 (1.80)
MOVEUP
N
1970-74 Arrivals
- .02 (-.]I)
1,809
1,809
1,809
Note: Other variables included in these equations-are education, experience, experience squared, marital status, health status, and ethnicity.
4.4
Summary
The major finding of this paper is that, although recent immigrants to the United States move between SMSAs at a rate that is comparable to or in some cases exceeds that of ethnic natives, there is little systematic evidence that this immigrant population becomes more geographically dispersed as time in the United States elapses. The only groups for whom we observed evidence of dispersion between 1975 and 1980 were the Indians, Japanese, and Koreans. The logit analysis of the determinants of migration between SMSAs showed that the more educated and younger immigrants are most likely to move. The variables that measure relative economic opportunities in the cities (unemployment rate, area wages, and welfare benefits) did not have significant effects on the probability that an immigrant changes SMSAs. For the Central and South American immigrants, we found that the concentration of fellow countrymen in the city was an important determinant of migration. These immigrants changed SMSAs in order to move to cities with higher concentrations of Central and South Americans. On the other hand, the high mobility rates of the Asian immigrants were unrelated to the percentages of Asians in the various cities, while the Europeans who moved actually experienced a decrease in the concentration of fellow countrymen. Obviously, there are important differences in the characteristics of these ethnic groups that could explain their different behavior, but with our data we have been unable to measure these factors. While it is difficult to explain satisfactorily why immigrants change locations in the United States, we can conclude that whatever migration does oc-
133
Internal Migration of U.S. Immigrants
cur is unlikely to lead to a substantial increase in the geographic dispersion of newer immigrants in the United States. It is important to note that this conclusion presumes that, if geographic dispersion occurs at all, it takes place within the first fifteen years of experience in the United States. At best, treating our cross section as a pseudo-panel, we have been able to observe the new immigrants only fifteen years after arrival in the United States. If we rely on our actual panel data, then our conclusion is extrapolated from only five years of data. Hence, it is possible that dispersion may occur but that we have been unable to observe it. An even larger question, of course, remains unanswered by our study. Is the lack of geographic dispersion a problem? On the one hand, some sociologists have argued that this will inhibit the process of assimilation. But an equally valid argument could be that ethnic enclaves provide the financial and emotional support necessary for immigrants to achieve success in their new country.
Notes 1. Adding more SMSAs increases the costs of data collection because our logit analysis reported in the next section requires information on city characteristics. The rate at which these costs rise far exceeds the rate at which the size of the immigrant sample increases. 2. The index is defined as S;, where S, is the proportion of individuals in the ith r=l SMSA. 3. The 1975-79 arrivals from Vietnam are more dispersed than the earlier cohorts because of the special placement program described earlier. 4.The samples of natives that are of Asian or Central and South American ethnicity each represent 25% of the actual number of observations in the Public Use B Sample. The non-Hispanic whites, or those of European ethnicity, are 1% of the actual number in the sample. 5. Note that LNWAGE is measured as of 1980, so we are implicitly assuming that high-wage cities in 1980 were also high-wage cities in 1975; in other words, that there was no systematic change in the ranking of cities according to wage level during this time period. 6. The 1975 ratio of PFOR for Central and South American movers relative to PFOR for Central and South American stayers is .55 for the 1965-69 cohort and .73 for the 1970-74 cohort. The 1980 ratio is .76 for the 1965-69 group and .85 for the 1970-74 group. 7. Regarding the construction of the native sample, see n. 4 above. 8. Greenwood (1975) shows that, while many studies have found a positive return to migration, others have been unable to support this conclusion. Bartel (1979) found that the wage gains from migration are dependent on the nature of the move (i.e., whether it is accompanied by a quit, a layoff, or an internal transfer within a company) and the age of the migrant. In particular, only those who were transferred by their companies experienced an increase in wages, which was significant for individuals under age 45.
2
134
Ann P. Bartel and Marianne J. Koch
References Baker, R. P., and D. S. North. 1984. The 1975 refugees: Theirjrstjveyears in America. Washington, D.C.: New Transcentury Foundation. Bartel, A. P. 1979. The migration decision: What role does job mobility play?American Economic Review 69(December):775-86. . 1989. Where do the new U.S. immigrants live? Journal of Labor Economics 7(0ctober):371-91. Dunlevy, J. A. 1980. Nineteenth-century European immigration to the United States: Intended versus lifetime settlement patterns. Economic Development and Cultural Change 29(0ctober):77-90. Fields, G. 1979. Place to place migration. Review of Economics and Statistics 6 1(February):2 1-32. Greenwood, M. J. 1975. Research on internal migration in the United States. Journal of Economic Liferafure 13(June):397-433. Massey, D. S. 1981. Dimensions of the new immigration to the United States and the prospects for assimilation. Annual Review of Sociology 757-85. Sjaastad, L. A. 1962. The costs and returns of human migration. Journal of Political Economy 70, suppl. (October):80-93.
5
Migration, Ethnicity, and Labor Force Activity Marta Tienda and Franklin D. Wilson
5.1 Introduction In this paper, we investigate how internal geographic mobility and the ethnic segmentation of jobs influence the employment and earnings of black, Mexican, Puerto Rican, Cuban, and American Indian men. Many of the men in our sample from the 1980 Census of Population are foreign immigrants, and the migration of minority men has parallels with the immigration of foreigners. Specific questions that guide our analysis are, (1) Do the labor force participation and unemployment of minority workers depend on whether they move between or within high- or low-ethnic-density labor markets? (2) Do ethnic job queues influence labor force behavior and unemployment risks? (3) Does migration within or between high- and low-ethnic-density markets and ethnic job queues also influence remuneration patterns? Evidence that hiring queues and remuneration depend either on spatial concentration of minority groups or on preferential hiring according to race and national origin is crucial for establishing direct links between the declining labor market status of Puerto Ricans (Bean and Tienda 1987; Tienda 1989); the economic bifurcation of the black (Wilson 1987), Puerto Rican (Tienda and Jensen 1988), and Native American (Sandefur 1986) populations; the slow economic progress of Marta Tienda is professor of sociology, University of Chicago, and associate director, Population Research Center of NORC and University of Chicago. Franklin Wilson is professor and chair of the department of sociology at the University of Wisconsin. This research was supported by grants from the Ford Foundation and the Department of Health and Human Services to the Institute for Research on Poverty. Computational support was furnished by a grant from the Center for Population Research of the National Institute for Child Health and Human Services and Human Development to the Center for Demography and Ecology (HD-05876). The authors gratefully acknowledge technical assistance from Sarah Rudolph and Gary Geisserer, programming assistance from Cheryl Knobeloch, and research assistance from Elaine Fielding, Jaewoong Shim, Brett Brown, and Albert0 Martini. The opinions expressed here are those of the authors, not of the sponsoring institutions.
135
136
Marta Tienda and Franklin D. Wilson
Mexican men (Portes and Bach 1985; Bean and Tienda 1987); and the rapid economic progress of Cuban men (Portes and Bach 1985). We focus on the relation between the economic outcomes of migration and two new variables, area ethnic density, which indexes the concentration and distribution of an ethnic group in a particular labor market, and the ethnic type of an individual’s industry and occupation. We find no evidence that spatial assimilation will necessarily promote socioeconomic integration and thereby reduce ethnic labor market inequities, and we find some indication that migration is associated with lower labor force activity, possibly because of the disruptive effects of the migration process per se. 5.1.1 Migration Types Depending on their direction, composition, and social underpinnings, migrant streams can promote ethnic consolidation or disintegration. For example, migration may strengthen ethnic solidarity in work and school domains by changing the racial/ethnic density of communities or institutional settings. Such outcomes also depend on the existence of ethnic labor market niches, the extent of school and neighborhood segregation, and the existence of ethnic power bases. Thus, it is conceivable that the benefits accruing to migrants who participate in flows leading to greater ethnic spatial concentration differ from those that produce ethnic spatial dispersion. Geographic moves involving dispersion could improve the employment and earnings prospects of migrants if market factors (i.e., the demand for migrants’ skills) rather than cultural and social factors (i.e., ethnic markers) dominate decisions to move and also influence choice of destination. Conversely, concentrated migration flows, which frequently are motivated by noneconomic considerations (such as the desire to reside in closer proximity to relatives and friends of like ethnicity), may render migrants less satisfactory employment outcomes, at least over the short run. This would follow especially if the reinforcement of cultural and ethnic bonds through concentrated flows involves a trade-off between psychic and economic rewards. However, if proximity to friends and relatives enables migrants to secure employment, then the gains from participation in concentrated flows might be greater than the economic penalties associated with ethnic crowding. Similarly, if the ability of dispersed migrants to secure a better-paying job is jeopardized by the absence of social networks in low-ethnic-density markets, then the potential economic gains from dispersed migration flows will be reduced. 5.1.2 Ethnic Hiring Queues The significance of geographic mobility for the labor market stratification of minority workers depends not only on the employment opportunities afforded movers but also on how individual ethnic traits circumscribe choices, are evaluated in the market place, and are used to organize the labor market. Specifically, if national origin is used as a criterion to define and maintain job
137
Migration, Ethnicity, and Labor Force Activity
queues-as demonstrated by previous research (Lieberson 1980; Hechter 1978)-then the economic costs and benefits of migration will derive not only from opportunities to interact with members of like ethnicity but also from the role of national origin in channeling minority workers to particular categories of jobs.' The viability of ethnic hiring queues, however, is related to ethnic spatial concentration patterns. Lieberson (1980, chap. 10) argues that the connection between the ethnic composition of labor markets and ethnic job queues reflects differences in opportunities over time and the force of history in stratifying the U.S. labor force according to race and national origin. He claims, furthermore, that a discriminatory hiring queue results when, given the existence of a queue, employers preferentially hire workers on the basis of ethnic traits rather than market skills. Two aspects of Lieberson's queuing premises have implications for our concerns. First, he argues that the job configuration of groups will vary in accordance with its share of the labor force in a given market.2 Second, he claims that, because of the existence of ethnic hiring queues, shifts in unemployment would be highest for the group(s) at the bottom of the queue during periods of rising ~nemployment.~ Applying these arguments about ethnic residential and ethnic occupational segmentation to the relation between migration and employment, we hypothesize that the market experiences of Mexican, Puerto Rican, Cuban, black, and American Indian men are influenced by their differential participation in concentrated versus dispersed migration streams and by their unequal placement in a job queue.
5.2 Data and Methods Our statistical analysis uses the 5% Public Use Microdata Samples (PUMS) of the 1980 Census. We limited our sample to men aged 25-64 who had valid responses to the migration questions and who self-reported their race or national origin as Mexican, Puerto Rican, Cuban, black, or American Indian.4 Restricting the lower end of the age distribution to 25 rather than 16 ensures that our sample of respondents was eligible to participate in the labor force prior to the beginning of the migration interval (1975). Additional sample restrictions purged our results from status changes that are systematically associated with migration probabilities. For this purpose, we excluded individuals who (1) never worked or were out of the labor force continuously during the migration interval, (2) were enrolled in school or in the military in either 1975 or 1980, or (3) resided outside the United States in 1975.5 Combined, these restrictions and additional random sampling of the Mexican and black populations (for computational efficiency) yielded the following subsample N's: 6,076 Mexicans, 6,630 Puerto Ricans, 4,134 Cubans, 5,827 blacks, and 5,810 American Indians.6 As can be seen in table 5.1, sizable proportions of our samples are foreign born: 36 percent of Mexicans, 93
138
Marta Tienda and Franklin D. Wilson
Table 5.1
Means and Standard Deviations of Selected Variables Included in Regression Analysis (standard deviations in parentheses)
Individual characteristics: Education: % Finishing high school % Finishing less than
high school Experience (years) % Good English ability % Foreign born % Work disabled
Weeks workeda Average hours’ Family status: % Household head % Children under 6 % Married
Labor market characteristics: Region: % North Central % south
%West % Metro residence
Area wage rate ($) Area unemployment rate
Mexican
Puerto Rican
Cuban
Black
American Indian
32.5 (46.9) 63.3 (48.2) 24.8 (12.5) 78.0 (41.5) 36.0 (48.0) 6.8 (25.1) 46.2 (11.0) 41.7 (10.2)
34.2 (47.5) 62.0 (48.5) 24.3 (11.7) 81.3 (39.0) 77.5 (41.8) 8.7 (28. I ) 46.7 (10.8) 40.0 (10.0)
40.7 (49.1) 42.1 (49.4) 29.2 (11.6) 62.6 (48.4) 93.2 (25.1) 5.2 (22.2) 48.0 (9.2) 42.6 (11.1)
45.4 (49.8) 47.8 (50.0) 25.1 (12.9) 99.8 (4.3) 3.6 (18.7) 10.7 (30.9) 45.6 (11.9) 40.0 (11.0)
51.2 (50.0) 41.4 (49.3) 23.1 (11.9) 97.1 (16.8) 2.5 (15.5) 14.5 (35.2) 43.2 (14.0) 42.0 (12.1)
85.2 (35.5) 34.3 (47.5) 81.3 (39.0)
79.6 (40.3) 26.9 (44.4) 71.7 (45.0)
89.6 (30.6) 14.7 (35.4) 81.8 (38.6)
77.8 (41.5) 20.0 (40.0) 64.5 (47.9)
81.6 (38.7) 25.1 (43.4) 72.0 (44.9)
10.6 (30.8) 35.0 (47.7) 53.6 (49.9) 86.0 (34.7) 7.22 ( .97) 6.19 ( I .98)
10.8 (3 1.O) 8.2 (27.4) 6.6 (24.9) 97.0 (16.9) 7.88 ~72) 6.69 (1.23)
4.1 (19.8) 61.8 (48.6) 8.5 (27.8) 98.2 (13.3) 7.38 (.61) 5.66 (1.30)
20.8 (40.6) 52.3 (50.0) 7.9 (27.0) 82.4 (38.1) 7.29 (1.01) 6.53 (1.92)
16.9 (37.5) 29.1 (45.4) 47.7 (50.0) 55.9 (49.7) 7.14 (1.39) 7.02 (2.51)
Source: 1980 5% A Sample Public Use Micro-data Samples, migrant subsample. a Based on subsample with positive annual earnings in 1979.
139
Migration, Ethnicity, and Labor Force Activity
percent of Cubans, and 78 percent of Puerto Ricans (where by “foreign born” we mean born on the island). Thus, for these minority groups, we analyze secondary choice of location of large numbers of immigrants. Restricting these samples to individuals with some wage and salary income in 1979 for the earnings analyses further reduced the population samples from 6 (Mexicans and Cubans) to 11 percent (Puerto Ricans). 5.2.1 Variables Our analyses focus on the relation among three variables: migration type (i.e., whether moves took place within or between high- and low-ethnicdensity types); ethnic j o b segmentation (i.e., whether jobs were ethnic typed, Anglo typed, or not ethnically differentiated); and labor market outcomes (specifically, whether respondents were in the labor force or unemployed in 1980 and their 1979 [logged] annual earnings). Migrants are defined as persons who changed residence during the five years prior to the census. We chose standard metropolitan statistical areas (SMSAs) and nonmetropolitan county groups (rather than states) to define migration status. Our distinction between high- and low-ethnic-density labor markets is derived from an analysis of both the ethnic composition of labor markets and the distribution of each ethnic group among them. Procedures used to classify labor markets (N = 414) into high- and low-ethnic-density areas are detailed in Appendix A. Briefly, a labor market area was defined as high ethnic density for a given reference group if the group was overrepresented relative to its share of the total population based on standardized (z) scores. Our hypotheses about the influence of geographic movement in altering the social environments and economic opportunities of migrants emphasize the direction of the flows. Ethnic residential dispersion involves moves from high- to low-ethnic-density labor market areas; flows from low to high ethnic density produce concentration; and flows within low- or high-ethnic-density areas, labeled intradensity moves, involve no changes in the ethnic composition of labor markets from the perspective of individual migrants.’ However, these moves usually alter economic opportunities. Since previous research (Tienda and Lii 1987) has shown that ethnic spatial concentration directly influences socioeconomic outcomes above and beyond productivity characteristics, we differentiate between moves within high- and low-concentration areas to detect the effects of concentration among intradensity movers. All totaled, we classified individuals into five categories according to whether they migrated and subsequently distinguished among those who participate in dispersed, concentrated, and intradensity moves within high- andlor within low-ethnic-concentration areas. Operationalizing our notion of ethnic job segmentation was more complicated than the coding of migration types. Because the statistical procedures we used are detailed in Appendix B, we only highlight the logic used in distin-
140
Marta Tienda and Franklin D. Wilson
guishing among workers classified in ethnic-typed, Anglo-typed, and nontyped jobs. We began with a thirty-cell-matrix representing a two-way classification of six industry sectors by five occupation groups using 1970 Census data.8 Sector-by-occupation matrices were computed for each of the five ethnic groups and non-Hispanic whites. Based on the results of a log-linear analysis, we classified job cells according to whether each ethnic group was overrepresented (ethnic typed), underrepresented (Anglo typed), or approximately equally represented (nontyped) relative to non-Hispanic whites and net of group differences in education and age composition. These results, summarized in Appendix tables 5B. 1 and 5B.2 below, are substantively informative, but we do not dwell on them in the interest of brevity. 5.2.2
Modeling
Our conceptualization of the employment experiences of minority men integrates two structural attributes of labor markets-the ethnic segmentation of jobs and the ethnic composition of markets-and assesses their influence on labor force participation, unemployment, and (logged) annual earnings. The segmentation of jobs along ethnic lines requires a critical mass of minority workers; hence, we hypothesize that labor market experiences of minority men may differ in high- and low-density labor markets. Accordingly, our empirical model, which assumes that both geographic mobility and the ethnic labeling of jobs influence the labor force participation and unemployment prospects of minority men, takes the form:9 (1)
PR(LF), =
CL
+ PI MI + y a P , + Z , + e , ,
where LF, = labor force status of individual i; 1 = in, 0 = out for the participation equation; 1 = unemployed, 0 = employed for the unemployment equation. M j = migration type, and k = 4, 3, 2, 1, and 0 represent whether individuals participated in dispersed, concentrated, intradensity low, or intradensity high flows, or were nonmigrants, respectively; P , = ethnic job segments, and k = 2, 1, and 0 for ethnic-typed, Anglotyped, and nontyped segments, respectively; z, = a vector of controls; e, = random disturbances.
The predicted effects of ethnic job segmentation are informed by economic logic as well as sociological insights about the significance of race and ethnicity in demarcating boundaries for social interaction. If the existence of ethnic hiring queues “reserves” jobs for minority workers (as in ethnic niches or enclaves), then y2 > 0 in the participation equation, and y2 < 0 in the unemployment equation. This result would show the influence of social (ethnicity) forces in defining paths of labor market activity for minority workers (Portes
141
Migration, Ethnicity, and Labor Force Activity
and Bach 1985).1° However, if workers destined for Anglo-typed jobs are more likely to be in the labor force than their (statistical) counterparts identified with nondifferentiated job categories, the y1 > 0 in the participation equation, and the y, < 0 in the unemployment equation. In this instance, factors other than ethnicity will govern the employment prospects of minority men. Our predictions about the influence of migration types on employment outcomes are informed theoretically by research and writing on the socioeconomic significance of ethnic spatial concentration (Tienda and Lii 1987). Extending ideas about ethnic density to geographic movement, we expect nonzero effects associated with participation in concentrated (p, # 0) and dispersed (p, # 0) migration flows, but the direction of these effects is an empirical question. For example, if p, > 0 in the participation equation (p, < 0 in the unemployment equation), then dispersed migration would appear to promote the labor market assimilation of minority men. A negative value of p, in the participation equation (positive in the unemployment model) would indicate that the investment properties of the migration decision either require a long time to mature or else may depend on the ethnic hiring queues at destination. Alternatively, a positive coefficient for p, in the participation equation (negative in the unemployment equation) would indicate that concentrated labor flows promote socioeconomic assimilation in the context of increasing ethnic pluralism. While not denying the importance of supply factors in determining minority labor market outcomes, this result is consistent with the premises of queuing and overflow perspectives of labor market dynamics (see nn. 3, 4). Negative returns to concentrated migration (p, < 0 in the participation equation and p, > 0 in the unemployment equation) suggest that, over the period considered, the disruptive aspects of the investment decision offset the investment gains from the decision. Intradensity moves presumably represent investment decisions in response to better employment prospects; hence, we expect p, and p2 > 0. Our assessment of the earnings consequences of migration in the context of ethnic residential concentration and ethnic job segmentation assumes the following form:
(2)
Y, =
OL
+ p j M j + ye P , + Z, + e , .
This model is analogous to our participation and unemployment equations, although the vector, Ziis not identical (see the definitions following eq. [ 11). Interpretationsof the ethnic queuing effects reflect our hypotheses about the underlying stratifying mechanisms. If minority workers destined for ethnically typed jobs gain financially compared to those destined for nondifferentiated jobs, then y2 > 0. This result is highly plausible in contexts where minorities are overrepresented among the self-employed (Lieberson 1980; Portes and Bach 1985). However, if minority workers destined for Anglotyped jobs earn more than their (statistical) counterparts holding nontyped jobs, then y1 > 0. These findings would indicate that the more desirable jobs,
142
Marta Tienda and Franklin D. Wilson
usually dominated by whites, hold the key to reducing wage disparities between minority and nonminority men. Our interpretation of migration effects will be similar to those discussed above. We introduce in all models a set of controls for individual and labor market characteristics known to influence labor market outcomes. Appendix C summarizes all variables included in the vector Z,, providing a brief operational description of the controls as well as the key dependent and independent variables. I 1 It also indicates whether the control variables were included in the labor supply equations, the earnings equations, or both. 5.2.3 Techniques Because the dependent variables in the labor supply equation, labor force participation and unemployment, are dichotomous, we use a maximum likelihood logistic regression to estimate the models. For ease of interpretation, we report only the partial derivatives of the probabilities using the procedure derived by Petersen (1985). The (log) earnings equations are estimated using OLS regression. 5.2.4 Evidence Before presenting results from the regressions, table 5.1 provides background information about our samples of minority men. The disadvantaged labor market status of Mexican and Puerto Rican men reflects their low stocks of human capital. Fully two-thirds of mature Mexican and Puerto Rican men had not completed high school, compared to 48 percent of black men and 41 and 42 percent of Indian and Cuban men, respectively. At the other extreme, 17 percent of mature Cuban men and approximately 7 percent of black and American Indian men were college graduates, compared to roughly 4 percent of Mexican and Puerto Rican men. Lack of proficiency in English is a problem confined largely to Hispanics. That only 63 percent of Cubans reported good to excellent proficiency in English reflects the predominantly foreign origins of our sample of mature men. Although roughly three-fourths of Puerto Rican men were born on the island of Puerto Rico, where Spanish is the predominant language, English is taught in the schools, and bilingualism is quite pervasive. Only 78 percent of Mexican men reported good to excellent proficiency in English, yet only 36 percent were immigrants. This results partly because of the higher rates of Spanish retention among the native born (Nelson and Tienda 1985) and partly because of the lower levels of schooling completed by mature Mexican men. The incidence of work limiting disability is highest among black and American Indian men and lowest for Cuban and Mexican men. Poor health is a corollary of the high poverty rates characteristic of these groups, particularly those isolated on remote reservations (Sandefur 1986) and inner-city ghettos (Wilson 1987). Not surprisingly, average weeks worked in 1979 were lowest
143
Migration, Ethnicity, and Labor Force Activity
for these two groups. Cuban men reported the highest average weeks worked in 1979 and the longest average work week. Our samples are further differentiated by family and household characteristics. Nearly four-fifths of Mexican and Cuban men were married, compared to less than 65 percent of black men and only 72 percent of Puerto Rican and American Indian men. Cuban and Mexican men were most likely, and black and Puerto Rican men least likely, to identify as households heads. Reflecting their younger age composition and higher fertility, approximately one-third of our Mexican sample had young children at home, compared to one-fourth of Puerto Rican and black men and less than 15 percent of Cuban men. Finally, minority men are differentiated by geographic characteristics. In 1980, less than 1 percent of mature Mexican men and 6 percent of American Indian men resided in the Northeast, compared to nearly three-fourths of mature Puerto Rican men. Mexicans were disproportionately concentrated in the South and West, Cubans in the South (Florida) and Northeast, while blacks were underrepresented in the West. Hispanics (Cubans and Puerto Ricans in particular) were largely a metropolitan population; in contrast, less than 60 percent of American Indian men reported living in metropolitan areas in 1980. These residential profiles have direct implications for employment and income opportunities. American Indians resided in labor markets characterized by the highest unemployment rates, followed by Puerto Rican men, while Cuban men confronted the lowest average unemployment rates in 1980. American Indian men faced the lowest average wage rates, largely because they were disproportionately located in nonmetropolitan and reservation areas, while the highest wages corresponded to labor markets where Puerto Ricans live. That high unemployment also characterized these high-wage markets is a key piece of information for decoding the declining economic status of Puerto Ricans. Table 5.2 shows that American Indians were the most mobile minority group during the late 1970s, as nearly one in five reported having changed county groups between 1975 and 1980. At the other extreme, only 8 percent of mature black men migrated between 1975 and 1980. Among Hispanics, Cubans were the most mobile, and Puerto Ricans were slightly less mobile than Mexicans, with 10-12 percent changing labor markets during the time period. More interesting are the differences in the direction of the migrant flows. Geographic movement between high-ethnic-concentration labor markets was the modal migration type for all groups, accounting for 50-65 percent of intermarket moves by Hispanic men and 35 and 43 percent of moves by American Indian and black men, respectively. Flows between labor markets with low levels of ethnic concentration were prominent only among American Indians. In the main, these moves capture the residential mobility of the nonreservation Indian population. Although the numbers of migrants engaged in dispersed and concentrated
144
Marta Tienda and Franklin D. Wilson
Table 5.2
Descriptive Statistics for Migration Qpe and Ethnic Job Segmentation Puerto Rican
Cuban
Black
6.5 1.2 1.2 1.7 10.6
4.9 1.4 1 .o 2.7 10.0
6.8 1.3 2.3 1.8 12.3
3.4 1.2 1.5 1.8 7.9
6.7 5.8 3.3 3.0 18.8
13.0 4.1 82.9
18.6 6.4 75 .O
13.1 14.1 72.8
36.6 5.6 57.8
19.9 15.0 65.1
6,076
6,630
4,134
5,827
5,810
Mexican Migration type: Intradensity high Intradensity low Concentrated Dispersed Migrants (totals) Ethnic job segmentation? Ethnic typed Anglo typed Nontyped
N
American Indian
Source: 1980 5% A Sample Public Use Micro-data Samples, migrant subsample. Nore: All tabulations exclude recent immigrants. a Compares each group to Anglos. See App. B.
migration flows did not differ greatly within groups (Puerto Ricans being a notable exception), concentrated moves were more pervasive among Cubans and American Indians, while Mexicans and blacks became slightly more dispersed. Puerto Ricans experienced the greatest residential dispersion during the late 1970s as a result of internal migration. Having always been a metropolitan population on the U.S. mainland, for them dispersion involved moves out of New York and the Northeast in general (Bean and Tienda 1987). The last three rows of table 5.2 support the view that blacks were especially likely to be in ethnic-typed jobs. Over one-third of all mature black men in the experienced civilian labor force identified with jobs where blacks were disproportionately concentrated. Appendix table 5B. 1 below shows that in 1970 blacks were uniformly overrepresented in the lower nonmanual and in upper manual jobs in the producer, social, and personal services sector. Auxiliary tabulations (available from the authors) reveal that these patterns persisted until 1980. Based on their disproportionate concentration in manual jobs, roughly 20 percent of Puerto Rican and American Indian men identified with ethnic-typed jobs. Finally, Mexican and Cuban men were least likely to be ghettoized in “ethnic jobs” in 1980. For Mexicans, ethnic job typing largely involved agricultural activities but also included lower manual jobs in the distribution and personal service sectors. Cubans, on the other hand, were overrepresented in some nonmanual as well as manual activities, but principally in the producer and personal services sectors. l2 At the other extreme, Cubans and American Indians were the only groups for whom the share engaged in "Anglo-typed" jobs exceeded 10 percent. For both groups, these jobs involved nonmanual activities in the transformative
145
Migration, Ethnicity, and Labor Force Activity
and distributive services sectors. Clearly, employment in nontyped jobs was the model form of labor market insertion for the overwhelming share of minority men, with blacks least likely and Mexicans most likely to be so situated as of 1980. However, these are very conservative measures of ethnic job segmentation (see App. B). Had our method to establish over- and underrepresentation not included statistical adjustments for age and educational differentials between the groups, the proportion of workers allocated to “ethnic-typed” jobs would have been considerably greater. Also, had we used more liberal cut points for stipulating the “tolerable” limits of ethnic job segmentation, the numbers of industry-occupation cells designated as ethnic typed and the share of workers so-allocated would have been higher. The three dependent variables analyzed are summarized in table 5.3. Even though our sample is limited to men aged 25 to 64,the labor participation rates range from a high of 96 percent for Cubans to a low of 86 percent for American Indians. The labor force participation rate for blacks was slightly higher than that of American Indians, but this does not necessarily mean that the latter group is situated at the bottom of the employment hierarchy. Rather, this reflects the extremely limited employment opportunities on reservations (Sandefur 1986; Snipp and Sandefur 1988). Auxiliary tabulations revealed that American Indian labor force participation was significantly lower and unemployment significantly higher in areas of high ethnic concentration, while the reverse was true for blacks. Differential participation and unemployment rates among groups reflect variation in employment opportunities by residence and differential placement of groups in the hiring queue. Among Hispanics, Puerto Ricans have especially low average participation rates and high unemployment (see also Tienda 1989). The ranking of minority groups based on 1979 annual earnings reaffirms the placement of black men at the bottom, where they are accompanied by Puerto Ricans rather than American Indians. At the opposite extreme stand 1980 Labor Force Participation and Unemployment Rates and la79 Average Annual Earnings of Minority Men Aged lbenty-Five to Sixty-Four (standard deviation in parentheses)
Table 5.3
Labor force participation Unemployment Logged annual earnings Average annual earnings ($) ~~
Mexican
Puerto Rican
Cuban
Black
American Indian
93.6 (24.4) 6.6 (24.3) 9.22 ( .89) 13,342 (9,414)
91.4 (28.0) 7.5 (26.3) 9.18 (37) 12,587 (8,647)
95.6 (20.6) 3.8 (19.2) 9.43 (.82) 16,368 (13,068)
88.9 (31.4) 8.6 (28.1) 9.14 (.97) 12,585 (8,485)
86.3 (34.4) 10.9 (31.2) 9.16 (1.06) 13,938 (1 1,269)
~
Source: 1980 5% A Sample Public Use Micro-data Samples, migrant subsample.
146
Marta Tienda and Franklin D. Wilson
Cubans, with average annual earnings of $16,400 in 1979. Mexican and American Indian men are situated in an intermediate position relative to blacks and Cubans. Although the average annual earnings of American Indians may seem high, recall that this figure is based on the subset with positive earnings in 1979. Since a lower share of this group participates in the labor market, the subset with earnings is a highly selected segment of this population. Overall these descriptive statistics provide useful background information for interpreting our multivariate results, which analyze, in sequence, the probability of being in the labor force or unemployed and logged annual earnings as a function of migration type, ethnic job typing, and a vector of control variables. 5.2.5
Labor Force Participation and Unemployment
The transformed logit coefficients in table 5.4 reinforce a picture of diversity in the determination of participation decisions and unemployment risks among minority men. Considering first the results for labor force participation, one general inference is that minority men who migrated during the latter part of the 1970s were consistently less likely, or no more likely than their nonmigrant counterparts to be in the labor force. Although only nine of the twenty estimated coefficients attained statistical significance, the reliable coefficients were uniformly negative. These results appear to contradict conventional wisdom about the investment properties of migration decisions. However, it is conceivable that, for minority men, migration may require a substantial period of adjustment before the returns accrue. There is no way for us to control for the exact timing of migration during the five-year interval; hence, it is not possible to ascertain to what extent our results may be capturing the initial disorganization following migration. l 3 The diverse effects on labor force decisions of migration type warrant more extensive discussion. At one extreme stand Native Americans, whose labor force behavior was relatively unaffected by geographic movement. These findings are consistent with those of Sandefur (1986), who found no significant differences in the effects of interstate migration on the labor force participation of American Indians. l4 Although Sandefur did not speculate why this result emerged, it is conceivable that the resettlement assistance provided movers does not stipulate that recipients secure employment prior to moving. Nonmigrants chose to stay either because they have employment or because, in its absence, they can rely on other forms of income maintenance and informal social supports to subsist. For blacks and Mexicans, the pattern of migration effects on labor force behavior is roughly similar. Moves within high-density labor markets lowered the probability of participation by 10 percent for mature black men and by approximately 5 percent for mature Mexican men. That dispersed migration substantially lowered rather than increased the probability of labor force par-
147
Migration, Ethnicity, and Labor Force Activity
ticipation for these groups illustrates both the disruptive aspect of moves and the employment difficulties experienced by men of color in sociocultural environments where members of like ethnicity are less available. The results for Cubans and Puerto Ricans were surprising for different reasons. Because the vast majority of concentrated flows of Cubans involve individuals moving to Miami, Florida (Bean and Tienda 1987), where a thriving enclave economy shields workers from general competition (Portes and Bach 1985), we expected that participation in concentrated flows would increase the probability of labor force participation for Cubans. Instead, the estimated effect is strongly negative, indicating participation probabilities 8.6 points below those of nonmigrants, on average. This result may indicate that the Miami labor market is already saturated with Cuban workers or that these migrants can afford to spend longer periods in job search precisely because the availability of ethnic compatriots (usually relatives) provides necessary supports during the interim period. The results for Puerto Rican men are cause for concern because of the sharp decline in the participation of mature men experienced during the 1960s and through the 1970s-a pattern less pronounced for other Hispanic populations (Tienda 1989). For Puerto Ricans, internal migration consistently lowered the employment prospects from 7 percent (for intradensity high moves) to 15 percent (for dispersed moves). This finding is all the more disturbing because Puerto Ricans have become more residentially dispersed than other Hispanic populations. Thus, the declining participation rates of mature Puerto Rican men result from many sources, including the disruptive aspects of moves in general but dispersed moves in particular, the greater prevalence of dispersed compared to concentrated flows, and the sharply declining employment opportunities in labor markets where Puerto Ricans traditionally have concentrated. The unemployment effects of migration were generally weaker than those on labor force participation in several senses. First, only three of the twenty estimated coefficients (for migration types) attained statistical significance; second, these coefficients were confined to Mexican and Native American men. Stated differently, migration does not appear to influence the unemployment prospects of Puerto Rican, Cuban, or black men, irrespective of the ethnic density of origin and destination labor markets. This is not the case for Mexican and American Indians, for whom migration increased the probability of unemployment. For Mexicans, the statistically significant effects corresponded to the migration types that generated reliable coefficients in the participation equation, except that the coefficients were oppositely signed (as predicted). Specifically, Mexicans who participated in dispersed migration flows experienced unemployment rates almost 10 percent higher than (statistically) equivalent nonmigrants. In contrast, migrants who participated in moves designated as concentrated were as likely as nonmigrants to be unemployed in 1980. Mexican
Table 5.4
Effects of Geographic Mobility and Ethnic Job Segmentationon Labor Force Participation and Unemployment (transformed logit coefficients) Mexican
LFP Migration type: Intradensity high lntradensity low Concentrated Dispersed Ethnic job segmentation: Ethnic typed Anglo typed Human capitaVfamily status: If < high school complete If high school complete Experience Experience squared If English good If foreign born If work disability If married
-.047** - ,014 - ,011 - .068*
... ,015 - ,027
,001 .003* ...
UNEM ,029 .029 -.035 .097*** -.024** -.050** .216*** .115* - ,002
...
-.020** .016* - ,009 - a*** .037* - .025** .007 - ,008
Puerto Rican
LFP
UNEM
Cuban LIT
-.068** -.091* -.118* - . l52***
.009 .003 - ,014 ,023
-.086** ,006
,007 ,011
- .010 - ,010
-.008 - ,003
- .199*** .loo** -.097* .050 .005** - ,002 ... ... - .020** ,011 ,005 - .017* - .446*** .070*** .019* - .022**
-.013 ,005
- .028
- ,014 ,003 - ,001 .009 .015 - .408*** .014
Black UNEM
American Indian
LFP
UNEM
LFP
-.012
-.103***
,010 -.001
-.040
,045 .025 ,028 .020
- .008 -.011 - .019 - .063
.058** ,039 ,053
,016
-.040**
,016 - .051***
-.025
-.008 .003 .102*** ,029 -.001
... ,004 ,004 .069*** -.014*
- ,042 -.143*** -.021* ,020
- ,045 -.036
.008***
...
.05 1 ,048 - .429*** .025*
-.050** .147*** .072* - ,003
... - ,029 .009 .041*
- .025**
.037* - .123***
- .094** .006*** ...
-.043 - ,034 - .352*** .033**
UNEM
.060*
,309*** .240*** - .005**
... -.027 ,025 .050** - .034***
If household head If children < 6 Labor market: If region = Northcentral If region = South If region = West Area unemployment rate Area wage rate If metro residence % Durable manufacturing % Nondurable manufacturing X2
df N
.045*** - .024** .020* ,019 ,015
...
,018
-.013 -.025 - ,011
- ,002
,011
.004 ,001 .Ooo
-.002 ,016
-.001 2,265 6,051 6,076
... -.001 2,576 5,663 5,688
.049*** .024* ,025 .035* .038** - ,012 ,008 ,026 ,001 .007*** 3,056 6,605 6,630
- ,015 ,002 - ,003 - ,041 - ,001 .009** - .013 .074*
...
.024*** ,017
- ,003 - .094* ,015
- ,005 - .025 .039*** - ,001
- .015* - .011 ,019 - ,022 ,015 - ,006
- ,009 - .026* - .002*
- ,001
- ,001
,001
3,081 6,035 6,060
1,152 4,109 4,134
1,205 3,925 3,950
.069*** .057*** ,020 .038** ,013 - .009** .015* .025 .001 .003** 3,159 5,802 5,827
- .037*** -.006
.088*** -.009
- .036***
.031 - ,002 .045* - .015***
- ,030
,029 - .029 - .017
,007
- ,002
- ,016
- .005 - ,003 - .013** .033**
.001
.005***
...
,001
,002
...
2,765 5,157 5,182
3,744 5,785 5,810
,011
.OlO***
,006 - ,019
3,179 4,989 5,014
Note: Evaluated at the following means: Mexicans, LFP = ,936, UNEM = ,066; h e r t o Ricans, LFF’ = ,914, UNEM = .075; Cubans, LFP = .956, UNEM = ,038; American Indians, LFF’ = .863, UNEM = ,109; blacks, LFP = ,889, UNEM = ,086. a Coefficient less than ,001. + p 5 .I0 (two-tailed test). * p 5 .05 (two-tailed test). **p 5 .01 (two-tailed test). ***p 5 ,001 (two-tailed test).
150
Marta Tienda and Franklin D. Wilson
migrants who moved between labor markets where Mexicans were disproportionately represented experienced unemployment rates only 3 percent higher than nonmigrants, on average, but this effect was statistically trivial. American Indian unemployment rates were most affected by residential mobility, as all migrants except those who moved between low-Indian-density labor markets and from low- to high-density labor markets experienced higher unemployment than their nonmigrant counterparts. The risks of unemployment for these American Indian migrants averaged approximately 6 percent more than those of (statistically) comparable nonmigrants. The influence of job segmentation on labor force participation and unemployment probabilities revealed very limited evidence that the ethnic typing of jobs operated to reserve positions for minority men or to shield them from unemployment. Only for blacks and American Indians, two groups identified as occupying the lower rungs of the hiring queue, did statistically reliable effects emerge in the participation equation. These results indicate that mature black and American Indian men destined for ethnic-typed jobs have lower rates of labor force participation than their (statistical) counterparts destined for nontyped jobs. l5 However, Native Americans associated with Anglo-typed jobs participated in the labor force at a rate 4 percent higher than their similarly skilled counterparts associated with nontyped jobs. This finding, coupled with evidence that American Indian men associated with Anglo-typed jobs experience significantly lower unemployment than their counterparts destined for nontyped jobs, possibly reflects the results of affirmative hiring of American Indians within the Bureau of Indian Affairs and in other nonmanual government jobs. The only shred of evidence that ethnic segmentation of jobs “reserves” slots for minority men emerges from the Mexican unemployment equation, which shows that adult men destined for “Mexican jobs” were slightly less likely to be unemployed in 1980 than their counterparts associated with nontyped jobs. But, as the results for Mexicans, blacks, and American Indians show, the lowest unemployment risks are associated with Anglo-typed jobs-predominantly nonmanual and skilled manual jobs. Apparently, the desirability of jobs rather than the ethnicity of incumbents influences the unemployment risks of minority men. Moreover, because the dominant group generally occupies the more desirable jobs, ethnic-typed jobs exhibit greater employment instability and hence expose their incumbents to greater risks of unemployment. With respect to labor market characteristics, none emerged as significant predictors of Mexican men’s labor supply, in contrast to results for the remaining groups. Participation rates of Puerto Ricans were lowest in the Northeast, the region where this group is disproportionately concentrated and where the dislocation effects of industrial restructuring have been particularly severe. That Puerto Rican unemployment risks were not responsive to our regional dummies reveals that the massive numbers of F’uerto Rican workers displaced
151
Migration, Ethnicity, and Labor Force Activity
from jobs in the Northeast have not been absorbed by other regions. The sharp impoverishment of the population during the 1970s and 1980s (Tienda and Jensen 1988) can largely be understood in these terms. Further evidence for this argument is found in the higher unemployment probabilities associated with residence in metropolitan areas and high-unemployment areas. Cubans residing in the South had lower participation rates than their counterparts residing elsewhere, but they also had lower unemployment rates. Black men residing in the South and in high-wage areas participated more in the labor force than other black men, and highest unemployment risks were associated with residence in the West, where blacks are less concentrated. Finally, participation rates of American Indians were slightly higher and unemployment rates lower in metropolitan areas. Among the set of human capital and family status variables included in our models, work disability produced consistently strong negative effects on labor force participation and strong positive effects on unemployment risks. English proficiency lowered unemployment probabilities only for Mexican and Puerto Rican men. The influence on the labor force participation and unemployment of minority men’s educational credentials was as expected for Mexicans and Puerto Ricans, but, surprisingly, we detected no schooling effects on the participation decisions of black and Cuban men. This result is all the more puzzling since, according to our previous discussion, these men occupy the top and bottom of the hiring queue and educational credentials presumably should not function identically for both. That they do highlights the importance of ascription over achievement (i.e., race over skill) in differentiating labor market experiences. To summarize, the evidence from our analyses of labor force participation challenges the widespread assertion that dispersed migration streams potentially can promote socioeconomic assimilation along with spatial assimilation. In fact, for three of the five groups analyzed (Mexicans, Puerto Ricans, and blacks), the effects on labor force participation of dispersed migration flows were negative rather than positive; for the remaining two groups (Cubans and American Indians), dispersed migrants were neither more nor less likely to participate in the labor force than nonmigrants. Also, identification with ethnic- or Anglo-typed jobs differentiated the participation decisions of American Indian men and to a lesser extent black men, but it had little to do with the participation decisions of Hispanic men. However, Mexican, black, and American Indian men associated with Anglo-typed jobs experienced lower unemployment risks compared to those associated with nontyped jobs.
5.2.6 Annual Earnings By and large, and in contradistinction to the results reported for the labor supply models, the earnings consequences of migration types are generally weak (see table 5.5). For Cubans and Puerto Ricans, the economic returns to
Table 5.5
Migration type: Intradensity high
Effects of Geographic Mobility and Ethnic Job Segmentation on 1979
(logged) Annual Earnings: Migrant and Nonmigrant Men Aged lkenty-Five to Sixty-Four (standard errors in parentheses) Mexican
Puerto Rican
- .078* (.039) .001
Intradensity low Concentrated Dispersed
(.087) - ,102 (.092) ,027 (.073)
Cuban
Black
,036
,036
(.a51
(.045)
(.ow
- .061 (.081)
- ,120 (.094) -.118+
(.063)
,081
,071 (.098) ,020 (.073) -.150+ (.082)
-.184+ (.101) ,145 (.091) - ,022 ( ,085)
American Indian
- .057 (.046) .056 - .060 (.063) - .138* (.067)
Ethnic job segmentation: Ethnic typed
- . I lo***
Anglo typed
(.029) .195***
- .145*** (.025) .077* (.039)
- .249*** (.033) .117*** (.031)
- .185*** (.023) ,001 (.047)
- ,017 (.029) .097** (.032)
- .361*** (.105) - .637*** (.114) ,025
- .287*** (.069) - .456*** (.077) ,015
- .268*** (.039) - .331***
- .243*** (.061) - .373*** (.078) - .020***
-
- .298*** (.044) - .518*** (.054) .022*** (.W)
- .0002**
- .0002*
- .0003***
- @)04***
Human capitaVfamily status: If < high school complete If high school complete Experience Experience squared If English good If foreign born If work disability If manied Weeks worked
(.ow 0@)4***
(.W1) (.W1)
.176*** (.028) - .071** (.024) - .120** (.046) .152*** (.030) .033*** (.001)
Usual hours A
Constant R2 N
(.ow
.007*** (.001) .o001 6.489 ,342 5,726
.116*** (.026) - .I1 I*** (.024) - .154*** (.042) .142*** (.025) .035*** (.@)I) .007*** (.001) .011*
6.448 .315 5,908
(.045) .010* (.004) (.OOO1)
(.OOO1)
(.OOO1)
.167*** (.028) - ,083 + (.045) - .159** (.061) .140*** (.032) .034*** (.001) .007***
- .I82 (.272) - ,036 (.059) - .205*** (.049) .119*** (.027) .034*** (.@)I)
.316*** (.071) - ,028 (.074) - .234*** (.042) .186*** (.034) .036***
(.001)
o@J***
(.001)
.011***
(.001)
(.001)
(.006)
.057* (.023)
- ,004 (.017)
6.656 ,340 3,895
6.566 ,376 5,235
5.590 ,428 5,324
.019***
Source: 1980 5% A Sample Public Use Micro-data Samples, migrant subsample. Note: Net of vector of labor market characteristics: region; area wage rate; metropolitan residence. + p 5 .I0 (two-tailed test). * p 5 .05 (two-tailed test). **p 5 .01 (two-tailed test). ***p 5 ,001 (two-tailed test). +
153
Migration, Ethnicity, and Labor Force Activity
different migratory destination are statistically trivial (p,-p, = 0; see eq. [ 2 ] ) ,although the income-depressing effects of dispersed streams border on statistical significance.l6 However, for each of the remaining groups, at least one migrant type emerged as a significant determinant of annual earnings. Participation in dispersed migration streams rendered American Indians a loss of 15 percent. Apparently, spatial dispersion does not automatically translate into economic gains for American Indian men. This questions the wisdom underlying efforts to relocate Native Americans as a strategy promoting social integration. For Cubans and Puerto Ricans, it appears that spatial dispersion is not associated with economic success, but other types of movers earn the same as equivalent nonmigrants. The (logged) annual earnings of migrant and nonmigrant black men were approximately equal, and our disaggregated migrant typology yields no further insights into the economic consequences of specific flows. However, Mexicans who participate in intradensity streams received negative returns on the migration decision. In contrast to our previous results, the regressions of (logged) annual earnings show large negative retums from incumbency in ethnic-typed jobs (with the exception of American Indians) and substantial positive returns from incumbency in Anglo-typed jobs (with the sole exception of blacks). Specifically, the earnings penalty from incumbency in ethnic-typed jobs ranged from a low of 11 percent for Mexicans to a high of 25 percent for Cubans, while the bonus from incumbency in Anglo-typed jobs ranged from a low of 8 and 10 percent for Puerto Rican and American Indian men, respectively, to a high of 20 percent for Mexican men. That blacks did not benefit financially from incumbency in Anglo-typed jobs reflects the legacy of discrimination in excluding them from these jobs and denying them equal returns for equal work. In sum, our findings on earnings challenge arguments about the economic benefits of ethnic segmentation for minority workers. Because the jobs in which minority men are concentrated disproportionately involve manual work in extractive, manufacturing, and personal services, their remuneration is lower, on average, compared to that of nonmanual jobs, particularly professional and managerial positions in the producer and social services sectors. The few minority men who do manage to enter Anglo-dominated jobs do remarkably well, financially. Nevertheless, as long as ethnicity continues to have exchange value in the labor market, not only will the earnings of minority men remain highly unequal, but so also will the financial rewards associated with ethnically segmented jobs.
5.3
Conclusion
On balance, our results provide some evidence about how and why the paths of labor market insertion differ among mature minority men, but our
154
Marta Tienda and Franklin D. Wilson
story about the role of migration and ethnic job segmentation in stratifying the minority work force is complex. First, when significant effects of migration on labor force participation emerged, they were uniformly negative (positive in the unemployment equation). This implies that the higher unemployment experiences and lower labor force activity rates of migrants may reflect the disruptive effects of the migration process per se. These effects, if they are associated with the process of movement per se, might disappear as migrants acquire experience and familiarity with their destination labor markets. Unfortunately, our cross-sectional Census data did not permit us to investigate whether migrants who had moved earlier were more likely to participate in the labor force than later migrants. This certainly is an important research priority for additional work on the economic consequences of residential mobility for minority populations. That the effects of migration were most pronounced for labor force behavior but not economic returns further challenges the premises of microeconomic theory, which presumes that decisions to move represent rational choices geared to improve economic well-being. But whether ethnic alliances are involved in explaining the prevalence of concentrated flows, and those within areas of high ethnic concentration, or in ameliorating the disruptive aspects of residential mobility is not clearly evident from our results. On the basis of the theoretical arguments presented at the outset, evidence for such claims should derive from the nature of ethnic segmentation and the existence of ethnic hiring queues. Although our ethnic segmentation effects on labor force decisions and unemployment risks were trivial at best, their effects were far more pronounced on (logged) annual earnings. One general implication from our results is that there is no evidence that spatial assimilation will necessarily promote socioeconomic integration and thereby reduce ethnic labor market inequities. Only Cuban men benefited from dispersed migration streams, while American Indians sustained substantial losses from disrupting their social ties and participating in dispersed migration streams. Certainly, this does not justify mandatory relocation programs as a strategy for promoting social equity.
Appendix A Analytical Procedures to Determine High-Density Labor Market Areas To determine which labor market areas contain an above-average concentration of a particular racial or ethnic group, we examined two relevant variables, the raciallethnic composition of each labor market area and the distribution of each group across the 414 labor market areas. These labor market areas were
155
Migration, Ethnicity, and Labor Force Activity
derived from the Census-defined county groups and consist of SMSAs or groups of nonmetropolitan counties within states. Population counts from the 1980 1/100 PUMSA were used to calculate these variables for the following groups: blacks, American Indians, Mexicans, Puerto Ricans, Cubans, and other Hispanics. The total population was divided in mutually exclusive categories as follows: anyone identifying himself or herself as “American Indian” on the race question was considered American Indian; non-Indian Hispanics were identified on the “Spanish origin” question, which contained separate spaces for Mexican, Puerto Rican, Cuban, and other; the remainder were placed into either the white, Asian, or black categories on the basis of their answer to the race question. A labor market area was defined as high density if the reference group was overrepresented in terms of both composition and distribution. Overrepresentation was determined by calculating a set of standardized scores for the two variables. For the compositional z-score, the group’s percentage for the country as a whole (the weighted mean across areas, e.g., 11.58 percent for blacks) was used to represent the value expected if that group was evenly distributed across labor market areas relative to all other groups. The simple mean, which is the same for all groups (.24 percent, or %M), was used for the distributional z-score. A labor market area was classified as concentrated if both these standardized scores were greater than zero. Therefore, a concentrated black labor market area would be one containing more blacks than the total U.S. average and a higher than average share of blacks. If only one of these conditions were met, the labor market area was not classified as high black density. Details about the classification of specific labor market areas are available from the authors. The results of this analytical procedure are available from the authors. Blacks are the most dispersed groups, with seventy-three concentrated labor market areas containing 75 percent of all blacks, and Cubans are the least dispersed, with 83 percent living in just seventeen areas. The percentage of each group living in concentrated labor market areas is fairly similar, ranging from a low of 68 percent for American Indians (sixty-two areas) to 85 percent for Mexicans (forty-nine areas). There were thirty-five concentrated labor market areas with 82 percent of the Puerto Ricans and forty areas with 72 percent of the other Hispanics.
Appendix B Estimation of Ethnic Job Queues For the estimation of the ethnic job typology, we used a sample from the 1970 Public Use Microdata Files of Hispanic, black, American Indian, and non-
156
Marta Tienda and Franklin D. Wilson
Hispanic white men aged sixteen to twenty-four who worked within the last five years preceding the Census and had nonmissing industry and occupation in 1970. First, we arrayed the data into a 6 X 6 X 5 X 3 X 3 matrix representing six ethnic groups, six industry sectors, five occupational groups, three education groups, and three age groups. Table 5B.2 presents the industry by occupation distribution for the five minority groups. To establish whether nonwhites are over- or underrepresented relative to non-Hispanic whites, we computed a log-linear analysis that establishes the associations among ethnicity, occupation, industry, education, and age groups. We estimated a saturated log-linear model of the form:
where A B C D E
= ethnicity (i = 1,
. . . , 6); industry sector ( j = 1 , . . . ,6); = occupation group ( k = 1, . . . 3; = education ( I = 1, . . . ,3); = agegroup(rn = 1, . . . ,3).
=
In its abbreviated notation, the hierarchical model is:
[ABCDE] From the 154 estimated parameters,I9 we computed the ethnic typing parameters for twenty-five occupation X industry cells for Mexicans, Puerto Ricans, Cubans, blacks, American Indians, and non-Hispanic whites using the following formula:
fori = 1 , . . . , 6 . The first three tau’s indicate the net effects of occupation by industry, and the second three tau’s denote the interaction effects of occupation by industry with ethnicity. All these parameters have been purged of the main and interaction effects of age and education. Ethnic-job-typing effects were calculated as the ratio of the total effects of occupation by industry for each ethnic minority group relative to that of non-Hispanic whites. Since the net effects of occupation by industry are uniform for all ethnic groups, our ethnictyping effects represent the ratio of interaction effects of occupation X industry x ethnicity. The results of these computations, reported in table 5B. 1, provided the ba-
157
Migration, Ethnicity, and Labor Force Activity
Table 5B.l Occupational Groups Mexican: Upper nonmanual Lower nonmanual Upper manual Lower manual Farmer Puerto Rican: Upper nonmanual Lower nonmanual Upper manual Lower manual Farmer Cuban: Upper nonmanual Lower nonmanual Upper manual Lower manual Fanner Black: Upper nonmanual Lower nonmanual Upper manual Lower manual Farmer American Indian: Upper nonmanual Lower nonmanual Upper manual Lower manual Fanner
Ethnic Job pping Effects by Ethnicity Extractive
Transformative
Distributive Services
Producer Services
Social Services
Personal Services
,971 2.040 2.143 2.158 1.146
,432 ,615 ,706 1.255
,552 ,550 1.117 1.500
.627 .430 1.817 1.135
,626 ,772 1.433 1.061
,722 1.282 1.379 1.528
...
...
...
...
...
,930 3.442 ,932 1.946 ,359
,217 ,649 ,618 1.021
,361 ,725 ,595 1.524
.588 .722 2.734 3.571
,513 1. w 2 ,983 1.173
,801 2.728 1.785 2.173
...
...
...
...
...
1.118 4.137 ,992 2.367 ,329
,456 ,600 ,657 ,883
,488 .693 .697 1.264
.771 ,848 1.883 1.812
,513 ,676 1.030 ,694
1.034 3.068 1.618 2.080
...
...
...
...
...
,399 .738 ,938 1.636 .542
,324 ,717 1.191 2.408
.289 ,529 1.146 2.494
,449 ,525 1.676 2.729
1.080 1.794 1.506 1.940
,823 ,890 1.622 1.931
. . .1
...
...
...
...
2.202 5.170 3.219 3.847 .920
.245 .405 .565 1.585
.334 ,307 .554 1.490
,393 ,388 2.353 2.210
,626 ,903 1.438 1.068
,613 1.667 1.646 ,884
...
...
...
...
...
Source: 1970 PUMS Files. All men who worked between 1965 and 1970 and had valid industry and occupation codes. Note: Effects are ratios of Tau ( 7 ) parameters for each ethnic group relative to whites, as described in App. B. Cut points for over, under, or equally represented are established at .5 and 1.5. *These cells are structurally impossible because farm laborers occur only in the extractive sector.
sis for the trichotomous representation of ethnic job queues analyzed in the text. Specifically, ethnic-typed jobs included those with tau ratios less than or equal to 1.5; Anglo-typed jobs included those with tau ratios less than or equal to .5; and nondifferentially preferred jobs categories obtained scores between .5 and 1.5 exclusive. These cut points provide very conservative profiles of ethnic job queues. A more liberal definition of the tolerable limits of ethnic job typing would use .33 and 1.67 as the relative cut points. However, we used the more conservative measures in our analysis.
158
Marta Tienda and Franklin D. Wilson
Table 5B.2
1980 Employment Classification of Men Aged Ikrenty-Five to Sixty-Four National Origin
Sector and Occupation Extractive Upper nonmanual Lower nonmanual Upper manual Lower manual Farmer Transformative Upper nonmanual Lower nonmanual Upper manual Lower manual Farmer Distributive services Upper nonmanual Lower nonmanual Upper manual Lower manual Farmer Producer services Upper nonmanual Lower nonmanual Upper manual Lower manual Fanner Social services Upper nonmanual Lower nonmanual Upper manual Lower manual Fanner Personal services Upper nonmanual Lower nonmanual Upper manual Lower manual Farmer Totals
Puerto Rican
Cuban
Black
12.6 .4
2.3 .2
2.0 .2
.o
.o
.o
2.8 2.0 7.4 44.8 3.2 1.5 32.5 7.6 NA 18.2 2.0 2.9 10.0 3.2 NA 3.8 .9 .7 1.1 NA 11.9 3.6 1.6 2.5 4.3 NA 8.8 1.1 .I 3.2 4.3 NA
.3 .8 1.1 42.4 2.4 2.6 32.5 4.8 NA 20.2 2.9 4.2 9.6 3.5 NA 9.7 2.7 I .8 1.2 3.9 NA 15.0 3.5 2.9 1.7 6.9 NA 10.4 1.3 .5 2.9 5.7 NA
.2 .8 .6 38.0 6.7 2.5 25.0 3.8 NA 26.3 6.7 7.1 9.9 2.6 NA 10.6 4.8 2.9 .7 2.2 NA 11.5 6.0 1.4 1.1 3.0 NA 11.5 2.3 .6 3.4 5.2 NA
3.8 .2 .I .5 .9 2.0 44.3 2.4 2.4 30.3 9.3 NA 20.1 1.8 3.6 10.6 4.2 NA 5.8 1.2 1.4 .8 2.5 NA 19.1 5.7 3.2 2.8 7.4 NA 6.9
100.0
100.0
100.0
Mexican
1 .o
American Indian
9.2 .6 .1 2.8 2.0 3.6 41.9 5.0 1.4 28.1 1.5 NA 17.0 2.8 2.9 8.7 2.7 NA 4.9 1.8 1.o
.8 1.3 NA 21.2 7.0 2.3 4.3 7.6 NA 5.8
1 .o
1.o
.2 2.2 3.4 NA
.2 2.0 2.5 NA
100.0
100.0
Source: 1980 5% A Sample Public Use Micro-data Samples, migrant subsample. Note: Upper nonmanual includes professionals, semiprofessionals, and managers. Lower nonmanual includes clericals and sales. Upper manual includes crafts and operatives. Lower manual includes service workers and laborers. Farmer includes farmers and farm laborers.
159
Migration, Ethnicity, and Labor Force Activity
Appendix C Variables Included in Logit, Probit and OLG Regressions Migration Type Categorical variables coded as dummies for four types of density: Intradensity High Intradensity Low Concentrated Dispersed Nonmigrants
Moves within high-ethnic-density SMSAs or nonmetropolitan county groups Moves within low-ethnic-density SMSAs or nonmetropolitan county groups Moves from low- to high-ethnic-density SMSAs or nonmetropolitan county groups Moves from high- to low-ethnic-density SMSAs or nonmetropolitan groups No residence changes across SMSA boundaries
Ethnic Job Typing Categorical variables coded as dummies for two preferential statuses; measures calculated separately for each national origin group (see App. B): Ethnic Typed
Anglo Typed
Nontyped
Denotes job cells in which ethnic workers were overrepresented relative to nonHispanic whites in 1970, unique for each group Denotes job cells in which ethnic workers were underrepresented relative to nonHispanic whites in 1970, unique for each group Denotes job cells in which ethnic workers were approximately equally represented relative to whites in 1970, unique for each group
Density Categorical variables coded as dummies for two levels of ethnic concentration: High
Low
If 1980 labor market areas meet criteria for high concentration of Mexican, Puerto Rican, Cuban, black, or American Indian groups (see App. A) Remaining labor market areas
160
Marta Tienda and Franklin D. Wilson
Human CapitallFamily Status Household Head and Young Children for labor supply equations only; Weeks and Hours for earnings model only: Education
Experience Experience Squared English Ability Foreign Birth Married Household Head Young Children Weeks Hours
Dummy variables designating completion of college, high school or some college or less than high school Labor market experience proxy derived as (age-education-6) Square of experience Dummy variable denoting good or excellent proficiency in English Dummy variable denoting foreign birth Dummy variable denoting if respondent was married Dummy variable denoting if respondent was household head Dummy variable denoting presence of children less than 6 years old Weeks worked last year Usual hours worked
Labor Market Definition of labor market in Appendix A; Area Unemployment Rate and Industry for earnings model only: Region
Area Unemployment Rate Area Wage Rate Metro Industry
Dummy variable designating four regions of residence: West, South, Northeast, and North Central Unemployment rate for SMSAs or nonmetropolitan county groups Mean wage rate for SMSAs or nonmetropolitan county groups Dummy variable coded 1 if metropolitan residence Share of area labor force engaged in durable and nondurable manufacturing
Notes 1. Affirmative action hiring on the basis of race, gender, and national origin is one mechanism through which minority populations may be designated as preferred hires for a specific set of jobs. The changes in minority employment patterns as a result of
161
Migration, Ethnicity, and Labor Force Activity
affirmative action laws and their enforcement are documented by Leonard (1984) and Smith and Welch (1984). 2. As an example, Lieberson explains that, if the most favored group is 10 percent of the population in one city and 20 percent in another city, the second-ranked group can begin to fill jobs ranked at the ninetieth percentile in the first city and the eightieth percentile in the second city. In other words, if the lowest-ranked group is large, then members of this group will find jobs higher in the occupational queue because the higher-ranked groups do not push as far down. 3. Alternatively, during periods of labor shortage, groups at the bottom would experience broadened employment opportunities because the preferred groups would not fill all the traditional employment opportunities. Thus, employment shifts usually will be most radical for groups at the bottom of the employment queue. 4. To save money, migration variables were coded for roughly half of all persons aged 5 and over. Because the A file is based on a 5 percent sample of the total population, we did not encounter sample size restrictions with the minority populations. 5. Restrictions pertaining to work status ensure that all individuals in the sample had valid occupation and industry codes, which are needed to derive our ethnic job categories. The restriction on U.S. residence in 1975 was necessary for computing the migration types. Although the ethnic concentration of origin countries of recent immigrants is uniformly high, it was not pertinent for the comparisons in our migration typology, which focuses on internal moves. Finally, men in the military or enrolled in college either in 1975 or 1980 were excluded because the groups have higher migration propensities de facto, independent of the social and economic motivations underlying these decisions. 6. The stringency of our sample restrictions prompted additional diagnostics of the social and demographic characteristics of the excluded population. These analyses revealed that men who never worked or who were in the military or enrolled in college in 1980 tended to be younger and were more apt to be unmarried than their respective subpopulations. Individuals who were not in the labor force in 1980 and who had last worked before 1975 were older, on average, than their source subpopulation. Also, since recent immigrants tend to be younger and less often married, the final Mexican and h e r t o Rican samples contained fewer men under age 30 and fewer unmarried men than would be true if recent immigrants were not excluded from the analysis. This was less true of Cuban immigrants, who tend to be older than all immigrants (on average). 7. Whether migration streams alter the ethnic composition of specific labor markets depends on the differential net migration of minority and nonminority groups. Thus, the ethnic density of a particular labor market may increase either because of net outmigration of non-Hispanic whites or net in-migration of minorities. Ethnic density may similarly decrease because of a net out-migration of minorities or net in-migration of whites. Our focus, however, is not on the changing residential configuration per se but rather on the significance of ethnic concentration for stratification processes or, more specifically, the direction of migrant streams relative to the ethnic density of origin and destination areas. 8. Our analysis of preference status categories was based on 1970-prior to the migration interval studied-rather than 1980 industry by occupation classifications to avoid endogeneity problems from confounding the causal ordering or migration and resulting ethnic job queues. However, the actual classification of individuals among the three categories is based on industry and occupation categories reported as of 1980, at the end of the migration interval. 9. Since the models used to predict unemployment probabilities are identical, we do not repeat them; we expect the effects of our key independent variables to be opposite those produced for the labor force participation models. 10. This implies that demand for workers can be ethnicity specific, especially in
162
Marta Tienda and Franklin D. Wilson
contexts where recruitment is governed by an ethnic job queue or where the assumptions of homogeneous labor and perfect competition do not hold. 1I . For the earnings models, we corrected for selection bias resulting from differences in the probability of being in the wage sample. Results are available on request. 12. This overrepresentation in producer and personal services reflects the large numbers of Cubans involved in the banking industry in Miami as well as the large share of Cuban-owned and -operated enterprises employing a highly ethnic (Cuban) labor force. See the discussion in Portes and Bach (1985). 13. A competing explanation is that the selection process governing the joint migration and employment decisions has not been adequately modeled. Elsewhere, we have modeled employment and migration outcomes using a bivariate probit specification that takes account of the selection processes and reached essentially the same conclusions. 14. Sandefur analyzed the effects of migration on labor force participation separately for endogamous, exogamous, and intermarried American Indian couples and found uniformly nonsignificant effects for all three groups. This finding obtained despite the very different rates of interstate migration for each of these three groups. 15. Although the result is not statistically reliable, association with ethnic-typed jobs appears to increase unemployment for these two groups. These results are consistent with evidence presented in Appendix table 5B. 1 showing that blacks and American Indians were differentially concentrated in nonmanual jobs even after adjusting for differences in their age and education composition. The alternative of not working may be more attractive than holding such low-status jobs, but this conclusion cannot be confirmed from these results. 16. To verify the robustness of these results, we substituted a single dichotomous variable indicating nonmigrant status in the earnings equation and confirmed that the logged annual earnings of migrants and nonmigrants did not differ significantly for any of the minority groups analyzed. However, by differentiating among migration types, we were able to detect some significant ethnic differences in the influence of migration on earnings. 17. For American Indians, dispersed flows were about as pervasive as concentrated flows and amounted to about 16 percent of all moves. 18. Determining boundaries of labor markets was a complicated process. The basic unit is the SMSA, which we reconstructed from county group codes. Then nonmetropolitan areas within states were divided up into two or three areas. The result was 414 labor market areas: 310 SMSAs and 104 nonmetropolitan areas. Individual area codes were determined not by the SMSA code but by a combination of the state and county group codes. This caused problems when county groups spread across two or more SMSAs or when SMSAs crossed state boundaries. The decision rules used to allocate county groups are available from the authors. 19. Because farm occupations are found only in the extractive sector, the industry occupation interaction yields twenty five rather than thirty cells. By deleting the five structurally impossible cells for all groups, we estimated thirty fewer parameters than model 1 implies.
References Bean, F. D., and M. Tienda. 1987. The Hispanic Population of the United States. New York: Russell Sage.
163
Migration, Ethnicity, and Labor Force Activity
Hechter, M. 1978. Group formation and the cultural division of labor. American Journal of Sociology 84(2):293-317. Leonard, J. 1984. The impact of affirmative action on employment. Journal of Labor Economics 2(4):439-63. Lieberson, S. 1980. A Piece ofthe Pie. Berkeley and Los Angeles: University of California Press. Nelson, C., and M. Tienda. 1985. The structuring of Hispanic ethnicity: Historical and contemporary perspectives. Ethnic and Racial Studies 8( 1):49-74. Petersen, T. 1985. A comment on presenting results from logit and probit. American Sociological Review 50(1): 130-3 1. Portes, A., and R. L. Bach. 1985. Latin Journey: Cuban andMexican Immigrants in the United States. Berkeley and Los Angeles: University of California Press. Sandefur, G. 1986. American Indian migration and economic opportunities. International Migration Review 20( l): 55-68. Smith, J. P., and F. Welch. 1984. Affirmative action and labor markets. Journal of Labor Economics 2(2): 269-301. Snipp, C. M., and G. Sandefur. 1988. Earnings of American Indians and Alaskan Natives: The Effects of Residence and Migration. Social Forces 66:994-1008. Tienda, M. 1989. Puerto Ricans and the underclass debate. Annals of the American Academy of Political and Social Sciences SOl(January): 105-19. Tienda, M., and L. Jensen. 1988. Poverty and minorities: A quarter-century profile of color and socioeconomic disadvantage. In Divided Opportunities: Minorities, Poverty and Social Policy, ed. G. Sandefur and M. Tienda, 23-61. New York: Plenum. Tienda, M., and D. T. Lii. 1987. Minority concentration and earnings inequality: Blacks, Hispanics and Asians compared. American Journal of Sociology 93(July): 141-65. Wilson, W. J. 1987. The Truly Disadvantaged. Chicago: University of Chicago Press.
This Page Intentionally Left Blank
Impact of Immigration, Trade, and Capital Flows on the U.S. Labor Market
This Page Intentionally Left Blank
6
Labor Market Adjustments to Increased Immigration Robert J. LaLonde and Robert H. Topel
During the 1970s, immigration to the United States was higher than in any decade since the 1920s, raising the number of immigrants in the U.S. labor market by 45 percent. The flow of new immigrants has actually increased during the 1980s, and in many areas immigration is a major component of labor force growth. These facts are central to the current debate over immigration policy since it is widely believed that new immigrants have deleterious effects on the labor market opportunities of native Americans.’ For example, if the main costs of immigration are borne by less-skilled natives through reduced earnings and employment opportunities, the case for immigration controls and redistributive policies is strengthened.2 In contrast, if the labor market easily absorbs new immigrants without serious distributional effects, these policy options are less attractive. Increased labor supply due to immigration enhances the welfare of the typical consumer, but it also creates adverse distributional effects among workers whose skills compete with those of immigrants. Yet it is difficult to argue that even the large flow of immigrants in the 1970s could have had a substantial Robert J. LaLonde is associate professor of industrial relations at the Graduate School of Business, University of Chicago, and a faculty research associate of the National Bureau of Economic Research. Robert H. Topel is professor of business economics at the Graduate School of Business, University of Chicago, and a faculty research associate of the National Bureau of Economic Research. This research was supported by the Olin Foundation through a grant to the Center for the Study of the Economy and the State and by the Ford Foundation through a grant to NBER. Topel also acknowledges support from the National Science Foundation and the William Fishman Research Fund. LaLonde also acknowledges support from the Graduate School of Business at the University of Chicago. The authors thank Dan Hamermesh, Richard Freeman, Arlene Holen, Kevin Murphy, Finis Welch, and workshop participants at the Board of Governors of the Federal Reserve, Chicago, Georgetown, Maryland, NBER, Northwestern, and the University of California, Santa Barbara and Los Angeles, for helpful comments. William Anderson assisted with the calculations. Errors are the authors’.
167
168
Robert J. LaLonde and Robert H. Topel
effect on the U.S. labor market. New immigrants of all ages contributed only about 2.5 million extra persons to the labor force over this period, compared to the concomitant increase of twenty million among workers aged 32 or less that was caused by the baby boom and increased labor force participation by young women.3 At this level of aggregation, immigration would have only a second-order effect on the labor market. However, nearly half of all new immigrants live in six metropolitan areas, so that the potential effects of increased immigration may be similarly concentrated in local labor markets. In fact, this feature of immigration is the focus of the current policy debate. Those who believe that immigration has important effects are concerned not with the earnings and employment of the typical worker, who probably gains, but instead with the prospects for certain groups who reside in specific areas, such as young blacks in Miami or native Hispanics in Los Angeles. Our analysis exploits this geographic diversity to study the effect of immigration on local labor markets. In our view, the empirical issue is how increased immigration affects labor market opportunities for workers who are close substitutes for immigrants. Since theory offers little guidance about which groups these are, our strategy is to analyze the effect of immigration on labor market outcomes for workers who are a priori similar to new immigrants-other members of current and past immigrant cohorts of similar ethnicity. Substitution effects for these workers will generally dominate those for nonimmigrant labor, so estimates of these effects will serve as upper bounds for the effect of immigration on labor market outcomes for native^.^ We also test these bounds by estimating corresponding substitution effects for young blacks and native Hispanics. The empirical analysis uses earnings and employment data for immigrants and native-born workers from the 1970 and 1980 Censuses. To estimate the effect of immigration on labor market outcomes, we rely on three distinct sources of variation in the relative importance of immigrants in local labor markets: (i) the share of all immigrants within a locale; (ii) the share of new immigrants; and (iii) the changes in these immigrant shares between 1970 and 1980. The first source of variation, the immigrants’ labor force shares, will generate corresponding differences in immigrant and nonimmigrant earnings if the geographic location of new immigration is exogenous and if nonimmigrant or other factor mobility does not fully arbitrage wage differences in market equilibrium. In other words, labor supply to a locale must be inelastic, at least in the short run. We find that earnings of both new and old immigrant cohorts are lower in areas where immigrants-especially new immigrantsform a large or growing portion of the local labor force. This relation suggests important effects of immigration on earnings, although an alternative explanation is that less-skilled immigrants locate in areas where immigrants form a large share of the labor force. The second, and more important, source of variation in immigration across areas is the labor force shares of different immigrant cohorts within locales.
169
Labor Market Adjustments to Increased Immigration
Variations over time in the rate and location of immigration have generated substantial differences between locales in average arrival times of immigrants. We find that an increase in the relative share of an immigrant cohort within an area (e.g., immigrants who arrived between 1970 and 1974) causes a corresponding decline in the wages and earnings of members of that cohort. Our best estimate is that a doubling of new immigration to a locale would reduce new immigrant annual earnings by less than 3 percent. This modest earnings disadvantage for members of large immigrant cohorts dissipates with time in the United States. It appears that immigrants assimilate into the broader labor market as they accumulate skills that are appropriate to the U.S. labor market. Our evidence also indicates that new immigration reduces the earnings of earlier immigrant cohorts. Thus, “new” and “old” immigrants are substitutes. However, as theory predicts, these substitution effects on wages are found to be smaller for older immigrant cohorts, which is also consistent with the assimilation of immigrants. We regard these results as evidence for the existence of within-market substitution effects of immigration on wages. In light of these results, it is not surprising that the effect of immigration on natives appears to be minor. For young (aged 16-34) blacks, we find a small negative effect of immigration on relative earnings. Our largest estimate is that a long-term doubling of immigration to an area may reduce the annual earnings of young blacks by about 4 percent, with much smaller effects on young Hispanics. Since market outcomes for young blacks and Hispanics are likely to be the most sensitive to changes in the supply of immigrants, we think this evidence weakens the case for serious distributional effects of immigration. These conclusions are reinforced by estimates derived from the third source of variation: within-market changes in the labor force shares of immigrant workers generated by new immigration. Interarea mobility will arbitrage geographic wage differentials in the long run, but we view the accelerated pace of immigration during the 1970s as an exogenous increase in supply that in the short run will generate relative wage adjustments in areas of unusually heavy immigration. Thus, we expect a decline in the relative earnings of immigrants (and close substitutes) between 1970 and 1980, and we expect that this decline will be concentrated in areas with unusually heavy immigration as well as among more recent immigrant cohorts. We find evidence for these effects: our best estimate from this experiment is that a doubling of the number of recent immigrants within a locale would reduce their relative earnings by about 3 percent. The strong correspondence between these panel estimates and those generated from a single cross section increases our confidence in the results. Our broad assessment of this evidence is that immigration flows do affect earnings and employment of immigrants and nonimmigrants. Members of large immigrant cohorts suffer slightly reduced earnings, especially on first arriving in the United States. But it appears that immigrants assimilate rapidly, and important effects on nonimmigrants are difficult to find. We conclude that
170
Robert J. LaLonde and Robert H. Topel
recent increases in the pace of immigration have been easily absorbed by the labor market so that distributional consequences are not a firm basis for policies that would further restrict immigration to the United States.
6.1 The Empirical Setting The geographic distribution of immigrants, especially new immigrants, is central to our analysis. Table 6.1 illustrates the geographic concentration of both new and old immigrants, showing the arrival date of the stock of immigrants in six “gateway” metropolitan areas in 1970 and 1980. These areas account for about 40 percent of all immigrants in both years and nearly half (47 percent) of all recent immigrants (those arriving within ten years of the survey date). Reflecting the increased flow of immigration, the population Table 6.1
Immigrants in the United States and Six Gateway Cities, 1970 and 1980 United States
Foreign born as % of population: 1970 4.8 1980 6.2 Immigrants in SMSA as % of immigrants in U.S.: 1970 1980 Recent immigrants in SMSA as % of all recent immigrants in U.S.: 1970 1980 Proportion of immigrants in SMSA arriving in past: 0-10 years: 1970 29.3 1980 39.5 10-20 years: 1970 18.1 1980 22.2
Chicago
Houston
Los Angeles
Miami
New York
San Francisco
8.1 10.5
2.6 1.6
11.2 22.3
24.4 35.6
15.0 21.3
11.0 15.7
5.8 5.3
.5 1.6
8.1 11.8
3.2 4.1
17.8 13.8
3.5 3.6
5.5 5.6
.8 2.6
11.6 17.1
6.9 3.7
18.4 13.6
4.5 3.9
37.9 41.9
41.7 64.5
42.0 57.0
63.9 35.6
30.4 38.7
38.2 43.1
22.8 19.8
19.9 18.3
21.8 21.8
10.7 45.2
16.0 24.6
20. I 24. I
Sources: U.S. Census of Population, 1980, General Social and Economic Characteristics, United States Summary and State Summaries, table 99; and U.S. Census of Population, 1970, Characteristics of the Population, table 144. The six cities in the table accounted for 40 percent of all immigrants in the United States. Note that these statistics, which report the importance of immigrants in the total population and in different SMSAs, understate the importance of immigrants in the work force of these cities since a larger share of immigrants is in the labor force than is the case for the native population as a whole.
171
Labor Market Adjustments to Increased Immigration
share of immigrants increased by 30 percent during the 1970s; by 1980, about 40 percent of all immigrants in the United States had arrived during the previous decade, up from 30 percent in 1970. This estimate is widely distributed across cities: in Los Angeles, more than half of all immigrants arrived during the 1970s, and the population share of immigrants doubled over the decade. Currently, immigrants account for nearly a quarter of the male labor force in the Los Angeles standard metropolitan statistical area (SMSA) and more than a third in Miami. Table 6.2 offers a more detailed picture of the geographic distribution of immigrants and their importance as a source of labor force growth. The first two columns report the distributions of “new” and “old’ male immigrants for SMSAs that account for at least 1 percent of all foreign-born persons in the work force. The remarkable correspondence between the flow distribution for new immigrants (col. 1) and the distribution of the stock that arrived before 1970 illustrates the importance of immigrant “enclaves”: new immigrants go where previous ones went. Separate distributions for persons of European, Mexican, and Asian origin confirm the relation and show that enclaves are primarily ethnic in origin. Because of this factor, the geographic distribution of immigrants tends to replicate itself through time. Thus, there is little evidence of wide swings in the geographic distribution of immigrants over time, which partly justifies our assumption, exploited below, that the locational decisions of new immigrants are exogenous. The last two columns of table 6.2 show the importance of immigration as a source of labor force growth in these areas. Though immigration is a minor factor in economy-wide labor force growth, it is the most important factor contributing to the growth of some markets. For example, in Los Angeles, immigration during the 1970s would in itself have caused a 31 percent increase in the local labor force, and new immigrants accounted for nearly twothirds of the actual increase in the labor force during this period. Of course, these estimates would be even larger if the base population were restricted to those with skills that are similar to those of new immigrants. Our econometric analysis will treat immigrants from different arrival cohorts as imperfect substitutes in production. This assumption will hold if either (i) immigrants “assimilate” with time in the United States in the sense of acquiring skills relevant to the American market or (ii) different arrival cohorts bring qualitatively different skills to the United States. Table 6.3 examines these possibilities, presenting differences between the mean log weekly wages of immigrants and white natives of the same age, by Census year. The table demonstrates three important facts. First, within a Census year, relative earnings profiles appear to reflect assimilation in the sense that earlier arrivals earn more (Chiswick 1978). Second, however, assimilation appears to be much less important if an arrival cohort is followed through time (Borjas 1985). For example, workers who were 25 to 34 years old in 1970 earned about 29 percent less than their native white counterparts, and by 1980
Table 6.2
New Immigration Flows and Stocks (the distribution of new and old male immigrants) All Immigrants
Anaheim Boston Chicago Cleveland DallaslFort Worth Detroit Houston Jersey City Los Angeles Miami New York Newark Philadelphia Seattle San Diego San Francisco San Jose Washington, D.C.
Europeans
Mexicans
Share of 1970-80
Share of pre-1970
Share of 1970-80
Share of pre-1970
3.4 1.8 8.3 .6 2.1 1.4 4.4 1.2 22.1 3.9 17.4 2.2 1.5 .7 1.9 4.1 I .8 2.2
2.4 2.3 7.7 1.2 1.o 3.1 2.0 1.4 12.7 6.1 19.7 2.3 1.9 1.2 1.8 4.1 1.6 1.6
1.4 4.6 8.8 1.7 1.1 3.2 1.4 1.6 6.8 1.6 22.0 4.1 3.6 1.1 .9 2.3 I .4 3.0
2.3 3.5 9.0 2.0 .6 4.9 .8 1.1 7.6 1
.o
22.4 2.9 2.6 1.7 1.4 3.6 1.3 1.4
1970-80 Immigrants as a Proportion of:
Asians
Share of pre-1970
Share of 1970-80
Share of pre-1970
1970 Labor Force
11.9
4.3 .1 10.3
...
...
4.3 .1 8.5
2.8 .7 6.1 .1 34.9 .1 .6 .1 .1 .2 5.6 4.1 2.6 .2
3.0 1.5 7.8 1.o 1.6 1.5 4.1 .6 19.0 .5 16.3 1.6 2.3 1.7 1.6 10.2 3.3 2.7
4.4 2.3 6.2 .6 .8 1.8 2.6 .4 14.7 .2 14.9 1.5 1.5 1.8 1.5 16.1 5.6 3.3
23.3 7.3 10.5 2.8 7.5 4.0 17.3 ... 31.1 38.0 5.7 10.8 3.4 6.9 18.9 15.9 19.2 10.4
Share of 1970-80 6.9 ...
... 42.2 .2 .5
.I .2 .2 3.6 2.3 1.7
...
1970-80 Labor Force Growth 14.1 17.6 21.5 23.6 6.5 8.8 9.5
... 65 .O 37.7 -56.4 22.4 16.2 8.1 14.7 30.4 14.7 16.8
Source: Census of Population, Public Use Sample B (see the Appendix for selection criteria used to create the sample for the table). Note: This table lists all SMSAs with at least a 1 percent share of either new or old immigrants in their work forces. All other U.S. SMSAs have a smaller than 1 percent share of the immigrant population.
173
Labor Market Adjustments to Increased Immigration
Table 6.3
Relative Weekly Wages of Male Immigrants by Place of Origin and Years Since Immigration, 1970 and 1980 Years Since Immigration
Place of Origin All immigrants: 1970: 25-34 35-44 1980: 3 5 4 45-54 EurOpe: 1970 25-34 3 5 4 1980: 35-44 45-54 Asia: 1970: 25-34 3 5 4 1980: 35-44 45-54 Mideast: 1970: 25-34 3 5 4 1980: 3 5 4 45-54 Mexico: 1970: 25-34 3 5 4 1980: 3 5 4 45-54 Other Latin American: 1970: 25-34 35-44 1980: 35-44 45-54
0-5
6-10
11-15
16-20
- .292 - .409
-.I19 - .244
- .075 - .095
0 - ,055
- ,473 - ,582
- ,349 - ,529
- .220 - ,385
-.122 - .206
-.133 - ,229
,017 - ,090
- ,022 - .008
.004
-.I44 - ,356
- .210 - ,259
- .032 - .214
- ,015
- ,206 - .400
- ,016 -.I52
.061 ,103
- ,486 - ,039
- .472 - .549
- ,065 - .360
.058 - .288
,036 .OH
- ,244 - .081
- .040 - .171
.147 .123
,122 - ,209
- .274
- ,001 - .243
,087 ,009
.123 ,129
- ,856
-.411 - ,623
- ,353 - ,398
- .322 - ,351
- ,983 - ,927
- .764 - ,930
- ,582
- ,399 - ,669
- ,436 - ,559
- .196 - .354
- ,150
- ,332
- ,083 - ,034
- .609
- ,492 - .740
- ,446 - .535
- .193 - ,355
- ,262
- .659
- ,895
- ,605
,085
- .050
Sources: Public Use Files from the 1970 and 1980 Censuses. For sample selection criteria, see the Appendix. Note: Estimates are differences between mean log weekly earnings of immigrants and of white natives in the indicated age category.
174
Robert J. LaLonde and Robert H. Tope1
they still earned 22 percent less. This estimate of relative assimilation is much smaller than what either cross section would imply, and it means that the average skills of successive arrival cohorts may have declined through time. The third point is that relative immigrant wages declined during the 1970s. This holds in virtually all age categories in the table. Again, one possibility is declining immigrant quality, but another is simple price adjustments in response to market forces, driven perhaps by the increased supply of new immigrants. Do wages respond to immigration flows? Table 6.4 reports relative weekly wages in 119 SMSAs for various immigrant groups in 1970 and 1980. The estimates are tabulated by the proportion of the local labor force in 1980 that is accounted for by immigrants who arrived during the previous decade. “Top third” in the table refers to the set of SMSAs with the largest immigration rates, which account for one-third of all immigration over the decade. “Middle third” refers to the next most immigrant-intensive SMSAs, which account for another third of total immigration, and so on. Relative weekly wages are calculated as the difference between the mean log wages of immigrants and of white males of the same age, within each SMSA. The data show that relative immigrant wages were dramatically lower in labor markets where new immigration flows were largest. For example, row 6 of the table shows that weekly wages of recent immigrants (those who arrived in the last ten years) fell by over 20 percent relative to white natives in cities with the highest immigration rates. The comparable estimate in cities with the lowest immigration rates is only 7.5 percent. One explanation for this pattern is that wages adjust to increases in supply, at least in the short run. An alternative explanation with much different implications is that less-skilled immigrants locate in immigrant enclaves, so that large immigrant populations are less skilled, on average. Our econometric approach seeks to isolate the first of these effects.
6.2 Theoretical Framework In what follows, we view immigration flows as exogenous shifts in the supply of labor to geographically defined labor markets. So long as immigrants form labor aggregates that substitute imperfectly for others in local production, these supply shifts will have their largest effect on immigrant earnings, with declining effects on other input aggregates as substitution possibilities decline. Thus, for example, an increase in new immigration to a locale will reduce the relative earnings of new immigrants and to a lesser extent the earnings of others for whom immigrants are substitutes. The empirical issue is the magnitude of these effect^.^ If immigrants have important effects on other market participants, it must be via the substitution effects just mentioned. An a priori restriction that we find reasonable is that the best substitute for the representative immigrant is another immigrant. At the other extreme, the representative native may be a
175
Labor Market Adjustments to Increased Immigration
Table 6.4
Relative Weekly Wages of Male Immigrants and White Natives in SMSAs Ranked by Shares of Immigrants in Work Forces Top Third
Middle Third
Bottom Third
1, Immigrants arriving after
1970, in 1980 25-34 35-44 45-54 2. All immigrants in 1980 3. Pre-1970 immigrants in 1980 4. All immigrants in 1970 5. Immigrants arriving after 1960, in 1970 6. Change in relative wages of recent immigrants (1-5) 7. Change in relative wages of all immigrants (2-4)
- ,500 - ,640 - .765 - ,422 - ,210 - .278
- ,491 - ,395 - ,537 - .662 - ,291 - ,159 - ,184
- ,284 - ,244 - ,257 - ,419 - ,078 - .046 - ,036
- ,438
- ,352
- ,208
- .204
- .I39
- ,075
-.I44
-.113
- ,042
- ,642
Sources: Public Use Files from the 1970 and 1980 Censuses. For sample selection criteria, see the Appendix. Note: Estimates in the table are differences between geometric mean wages for immigrants and white males within each SMSA. SMSAs are ranked by the share of the male labor force accounted for by immigrants who arrived between 1970 and 1980. “Top third” refers to SMSAs with the largest immigration rates that together account for one-third of all post-1970 immigrants. “Middle third‘’ refers to those SMSAs with the next largest rates of immigration that together account for another third of all new immigrants. Finally, the column labeled “bottom third” refers to SMSAs with the lowest rates of immigration that together account for the remaining one-third of new immigrants.
very poor substitute for new immigrants, who enter the United States with skills (e.g., language and institutional knowledge) that typically are less valued in the American market. Yet over time the immigrants assimilate. In our analysis, this assimilation entails greater ease of substitution between an immigrant cohort and native workers as their time in the United States accumulates.6 Thus, substitution between old and new immigrant cohorts is also imperfect. We further expect these intercohort substitution effects to dominate those between new immigrants and (most) native workers. The following model formalizes these ideas and serves to guide the subsequent empirical work. We assume the existence of a large number of geographically distinct labor markets.’ Immigrant and nonimmigrant labor are combined in a concave local production function represented by
In equation (l), Y , refers to total output produced in locale c (empirically an SMSA), and M,. is total human capital supplied by labor aggregatej in locale c. In our discussion, we assume that the labor aggregates in g(.) include immigrant arrival cohorts ( j = 1, . . . , k - 1) plus nonimmigrant labor ( j =
176
Robert J. LaLonde and Robert H. Topel
k) as inputs, though empirical implementation requires further judgments about substitution possibilities. Thus, some natives who are thought to be close substitutes for immigrants-young Hispanics or blacks, for examplecan be included as separate factors. Another possibility is to allow immigrant groups of different ethnicities to form separate inputs in the production function. The (weak) separability assumption in equation (1) is maintained throughout. Given our specification of g(.), h(.) contains capital and other resources that are incidental to the analysis. The parameters 8, and acare locale-specific factor-neutral shifters of the effective quantities of labor and other factors. For example, these shifters may represent forces that shift the local demand for labor. Given varying sizes of cities to which the model may be applied, a plausible assumption about g(.) is that it has constant returns, so that doubling all labor quantities leaves relative wages unchanged within any locale.8 Assume for the moment that each member of arrival cohort j supplies one unit of relevant human capital. Then the marginal product (wage) of group j workers at locale c is
(2)
w, = F1(.)8,gj(Mc1,. .
f
9
MC,h
where subscripts to functions denote partial derivatives with respect to the indicated argument. The separability assumption in equation (1) implies that other inputs enter the marginal product of labor only through F , ( . ) . Thus, shifts in 8,, a,,or other nonlabor components leave relative wages for labor inputs unchanged. Given (2), the log wage of groupj workers in locale c is
(3)
wcj = In [F,(.)8,1 + In gj(Mc,, .
. . ,Mck).
The first term on the right-hand side of equation (3) is an area-specific term and is independent ofj; it is fixed for all labor inputs within a locale. Equation (3) is the basis for our empirical analysis. The first step toward an empirical specification of (3) is to replace In [F,(.)8,]with an area-specific fixed effect, p, and to expand In gj(Mc,,. . . , Mck)to first order in logs: (4)
p, + &yji In M c i . 1, 2, . . . , k ) are “elasticities of complemen-
wcj =
In (4), the parameters yji (i = tarity” (8 In W, / 8 In M C i )that satisfy Xi yji = 0 if there are constant returns. Aside from this homogeneity condition, the only restriction implied by theory is y, < 0-an increase in the supply of groupj workers reduces their wage.9 However, if j indexes cohorts by their time in the United States, we expect yij < 0 (for i # j ) with effects that dissipate as I j - il increases. In the language of demand theory, adjacent immigrant cohorts should be q-substitutes (see Hamermesh 1986). In other words, recent immigrants offer the greatest substitution possibilities for new immigrants, so the yij will trace out an assimilation profile of wage adjustments. Complementarity is also a possibility
177
Labor Market Adjustments to Increased Immigration
( y i j > 0). For example, a large enclave of past immigrants may improve market opportunities for new immigrants, especially when language and cultural ties are important. These restrictions are tested below. To complete the empirical specification, we drop the assumption that each individual contributes a single unit of human capital to the stock Mj. We assume that an individual’s stock of human capital, m, depends on his characteristics, X, so that, for person 1 in cohortj and city c, mcjl = exp(X,G Pj Po E,~~}.The cohort effects Pj allow for both assimilation (earlier cohorts have acquired skills relevant to the U.S.market) and differences in the quality of immigrants over time. Similarly, the “origin effects” Pocontrol for differences in average immigrant characteristics between broadly defined places of origin. With this assumption, the log wage of individual 1 is
+ +
+
wcjl =
P,
+ Pj + Po + X,G + Ziyji In Mci + E,~,.
Three points about (5) are noteworthy. First, the appearance of locale effects (p,) in (5) implies that the yji’s capture shifts in the relative earnings of different immigrant groups within a locale that are induced by changes in the relative shares of immigrants. More precisely, with fixed city effects, the estimable substitution parameters are yji - y k i ,j = 1 , . . . , k - 1. Since k refers to native workers as an aggregate, our maintained assumption is that yki = 0; changes in the stocks of immigrants do not affect the wages of the typical native worker, so relative wage adjustments capture the effects of interest.I0For example, y I 1< 0 implies that, in a locale where the market share of new immigrants is large, wages of new immigrants will be low relative to the earnings of other workers in that area. Thus, our analysis examines the effect of immigration on rotations of the assimilation profile of immigrant wages within locales. Sample selection due to unobservable differences across areas in immigrant “quality” will not affect our results. For the same reason, controlling for locale effects implies that our results are not affected by differences in demand conditions, local amenities, or the cost of living across markets, so long as these conditions have factor-neutral effects on the wages of separate labor categories within a locale. Therefore, demand-induced shifts in immigration to a locale are not an issue unless they have differential effects on certain immigrant cohorts. A second noteworthy point about equation (5) is that, while the estimated area effects, pC,subsume wage adjustments for each locale, it is still true that an increase in the total supply of immigrants will normally reduce immigrant earnings relative to those of nonimmigrants. These relative wage adjustments can be evaluated from ( 5 ) . Because the cross-substitution effects yji (for j # i ) typically will be nonzero, an increase in the supply of all immigrants may have a larger negative effect on immigrant earnings than would be implied by own substitution effects (y,) alone.” Finally, equation (5) controls for cohort (time of arrival) effects directly, so that differences in immigrant quality over time do not influence the estimates
178
Robert J. LaLonde and Robert H. Topel
of yji. For example, if recent immigrant cohorts are less skilled than their predecessors, a model like (5) that did not control for time in the United States might attribute the entire decline of relative earnings among recent immigrant cohorts to the increased relative supply of new immigrants. Model ( 5 ) is not subject to this bias as long as within-cohort average quality is neutral with respect to locale. Similar arguments apply to the presence of place of origin effects, Po. 6.2.1 Relative Wage Adjustments within Areas Despite these controls, model ( 5 ) arguably is inappropriate since mobility of either natives or other factors may arbitrage geographic wage differentials in the long run. Differences in immigrant shares can persist in equilibrium, but, if factors are mobile, these differences have no implications for wage differentials. This argument is less persuasive when applied to short-run adjustments to changes in the flow of immigrants. Our evidence in section 6.1 documented the large increase in the flow of immigrants during the 1970s and showed that the direction of this flow was mainly toward existing immigrant enclaves. We assume that these facts represent an exogenous increase in the supply of immigrants to these areas and that the effects of this supply shift on wages cannot be arbitraged in the short run by mobility of other factors. This suggests a comparison of within-area wage changes between the 1970 and the 1980 Censuses in response to changes in the stock of immigrants. More formally, we effectively difference ( 5 ) within areas by including city by cohort effects in the model:’*
where t indexes Census year (1970, 1980). In equation (6), differences in immigrant earnings across areas are subsumed in the P,’s, which vary by area and entry cohort (but not by year). In this model, parameters yi, are identified from within-area changes in relative immigrant shares over time. For example, y I , < 0 implies that areas experiencing an increase in the share of new immigrants over the decade will also show declining wages of new immigrants relative to other workers in those locales.
6.3 Empirical Results: The Effects of Immigration on Wages and Earnings 6.3.1 Results from the 1980 Census In this section, we report parameter estimates for versions of models (5) and (6). The basic sample consists of 26,681 immigrants derived from the 1% Sample of the 1980 Census. Immigrant arrival cohorts in these data are de-
179
Labor Market Adjustments to Increased Immigration
fined by date of arrival in the United States as recorded by the Census. The six identifiable cohorts are immigrants with zero to five, six to ten, eleven to fifteen, sixteen to twenty, twenty-one to thirty, and more than thirty years in the United States. All our results are for men between the ages of 16 and 64 who were labor force participants (employed or unemployed) at the time of the Census survey (roughly, April 1980) and who had positive earnings during the previous calendar year. These men resided in 119 SMSAs (listed in the Appendix). Details of selection criteria, variable definitions, and summary statistics are also appended. The dependent variable in all the models estimated below is the natural logarithm of the average weekly wage (annual earnings divided by weeks worked) for each individual.l 3 Judgments about which labor aggregates to include in the model determine the base group against which relative earnings adjustments of immigrants are measured. We have tried several aggregation schemes, with very similar results, of which two are noteworthy. First, when the base group is defined to be natives as an aggregate, the estimable substitution matrix r = [y,,] has forty-two indcpendent elements. Using this model, the substitution effects that we estimate are quite small. Further, we were unable to reject (either jointly or individually) the hypothesis that y6, = 0, i = 1, , . . , 6, in this case, which indicates that the wages of immigrants with more than thirty years in the United States are fully assimilated insofar as substitution effects are concerned. This suggests a second aggregation scheme that restricts attention to immigrants only and where the normalizing group is immigrants with more than thirty years in the United States. In this case, r contains thirty independent elements. This sample produced slightly larger estimates of the effect of immigration on relative wages. Since our main finding is that these effects are small in all relevant cases, we report only the results using the second approach. l4 Column 1 of table 6.5 reports the estimated diagonal elements of r = [y,,] from a completely unrestricted model. Taken literally, these “own” effects imply that a 10 percent increase in the flow of new immigrants to an area would reduce new immigrant weekly wages by about 1 percent ( - .098 x . l ) relative to immigrants in the United States for more than thirty years.I5This estimate is not very precise, and, in the unrestricted models of columns 1-3, it is the only effect that is larger than its standard error. Off-diagonal terms in r (not reported) are also imprecisely estimated. In part, this imprecision reflects a vain effort to estimate the thirty free parameters of r from immigrant shares in only 119 SMSAs. The problem is colinearity. One way to impose further structure and summarize the overall effect of immigration on wages is to estimate the effect of a proportional increase in the size of all immigrant groups. Since d In w = T d In M ,the estimated effect on each cohort is simply the row sum from the substitution matrix, r. Estimates of these effects for the unrestricted models of columns 1-3 of table 6.5 are shown in table 6.6. In general, these estimates imply that immigration reduces wages, especially among
180
Robert J. LaLonde and Robert H. Topel
Table 6.5
The Effects of Immigration on the Wages of Immigrants (dependent variable is log average weekly earnings of males in 1979) Restricted Version: Cross Effects Constrained to Zero
Unrestricted Version: All Parameters Free (3) Own effects: y:, Years in the U.S.: 0-5 6-10
11-15 16-20 21-30 Arrival cohort effects: Years in the U.S.: 0-5 6-10
11-15 16-20 21-30 Regression includes: Cross effects: years in U.S. interacted with other city cohort shares Occupation controls Industry controls Place of origin controls
- ,098 C.043) - ,052 (.053) - ,001 .018 (.045) .050
(.065) - ,519 (.106) - .292 (.107) - ,294 (.109) - ,180 (.118) ,013 (. 105)
- ,099
(.M) - ,064
- .099 - ,072
,061 (.066)
(.055) .006 (.049) ,031 (.046) ,064 (.OW
- ,567 (.107) - .375 (.log) - ,346 (.111) - .216 (.119) - ,032 (.107)
- ,579 (.108) - ,375 (.109) - ,333 (.112) -.189 (.120) - ,012 (. 108)
(.054) - ,001 (.049) .017
(.ow
no yes yes
no no Yes
(7)
(4)
- .045 (.014) - .045 (.014) - .031 (.014) - .002 (.015) .020 (.020)
- .376
- .047 - ,051 (.014) - ,036 (.014) - .002 (.015) ,015 (.021)
- ,412
- .414
.076 (.071)
(.051) - ,316 (.050) - .201 (.053) - ,019 (.059) ,042 (.072)
no
no
(.050)
- .267 (.049) -.164 (.052) - ,006 (.058)
- ,045
(.015) - ,047 (.014) - ,033 (.015) ,003 (.015) ,021 (.021)
(.015)
(.051)
- ,312
- ,045 (.015)
- ,050 (.014) - ,037 (.015) - ,005 (.015) .014 (.021)
- ,443 (.051) - ,347
(.050)
(.050)
- ,186 (.053) - .003 (.059) ,065 (.073)
- .231 (.053) - .041 (.059) .039 (.072)
no
no
no no Yes
no Yes no
Nore: Regressions control for years of schooling, potential experience and experience squared, race, the presence of children, two marital status dummy variables, and a dummy variable for a disability that limits a person’s work. There are nine occupation controls, eighteen industry controls, and six place of origin controls: Europe, Canada, Australia, and New Zealand; Asia; the Middle East; Mexico; other Latin America; and other immigrants. Standard errors are in parentheses. The cross-substitution estimates associated with columns 1-3 are not reported in the table.
more recent arrivals: a sustained increase in immigration of the indicated magnitude would reduce the wage of new arrivals by about 9 percent, with smaller but still substantial effects on the wages of earlier cohorts.I6 Only the earliest arrivals (twenty-one to thirty years) are insulated from relative wage adjustments.
181
Labor Market Adjustments to Increased Immigration Estimated Effects on Log Weekly Wages of a Proportional Increase (d In Mi = 1) in All Immigrant Cohorts, Unrestricted Substitution Effects, 1980
Table 6.6
Years Since Immigration Model 1 2 3
0-5
6-10
11-15
16-20
- ,091
- ,046
- ,054
- ,040
(.028) - ,096 (.029) - ,098 (.029)
(.029) - .059 (.029) - ,057 (.030)
(.030) - ,065 (.030) - ,060 (.031)
(.032) - ,046 (.032) - ,039 (.033)
2 1-30 ,006 (.028)
.o (.029) ,004 (.029)
Nore: Calculated from estimated substitution matrix for the unrestricted models in columns 1-3 of table 6.5. For each immigrant group, the estimated effect of a proportional increase in all immigrants in a local labor market is the sum of coefficients in the corresponding row of the substitution matrix. For other controls in each model, see table 6.5. Standard errors are in parentheses.
Columns 4-7 of table 6.5 report more parsimonious specifications that constrain the off-diagonal terms of r to zero. In these specifications, arrival cohorts are assumed to be independent inputs in local production, so there are no intercohort crowding effects on wages. In these models, a larger own labor force share for an immigrant cohort tends to reduce wages for that cohort but has no effect on other cohorts. These own-substitution effects also tend to die out as time in the United States accumulates. Thus, there appears to be significant crowding among recent arrivals, but the effects of own cohort size dissipate over time. The parsimony of the specification in columns 4-7 of table 6.5 was purchased with a substantial loss of generality: cross-cohort substitution was assumed away. We next reintroduce these substitution effects with additional structure. We hypothesize that, for each immigrant cohort, cross effects are smaller than own effects (yjj < yji for i f 1) and that these substitution effects dissipate as [i - j ] increases. That is, members of adjacent arrival cohorts are better substitutes than are members of distant ones. Under this hypothesis, in each row of r the largest negative element is along the diagonal, while other effects should be smaller moving away from the diagonal in either direction. To test this hypothesis, we allow (7)
yji = 7,
+ AAi - jl,
j = 1 , 2,
. . . ,5.
If adjacent cohorts are imperfect substitutes, then we expect yjj < 0 and Aj > 0, with [Aj I < ly,l. The linear restrictions (7) reduce the number of estimated substitution parameters from thirty to ten while retaining substantial flexibility. In fact, the restrictions imposed in equation (7) cannot be rejected-either individually or jointly-in any form of the model that we have estimated. Estimates based on (7) are shown in table 6.7 for various combinations of other controls.
182
Robert J. LaLonde and Robert H. Topel
Table 6.7
The Effects of Immigration on Wages, Linear Restrictions on Intercohort Substitution (dependent variable: log average weekly earnings in 1979) (1)
Own Cross Effects: Years in the U.S.: 0-5 years: Own effect (y,,)
Cross effect (A,) 6 1 0 years: Own effect (y,,)
Cross effect (A,) 11-15 years: Own effect (yaJ
Cross effect (A?) 1 6 2 0 years: Own effect (y4)
Cross effect (A,) 21-30 years: Own effect (y,,)
Cross effect (A,) Regression includes: Occupation controls Industry controls Place of origin controls Arrival cohort controls
(2)
(3)
(4)
- ,030
(5)
- .032 (.all) ,011 (.005)
- .035 (.012) .012 (.005)
- ,032 (.012) .011 (.@35)
- ,037 (.012) ,018 (.@37)
- ,038 (.012) ,017 (.@37)
- .036
(.012) ,016 (.007)
(.012) ,016 (307)
- ,010 (.015) ,003 (.011)
- ,013 (.015)
- ,009 (.016) ,002 (.01)
- .012 (.016) ,003 (.011)
- ,015
,015 (.019) - ,011 (.012)
,015 (.019) - ,012 (.012)
.01 (.02)
- .01 (.01)
.013 (.019) - .010 (.013)
.011 (.019) - .010 (.012)
,027 (.014) - .013 (.007)
,028 (.015) - ,011 (.007)
.03 (.01) - ,013 (.007)
.027 (.015) - ,013 (.@37)
,026 (.015) - ,013 (.@37)
Yes
no
no
no
no
Yes Yes
Yes Yes
no Yes
no no
Yes no
Yes
Yes
Yes
Yes
Yes
,004
(.011)
(.012)
,010 (.@35) - .036
- ,033 (.012) .011 (.005) - ,037 (.012) ,017 (.007) (.015) ,005 (.011)
Nore: See note to table 6.5. The own effects (y,) for each cohort j are unrestricted. The crosssubstitution effects are restricted to follow y, = y, + A, (i - jl, where i indexes the time of arrival of cohort i relative to cohort j .
The key finding in table 6.7 is that own effects of cohort size (yjj) are negative and significant for recent arrivals, while cross-cohort substitution effects die out with the difference in time of entry to the United States (A > 0). Especially for recent arrivals, we find that adjacent cohorts are q-substitutes.” Both the own and the cross effects of cohort size tend to diminish with years since entry. We take this finding as evidence of assimilation: the effect on
183
Labor Market Adjustments to Increased Immigration
wages of a large cohort is diluted as immigrants melt into the broader market of native workers. To pursue this point, we add additional structure to (7) by assuming that
Together, restrictions (7) and (8) express the form of the substitution matrix in terms of just four parameters. If assimilation means increasing substitution between past immigrants and the labor market as a whole, as time in the United States accumulates, we expect y < 0, p > 0, A > 0, and < 0. Furthermore, the parameters should also satisfy 141< IyI and (A1 < IyI if there is (imperfect) substitution among immigrant cohorts. Estimates of this parameterization of the substitution matrix are shown in table 6.8 for three illustrative sets of other controls. Other specifications differ trivially from these. All parameters are of the anticipated signs and relative magnitudes, and they are significantly different from zero by conventional standards. The reported F-statistics in the table test the four-parameter structure given by (8) against the unrestricted, thirty-parameter model of r. Remarkably in a sample of this size, the restrictions are never rejected.'* Thus, (8) offers a good summary of the data. The estimates imply that an increase of roughly 170 percent (d In M = 1) in the stock of new immigrants would reduce the relative weekly wages of new immigrants by about 3 percent (y < 0). The immediate effect on earlier immigrant cohorts of this increase would be smaller (A > 0). As time in the United States accumulates for an arrival cohort, the earnings disadvantage caused by being a member of a large cohort evaporates (p. > 0), as do the cross effects of cohort size on adjacent
+
The Effects of Immigration on Wages: Linear Restrictions on Own and Intercohort Substitution (dependent variable: log average weekly earnings in 1979)
Table 6.8
Parameter Model
Y
A
CL
4
F-Statistic for Restrictions
Nore: Model 1 controls include cohort, origin, occupation, and industry effects in addition to the
demographic controls listed in table 6.5. Model 2 drops occupation from the set of controls, and model 3 drops industry and occupation. The reported F-statistics test the restricted four-parameter model relative to the completely unrestricted model with thirty parameters. Dependent variable is log weekly wages; standard errors are in parentheses. For definitions of parameters, see the text.
184
Robert J. LaLonde and Robert H. Tope1
arrivals (4 < 0).I9All these estimates are consistent with immigrant crowding in local markets, tempered by assimilation and imperfect substitution. 6.3.2 Annual Earnings versus Wages: Do Quantities Matter? The analysis to this point has focused only on market clearing price adjustments with inelastic labor supply. However, if immigration also causes quantity adjustments in terms of unemployment, hours, or weeks worked, then annual earnings may be a more appropriate measure of welfare. A detailed analysis of adjustments on each of these margins is beyond the scope of this paper (see Altonji and Card, in this volume). Yet quantity and price adjustments are likely to be correlated, so the effects of immigration on annual earnings may be larger than on wages. The estimates in tables 6.5-6.8 would then underestimate the distributive effects of immigration. To explore this possibility, table 6.9 reproduces the estimates in tables 6.6 and 6.8 when log annual earnings in 1979 instead of log average weekly earnings is used as the dependent variable. The estimates from the unrestricted model of the effect of a proportional increase in all immigrant groups, in part A of the table, are slightly larger than the corresponding estimates in table 6.6 (the most recent arrival cohort is an exception). For the specification in row 2 of table 6.9, part A, the effects on earnings exceed those on wages, on average, by about a third, though the standard errors are large enough that equality of effects cannot be rejected. Thus, it appears that the main distributive effects of immigration operate through price flexibility rather than through adjustments in unemployment or participation. This conclusion is reinforced by a comparison of the estimates in part B with those of table 6.8, which report restricted estimates of substitution parameters. The estimates for wages and annual earnings differ only trivially. On this evidence, we conclude that the main actor in market adjustments to immigration must be wage flexibility. Adjustments in unemployment or participation are negligible. 6.3.3 The Effects of Immigration on Young Native Blacks and Hispanics
To this point, we have treated all nonimmigrants as a single aggregate, while focusing on substitution possibilities among immigrants. For these groups, the effect of immigration on measures of welfare are quite small. Even so, some groups of native Americans may be more sensitive to the crowding effects of immigration than others, and for them the implied redistributive effects are of some concern. Here, we focus on two identifiable groups who may face the most important crowding effects of immigration: young (aged 16-34) blacks and Hispanics. We treat young blacks and Hispanics as separate inputs that interact with immigrants in local production (see eq. [ 13 above). The unrestricted matrix of estimated substitution effects now contains fifty-six parameters, and it is not very informative. As above, we may calculate the effect of a scale (d In M i=
185
Labor Market Adjustments to Increased Immigration
Table 6.9 A. Effects on Log Annual Earnings of a Proportional Increase (d In M i= 1) in All Immigrant Cohorts, Unrestricted Substitution Effects (dependent variable: log annual earnings in 1979) Years Since Immigration Model
0-5
6-10
11-15
16-20
2 1-30 ~~
1 2 3
- .089
(.032) - .093 (.033) - ,091 (.033)
- .064 (.032) - .079 (.033) - ,074 (.033)
- .077 (.034) - ,085 (.034) - .078 (.034)
- ,066 (.036) - .071 (.037) - ,062 (.037)
~
- ,008 (.032) - .014 (.033) - ,008 (.033)
B. The Effects of Immigration on Earnings: Linear Restrictions on Own and lntercohort Substitution (dependent variable: log annual earnings) Parameter Model 1
2 3
Y - .026 (.008) - ,030 (.009) - ,028
(.ow
A ,008
CL
.009
(.ow (.ow ,009 .011 (.ow (.ow ,009 .011 (.ow (.ow
4
F-Statistic for Restrictions
- ,003
,721
- ,004
,756
- ,004
,822
Note: See notes to tables 6.6 and 6.8. Standard errors are in parentheses.
1) increase in all immigrant cohorts on the wages or earnings of blacks and Hispanics. These estimated effects are shown in part A of table 6.10 for two specifications of the model.20Overall, there is only weak evidence that immigration reduces the wages and earnings of these natives. The largest estimates that we obtained are shown in row 1: the point estimate of the effect of a 170 percent increase in the size of all immigrant cohorts on black wages is only 2.4 percent, though the estimate is smaller than its standard error. The corresponding estimate for Hispanics is less than 1 percent. Surprisingly, in light of our previous results, the effects on earnings are slightly larger than on wages. Thus, there is some evidence of reinforcing adjustments on time worked, especially among young blacks. Again, however, these effects are not precisely estimated. An alternative strategy for examining these effects is to impose the restrictions given by (7) and (8) on the matrix of intercohort substitution terms among immigrants, while leaving own and cross effects for blacks and Hispanics as free parameters. To impose some structure, we allow black and Hispanic wages to be affected separately by immigrant cohorts that arrived before and after 1965. The hypothesis is that crowding effects of immigration are
186
Robert J. LaLonde and Robert H. Tope1
Table 6.10
The Effects of Immigration on Wages and Earnings of Young Blacks and Hispanics
A. The Effects of a Proportional Increase in All Immigrant Cohorts (unrestricted models) Effect On: Model
Black Wages
Black Earnings
Hispanic Wages
1
- .024
- ,059
- .009
2
(.030) - .020 (.030)
(.035) - ,046 (.036)
(.032) - ,008 (.032)
Hispanic Earnings
- ,015 (.037) - .012 (.038)
B. Estimated Cross Effects of Immigrants on Blacks and Hispanics (linear restrictions imposed) Effect on Blacks of an Increase in:
Model Earnings: 1 2 Wages: 1
2
Native Blacks
Post-1965 Immigrants
Effect on Hispanics of an Increase in:
Pre-1965 Immigrants
Native Hispanics
Post-1965 Immigrants
Pre-1965 Immigrants
- .042 (.018) - .028 (.018)
- ,005
- ,014 (.018) - ,008 (.017)
,015 (.010) ,018 (.OlO)
- ,025
- ,042 (.015) - ,031 (.015)
,008 (.010)
- ,020 (.015) - ,014 (.015)
.007
- ,013
(.OlO) ,010 (.009)
- ,016
(.012)
- ,008
(.012)
,005 (.010)
(.015)
- ,030 (.015)
(.013) (.013)
Note: Part A parameter estimates refer to the effect of a unit change in log employment of all immigrant cohorts (d In M,= 1 for all i) on log wages or earnings of blacks and Hispanics. Part B estimates represent the effect of a unit change in log employment of the indicated group. Model 1 contains all demographic controls listed in table 6.8. Model 2 adds industry and occupation controls. Standard errors are in parentheses. Part B models constrain intercohort substitution matrix for immigrants to follow eqq. (7) and (8). Black and Hispanic effects are free parameters.
concentrated on these demographic groups and that recent immigration is the most important factor. We report (in pt. B of table 6.10) the own effects for both blacks and Hispanics as well as estimated cross effects with immigrants. In each case, we find crowding effects of blacks and Hispanics on their own wages; increases in the labor force shares of these groups reduce their wages, though only the estimate for blacks is significant. We also find that recent immigrants are substitutes for young blacks, though the effect is small ( - .01 is the largest estimate we obtained) and imprecisely estimated. It is substantially smaller than the own effect on black wages ( - .042). The estimates for Hispanics are more mixed. Finally, for neither group do we find important differences between the wage and the earnings estimates, suggesting that employment and hours adjustments are also minor concerns. Overall, these estimates do not suggest to
187
Labor Market Adjustments to Increased Immigration
us that immigration is a prime factor affecting labor market outcomes for these young natives. 6.3.4
Results from the 1970 Census
According to Census data, immigration to the United States in the 1970s was roughly double its level in the 1960s. Because this sharp increase in the flow of new immigrants was highly geographically concentrated (see sec. 6. l ) , it is plausible that short-run labor market adjustments would generate a stronger relation between immigration and relative wages in the 1980 Census than in the 1970 Census. We examine this point in table 6.11, which summarizes estimates of the substitution effects from the 1970 Census. Because the story is not much different in these data, we report only the substitution effects of a proportional increase in all immigrant cohorts from the unrestricted model (eq. [5]) in part A of table 6.11 and the restricted form of intercohort substitution effects (eq. [S]) in part B. Table 6.11 A. Effects on Relative Log Weekly Wages of a Proportional Increase in All Immigrant Cohorts (unrestricted substitution effects, 1970) Years Since Immigration Model 1
2 3
0-5 ,012 (.022) .008 (.023) ,009 (.023)
6-10
11-15
16-20
2 1-30
- ,024
- ,007
- ,017
.ow
(.024) - ,025 (.024) - ,028 (.024)
(.027) - ,003 (.028) - ,006 (.028)
(.028) - ,024 (.029) - .017 (.029)
(.024) ,012 (.025) ,015 (.025)
B. Estimated Substitution and Assimilation Parameters for Log Weekly Wages of Immigrants (linear restrictions imposed, 1970) Parameter Model 1 2 3
Y - ,019
h
P
,008
,001
(.007)
(.003) .008 (.003)
- ,021 (.007)
,008 (.003)
- ,020
4J
F-Statistic for Restrictions
- .0005
,861
- .o004
,834
(.001)
,001
(.001)
,001
- .oO04
,814
(.001)
Nore: Part A calculated from estimated substitution matrix for unrestricted models analogous to those in columns 1-3 of table 6.5. These results are comparable to those in table 6.6. For other controls in each model, see table 6.5. Calculations are based on a sample of 17,158 immigrants in 119 large SMSAs from the 1970 Census. Standard errors in parentheses. For part B, see notes to table 6.8. Dependent variable is log weekly wages; standard errors are in parentheses. The results when the dependent variable is log annual earnings are similar.
188
Robert J. LaLonde and Robert H. Tope1
The estimates in part A should be compared to the corresponding estimates for 1980 in table 6.6. Whereas the 1980 estimates implied sharply lower earnings among new immigrants, the corresponding estimates for 1970 are negligible. For earlier arrivals, the estimates are negative though generally smaller than in 1980, and none are significant by conventional standards. These points are also apparent in part B; while all the substitution relations take the anticipated sign, only y is significant. The key point is that all these effects are substantially smaller than in the 1980 data (see table 6.8). The relation between the estimates generated by the 1970 and 1980 cross sections raises an important issue. Did the increased immigration of the 1970s generate the substantial crowding effects that seem to show up in the 1980 cross section? To answer this question, we create a pseudo-panel from the combined 1970 and 1980 Census files and analyze within-market changes in immigration, wages, and earnings. 6.3.5 Panel Estimates: Relative Wage Adjustments within SMSAs, 1970-80 The preceding econometric results rely on cross-sectional differences in labor force shares to generate price adjustments. Since labor is mobile in the long run, the existence of these wage differentials appears inconsistent with spatial equilibrium, so our interpretation of these results may be suspect. In light of this problem, we estimate equation (6), which pools the data from the 1970 and 1980 Censuses. We add to the model six hundred fixed effects that control for entry cohort (time in the United States) within each SMSA. Thus, the variation used to estimate substitution effects occurs over time and within SMSA-cohort cells. In effect, we ask whether areas that experienced unusually rapid immigration over the decade also experienced falling relative wages and earnings of recent immigrants and whether there were spillover effects of these changes on other groups.21 Results are summarized in tables 6.12 and 6.13. In table 6.12, we report models for the determination of log weekly wages and annual earnings that constrain intercohort substitution effects to follow (7). Each row of the substitution matrix is summarized by two parameters: an “own” effect of increasing cohort size on members of the cohort and a cross-cohort substitution effect that allows each cohort to have the largest effects on adjacent arrival cohorts. As above, we expect the former effect to be negative and the latter to be positive. The results are surprisingly similar to the cross-sectional estimates (see table 6.7), though standard errors are somewhat larger. In four of five cases, the point estimate of the own effect of cohort size is negative, with smaller effects on adjacent cohorts. Differences between the estimates for log weekly wages and annual earnings are small, which indicates again that the main effects of immigration are on wages rather than employment (weeks worked). Furthermore, the estimates show a tendency to “die out” as time in the United
189
Labor Market Adjustments to Increased Immigration
Table 6.12
Wages Changes within Locales: The Effect of Immigrationon Changes in Wages and Earnings within SMSAs, Linear Restrictions on Intercohort Substitution, 1970-80 Dependent Variable Log Weekly Wage
Cohort: Years Since Immigration 0-5: Own effect
Cross effects
6-10: Own effect Cross effects 11-15: Own effect
Cross effects
16-20: Own effect Cross effects 21-30: Own effect
Cross effects Origin effects Cohort x SMSA effects Industry effects Occupation effects R2
(1)
- ,039 (.014) ,018
- .041 (.014) ,020
- .007
(2)
- ,034 (.014) ,016 (.006)
- ,036 (.014) ,018
- .009
Log Earnings (3) - ,049
(.016) .021
(.ow
- ,065
(.015) .036 (.009)
- ,002
(4)
- .045 (.015) ,020 (.006)
- ,061 (.015) ,020
- ,003
(.013) ,006 (.009)
(.013) ,007
(.014)
(.009)
(.010)
(.014) .003 (.010)
,032 (.018) - ,011 (.011)
,027 (.018) - ,008 (.011)
,047 (.021) - .019 (.013)
,042 (.021) - ,015 (.012)
- .010
- .010
- ,006
- .005
(.015) ,003
(.014) .001
,002
(.016) (.007)
(.016) .001 (.007)
,002
Yes Yes
Yes Yes
Yes Yes
Yes Yes
no no
Yes no
no no
Yes no
,258
,272
.243
,257
Note: For other regressors, see note to table 6.5. The models include a dummy variable for 1980. Standard errors are in parentheses.
States accumulates: effects of within-city changes in shares are stronger for more recent arrivals. In light of the last point, table 6.13 shows estimates for the most parsimonious specification, which restricts substitution terms to follow (8). These “panel” estimates should be compared to the cross-sectional results reported
190
Robert J. LaLonde and Robert H. Tope1
Table 6.13
Wage Changes within Locales: The Effect of Immigration on Changes in Log Wages and Earnings within SMSAs: Linear Restrictions on Own and Cross-Substitution Effects, 1970-80 Log Weekly Wage (1)
(2) - ,020
(.009)
,012 (.004) ,003 (.002) - ,002 Origin effects Cohort x SMSAeffects Industry effects Occupation effects R2
Log Earnings (3)
(4)
- ,029 (.014) ,016 (.004) ,005
- ,027
- ,002
- ,002
(.010) ,015
,005
(.001)
(.001)
(.001)
Y e5 Yes no no
Ye5 Yes Yes no
Yes Yes no no
Yes Yes Yes no
.257
,272
,243
.256
Note: See note to table 6.8. Standard errors are in parentheses. N = 44,004
in tables 6.8 and part B of table 6.9. In light of our previously stated concerns, we are surprised that the panel and cross-sectional results are almost identical. All parameters are of the anticipated signs, with relative magnitudes that accord with theory. Our point estimates imply that a rough tripling (d In M = 1) of the rate of new immigration to an area would reduce the relative wages and earnings of new immigrants by 2-3 percent. Again, this crowding effect of membership in a large cohort dies out as U.S. experience accumulates, which indicates assimilation. Effects of new immigration on previous immigrants are smaller than the direct effects, which is indicative of imperfect substitution. 6.4
Conclusion
This paper has examined the effect of immigration on the labor market. Our basic finding is that increased immigration reduces the wages and earnings of immigrants and their close substitutes, though in our view the effects are not large. For immigrants themselves, a sustained doubling of the rate of new immigration may reduce relative earnings of new immigrants by about 3 percent, but even this effect tends to die out over time as immigrants assimilate to the American market. Labor market effects on nonimmigrants appear to be quantitatively unimportant: the wages and earnings of young blacks and Hispanics are not very sensitive to immigration. In short, our estimates imply that immigrants are rather easily absorbed into the American labor market. There
191
Labor Market Adjustments to Increased Immigration
is little here to indicate that the redistributive effects of immigration should be a major policy concern. These conclusions are tempered by at least two points. First, our analysis has relied heavily on differences in wages across geographic areas. These differentials are difficult to rationalize as an element of a long-run equilibrium of the labor market. We argued that the upsurge of immigration in the 1970s was a change in labor supply that generated short-run wage adjustments among areas, and comparison of time-series and cross-sectional results tended to support this assumption. Second, our analysis mainly treated immigrants as a homogeneous group, and so we ignored the effect that specific immigrant groups may have. For example, in light of our results, it is plausible that illegal immigration from Mexico affects mainly young Hispanics. These points deserve attention, but we defer them to later research.
Data Appendix Selection and Construction of Variables The data used in this study were drawn from the 1970 and the 1980 U.S. Census of Population and Housing, Public Use Samples (see U.S. Bureau of the Census 1973, 1983). The samples include males sixteen to sixty-four years old, who were not attending school, who were currently in the labor force at the time of the Census, who had worked for pay during 1979, who were not institutionalized, and who were living in SMSAs identified on both the 1970 and the 1980 Public Use Samples. For 1980, we used the 1%-B Public Use Sample. For 1970, we used the l%5% questionnaire-county Group Public Use Sample. SMSA Definitions During the 1970s, the Office of Management and Budget changed the definitions of many SMSAs based on population and commuting patterns in the 1970 Census. These changes are published in “Standard Metropolitan Statistical Areas” (Office of Management and Budget 1976). We used this information to make the SMSA definitions in the 1980 and 1970 samples as comparable as possible. In principle, there are two ways to make these adjustments: (i) the SMSA definitions in the 1980 sample can be adjusted so that they conform to the 1970 definitions (see Altonji and Card, in this volume); (ii) the SMSA definitions in the 1970 sample can be adjusted so that they conform to those in 1980. Neither Public Use Sample provides enough information so that a user can redefine the SMSA definitions to make the two years exactly comparable. However, for most SMSAs, the changes do not add or
192
Robert J. LaLonde and Robert H. Topel
subtract many persons from the sample. The first procedure (i) is a little more precise, although it leads to a smaller sample size, while the second procedure (ii) is less precise but leads to a larger sample size. We tried both procedures and found that the results were robust to either method. All the results reported in the paper are based on the second procedure, where we redefined the 1970 SMSA definitions to make them comparable to the 1980 definitions. Adjusting the SMSA definitions is difficult because the Public Use Samples do not provide enough information on a household’s county group. Therefore, a user often does not know for sure whether some households are in a particular SMSA after a county (or a portion of a county) has been either added or subtracted between two Census years. In many cases, we drew a random sample of persons from a particular county (or group of counties if this was the finest level of identification) that corresponded to the share of persons in the area that was actually added or subtracted from the SMSA definition. This task is particularly difficult in New England and eastern Virginia. In a few cases, it was simpler and more precise to use the 1970 SMSA definitions as opposed to the 1980 definitions as the standard. This poses no problems for the analysis as the important thing is to have comparable SMSA definitions for the two years. Table 6A.1 presents a list of the 119 SMSAs used in the analysis, along with the shares of all immigrants and recent immigrants in both 1970 and 1980 and the 1980 shares of young (16 to 34 years old) blacks and Hispanics in each SMSA’s employed labor force. Note that the share of employed young blacks seems small in large SMSAs. This fact, however, is due to the concentration of blacks in the central cities. For example, in Chicago, blacks are concentrated in the city, whereas there are fewer blacks in heavily populated suburban Cook, Lake, and DuPage counties. In southern SMSAs, a much larger share of the outlying population is black. Variable Definitions We used two measures of earnings as dependent variables, weekly wages and annual earnings. Annual earnings is the sum of wage and salary income and self-employment income. We excluded persons who reported that their self-employment earnings where negative. Weekly wages are defined as annual earnings divided by weeks worked in 1969 and 1979. Two potential problems with these earnings data are that (i) earnings are reported up to a maximum of $50,000 in 1970 and $75,000 in 1980 and (ii) in 1970 weeks worked is reported in discrete intervals. For practical purposes, the “top coding” problem seems to be minor. In 1980, 1.2 percent of the immigrants, ,1 percent of the young black males, .1 percent of the young Hispanic males, and 1.2 percent of all other native workers had either wage or salary income or self-employment income that was greater than $75,000. In 1970, .6 immigrants had wage or salary income or self-employment income that was greater than $50,000. To resolve the problem in the weeks worked data for 1970, we inputted weeks worked for each person based on the mean
193
Labor Market Adjustments to Increased Immigration
Table 6A.1
Share of Immigrants and Young Native Blacks and Hispanics in Large SMSAs Proportion of Employed Male Labor Force ~~
Immigrants in
Immigrants in
1970
1980
Natives 16-34 Years SMSA
All
Recent Arrivals
AKRON,OH
,045 .046 ,027
,013 ,010 ,012
.030 .040 ,043
,003 ,012 ,021
.03 1 .012 ,005
,002 ,003 .183
,034
,005
,036
,010
,008
,012
,082 ,023 ,009 ,006 ,021 ,066 ,034 .008
,033 ,007 ,004 ,002 .005
,161 ,022 .025 ,017 ,036 ,097 ,033 ,018
,096
,007 .Ooo ,102 ,126 ,041 ,016 ,089 ,123
.036 .Ooo ,002 ,006 ,078 ,077 ,002 ,003
.077 ,001 ,099 ,016 ,020 ,023 .020 ,122 .092 ,056 ,060 .049 ,055 .143 ,047 .013 .058
,012 ,001 ,002 ,005 .036 ,005 ,003 ,007 ,001 ,001 .019 .002 .007 ,006 .002 .228 .029
.019 .035 .02 1 ,021
,014 .001 .040
ALBANY-SCHEN-TROY ,NY ALBUQUERQUE,NM
All
Recent Arrivals
Blacks
Hispanics
ALLENTOWN-BETHEASTON ,PA-NJ ANAHEIM-SANTA ANA-GRDN GVE,CA APPLETON-OSHKOSH ,WI ATLANTA,GA AUGUSTA,GA-SC AUSTIN,TX BAKERSRELD,CA BALTIMORE,MD BATON ROUGE,LA
,018 .010 ,003
,009
,011
.006
,014 .047 .012 ,006
BEAUMONT-FT ARTHURORANGE,TX BINGHAMPTON ,NY-PA BIRMINGHAM,AL BOSTON,MA BRIDGEPORT,CT BUFFAL0,NY CANTON,OH CHARLESTON,SC CHARLOTTE,NC CHATTANOOGA,TN
,014 ,041 ,003 ,089 ,100 .059 ,023 ,014 ,015
.004
CHICAGOJL
,096
CINCINNATI ,OH-KY-IN
,018 ,067 .003 ,012 ,031 ,020
CLEVELAND,OH COLUMBIA,SC COLUMBUS,OH CORPUS CHRIST1,TX DALLAS-FORT WORTH ,TX DAVENPT-ROCK I S MOLINEJA-IL DAYTON'OH DENVER,CO DES MOINES,IA DETROIT,MI DULUTH-SUPERIOR,MI-WI EL PAS0,TX ERIE,PA FLINT,MI
(continued)
,019 .014 ,036 ,021 ,074 ,033 ,196 .025 .045
,001 ,013 ,002 ,029 .033 ,011
,006
.002 .007 .001 .034 ,005 .017 .Ooo .003
,006 ,006
.004 ,006 ,013
.004
,016 .002 .049 ,005 ,017
,023 ,034 ,010 ,095 ,086
,064
.018 ,015 ,022 ,010 ,132 ,024 ,059 .014 .021 .063 .054
,027 ,018 ,046 ,027 ,065 ,023 ,233 .017 .014
,013 ,012 ,003 .037 .026 .017
,004 ,004 ,010 ,007 .067 ,007 .018 .006 .009 .026 ,035 ,012
,004
,020 ,020 ,017 ,005 ,127 .005
,004
,068
.Ooo ,006 .019 ,061
.004 ,005 .Ooo .209 ,002 ,010
194
Robert J. LaLonde and Robert H. Topel ~
Table 6A.1
~____
~
(continued) Proportion of Employed Male Labor Force
Immigrants in 1970
Immigrants in 1980
Natives 16-34 Years SMSA
All
Recent Arrivals
Recent All
Arrivals
Blacks
,015
Hispanics
FTLAUDERDALEHOLLYWOOD,FL FRESN0,CA
,053 ,080
.026 ,019
.084 ,137
.034 ,069
.062 ,032
,012 .004
,055 .03 1
,016 .007
,071
,005 .016 ,019 ,114 ,032
.001
.016 ,019 .019 ,095 .I07
.007 .007 .011 .028 ,074
,017 .055 ,026 ,020 .075
,005 .014 .043
,044
.013 .099
GARY-HAMMONSEAST CHICAGOJN GRAND RAPIDS,MI
.026
.026 ,007
GREENSBOReWSTN-SLMHIGH PT,NC GREENVILLE,SC HARRISBURG,PA HARTFORD,CT HOUSTON,TX
,005 .001 ,040
,014
.002 ,001
HUNTINGTON-
LANSING,MI
,007 .013 ,006 ,020 ,193 .009 ,016 .005 ,011 ,023
LAS VEGAS,NV
,040
ASHLAND,WV-KY4H JNDIANAPOLIS,IN JACKSON,MS JACKSONVILLE,FL JERSEY CITY,NJ JOHNSTOWN,PA KANSAS CJTY,M*KA KNOXVILLE,TN LANCASTER,PA
.002
,003
,002 .006 ,128 .Ooo .006 .Ooo .003 .Ooo ,020
.012
.003
,009
.004
,013 ,015
,004
,051 .I64
,001
.019 ,294 ,009 .022 ,018 .023 ,025 .085
,005
,006 ,153 ,004 .009 ,007 ,010
,009 ,036
,074 ,037 .004 ,044
.017 ,003 ,019 ,046
,002 .007 ,059 .002 .008 .002 .012
,012 ,024
LITTLE ROCK-N LITTLE ROCK, AR LORAIN-ELYRIA,OH
,009 ,027
LOS ANGELES-LONG BEACH,CA LOUISVILLE,KY-IN MADISON ,WI MEMPHIS,TN-AR MIAM1,FL MILWAUKEE,WI MNPLS-ST PAUL,MN MOBILE,AL NSHVL-DAVIDSON,TN NEW HAVEN,CT NEW ORLEANS,LA NEW YORK,NY NEWARK,NI NWPTNWS-HAMPTON,VA NFOLK-F"TSMTH,VA OKLAHOMA CITY,OK ORLAND0,FL
.I35 ,008 ,038 ,013 ,264 .039 ,027 .011 .006 ,092 ,032 ,158 ,098 ,010 ,009 .009 .032
,002 ,005
.010
,004
,033
,005
,202 ,004 ,011 ,003 .230 ,009 ,011 ,003 .006 .020 ,020
,067
,266
.001
.008
,014 ,004 .249 ,006 ,008 .Ooo .003 ,029 ,016 ,065 ,036 ,002 ,001 ,003 .016
,033 ,010
.420 ,039 ,027 ,008 .013 ,072 ,044 ,205 ,137 .024 ,023 ,019 .044
.I01 ,064 .007 .009 .011 .021
,102 ,026
.044 ,041
,006 ,168 ,062 ,034 ,009 ,120 ,060 ,043 . I13 ,038 ,059 ,110 ,134 ,043 ,042
,003 .020 ,055
.Ooo ,006 ,001
.025 ,010 .004 .006 ,001 ,008
,013 ,032 .015 ,003 ,005
,009 ,011
195
Labor Market Adjustments to Increased Immigration
Table 6A.1
(continued)
Proportion of Employed Male Labor Force
Immigrants in 1970
Recent Arrivals
Immigrants in
1980 All
Natives 16-34 Years
Recent Amvals
Blacks
SMSA
All
Hispanics
OXNAR>VENTURA,CA
,105
,053
,161
,078
,011
.053
,127 ,009 ,049 .041 ,036 .038 ,072 .026 .015 ,072 ,040
,046 ,001 ,011 ,010 ,006
,177 ,021 ,048 ,059 ,025 ,048 ,085 ,027 .023 ,056 ,031 ,068 ,019 ,204 ,032 ,085 ,138
,084 ,012 ,018 ,025 ,006 ,019 .041 ,012 ,007 ,014 .009 ,021 ,006 ,101
.044 .Ooo
.036 ,065
.033 ,020 ,054 .010 ,022 ,011 ,007 .006 ,128 ,028 ,018 ,020 ,055 ,010 ,007 ,023 .02 1
.027 ,200 .040
,164 ,142 ,107 ,065 .013 .042
.076 .076 .038 .021 .001 ,006
,040 ,017 .016 ,015 ,140 ,007
.032 ,063 .072 ,007 ,008 .006
,021 ,064 ,007 ,015 ,016 ,007 ,011 ,016
.010 ,011 ,020 ,019 ,038 .022
.014 ,066
PATERSON-CLIFTONPASSAIC,NI PEORIAJL PHILADELPHIA,PA-NI PHOENIX,AZ PITTSBURGH ,PA PORTLAND,OR-WA PROVIDENCE, RI READING,PA RICHMOND,VA ROCHESTER,NY ROCKFORD,IL SACRAMENT0,CA ST LOUIS,MC+IL SALINASMONTEREY ,CA SALT LAKE CITY,UT SAN ANTONI0,TX SAN DIEG0,CA
.064 ,016 ,121 .037 ,061 .079
,008 ,025
,008 ,005
,027 .006 ,018 ,004 ,038
,008
,014 ,033
,005
,008
.055 ,001
,006 ,002 .009 ,007 ,005 ,006 .037 ,003 ,058
SAN FRANCISCOOAKLAND,CA SAN JOSE,CA SANTA BARBARA,CA SEATTLE-EVERETT,WA SHREVEPORT,LA SPOKANE,WA
.126 ,098 ,104 ,066 ,007 ,045
,054 ,044
.042 ,020 .003 .012
SPRINGRELHHCPEESTOCKTON,CA
,069 ,121
SYRACUSE,NY
,040
TACOMA ,WA
.049 ,038 ,026 .076 .057
HLYKE,MA
TAMPA-ST PETE,FL TOLED0,OH-MI TRENTON,NJ TUCSON,AZ TULSA,OK UTICA-ROME,NY
.008 .045
WASHlNGTON,DC-MD-VA
.054
WEST PALM BEACH,FL
.066 .012 ,037 .079 ,013 .027
WICHITA,KA WLMNGTN ,DEL-NJ-MD WORCESTER,MA YORK,PA YNGSTWN-WRN ,OH
,028 ,041 ,012 ,011 ,012 ,005 .02 I ,016 ,002 .011 ,024 .029 .001 ,015 ,021 .Ooo ,003
.058 ,123 .040
,036 ,056 ,022 ,078 ,058 ,011 .02 1 ,086 ,090
,026 ,033 ,049 ,019 ,036
,004
,002 ,045 ,034 ,010 ,007 ,012 ,010 ,006
.040
.011 ,030 .013 ,108 .a9 ,030 ,052 ,003 ,011 ,028
,001
.010 ,017
,009
,016 .088 ,005 .Ooo
,006 .014
,008 ,004 ,007 ,001 .003
Table 6A.2
Coefficients from Wage Equation Reported in Table 6.5, Column 1 (dependent variable is log average weekly earnings) Coefficient
Education Experience Experience squared Married Divorced Children Disability Black Hispanic Place of origin: Europe, USSR, Canada, New Zealand, Australia India, South and East Asia Pakistan, Mideast, North Africa Mexico Other Latin America and Caribbean All other areas Occupation: Professionals and technical workers Managers and administrators Sales workers Clerical workers Services (nonhouse) Craft and repair Nontransportation operatives Transport operatives Laborers, handlers All others, including farm workers Industry: Agriculture Mining Construction Food, tobacco, textile, apparel, leather Chemicals, petroleum products, rubber, and plastics Paper, lumber, stone, glass, or clay products Primary and fabricated metals Electrical and nonelectrical machines Transportation equipment Other manufacturing Transportation Printing and publishing, communications, utility Wholesale trade Retail trade Finance, insurance, real estate Business repair Personal and entertainment Professionaligovemment administration Standard errors of regression Note: N = 26,844.
,035 ,027 - .oO04 ,167 ,112 ,058
- ,141 - ,155 - ,089
... -.130 - ,045 - .095 - ,092 -.I17
...
,059 - .069 - .337 - ,386 - ,194 - ,323 - ,221 - .333 - ,360 ...
,350 ,152 ,056 ,145 .023 ,185 ,151
.247 ,100 .184
.I96 ,121 - ,077 ,075 ,017 - ,059 ,087 ,684
(Standard Error) (.001) (.001)
(.oooo3) (.014) (.020) (.010) (.025) (.024) (.020)
...
(.015) (.023) (.023) (.021)
(.021)
... (.018)
(.023) (.023) (.019) (.017) (.019)
(.026) (.023) (.045)
...
(.077) (.046) (.048) (.050) (.050) (.048) (.047) (.049) (.051)
(.048) (349) (.048) (.045)
(.048) (.047) (.047) (.046)
197
Labor Market Adjustments to Increased Immigration
number of weeks worked for persons in the same interval in 1980. This procedure potentially affects only estimates where weekly wage, not annual earnings, was used as the dependent variable. The Public Use Samples allow a user to identify whether a person is an immigrant and when he arrived in the United States. In 1980, immigrants are classified into six cohorts based on when they arrived in the United States: 1975-80, 1970-74, 1965-69, 1960-64, 1950-59, and before 1950. In 1970, immigrants are classified into ten cohorts based on when they arrived in the United States: 1965-70, 1960-64, 1955-59, 1950-54, 1945-49, 1935-44, 1924-34, 1915-24, before 1915, and a category for those who do not report when they arrived in the United States. The Public Use Samples record in both Census years the highest year of schooling; age (which we used to construct potential experience as age minus schooling minus 6); marital status (which we use to construct dummy variables for those who are married and those who are separated, widowed, or divorced); whether there are children in the household; whether a person has a disability that limits his work (which is defined differently in 1970 than in 1980); race (which we use to construct our samples of young blacks and Hispanics); and, finally, place of origin, occupation, and industry. All these variables are used in the regressions reported in tables 6.6-6.10, unless the table indicates otherwise. In table 6A.2, we present estimates of the coefficients that are not reported in table 6.5, column 1 . The estimates of these coefficients in other tables are similar. These estimated coefficients are of some interest in themselves when comparing our findings to other research on the economic effects of immigration. Note that the returns to education and experience tend to be lower for immigrants than the returns that are estimated for natives, that the effects of marital status, children, and disability are consistent with other studies of the wages of natives, and that an individual’s place of origin has an effect on earnings even after controlling for all other observable variables. Immigrants from Europe and the Mideast have the highest earnings, while immigrants from Asia have the lowest earnings. An immigrant’s occupation and industry also had a significant effect on earnings.
Notes 1 . Several recent papers examine the changing skills of different immigrant cohorts (see Chiswick 1986; and Borjas 1985, 1987b). But the issues addressedin those papers are not new. Between 1900 and 1910, new immigrants accounted for 10 percent of the labor force, which sparked a similar policy debate at that time (see Douglas 1919). This debate culminated in the Immigration and Naturalization Act of 1923, which attempted to control the flow of immigrants through country-specific quotas. In 1965, amendments to the immigration laws changed the principal criteria used to control entry into the United States from the “national origin” quotas to a system based on
198
Robert J. LaLonde and Robert H. Topel
kinship with an American citizen or resident. The most recent legislation is the Immigration Reform Act of 1986, which attempts to regulate the flow of illegal aliens into the United States. 2. For a discussion of the potential distributional implications of immigration, see Johnson (1980) and Greenwood and McDowell(l986). 3. The estimate of 2.5 million assumes a labor force participation rate of .45 among all foreign-born persons in the 1980 Census who reported that they arrived in the United States between 1970 and 1979. 4. In this sense, our analysis is similar to earlier studies of the effect of the baby boom on wages, e.g., Welch (1979) and Bloom and Freeman (1987). These studies estimate the “own effect” of increased cohort size on earnings and employment of members of large cohorts. 5. Borjas (1987a) estimates a model of wage determination based on differences in demographic group labor force shares across geographic areas. He finds, as do we, that immigrant earnings are lowest in large immigrant enclaves. 6. Greater ease of substitution need not imply that immigrant wage levels converge to those of natives. The long-run stock of skills for the representative immigrant may remain below that of natives, so that immigrants earn less than natives, yet immigrant and native skills may substitute perfectly. In fact, in 1980 immigrants who arrived in the United States before 1950 earned more than natives. 7. Equilibrium among these markets is largely ignored in what follows, so we implicitly assume that mobility costs form a significant barrier to intermarket arbitrage in the short run. Topel (1986) contains an alternative approach that allows for migration among geographic markets. 8. Later in the paper, we examine the effect of a doubling of all immigrant cohorts on relative wages. Since native labor is being held constant for those calculations, the relative marginal products of immigrant labor can change. 9. Symmetry of cross-substitution effects is not implied; i.e., y,l # yl,. Symmetry of signs is a restriction of the theory; i.e., sign (yl,), as is negative definiteness of the full matrix of substitution effects. 10. Symmetry of effects implies dWcJltMc,= dWJMcJ,, so y k ,is proportional to Y , ~ , i = 1, . . . , k - 1. Thus, under the assumption that ykk = 0-an increase in the number of white natives at a locale has a negligible effect on wages-we may test yki = 0 if symmetry is assumed. Our tests indicate ytt = 0 in all cases. 11. This assumes that the matrix [ y i j ]is negative definite. 12. Cohort refers to years in the United States. 13. Estimates for the determination of hourly wages differ trivially from those reported here. 14. Estimates when natives are included in the data are available on request. 15. We remind the reader that our estimates measure relative wage adjustments, where the normalizing group is persons with more than thirty years in the United States. 16. This roughly corresponds to a tripling of the stock of immigrants. For example, a d In M = 1 change in all immigrant quantities would roughly correspond to an SMSA whose shares of the six immigrant arrival cohorts relative to native workers increased from .033, .022, .022, .022, ,011, ,011, and ,011 to .107, ,067, .067, ,034, ,034, and ,034, respectively. Multiply these estimates by .1 for the effect of a 10 percent change in the immigrant stock. 17. Though not reported separately, corresponding estimates for hourly wages show larger effects than for weekly wages. The implication is that adjustments in hours worked plays a minor and unsystematic role: all the effects reported here are due to price adjustments. We reach a similar conclusion with regard to weeks worked below. 18. We also computed F-statistics to test the four-parameter structure against the ten-parameter structure given by (7). These additional restrictions are also not rejected.
199
Labor Market Adjustments to Increased Immigration
19. For comparison with the unrestricted estimates in table 6.6, these parameters imply that a proportional increase in the size of all immigrant cohorts would reduce the earnings of the most recent immigrants by 6.5 percent. This effect dies out by 1 percentage point for each prior entry cohort; e.g., - .055 for those with six to ten years in the United States. 20. Effects on immigrants themselves are nearly identical to those reported in table 6.6. 21. Many SMSAs in our sample experienced a significant increase in the share of new immigrants in their work forces during the 1970s. Miami and Jersey City are exceptions. Although the flow of new immigration into these cities was substantial during the 1970s, this influx represented the continuation of a trend begun the decade before.
References Bloom, David, and Richard Freeman. 1987. The labour-market consequences of generational crowding. European Journal of Population 3: 131-76. Borjas, George. 1985. Assimilation, changes in cohort quality, and earnings of immigrants. Journal ofLabor Economics 3, no. 4:463-89. . 1987a. Immigrants, minorities, and labor market competition. Industrial and Labor Relations Review 40 (April):382-92. . 1987b. Self-selection and the earnings of immigrants. American Economic Review 77(September):53 1-53. Chiswick, Bany R. 1978. The effect of Americanization on the earnings of foreign born men. Journal of Political Economy 36(0ctober):897-922. . 1986. Is the new immigration less skilled than the old? Journal of Labor Economics 4(April): 168-92. Douglas, Paul. 1919. Is the new immigration more unskilled than the old? Journal of the American Statistical Association 15, no. 126(June):393-403. Greenwood, Michael J., and John M. McDowell. 1986. The factor market consequences of U. s. immigration. Journal of Economic Literature 24(December): 1738-76. Hamermesh, Daniel S. 1986. The demand for labor in the long run. In Handbook of Labor Economics, vol. 1, ed. 0. Ashenfelter and R. Layard, 429-71. Amsterdam: Elsevier. Hicks, John R. 1963. The Theory of Wages (1932). 2d ed. New York: St. Martin’s. Johnson, George. 1980. The labor market effects of immigrants. Industrial and Labor Relations Review 33(April):33 1-41. Office of Management and Budget. 1976. Standard metropolitan statistical areas. Washington, D.C.: U.S. Government Printing Office. Topel, Robert H. 1986. Local labor markets. Journal of Political Economy 94, no. 3, pt. 2(June):Slll-S143. U.S. Bureau of the Census. 1973. Technical documentation for the 1970 census of population and housing. Public Use Samples. Washington, D.C.: U.S. Government Printing Office. . 1983. Technical documentation for the 1980 census of population and housing. Public Use Samples. Washington, D.C.: U.S. Government Printing Office. Welch, Finis. 1979. The effect of cohort size on earnings: The baby boom babies’ financial bust. Journal of Political Economy 87, no. 5 , pt. 2(0ctober):S65497.
This Page Intentionally Left Blank
7
The Effects of Immigration on the Labor Market Outcomes of Less-skilled Natives Joseph G . Altonji and David Card
One of the most controversial aspects of immigration policy is the extent to which the arrival of immigrants helps or harms less-skilled natives. Although economists have developed a variety of theoretical models to analyze this question (see, e.g., Johnson 1980a, 1980b; Chiswick 1982; or Borjas 1987), relatively little empirical evidence is available. In this paper, we use variation in the fraction of immigrants across different cities to measure the effects of immigration on the labor market outcomes of less-skilled natives. We assemble information from the 1970 and 1980 Censuses on labor market outcomes of natives in 120 major cities. Information from consecutive Censuses allows us to correlate changes in immigrant fractions with changes in native outcomes within cities-thereby abstracting from differences across cities that might bias a simpler cross-sectional analysis. We also provide a variety of information on the industry distributions of natives and immigrants and analyze the changes in these distributions that have occurred in cities with higher and lower immigrant shares. In the first section of the paper, we present a simple theoretical model that describes the effects of immigration on the domestic labor market. We assume that the labor market within each city consists of skilled and unskilled workers and that immigration adds workers to both sectors, with relative additions David Card is professor of economics at Princeton University and a research associate of the National Bureau of Economic Research. Joseph G. Altonji is professor of economics at Northwestern University and a faculty research fellow of the National Bureau of Economic Research. The authors are grateful to Brain McCall and Sarah Turner for assistance with this research and to the National Bureau of Economic Research, the Center for Urban Affairs and Policy Research, Northwestern University, and the Industrial Relations Section, Princeton University, for research funding. They thank John Abowd, Francine Blau, George Borjas, Gregory DeFreitas, Richard Freeman, Peter Kuhn, and participants in seminars at Columbia University, the University of Minnesota, Princeton University, and the National Bureau of Economic Research for comments on earlier drafts.
201
202
Joseph G. Altonji and David Card
depending on the nature of immigrant inflows to the city in question. Our theoretical framework departs from earlier models in two ways. On the one hand, we disaggregate labor along skill lines rather than along the lines of national origin. On the other hand, we allow for demand-side effects associated with increases in the local population and for supply-side effects associated with the possible crowding out of native workers in response to lower wage rates. The model leads to a simple empirical specification in which wage and employment outcomes of less-skilled natives (either in cross section or within cities over time) vary with the share and skill composition of immigrants in the local labor market. In the second section of the paper, we address the question of whether immigrants and natives within the same city compete in the same labor market. Given the size of immigrant flows during the last two decades, our theoretical analysis implies that large adverse effects on less-skilled natives are unlikely unless increases in immigration lead to proportionately larger increases in the supply of labor to less-skilled jobs. We focus on industry-specific labor markets within cities. We develop a simple index that measures the effect of a given inflow of immigrants on the labor market of natives. We find that a 1 percentage point increase in the share of immigrants in a city generates approximately a 1 percent increase in the supply of labor to industries in which less-skilled natives are employed. The degree of competition between immigrants and less-skilled natives varies somewhat by race and sex group, being highest for black females and lowest for black males. Overall, however, the results suggest that immigrants are not sufficiently concentrated in the industries that employ less-skilled natives to have large effects on the less-skilled native groups. We go on to investigate whether immigrant inflows have displaced lessskilled natives from certain industries. Here, we compare the industry distributions of less-skilled natives in cities with relatively high and relatively low immigrant densities. We find some evidence that less-skilled natives in highimmigrant cities have moved out of immigrant-intensive industries. We also find that the nationwide trend of falling employment in these industries has been slower in high-immigrant cities, suggesting that the availability of immigrant labor has enabled certain low-wage industries to survive in highimmigrant cities.2 In the third section of the paper, we turn to a regression analysis of the relation between immigrant shares (or the change in immigrant shares) and employment outcomes of natives (or the change in these outcomes) across major cities. The results vary somewhat between the cross-sectional and firstdifference analyses. We argue, however, that the first-difference analysis is less likely to be contaminated by city-specific factors that affect immigrant densities and native outcomes. The analysis of changes shows no effect of increased immigration on participation or employment rates of less-skilled natives. It does reveal a systematically negative effect on native wages, al-
203
Immigration and the Labor Market Outcomes of Less-skilled Natives
though the specific estimates depend on the group and on whether we use an instrumental variables procedure to account for the fact that immigration inflows may depend on local labor market conditions. For the four racehex groups that we consider, the instrumental variables estimates (which we prefer) imply that an inflow of immigrants equal to 1 percent of the population of a standard metropolitan statistical area (SMSA) reduces average weekly earnings of less-skilled natives by about 1.2 p e r ~ e n tThe . ~ least squares estimates, by comparison, imply a more modest .3 percent reduction. 7.1 Analytical Framework
Our framework for analyzing the effect of immigration on the labor market outcomes of less-skilled natives is to view the inflow of immigrants to each city (or, more precisely, SMSA) as an outward shift in the supply of labor. Since we are specifically interested in the effects of immigration on lessskilled natives, we consider a two-sector labor market consisting of skilled and unskilled labor. Within skill categories, we make no distinction between native and immigrant labor or between earlier and later cohorts of immigrants. We assume that the demands for skilled and unskilled labor in each city are decreasing functions of their respective wage rates and that prices of capital and other inputs are exogenous to the local labor market. This framework contrasts with the one adopted by Borjas (1987), for example, who treats immigrants and natives as separate factors of production and assumes that locally produced output is sold at an exogenous price. In this case, the conventional elasticities of labor demand are undefined since an increase in the wage rate of one type of labor with other factor prices held constant leads to an increase in marginal cost that drives local firms out of busin e ~ s Given .~ that many of the goods produced within a city are nontraded services, however, and that many others enjoy some degree of imperfect substitutability due to transportation costs, we believe that it is more reasonable to posit the existence of downward-sloping labor demand functions at the local level. The observation that the demand for labor within a local economy arises in part from the demand for location-specific goods and services implies that a partial equilibrium model of the labor market is potentially misleading. In the extreme case, if all output is locally consumed, and if new immigrants arrive in the same skill proportions as the existing labor force, then an influx of immigrants leads to a new equilibrium at the original wage rates, with proportionately higher levels of employment, output, and cons~mption.~ More generally, the arrival of new immigrants shifts the demand for city output and hence the demand functions for skilled and unskilled labor. The size of this effect depends on the share of output consumed locally and on the relative skill composition of the existing and immigrating labor forces. To illustrate these propositions and establish a framework for our empirical
204
Joseph G. Altonji and David Card
analysis, consider an urban economy with two goods: a locally produced good (or service), Y, that is consumed locally and exported to other cities and an imported national good.6 Assume that Y is produced by a competitive industry with a constant-returns-to-scaletechnology using skilled labor, unskilled labor, and other inputs (capital apd/or raw materials) whose prices are exogenous and fixed.' Under these conditions, total industry cost (in units of the imported good) is described by a function of the form
C(W,,w,, Y ) = Yc(ws,W J , where w, and w, represent the real wages of unskilled and skilled labor (in units of the imported good), and c(.) is a unit cost function.* Let q represent the unit price of local output (denoted in units of the imported good). The assumptions of constant returns and perfect competition imply that q =
c(w,,W").
Demand for Y arises from three sources: local demand from skilled workers, Y,;local demand from unskilled workers, Y,;and export demand from the rest of the economy, Y,. Let D,(q, w,) and D,(q, wu) represent the per capita demand functions of skilled and unskilled workers, respectively, and let D,(q) represent the demand function for locally produced output from the rest of the economy. Let P, and P, represent the populations of skilled and unskilled workers in the city, and denote the total population by P = P, + P,. Product market equilibrium requires (1)
y = p , . D,(q, w,) + P" * D, (4, w,) + D,(q).
Let Ls(ws, q) and L,(w,,q) represent the per capita labor supply functions of skilled and unskilled workers, respectively. Equilibrium in the local labor market requires (2a)
P, - Ls(w,,4)
=
y * c, (ws,w,)
P" * L,(w,,4)
=
y . cz(w,,W"),
and (2b)
where c,(.) and c,(.) denote the partial derivatives of the unit cost function with respect to unskilled and skilled wage rates, respectively. Suppose that in an initial equilibrium the fraction of unskilled workers in the local population is a = PJP We wish to analyze the effect of an inflow of immigrants of size AZ. Let ci represent the share of unskilled workers in the new group. The effects of an immigrant inflow can be obtained by differentiating equations (l), (2a), and (2b) and making use of the fact that the proportional change in the price of output, Aq/q, equals the share-weighted sum of the proportional changes in all factor prices. For simplicity, assume that the cross-elasticities of the output demand and labor supply are zero.9 Then the proportional changes in skilled and unskilled wage rates satisfy the following pair of equations:
205
Immigration and the Labor Market Outcomes of Less-skilled Natives
A,
(3a) (3b)
+ q , A log w S , = q,,,A log wu + (qSs- E * ) Alog w,,
( d a ) AIIP = (quu- € , ) A log
A, [(l - or)/(l-a)] AUP
W,
where q,,is the elasticity of labor demand for skill group i with respect to the wage of group j, E, is the elasticity of labor supply of group i, and A, and A, are a pair of numbers between zero and one:
A, A,
- Y , - k, * Y,) I Y , (Y - k, * Y , - Y , ) / Y ,
= (Y =
k, = ~ ( l a)/[ ~~(l-a)], k2 = a(l-a)/[ ~ ( 1 - a ) ] .
The labor demand elasticities in equations (3a) and (3b) are determined by the conventional Marshall-Hicks formulas: %, = e,cw,, -
Y)?
where 8, is the share of the value of output paid as wages to skill group i, u,] is the partial elasticity of substitution of skill group i with respect to group j, and y is the elasticity of demand for Y with respect to its relative price q (a weighted average of the elasticities of demand exhibited by consumers in the local market and those elsewhere in the economy). The expressions A, (a/a)AI/Pand A, [(l I a)/(l - a)]AI/P in equations (3a) and (3b) give the effective percentage increases in unskilled and skilled labor resulting from an inflow of immigrants AI. The increases in skilled and unskilled populations are a A l and (1 - a ) A l , respectively. The proportional increases in the populations of unskilled and skilled workers are therefore (a/u)Al/P and [(l - a)/(l - u ) ] A l / P ,respectively. The factors Xu and A, adjust the gross increases in labor supply for the net increases in demand generated by the new immigrants. If local output is consumed entirely within the city and immigration is balanced in the sense that a = a, then Xu = A, = 0. Otherwise, the effective increases in labor supply depend on the fraction of local output sold outside the city and on the imbalance of skill ratios between the existing and the newly arriving population. In the simple case where newly arriving immigrants have the same skills as the existing population, A, = A$ = Y J X the fraction of output exported. If newly arriving immigrants are less skilled, however, A, > Y J Y > A,, accentuating the effective increase in unskilled labor supply. Using equations (3a) and (3b), changes in wages rates can be related to changes in the fraction of immigrants in the local population ( f ) by noting that Af = A(I/P) = (1 - f ) A l / P . In the special case that the demand for unskilled labor is independent of the wage rate of skilled labor (i.e., qu,= 0), equation (3a) can be simplified to
A log (4)
W,
=
(a/u)Al/P,
~
&u
- rl",
206
Joseph G. Aitonji and David Card
which specializes to the formula derived by Johnson (1980a) when Xu = 1 and (Y = a. l o Our model extends Johnson's earlier analysis in two directions: by allowing for skilled and unskilled workers in the existing and immigrating populations and by accounting in a very simple manner for the effect of added population on the demand for local output. If the demand for unskilled workers depends on the wage rate of skilled labor (i.e., qu, # 0), then the expression for the change in unskilled wage rates takes the more general form
(5)
A log w,,= B,,A l i p ,
where
Using the labor supply function, the change in the per capita labor supply of unskilled natives can then be written as (6)
A log Lu =
E,
. B,A lip.
To get some idea of the magnitude of the coefficient BY relating wage changes to immigrant inflows, suppose that (Y = a, so that Xu = A,. In this case, equation (5) can be rewritten as A log w, = AbUAlip, where the coefficient bU(b, < 0) is a function only of the supply and demand elasticities for skilled and unskilled labor, and A equals the fraction of local production exported to other cities. Values of the coefficient b, corresponding to alternative values of the supply and demand parameters of the model are displayed in table 7.1. The rows of the table present alternative choices for the ratio between the partial elasticity of unskilled labor with respect to nonlabor inputs (uUk) and the partial elasticity of skilled labor with respect to nonlabor inputs (a,,).The share-weighted average of these two elasticities is constrained to equal .6." The columns of the table present alternative choices for the partial elasticity of substitution between skilled and unskilled labor (us"). For each choice of the technological parameters, two values of bu are reported, corresponding to alternative choices for the elasticities of labor supply: . 1 and 1.O. Other parameters in the model are set as follows: the share of skilled labor (0,) = .4, the share of unskilled labor (0,) = .3, and the elasticity of demand for city output (y) = - 2.5. The first row of the table presents calculated values of buunder the assumption that capital is a substitute for unskilled labor and a complement for skilled labor.I2 As Hamermesh (1986, 460-62) has noted in his review of the literature on labor demand, many empirical studies based on the distinction be-
Immigration and the Labor Market Outcomes of Less-skilled Natives
207 Table 7.1
Predicted Effect of an Increase in Immigration on Unskilled Wage Rates
Ratio of Partial Elasticities of Substitution with Capital (u&J 1. -.25
2. 0 3.
.5
4. 1.0
Partial Elasticity of Substitution of Skilled for Unskilled Labor (u,J Labor Supply Elasticity ( E ) ~
.25
1 .o
3.0
.1
...
- .31
- .42
...
- .21 - .39 - .30 - .46 - .33 - .49 - .34
- .30 - .45 -.31 - .48 - .33 - .49 - .34
1.o .1
- .21
I .o
- .29
.1
- .42
I .o .1 1 .o
- .32 - .49 - .34
Note: For notation and assumptions, see the text. a Share-weighted average of substitution elasticities of skilled and unskilled labor with capital is constrained to equal .6. Labor supply elasticities of skilled and unskilled workers are constrained to be equal.
tween blue-collar and white-collar workers in manufacturing have confirmed this hypothesis. In contrast, the last row of the table presents values of b, under the assumption that skilled and unskilled labor are equally substitutable with capital.I3 Despite the wide variation in demand and supply parameters represented in the table, the range of the coefficient b, is relatively modest: from - .49 to - .27.14Under the assumption that immigrants add nothing to the demand for locally produced output (i.e., A = l), these coefficients imply that a 1 percent increase in the population of a city due to an influx of immigrants with the same skill composition as the existing labor force reduces unskilled wages by .3-.5 percent. The implied reduction in the per capita labor supply of natives (and existing immigrants) is proportional to this reduction in wages, multiplied by the elasticity of labor supply. If the elasticity of labor supply is in the range of zero to one, the implied reduction in per capita labor supply of natives is 0-.5 percent. The magnitude of these predicted effects is dampened by any expansionary effect that immigrants have on the demand for locally produced goods. For example, if one-third of output is consumed locally, then the implied wage effects of a given immigrant inflow are reduced by approximately one-third.I5 Any imbalance in the skill distribution of arriving immigrants, on the other hand, accentuates their effect on the local labor market. In the most extreme case, if newly arriving immigrants are all unskilled and the proportion of skilled workers in the existing labor force is . 5 , then the predicted value of b, ranges from - 2.0 to - 1.O, implying roughly two to three times larger effects on unskilled wage rates. Our empirical strategy in section 7.3 below is to correlate variation in the
208
Joseph G. Altonji and David Card
share of immigrants in the local labor market with variation in the employment and wage outcomes of less-skilled natives. We interpret the coefficient relating wages to immigrant shares as an estimate of the expression B, in equation ( 5 ) and the coefficient relating employment rates (or participation rates) to immigrant shares as an estimate of the product of B, and the elasticity of labor supply of unskilled native workers. As the previous discussion makes clear, the value of B, depends on the nature of immigrant flows to each city and on the characteristics of the demand for output produced in each city. Even ignoring these issues (as we do), it is important to keep in mind the potential endogeneity of immigrant inflows to different cities. If the supply of immigrants is wage elastic, then the covariation across cities between the labor market outcomes of natives and the share of immigrants in the labor market will be a positively biased estimate of the expression BU.In our analysis, we address this issue with an instrumental variables scheme that isolates the component of immigrant inflows associated with the predetermined characteristics of each city. Before turning to the empirical work, two limitations of the model deserve discussion. First, the model assumes that the existing native population is immobile. However, one might loosely interpret the supply elasticity of natives to reflect both labor supply changes of the current population of the city and out-migration (or in-migration) of natives to (or from) other cities.16 If one interprets the intercity mobility of natives as raising the long-run elasticity of labor supply, then one would conclude that migration by natives in response to immigrant inflows would lower the effect of immigration on wages. It would also lower the effect on per capita labor supply of natives, as measured by a variable such as the employmenVpopulation rati0.I’ However, intercity migration would imply spillover effects on wages and employment/population ratios in other cities, which we ignore in our empirical work. Second, the model assumes that the local labor market clears. Within the model, unemployment can be viewed as depending on the wage rate relative to the benefits of being unemployed. This view is most sensible in the long run. Barriers to wage adjustment (such as binding minimum wage levels or fixed welfare benefits) might be expected to strengthen the effect of an increase in immigrants on the employment and unemployment outcomes of natives while weakening the effects on wage levels relative to those implied by equations (6) and (7). The employment effects for natives could be especially large if employers of immigrants are less likely to comply with minimum wage laws or to be unionized.’”
7.2 Industry Distributions of Natives and Immigrants Our empirical analysis is based on the labor market outcomes of less-skilled natives in 120 major SMSAs in the 1970 and 1980 Censuses. We consider four groups of “less-skilled” natives: white males with less than twelve years
209
Immigration and the Labor Market Outcomes of Less-skilled Natives
of completed education; white females with less than thirteen years of completed education; black males with less than thirteen years of completed education; and black females with less than thirteen years of completed education. Our data base consists of samples of each race/sex group drawn from the 1/100 Public Use Sample of the 1970 Census and the 5/100 “A” sample of the 1980 Census. A description of our sampling procedures and information on our procedures for matching SMSA definitions between the 1970 and the 1980 Censuses are provided in Appendices A and B. Table 7.2 provides an overview of our samples of less-skilled natives. The samples are restricted to individuals between the ages of nineteen and sixtyfour who report themselves as not in school during the Census week.I9 Because of the age and education requirements, the average age of our lessskilled native groups is close to 40. The average years of complete schooling is less than eight for white male high school dropouts and between ten and eleven for the other groups. The labor market outcomes that we consider are the labor force participation rate during the Census week; the employment rate during the Census week (measured for those in the labor force in the Census week); the Table 7.2
Descriptive Statistics for Native Samples
Demographic and Economic Characteristics: 1. Age 2. Education 3. Labor force participation rate ( X 100) 4. Employment rate (X
100)
White Male Dropouts
White Females No College
Black Males No College
Black Females No College
1970
1980
1970
1980
1970
1980
1970
1980
44.3 8.5
43.5 8.8
40.9 10.6
40.8 11.0
39.1 9.2
37.4 10.2
38.7 9.6
38.3 10.4
88.8
81.0
47.3
56.5
83.6
78.4
55.1
59.1
96.0
91.1
95.6
94.0
94.4
86.9
92.6
87.9
85.2
73.7
45.2
53.3
78.9
68.3
51.1
52.1
91.6
82.9
54.5
61.1
86.7
78.0
60.8
60.1
5. Employment pop-
ulation rate Census week(X 100) 6. Employment population rate last y e a r ( x 100) 7. Logarithm of weeks worked last year 8. Logarithm of weekly earnings last year (current $) 9. Sample size
3.81
3.75
3.57
3.60
3.77
3.69
3.58
3.60
4.95
5.52
4.26
4.96
4.61
5.29
4.03
4.90
84,068
24,925
99,488
81,151
27,779
29,723
34,013
34,540
Nore: Samples consist of individuals age 16-64 in 120 major SMSAs. Individuals enrolled in school in Census week are excluded. White male dropouts sample includes individuals with less than 12 years of completed education. Samples for other groups include individuals with less than 13 years of completed education. For further information, see App. A.
210
Joseph G. Altonji and David Card
employment-population ratio in the Census week; the fraction of people who reported working at any time in the previous year (for simplicity, we refer to this as the employment-population ratio last year); and the logarithms of weeks worked and average weekly earnings during the previous year (measured for those individuals who report positive weeks of work and positive earnings in the previous year). Precise definitions of these outcomes are presented in Appendix A. The model of the previous section treats the market for less-skilled workers within each city as homogeneous. Even within a particular city, however, the market for less-skilled workers may be segmented along industry lines. If immigrants and natives tend to work in different industries, then the firstround effects of new immigration will be mainly concentrated among existing immigrants. If immigrants tend to work in the same industries as a particular subgroup of natives, however, then the effects of immigration on this subgroup of less-skilled natives will be magnified. Some simple evidence on the correspondence between industry distributions of natives and immigrants is presented in table 7.3. For the ten two-digit industries with the highest immigrant employment shares and the ten industries with the lowest immigrant shares, this table shows the fraction of each of the four less-skilled native groups in the industry in 1980.20High-immigrantshare industries include several low-wage manufacturing industries (apparel, leather, furniture, miscellaneous manufacturing, and textiles) as well as low-wage service industries (private household services, hotels and motels, restaurants and bars, and transportation services) and agriculture. Lowimmigrant-share industries include the government sector as well as railroads, communications, and several regionally based industries (tobacco, pipelines, coal mining, and oil and gas extraction). A comparison of the second and third columns of the table shows that industries with high or low immigrant shares in 1980 exhibited the same characteristic in 1970, although the immigrant fractions in many industries increased sharply between 1970 and 1980.21The immigrant share of total employment in all industries in our sample of 120 cities increased from 6.0 percent in 1970 to 9.6 percent in 1980.22 The data in table 7.3 suggest that immigrants are most directly competitive with native women-particularly black women. In fact, the proportion of black females in the ten highest-immigrant-share industries in 1980 was almost as high as the fraction of immigrants in those industries. By comparison, black males are the least concentrated in high-immigrant-share industries and the most heavily concentrated in low-immigrant-share industries. One way to evaluate the effect of immigration on a particular native group is to calculate the overlap in the industry distribution of the group with the industry distribution of immigrants. Assuming that interindustry mobility costs are large, the effects of immigration on native wages will be directly proportional to the average increase in labor supply to industries in which natives are employed. To formalize this measure, let S,, represent the share of the native group in the ith industry, let E , represent the initial level of total
211
Immigration and the Labor Market Outcomes of Less-skilled Natives
Table 7.3
Distributions of Natives in High- and Low-Immigrant-ShareIndustries, 1980 ~-
% of Natives in Industry
Industry High immigrant share: 1. Apparel 2. Leather 3. Agriculture, crops 4. Furniture 5 . Miscellaneous manufacturing 6. Private household services 7. Hotels and motels 8. Transportation services 9. Restaurants and bars 10. Textile mills Total: 10 industries
Low immigrant share: 1. Pipelines 2. Gov’t.: justice and public safety 3. Gov’t.: revenue and taxation 4. Coal mining 5 . Railroads 6. Tobacco 7. U.S.Post Office 8. Oil and gas extraction 9. Communications 10. Gov’t.: economic programs Total: 10 industries
% % % of All Immigrant Immigrant Immigrants 1980 1970 InIndustry
White White Black Black All Males Females Males Females
38.4 21.3 25.8 21 .o
21.1 14.4 10.0 11.0
5.1
.6 1.5
1 .O
1.3 .2 .6 .4
.6 .3 1.2 .7
2.0 .3 .4 .4
.5 .1 .5 .6
2.3 .3 .3 .4
20.9
10.6
2.3
1.1
1.2
1.3
1.0
1.4
20.2 18.2
9.5 10.6
1.4 2.2
.7 1.2
.2 .7
.8 1.7
.2 1.2
6.0 3.5
15.8
11.2
.5
.3
.I
.4
.I
.1
15.6 15.6
9.3 8.8
6.4 .8
3.9
2.5 .I
7.6
3.1
.6
5.5 .8
...
...
21.8
10.1 8.2
15.6
7.9
20.6
1.5
1.9
.o
.o
.o
.o
.o
.o
2.8
2.3
.4
1.4
.9
.8
2.0
1.o
2.8 3.5 3.8 3.9 4.1
3.4 2.4 3.5 1.8 2.4
.1
.5
.2 .I
.5
.I .I .4
1.1
.4
.4 .o .1 .I .6 1.4 .1 .I 1.0 .9
.2 2.6
.2 .2 1.3
4.2 4.4
2.0 3. I
.2 .8
.4 1.7
.4 .5
.2 2.1
.2 1.1
1.9
4.5
2.9
.3
.6
.2
.5
.9
.8
...
...
2.5
6.1
4.5
4.7
8.4
6.0
.O .3
.o
.5
.I
.o
.o
.1
Note: Based on the industry distributions of 19- to 64-year-olds in 120 major SMSAs in the 1980 Census, a All natives include all education groups. Other groups are defined in the note to table 7.2.
employment in industry i , and let AE, represent the increase in labor supply to the ith industry associated with the amval of a fixed number of new immigrants AE. The average proportional increase in labor supply experienced by the native group is
212
Joseph G. Altonji and David Card
Suppose that new immigrants sort themselves into industries in the same proportions as existing immigrants. Then A E , = S,,AE, where s,, is the share of existing immigrants employed in industry i. Finally, E , = S,E , where S, is the share of all workers in industry i, and E is level of total employment in the labor market. Thus, the average proportional increase in labor supply experienced by the native group is PAEIE, where
This expression reduces to one in the case of a homogeneous labor market, in which S,, = S,, = S , . In a heterogeneous labor market, however, the average proportional increase in labor supply experienced by a particular native group may be more or less than AEIE, depending on the degree of similarity between the industry distributions of immigrants and the native group. Estimates of this index of labor market competition are presented in table 7.4 for the four groups of less-skilled natives. We have calculated the index separately using the 1970 and 1980 industry distributions of natives and immigrants. We have also calculated the index separately over two subsets of cities: the twenty cities with the highest fraction of less-skilled immigrants in 1980 and the forty cities with the lowest fraction of less-skilled immigrants in 1980. These cities are identified in Appendix D. Estimates of the index of labor market competition are very similar using the 1970 and 1980 industry distributions. The values of the index range from a low of .85 in 1980 for white males in low-immigrant cities to 1.28 in 1970 for black females and are consistently below one for black males. The results confirm the impression that black females are in most direct competition with immigrants, whereas black males are most isolated from immigrant competition. Nevertheless, the values of the index are not far from one for any of the groups, suggesting that increases in the Share of immigrants in the labor market have roughly proportional effects on the labor markets of unskilled nat i v e ~ . *The ~ differences in the index between high- and low-immigrant cities are positive for males and negative for females, suggesting that immigrants and native males are in more direct contact in high-immigrant cities while immigrants and native females are in less direct contact. One interpretation of this finding is that less-skilled native females have been displaced from immigrant-intensive industries in high-immigrant cities. We explore this hypothesis next. Evidence on the extent of industry displacement is presented in tables 7.5 and 7.6, which give the cross-sectional and time-series patterns of differences in the industry distributions of less-skilled natives in high-immigrant and lowimmigrant cities. For ten high-immigrant-share industries and ten major immigrant-employing industries, table 7.5 displays the relative share of unskilled natives in high- versus low-immigrant cities. Specifically, let EE, and Ef;, represent the employment of native group N in industry i in highimmigrant and low-immigrant cities, respectively. Let EY and Ef represent
213
Immigration and the Labor Market Outcomes of Less-skilled Natives
Table 7.4
Estimated Index of Labor Market Competition between Immigrants and Natives All Cities
Native Group
Low-Immigrant Cities
1970
1980
1970
1980
1970
1980
1.06
1 .oo
1.09
1.03
.99
.85
1.09
1.08
1.05
1.03
1.10
1.12
.94
.94
.91
.93
.91
.91
1.24
1.15
1.28
1.06
1.20
1.16
1. White male
dropouts 2. White female no college 3. Black males no college 4. Black females no college
High-Immigrant Cities
Nore: For definition of index, see the text. High-immigrant cities include 20 SMSAs with highest fraction of less-skilled immigrants. Low-immigrant cities include 40 SMSAs with lowest fraction of less-skilled immigrants.
total employment in industry i in these cities, and let E; and Ek represent total employment of the native group in these cities. For each industry and native group, table 7.5 displays the ratio E ,H/EH _; , l E B_ E_ EktlEf. ’ E k / E L ’
~
which represents the relative employment share of natives in the ith industry in high- versus low-immigrant cities, divided by the relative shares of natives in total employment in those cities. A value of unity indicates that natives have equal shares of employment in the industry in the two groups of cities, controlling for their relative shares in total employment. A value of less than unity, on the other hand, indicates relative displacement in the highimmigrant-fractioncities. For most of the high-immigrant-share industries, there is evidence of displacement of natives in the high-immigrant-share cities. The displacement effects are less apparent for white males, with ratios in excess of unity for four industries.24 For the other three groups, however, relative employment shares in the set of high-immigrant cities are generally less than unity. By comparison, the evidence of displacement of less-skilled natives from the major immigrant-employing industries in the lower panel of table 7.5 is mixed. On balance, these data suggest that the industry displacement of natives is restricted to low-wage service and manufacturing industries and agriculture. As the ratios in the right-hand column of table 7.5 suggest, these industries are generally more important in high-immigrant than low-immigrant cities, although in cross section it is difficult to distinguish alternative explanations for this effect.25 Table 7.6 repeats the analysis in table 7.5, taking the ratio of the relative
214 Table 7.5
Joseph G. Altonji and David Card Relative Industry Distributions of Natives in High- and Low-Immigrant Cities, 1980 Relative Share of Native Group: High- vs. Low-Immigrant Cities' % of All
Industry High immigrant share: I . Apparel 2. Leather 3. Agriculture, crops 4. Furniture 5. Miscellaneous manufacturing 6. Private household services 7. Hotels and motels 8. Transportation services 9. Restaurants and bars 10. Textile mills Other major immigrant employers: 1. Hospitals and health services 2. Construction 3. Education 4. Business services 5. Electrical equipment 6. Machinery 7. Transportation equipment 8. Grocery stores 9. Wholesale trade: nondurables 10. Food products
Immigrants in Industry
White Males
5.1 .6
1.43 1.33
1.5 1 .o
2.3
White Females
Highvs. LowImmigrant Citiesb
Black Males
Black Females
.49 .71
1.29 .62
.44 .97
2.64 1.40
.56 .64
.86 .68
.84 .68
.74 .36
1.71 .94
.83
1.04
.65
.66
1.89
1.4
...
.65
.35
.79
1.25
2.2
1.42
.91
.67
.54
1.25
.5
.59
1.12
.09
1.33
2.29
6.4 .8
1.32 .73
.80 .77
.95 1.22
.50 .65
1.01
8.4 5.7 4.5 3.3
1.71 .97 .94 1.51
.89 1.04 1.15 .81
1.48 .83 1.07 1.18
I .07 .81 I .oo .99
3.3 3.2
.75 .91
1.13 1.62
.61 .84
.82 1.32
1.17 .68
2.7 2.6
.78 1.61
1.52 .89
.74 1.89
.72 .98
.74 1.03
2.5 2.1
1.27
.94 1.35
.96 .65
I .33 .70
1.17 .79
.81
.51
.91 1
.oo
.89 1.51
Note: Based on the industry distributions of 19- to 64-year-olds in 120 SMSAs in the 1980 Census. High-immigrant cities include 20 SMSAs with the highest fraction of less-skilled immigrants. Lowimmigrant cities include 40 SMSAs with the lowest fraction of less-skilled immigrants. a For each industry and native group, the relative share is the proportion of industry employment contributed by the native group in high-immigrant cities, divided by the same proportion in low-immigrant cities. This ratio is then divided by the ratio of the shares of the native group in total employment in the two groups of cities. Ratio of industry share of total employment in high-immigrant cities to industry share of total employment in low-immigrant cities.
215 Table 7.6
Immigration and the Labor Market Outcomes of Less-skilled Natives Relative Growth of Employment Shares of Natives in High- and LowImmigrant Cities, 1970-80 Relative Growth of Native Group: High- vs. Low-Immigrant Citiesa
Industry High immigrant share: 1. Apparel 2. Leather 3. Agriculture, crops 4. Furniture 5. Miscellaneous manufacturing 6. Private household services 7. Hotels and motels 8. Transportation services 9. Restaurants and bars 10. Textile mills Other major immigrant employers: 1. Hospitals and health services 2. Construction 3. Education 4. Business services 5. Electrical equipment 6. Machinery 7. Transportation equipment 8. Grocery stores 9. Wholesale trade: nondurablesd 10. Food products
White Females
1.73 1.33
.85 1.72
.82 .I9
.39 .43
I .30 3.10
.67 .62
.43 .77
.72 .88
1.29 1.26
1.45 1.59
1.88 1.06
.95 .85
.67
.91
.75
.33
1.11
.96
. .
.72
.38
.83
1.55
.52
1.47
1.15
.71
.72
.93
.I6
.61
2.23
.04
2.16
.68
.39
1.36 .94
.97 .88
.89 2.01
.98 .95
.94 .82
1.05 .56
1.75 .89
1.oo
1.04 .77 1.15
.77 1.27
1.08 .72 I .52
.91 1.13 .82
1.17 1.03 .89
1.28
.89
.56
.67
.97
I .32
.60 .79
1.08 1.05
.66 .65
.64 .52
1.38 1.40
.75 .93
.83 1.33
1.54 I .04
1.07 1.15
.87 1.14
1.11 I .07
.78 .92
..
..
..
1.09
.89
.72
.93
.78
.75
1
.oo
Black Females
Growth of Total Employment All Citiesc
White Males
..
Black Males
Relative Growth of Total Employment: High- vs. LowImmigrant Citiesb
Note: For definitions of high-immigrant and low-immigrant cities, see the note to table 7.5. a For formula, see the text. Relative ratio of 1980 to 1970 employment totals for industry in high-immigrant vs. low-immigrant cities. Ratio of 1980 to 1970 employment totals for industry in all cities. Data for wholesale trade nondurables industry not available.
216
Joseph G. Altonji and David Card
employment share of natives in 1980 to the relative employment share in 1970. A value of unity for this ratio suggests that natives have maintained their relative share of industry employment, controlling for the relative growth of total employment of natives in the two sets of cities. A value of less than unity, on the other hand, suggests that natives have lost relative share in the industry in high-immigrant versus low-immigrant cities.26 The results in table 7.6 are generally consistent with those in table 7.5 and suggest some movement of less-skilled natives out of high-immigrant-share industries in the high-immigrant cities between 1970 and 1980. The fifth column of the table indicates the relative growth of total employment by industry in high- versus low-immigrant-share industries, while the sixth column gives the ratio of total employment in the industry in 1980 in all cities to total employment in all cities in 1980. Although several high-immigrant industries were declining relatively quickly between 1970 and 1980, in most cases the relative decline was slower in high-immigrant cities. This suggests that the availability of immigrant labor may allow certain industries to survive in highimmigrant cities even at the same time as natives continue to exit from these industries. Our analysis of the industry distributions of immigrants and 1ess:skilled natives suggests three conclusions. First, a 1 percentage point increase in the share of immigrants generates approximately a 1 percent increase in the supply of labor to industries in which less-skilled natives are employed. There is no indication that immigrants and less-skilled natives are concentrated in particular industries in a manner that would greatly accentuate the labor market competition between them or, on the other hand, substantially reduce the degree of labor market competition between them. Second, among the four native groups that we consider, immigrants are most directly competitive with black females and least competitive with black men. Third, differences in industry distributions between high- and low-immigrant cities suggest that natives have been displaced from some low-wage service and manufacturing industries and that these industries have declined less quickly in cities with more immigrants.
7.3 An Analysis of the Effects of Immigration on Less-skilled Natives In this section, we examine the correlation across cities between the labor market outcomes of less-skilled natives and the fraction of immigrants in the city. We present cross-sectional analyses for 1970 and 1980 as well as a firstdifferenced analysis of changes between 1970 and 1980. Our basic approach is very simple. We regress SMSA averages of the labor market outcome variables for our four racehex groups against measures of the immigrant fraction in the SMSA and a variety of controls for the characteristics of each city. Before turning to the results of the analysis, however, we first discuss the construction of SMSA means for the outcome variables. We then briefly dis-
217
Immigration and the Labor Market Outcomes of Less-skilled Natives
cuss potential econometric problems with the cross-sectional and firstdifferenced analyses and offer some comments on the interpretation of our estimates. 7.3.1
Construction of SMSA-Level Outcome Measures and Control Variables
The first step in our analysis is to construct SMSA-specific means of the outcome variables that are purged of differences in the observable characteristics of the native population across different cities. Given the limited information collected in the Census, this step amounts to regression adjusting the outcome variables for differences in age and education. Such an adjustment has two potential advantages. First, it should reduce the sampling variation associated with the means of the outcome variables across different cities. Second, it should eliminate any bias arising from correlations between the fraction of immigrants in a city and the age and educational attainment of natives. For each racehex group in each of the two Censuses, we regress each of the outcome variables against a full set of SMSA dummies and a flexible function of age and education. Specifically, we include a cubic polynomial in age, a detailed set of dummy variables for different education levels, and a full set of interactions of age and education up to the second order. We then use the estimated SMSA dummies as our regression-adjusted outcome measures.27 The explanatory variables in the second step of our analysis include the fraction of immigrants in each SMSA and three additional control variables: the logarithm of SMSA population and SMSA-specific means of age and education for the particular racehex group under consideration. Although the outcome variables are adjusted for age and education, we found in preliminary work that the mean of adjusted weekly earnings is correlated across cities with the mean of education, particularly for blacks. We have no explanation for this phenomenon, although it may indicate a correlation across cities between the quality and the quantity of education among blacks or possibly a market externality associated with higher levels of education among the less-skilled black population. In any case, we include SMSA-specific means of age and education for the particular racehex group in all our SMSA-level regressions. These means are calculated directly from our native extracts. Our measure of the fraction of immigrants in each SMSA is the fraction of foreign-born residents, taken from published tabulations of the 1970 and 1980 Censuses. From the standpoint of the theoretical model, it would be preferable to use the fraction of immigrants in the local labor force. Since our sample sizes for 1970 are too small to provide reliable estimates of the fraction of immigrants in many of the smaller cities, we have relied instead on the published population data. Provided that changes in the immigrant labor force are proportional to changes in the population of immigrants, the use of fraction of immigrants in the population will not affect our results.
218
Joseph G . Altonji and David Card
7.3.2 Econometric Issues We next turn to a brief discussion of our estimating equations. We focus on three issues: possible sources of bias in the estimating equations; the interpretation of differences between cross-sectional and first-differenced estimates of the effects of immigration; and the use of weighted least squares in the estimation. Our cross-sectional estimating equations have the form
(7) where 9, is the adjusted labor market outcome for native group N in city j , X, is a vector of control variables for the racehex group and city (the mean of age and education for the group and the logarithm of SMSA population), f , is the fraction of immigrants in the city, and e,, is a residual term. Similarly, our first-differenced estimating equations have the form
AFNJ = AXNjb + Af, c + Ae,,,,, where AZj refers to the change in the variable Z in cityj between 1970 and 1980. Depending on the choice of outcome measure I: these equations have the form of equations ( 5 ) or (6) derived from our theoretical model. The interpretation of estimates of the coefficient c obtained from equation (7) or (8), however, depends on the nature of the residual terms in these equations. These residuals can be decomposed into two conceptually distinct components: (1) a market-level SMSA effect due to factors other than immigration (e.g., unmeasured characteristics of natives or demand shocks affecting the local economy) and (2) sampling variation arising from the fact that we observe only a sample of natives in each SMSA. Let Y, represent the true population value of the outcome variable for natives in city j . Then we may decompose eNjas
where a,,,,represents the SMSA effect due to factors other than immigration, and 9, - Y, is the component of e, attributable to sampling variability. Only if a,, is orthogonal to the fraction of immigrants in the city will estimates of the coefficient c from the cross-sectional regression (7) yield unbiased estimates of BU or E * B,, as described by equation ( 5 ) or (6). In the first-differenced specification, the corresponding requirement is that changes in the unmeasured SMSA effects be uncorrelated with changes in the fraction of immigrants in the city between 1970 and 1980. Clearly, the main advantage of the first-differenced analysis is that it eliminates any bias introduced by city-specific fixed effects that are correlated with the fraction of immigrants in a city and the labor market outcomes of natives. Transitory effects (associated with transitory fluctuations in the demand for the output of specific cities, e.g.) will still lead to biases in the differenced
219
Immigration and the Labor Market Outcomes of Less-skilled Natives
analysis if they influence the inflow rate of immigrants. Bartel’s (1989) recent analysis suggests that economic conditions have a relatively small effect on the destination city chosen by immigrants. Instead, Bartel’s findings suggest that immigrants are mainly attracted to cities with large concentrations of previous immigrants from the same country (see also Greenwood and McDowell 1986). Nevertheless, her research leaves open the possibility that the timing and size of immigrant inflows are affected by economic conditions in particular cities. We attempt to control for any potential correlation between immigrant inflows and local economic conditions in our first-differenced analysis by an instrumental variables procedure. As suggested by Bartel’s (1989) work, we use the fraction of immigrants in a city in 1970 to predict the change in the fraction of immigrants over the following decade.28 Immigrant inflows are strongly correlated with the initial fraction of immigrants in a city, and these variables are reasonably strong predictors of the change in immigrant fraction. In comparing the cross-sectional and first-difference results, one should also keep in mind that the first-difference analysis is more likely to capture the short-run effects of immigration, in which the capital stock and the industry/ skill composition of labor demand have not had time to adjust fully. The effects of immigration on per capita employment rates and wages may weaken over time as natives move to other cities or to labor market sectors that are less affected by immigrant competition. Dynamic issues are not addressed in our formal model, but we suspect that the short-run effects of immigration on employment of less-skilled natives will be larger than the long-run effects. The relative magnitude of the short-run and long-run effects on wages depend on whether there are barriers to wage adjustments in the short run. In fact, we find that the cross-sectional estimates of the effect of immigration on employment outcomes of natives are larger than the differenced estimates, whereas the opposite is true of the estimated effects on wages. This leads us to suspect that the differences between the cross-sectional and the differenced results are primarily due to correlations between city-specific effects and immigrant shares that are eliminated in first-differences rather than to a distinction between long-run and short-run effects. A final econometric issue arises from the relatively small samples of black natives in many cities, particularly in our 1970 sample. We restrict our crosssectional and differenced analysis of each racekex group to the set of cities for which we have at least thirty group members in both 1970 and 1980. Consequently, we work with a set of ninety-one cities for black males, a set of ninety-four cities for black females, and a full set of 120 cities for white men and women. We also use weighted least squares methods to estimate our equations, using the square root of the number of observations for the race/sex group in the city as a weight. In our first-differenced specifications, we use as a weight (N,’ N G * ) - ~ where ’ ~ , N,, and N,, are the number of observations for the native subgroup in the SMSA in 1970 and 1980, respe~tively.~~ This
+
220
Joseph G. Altonji and David Card
weighting scheme assumes that the residual eN arises mainly from sampling variability associated with the estimated outcome measure. Even controlling for the covariates in our models, however, the labor market outcomes of different racehex groups are correlated across cities, suggesting the presence of omitted city-specific effects. We have not adjusted our standard errors or estimation procedures to take account of such error components. 7.3.3 Empirical Results
To provide an introduction and overview of our results, table 7.7 presents weighted least squares estimates of the effects of immigration on the labor market outcomes of the pooled set of four race/sex groups. The estimated equations include unrestricted intercepts for the four groups as well as groupspecific coefficients on the means of age and education. The coefficients on the immigrant share variable and the population variable, however, are restricted to be the same across the four native subgroups. The cross-sectional results for 1970 show significantly negative effects of an increase in immigrant shares on the labor force participation rates and employment rates of less-skilled natives. The results imply that a 10 percentage point increase in the fraction of immigrants in an SMSA would lead to a reduction in the employment/populationratio of less-skilled natives of roughly 2 percent. The employment rate would also fall by 1 percent, implying an increase in unemployment rates of about 1 percent. Among those who work, average weeks per year would fall by about 2 percent.
Table 7.7
Effects of Immigration on Four Groups of Less-Skilled Natives, Pooled Sample (standard errors in parentheses) Cross-sectional
Outcome Variable 1. Labor force/ population 2. Employment/ population 3. Employment/labor force 4. Fraction worked last Ye= 5 . Log weeks worked 6. Log earningslweek
First-Differenced
1970
1980
- ,173
- .083
,080
(.066) - ,240 ( ,074) - ,109 (.036) -.161 (.063) -.191 (.078) .467 (.165)
(.049) - .054
(.083) ,404 (.097) .461 (.077) .090 (.084) .232 (.132) - ,262 (.228)
(.ow ,019
- .I58 (.050) - ,088 (.061) ,018 (.112)
1980-70
1980-70 IV’
- ,102 (. 122)
,085
(. 144)
,231 (.113) - ,246 (.125) ,142 (.193) - 1.205 (.342)
Note: All equations included subgroup-specific intercepts, the total population in the SMSA, and the average education and age of the subgroup in the SMSA (with subgroup-specific coefficients). The sample size is 424. a Estimated by instrumental variables. The change in the fraction of immigrants in the SMSA is instrumented with the fraction of immigrants in 1970 and its square.
221
Immigration and the Labor Market Outcomes of Less-skilled Natives
These negative employment effects contrast sharply with the finding that immigration has a positive effect on weekly wages. The estimated coefficient in row 6 implies that a 10 percentage point increase in the immigrant share would lead to a 4.7 percent increase in weekly earnings. Within the context of our model, these results can be reconciled only if the labor supply elasticity of less-skilled natives is negative.3o The 1980 cross-sectional results for the various employment outcomes also indicate a negative effect of immigration, although the estimated coefficients are smaller in magnitude than those for 1970. In the 1980 data, however, the estimated effect of immigrant densities on the average weekly earnings of natives is essentially zero. This gives further reason for caution in the interpretation of the 1970 results. Weighted least squares estimates of the first-differenced specification are presented in the third column of table 7.7. In contrast to the cross-sectional results, these estimates suggest a modest positive effect of the fraction of immigrants on the employment outcomes of natives. The estimated effect on earnings per week is negative ( - .267) but not statistically different from zero. Instrumental-variables estimates of the first-differenced specification are presented in column 4. These estimates give an ambiguous picture of the effect of immigration on the employment outcomes of natives. A marginally significant positive effect on the employment rate in the Census week is counterbalanced by a marginally significant negative effect on the employmentpopulation ratio last year. Nevertheless, the instrumented first-differencedresults indicate a significantly negative effect of immigration on wages. The coefficient is - 1.2 with a standard error of .242. The more negative effect associated with the instrumental variables estimation scheme is consistent with the hypothesis that the least squares estimate is positively biased by endogenous immigration inflows. On balance, the pooled data suggest that the effect of immigrant densities on the employment and participation rates of natives is small and potentially zero. If the instrumented first-differenced specification is taken at face value, however, the effect on wages is apparently negative. For the most part, these conclusions carry over to the detailed results for the four subgroups, to which we now turn. Results for Individual RacelSex Groups
Estimates of the relation between immigrant fractions and the labor market outcomes of black males are presented in table 7.8, which has the same format as table 7.7. As in the pooled analysis, the cross-sectional results for black men suggest a negative correlation between the fraction of immigrants and employment outcomes. In the differenced analysis, however, the relation is much less consistent. Likewise, although the 1970 cross-sectional analysis suggests a positive effect of immigration on black male wages, the 1980 crosssectional results and the differenced results indicate a negative effect.
222
Joseph G. Altonji and David Card
Table 7.8
Effects of Immigration on Black Males with Less than Thirteen Years of Education (standard errors in parentheses) Cross-sectional
Outcome Variable 1. Labor force/population
2. Employment/population 3. EmploymenVlabor force 4. Fraction worked last year 5. Log weeks worked 6. Log earningsiweek
1970
1980
- ,145 (.126) - ,264 (.156) - ,165 (.090) - .183 (.loo) - .I54 (.121) ,736 (. 346)
- ,136
(.084) - ,068 (.115) ,046 (.098) - .214 (.081) - ,051 (.Ill) - .153 (.248)
First-Differenced
1980-70 - ,040
(.170) .658 (.234)
,864
(.210) .I01 (.168) - ,447 (.252) - .806 (.494)
1980-70 IVa
- ,273 (.240) ,285 (.234) ,623 ( .294) - ,268 (.168) ,272 (.351) - 1.910 (.7W
Note: All equations include average age and education in the SMSA as well as total population. The sample size is 91. a Estimated by instrumental variables. See the note to table 7.7.
Table 7.9
Effects of Immigration on White Males with Less than 'helve Years of Education (standard errors in parentheses) Cross-sectional
First-Differenced
1970
1980
1980-70
1980-70 IV'
1. Labor force/population
- ,193
2. EmploymenVpopulation
(.075) - .279
- ,079 (.083) - ,159 (.112) -.I10 (.074) - ,215 (.078) - ,312
,066 (.149) ,349 (.186) .343 (. 134) - .145 (.136) - ,018 (.211) - ,356 (.406)
,036 (.231) ,109 (.289) .086 (.211) - ,609 (.211) - ,190 (.328) -1.103 (.637)
Outcome Variable
(.101)
3. EmploymenVlabor force 4. Fraction worked last year
5. Log weeks worked 6. Log earningsiweek
- ,107 (.053) -.151 (.070) - ,223 (.074) - ,264 (.201)
(.I%) -.I78 (.212)
Note: All equations include average age and education in the SMSA as well as total population. The sample size is 120. a Estimated by instrumental variables. See the note to table 7.7.
The results for white male dropouts are presented in table 7.9. These results are very similar to those for black males, although the point estimates of the effects of immigration on wages are somewhat smaller in magnitude. Again, the differenced specifications in particular suggest a negative effect of immigrant densities on native wage rates, while the effects on employment and
223
Immigration and the Labor Market Outcomes of Less-skilled Natives
participation rates are smaller and vary with the precise measure of employment. The regression results for black females in table 7.10 are of particular interest, given the evidence in section 7.2 that black women are in closer competition with immigrants than the other three groups. Nevertheless, the estimated coefficients for this group are not much different than those for the other groups. The cross-sectional results suggest a small negative effect of immigrant shares on employment outcomes and a modest positive effect on weekly wages. These conclusions are reversed, however, in the first-differenced analysis, which suggests a generally positive effect on employment rates and a negative effect on wage rates. The differenced results for black females are not particularly sensitive to choice of least squares or instrumental variables estimation, although as in previous tables the strongest negative wage effect is obtained by the instrumental variables procedure. Table 7.11 presents our results for white females. Again, the crosssectional results for 1970 indicate a negative relation between immigrant shares and employment outcomes, while the differenced analysis indicates much weaker effects. The cross-sectional and first-differenced specifications fit by least squares suggest a positive effect of immigrant shares on wage rates. When the change in immigrant share is instrumented, however, the estimated wage coefficient is negative and consistent with the results for the other native groups. A check on the wage effects reported for the different native groups in tables 7.7-7.11 is contained in table 7.12. Here, we estimate the same specifications using the wage outcomes of immigrant workers as the dependent variable. We Table 7.10
Effects of Immigration on Black Females with Less than Thirteen Year of Education (standard errors in parentheses) Cross-sectional 1970
1980
1. Labor force/population
- ,216
- .063
2. Employment/ population 3. Employmentflabor force 4. Fraction worked last Yea 5. Log weeks worked
- ,221
Outcome Variable
First-Differenced 1980-70
1980-70 IVa ~
(.179)
6. Log eamingsiweek
(.192)
- ,037
(.105)
- ,165 (.169) - ,247 (.232) 1.213 (.402)
(.119) ,003 (.128) .073 (.086) - ,127 (.120) .143 (. 143) .533 (.236)
- .154 (.256) .149 ( ,269) .457 (.186)
,054
(.272) .735 (.387) - .838 ( ,609)
- .221
(.357) .032 (.374) ,320 (.259) - ,219 (.379) .217 (.542) - 1.369 (.848)
Note: All equations include average age and education in the SMSA as well as total population. The sample size is 94. a Estimated by instrumental variables. See the note to table 7.7.
224
Joseph G. AItonji and David Card Effects of Immigration on White Females with Less than Thirteen Years of Education (standard errors in parentheses)
Table 7.11
Cross-sectional Outcome Variable
I . Labor force/population 2. Employment'population 3. Employment'labor force 4. Fraction worked last Year 5. Log weeks worked
1970
1980
- .037 (.144) - ,095 (. 150) -.132
,058 (.097) .027
(.058)
- ,047 (.145) - .094 (.170) ,667 ( ,245)
6. Log eamingslweek
First-Differenced 1980-70 ,273 (.137) .420 (.154) ,306 (.125) .I89 (.146) ,133 (.270) .309 (.430)
(.105)
- .045
(.045) .005
(.098) -.I18 (.110) ,397 (.132)
1980-70 IV" - ,044
(.207) - ,089 ( ,240) - ,017 (.190) - ,162 (.222) ,335 ( .399) - ,955 (.663)
Note: All equations include average age and education in the SMSA as well as total population. The sample size is 120. a Estimated by instrumental variables. See the note to table 7.7.
Effects of Immigration on Male Immigrant Wages (standard errors in parentheses)
Table 7.12 ~
~
~~
Cross-sectional Outcome Variable 1. Log eamings/week
(unadjusted) 2. Log earningdweek (adjusted)
1970
1980
- ,459
- ,741
(.357) ,116 (.302)
(.181) - ,499 (.167)
First-Differenced 1980-70
1980-70 IV'
- ,504 (.381) - ,958 (.354)
- ,823 (S12) - 1.492 (.481)
Nare: Immigrant group includes males age 16-64 not in school in Census week. All equations include average age and education in the SMSA as well as total population. The sample size is 74. a Estimated by instrumental variables. See the note to table 7.7
use two measures of immigrant wages: the mean of actual log weekly earnings for male immigrants and an adjusted mean that controls for the average levels of age and education of immigrants in each city. The results reveal three findings. First, unadjusted mean earnings of immigrants are more strongly correlated in cross section with the fraction of immigrants than mean earnings that have been adjusted for measured skill attributes. This suggests a negative correlation between the skill level of immigrants and their fraction in the population. Second, as we found for the native groups, the instrumental variables estimate of the first-differenced specification leads to the largest negative estimate of the effect of immigrant densities on wages. Finally, the instrumental variables estimates of the effect of immigrant shares on immigrant wages is very similar to the corresponding estimate for native wages. There is no evi-
225
Immigration and the Labor Market Outcomes of Less-skilled Natives
dence that immigrants have a stronger negative effect on their own wages than on those of less-skilled natives.
Other Results We estimated many of our least squares models for the 1970, 1980, and differenced samples with a control for the fraction of blacks in the SMSA population. This addition made little difference to the results. We also reestimated many of our specifications using the fraction of “lessskilled” immigrants in the SMSA population in place of the overall fraction of immigrants in the SMSA population. We defined the fraction of “less-skilled” immigrants as the product of the fraction of immigrants in the SMSA population and the fraction of male immigrants in the SMSA whose predicted e m ings are less than the national median for male immigrants (see App. D). The (unweighted) correlation across 120 cities between the “less-skilled” immigrant fraction and the total immigrant fraction is .94 in 1970 and .95 in 1980. The correlation of changes in the two immigrant measures is .82. Perhaps as a result, least squares results using the fraction of less-skilled immigrants are similar to those reported in tables 7.7-7.11. The regression coefficients typically increase in absolute value, reflecting the fact that the scale of the lessskilled immigrant variable is compressed relative to the other variable. It is worth noting that instrumental variables estimates (using the fraction of immigrants in the SMSA in 1970 and its square as instruments) point to a somewhat larger negative effect of the fraction of less-skilled immigrants on the weekly earnings of natives. The coefficients for black males, white males, black females, and white females are -7.0, -4.8, - 12.9, and - 12.3, respectively. These estimates are very imprecise, however, perhaps because the correlation between fraction of immigrants in 1970 and the change in fraction of less-skilled immigrants in the SMSA is only .27.31 Finally, we reestimated the 1980 cross-sectional specifications and the firstdifferenced specifications for each of our labor market outcome variables using the SMSA-specific mean of the corresponding labor market outcome for white males age 31-64 with thirteen or more years of schooling as a control variable. We view this approach, which uses the labor market outcomes of highly skilled workers to control for general labor market conditions within each city, as an alternative to our instrumental variables procedure. It is strictly correct only if, in contrast to the implications of our model, immigration has no effect on more highly educated whitemales. The results from this alternative procedure are generally similar to our ordinary least squares estimates and suggest smaller negative effects of immigration on less-skilled native wages than the instrumental variables procedure. 7.4
Conclusions
This paper presents a variety of evidence on the effects of immigration on the labor market outcomes of less-skilled natives. Working from a simple
226
Joseph G. Altonji and David Card
theoretical model of a local labor market, we show that the effects of immigration can be estimated from the correlations between the fraction of immigrants in a city and the employment and wage outcomes of natives. We go on to compute these correlations using city-specific outcomes for individuals in 120 major SMSAs in the 1970 and 1980 Censuses. We also use the relative industry distributions of immigrants and natives to provide a direct assessment of the degree of labor market competition between them. Our empirical findings indicate a modest degree of competition between immigrants and less-skilled natives. A comparison of industry distributions shows that an increase in the fraction of immigrants in the labor force translates to an approximately equivalent percentage increase in the supply of labor to industries in which less-skilled natives are employed. Based on this calculation, immigrant inflows of the magnitude observed between 1970 and 1980 generated 1-2 percent increases in labor supply to these industries in most cities. A comparison of the industry distributions of less-skilled natives in high- and low-immigrant-share cities between 1970 and 1980 shows some displacement of natives out of low-wage immigrant-intensive industries. We find little evidence that inflows of immigrants are associated with large or systematic effects on the employment or unemployment rates of less-skilled natives. Our estimates of the effect of immigration on native wage rates are sensitive to the choice of specification and estimation procedure. When we consider first-differencesbetween 1980 and 1970 and use an instrumental variables estimation procedure to control for endogeneity of immigrant inflows, we find that a 1 percentage point increase in the fraction of immigrants in an SMSA reduces less-skilled native wages by roughly 1.2 percent. The least squares estimates imply a wage reduction of . 3 percent. We point out a number of reasons to prefer the instrumental variables procedure, but additional research, perhaps with the 1990 Census, will be required before one can draw strong conclusions about the response of wages to immigration.
Appendix A Sampling Procedures and Variable DeJnitions Sampling Procedures Our 1970 samples are drawn from the 1/100 County Group Public Use Sample based on the 5% version of the 1970 Census questionnaire. The sample universe consists of all individuals age 19-64 currently residing in one of 120 SMSAs. (The samples actually contain 121 SMSAs, but, for comparability with the 1980 Census, Dallas and Fort Worth are considered as one SMSA). As described in the text, our analysis is limited to individuals not
227
Immigration and the Labor Market Outcomes of Less-skilled Natives
currently enrolled in school and in specific race/sex/education and national origin groups from this universe. Our 1980 samples are drawn from the 9100 Public Use “A” Sample of the 1980 Census. The sample universe consists of all individuals age 19-64 currently residing in one of 120 SMSAs (adjusted to 1970 boundaries: see App. B). To limit the size of the samples, stratified random samples of individuals meeting the above requirements were drawn by SMSA. Samples of nativeborn nonblacks (i.e., race coded as white, American Indian, Asian, or other) were drawn to generate approximately twenty-three hundred observations per SMSA for all agelsexleducation levels. The samples were then further restricted to two subsets of observations: females with twelve or fewer years of completed education and males with eleven or fewer years of completed education. Samples of native-born blacks were drawn to generate a maximum of 500 observations per SMSA for black females with twelve or fewer years of completed education and 500 observations per SMSA for black males with twelve or fewer years of completed education. One hundred percent samples of foreign-born individuals were taken for all but five large SMSAs, which were sampled with the following probabilities: Chicago, .400; Los Angeles, .170; Miami, .500; New York, .137; and San Francisco, 3 0 . Labor Market Outcome Variable Definitions The following labor market outcome variables are defined for all individuals in the sample universe: employed in the previous year (P35 = 0 in 1970; P94 = 1 in 1980); in the labor force in the Census week (P31 = 1, 2, 4, 5 in 1970; P81 = 1, 2, 4,5 in 1980); employed in the Census reference week (P31 = 1, 2, 4, 5 in 1970; P81 = 1, 2, 4, 5 in 1980). For individuals in the labor force in the Census week, a fourth variable is defined to be one if the individual was employed in the Census week and zero otherwise. For individuals who worked in the previous year and who reported strictly positive values for the number of weeks worked in the previous year (P36 = 0-5 in 1970; P95 > 0 in 1980) and earnings in the previous year (P37 = 0-500 in 1970; PlOl > 0 in 1980), two additional variables are defined: weeks worked in the previous year and earnings per week in the previous year. For 1980, these variables are constructed directly: weeks worked is measured by variable P95; and earnings per week is measured by P101/P35. (These calculations make no adjustments for allocated responses or truncation of the reported earnings figure.) For 1970, only interval measures of weeks worked and total annual earnings are available. We assigned midpoints of the intervals to the weeks and earnings figures and then constructed earnings per week as the ratio of the assigned values.
228
Joseph G . Altonji and David Card
Appendix B Matching SMSA Dejinitions between I970 and 1980 The Public Use Samples of the 1970 Census identify 125 individual SMSAs (see pp. 123-26 of the Description and Technical Documentation for the Public Use Samples of Basic Records from the 1970 Census). A total of 120 of these are used in our statistical analysis. Four SMSAs were deleted because of difficulty matching between 1970 and 1980 or because of too small sample sizes: Lorain-Elyria, Ohio; Johnstown, Pennsylvania; San BernadinWRiverside, California; and Wilkes Barre-Hazelton, Pennsylvania. The Fort Worth SMSA was merged with Dallas (see below). The Census Bureau publication Geographic Ident$cation Code Scheme (1983, 11-17) gives a detailed list of changes in the county-level definitions of SMSAs between 1970 and 1980. In most cases, these changes involve the addition of surrounding counties or parts of these counties to the SMSA. The major exceptions are (1) the combination of Dallas and Fort Worth into a single SMSA; (2) the creation of a separate SMSA consisting of Nassau and Suffolk counties of New York State (formerly part of the New York SMSA); and (3) the reclassification of Bergen County, New Jersey, from the PatersonClifton-Passaic SMSA to the New York SMSA. Our general matching strategy was to redefine 1980 SMSA boundaries to the 1970 boundaries. With only a few exceptions, this involved deleting individuals from the 1980 Census file who resided in counties that were classified as part of the SMSA in 1980 but not in 1970. For example, Montgomery County, New York, was added to the Albany-Schenectady-Troy SMSA in 1973. Individuals in this county were therefore deleted from the 1980 file. County-level information for each household is coded in the variable COGRP (location 6-8 of the household record) of the Public Use “A” Sample of the 1980 Census. County group codes are obtained from the 1980 County Group Equivalence File (1980 Census of Population and Housing, Public Use Micro Data Sample, part 77) and Appendix M of the 1980 Census Public Use Microdata Samples Technical Documentation. In most cases, individual counties are identified by one or more county group codes. For these cases, the deletion is accomplished by specifying the county group code(s) of those counties added to the SMSA after 1970. In some cases, only parts of a surrounding county group were added to the SMSA. In these cases, we randomly deleted a fraction of individuals from the added county or county group. The fraction of individuals deleted was set equal to the relative population of the part of the county added to the SMSA. Estimates of population for county subgroups were obtained from the 1980 County Group Equivalency File. In all, a total of forty-nine counties or county subgroups were deleted from the definitions of the 120 SMSAs. Another forty counties or county subgroups
229
Immigration and the Labor Market Outcomes of Less-skilled Natives
were partially deleted. The number of individual records actually affected by these deletion procedures is small. For example, of 244,941 immigrants identified on the 1980 Public Use A Sample using the 1980 SMSA definitions, 2,609 (1.07 percent) were deleted in the change to the 1970 definitions. A copy of the computer instructions that performed the deletions is available from David Card on request. To account for changes in the classification of Nassau and Suffolk counties in New York State, we added individuals in the Nassau-Suffolk SMSA in 1980 to the New York SMSA sample. To account for the changes in definition of the Paterson-Clifton-Passaic SMSA, we added individuals in the 1980 sample living in Bergen County, New Jersey (classified as part of the New York SMSA in 1980), to the Paterson-Clifton-Passaic SMSA sample and deleted them from the New York SMSA sample. To account for the reclassification of Dallas and Fort Worth into a single SMSA, we combined individuals from the Dallas and Fort Worth SMSAs in the 1970 Census file into a single DallasFort Worth sample. No attempt was made to deal with minor reclassifications affecting the Boston and Providence SMSAs and the Detroit and Flint SMSAs.
Appendix C Industry Definitions Matching of 1970 and 1980 Three-Digit Codes Our procedure was to reclassify the three-digit industry codes of individuals in the 1970 Census to 1980 industry codes. The Census Bureau provided us with cross-tabulations of 1970 and 1980 three-digit industry codes for samples of males and females who had been coded under both systems. These cross-tabulations were used to estimate the probability that an individual with a given 1970 code would be classified in a particular industry under the 1980 coding scheme. Using these probabilities, a computer program was developed that reclassifies individuals probabilistically from their 1970 three-digit industry to a particular 1980 three-digit industry. The computer program processes males and females separately. A copy of the program is available from David Card. Industry Classifications Used in Tables 7.3-7.6 Using the three-digit industry titles in Appendix H of the Public Use Microdata Samples Technical Documentation, we developed a “two-digit” classification consisting of seventy-six individual industries. (There are 23 1 separate industries in the 1980 Census industry coding system.) This classification combines many smaller three-digit industries: for example, “agricultural ser-
230
Joseph G. Altonji and David Card
vices except horticulture” (industry 020) and “horticultural services” (industry 021). A listing of the computer instructions used to classify three-digit industries into this two-digit system is available from David Card.
Appendix D Classification of High- and Low-Immigrant Cities In order to determine average immigrant skill levels by SMSA, a regression equation was fit to the log of average weekly earnings for the 1980 sample of male immigrants. The equation included the same flexible function of age and education used to regression adjust native outcomes (see the text description) as well as a set of forty-six countryhegion dummy variables and their interactions with an indicator variable for having entered the United States after 1970 and a variable representing years in the United States. (Chiswick [ 19781, Borjas (1985, 19871, and others have shown that country of origin, immigration cohort, and years since immigration affect earnings in the United States.) This equation was then used to assign a predicted wage to each male immigrant. Immigrants with a predicted wage less than the median predicted wage for the entire United States were classified as “low skill.’’ Finally, the fraction of low’lkenty Cities with Highest Fraction of Low-Skill Immigrants
Table 7D.1 City
Miami El Paso Los Angeles Salinas Jersey City Oxnard-Ventura New York Honolulu Paterson Fresno San Diego Anaheim Bakersfield Stockton Santa Barbara San Francisco San Jose Houston San Antonio Providence
Fraction Immigrants
Fraction Low-Skill Immigrants
.36 .21 .22 .19 .24 .13 .21 .15 .15 .ll .13 .13 .09 .ll .12 .16 .14 .08 .07 .09
.20 .20 .16 .16 .15 .10 .10 .10 .09 .09 .08 .08 .08 .08 .07 .07 .07 .06 .06 .06
231
Immigration and the Labor Market Outcomes of Less-skilled Natives
Table 7D.2
Forty Cities with Lowest Fraction of Low-Skill Immigrants
City Huntington-Ashland, KY Chattanooga Birmingham Knoxville York, PA Canton Jackson, MS Cincinnati Dayton Flint Appleton Louisville St. Louis Nashville Indianapolis Richmond Duluth Memphis Akron Greensboro South Bend Utica-Rome, NY Erie, PA Pittsburgh Harrisburg Binghampton Greenville Peoria Wilmington Fort Wayne Mobile Madison Lancaster Toledo Youngstown Lansing Columbus Atlanta Minneapolis Shreveport
Fraction Immigrants .01 .01 .01 .01 .01
.02 .01
.02 .02 .03 .02
Fraction Low-Skill Immigrants
.oo
.oo .oo .oo .oo .oo
.oo .01 .01 .01 .01
.01
.01
.02
.01 .01
.01
.02 .02 .03 .01 .03 .01
.03 .04 .03
.01 .01 .01 .01 .01 .01 .01 .01 .01
.03 .02 .04 .02 .02 .03 .02
.01
.01
.01
.03 .02
.01 .01 .01 .01
.03 .04 .03 .02 .02 .03 .02
.01
.01 .01
.01 .01 .01
.01
.01 .01 .01 .01
skill immigrants in each SMSA was determined by multiplying the fraction of immigrants in the SMSA by the fraction of immigrants who are classified as low skill. Table 7D.1 lists the twenty cities with the highest fraction of lowskill immigrants. Table 7D.2 lists the forty cities with the lowest fraction of low-skill immigrants.
232
Joseph G. Altonji and David Card
Notes 1. Most of the available evidence is summarized by Greenwood and McDowell (1986), General Accounting Office (1988), and Papademetriou et al. (1989). Two studies of particular relevance to ours are Grossman (1982) and Borjas (1987). Lalonde and Topel (in this volume) provide a parallel study to ours, focusing on the effects of recent immigrants on the labor market outcomes of earlier immigrants. Muller and Espenshade (1985) analyze the effect of immigrants on various California cities. 2. A similar conclusion is reached by Kuhn and Wooton (in this volume) and Papademetriou et al. (1989, ch. 4). 3. The average change in the percentage of immigrants between 1970 and 1980 in the 120 SMSAs in our sample is 1.4 and ranges between 0 and 11.4 percent. 4. If the price of output is exogenous, it is more convenient to work with the elasticities of factor prices with respect to factor quantities, holding constant marginal cost. These are usually known as elasticities of complementarity (see, e.g., Hamermesh 1986). 5. This depends, of course, on constant returns to scale and on perfectly elastic supplies of capital and other inputs. 6. In order to avoid the theoretical prediction of factor price equalization across cities, it is necessary to assume that the number of goods produced within a city is less than the number of locally supplied factors. For further discussion of this point, see Kuhn and Wooton (in this volume). 7. We ignore land or any other locally supplied factors. 8. For notational simplicity, we suppress the dependence of c(.) on the prices of nonlabor inputs. 9. In the notation of eqq. (1) and (2), dD,(9, w,)/dw, = 0, and dL,(w,, 9)/d9 = 0, for j = ( u , s). 10. Johnson (1980a) makes the further assumption that the elasticity of labor supply among existing immigrants is zero, so that the effective supply elasticity in the market for unskilled labor is ( I - f,)~, wheref, is the fraction of immigrants in the existing pool of unskilled workers, and E is the labor supply elasticity of natives. 11. That is, O p , + Osust= .6(O, + Os), where 8, represents the value share of labor in thejth skill group. 12. No entries are included in the first row under the column for a,, = .25. In this row of the table, a,, is strongly negative ( - S25). Thus, skilled and unskilled labor must be relatively strong substitutes (i.e., a,, > .8) to satisfy the restrictions on the matrix of partial elasticities. 13. If a,* = ask, eq. (5) implies that the value of the coefficient b, is independent of the substitutability between skilled and unskilled labor. 14. The elasticities of demand for unskilled labor with respect to its own wage rate (qJ implied by the parameter choices in table 7.1 range from - 1.O (in the lower-lefthand entries of the table) to - 2.6 (in the upper-right-hand entries of the table). 15. Estimates of the fraction of output produced in a city that is consumed locally are not easily obtained. Roughly 35 percent of consumer e tures are allocated to personal, health, business, and education services, public s, transportation services, and other goods with a high local content. 16. If the immigrants are primarily unskilled, then one might expect out-migration of unskilled natives and in-migration of skilled natives. 17. Filer (1988) shows that the net migration rate of natives to an SMSA between 1975 and 1980 is negatively related to the migration rate of immigrants into the SMSA between 1970 and 1974 and to the migration rate of immigrants into the SMSA be-
233
Immigration and the Labor Market Outcomes of Less-skilled Natives
tween 1975 and 1980. The negative relation appears to be strongest for low-skilled and less-educated natives. 18. Papademetriou et al. (1989, chap. 4) summarize evidence from a few industry studies suggesting that in some cases immigrant labor has been used to undercut union firms paying higher wages and employing native workers. 19. By “Census week” we mean the week immediately preceding the administration of the Census, for which individuals report their major activity. The Census is administered on 1. April. 20. Our two-digit industry classification is explained in App. C. 21. Of the ten highest-immigrant-share industries in 1980, seven were in the top ten industries by immigrant share in 1970. The rank-order correlation across industries between the 1970 and 1980 immigrant shares is .86. 22. The average fraction of immigrants in the total population in our sample of cities in 1970 was .044 and ranged from .003 to .242. The average fraction of immigrants in the total population in 1980 was .058 and ranged from .008 to .357. 23. It should be pointed out that the index is computed from the industry distribution of existing immigrants and cannot be used to assess the effects of an inflow of immigrants that are much different from the existing stock. 24. The number of white males in private household services is so low that the index cannot be calculated. 25. For example, many high-immigrant-share cities are also major transportation centers (New York, Los Angeles, Miami). This fact may partially explain the relatively high share of the transportation services industry in the high-immigrant-share cities. 26. It is interesting to note that total employment growth rates between 1970 and 1980 for the twenty high-immigrant-share cities and the forty low-immigrant-share cities were virtually identical-the ratio of 1980 to 1970 employment was .92 for the high-immigrant-share cities and .91 for the low-immigrant-share cities. The relative growth rates of less-skilled native employment, however, were somewhat different in the two sets of cities. The relative ratios of 1980 to 1970 employment totals in highversus low-immigrant cities were .96 for white males, .90 for white females, 1.02 for black males, and .87 for black females. 27. A similar approach is used by Borjas (1987). 28. An alternative strategy is to study the effect of immigrant flows to particular SMSAs that one can identify as exogenous. For example, Card (1990) examines the effect of the Maria1 boat lift on the Miami labor market and finds little effect on the wages and unemployment rates of less-skilled blacks and other non-Cuban groups. His results for wages are somewhat at variance with the instrumental variables estimates we report below. 29. The instrumental variables estimation of the first-difference equation also uses these weights. 30. The implied per capita labor supply elasticity is roughly minus one. An alternative explanation, which might be consistent with an extended version of the model allowing for heterogeneity within the population of less-skilled natives, is that a downward shift in the wage distribution induced by immigration results in the exit from the labor force of natives with the lowest skill levels. However, given that the decline in the employment population ratio is small, a compositional shift cannot explain the results even if the wages of those who left employment were essentially zero prior to their departure. 31. In contrast, the correlation between the fraction of immigrants in 1970 and the change in fraction of all immigrants in the SMSA is .60. These correlations refer to the unweighted sample of 120 SMSAs.
234
Joseph G. Altonji and David Card
References Bartel, Ann. 1989. Where do the new U.S. immigrants live? Journal of Labor Economics 7 (October):371-91. Borjas, George. 1985. Assimilation, changes in cohort quality, and earnings of immigrants. Journal of Labor Economics 3(0ctober):463-89. . 1987. Immigrants, minorities, and labor market competition. Industrial and Labor Relations Review 40(April):382-93. Card, David. 1990. The impact of the Mariel boatlift on the Miami labor market. Industrial and Labor Relations Review 43(January):245-57. Chiswick, Barry. 1982. The impact of immigration on the level and distribution of economic well-being. In The gateway: US. immigration issues and policies, ed. Barry Chiswick. Washington, D.C.: American Enterprise Institute. Filer, Randall. 1988. The impact of immigrant arrivals on migratory patterns of native workers. Typescript, Department of Economics, Hunter College-CUNY. General Accounting Office. 1988. Illegal aliens: Influence of illegal workers on wages and working conditions of legal workers. Washington, D.C.: U.S. Government Printing Office. Greenwood, Michael, and John McDowell. 1986. The factor market consequences of U S . immigration. Journal of Economic Literature 24(December): 1738-72. Grossman, Jean. 1982. The substitutability of natives and immigrants in production. Review of Economics and Statistics 64(November):596-603. Hamermesh, Daniel. 1986. The demand for labor in the long run. In Handbook'of labor economics, ed. Orley Ashenfelter and Richard Layard. Amsterdam: NorthHolland. Johnson, George. 1980a. The labor market effects of immigration. Industrial and Labor Relations Review 33(April):33 1-41. . 1980b. The theory of labor market intervention. Economica 47(August): 309-30. Muller, Thomas, and Thomas Espenshade. 1985. Thefourth wave: California's newest immigrants. Washington, D.C.: Urban Institute Press. Papademetriou, Demetrios, et al. 1989. The effects of immigration on the U.S. economy and labor market. U.S. Department of Labor Bureau of International Labor Affairs Immigration Policy and Research Report no. 1, May.
8
Industrial Wage and Employment Determination in an Open Economy Richard B. Freeman and Lawrence F. Katz
The increasing internationalization of the U.S. economy, evinced in the growth of trade, immigration, and (post- 1982) trade imbalance-induced capital flows, raises questions about the responsiveness of the labor market to shocks produced by open economy developments. How do trade-induced changes in product demand and immigration-induced changes in labor supply affect relative wages and employment? Do industrial labor markets respond to shocks generated by international flows of goods and labor as they do to those generated by domestic developments? Does a decline in demand due to international trade (and other factors) reduce wages in an industry relative to those elsewhere? To what extent do wages respond differently in union than in nonunion settings? To what extent do wages respond differently to increases as opposed to decreases in relative demand? To answer these questions, we analyze cross-section time-series data on imports, exports, immigrant shares of employment, annual and hourly earnings, and employment for detailed U.S. manufacturing industries over the period 1958-84 and contrast the responsiveness of the industry earnings in more and less highly unionized industries and between industries facing greater and lesser shocks in sales. In contrast to studies that focus on the direct and indirect effects of the trade balance or immigrant flows on the aggregate economy (using general equilibrium models or input-output analysis), our concern is with direct trade effects on disaggregated industries. I The principal finding is that the industry wage structure responds to Richard B. Freeman is professor of economics at Harvard University and director of the Labor Studies program at the National Bureau of Economic Research. Lawrence F. Katz is associate professor of economics at Harvard University and a research associate of the National Bureau of Economic Research. The authors are extremely grateful to Dan Kessler for expert research assistance and to Lawrence Summers for helpful comments.
235
236
Richard B. Freeman and Lawrence F. Katz
changes in product market sales, with trade-induced changes in sales having approximately the same effect on earnings as sales due to domestic market developments: on average, a 10% annual change in relative industry revenues resulting from trade or other factors alters relative earnings by about .5% over the long run. In addition, we find enough variation in changes in wages, sales held fixed, to trace out a demand curve “trade-off’ between wages and employment across industries. Surprisingly, perhaps, we also find that wages respond more to sales in unionized industries than in nonunionized industries and more to relative declines in sales than to relative increases in sales. Finally, industries with growing or large immigrant shares of employment tend to fall in the industrial wage structure, apparently for reasons beyond any immigrant-native pay differential within industries.
8.1 Potential Labor Market Responses to Ikade-induced Changes in Product Demand When product demand changes in an industry, one expects employment to change in the same direction, with wage adjustments “buffering” the magnitude of job losses or gains. The extent of wage responses to shifts in demand is likely to depend on the mechanisms for wage setting-in particular, on whether wages are set in a decentralized fashion in industry labor markets by supply and demand to clear labor markets; by collective bargaining/administered wage setting that produces premium wages in some industries; or on a national basis with little scope for industry variation. In competitive decentralized wage setting, the extent of wage response to shifts in demand or supply of labor depends on elasticities of demand and supply.*Formally, write the industry demand for labor curve in first-difference form as (1)
dE = -kdW
+ dX,
where E = In employment, W = In wage, X = In shift in the derived labor demand curve due to shifts in product demand, and d = the difference operator. Let the industry labor supply curve be represented as dE
(2)
=
edW
+ dS,
where S = In factors that shift supply. Market clearing produces the following reduced-form relations: dW
(3b)
=
(dX - dS)/(k + e )
dE = (edX - kdS)/(k + e )
Since elasticities are likely to be greater in the long run as factor mobility increases and as firms move in and out of industries, wage responses will be smaller and employment responses greater to any exogenous shock as time
237
Industrial Wage and Employment Determination in an Open Economy
proceeds. In the extreme, when the elasticity of labor supply to an industry approaches infinity, there is no wage response to shifts in either schedule, and the sole change is in employment. Under collective bargainedladministered decentralized wage setting, wages in particular industries diverge from market-clearing rates for any of a number of possible reasons (efficiency wages, rent sharing, collective bargaining), producing a queue of workers at going rates and an effective infinitehear infinite elasticity of labor supply to the ~ e c t o r .Since ~ neither employers nor unions are directly constrained by labor supply conditions, there are several possible wage responses to shifts in demand. Some argue that, in markets where senior workers have a disproportionate influence on wage setting, wages are less responsive to trade-induced changes in demand than competitively determined wages. Grossman (1984) analyzes the conditions for this to be true in a model in which wages are set (subject to a labor demand constraint) by a majority-voting union with a seniority layoff rule and free entry into the union. Modeling an increase in international competition as a exogenous decline in the world price of the product produced by the unionized sector, he shows that international competition has two offsetting effects on the wage. For a union of a given size, a lower wage will be desired because greater international competition increases the risk of layoffs. On the other hand, declines in union membership will raise the average seniority level, which, he argues, produces a median member who wants higher wages. The net effect is ambiguous in general, but, in the case of a constant elasticity labor demand schedule, his model predicts that the union wage will be completely unresponsive to the international price.4 Others note the possibility (and existence in some industries in some periods) of “endgame bargaining” in which unions, seeing little future to an industry, seek to extract as much as they can in a short period (Lawrence and Lawrence 1985). Our analysis emphasizes the possibility of greater-than-competitive downward wage adjustments when wages exceed outside alternatives in a unionized-administered wage sector, so that we expect larger downward wage adjustment for union than for nonunion workers. The view that wage responses to increased product market competition may be greater under unionism is consistent with research on the effects of trucking deregulation on wages that finds substantial relative wage reductions for union truckers and much less wage response for nonunion truckers following deregulation (Rose 1987). Consider, finally, industry wage responses under a centralized system in which wages are set nationally (corporatist economies) or in which there is considerable “spillover” or “flow on” of changes in wages across industries. In these settings, we would anticipate shifts in demand to have little effect on wages but substantial effects on employment. The existence of such wagesetting systems outside the United States provides potentially fruitful controls for evaluating the effect of decentralized wage setting in the United States on wage responsiveness.
238
Richard B. Freeman and Lawrence F. Katz
8.1.1 Modeling Union Behavior Consider first the wage policy of a union concerned with both wages and employment that is subject to a labor demand constraint. In simplest form, its decision making can be viewed as maximizing a utility function U ( W , E ) subject to a labor demand constraint E = E (W). Then the maximizing condition is to set wages so that the ratio of the marginal value of employment to the marginal value of wages equals the elasticity of labor demand. If trade (or other factors) increases the elasticity of demand, wages are likely to drop, as the wage-employment trade-off facing the union is worsened. Huizinga (1987) shows that, in imperfectly competitive product markets, an increase in international competition is likely to increase the elasticity of the product demand elasticity facing domestic firms and lead to wage concessions by a monopoly union. If trade (or other factors) leaves the elasticity unchanged but shifts the demand curve downward, the union is also likely to lower wages when demand declines and raise them when demand increases. Similar implications can be derived for the standard utilitarian union model (McDonald and Solow 1981) in which a union with a fixed membership maximizes the welfare of the representative member subject to the labor demand ~ o n s t r a i n tThe . ~ union maximizes
where N is the fixed membership, u(.) is the utility function of the representative member, e is the level of employment, w * is the opportunity (or alternative) wage, and w is the wage level. The maximand can be rewritten as e(w)[u(w)- u(w*)]dropping a constant term. This formulation yields the familiar optimizing condition in which the elasticity of the gain from employment is equated to the elasticity of labor demand:
-u’(w)w/[u(w)- u(w*)]= e’(w)w/e. Here, the union wage depends only on the elasticity of labor demand and the degree of risk aversion of the representative union member. Changes in international competition that affect the elasticity of labor demand will affect wages in the same direction, while shifts in labor demand not changing the elasticity will not affect wages. In the case of efficient bargains between the union and firms, the level of labor demand will affect wages as well as the elasticity.6 When senior workers play an especially important role in the union, there is a strong possibility of a more complex response pattern. Faced with a positive shock in demand, existing union members are likely to weigh wage gains highly relative to employment gains, producing sizable increases in wages. Faced with modest negative shocks, they are less likely to sacrifice rents to save the jobs of marginal employees, producing wage inertia. Faced with siz-
239
Industrial Wage and Employment Determination in an Open Economy
able negative shocks and threats of plant shutdowns, on the other hand, existing unionists may be willing to offer large wage concessions. As there is reasonable a priori logic for expecting unions to respond less, more, or even “perversely” to shocks due to trade or other factors, the question of which response pattern dominates actual wage setting in the United States is an empirical one.
8.1.2 Sales and Shifts in Demand The models described thus far relate changes in wages and employment to exogenous shifts in demand and supply. To apply them to data, it is necessary to measure the exogenous shifts, which, given our focus on trade, requires that we obtain appropriate indicators of shifts in product demand and the contribution of trade to such shifts. In this study, our primary indicator of shifts are industry sales and its price and quantity components and sales decomposed into domestic market sales, exports, and imports, appropriately weighted to take account of their relative magnitudes. As sales depend on industry supply as well as demand conditions, however, simply replacing the X terms in (3a)-(3b) with sales and regressing wages/employment on sales does not yield the desired response parameters. There is a potentially important simultaneity bias due to the effect of wages on industry prices and output. As a first cut at the simultaneity problem, assume that the supply curve of industry output is flat, so that prices depend solely on costs of production. Then we can model the simultaneous relation between wages and sales with the following simple market model: a) product demand: (4)
dQ = -hdP
+ dX,
where Q = In output, P = In price of output, and X = In shifts in demand, as in (3a); b) the effect of wages on cost of productiodproduct price: dP
=
adW,
where a is a labor’s share of cost; c ) wage determination equation: dW
=
qdX,
where q is the parameter of interest to us ( = l/[k
+ el in eq. [3a]).
Note that this equation makes wage changes depend not on observable changes in prices or quantities (which are affected by wages) but rather on the unobserved exogenous shift in product market conditions. Substituting (5) into (4) yields a relation between output and wages: (7)
dQ = -hadW
+ dX,
240
Richard B. Freeman and Lawrence F. Katz
which, in turn, yields a relation between sales (dS = dQ dS = (1 - h)adW
+ dP) and wages:
+ dX.
Solving for dX in (8)' and substituting into (6) yields an equation between changes in wages and changes in sales?
(9)
dW
=
{q/[l
+ q(1 - h ) ~ ] } d =S AdS.
Adding an error term to (9) with the usual properties, we can estimate the parameter A by least squares regression of observables on observables. We are interested, however, not in A but in q , the response of wages to changes in product market conditions. Rearranging terms we see that, for any estimate of A: (10)
= A/[1 - Aa(1 - h)].
Equation (10) shows us that the estimated parameter of wages on sales yields the correct response coefficient only if the product demand elasticity is unity. If h is less than unity, A will understate q, while, if h is greater than unity, it will overstate q , with the magnitude of the difference between A and q dependent on the magnitude of Aa(1 - h). For reasonable values of the parameters, however, it turns out that the difference between A and q will be small. For example, with the mean value of a (labor's share of cost of sales) in our data of .25 and estimated values of A below .lo, the bias is modest for anything short of huge elasticities of product demand.9 In the context of the model of equations (3a)-(3b), moreover, the difference between the parameter relating employment to shifts in demand and the regression coefficient of employment on sales will also be small.'O The econometrics gets more complicated, however, if, rather than adding an error term to equation (9), we allow for error terms in each of the underlying equations as well. As we substituted for dX to get (9), the error terms in the price and wage determination equations become part of the error structure in (9), with the result that dS is correlated with the error.II In this case, it is necessary to instrument dS to obtain a consistent estimate of A.I2 As an alternative way of modeling the relation between wages, prices, and output in an open economy, consider the situation when prices are determined on world markets so that an industry in a given country can sell as much as it produces at the going world market price. Here, there is still likely to be a feedback of wages on sales, as increases in wages increase costs of production and reduce output, thereby reducing sales. We model this market pattern by assuming an upward-sloping industry supply curve with a fixed elasticity. Following a logic analogous to that in (4)-(10) above, we can show that the regression of wages on sales leads to an understatement of the parameter of wage responsiveness of exogenous shifts in market demand, essentially because the reverse causality is negative.
Industrial Wage and Employment Determination in an Open Economy
241
8.1.3 The Trade Component of Sales Turning to the effect of changes in trade, we decompose sales into its component parts-the size of the domestic market (DOM = sales - eximports); exports; and the import share of domestic market sales ports (MSHR = imports/DOM)-and use a first-order approximation to obtain
+
(11)
dS = w,d In (DOM)
+ w,d In exports - w,d(MSHR),
where w, = (sales - exports)/sales, w z = exports/sales and, w, = DOWsales. The weights are obtained by considering the effect of small changes in domestic-generated revenues, export-generated revenues, and the import share of revenues on changes in total revenues in a decomposition that ignores interaction terms, and will, accordingly, be more accurate for small than for large changes. The purpose of the weighting is to adjust the relevant changes for the difference in absolute magnitude of sales generated by domestic demand (90% or so of sales) and trade. When there is trade balance in an industry, the weight on changes in the import share of the domestic market becomes unity [DOWsales = (sales imports - exports)/sales = sales/ sales]. When, as in the 1980% imports exceed exports, the weight placed on this term exceeds unity. Substituting (1 1) into (9), we obtain a relation between wages (or employment) and weighted In changes in the domestic and foreign components of revenues:
+
(12)
dW
=
+ +
Aw, d In (DOM) Aw,(d In exports) - Aw,[d(MSHR)] other factors.
Note that this model makes a strong implicit assumption about market behavior: it postulates that the labor market responds similarly to (weighted) changes in sales due to trade-related factors as to those due to domestic factors. While in the short run there may be some differences in market responses to trade-generated as opposed to domestic market-generated changes due, say, to differing assessments of whether changes will persist over time (e.g., because foreign competition depends on highly volatile exchange rates), in the long run we see no compelling argument to expect industry labor markets to react any differently to changes in revenues from different sources: a 10% shift in demand is a 10% shift in demand. If the assumption that foreign- and domestic-based changes have the same effects on the labor market is valid, then the coefficients on the trade and domestic revenue terms will be similar in regression analysis.
8.1.4 Shifts in Supply and Immigrant Labor The impact of immigrant labor on industry wage levels is twofold. To the extent that immigrants are paid differently than otherwise comparable native-
242
Richard B. Freeman and Lawrence F. Katz
born workers, average wages in an industry will depend on the immigrant share of labor with a coefficient equal to the wage-differential between immigrants and native workers. Changes in the immigrant share of labor in an industry will, accordingly, be associated with changes in industry wages:
(13)
dW = bd(IMS),
where IMS = immigrant share of the work force and b = wage differential between native and immigrant labor. In addition, however, if immigrant labor is a good substitute for native labor in immigrant-intensive industries, supply-induced changes in immigrant shares will alter the wages of natives in the industry as well, further reducing the position of the industry in the industry wage structure.
8.2 Cross-Industry Analysis for U.S. Manufacturing We estimate wage-sales and employment-sales equations using the NBER Trade and Immigration Industry data set for manufacturing production workers.I4 The data set provides information on the wages and employment of production and all workers, trade flows and immigrant shares of employment, as well as other control variables in 428 four-digit SIC manufacturing industries from 1958 to 1984.15We examine the data in three ways. First, we analyze changes over the twenty-six-year period from 1958 to 1984, which can be viewed as reflecting changes in long-run comparative statics for the “average” industry. Second, we relate annual changes in wages and employment across industries to changes in revenues and the part of those changes due to trade and domestic demand, with individual year dummy variables entered in our regressions to capture economy-wide cyclic-type phenomena. I 6 Third, we explore responses over different time periods to see whether responses in the 1980s period of large trade imbalances differ from those in earlier periods. As a check on the results from our establishment-based analysis, we also estimate wage change equations utilizing industry wage differentials estimated from the 1974 and 1984 Current Population Surveys (CPSs) for the fifty-eight three-digit 1980 Census industries (CICs) that can accurately be matched to the 1970 CIC system used in the 1974 CPS and to the NBER trade and immigration figures. Our pooled cross-section time-series industry analysis differs, it should be noted, from the time-series analyses for particular industries that other researchers have used to investigate the effect of trade on the labor market.” We examine the relative responsiveness of industries to the particular shocks that face them, exploiting the differential patterns of change among industries rather than the time-series patterns of change for a particular industry. As a consequence, our estimated response parameters are average elasticities of response across industries. Formally, in terms of the model of (12), if each industry has its own response parameter a ai, where a is the mean of the
+
243
Industrial Wage and Employment Determination in an Open Economy
industry response parameters, the form of our basic equation can be written as (14)
dW,, = (a
+ a,)dS,, = ads,, + a@,,,
Where the latter term becomes part of the error structure. This term reflecting the heterogeneity in industry responses creates heteroskedasticity in the errors but does not bias estimation of the average response coefficient as long as the individual industry component of the response (a,) is independent of other variables in the equation.I8 In section 8.3, we consider potential differences in response coefficients among industries. 8.2.1 Data Description Table 8.1 gives the 1984 level of variables, 1958-84 changes in variables, and standard deviations for variables of concern to us (pt. A) and selected correlations of the changes for the period 1958-84 (pt. B). The descriptive statistics reveal several characteristics of industry labor and product markets that underlie the ensuing econometric results: The shares of imports and exports relative to the size of the domestic market are relatively modest even in 1984 after two or so decades of rapid growth of trade, with imports averaging 14% of domestic demand and exports averaging 8% of sales. Immigrant shares of labor are also modest, averaging 8% in 1984.19 The principal dependent variables of concern to us-changes in In annual earnings (obtained by dividing payroll by employment) and in In hourly earnings (obtained by dividing payroll by person-hours)-show nearly identical industry variation over the period 1958-84, indicating virtually identical patterns of change in hours per employee. The correlation in the 1954-84 change in In annual earnings and change in log hourly earnings is .95. When, by contrast, we examine short-run year-to-year changes, we find considerable variation in hours per worker across industries and thus differences in changes in annual and hourly earnings. The standard deviation of In changes of industry employment exceeds the standard deviation of In changes in hourly earnings by a factor of 3.9 (.67/ .17), documenting the fact that quantity adjustments dominate industrial labor markets, possibly because workers are good substitutes across industry lines. In the goods market, the standard deviation of In changes in physical output (deflated sales) exceeds the standard deviation of In changes in prices by a similar proportion (3.0 = .73/.24). The standard deviations of our major independent variables-weighted changes in domestic demand, foreign demand, and the import shareshow considerable interindustry variation, as is necessary if we are to estimate their effects on the labor market with any precision. Turning to the correlations in part B of the table, note first the .42 positive correlation between changes in the level of imports and changes in In employ-
244
Richard B. Freeman and Lawrence F. Katz
Table 8.1
Descriptive Statistics for NBER 'kade and ImmigrationData Set, 428 U.S. Manufacturing Industries A. Major Variables
Mean Levels, 1984: Imports/domestic demand Exports/sales ImmigrantslEmployment Changes, 1958-84: DLN(hour1y wages for production workers) DLN(annua1 wages for production workers) DLN(production employment) ImmigrantsiEmplo yment DLN(sales) DLN(price) DLN(output) Weighted changes, 1958-84: * DLN(domestic demand) DLN(foreign demand) Import share
SD
.14 .08 .08
.I6 .10 .05
1.43 1.43 - .02 - .OO 1.74 1.06 .67
.I1 .18 .67 .03 .72 .24 .73
1.71 .16 .I5
.65 .23 .43
B. Correlations, 1958-84: Log Changes Weighted Changes Production Hourly EmployWages ment Hourly wages Production employment % Immigrantsb
1.00
.04
.04
1.00 .06
-.I2
%I&grants
Domestic Foreign Import Sales Imports Exports Demand Demand Share
-.12
.26
-.08
.I0
.13
.I5
.90 .04
.42 .13
.44 .05
.83 .10
.23 - .04
.06 1.00
-.23 -.18 .06
Note: DLN(.) = change in logarithm of the variable. a Weights utilized are the average of the 1958 and 1984 weights. Absolute change in percent immigrant.
ment. If industries with increasing imports expand employment, why is there such public concern over the effect of imports on jobs? The reason for the seeming paradox is that imports and domestic production tend to increase in the same industries-namely, those where domestic demand is growing. This highlights the need to control for total demand or, alternatively, to focus on the import share of the domestic market, in estimating the effect of imports on the labor market. Indeed, the (weighted) change in the import share of domestic sales is negatively correlated with changes in employment at a highly significant - .18. Similarly, with respect to earnings, while the correlation between changes in imports and earnings is a modest - .08, the correlation between changes in earnings and the (weighted) change in import share is a
245
Industrial Wage and Employment Determination in an Open Economy
hefty - .23. Note finally that changes in the immigrant share of employment in an industry are negatively related to changes in wages and positively related to employment changes, indicating that immigrants have gone into industries with growing employment but declining wages.*O 8.2.2
Basic Regression Results
Table 8.2 presents coefficients and standard errors on the major variables of concern from our regression analyses of long-term (1958-84) changes in hourly earnings and personhours worked. All the regressions contain the set of controls listed at the bottom of the table, including two-digit SIC industry dummies (to allow for broad industry differences in responses). Columns 1-3 show the estimated effect on hourly earnings of sales, prices, and physical output taken separately and weighted trade and domestic revenues; columns 4-6 show the estimated effects on hours of sales, price, and quantities and of weighted trade and domestic revenues. In addition, we report the coefficients and standard errors on the change and base year level of immigrant shares in each regression. We do not report the results of comparable regressions for Table 8.2
Coefficients(standard errors) for Effects of Changes in Sales, Trade Variables, and Immigration Ratios on Wages and Employment of Production Workers in U.S. Manufacturing Industries: Long Period Log Changes (N = 428), 1958-84 Hourly Wages
Variables
(1)
(2)
Annual Hours (3)
(4)
(5)
(6)
~
Change in log sales Change in log output Change in log price Change in % immigrant % Immigrant in 1960
,886 (.017)
,049 (.011)
-.445
(.286) -1.164 (441)
,049 (.011) ,108 (.035) -.389 (.287) -1.198 (41)
Weighted log change in domestic demand Weighted log change in foreign demand Weighted change in import share
RZ
- ,432 (.286) -1.164 ( ,440) ,040 (.011) .076 (.034)
1.082 (.462) ,900 (.713)
,886 (.017) .884 (.057) 1.080 (.466) ,901 (.714)
.89
.89
- ,064 (.017)
.38
.39
.39
1.002 (.497) ,545 (.765) ,894 (.020) ,710 (.059) - .479 (.030) .88
~~
Nore: The reported regressions include two-digit SIC dummies, change in percent union, change in percent production workers, and the initial (1958) values of the following variables: percent union, percent production workers, and log of value added per worker. Weights utilized in weighted log changes are means of 1958 and 1984 weights.
246
Richard B. Freeman and Lawrence F. Katz
annual earnings and production worker employment as they yielded virtually identical coefficients to those in the table because of the lack of industry variation in hours per employee over the long run pointed out on page 243.21 There are three principal findings. First, the calculations show that changes in revenues significantly affect relative wages, implying that the industry wage structure is “flexible” with respect to changes in the market conditions in particular industries. Using the reduced-form model of (3a)-(3b) to interpret the results, the ratio of the estimated effect of sales on hours to the estimated effect of sales on earnings provides a measure of the magnitude of the elasticity of labor supply across industries: from columns 4 and 1, it is 18.0, which implies considerable labor mobility across industry lines in response to changes in wages. Second, despite the long period under study, changes in the (weighted) import share of the domestic market and in exports have roughly comparable effects on‘wages, as do weighted changes in domestic shipments, indicating that, as a first approximation, the industrial wage structure in the United States responds to open economy developments to the same extent as to domestic developments. In addition, the trade-generated and domestic market-generated changes in sales have significant effects on person-hours, though with noticeably different estimated coefficients. Third, the long period change estimates show that both the change and the initial level of the immigrant ratio are negatively related to changes in hourly earnings and positively related to changes in employment. The magnitudes of the coefficients on immigrant shares are, however, too large to be attributed to pure immigrant-native wage differentials given the small proportion of immigrant workers and likely modest differentials correcting for worker skill (Borjas 1985; Chiswick 1978). They are more likely to reflect the concentration and movement of immigrants into low- and declining-wage industries. That immigrants find jobs in industries that are increasing employment but falling in the wage structure is consistent with the basic fact that employment in the United States has been growing in industries with low and relatively declining wages. 8.2.3 Annual Changes Next we examine the effect of domestic- and foreign-generated changes in sales on earnings using annual rather than long period changes in the variables. Because annual earnings and hourly earnings differ in the short run, owing to short-run variations in person hours worked relative to employees, we report results for both earnings variables. Because our focus is on interindustry responses, we include year dummies in these regressions to control for general cyclic phenomena. As can be seen in table 8.3, we obtained somewhat different results between annual and hourly earnings. First, changes in sales had much larger effects on annual earnings than on hourly earnings, owing to the implicit re-
247
Industrial Wage and Employment Determination in an Open Economy
Table 8.3
Coefficients(standard errors) for Effects of Changes in Sales and Wade Variables on Wages in U.S. Manufacturing Industries: Annual Log Changes (N = 11,165), 1959-84 Annual Earnings
Variables Change in log sales
(1)
(2)
(3)
(4)
,056
(.007)
Weighted log change in domestic demand Weighted log change in foreign demand Weighted change in import share
,066
.029
(.004)
.008
,071 (.015) - ,059 (.011) .22
(6)
,026 (.@J7) .032 (.007)
,069
(.004)
Change in log price
(5)
,027 (.004)
,068 (.004)
Change in log output
R2
Hourly Earnings
.22
.22
(.016) - .01I (.012)
.19
.20
.20
Note: The reported regressions include year dummies, two-digit SIC dummies, change in percent union, and change in percent production workers as well as the variables listed above.
sponse of hours per employee to changes in sales (the difference in the effect of a variable on annual and hourly earnings is its effect on hours per worker) as firms responded to relative declines in sales by reducing work hours and/or temporarily laying off some workers and responded to increases in sales by increasing work hours, including over time. Comparing the effect of sales on hourly earnings in table 8.3 with the effect of sales on hourly earnings in table 8.2, we find that hourly earnings adjustments are greater in the long run, contrary to the purely competitive model of (3a)-(3b). This can be taken as evidence for collectively bargained/administrated wage settlements as opposed to spot market settlements. Second, and more disturbing to our analysis, table 8.3 shows that, while annual changes in sales due to open economy developments have substantial and well-defined effects on changes in annual earnings, they have statistically insignificant effects on changes in hourly earnings, contrary to the findings of table 8.2. To reconcile the findings on the magnitude of the effect of sales and tradeinduced changes in sales in the short run and long run, we made two further calculations. First, we examined the determination of hourly earnings over three intermediate periods: 1958-70, which covers the 1960s strong job market; 1970-80, when the economy was sluggish; and 1980-84, when the country developed an extraordinary trade imbalance. The results of this analysis are given in table 8.4. Consistent with the results for the entire period, they show substantial and significant effects for import-induced changes in sales on
248
Richard B. Freeman and Lawrence F. Katz
Table 8.4
Long Period Log Hourly Wage Change Regressions by Time Period, U.S. Manufacturing Industries, 1958-70, 1970-80, 1980-84 Time Period
Variables Change in log sales Weighted log change in domestic demand Weighted log change in foreign demand Weighted change in import share Change in % immigrant R2
N
1958-70
1970-80
.022
.035
(.ow (.015)
1980-84
1958-70
1970-80
1980-84
,084 (.018)
- .148 (.271)
.031 (.017) ,043 (.033) -.076 (.057) - ,342 (34)
.28 428
.26 428
.086 (.016) ,017
(.009)
,095 (.039) - ,141 (.055)
- ,263 (.269)
-.325 ( .262)
.27 428
.26 428
.. . .20 428
,040
(.066) -.067 (.018)
... .19 428
Note: The reported regressions include two-digit SIC dummies, change in percent union, change
in percent production workers, and the initial values of the following variables: percent union, percent immigrant, percent production workers, and log of value added per worker. Weights utilized in weighted log changes are mean of initial and final period weights. The 1958-70 regressions do not include change in percent union, and the 1980-84 regressions do not includz change in percent immigrant because the required data are not available for these variables over these periods.
hourly earnings but weaker effects for export-induced changes. The period regressions also show a marked pattern of differences in wage responsiveness among the periods, with the effect of sales on hourly earnings greatest in the 1980s, as might be expected given the wage concessions of that period.22Second, we have explored the timing of the effect of sales on hourly and annual earnings by including lagged sales variables in our regressions of annual changes in hourly earnings on annual changes in sales. The results of these calculations (see table 8.5) suggest that within three years the effect of changes in sales on hourly earnings rises to the long-run level and is roughly equal to the effect of changes in sales on annual earnings. The differences in timing of the effect of sales on hourly and annual earnings suggest that hours worked (which vary because of both layoffs and overtime or short time) may be an important indicator to workers of the need to adjust hourly pay in the face of demand shocks. For our purposes, what matters is that these calculations show that the long period hourly earnings results are the valid ones for assessing adjustments beyond a year or so.
8.2.4 Current Population Survey Data
As a check on our findings from establishment data, we have also estimated the effect of trade and immigration on industry wages using household data from the CPS tapes. These calculations have the advantage of letting us con-
249
Industrial Wage and Employment Determination in an Open Economy Regression coefficients (standard errors) for the Effect of dS on dw
Table 8.5
Hourly Earnings
Annual Earnings
.029
dS
,069
(.ow
(.ow - ,027 (.ow .005 (.ow
- .001
&(-I)
(.ow
ds( - 2 )
.017
(.ow
Sum
,045
.
,047
Nore: The coefficients are based on the same specifications as in table 8.3, with the addition of the lagged sales variables.
Table 8.6
Long Period CPS Industry Wage Differential Change Regressions, 1974-84: Fifty-Eight Three-Digit 1980 CIC U.S. Manufacturing Industries All
Union
Variables Change in log sales
(3)
R2
N
-.869 (.409) .21 58
(4)
- .ow (.031)
.I31 (.047)
,057 (.025)
Weighted log change in domestic demand Weighted log change in foreign demand Weighted change in import share Change in % immigrant
Nonunion
,010 (.029) ,019 (.095) - ,228 (.050) -.523 (.383) .39 58
.097 (.055) -.180 (.181)
- 1.053 (.751)
- ,367 (496) - .737 (.731)
.18 58
.31 58
-.612 (.501) .09
58
- ,089 (.034) ,067 (.112) - ,221 (.059) -.071 (.451) .35 58
Note: Reported regressions include change in percent union and change in percent production workers.
trol for individual characteristics that affect earnings at the cost of limiting the sample to fewer and more aggregated industries. We proceeded in a two-part analysis. In step 1, we estimated industry wage effects by regressing the In of average hourly earnings of individuals on their characteristics and dummy variables for the In step 2, we regressed the change in the estimated industry effect on changes in sales, immigrant ratios, and sales decomposed between trade and domestic factors. The basic results, shown in columns 1 and 2 of table 8.6 (we will discuss the findings in cols. 3-6 shortly), confirm the table 8.2 finding that, over an extended period of time, changes in revenues due to trade substantially aEect industry hourly earnings and also confirm the finding that industries with growing immigrant shares of the labor
Richard B. Freeman and Lawrence F. Katz
250
force fall in the wage structure. Indeed, in these calculations, the dominant factor in changes in wages in the period 1974-84 is the import part of sales. 8.2.5
The Wage-Employment Trade-off
While our analysis shows that wages respond to changes in sales resulting from trade and other factors, there is also a significant independent or unexplained component to changes in wages as well. Does this component of the change in wages affect the quantity of labor used, sales held fixed? To what extent do cross-industry data show a trade-off between wage responsiveness and the employment of labor? To answer these questions, we have performed the regression calculations summarized in table 8.7, in which we relate annual and long-run changes in In annual production hours to changes in wages and sales (cols. 1, 4),product wages and deflated output (cols. 2, 5), and wages and sales decomposed into trade and domestic market determinants (cols. 3, 6), using both annual change and long period change data. The results offer strong support for the notion that, sales held fixed, there is a significant wageemployment trade-off across industries in the U.S. labor market. Employment Table 8.7
The Wage-Employment 'kade-off in U.S. Manufacturing Industries: Dependent Variable = Change in Log Annual Hours of Production Workers Annual Changes, 1959-84
Variables Change in log wage
(1)
(2)
- ,629 (.013)
- ,628
(.013) Change in log product wage Change in log sales Change in log output
N
(4)
- ,612 (.079) (.045)
.921 (.016)
,670
,921 (.016)
,699 (.005)
,922 (.018) ,754
.649 (.006) ,676 (.022) - ,545 (.017) .71 11,165
(6)
- .887
(.008)
(.om
(5)
- ,683 (.071)
- ,545
Weighted log change in domestic demand Weighted log change in foreign demand Weighted change in import share
R'
(3)
Long Period Changes, 1958-84
.72 11,165
.69 11,165
(.055)
- ,524 (.028) .9 1 428
.91 428
.89 428
Nure: The reported annual change regressions include year dummies, two-digit SIC dummies, change in percent union, and change in percent production workers as well as the variables listed above. The reported long period change regressions include the same variables except for the time dummies and in addition include the change in percent immigrant. The wage variable utilized in all the regressions is the hourly wage of production workers. The change in the log product wage is given by the difference in the change in the log wage and the change in the log shipments deflator.
251
Industrial Wage and Employment Determination in an Open Economy
growth is lower, sales fixed, by roughly 5%-9% in industries where wages rise by lo%, relative to other industries. While we recognize that one cannot interpret the estimated relation as a demand curve, particularly in light of our analysis of wages as dependent on sales, the inverse relation is nonetheless impressive. Unfortunately, we lack good measures of shifts in labor supply to industries and of factors that lead to different union/employer wage-setting policies to estimate a structural demand equation given our model, in which wages are endogenous.
8.3 Differences in Wage Responses among Sectors Are the wage responses found in section 8.2 the same across all industries, or do different wage-setting institutions or economic conditions produce different responses to changes in sales? Do the more heavily unionized industries respond more, or less, to changes in sales due to trade and other factors than do the less unionized industries? Is there evidence of asymmetric wage responses to increases and decreases in sales? 8.3.1
Union and Nonunion Responsiveness
To evaluate the effect of trade unionism on the degree of wage responsiveness, we have performed two related analyses. First, we estimated earnings and employment response equations separately for industries whose union density made them high (upper third), medium (middle third), or low (lower third) in the distribution of union density as of 1973-75. Second, we added interaction terms between changes in sales and dummy variables for high, medium, and low union status to our basic regressions. As the findings for the two analyses were quite similar, we present for purposes of parsimony the separate union-density category regression results in table 8.8. Row 1 of the table records the estimated effects of sales on In hourly earnings by union class from annual change regressions. It shows that earnings responses tend to be higher in the more highly unionized Given that the scope for union wage responsiveness is likely to be greater the greater the gap between union and other wages, we take the analysis a step further in rows 2-5, by estimating a single change in In hourly earnings regression with more complex interaction terms that distinguish not only between high, medium, and low union density but also between industries with high (upper third), medium (middle third), and low (lower third) hourly earnings in 1958. These calculations show that it is the responsiveness of union wages to changes in sales in high-wage industries that underlies the greater elasticity of union wages to sales in row 1. To the extent that wages are high in these industries because of large union wage effects, this finding supports the notion that wage adjustments are greater where wages exceed competitive market levels. Finally, we note that regressions comparable to those in table 8.8 with employment as the dependent variable show that the pattern of change in employment
Richard B. Freeman and Lawrence F. Katz
252 Table 8.8
Wage Responsiveness to Changes in Sales by Union Density: Annual
Log Wage Change Regressions, Hourly Wages, 1959-84 Union Density
Wage Class 1. All industries
2. High initial wage 3. Medium initial wage 4. Low initial wage
High ,041
(.ow
,088 (.025) .035 (.026) ,027 (.022)
Medium
Low
,021 (.007) ,066
.019 (.007) - ,010
(.019)
(.013)
,003 (.022) .023
,032 (.014)
,031
(.012)
Note: The union density and initial wage classes are derived by dividing the industries into thirds on the basis of initial union density and initial hourly wage. Row 1 presents the coefficients (standard errors) on change in long sales for separate log wage change regressions by union density class, which also include change in percent production workers, change in percent union, and time dummies. The regression used for rows 2, 3, and 4 included the same controls plus initial wage class and union class dummies, two-digit SIC dummies, and a full set of interactions of the union class and wage class dummies with change in log sales. Numbers in parentheses in rows 2, 3, and 4 are standard errors for difference between reported coefficient and the coefficient on the base group, low union, low initial wage.
responsiveness is opposite that for wage responsiveness, with employment responsiveness declining with union density, as would be expected if the wage adjustments serve to “buffer” employment in the market.25 Because the establishment-based data set does not permit us to differentiate between union and nonunion firms or workers within an industry, it is possible that the differences found in table 8.8 are not due to genuine differences in behavior between union and nonunion firms.26Accordingly, we have also estimated response parameters for union and nonunion parts of industries using our CPS data set, where it is possible to distinguish between union and nonunion workers. Here, we estimated industry wage effects by regressing In wages on worker characteristics and industry dummy variables for union and nonunion workers taken separately and then regressed the change in the industry effects on changes in industry sales, immigrant ratios, etc. These results, given in the regression coefficients in columns 3-6 of table 8.6, support the finding that unionization increases rather than reduces wage responsiveness: the industry differentials for unionized workers are significantly influenced by changes in sales, and the import component of changes in sales has an especially large effect in the union sector. By contrast, there is no noticeable effect of changes in sales or the import component of those changes on the wages of nonunion workers.
8.3.2 Responses of Industries at the Extremes The finding that wages as well as employment respond to open market shocks does not mean that those responses are a major element in industrial
253
Industrial Wage and Employment Determination in an Open Economy
wage and employment determination. Since the bulk of revenues are generated in domestic markets and most workers are native born, changes in trade and immigrant flows cannot possibly be a dominant force in altering the industrial wage structure or the composition of employment. Still, the wage responses to trade flows can have nonnegligible effects on wages and employment particularly at the extremes. A one standard deviation change in the import share (.43 by table 8.1), for example, induces a .028 change in wages and a .21 change in industry personhours, according to the coefficients in table 8.2. More strikingly, industries that faced massive changes in sales for either trade or domestic market reasons experienced large changes in wages as well as in employment. Figure 8.1 documents this by contrasting wage and employment changes from 1958 to 1984 between the ten industries experiencing the greatest positive and negative weighted changes in import shares, exports, and domestic market sales and changes in In total sales. The figure shows a wide spread in wage responsiveness between the extremes. For imports, the industries with the greatest increase in import share had wage increases some nineteen In points below the average for all industries and some thirty-three In points below wage increases for the industries with the largest decrease in import shares. Industries with the most/least rapid growth of domestic market sales show a smaller though still pronounced range of variation, while industries with the most/least rapid growth of exports show the least pronounced range. As for employment, the figure shows declines of .81 In points in industries with the most rapid rises in import shares, but it also shows above average declines in employment in industries where import shares fell-a seeming paradox that is due to the fact that import shares dropped most in industries with falling domestic market sales.27By contrast, the figure shows a monotonic relation between extreme changes in exports and in domestic market sales and employment: here, employment rises at rates far above average in the ten industries where exports or domestic sales increase most while falling at rates far below average in the ten industries where exports or domestic sales increase the least. Finally, putting all the components of change in sales together, we see sizable differences in changes in wages as well as in employment between industries experiencing the extremes of the change in sales. 8.3.3 Asymmetric Responses to Changes in Sales An important issue in decentralized labor markets where wages are flexible to industry conditions is whether wage responsiveness is symmetrical to declines in demand and increases in demand. In an economy in which wages are above the reservation wages of unemployed workers, greater responsiveness of wages to declines in demand than to increases in demand can increase employment.28To examine the symmetry of response, we have divided the change in the In sales variable into two parts: changes in excess of the mean change and changes below the mean change. We then regressed changes in In hourly wages over the twenty-six-year period 1958-84 on changes in sales interacted with a dummy for above average changes and a dummy for below
Richard B. Freeman and Lawrence F. Katz
254
-
.... ....' ..... .'*.*.... .*. ..... .... ..... .... ..... .... '..... :.:.:.' ....
-0.15 -
7
'
-'I3'
Low
High
Low Sales 'Domestic' Foreign Demand Demand
'
'
''
Import Share
Deviations from average LN change in production employment
1
0,4 -02 -0,8
Low
- 1,4
-2 Demand
Demand
Share
Fig. 8.1 Mean changes for the ten industries with the highest and lowest changes in each category relative to the overall mean change-1958-84
'
255
Industrial Wage and Employment Determination in an Open Economy
average changes. The coefficients on the two sale interaction terms (standard errors) given below show greater responses to the below-mean than to abovemean changes on changes in In hourly earnings: effect of below-mean changes: .080 (.017); effect of above-mean changes: .011 (.020). This finding suggests that relative wage flexibility in U.S. manufacturing has taken the form of “concessions” in industries doing more poorly to a greater extent than it has of large relative wage gains in industries doing better and thus may have contributed to the job growth in the country. We did not find any evidence that industries paid exceptionally large wage gains in booming markets: most of the adjustment is on the down side. 8.4
Conclusion
This paper has documented with two different data sets that, in detailed manufacturing industries in the United States, wages respond significantly to changes in industry sales, whether generated by domestic market or traderelated developments, and found that changes in immigrant shares are also related to industry wages. It has also found that the wage-setting institutions in the labor market condition wage responsiveness, with unionized high-wage industries showing the greatest response to changes in sales. While far from the dominant force in altering the industrial wage structure, shifts in product demand due to changes in sales from trade as well as domestic market developments have contributed to changes in earnings by industry. As the observed inverse relation between changes in the immigrant share of the work force and changes in industry wages cannot be readily explained by compositional factors, this finding evidently merits more detailed analysis of the market for immigrant labor.
Notes 1. Dickens et al. (1985) provide a detailed review of aggregate studies of the employment effect of trade in the United States. Lawrence (1984) provides a good example of a study utilizing an input-output framework to analyze the effect of trade on employment in U.S. manufacturing. 2. We define competitive decentralized wage setting to be a system in which wages are set to equate labor supply and labor demand, as in a textbook perfectly competitive labor market. We defer until later in this section a discussion of the possibility that efficiency wage considerations may yield above market-clearing wages even with decentralized wage setting in competitive labor markets. 3. Katz (1986) surveys efficiency wage and rent-sharing models of wage determination and discusses some of the implications of these models for interindustry wage differences. Dickens and Katz (1987), Katz and Summers (1989), Krueger and Sum-
256
Richard B. Freeman and Lawrence F. Katz
mers (1988), and Murphy and Topel (1987) provide evidence on the role of efficiency wage, rent-sharing, and standard competitive factors in explaining interindustry wage differentials. Lewis (1986) and Freeman and Medoff (1984) survey the vast literature on the effect of collective bargaining on relative wages. 4. Grossman assumes that the unionized sector faces an infinitely elastic product demand schedule at the exogenous international price. Labor demand elasticity depends only on the production technology and factor substitution possibilities. International competition is not viewed as changing product market structure and/or the elasticity of product demand. 5. A key assumption implicitly made in this model is that the membership is greater than the employment level that would prevail at the competitive wage rate w*. 6. In the case of Nash bargains with a utilitarian union and strongy efficient bargains, the wage is the mean of the average product of labor and the opportunity wage (McDonald and Solow 1981). In this case, a decrease in rents from increased international competition will lead to wage decreases in the union sector. 7. We can also substitute for dQ using (7) to obtain a relation between wages and output, dW = [ q / ( l + haq)]dQ. 8. While we have data on prices and quantities, the likelihood that price indices are inaccurate leads us to focus on sales. In empirical work, however, we also examine the separate effect of measured quantities and prices. 9. Consider the value of q when A is .I0 and n is .25. By eq. ( l l ) , we have q = .10/[1 - . l o . .25 . (1 - h) = .10/[1 - .025(1 - h)]. For the minimum value of h of zero, q differs from A by a bare 2.5%. For a high elasticity of, say, five, the difference is less than 10%. 10. To see this, we substitute dS - (1 - h)ndW for dX in eq. (3b) and then substitute (1 1) for dW.This yields the following: dE = {v/[l
+ q(1
- h)a]}dS,
where v is the desired parameter. 11. Rewriting eq. (5) and (6) as dP = udW + u and dW = qdx v, where u and v are error terms, one can easily derive a relation between wages and sales analogous to eq. (9):
+
dW = AdS
+ [V
-
q(1 - h)u]/[l
+ q(1 - h ) ~ ] .
12. Revenga (1989) uses an instrumental variables technique with the change in the (import-source-weighted) industry exchange rate used as an instrument for dS to reanalyze the data set that we examine in sec. 8.2. Her findings are qualitatively quite similar to those we report in sec. 8.2 and suggest that our least squares estimates may understate the response of industry wages to changes in industry sales. 13. This decomposition of change in log sales can be derived by writing sales as S = [(S - x)/DOM]DOM x, where x equals exports. The application of the difference operator to this decomposition of sales yields
+
dS = [S - x)/DOM]d(DOM) - DOM . [d(MSHR)] - d(D0M) . d(MSHR)
+ dx.
Equation ( I 1) in the text can then be derived from the above expression by dividing through by S to yield an expression in percentage changes, approximating percentage changes as In changes, and dropping the interaction term. This approximation is almost exact for annual changes. We have experimented with the exact decomposition using percentage changes and including the interaction term in several of our specifications and have found results in all cases quite similar to those obtained with our In change approximation.
257
Industrial Wage and Employment Determination in an Open Economy
14. We have also examined all workers and nonproduction workers. As production workers are the majority of workers, the results for all workers are quite similar to those for production workers. For nonproduction workers, there are some modest differences, but nothing substantial enough to change the tone of the findings. 15. Abowd (in this volume) provides detailed descriptions of the data set and its construction. 16. We have also estimated annual change equations replacing the time dummies with observed cyclical variables, such as the aggregate unemployment rate, and allowed different industries to have different cyclical sensitivities. The results are quite similar to the reported estimates based on equations with time dummies. 17. Branson and Love (1988) analyze the effect of time-series changes in the real exchange rate on employment in U.S. manufacturing industries. Eichengreen (1988) performs a similar time-series analysis of the effect of the real exchange rate on employment in four U.S. basic industries with the industry-specific wage treated as exogenous and included as an explanatory variable. Grossman (1987) takes changes in the price of import substitutes as exogenous and analyzes the effect of time-series changes in import prices on wages and employment in nine trade-affected U.S. manufacturing industries. The lack of import price data for a large number of industries at the level of aggregation of available wage and employment data prevents us from following a similar strategy. Grossman (1986) also uses a similar methodology to analyze the effect of international competition on employment in the steel industry but treats the wage as exogenous. 18. We have computed White (1980) heteroskedasticity-consistentstandard errors for several of our specifications. The White standard errors typically differ from the reported standard errors by less than 5%. 19. The immigrant share data for 1984 are based 1980 Census of Population data. We utilize 1960, 1970, and 1980 Census of Population data on immigrant shares of population in the ensuing empirical analysis. 20. The change in immigrant share from 1960 to 1980 has a correlation of - .25 with the initial (1958) industry In hourly wage level, and the initial (1960) immigrant share has a correlation of - .35 with the change in the In hourly wages over the period 1958-84. Immigrants have moved into low-wage industries, and initially immigrant intensive industries have experienced relatively low wage growth. 21. For example, the estimated effect of sales on annual earnings was ,045with a standard error of .011, while the effect of sales on employment was ,890 with a standard error of ,018. 22. We have also analyzed annual changes within the periods and found the same result: greater responses of hourly earnings to sales in the 1980s than in earlier periods. 23. These regressions utilized samples from the May 1974 CPS and the full-year 1984 CPS for workers in the fifty-eight three-digit CIC manufacturing industries with consistent industry classifications over this period. The controls included in the In earnings regressions in addition to industry dummies were education and education squared; experience and experience squared; nonwhite, female, SMSA, region, part time, marital status, married times female, and occupation dummies; and interactions of the education and experience variables with the female dummy. The regressions for all workers (both union and nonunion) included a union dummy. Industry differentials for all workers from regressions without a union dummy yield quite similar results. 24. The same pattern of larger wage responses to sales of the high-union density class is also apparent when long changes for 1958 to 1984 are analyzed rather than annual changes. 25. In separate annual change regressions by union category for 1959-84 of changes in In annual production worker hours on changes in In sales and our usual set
258
Richard B. Freeman and Lawrence F. Katz
of controls, the coefficients (standard errors) on change in In sales were .63 (.01) for high union density, .61 (.01) for medium union density, and .71 (.01) for low union density. 26. For a discussion of the potential pitfalls in interpreting estimated coefficients on union density in industry regressions, see Lewis (1986). 27. The industries with the lowest growth in import shares also experienced well above average wage growth, suggesting possible endgame behavior of the type modeled by Lawrence and Lawrence (1985). Still, fig. 8.1 does show that the industries with the biggest declines in domestic market size had below average wage increases, indicating that endgame is far from the norm for declining industries. 28. This assumes no difference in the size of sectors with increasing/decreasing demand.
References Borjas, George J. 1985. Assimilation, Changes in Cohort Quality, and the Earnings of Immigrants. Journal of Labor Economics 3 (October): 463-89. Branson, William H., and James P. Love. 1988. U.S. Manufacturing and the Real Exchange Rate. Misalignment of Exchange Rates: Effects on Trade and Industry, ed. Richard C. Marston. Chicago: University of Chicago Press. Chiswick, Barry R. 1978. The Effect of Americanization on the Eamings of Foreignborn Men. Journal of Political Economy 86 (October): 897-921. Dickens, William T., and Lawrence F. Katz. 1987 Inter-industry Wage Differences and Theories of Wage Determination. NBER Working Paper no. 2271. Cambridge, Mass.: National Bureau of Economic Research, June. Dickens, William T., Phillip Shapira, Laura Tyson, and John Zysman. 1985. The Employment Effects of International Trade: A Review of the Literature. University of California, Berkeley, February. Mimeo. Eichengreen, Barry. 1988. International Competition in the Products of U.S. Basic Industries. In The United States in the World Economy, ed. Martin Feldstein. Chicago: University of Chicago Press. Freeman, Richard B., and James L. Medoff. 1984. What Do Unions Do? New York: Basic. Grossman, Gene M. 1984. International Competition and the Unionized Sector. Canadian Journal of Economics 17 (August): 541-56. . 1986. Imports as a Cause of Injury: The Case of the U.S. Steel Industry. Journal of International Economics 20: 201-23. . 1987. The Employment and Wage Effects on Import Competition in the United States. Journal of International Economic Integration 2: 1-23. Huizinga, Harry. 1987. Union Wage Bargaining in the International Economy. Harvard University, April. Mimeo. Katz, Lawrence F. 1986. Efficiency Wage Theories: A Partial Evaluation. In NBER Macroeconomics Annual 1986, vol. 1, ed. S. Fischer, 235-75. Cambridge, Mass.: MIT Press. Katz, Lawrence F., and Lawrence H. Summers. 1989. Industry Rents: Evidence and Implications. Brookings Papers on Economic Activity: Microeconomics, 209-75. Krueger, Alan B., and Lawrence H. Summers. 1988. Efficiency Wages and the Interindustry Wage Structure. Econometrica 56 (March): 259-93. Lawrence, Colin, and Robert Z. Lawrence. 1985. Relative Wages in U.S. Manufactur-
259
Industrial Wage and Employment Determination in an Open Economy
ing: An Endgame Interpretation. Brookings Papers on Economic Activity, no. 1: 47106. Lawrence, Robert 2. 1984. Can American Compete? Washington, D.C.: Brookings. Lewis, H. G. 1986. Union Relative Wage Effects: A Survey. Chicago: University of Chicago Press. McDonald, Ian, and Robert Solow. 198 1. Wage Bargaining and Employment. American Economic Review 7 1 (December): 896-908. Murphy, Kevin M., and Robert Topel. 1987. Unemployment, Risk, and Earnings. In Unemployment and the Structure of Labor Markets, ed. K. Lang and J. Leonard. London: Blackwell. Revenga, Ana L. 1989. Wage Determination in an Open Economy: International Trade and U.S. Manufacturing Wages. Harvard University, December. Mimeo. Rose, Nancy L. 1987. Labor Rent-Sharing and Regulation: Evidence from the Trucking Industry. Journal of Political Economy 95 (December): 1146-78. White, Halbert. 1980. A Heteroskedasticity-consistentCovariance Matrix Estimator and a Direct Test for Heteroskedasticity. Econometrica 48 (July): 8 17-38.
This Page Intentionally Left Blank
9
Foreign-Owned Businesses in the United States Jonathan S. Leonard and Rachel McCulloch
For the United States, concern about foreign direct investment (FDI) has historically centered on the costs and benefits to the nation from the establishment of subsidiaries abroad by U.S. multinational firms. Since the mid-l970s, however, the United States has emerged as the world’s leading host to inward direct investment. Along with record purchases of U.S. securities, individual and institutional investors across the globe have purchased U.S. farmland, department stores, and luxury hotels. Foreign manufacturing firms have expanded U.S. distribution and service facilities and local production capacity. Japanese and European banks have opened branch offices in New York, Chicago, and Los Angeles. By 1984, the total value of foreign direct investments in the United States amounted to about 10 percent of the value of all New York Stock Exchange stocks. Moreover, these holdings had increased at a dramatic rate, from $35 billion in 1977 to $160 billion (nominal) in 1984. This flood of inward foreign direct investment represents a dramatic shift from the established pattern of the earlier postwar period. Until the 1970s, FDI globally was dominated by the outward thrusts of U.S. firms: the multinational corporations doing the investing were viewed by many writers as a peculiarly American phenomenon. But now foreign direct investment appears to be one more area in which the nation’s industrial competitors have been catching up to the United States. After years of relative stability, the ratio of inward to outward direct investments rose from less than one-quarter in 1977 to more than three-quarters by 1985 (Lipsey 1987). Jonathan S. Leonard is associate professor in the Organizational Behavior and Industrial Relations Group of the Haas School of Business Administration, University of California, Berkeley, a research associate of the Institute of Industrial Relations at Berkeley, and a faculty research fellow of the National Bureau of Economic Research. Rachel McCulloch is Rosen Family Professor of Economics and director of the Lemberg Program in International Economics and Finance at Brandeis University, and a research associate of the National Bureau of Economic Research.
261
262
Jonathan S. Leonard and Rachel McCulloch
As the foreign presence grows, U.S. policy concerns shift accordingly. Policymakers in the United States once focused primarily on the effects of direct investments abroad by U.S.-based corporations. The central issue in that policy debate was the relationship of outward investments to U.S. trade and domestic employment. Researchers assessed the degree of complementarity or substitution between U.S. exports and host-country production by U.S. subsidiaries and, to a lesser extent, between U.S. production for domestic markets and imports from subsidiaries abroad (e.g., Musgrave 1975; Dewald et al., 1978; Bergsten, Horst, and Moran 1978, chaps. 3,4). From a theoretical perspective, the key question was whether U.S. investments abroad were “defensive”-necessary to secure markets that would otherwise be lost to foreign rivals. Sympathetic observers like Vernon (1971) inferred from case studies that investments abroad by U.S. multinationals were largely defensive, but most American labor unions and some academic researchers (e.g., Frank and Freeman 1978) took a less optimistic view. Empirical testing was complicated by the product-cycle character of most U.S. direct investments abroad; the industries and firms with above-average propensities to invest abroad were also those with above-average propensities to export. Today, while the potential loss of U.S. jobs associated with establishment of foreign subsidiaries by U.S. firms remains an important policy concern, attention has moved to issues raised by inward direct investment-issues that are novel from the U.S. perspective but widely discussed during the postwar period in other major host countries, both industrialized and developing. The fundamental concern is the extent to which the investing firm, rather than the host country, derives the lion’s share of economic benefits from controlled local production. The distinguishing feature of foreign direct investment is an ownership stake sufficient to permit a local management role for the investor.’ In dollar terms, direct investment inflows from abroad remain small relative to the nation’s total foreign borrowings. However, this measure may understate the potential economic, political, and social impact. By definition, direct investments represent the extension of foreign firms’ managerial control into the U.S. economy, just as U.S. direct investments abroad have allowed American firms to control significant parts of the domestic economies of other nations.2 The recent pattern is controversial in the United States for much the same reasons that the former one has been controversial in host nations abroad. What accounts for the rapid increase in foreign direct investments in the United States? Is the flood of capital from abroad linked to U.S. trade problems? Should the United States welcome foreign investors, or are there reasons to limit the sale of U.S. assets? This paper considers the theoretical explanations for the growing foreign direct investment in the United States and explores some of the empirical regularities associated with foreign-owned companies. Section 9.1 discusses the
263
Foreign-Owned Businesses in the United States
many motivations underlying FDI in general and in the United States specifically. It provides a synthesis of theoretical explanations for the effects of these investments on the domestic labor market. Section 9.2 presents our data sources and methods. Section 9.3 analyzes the differences and similarities between U.S. and foreign-owned businesses in the United States. Section 9.4 offers some conclusions from the analysis.
9.1 The Motivations for Foreign Direct Investment 9.1.1 Foreign Investment and Internationalization Once largely insulated from developments abroad by its size and distance from other industrial powers, the United States has in recent years been drawn into increased economic intimacy with other nations. Foreign direct investments, first outward and more recently inward, have played a central role in establishing these linkages, bringing production and sales of enterprises in the United States and other nations under the control of a single management. The internationalization of the U.S. economy can be measured in a number of ways. The most obvious is that U.S. markets for goods and services are far more open than in the past. Almost every manufacturing industry has experienced a dramatic rise in the ratio of imports to domestic production-perhaps no surprise at a time when the nation has run trade deficits of record proportions. Less well known is that almost every U.S. manufacturing industry has also experienced a rise, albeit not as dramatic, in its exports. The same is true for agriculture, for the extractive industries, and for many of the service activities that now dominate U.S. employment. Even more striking than the increased flow of goods across U.S. borders is the growth in the volume and variety of asset transactions with other nation^.^ U.S. investors have long dominated international financial markets as lenders. In the 1980s, however, the United States became a major international borrower. Indeed, as a result of increased foreign borrowing and reduced foreign lending, the nation is today billed as the world’s leading debtor nation-an unfamiliar role, and one that many Americans find troubling. While the nation’s overall dependence on capital from abroad is itself womsome, the primary focus of concern is direct investment and the associated control by foreign enterprises over U. S. productive activity. The increased trade in financial assets reflects several independent developments. The United States and most other industrial nations have greatly reduced legal barriers to both inward and outward financial flows, part of a more general trend toward deregulation of financial markets. The recycling of large petrodollar surpluses in the 1970s contributed to the growth of institutions capable of handling huge international capital flows. Revolutionary changes in global communications have facilitated the integration of national financial markets into a single worldwide network of lenders and borrowers.
264
Jonathan S. Leonard and Rachel McCulloch
Together with expanded options for international communication, reduced costs of transporting goods and people have promoted a shift by many firms from national to global management of innovation, production, and distribution. Although the recent cries of alarm suggest otherwise, U.S. borrowings abroad are themselves nothing new. The United States was a borrower and net debtor for most of the period between the nation’s founding and World War I. What is new is the emergence of the United States as the major host country for inward foreign direct investments by multinational corporations based in other countries. From 1961 through 1967, less than 3 percent of the world’s flow of new FDI came to the United States. That fraction rose gradually during the 1970s and peaked in 1981, when the United States attracted nearly half of total direct investment inflows worldwide and two-thirds of total inflows going to developed countries (Lipsey 1987). Around 1980, the United States displaced Canada as the world’s leading host to foreign subsidiaries. The growth in U.S. trade and the growth in foreign direct investment have been linked developments. Transactions between multinational firms and their foreign subsidiaries have accounted for a major share of the overall rise in the volume of U.S. trade.4 Coordinated by a single global management, trade in intermediate inputs as well as completed goods allows international comparative advantage to operate not only at the level of individual products but also in determining the location of different steps in the production of a single product. As of 1985, about 15 percent of all foreign direct investments in the United States (and nearly two-thirds of all Japanese investments) were in the wholesale trade sector. The main task of such investments is to promote the parent’s imports to or exports from the United States. Even when local production by a manufacturing subsidiary replaces goods previously imported, as has occurred in autos and electronics, the resulting fall in imports of finished goods is typically offset by an associated increase in imports of component^.^ 9.1.2 Why Firms Go Abroad To understand why foreign firms have been increasing their ownership stakes in the U.S. economy, it is useful to review the basic conditions required for profitable foreign direct investment. These conditions provide a context for identifying specific changes in the global environment that may underlie the recent surge in establishment of U.S. subsidiaries by firms abroad. In one sense, the central puzzle concerning FDI is why it takes place at all, given the significant competitive disadvantages faced by a firm entering a foreign market. To make the strategy viable, the investing firm must possess an “advantage” in terms of product, process, or management sufficient to outweigh its obvious disadvantages relative to actual or potential domestic competitors in the host country. The existence of additional hurdles and risks fac-
265
Foreign-Owned Businesses in the United States
ing a foreign entrant suggests that foreign subsidiaries should on average yield higher profits than domestic operations in the same industry. These conditions imply that direct investments will be concentrated in markets that do not conform to the paradigm of perfect competition-markets characterized by incomplete information, barriers to entry, or other “imperfections.” A rent-generating competitive advantage is still only a necessary condition for the viability of foreign direct investment. Since the firm’s competitive advantage could in some circumstances be better exploited by exporting from the home country, an additional requirement for setting up foreign operations is a locational advantage. This could reflect the usual considerations of production and transport costs as well as advantages arising from national policy at home and abroad, for example, taxation, regulation, and barriers to trade. In the absence of a significant locational advantage, the potential investor is likely to choose exporting over the more costly and risky option of establishing a local subsidiary. Like the competitive advantage, the locational advantage is necessary but not sufficient. Even with a locational advantage, there must be an organizational advantage of FDI over alternative strategies such as licensing or other long-term contractual arrangements with firms in the host country. In the language of industrial organization, there must be an internalization advantagean advantage to substituting internal modes of resource coordination within a single firm for an external market-based arrangement between independent firms. In other words, there must be an advantage of integrated global management .’
9.1.3 Internalization and Internationalization Foreign direct investment is precisely a firm’s internalization of economic activity across a national boundary-internalization of management. The underlying motives are essentially the same ones that promote expansion of a firm’s activities within a single domestic market, but with a larger anticipated benefit required to offset the larger costs of international expansion. This perspective is supported by the empirical finding that firms investing abroad are on average larger in their domestic operations than other firms in the same industry. Multinational firms based in small countries are also typically smaller in absolute size than their counterparts from large countries. Both observations are consistent with the hypothesis that firms should exhaust the gains from domestic internalization before going abroad. Since the investments abroad of U.S. multinationals dominated the global picture for several decades after World War 11, most of the empirical research has focused on these. Studies of I 3 1 by U.S. manufacturing firms have verified that investment activity is clustered in the industries where research and development and advertising expenditures are important. Such expenditures presumably create the competitive advantage necessary for a U.S. firm to operate profitably in a foreign environment. Competitive advantages interact
266
Jonathan S. Leonard and Rachel McCulloch
with potential gains from internalization. For example, high-technology firms tend to exploit their newest technologies via subsidiaries, while older products and processes are licensed to independent foreign producers. Evidence on locational advantage is less compelling, except for resourcebased industries. Predictable factors such as availability of suitable labor at lower cost, a large and protected domestic market, favorable tax or regulatory treatment, and stable political environment all appear to have some influence on location decisions. Other locational considerations cannot be separated from the benefits of a single global management structure (internalization). Some FDI is primarily for the purpose of enhancing local sales of goods imported from a firm’s production facilities elsewhere, as with distribution and service facilities.* Local operations can also enhance exports by the parent to the host market by providing the parent with up-to-date market information when conditions are changing rapidly, thus keeping the parent in close touch with market trends. In a concentrated industry, the establishment of local production capacity may be central to the investing firm’s competitive strategy (e.g., Graham 1978). The subsidiary represents the firm’s precommitment to a substantial presence in the local market. Other advantages of multinational activity are associated with being multinational rather than with any specific host location. A global production network permits the firm to diversify risk and, more generally, increases its options when conditions are volatile (e.g., Kogut 1983). The risk-handling motive may be relevant in explaining not only investments in the post-BrettonWoods era of volatile exchange rates but also the classic foreign direct investments in extractive industries. Enhanced opportunities for tax avoidance are another widely cited benefit of multinational operations. The modest reported financial success (and correspondingly low tax burdens) of many foreignowned U.S. plants may reflect such accounting manipulation.
9.1.4
Exchange Rates and Direct Investment
Other things equal, a lower dollar makes U.S. products a better buy in world markets. Should the same hold for assets? If a U.S. asset is seen as a claim to a future stream of dollar-denominated profits, and if profits will be converted back into the domestic currency of the investor at the same exchange rate, the level of the exchange rate does not affect the present discounted value of the investment. Thus, dollar undervaluation (or overvaluation) is irrelevant unless a major motive for the investment is speculation on future movements in exchange rates. Speculative motives may influence portfolio investments but are unlikely to play an important role in direct investments, where the planning horizon usually extends over years or even decades. A more relevant consideration is that a weak dollar makes the United States more attractive as a production site. By lowering U.S. production costs rela-
267
Foreign-Owned Businesses in the United States
tive to those in Europe or Japan, a fall in the dollar might shift locational preference for direct investors toward the United States. Even so, advantages of internalization would be required to make direct investment a profitable response to the new currency values. In the absence of such advantages, foreign firms would be unable to compete with U.S. firms in exploiting the locational advantage of lower production costs. A more basic problem with attributing investment flows to exchange-rate levels is that the post-Bretton-Woods regime of generalized floating has been characterized by large swings in key rates, from apparent undervaluation to apparent overvaluation and back again.9 Thus, the motive for increased U.S. investments may lie less in the specific level of the exchange rate around the time of the investment than in the high probability of future large movements. Here internalization does play a key role-allowing increased costs in one location to be offset by reduced costs elsewhere and permitting some flexibility in shifting marginal production between locations on different sides of a major rate alignment. 9.1.5 The Role of Protection No other incentive for foreign direct investment has received as much attention as import barriers. It seems almost self-evident that tariffs or quotas will stimulate direct investments in the protected markets. Recent developments in the U.S. auto and electronics industries offer visible support for the proposition. Yet statistical analyses of Canadian and U.S. data have failed to confirm a systematic relationship between direct investment and protection. The likely reason for the weak empirical findings is that protection by itself confers only a locational advantage. Whether that locational advantage leads to inward investment or simply alters conditions of domestic entry and exit depends on other industry characteristics. In the absence of a firm-specific competitive advantage optimally exploited through internalization, domestic producers will be better able than subsidiaries of foreign companies to capture the benefits of local production. Important though they are in their own right, autos and electronics may be exceptions to the general rule. In these industries, technological know-how and managerial know-how are firm-specific advantages that allow foreign producers (notably Japanese) to compete effectively with established domestic firms.l0 By contrast, the highly protected U.S. apparel and footwear industries have seen almost no direct investments from abroad. For these lowtechnology industries, firm-specific advantages are apparently too small to offset the greater costs incurred by foreign investors. Evidence at the country rather than the industry level also casts doubt on the hypothesis that protection is a strong magnet for inward direct investments. Among the less-developed countries, open, export-oriented economies have been more successful in attracting new investments than nations pursuing import-substitution strategies. For U.S. outward investments, Canada, the
268
Jonathan S. Leonard and Rachel McCulloch
United Kingdom, and Germany, all with relatively liberal trade regimes, have been the most important host countries. Bhagwati (1985) suggests a more subtle link between protection and direct investment using the concept of “quid pro quo” foreign investment-investment made to defuse protectionist pressure rather than to circumvent actual current or anticipated protection. On this interpretation, Japanese investments in the automobile industry were intended, at least in part, to avoid future increases in protection (e.g., local-content requirements) rather than to circumvent existing import restrictions. Presumably, such investments would lessen the perceived need for protection and also would shift the domestic political balance toward a more liberal trade stance. An alternative “strategic” interpretation is suggested by the oligopolistic structure of the auto industry and the extensive publicity surrounding Japanese entry into U.S. production. Japanese firms may wish to demonstrate to their U.S. rivals that Washington cannot protect them from yielding part of their customary shares in the U.S. market. The new Japanese entrants could actually benefit from future increases in protection if their competitive advantage translates into lower costs in U.S. production while trade barriers prevent the Big Three from putting their own nameplates on captive imports from Japan or elsewhere. 9.1.6 Foreign Investment and U.S. Labor For host countries worldwide, the most important anticipated benefit from foreign investment is the creation of new jobs. In this, the United States has been no diEerent. Holding out the prospect of hundreds or even thousands of new jobs, U.S. state and local officials have mounted formidable campaigns to lure foreign plants, usually offering sizable financial incentives to supplement the region’s other attractions.’’ Yet, as with the presumed job losses associated with outward investments of U.S. multinationals, the overall effects of inward investments on domestic employment are far from clear. First, while there are obviously “new jobs” created by new subsidiaries, to some extent these new jobs will be offset by employment losses elsewhere. In the most optimistic scenario, local production will simply substitute for goods previously imported.’* At least for the industry, the effect on total employment should be positive. But local production by foreign affiliates can also cut into the market share of domestic competitors, so that the new jobs are matched by layoffs elsewhere in the same industry. If the affiliates use more imported inputs than their domestic counterparts, production and employment may be reduced accordingly in the supplier industries. A second concern is about the types of jobs created. Will foreign multinationals use U.S. labor for routine assembly operations, keeping the “good jobs” at home? Reich and Mankin (1986) interpret Japanese joint ventures in the United States as “part of a continuing, implicit Japanese strategy to keep the higher paying, higher value-added jobs in Japan and to gain the project
269
Foreign-Owned Businesses in the United States
engineering and production process skills that underlie competitive success.” A related concern shared with other host countries is that foreign affiliates allow little opportunity for local workers to rise into management ranks. While the Japanese presence in U.S. manufacturing is still too small and too new to offer much evidence on this issue, most analysts agree that U.S. operations abroad have benefited U.S. managers and skilled workers at the expense of less-skilled U.S. production workers, whose jobs have moved offshore. 9.1.7 U.S. Competitiveness and Inward Direct Investment The close link between FDI and the investing firm’s competitive advantage suggests that the rise in inward foreign investments in the United States as well as the slowing of U.S. direct investments abroad reflect the industrial catch-up of other nations to the United States. Where the competitive advantages were once controlled almost exclusively by U.S. companies, new rivals have emerged in Europe, Japan, and even some of the developing countries. As with the successful U.S. multinationals of earlier decades, these firms have exploited their competitive advantages first through exports and later through direct investment in the market countries. Like other host countries over the years, the United States is reevaluating the potential gains and losses from allowing free entry to foreign subsidiaries.
9.2 Data Sources and Methods Section 9.3 compares foreign-owned U.S. firms with their U.S.-owned counterparts. Most of the data used are derived from foreign direct investment series published by the Bureau of Economic Analysis of the U.S. Department of Commerce. Extensive cross-sectionaldata for 1980 are from Foreign Direct Investment in the United States, 1980, a survey of U.S. business enterprises in which foreign ownership, either direct or indirect, was at least 10 percent. A number of caveats apply to these data, particularly where comparisons are made to domestic aggregates. First, FDI data are reported in consolidated form for the U.S. affiliates. The activities of each establishment within multiestablishment enterprises are not classified separately by their own industry (as in the National Income and Product Accounts) but, rather, are classified by the industry group accounting for the largest percentage of the enterprise’s sales. This undercounts data for industries with many “owned” establishments and overcounts for industries with many “owner” establishments. It overstates cross-industry variance in sales. Second, FDI data aggregate petroleum-related activities including extraction, refining, and retailing. These have been removed from their respective industries and aggregated into a separate category that has been suppressed in the industry detail presented here. In consequence, these foreign-owned activities are undercounted in their respective subindustries.
270
Jonathan S. Leonard and Rachel McCulloch
Third, enterprises that are entirely foreign owned but in which no single foreign person owns at least 10 percent are not classified as foreign owned. For this purpose, “person” is defined to include any individual, partnership, associated group, or corporation, including members of a syndicate or joint venture. Fourth, compensation and employment data are collected only for U.S. affiliates whose assets, sales, or net income exceeded $1 million or whose land ownership exceeded two hundred acres. Fifth, data are annual averages for each enterprise’s fiscal year. Sixth, employment is reported as annual average number of employees, not as full-time equivalents. This will cause an understatement of compensation or wages per employee as reported here, but it should not affect the comparison of foreign direct with domestic because analogous concepts are used for domestic. Compensation per employee is also understated in some cases by the use of part-year compensation and year-end employment in some newly acquired enterprises and establishments. Finally, all the data discussed here are aggregates of enterprise data. The composition of the underlying sample changes over time, as the section on new acquisitions and establishments shows. Any change in, say, compensation per employee may then be due to (1) pay raises within previously sampled establishments, (2) deletion of low-wage establishments from the sample, (3) addition of high-wage establishments, or (4) purchase of a high-wage establishment by an enterprise with greater sales in another industry.
9.3 Buying American Foreign direct investment in the United States includes all firms in which 10 percent or more of the equity is foreign owned. The stock of FDI increased more than fourfold (in nominal terms) from $34.6 billion in 1977 to $159.6 billion in 1984. In 1977, the value of FDI in the United States was equal to 23.6 percent of the value of U.S. direct investment in foreign countries built up during earlier decades. By 1984, the reversal of net investment flows was well along. FDI in the United States was 68.4 percent of U.S. FDI abroad. Between 1977 and 1984, FDI in the United States more than doubled in proportion to the total value of all stocks listed on the New York Stock Exchange, the proportion rising from 4.3 percent to 10.1 percent. In part, this reflects growing foreign investments in all forms of U.S. assets. However, table 9.1 also shows that foreign direct investments increased in value relative to foreign-owned stocks and to U.S. investment abroad. Both developments indicate that potential foreign investors see greater competitive, locational, and organizational advantages to establishment of U.S. subsidiaries than in earlier periods. High interest rates affected many foreign investments in U.S. financial in-
271 Table 9.1
Foreign-Owned Businesses in the United States International Assets, 1977-84 1977
1. Private foreign investment in private U.S. assetsa 2. Row 1 as % of private U.S. fixed nonresidential gross capital 3. FDI in the US.” 4.Row3as%ofNYSEValue 5. Row 3 as % of foreignowned U.S. stocks 6. Row 3 as % of U.S.direct investment abroad
1978
1979
1980
1981
1982
1983
1984
157.9 189.8 242.1 308.7 380.1 477.0 559.2 630.5
5.3 34.6 4.3
5.6 42.5 5.2
6.2 54.5 5.7
6.9 7.7 9.0 10.1 10.9 83.0 108.7 124.7 137.1 159.6 6.7 9.5 9.6 8.7 10.1
86.9 101.0 112.8 128.5 168.3 162.4 140.9 166.4 23.6
26.1
29.0
38.5
47.6
56.2
60.4
68.4
Sources: Survey of Current Business, various issues; Economic Report of the President, various issues; Statistical Abstract of the United States. a Billion current dollars.
Table 9.2
Flows of Foreign Investment in the United States, 1960-86 (billions of U.S. dollars)
Year
Total Inflow
1960 1970 1972 1974 1976 1978 1980 1981 1982 1983 1984 1985 1986
2.3 6.4 21.5 22.5 42.7 65.4 84.7 78.2 109.7 96.9 108.2 127.1 213.3
Total FDI
FDI as % of Total
Japanese FDI
% of FDI from Japan
.3 1.5 .9 4.8 4.3 7.2 13.7 22.0 10.4 11.9 25.4 17.9 25.6
13.7 23.0 4.4 21 .o 10.2 12.1 16.1 28.1 9.5 12.3 23.4 14.0 12.0
NA
NA 3.6 2.0 4.4 13.5 12.5 5.3 12.6 16.8 13.8 17.2 17.3 18.5
~
.o .o
.2 .6
1.a .7 2.8 I .7 1.7 4.4 3. I 4.7
Sources: Survey of Current Business, various issues (for 1972-86); Business Statistics, 1984 (for 1960 and 1970). Note: Percentages calculated from unrounded flow data. Data for 1986 are preliminary.
struments. The composition effect created by the increase in foreign ownership of U.S. financial assets overshadows a less noticed shift in foreign investment toward direct corporate ownership. Since 1977, foreigners have increasingly been purchasing control of U.S. corporations. The value of inward FDI, as a percentage of foreign ownership of U.S. stocks, rose from 87 percent in 1977 to 166 percent in 1984. Table 9.2 shows the trends in the flow of new FDI. The total inflow of foreign investments increased in the 1970s and 1980s. Although FDI as a
272
Jonathan S. Leonard and Rachel McCulloch
percentage of the total inflow of investment does not show a distinct trend over this longer period, Japanese FDI has increased substantially. The flow statistics offer less support for the hypothesis of changing competitive, locational, or organizational advantages to establishment of U.S. subsidiaries by foreign-based multinational corporations. The absence of a trend in FDI as a percentage of total investment inflow indicates that the growth in FDI may be simply a manifestation of the growth in all forms of foreign investment in the United States. The trend in Japanese FDI in the United States suggests that economic advantage arguments may, however, apply to Japan and to the industries in which Japanese companies are highly visible. 9.3.1 Acquisitions and New Establishments Newly acquired or established enterprises show one form of increased investment by foreigners in the United States (see table 9.3). The Bureau of Economic Analysis (BEA) classifies as a new acquisition an existing U.S. enterprise in which foreign ownership (directly or through U.S. affiliates) passes 10 percent. However, this is only a small part of total investment, because additional equity investments in existing U.S. affiliates are not counted once the 10 percent threshold has been passed, and because only enterprises with assets exceeding $1 million or two hundred acres of U.S. land are included. To illustrate, of the 2.1 million U.S. employees of foreign-owned companies in 1980, 13 percent were in newly acquired enterprises, and .6 percent were in newly established enterprises. Compared to analogous rates for total domestic industry, the acquisition rate is high and the start-up rate low. Of the $522 million of foreign-owned 1980 assets, 8 percent were newly acquired and 1.4 percent newly established. Roughly 80 percent of these investment funds came through existing U.S. affiliates. In 1980, 37 percent of these investments were financed by U.S.-source funds. Only 2 percent of Table 9.3
Outlays and Employment in U.S. Enterprises Newly Acquired or Established by Foreign Direct Investors, 1979-85 Outlays ($ million current)
1979 1980 1981 1982 1983 1984 1985
15,317 12,172 23,219 10,817 8,091 15,197 19,547
Employment New Establishments 15,467 13,022 14,072 8,169 5,556 4,139 7,772’
Employment New Acquisitions 314,548 279,459 428,745 225,673 102,557 168,406 235,667’
Sources: Outlays: Shea (1986, 47, table 1). Employment: same 1979, 1980, 1981, 1982, 1983, 1984, and 1985. preliminary
273
Foreign-Owned Businesses in the United States
these investments (1982) were reported to receive specific state or local investment incentives or subsidies. Measured either by employment or assets added in new acquisitions or establishments, inward FDI has been volatile and shows no clear trend. This is misleading. Indeed, both assets and employment in foreign direct investments have been growing steadily in the 1980s. The difference arises because most of the growth has occurred in ongoing foreign-owned businesses. This is similar to the growth process of domestic industry generally, which is also dominated by the expansion of ongoing concerns. 9.3.2 Foreign Ownership by Industry The nature of the industries in which foreigners invest does differ substantially from domestic industry as a whole. Table 9.4 shows the industrial distribution of employment in foreign-owned businesses. It bears greater resemblance to U.S. direct investment in other developed economies. Overall, foreigners invest predominantly in U.S. manufacturing industries. While 22.1 percent of 1980 U.S. employment was in manufacturing, this sector accounted for fully 54.3 percent of FDI. The service and retail trade sectors show the fastest growth in FDI, but manufacturing still dominates. Employment in foreign-owned manufacturing doubled between 1977 and 1984. Measured as a percentage of total industrial employment, foreign ownership has advanced farthest in chemicals, where 39 percent of all employment is in foreign-owned establishments; stone, clay, and glass (1 1 percent); primary metals (1 1 percent); food (9 percent); and electrical machinery (8 percent).I3The chemical industry stands out as a case in which foreign ownership is approaching a majority of the industry. These are all manufacturing industries in which the foreign parent may have a competitive advantage due to the importance of technology in determining business success. Of the sectors Table 9.4
Industrial Distribution of Employment in Foreign-owned Business, 1977-84. (%) 1977
1978
1979
1980
1981
1982
1983
1984
Mining Manufacturing Wholesaletrade Retail trade Construction Services Residual
1.31 56.28 12.55 11.65 1.07 3.04 14.11
1.12 56.22 12.03 11.96 1.61 3.57 13.50
1.03 57.39 11.18 13.46 1.60 3.76 11.58
1.23 54.33 10.67 14.95 2.11 4.18 12.54
1.65 53.79 10.51 14.23 2.40 5.13 12.29
1.67 50.74 11.44 16.26 2.12 5.43 12.34
1.46 51.54 10.61 16.51 1.98 5.34 12.55
1.18 50.76 10.79 16.72 1.55 7.07 11.93
Total (thousands)
1,219
1,430
1,753
2,034
2,417
2,448
2,526
2,715
Sources: U.S. Department of Commerce (1983) and Shea (1986). Note: Foreign employment in industry i over total foreign employment (in %), 1977-84.
274
Jonathan S. Leonard and Rachel McCulloch
with highest foreign ownership, two (electronics and chemicals) are commonly considered to embody advanced and rapidly progressing technology. In most other industries, the share of employment in foreign-owned businesses remains under 5 percent. Foreign employment as a share of total employment is notably low in communications and public utilities (.4 percent), services ( . 8 percent), agriculture (.5 percent), and construction (.9 percent). Regulation limits access to the first of these markets. The others are all nonmanufacturing industries with low domestic sales concentration. Foreign ownership rates are highest within manufacturing. Foreign ownership is increasing in almost every industry, including stagnant industries such as primary metals. In recent years, it has increased fastest in such home-goods industries as service, real estate, and retail trade.
9.3.3 Location Decisions As new entrants to the U.S. employment market, foreign direct investors have at times been characterized as locating in the low-wage South, the growing West, or the technologically advanced Northeast. From the perspective of any of these stories, the surprising fact is just how closely the geographic distribution of FDI employment parallels that of all domestic firms. Table 9.5 shows, for each of the nine major Census geographic divisions, the shares of foreign direct and of domestic employment for all sectors and for manufacturing. The largest difference between FDI and domestic location occurs in the Middle Atlantic states (New York, Pennsylvania, and New Jersey), which account for 21 percent of FDI employment but just 17 percent of doTable 9.5
Geographic Distribution of Employment, 1980 (in thousands) Fraction of Total
Geographic Region New England Middle Atlantic South Atlantic East North Central East South Central West North Central West South Central Mountain Pacific Total
Manufacturing
Fraction of Total Manufacturing
Foreign Domestic Foreign Domestic Foreign Domestic Foreign Domestic 122.9 5,474.5 414.5 15,011.6 363.1 14,625.2
.06 .21 .I8
.06 .17 .16
72.2 217.2 192.1
1,524.6 3,554.2 3,041.5
.07 .20 .I8
.07 .I7 .I5
368.1
16,826.8
.18
.I9
226.4
4,687.6
.21
.23
99.5
5,145.5
.05
.06
61.5
1,362.7
.06
.07
103.2
6,903.0
.05
.08
63.2
1,381.3
.06
.07
209.8 64.0 274.8
9,313.3 4,488.1 13,058.8
.10 .03 .14
.10 .05 .14
93.2 29.1 142.7
1,669.5 563.2 2,569.0
.08 .03 .13
.08 .03 .13
2,019.9 90,846.8
1.00
1.00
1,097.7 20,353.6
1.00
1.00
Sources: U.S. Department of Commerce (1983, table F-7); U.S. Bureau of Labor Statistics.
275
Foreign-Owned Businesses in the United States
mestic. FDI employment is also more prevalent in the South Atlantic states and less prevalent in the relatively depressed East North Central region. Aside from these differences, the location decisions of foreign direct investors in the United States look much like those of domestic employers. As we noted above, much FDI involves the creation of wholesale, retail, and service establishments to support the international trade of the parent. Apparently, the geographic distribution of population in the United States is more important than regional labor market differences in determining the location of these establishments. 9.3.4 Compensation Differences U.S. production workers do not appear to suffer under foreign ownership. In the manufacturing sector, the ratio of compensation per worker in foreignowned to the same measure for employees of U.S .-owned businesses increased from .94 in 1977 to 1.08 in 1984 (table 9.6). Foreign ownership has little effect on the mix of fringes to wages, so wages show a similar pattern. While compensation per worker is 10 percent lower in the foreign-owned establishments of the food, primary metals, and instruments industries, in general foreign gains in domestic industries have not been accompanied by relatively low-wage labor. Overall, workers in foreign-owned businesses enjoy a 20-30 percent advantage in compensation over employees of U.S.-owned businesses. The compensation differential can be decomposed into a within-industry differential and a composition effect. Only about a third of the overall difference is due to higher compensation in foreign direct employment within industry. For the most part, the higher compensation found in the aggregate in foreign direct employment is explained by the greater concentration of FDI in the highwage manufacturing sector. For manufacturing, the BEA provides separate data on wages, hours, and occupational structure. The hourly wages of production workers in foreignowned enterprises are 8 percent greater than the domestic average in 1980, although the hours worked are 8 percent less. For production workers, increased hourly pay is balanced by shorter hours. Foreign-owned firms in manufacturing appear more top heavy, employing 64 percent production workers in 1980 compared to the domestic average of 70 percent (see table 9.7).14 Since the overall 1980 compensation ratio is less than one, non-production workers in foreign-owned firms appear to be paid less than their counterparts in U. S .-owned firms. As table 9.6 shows, foreigners appear to invest in high-wage industries. In our discussion of the sources of advantage to FDI, we noted that locational advantages may be conferred by protective trade policies, as in the automotive industry, for example. However, successful direct investment requires that the foreign investor also possess competitive advantage. Otherwise, only domestic entry and pricing decisions are altered by the protection. When some trade
276 Table 9.6
Jonathan S. Leonard and Rachel McCulloch Compensation Ratios (foreign compensation per worker over domestic compensation per worker)
Industry
1977
1978
1979
1980
1981
1982
1983
1984
All industries All nonpetroleum industries Mining Manufacturing Durable Transportation & equipment Primary metal industries Fabricated metal products Machinery, except electrical Electric & electronic equipment Nondurable Textile products & apparel Lumber & furniture Paper & allied products Printing & publishing Rubber & plastics products Stone, clay, & glass products Food & kindred products Wholesale trade Retail trade Construction Services Finance Real estate Insurance Communication Transportation
1.20 1.17 .94 .94
1.23 1.20 1.02 .96
1.20 1.18 .98 .94
1.20 1.19 1.02 .95
1.27 1.42 1.09 1.06
1.30 1.28 1.16 1.11
1.29 1.28 1.16 1.08
1.26 1.24 1.14 1.08
.81 .83 .96 .96
1.07 .86 .98 .99
.93 .80 .98 .98
.97 .86 .90 .97
.97 .86 .98 .99
.98 .90 1.09 1.10
.90 .95 .96 1.05
.96 .99 1.12 1.01
.83
.86
.87
.90
.99
1.OO
.96
.98
1.21 1.01 .99 1.02 .82 .90 .89 1.03 1.22 .80 1.10 1.33 1.02 1.00 1.31 1.05
1.13 .93 1.00 1.03 .83 .97 .91 1.03 1.21 1.10 1.17 1.67 .96 1.02 .86 1.07
1.07 .81 .98 1.07 .87 .97 .78 1.01 1.29 1.16 1.11 1.68 1.40 1.01 1.13 1.09
1.07 .89 .96 1.07 .95 1.08 .83 1.02 1.24 1.06 1.04 1.54 1.38 .99 .74 .90
1.14 .84 1.22 1.10 .91 1.11 .89 1.01 1.18 1.16 .94 1.54 1.34 .94 .76 .88
1.21 .83 1.24 1.11 .94 1.21 .90 1.11 1.21 1.23 .97 1.91 1.19 .93 .63 .93
1.17 1.03 1.30 1.14 .92 1.17 .87 1.19 1.13 1.26 1.03 1.87 1.12 .91 .55 1.14
1.21 1.01 1.18 1.14 1.06 1.12 .91 1.13 1.05 1.14 1.05 1.71 1.12 .91 .57 1.07
Sources: U.S. Department of Commerce (1983); Survey of Current Business; National Income and Product Accounts.
flows are restricted through commercial policy, factor flows are magnified. A partial explanation of the tendency of foreigners to invest in high-wage U.S. industries (and to pay the U.S. wage in those industries) is the export of capital to the United States as an alternative to export of the products of their domestic high-wage industries.
9.3.5 Research and Development Just as U.S. direct investment in other countries is dominated by industries with substantial research and development (R&D), foreigners investing in U.S. domestic industries use R&D to generate a competitive advantage. While R&D scientists and engineers constitute only .5 percent of employees of domestic-owned firms, they are 2.1 percent of the employees in foreignowned companies (see table 9.8). By this measure, R&D intensity is half
277
Table 9.7
Foreign-Owned Businesses in the United States
Proportion of Production Workers in Manufacturing, 1980 Ownership Foreign Manufacturing Food & kindred products Chemicals & allied products Primary & fabricated metals Primary metals Fabricated metals Machinery Machinery, except electrical Electric & electronic Textile products & apparel Lumber & furniture Paper & allied products Printing & publishing Rubber & plastics Stone, clay, & glass Transportation equipment Instruments & related products
.64 .68 .48 .72 .71 .74 .61 .57 .63 .81 .79 .74 .57 .75 .75 .66
Domestic .70 .69 .57 .75 .77 .74 .64 .64
.64
.64
.86 .82 .75 .56 .77 .71 .65
.60
Sources: U.S.Department of Commerce (1983); Statisrical Abstracr of the Vnired Srares.
Table 9.8
Research and Development Intensity, 1980 9% of Total Employees Who are R&D Scientists and Engineers
All industries Manufacturing Food & kindred products Chemicals & allied products Primary metal industries Fabricated metal products Machinery Electric & electronic equipment Other manufacturing Textile products & apparel Paper & allied products Stone, clay, & glass products Transportation & equipment Instruments & related products
Foreign
Domestic
2.13
.46
SO
.42 4.48 .78 .61 3.03 3.91
6.14 .72 2.40 3.62 3.78 .39 .43 .94 2.34 4.10
Sources: U.S. Department of Commerce (1983); National Science Foundation.
.09 1.09 .81
1.88 3.86
278
Jonathan S. Leonard and Rachel McCulloch
again as high in foreign-owned compared to domestic-owned manufacturing businesses (3.1 vs. 2.0 percent) and more than ten times greater in nonmanufacturing (.9 vs. .07 percent). One need not be xenophobic to wonder about the fate of U.S. comparative advantage in R&D-intensive industries when the declining share of U.S. citizens in U.S. graduate science and engineering education is coupled with the declining domestic-ownership share in U.S. R&Dintensive industries. 9.3.6 Collective Bargaining Among industrialized countries, the United States now has one of the lowest union representation rates. Among the many explanations proffered, some have pointed to differences in management attitude. Managers from some European countries and Japan are often surprised at the unquestioned and vehement antiunion animus of their U.S. counterparts. Higher unionization rates in home countries may be associated with greater management tolerance of unionization and perhaps with greater skill in developing cooperative arrangements with unions. If foreign owners really take a less antagonistic position toward unions, one might expect this to carry over to their U.S. operations and reveal itself in higher unionization rates than in U. S .-owned domestic operations. On the other hand, once geographically removed from home country approbation and leverage, the same cost considerations that drive U.S. companies may dominate. To the extent that cost disadvantages of union firms can be capitalized in the sales price of corporate assets, no difference in unionization is expected on the basis of foreign ownership. In 1980, 23 percent of U.S. employees were union members. Among foreign-owned companies, 29 percent of employees were covered by collecTable 9.9
Union Density at Home and Abroad, 1980 % Covered by Collective Bargaining in Foreign Owned
All Industry United States All foreign owned Canada United Kingdom Japan Netherlands Sweden France Germany Switzerland
29.21 32.05 26.07 20.26 24.52 31.90 47.49 30.96 17.35
Manufacturing
35.9 31.13 35.39 29.11 28.82 34.03 55.05 20.06 22.55
% Unionized Home Country
23.1 30.50 53.10 30.80 37.10 87.80 19.20 38.60 33.50
Sources: U.S. Department of Commerce (1983); Kokkelenberg and Sockell (1985); Troy and Sheflin (1985).
279
Foreign-Owned Businesses in the United States
tive bargaining contracts (see table 9.9). Again, this difference is almost entirely due to the greater concentration of FDI in the manufacturing sector. Within manufacturing, foreigners leave any pro-union sentiments at home. Whereas 36 percent of U.S. employees are union members, only 31 percent of the employees in foreign-owned companies are covered by collective bargaining contracts (a more inclusive measure). Part of the difference in manufacturing union density may reflect union avoidance on the part of foreigners. Part may reflect compositional differences within manufacturing, and part may be due to a vintage effect. Newer firms and industries are less unionized, and foreign ownership is presumably concentrated among these. In any case, owner attitudes inferred from homecountry unionization can be dismissed as an important factor. The growing internationalization of the world economy has so far presented greater competition attacking local rents and greater opportunities for union avoidance than for the international application of union leverage to enforce union standards.
9.3.7 Home-Country Effects The characteristics of FDI in the United States may differ systematically by home country of the investor, although in general one may suspect that such differences either are transient or represent industry-specific or firm-specific effects. For the countries that are home to most of the ultimate beneficial owners of FDI in the United States, table 9.10 compares a number of characteristics. Of these countries, investments owned by Australians and Dutch appear the most successful. They show the highest return on assets and the highest pay. The Dutch investments are also R&D intensive. Countries such as Germany and Switzerland, with the greatest share of their investment in manufacturing, show the worst rates of return. Japanese investments stand out only in their avoidance of manufacturing, compared to the investments of other foreigners. In general, the measures in table 9.10 appear to tell more about the common characteristics of FDI in the United States than about differences systematically related to home-country factors. 9.4
Conclusions
The close link between foreign direct investment and the investing firm’s competitive advantage suggests that both the rise in inward direct investment in the United States and the slowing of U.S. direct investment abroad reflect the industrial catch-up of other nations to the United States. Where the competitive advantages that underlie successful foreign investment were once controlled almost exclusively by U.S. companies, new rivals have emerged in Europe and Japan, and even in some of the developing countries. As with the U. S .-based multinationals that dominated global direct investment flows in earlier decades, these foreign-based firms have exploited their competitive
Table 9.10
Characteristics of Foreign Direct Investment by Country of Origin, 1980
1. Net income/assets 2. Wagesiworkers* 3. Compensation/workers* 4. R&D employedlworkers 5. Manufacturing employed/workers 6. Wages/workers** 7. Compensation/workers**
All Foreign Countries ,031 16.28 19.69 .021 .54 16.92 20.67
Canada
United Kingdom
Japan
.022 16.92 20.68 .011 .52 17.19 21.90
,056 14.59 17.80 ,017 .52 15.13 18.50
,026 16.26 18.57 ,015 .31 16.35 19.25
Sources: U.S. Department of Commerce (1983, tables B-8, E-2, F-2, F-13, F-14). Nore: U.S. column gives comparable domestic averages. *In thousands of dollars. **In manufacturing.
Netherlands ,049 18.84 23.00 ,040
.55 17.60 21.13
France
Germany
,011 18.58 22.76 .016 .58 20.33 25.64
15.40 18.73 .024 .64 15.76 19.38
,006
Switzerland
Australia, New Zealand, South Africa
United States
.008 15.99 19.04 ,038 .65 17.88 21.50
,097 18.17 21.35 ,019 .29 17.97 21.12
.059 13.91 16.39 ,005 .20 17.36 21.68
281
Foreign-Owned Businesses in the United States
advantages first through exports and later through direct investments in market countries. As with U.S. companies investing abroad, foreign firms establishing subsidiaries in the United States often rely on superior technology for the cornpetitive advantage necessary to make their investments profitable. Although these firms come to the United States to exploit an already-established cornpetitive advantage, their U.S. operations employ a larger proportion of scientists and engineers than U.S.-owned businesses in the same industry; foreignowned enterprises in the United States are on average more R&D intensive than their domestic counterparts. However, R&D intensity is the only large difference between foreignowned and US.-owned businesses that emerged from our statistical comparison. Indeed, it is striking how similar foreign-owned and U.S.-owned businesses appear statistically. Although foreign-owned companies have a different industrial mix favoring manufacturing, retail and service establishments are growing fastest. Wages and compensation of foreign-owned businesses are very similar to those of U.S.-owned businesses in the same industries.
Notes 1. In U.S. statistics, the line is drawn at 10 percent of equity, although most U.S. direct investments abroad and many foreign direct investments in the United States are wholly owned subsidiaries of the parent. Other asset purchases, e.g., of private or government bonds or smaller blocks of stocks, are termed portfolio investments. 2. The size of the ownership stake need not indicate the total size of the controlled activity. A fall in measured direct investment could, in principle, be accompanied by an increase in the extent of controlled activity. Alternative measures of foreign influence include sales, employment, and profits of the controlled enterprise. 3. Although appropriately classed as asset transactions, direct investments are distinctive in that no international transfer of financial capital, i.e., purchasing power, need be entailed. In many cases, the contribution of the investing firm to a joint venture consists primarily of proprietary technology or managerial expertise rather than financial capital. Even when financial capital is part of the investment “package,” the required funds may be borrowed locally in the host country. This was a common practice in the 1960s for the U.S. firms establishing European subsidiaries. 4. In 1983, US.multinational corporations accounted for more than three-quarters of U.S. exports and almost half of U.S. imports. However, these shares have been declining from their peaks in the 1970s (the comparable percentages for 1977 were 84 and 58), while the U.S. trade role of foreign multinational firms appears to have grown over the same period. (See Barker 1986.) 5. This has been alleged in recent years about Japanese investments in the United States but is also a longstanding complaint of less-developed host countries. These nations invite foreign investments with the hope of reducing chronic balance-ofpayments difficulties. The usual experience is that induced imports of machinery and components tend to offset any direct reduction in imports or rise in exports of the
282
Jonathan S. Leonard and Rachel McCulloch
product itself. This is, of course, consistent with the notion that balance-of-payments difficulties are fundamentally macroeconomic problems requiring macroeconomic solutions. On the macroeconomic roots of the U.S. trade deficit, see McCulloch and Richardson (1986). 6. The idea that FDI requires a significant departure from conditions of perfect competition was advanced by Hymer (1960) and expanded by Kindleberger (1969), Caves, and many others. For a comprehensive survey of the literature, see Caves (1982, chap. 2). 7. The three “necessary conditions” are elaborated by Dunning (198 1, and earlier papers). Dunning uses this classification to explain the distribution of investment by home and host country and industry. For a detailed analysis of direct investment as international internalization, see Rugman (1980). 8. The U.S. investments by Japanese firms are a case in point. While this kind of complementary relationship between exporting and direct investment was suggested by Bergsten, Horst, and Moran (1978) for manufacturing investments, it may be even more important in the case of service industries such as banking and insurance. 9. As long as rates are determined mainly by market forces, in one sense under- or overvaluation cannot occur. These descriptions usually refer to a deviation of marketdetermined rates from rates calculated using relative price levels (purchasing power parity). 10. As discussed below, local policies to attract new investment may also play a role. 11. These benefits include tax breaks, cheap loans, worker training, and free infrastructure. According to a Kentucky legislative study, the state will spend $125 million, or about $42 perjob, to attract Toyota’s new plant (Lore 1987). 12. In early 1987, the president of Ford Motor Co. called for further reductions in auto imports from Japan to compensate for increased production by Japanese plants in the United States. 13. Statistics are for 1984. Sources are U.S. Department of Commerce, Foreign Direct Investment in the United States, and the Survey of Current Business. 14. This difference may be understated by the use of full-time equivalent counting only for the domestic figure.
References Barker, Betty L. 1986. U.S. Merchandise Trade Associated with U.S. Multinational Companies. Survey of Current Business 66 (May):55-72. Bergsten, C. Fred, Thomas Horst, and Theodore H. Moran. 1978. American Multinationals and American Interests. Washington, D.C.: Brookings. Bhagwati, Jagdish. 1985. Investing Abroad. Esmee Fairbairn Lecture, University of Lancaster, 27 November. Caves, Richard E. 1982. Multinational Enterprise and Economic Analysis. Cambridge: Cambridge University Press. Dewald, William, Harry Gilman, Harry Grubert, and Larry Wipf, eds. 1978. The Impact of International Trade and Investment on Employment. Washington, D.C. : U.S. Department of Labor. Dunning, John H. 1981. Explaining the International Direct Investment Position of Countries: Towards a Dynamic or Developmental Approach. Weltwirtschaftliches Archiv 117:30-64.
283
Foreign-Owned Businesses in the United States
Frank, Robert H., and Richard T. Freeman. 1978. The Distributional Consequences of Direct Foreign Investment. In The Impact of International Trade and Investment on Employment, ed. William Dewald, Hany Gilman, Harry Grubert, and Larry Wipf. Washington, D.C.: U.S. Department of Labor. Graham, Edward M. 1978. Transatlantic Investment by Multinational Firms: A Rivalistic Phenomenon? Journal of Post Keynesian Economics 1(Fall):82-99. Hymer, Stephen. 1960. The International Operations of Firms. Ph.D. diss., Massachusetts Institute of Technology. Kindleberger, Charles P. 1969. American Business Abroad: Six Lectures on Direct Investment. New Haven, Conn.: Yale University Press. Kogut, Bruce. 1983. Foreign Direct Investment as a Sequential Process. In The Multinational Corporation in the 1980’s ed. C. P. Kindleberger and D. B. Audretsch, Cambridge, Mass.: MIT Press. Kokkelenberg, E. C., and D. R., Sockell. 1985. Union Membership in the United States, 1973-1981. Industrial and Labor Relations Review 38:497-543. Lipsey, Robert E. 1987. Changing Patterns of International Investment in and by the United States. In The United States in the World Economy, ed. Martin Feldstein, 475-545. Chicago: University of Chicago Press. Lore, Dave. 1987. Japanese Auto Investments: Great Today but What about Tomorrow? MidAmerican Outlook 1O(Spring):2-4. McCulloch, Rachel, and J. David Richardson. 1986. U.S. Trade and the Dollar: Evaluating Current Policy Options. In Current US.Trade Policy: Analysis, Agenda and Administration, ed. R. E. Baldwin and J. D. Richardson. Cambridge, Mass.: National Bureau of Economic Research. Musgrave, Peggy B . 1975. Direct Foreign Investment Abroad and the Multinationals: Effects on the US. Economy. Washington, D.C.: Senate Foreign Relations Committee. Reich, Robert B., and Eric D. Mankin. 1986. Joint Ventures with Japan Give Away Our Future. Harvard Business Review 64(March/April):78-86. Rugman, Alan M. 1980. Internationalization as a General Theory of Foreign Direct Investment: A Re-appraisal of the Literature. Weltwirtschaftliches Archiv 116:36579. Shea, Michael A. 1986. U.S. Business Enterprises Acquired or Established by Foreign Direct Investors in 1985. Survey of Current Business 66(5):47-53. Troy, L., and N. Sheflin. 1985. Union Sourcebook. West Orange, N.J.: IRDIS. U.S. Department of Commerce. 1983. Foreign Direct Investment in the United States, 1980. Washington, D.C.: U.S. Government Printing Office. Vernon, Raymond. 1971. Sovereignty at Bay. New York: Basic.
This Page Intentionally Left Blank
10
Immigration, International Trade, and the Wages of Native Workers Peter Kuhn and Ian Wooton
The purpose of this paper is to develop and apply to U.S. data a theoretical model with the following features. First, it should yield a set of predictions regarding the effects of international factor movements, such as immigration, on the rewards of all factors employed in the country, including labor disaggregated by skill level. Second, it should be consistent with the following stylized facts: (i) the U.S. economy is “partially open” in the sense that it produces both internationally traded and nontraded goods; and (ii) international trade in goods has apparently not equalized factor prices between the United States and the rest of the world. These requirements play an important role in this paper because few of the existing models that consider general-equilibrium effects of factor endowments on factor prices satisfy them. For example, Hicks’s (1932, chap. 6) classic analysis assumes that no goods are traded internationally. This work predicted that the effect of factor quantities on factor prices was determined by a set of within-industry elasticities of substitution as well as substitution elasticities in consumption, and it stimulated several empirical attempts to estimate these parameters (e.g., Fallon and Layard 1975). On the other hand, the basic two-good, two-factor (2 X 2) trade model (Samuelson 1948) assumes that all produced goods are traded. It predicts, unrealistically, that trade alone should eliminate all factor price differentials between countries and thus that factor endowments should have no effect on factor prices. Finally, among the trade models that do allow for international factor movements to affect factor prices (e.g., the 2 X 2 models of Kemp 1966; Markusen and Melvin 1979; Brecher and Choudhri 1982; and Rivera-Batiz 1982; the two-good, three-factor (2 X 3) models of Batra and Casas 1976; Ruffin 1981; and Jones Peter Kuhn is associate professor of economics, Department of Economics, McMaster University. Ian Wooton is associate professor of economics, Department of Economics, University of Western Ontario.
285
286
Peter Kuhn and Ian Wooton
and Easton 1983; as well as the higher-dimensional treatments of Jones and Scheinkman 1977; or Chang 1979), the only one that includes any nontraded goods is Rivera-Batiz (1982). The model in this paper can be thought of as an extension of Rivera-Batiz (1982), which adds an extra traded good and an extra factor. We thus have three factors, two traded goods, and one nontraded good. The additional traded good allows us to have both an exporting and an import-competing industry in the analysis and to compare how these two sectors are affected by changes in factor endowments. The additional factor allows us to distinguish between workers with different investments in human capital. Specifically, we subdivide the labor force into skilled workers and unskilled workers. Immigrants of a particular type are considered to be perfect substitutes for native workers of that same category. Our model may also be thought of as the addition of a nontraded sector to Ruffin’s (1981) 2 X 3 model. The paper’s main theoretical results are twofold. First, we find that the directions of the effects of factor endowments on factor prices, while not zero as in the “standard” trade model, are still independent of the within-industry technical substitution elasticities between inputs in production. This independence property (which incidentally also holds in Rivera-Batiz’s lowerdimensional model) dramatically illustrates the effects of allowing international trade in even a subset of commodities on models of the functional distribution of income. It arises because, contrary to the closed-economy model, the fundamental determinants of factor price changes are not the ability to substitute factors in production; they are, instead, the tendency for factor prices to change in such a way as to maintain the international competitiveness of the country’s exporting and import-competing industries, as long as those industries continue to operate. Second, providing that a relatively weak “normality” condition holds, the directions of all the factor quantity-factor price effects in our model can be deduced directly from the relative intensities of factor use within the traded sector of the economy only, as follows. First, an increase in the supply of any factor lowers its own price. Second, with three factors, one will be “extremely” intensively used in exports, another in imports, and the third will be the “middle” factor in the traded sector of the economy. Our model predicts that an increase in the supply of either extreme factor lowers the price of the other extreme factor and raises the price of the middle factor. Third, an increase in the supply of the middle factor benefits owners of both extreme factors. Interestingly, the results given above are identical to those obtained by Ruffin (1981) without a nontraded sector. The paper’s main empirical result, based on factor intensities in 430 fourdigit U.S. manufacturing industries for the years 1960, 1970, 1980, and 1984, is the following. For all definitions of traded versus nontraded goods considered, and for all years except 1960, skilled labor is extreme in exports and unskilled labor is extreme in imports, with capital as the middle factor. Thus,
287
Immigration and Wages
our model predicts that, in the long run, the interests of both types of labor in immigration issues should coincide and should conflict with those of capital. Workers of both types should oppose all immigration but favor foreign investment in the United States, while owners of capital should favor immigration of both types of workers. Section 10.1 of the paper outlines the structure of the model. Section 10.2 solves the model for the effect of factor endowment changes on factor prices. Section 10.3 characterizes the properties of that solution. Section 10.4 presents our empirical estimates of factor intensities for the United States and their implications, while section 10.5 concludes.
10.1 The Model Each of the three goods X , , X,, and X , is produced using the services of the three factors of production, V , , V2,and V,, according to linearly homogeneous production functions. We adopt the convention that good 3 is nontraded and that, of the two traded goods, XI is imported and X , exported. Let qJbe the quantity of factor i required to produce a unit of good j , where a,Jdepends on the prices of the three factors w,,w,, and w,.Without loss of generality, we number factors in such a way that alliu12
‘2Iia22
u31iu32
in the initial equilibrium and assume the inequalities are strict. Thus, in Ruffin’s (1981) terminology, when comparing factor intensities of the two traded sectors, factor 1 is extreme in imports (XI),factor 2 in exports ( X J , and factor 3 is the middle factor in the traded sector. If the nominal prices of the three goods are p , , p,, and p , , then the zeroprofit conditions for production are (1)
allwl
+
%Iw2
+ ‘3Iw.3
=
(2)
a12wl
+
‘2Zw2
+
= p29
PI?
+ a23w2 + a33w3 = p3. It is assumed that all three goods are produced in positive amounts, so the market prices exactly reflect the costs per unit of output. Full employment of the stocks of the three factors of production would entail (3)
a13wl
Were all three goods prices exogenously determined, then equations (l), ( 2 ) , and (3) would uniquely determine the factor-price vector, and factor prices
288
Peter Kuhn and Ian Wooton
would be independent of the factor endowments. However, while goods 1 and 2 are considered to be traded internationally at exogenously given world prices, it is assumed that good 3 is nontraded, its price being endogenously determined by domestic demand. In consequence, changes in factor endowments, through immigration, may influence the returns to factors in the economy. Let domestic demand for good 3 be represented by a Hicksian compensated demand function, that is,
Equilibrium in the market for the nontraded good occurs when domestic supply exactly meets domestic demand, (7)
and “national’ utility is a function of the quantities of goods consumed in the country by native and immigrant factors together:
u=
(8)
U [ C , ,c,, q.
National utility is maximized subject both to the balanced trade constraint,’ (9)
PI CI + P2
c 2
=
PIX, + PIX,,
and to the constraints of technology and endowments, (10)
x3
=
g [ X , ,x,; v,, v,,
v31.
Equations (I)-( 10) provide a complete description of the static general equilibrium of the economy.
10.2 Factor Migration Consider a change in the domestic supply of a factor of production as a result of migration. This will directly affect production activity through the change in the total factor supplies available for production. It will also affect the (endogenous) price of the nontraded good, with consequent further changes in output and induced changes in factor rewards. Differentiating equations ( l ) , ( 2 ) , and ( 3 ) reveals the way in which the equilibrium is disturbed by small changes in commodity prices: e l l G I + e2]G2 + e31G3= pI, e12Gl
+ e22G2+ e32G3= b2,
e13Gl+ ez3G2 + e33G3 = p3, where O,, =a,w,lp,, the distributive share of factor i in industry j , and a “hat” (*) over a variable denotes a relative change (e.g., = dw/w). Similarly, by ~
289
Immigration and Wages
differentiating equations (4), ( 5 ) , and (6), the response to changes in factor endowments can be determined: (4’)
(5’) (6’)
+ A$, + AIg3 = C, A,$, + AzzX2 + AZg3 = Qz A31Xl + A3zrZz + A3g3 = A,,rZ,
QI
- {at GI
+ a: Gz + a: G3}, - {u; GI + U: Gz + G3}, - {a: GI + U: G, + U: G3},
where A, = a,Xj /Vi, the fraction of the total supply of factor i used in the jth industry, and @ denotes the economy-wide substitution toward or away from the use of factor i when factor k becomes more expensive, under the assumption that each industry’s output is held constant. That is, a: =EjA,Ei, where Ei is the elasticity of demand for factor i with respect to w, in industry j , holding output and other factor prices constant. Were commodity prices to remain unchanged after the factor movement (as would occur if all goods were traded at exogenously given world prices), then, from equations (1’), (2’), and (3’), factor earnings would also remain constant, and hence the bracketed terms of equations (4’), ( 5 ’ ) , and (6’) would all be zero, as there would be no substitution between factors in production. Output change in response to changes in factor endowments would be influenced only by the relative intensities with which factors are used in each of the three industries. This behavior results from a higher dimensional analogue of the familiar Rybczynski theorem (which was derived for a model with two factors and two goods), and we shall call it a “pure Rybczynski effect.” Good 3 is, however, not traded internationally, and the inflow of factors will induce changes in both the demand for and the supply of that good. Differentiating equation (8), and using equations (9) and (10) to determine the change in utility resulting from factor immigration at constant commodity prices,
where p is the marginal utility of money income, and I is national income:
Rewriting equation (1 1) in terms of relative changes yields (11‘)
0
= ox
8’ Qi,
where w = pl/U, and 8‘ is the share of factor i in national income,
290
Peter Kuhn and Ian Wooton
By the appropriate choice of utility scale, let o = 1 locally. Then equation ( 1 1’) becomes
0 = 8lV1 1-
(12)
e2V2
+
e3V3.
Differentiating the market-clearing condition, equation (7), to determine the equilibrium responses to disturbances in the market for good 3,
x, = v,p, + v d 2 + v3p3+ KO,
(13)
where u I , v2, and v3 are the compensated price elasticities of demand for X,, and IJ. is the income elasticity of demand for X,. It will be convenient to rewrite (13), using (12), as
p,p, + pd2+ p3x3+ +{elV, + e2V2 + e3P3}, where PI -vlIv3, p, -v2/v,, p3 l/u3 < 0, and (if good 3 is normal) + .. - phJ3> 0. (13’)
p,
=
’’
”
“
Rewriting the output-response equations (4’)-(6’) in matrix form,
[
’I1
’I2
’13
’21
’22
’231
’31
‘32
’33
u; 0; a;
=
-
Solving this for X,, (14)
f!] [ [2 2 $1 [
x3 = s,Q,+ S2V2 + s3Q3 + ylGl + y2G2 + y3G3,
where (15)
yk
=
-{s,a; +
for k = 1, 2, 3; Sk = IAtl/lhl, for k lambda matrix
=
+ S,a:},
6,a;
1, 2, 3; \A1 is the determinant of the
‘I1
’I2
‘13
’31
’32
’33
and
from the definitions of the “extreme” and “middle” factors. Note that the pure Rybczynski effect (i.e., when factor prices are constant) of factor endowment changes on output of good 3 is then
Substituting equations (3’) and (14) into (13), and solving, yields
291
Immigration and Wages
(17) A I G l+ A2G2+ A3G3=
PI@,+ Pd2+ Zlpl + Z2p2+ Z3p3,
where (18)
A, = (0,3 - P3YJ9
fori = 1, 2, 3; and (19)
z,= P,@, - CLO'),
for i = 1, 2, 3. The comparative statics of factor rewards in this economy are now completely determined by equations (1 '), (2'), and (17):
which is easily solved for G I , G2,and G3. Since, in this paper, we focus on the effects of factor endowments on factor prices, we henceforth set @ = @ = 0 and solve (20) for Gkin terms of pi,for i , k = 1 , 2 , 3 . The induced change in factor rewards when factor endowments change is then given by
for i, k = 1,2, 3,
The properties of equation (21) are analyzed in the following section.
10.3 Consequences for Factor Rewards The consequences of (21) for the effects of factor endowment changes on both nominal factor prices and the welfare of factor owners are analyzed in turn below. We begin with properties of price changes that are independent of a certain "normality" condition and then consider the additional restrictions imposed by that condition. Finally, welfare effects are considered. 10.3.1 General Results To develop an intuition for the effects of endowment changes on factor rewards, consider the effect of a change in the endowment of factor i on the rewards paid to factor k in equation (21). The Zj term gives the effect of im-
292
Peter Kuhn and Ian Wooton
migration (pi> 0) or emigration (pi< 0) on the price of the nontraded good X,. Recalling equation (19), (19)
z, = p3(s,-
pet).
The first term in the parentheses is the supply effect of the change in output of good X , resulting from the immigration of factor i, that is, the pure Rybczynski effect of equation (16). The second term is a demand effect, reflecting the increased demand for good X, resulting from a higher level of national income created by the increase in the economy’s endowment. We call the difference between the terms the “modified Rybczynski effect.” Should this be positive, then an excess supply of the nontraded good has been induced by the migration, triggering a fall in its price, p 3 (because 6, < 0), and vice versa. The other component of the effect of pion w, in (21) is the ratio of the two determinants. Changes in commodity prices in tbe traditional two-factor, twogood trade model induce changes in factor prices according to the StolperSamuelson Theorem. Were all three commodities in our model traded internationally, then the response of factor prices to a change in p , would be a higher dimensional analogue to these familiar magnification effects. We shall call it the “pure Stolper-Samuelson effect.” The magnitude of this response can be measured by solving equations (l’t(3’) (letting$, = ~3, = 0), yielding
[ I
fork = 1, 2, 3, where 1 0 1 is the determinant of the theta matrix,
8
=
‘11
‘21
‘31
‘12
‘22
‘32
‘I3
‘23
‘33
.
Good 3 is not traded, however, and faces a less than infinitely elastic demand. As a result, changes in factor earnings have repercussions on the amount of X , supplied, through within-sector substitution among factors, and this will induce a further change in the commodity price. This effect is captured by the P3y, terms that modify the denominator from 181, in the pure StolperSamuelson effect of equation (22), to A in equation (21). We therefore call the ratio of determinants in (22) a “modified Stolper-Samuelson effect” of p , on w,. It can be shown, for any negative semidefinite economy-wide substitution matrix, that (23)
sign(A) = sign((81),
and so the modified Stolper-Samuelson effect will be qualitatively identical to the pure Stolper-Samuelson effect.2 The entire effect of a change in the endowment of factor i on the earnings
293
Immigration and Wages
of factor k is thus simply the product of the modified Rybczynski effect of V , on p , and the modified Stolper-Samuelson effect of p , on w,. This interpretation and decomposition of (21)emphasizes a major feature of the model that is independent of any assumptions regarding the structure of demand and factor-intensity rankings and is true for changes in the rewards to all factors. The only way that factor endowment changes can affect factor prices is through inducing changes in price of the nontraded good, p,; that is, factor prices cannot change unless p , changes. Otherwise, were p , fixed, because either the good was traded internationally or domestic demand was infinitely elastic (p, = 0 ) , then factor prices would be uniquely determined by equations (1)-(3), independently of the endowment.,
10.3.2 The Normality Condition Consider more closely the excess supply of X, induced by immigration of factor i (6, - pel). The level of national income will always rise with factor inflows, and, if this is compounded by a negative Rybczynski effect (6, < 0) lowering the supply of X,, then there will undoubtedly be excess demand for the nontraded good, inducing an increase in its price. However, suppose immigration of factor V, induces an expansion of X, production because 6, > 0. The change in factor endowment has then raised both the demand for and the supply of the nontradable. Thus, the potential exists for what we shall call a “perverse demand” result that, with a sufficiently high national income elasticity of demand for the nontraded good, an increase in the output of good 3 induced by a change in endowment will be accompanied by an increase in its price. In this subsection, we develop a pair of sufficient conditions, together called the “normality condition,” that rule this out. One of these is a fairly weak restriction on the structure of demand, while the other is a constraint on supply effects that, as we shall see, is clearly satisfied by our empirical evidence on factor intensities. Together they ensure that (24)
sign(6, - kel) = sign@,);
thus, the modified Rybczynski effect, like the Stolper-Samuelson effect, is qualitatively the same as the “pure” effect. First, rewrite the expression for 8, in equation (16)in terms of distributive shares:
where 0, = pjXj/I,the share of good j in national expenditure. From our numbering of factors in terms of relative intensity of their use in the traded sector, we know that
(26)
sign(+,)
=
sign(+,) = -sign(+J.
294
Peter Kuhn and Ian Wooton
Thus, immigration of either of the two “extreme” factors will have opposite effects to immigration of the “middle” factor on the production of the nontraded good. Furthermore, it is helpful to note that (27)
sign(6,)
=
signlo(,
since l0,l > 0. A sufficient condition for 6, > 0 is then that
-‘ 3 2< - ‘12 -‘22o r - <‘31‘33
‘13’
‘23
‘33
‘I1
-.‘21
‘13’
‘23
In other words, an increase in the supply of the middle factor in the traded sector (V,) raises the supply of the nontraded good if, relative to either one of the traded goods, nontraded goods are extremely intensive in V , . Conversely, if nontraded goods are extremely unintensive in V , relative to either X , or X , , the supply of X , will fall when V3rises. Thus, through (27) and (26), the signs of all the Rybczynski effects hinge on whether the nontraded sector uses the middle “traded” factor intensively or unintensively. Now assume that the two traded goods, XI and X,, are not, as a bundle, inferior in demand, in the sense that the total expenditure on the two goods ( p , X , + p , X , ) does not decline when national income increases, at constant prices. Differentiating the national income identity yields
i = elxl + 0,x, + 03x,. XI and X , then are not inferior as a bundle if and only if
i - 0,x, > 0. Noting that p = X3/f,this may be rewritten as 1 > k0,.
Multiplying by 0‘ and rewriting yields
Thus, the noninferiority of goods 1 and 2 ensures that the demand term modifying the Rybczynski effect is less than the wage bill of the immigrant factor relative to the value of output of X , . We can therefore rule out the perverse demand result if, when 6, > 0, 6, > 0Y0,. Using equation (25), this may be rewritten to state: sign 10.1= sign 101 Suppose that factor 3 has a positive Rybczynski effect, that is, S3 > 0. As are neganoted earlier, this means that 1 01 is positive. As both l0,l and
295
Immigration and Wages
tive, we are assured that l9,l exceeds 191, and, hence, the perverse demand result cannot occur if goods 1 and 2 are noninferior. If, instead, immigration of factor 3 directly reduces the production of the nontraded good, that is, 6, < 0, then the potential for a perverse demand result arises with immigration of either factor 1 or factor 2. Now, as 191 < 0 and only I@,[ > 0, it is not possible to rule out l9,l/l9l
1 for 191 < 0 a n d j = 1 or 2, which ensures a large enough supply effect in the cases where the nontraded sector is extremely unintensive in the middle “traded” factor. The latter condition is clearly satisfied by the data on factor intensities that we consider below. 10.3.3 Effects of Factor Endowments on Factor Prices under the Normality Condition Under the normality condition, modified Rybczynski effects are qualitatively the same as the pure effects, as is also true for modified StolperSamuelson effects. Furthermore, it is straightforward to show that the Rybczynski effect of Vt on X , is always the same sign as the StolperSamuelson effect of p , on wi.This can be seen by manipulating (25) and substituting into (23), yielding (30)
sign(6,) = sign
[TI.
19.1
Substituting (24) and (30) into (21), and noting that p, < 0, we have
(31)
=
sign( - s,s,~J,
for i, k = 1 , 2, 3. Irrespective of whether 191 S 0, it is clear from (31) that any factor is its own “enemy,” in that immigration of a factor will always lower its own nominal wage, while emigration raises its earnings. Further, as w , and w, have the same sign, the extreme factors, 1 and 2, will be mutual enemies while being friends with the middle factor, 3. The qualitative effects of factor quantities on factor prices thus depend only on which factors are extreme in the traded sector of the economy, not on the relative factor intensities of the traded versus nontraded sector. The reason is that, whenever a factor supply shift raises the price of nontraded goods (by lowering supply), higher nontraded goods prices lower the equilibrium price of that factor, and conversely. 10.3.4 Welfare Effects The effectof the above nominal changes in returns on the welfare of a factor depends on the induced changes in p , ( p , and p , being fixed, by assumption). Assessment of welfare effects is, however, made easy if we note that, by substituting (22) into (21),
296
Peter Kuhn and Ian Wooton
p3 = p, 101(6, A
kf?)+,.
Since the ratio of determinants is always positive, the general equilibrium effects of V, on p 3 are always the same sign as when factor prices are held fixed. In more detail, immigration of factor Viwhose 6, > 0 will result in a fall in p , , raising the real income of those factors whose nominal earnings have increased. The other factors are faced with both reduced nominal earnings and lower prices, with apparently ambiguous consequences for their welfare. However, consider the relative change in the ratio of earnings of factor k to the price of good 3, (33)
[3
= w -
p,
[m 1%
Substitute equations (21) and (32) into this expression and rearrange to find (34)
t]
=
8 3
- 1).
Using (29), (31), and (32), equation (34) implies that sign [6,1p3] = sign 6,. Thus, under the normality condition, changes in nominal earnings will outweigh changes in the price of the nontraded good, ensuring that changes in welfare will have the same signs as the changes in factors’ nominal returns. Table 10.1 gives a complete listing of the predicted responses of variables to changes in migration.
10.4 Empirical Estimates of Factor Intensities Our estimates of U.S. factor intensities are based on an analysis of 430 four-digit manufacturing industries during the period 1960-84. These indusTable 10.1
Predicted Effects of Factor Endowment Changes under the “Normality”Condition Endowment Change
297
Immigration and Wages
tries constitute all the industries in the NBER Data Set on Trade, Immigration, and Foreign Direct Investments (450 in all) with, data on both exports and imports.6 This data set has the advantages of being at a finer level of aggregation than most existing empirical studies of factor intensities and trade and of being available for a long series of years, as recently as 1984.’ It has the disadvantage of being limited to manufacturing industries only. Thus, we view our results here as (a)likely to be quite accurate for the manufacturing sector relative to other studies but (b) relevant to the whole economy only to the extent that the pattern we identify in manufacturing generalizes to other industries. Our basic procedure begins by allocating the total value added in each fourdigit industry into payments to one of the three factors in the model. Total income of “unskilled” workers was defined as the industry’s total productionworker payroll for the year. Income of “skilled” workers is the difference between this and the industry’s total payroll.8 Finally, subtracting total payroll from total value added yielded the payments to “capital” in the industry. Thus, in keeping with our model’s assumptions, we ignore indirect factor inputs and assume no intermediate inputs in produ~tion.~ Next, the 430 four-digit industries were aggregated into three as follows. First, nontraded goods were defined as those industries whose level of “openness” to international trade, measured by the sum of exports plus imports divided by total industry output, was under k%, where k took on alternative values of 0, 5, 10, and 20. Then the remaining, traded goods were divided into import-competing goods and exports, depending on whether the net imports of the industry (imports less exports) were positive or negative, respectively. This yielded 3 X 3 table of factor incomes, from which it was then straightforward to calculate the matrix of factor intensities, 0. Table 10.2 shows the factor intensities calculated in this manner for the year 1984 and with k = 10%. This criterion yielded a nontraded sector producing $351 billion of value added, which is 36.6% of the total value added of $959 Table 10.2
Estimated Factor Intensities for U.S. Manufacturing, 1984 V,, Unskilled Labor
V,,Skilled Labor
V,, Capital
1. “Absolute” factor intensities (9J:
a. XI (import competing) b. X , (exporting) c. X, (nontraded) 2. Relative factor intensities (9,/9,J: a. 9,,/9,, (imports vs. exports) b. 8,,/9,,(imports vs. nontradeds) c. 9,,/9,, (exports vs. nontradeds)
,2615 ,2092 ,2205
,1504 ,1858
.I408
1.2502
.SO96
1.1861
1.0681
.9487
1.3192
,5881 ,6051 ,6387 ,9719 ,9208 ,9473
Sources: NBER Data Set on Trade, Immigration, and Foreign Direct Capital Investments. Note: Nontraded sector is defined as industries with (exports + imports)/output under 10%. Determinant of 9 equals ,0019528.
298
Peter Kuhn and Ian Wooton
billion in the sample. Import-competing and exporting industries, respectively, produced $361 billion (37.6%) and $247 billion (25.7%) of value added. A list of industries producing more than $10 billion of value added, by trade category, is given in table 10.3. Row 2a of table 10.2 clearly indicates that, within the traded sector, unskilled labor is extremely intensive in imports and skilled labor is extreme in exports, with capital as the middle factor. This accords well with the widely held notion that the United States is well endowed with skilled labor and poorly endowed with unskilled labor, relative to the rest of the world (e.g., Baldwin 1971).1° The list of industries in table 10.3 supports this notion, with what are commonly considered “low-technology’’ industries such as paper and steel appearing only in the import category and “high-technology’’ industries like computers dominating in exports. Finally, row 2b of table 10.2 indicates that the middle “traded” factor, V,, is used extremely intensively in the nontraded sector relative to imports. Thus, in a well-defined sense, the nontraded sector is intensive in capital and unintensive in both types of labor, relative to the traded sector. By previous results, this means that 1 01 > 0 and that the “perverse demand” result cannot occur as long as XI and X , are normal goods (either individually or as a bundle). Thus, in combination with the results in table 10.2, our model predicts that immigration of either type of labor will, in the long run, hurt doTable 10.3
Four-Digit Industries with over $10 Billion of Value Added, by Wade Category, 1984
1. Exports:
2869 3573 3721 2. Imports: 2621 291 1 3312 3662 3674 3679 371 1 3714 3861 3. Nontraded goods: 271 1 2752 2834 3079 3761
Industrial organic chemicals, n.e.c Electronic computing equipment Aircraft Paper mills Petroleum refining Blast furnaces Radio and TV transmitting, signaling and detection equipment Semiconductors and related devices Electronic components, n.e.c. Motor vehicles and passenger car bodies Motor vehicle parts and accessories Photographic equipment and supplies Newspapers, publishing Commercial printing, lithographic Pharmaceutical products Miscellaneous plastics products Guided missiles and space vehicles
Note: Nontraded goods deflned as (exports where classified.”
+ imports)/output < 10%. N.e.c.
means “not else-
299
Immigration and Wages
mestic owners of both types of labor. The reason is that immigration raises the price of nontraded goods, which in turn benefits owners of capital. This higher return to capital must be compensated for by a reduction in wages of both types of workers if U.S. exports and import-competing industries are to remain internationally competitive. How robust are these conclusions to changes in the definition of traded versus nontraded sectors and to changes in factor intensities over time in the U.S. economy? This question is explored in table 10.4, which summarizes our estimates of factor intensities for various levels of k and for other years. What is most striking about this table is that, for every year but 1960 and for every level of k, the factor intensity rankings in the traded sector are the same as in table 10.2. Also, (81is the same sign, indicating that the relative factor intensities of the traded versus nontraded sectors are unchanged and that normality of X , and X , in demand continues to be sufficient to rule out perverse factor-demand results. Only in 1960 are the results mixed; we feel that this is connected with changes in the U.S. economy over time that produced “Leontief paradox”-type results in earlier years but no longer do so, as the United States specializes more and more in knowledge-intensive industries. We conclude that our estimates of direct relative factor intensities (and the resulting predictions about the effects of immigration) reliably summarize a broad underlying pattern that has persisted in manufacturing since 1970 or even somewhat earlier. Table 10.4
Factor-IntensityRankings for Various Definitions of the Nontraded Sector, 1960, 1970, 1980, and 1984
+
Maximum Level of (Exports Imports)/ Output in the Nontraded Sector
.o 1960: Extreme factors in imports and exports, respectively (within traded sector) Sign I 0 I 1970: Extreme factors in imports and exports, respectively (within traded sector) Sign I 0 I 1980: Extreme factors in imports and exports, respectively (within traded sector) Sign I e I 1984: Extreme factors in imports and exports, respectively (within traded sector) Sign 10 I
.05
.I0
.20
K, S
+
u, s
u, s
+
NA
Note: K = capital, S = “skilled (nonproduction) labor, and U
u, s
+
u, s
+
=
“unskilled” (production) labor.
300
Peter Kuhn and Ian Wooton
How robust are our estimates to the inclusion of nonmanufacturing industries in the analysis? This question cannot be answered with the NBER data but is explored using two-digit industry data from another source in the Appendix. Interestingly, this analysis indicates that the basic results do generalize. The basic reason for this is that services (which constitute the bulk of the excluded, nonmanufacturing industries), contrary to widespread beliefs, are not always nontraded and not always labor intensive. For example, real estate (a very important nontraded service) is highly capital intensive, while education (which is very intensive in skilled labor) is a substantial U.S. net export. Still, owing to the considerable practical and conceptual problems involved in estimating service trade (see the data sources listed in the Appendix), we view this finding with considerable caution. Clearly, there is an outstanding need for careful empirical analysis based on better-quality trade data for service industries. The consistency of our results across definitions of sectors and years is, however, suggestive and proves a useful illustration of the ease with which our theoretical model can be implemented. 10.5
Conclusion
This paper has developed a simple, general equilibrium model of how factor endowments affect factor prices when a subset of the goods produced in the economy is traded at internationally fixed prices. The result is a model that makes unambiguous predictions that are independent of estimated elasticities of substitution among factors in production. Our empirical analysis, based on 430 four-digit manufacturing industries in the years 1960, 1970, 1980, and 1984, indicated that, at least since 1970, factor intensities in U.S. manufacturing follow a very consistent pattern: skilled labor is used extremely intensively in exports, while unskilled labor is extremely intensive in import-competing industries. Furthermore, the middle factor in the traded sector, capital, is used intensively in the nontraded sector relative to the traded sector. Thus, to the extent that the relative factor intensities we find in manufacturing generalize to the whole economy, our model predicts the following. Increased immigration of either skilled or unskilled workers to the United States will, in the long run, hurt U.S. workers of both types and benefit owners of capital. These effects should be associated with an increase in nontraded goods prices, which, by reducing the international competitiveness of the country’s traded goods, causes the reduction in wages. They are also independent of the technical substitution elasticities in production that so many analysts have attempted to estimate (for a recent summary, see Hamermesh and Grant 1979). Our model can, of course, be extended and improved on in various ways. These include incorporating intermediate inputs (as is done in empirical work by Baldwin 1971; and Stem and Maskus 1981), allowing for capital mobility (as in Gerking and Mutti 1983), considering shorter run effects (as in Rivera-
301
Immigration and Wages
Batiz 1987), relaxing the extreme assumption that U.S. consumption and production have no effect on the prices of traded goods, and considering nonbalanced trade. In addition, better data, which consider nonmanufacturing industries as well, may also be available in the relatively near future as more trade statistics on services are collected. All these extensions could, of course, change our specific predictions about the directions of factor price changes here, which we view as suggestive but tentative. It is clear that they will not, however, change what we view as the fundamental lesson of this paper. This lesson is that the effects of immigration in a partially open economy may be determined by a fundamentally different set of factors than in a closed economy, where technical substitutability between factors in production plays the key role. In the partially open economy, factor prices are constrained to change in a way that preserves the international competitiveness of its traded goods, as long as those goods continue to be produced. This places tight restrictions on the kinds of factor-price changes that can occur. These restrictions deserve, we feel, greater prominence in the work of empirical and policy-oriented researchers studying the effects of immigration to the United States.
Appendix Factor Intensity Estimates or the Entire U S . Economy from Two-Digit ndustry Data, 1983
f
This appendix explores the generalizability of our results based on manufacturing only to the entire U.S. economy by constructing a factor-intensity matrix for the entire U.S. economy from two-digit data for 1983. These data were obtained from the following sources: total value added and compensation of employees from the Survey of Current Business (66 [July 19861: 1986, table 6); merchandise trade data from the 1985 Statistical Abstract of the United States (tables 1448, 1449); and rough estimates of service trade from the Office of Technology Assessment (1986, table 2, using midpoints of intervals). Thus, some services are classified as exports (e.g., education), others as imports (e.g., insurance), and still others as nontraded (e.g., retailing). Skilled workers were defined as those who completed high school; the percentage of the work force that was skilled was then taken from a 1980 Census tabulation of occupation by industry (U.S. Department of Commerce series PC-80-2-70, pp. 1-80) and their relative wage rates in 1983 from U.S. Department of Commerce series Money Income of Husbands, Families and Persons in the United States (P-10, no. 146, table 48). Payments to land as an input are included by definition with capital, which functions as a kind of “residual” factor here. Because of the small number of industries. we chose
302
Peter Kuhn and Ian Wooton
criteria of openness to trade and a critical balance-of-trade level that yielded nontraded, exporting, and importing sectors of roughly equal sizes (this meant that some industries with low trade deficits were classed as exports). This procedure yielded the following factor-intensity matrix, 8:
where the three factors are, respectively, unskilled labor, skilled labor, and capital, as in table 10.2. Noting that, as in table 10.2, the nontraded sector is again relatively intensive in capital, w e conclude that the factor-intensity rankings w e find in manufacturing may well generalize completely to the entire U.S. economy when better data on service trade are available.
Notes 1. Since this is an undistorted economy, we know that it maximizes a weighted sum of the utilities of its members subject to the balanced trade and resource constraints. Under certain conditions (see, e.g., Gorman 1953), this is equivalent to maximizing a community utility function of the type used here. We assume that such conditions are satisfied here, which does not necessarily imply that all agents (including immigrants) are equally endowed. This latter condition would of course make the analysis of immigration much less interesting. 2. To see this, note that the denominator can be written as
Through algebraic manipulation, the second term can be reduced to [p3/(0,10/)]u’Tu, where T is a negative semidefinite matrix derived from the economy-wide substitution matrix (see Jones and Scheinkman 1977), and u is a column vector. The product U‘TU is nonpositive, while p, is negative, and 8, is positive. By simple manipulation it follows that lel/A is always positive. 3. It is assumed that endowment changes do not move the economy out of the “cone of diversification.” 4. The national income elasticity of demand equals the national utility elasticity of demand since, locally, units of income and utility are equivalent. Also, it is easily shown that, in the “perverse demand” case, the demand curve for factor V ,in general equilibrium is upward sloping. While this might be a fortunate situation for owners of V,,it appears quite unlikely given the fairly weak sufficient conditions needed to rule it out. 5. In the standard 2 X 2 model of international trade, the Rybczynski effect of the immigration of one factor resulted in the output of one industry increasing while production in the other industry diminished. Thus, the value of output of the expanding
303
Immigration and Wages
industry had to increase by a greater amount than national income. If both of the goods were normal, then the increase in the value of production for the expanding industry would then necessarily be greater than the increase in the value demanded of the good, ensuring a nonperverse demand result. Our results show that, when there are three goods and three factors, nonperversity is not guaranteed. To see this, suppose that immigration of both factors 1 and 2 increases output of X,, through the Rybczynski effect. Even with factor intensities unchanged (as product prices are constant), this need not result in a fall in the output of both of the traded goods. Indeed, the increase in the value of production of X , might be quite small relative to the increase in national income when output of one of the traded goods also increases. Thus, the potential does exist that the expansion in output of good 3 may be less than the increase in its demand, even when goods 1 and 2 are jointly noninferior in consumption. 6. The factor income information in the NBER data set was taken from the Annual Survey ofManufactures for various years. This could be matched with trade figures from the Trading Monitoring System of the Bureau of Labor Statistics for 430 of the 450 four-digit industries. The data were collected and made available to us by John Abowd and Richard Freeman. 7. See table 4.2 in DeardorfF (1984). The finest level of detail used in the studies cited there is in Stem and Maskus (1981), who use 128 three-digit industries. Our sample of 430 four-digit industries contrasts very favorably with all three studies. 8. An alternative, and perhaps preferable, definition of skilled vs. unskilled workers might be based on total years of education. Unfortunately, information on years of education by industry is not available in this data set. Our experiments using high school completion rates with two-digit data, reported in the Appendix, lead us to expect that this would not change the results. 9. Ideally, our theoretical model should incorporate (both traded and nontraded) intermediate inputs and specify how these should fit into the empirical analysis. Since it appears that this extension of our model would significantly complicate our analysis, it is not undertaken here. We feel that it is, however, an important area for further research on this topic. 10. This notion has often been advanced as a possible explanation of the “Leontief paradox” of capital-intensive U.S. imports (Leontief 1953). Interestingly, this paradox does not arise in table 10.2 here (the combined labor shares in imports exceed those in exports, and the capital share is greater in exports than imports). Three possible reasons for this are the fact that we use four-digit, not two-digit, data, the fact that our focus on manufacturing excludes natural resource industries, and the later time period. Of these, the “time” explanation seems to be most convincing, for the following reason: for the four different values of k considered in the experiments of table 10.3, the capital share in exports exceeded the capital share in imports three out of four times in 1984, four times in 1980, twice in 1970, and zero times in 1960. This seems fairly strong evidence of an increasing relative capital intensity of U.S. exports over time.
References Baldwin, R . E. 1971. Determinants of the Commodity Structure of U.S. Trade. American Economic Review 61:126-46. Batra, R. N., and F. R. Casas. 1976. A Synthesis of the Heckscher-Ohlin and the Neoclassical Models of International Trade. Journal of International Economics 6:21-38.
304
Peter Kuhn and Ian Wooton
Brecher, R. A., and E. U. Choudhri. 1982. The Factor Content of International Trade without Factor-Price Equalization. Journal of International Economics 12:277-84. Chang, W. W. 1979. Some Theorems of Trade and General Equilibrium with Many Goods and Factors. Econometrica 47:707-26. Deardorf€, A. 1984. Testing Trade Theories and Predicting Trade Flows. In Handbook of International Economics, vol. 1, ed. R. W. Jones and P. B. Kenen. Amsterdam: North-Holland. Fallon, P. R., and P. R. G. Layard. 1975. Capital-Skill Complementarity, Income Distribution, and Output Accounting. Journal of Political Economy 83:279-302. Gerking, S., and J. Mutti. 1983. Factor Rewards and the International Migration of Unskilled Labor: A Model with Capital Mobility. Journal of International Economics 14:367-80. Gorman, W. 1953. Community Preference Fields. Econometrica 21:63-80. Hamermesh, D. A., and J. Grant. 1979. Econometric Studies of Labor-Labor Substitution and Their Implications for Policy. Journal of Human Resources 1 4 518-42. Hicks, J. R. 1932. The Theory ofWages. New York: Macmillan. Jones, R. W., and S. T. Easton. 1983. Factor Intensities and Factor Substitution in General Equilibrium. Journal of International Economics 15:65-99. Jones, R. W., and J. A. Scheinkman. 1977. The Relevance of the Two-Sector Production Model in Trade Theory. Journal of Political Economy 85:909-36. Kemp, M. C. 1966. Gain from International Trade and Investment. American Economic Review 56:788-809. Leontief, W. 1953. Domestic Production and Foreign Trade: The American Capital Position Re-examined. Proceedings of the American Philosophical Society 97: 332-49. Markusen, J., and J. Melvin. 1979. Tariffs, Capital Mobility, and Foreign Ownership. Journal of International Economics 9:395-410. Office of Technology Assessment. Congressional Board of the Ninety-ninth Congress. 1986. Trade in Services: Exports and Foreign Revenues, Summary. Washington, D.C.: U.S. Government Printing Office. Rivera-Batiz, F. 1982. Nontraded Goods and the Pure Theory of International Trade with Equal Numbers of Goods and Factors. International Economic Review 23: 401-9. . 1987. Modeling the Short-Run Economic Effects of Immigration: Some General Equilibrium Simulations. In Modeling andSimulation, Vol. 17, ed. W. G. Vogt and M. H. Mickle. Research Triangle Park, N.C.: Instrument Society of America. Ruffin, R. J. 1981. Trade and Factor Movements with Three Factors and Two Goods. Economic Letters 7: 177-82. Samuelson, P. A. 1948. International Trade and the Equalization of Factor Prices. Economic Journal 58: 163-84. Stem, R. M., and K. E. Maskus. 1981. Determinants of the Structure of U S . Foreign Trade, 1958-1976. Journal of International Economics 11~207-24.
11
Immigrants, Labor Market Pressures, and the Composition of the Aggregate Demand Susan M. Collins
The purpose of this paper is to examine the effect of changes in the composition of aggregate demand on total labor requirements and on the requirements for jobs typically held by immigrants, using the input-output (10)tables for the United States. The paper asks two sets of questions. First, how are labor requirements affected by a dramatic turnaround in the trade balance? Does it matter whether the deficits are accompanied by an investment or a consumption boom, and how are the resulting labor market pressures distributed across industries? Second, how are immigrants distributed across domestic industries, are they differentially affected by shifts in the composition of demand, and are the recent changes likely to have made immigrant workers more “visible,” providing one explanation for the increased attention they have received in recent years?’ The basic approach is as follows. Changes in the composition of aggregate demand will alter the distribution of labor requirements across sectors and industries. While these shifts do not imply changes in labor demand or in actual employment, they can be interpreted as indicating labor market pressures in those sectors where demand has decreased. Because immigrants and native workers are distributed quite differently across jobs, these pressures will influence the two groups differently.* Input-output analysis provide a useful framework to explore the linkages between aggregate demand, labor market pressures, and immigrants because it integrates both microeconomic and macroeconomic aspects. On the microeconomic side, it considers the output and employment responses of particular industries. On the macroeconomic side, it incorporates the key identity from Susan M. Collins is associate professor of economics at Harvard University and a faculty research fellow of the National Bureau of Economic Research. The author would like to thank 1. Abowd, H. Bowen, R. Freeman, and K. Lang for comments and A. Revenga for excellent research assistance.
306
Susan M. Collins
the National Income and Product Accounts, which highlights the linkages between net exports and the other components of aggregate demand. The counterpart to an external imbalance (i.e., a deficit in U.S. goods and services vis i vis the rest of the world) must be an excess of investment over domestic savings. The macroeconomic focus is important because it points to a different set of issues and conclusions than many of the industry studies. For example, suppose that a researcher concluded that imports had been a “cause of injury” in a particular industry and that restricting imports would be likely to raise domestic output and employment. From the macroeconomic perspective, unless the policies to restrict imports were expected to reduce the savings investment imbalance, thereby reducing the total trade deficit, these policies merely shift the trade deficit between sectors. There are also some drawbacks to the I 0 analysis. By maintaining constant input-output coefficients and fixing the commodity composition of each component of aggregate demand, it rules out substitution on both the production and the consumption sides. A related point is that it does not specify why aggregate demand changes and how relative prices (including interest rates and exchange rates) are affected. The answers to these questions will in turn have implications for the composition of imports, consumption, and the other components of demand. In order to incorporate these factors, it would be necessary to imbed the I 0 framework into a macroeconomic model, which is beyond the scope of the current paper. The paper is composed of four remaining sections. Section 11.1 asks where the immigrants are and examines the distribution of immigrant workers across sectors. Section 11.2 turns to the key macroeconomic issues and discusses changes in the composition of aggregate demand. Section 11.3 analyzes the effect of shifts in the composition of aggregate demand on labor requirements by sector and for immigrants and nonimmigrants. The section first spells out the methodology and then discusses results. Concluding remarks are given in the final section.
11.1 Where Are the Immigrants? A number of authors have pointed out that immigrant workers tend to be concentrated in different industries than native workers. In particular, immigrants tend to enter the labor market in low-wage, relatively unskilled positions, but the distribution of immigrants over industries and occupations becomes more similar to that of natives the longer they remain in the United state^.^ Table 11.1 compares the 1982 employment distribution of foreign-born (immigrants plus refugees) and native workers across sectors. These data are derived from a special matched sample of respondents to supplementary CPS surveys conducted in March and April 1983. Unfortunately, only about 75%
307
Immigrants, Labor Market Pressures, and the Aggregate Demand
Table 11.1
Employment Distribution by Sector % of Total Group Employment
Total Agriculture Mining Construction Manufacturing Transportation and public utilities Wholesale trade Retail trade Finance, insurance, real estate Private household Other service Public administration Total (thousands)
Foreign
Native
Foreign as % of Total Sector Employment
4.7 25.3
3.4 1.o 6.0 19.2
8.4 5.7 6.4 10.2
1.0 4.3 16.3
4.2 3.9 16.5
1.2 4.4 16.3
4.8 7.1 8.1
6.2 1.3 30.3 4.7
6.6 1.8 30.5 2.4
6.2 1.2 30.3 4.9
8.4 11.5 8.0 4.1
109,064
8,694
100,370
7.97
3.4 .9 5.9 19.6
3.6
.I
Source: Sehgal(1985).
of the April sample matched with the March sample, and no corrections were made for missing value^.^ The first three columns of the table report the shares of total, foreign-born, and native employment in each of eleven sectors. The final column gives the percentage of foreign born in total sectoral employment for each sector. The table shows that foreign born accounted for 7.97% of total employment but that these workers were not evenly distributed across sectors. The largest difference between the two groups is in manufacturing, which accounted for 25% of foreign-born workers but only 19% of native workers. Furthermore, manufacturing has the second highest concentration of immigrants, behind private household services. Immigrants are relatively underrepresented in public administration. However, it is not surprising that natives are twice as likely to hold these jobs since many of them require citizenship. Immigrants are also underrepresented in construction and in transport and public utilities. Within manufacturing, immigrants are disproportionately located in nondurable goods: 9.5% of total employment in nondurables was immigrant compared to less than 7% of total employment in durables. Immigrants are disproportionately located in apparel, where they account for over 19% of total employment. They also account for large employment shares in textiles, footwear, leather, drugs, and cleaning and toilet preparations. They are relatively scarce in tobacco, petroleum refining, and chemical product industries. The various durable goods industries each account for a small share of total immigrant employment. Immigrants are relatively visible in some sectors, such as miscellaneous manufacturing.
308
Susan M. Collins
In summary, immigrants account for less than 8% of total employment. However, they are distributed across industries quite differently than native workers are. Furthermore, they are extremely visible in some industries, amounting to 12%-20% of the total work force. The remainder of the paper explores the implications of recent changes in aggregate demand on the distribution of job requirements across industries and asks whether reductions in job requirements have been concentrated in industries where immigrants are also concentrated. 11.2 Shifts in the Composition of Aggregate Demand In fact, there have been large recent changes in the composition of aggregate demand. The two identities from national income accounting given in (1) and (2) are very useful for documenting the shifts in key macroeconomic variables and for highlighting the linkages between the foreign sector and domestic demand:
C
+ 1 + G + (X - M),
(1)
Y
(2)
(X - M ) = S,
=
+ S, - 1 - R .
As usual, Y denotes GNP; C , I , and G denote private consumption, investment, and government spending, respectively; and X and M refer to exports and imports of goods and services, S, and S, to private and government savings, and R to net other international transaction^.^ Equation (2) says that foreign savings must equal the difference between domestic savings and investment. Table 11.2 shows the U.S. experience during 1973-86. The top panel gives the composition of aggregate demand as shares of GNP during each of four subperiods, while the bottom panel shows the domestic savings and investment counterparts to net export performance. As shown, net exports declined during 1977-79, improving somewhat during 1980-82 before the substantial deterioration during 1983-86. The two periods of poor trade performance differ in more than simply the magnitude of the deficit. In 1977-79, investment rose by 2.2% of GNP relative to 1973-76, requiring additional domestic and/or foreign savings. Approximately one-third was met by foreign savings, as the trade deficit declined by .8% of income. The remaining two-thirds was met by an increase in government savings. During 1980-82, investment fell, as did all three components of savings. However, because of the larger decline in government savings, foreign savings did not return to its 1973-76 level-the trade balance recovered only partially. The 1983-86 period stands in stark contrast to 1977-79. Although the trade balance deteriorated by 2.6% of GNP, investment rose by little more than .5 percent. Instead of increased domestic savings, government savings fell pre-
309
Immigrants, Labor Market Pressures, and the Aggregate Demand
Table 11.2 Year
The U.S. Experience, 1973-86 Consumption
The shifting composition of aggregate demand (as a percentage of GNP): 1973-76 62.6 1977-79 62.7 1980-82 63.6 1983-86 65.2
Investment
15.8 17.9 15.7 16.3
Gov’t
20.2 18.9 19.6 20.1
(Defense)
Net Exports
(5.5)
1.3
(4.9) (5.6) (6.4)
1.o - 1.6
.5
Savings Private Decomposition of net exports: savings investment (as a percentage of GNP): 1973-76 1977-79 1980-82 1983-86
18.1 17.9 17.7 17.2
Gov’t
- 1.5 - .2 - 1.9 -3.3
Private Investment
15.8 17.9 15.7 16.3
Net Exports
Other
1.3 .5
.6
1.o
.9
- 1.6
.8 .8
Source: Economic Report of the President, 1987
cipitously, by 1.5%of GNP, while private savings continued its trend decline. From table 11.2, government spending rose by just .5 percent of GNP (although this figure masks the large shift toward defense spending). Private consumption, on the other hand, ranged from 64.8% to 65.6% of GNP during 1983-86. The jump is especially notable because private consumption has been relatively stable at 63% of income since 1950 and has exceeded 64% in only four years between 1950 and 1980. Thus, the 1983-86 trade deficit coincided with the large reduction in government revenues, which lowered government savings but raised private consumption. The next section of the paper explores the labor market implications of these compositional shifts in demand.
11.3 Final Demands and Labor Requirements 1 1 .3.1 The Framework The I 0 tables provide a useful way to link changes in the composition of demand to shifts in industrial output and labor requirements. As already discussed, the major shortcomings of the approach are that it does not consider whether the demand shifts are associated with relative price changes and that it rules out substitution-on both the demand and the supply side-by assum-
310
Susan M. Collins
ing constant coefficients. The results provide information about the labor that would be required to produce the sectoral outputs consistent with a particular final demand. These labor requirements may be very different from sectoral employments-especially in the short run. Although the results of an I 0 analysis cannot be interpreted as indicting shifts in actual labor demands or employments, they do provide information about the likely labor market pressures. The analysis uses the eighty-five-industry-level disaggregation of the 1980 I 0 tables.'j Final demands and data used to compute the technical input-output coefficients are valued in producer prices. The vector of labor requirements per dollar output is matched to the 1977 10 tables.' The exercises discussed below will consider different compositions of final demand, so that the index k refers to the kth scenario, or aggregate demand composition. The central relation is given in equation (3):
(3)
L,
=
Y [ I ] . Q . dk9
where
L, = the total labor required in each industry, given the kth aggregate demand composition (85 x 1); y = the vector of labor requirements for a dollar of output industry
(85 X 1); I = the identity matrix (85 x 85); Q = the total requirements (direct and indirect) matrix of the output from each industry required to prduce a dollar's worth of each commodity (85 x 85); d, = the vector of final demands for each commodity, given the kth aggregate demand composition (85 X 1). Equation (4)divides final demand into three parts:
(4)
d, =
n
*
T,.D,
where
D = total aggregate demand;
TT, =
n=
the vector of shares in total demand of each of the nine componentsprivate consumption, investment, inventory accumulation exports, imports, and four types of government expenditures (9 X 1); the matrix of demands for each of the eighty-five commodities per dollar of each component of final demand (85 X 9).
To focus on the implications of recent shifts in demand composition, alternative final demand vectors were computed by varying the shares in aggregate demand (T,)while holding everything else constant. In other words, total demand and the commodity composition of each piece of final demand were held constant as the shares of investment, consumption, imports, etc. were varied.
311
1 1.3.2
Immigrants, Labor Market Pressures, and the Aggregate Demand
The Components of Final Demand
Variations in the composition of aggregate demand will influence labor requirements even when total demand is held constant. This is because each type of demand has a different commodity basket so that each concentrates its spending on commodities with different labor requirements. To help interpret the results in the next sections, table 11.3 compares the total labor requirements to produce one million (1980) dollars worth of the commodities in each demand component. (Thus, the entry for imports is positive .) Inventories are excluded because the commodity composition of inventories varies substantially from year to year so that it is not particularly useful to think of them as a fixed basket of commodities. The first column of table 11.3 reports the total number of jobs required per million dollars. The second column gives the average expenditure per job to produce each commodity basket. The figures point out that there is substantial variation. Government spending requires the most labor per dollar spent. A million dollars of government expenditure requires 69 jobs when spent on education and 55 jobs when spent on defense. (Equivalently, the figures imply a total of $14,400 of educational expenditure per job and $18,200 of defense expenditure per job .) Government expenditures are followed by fixed investment and private consumption, which require 42 and 41 jobs per million dollars, respectively. It may seem surprising that the external sector has the smallest labor requirements and that labor requirements are slightly higher for exports than for imports. Because of the relative capital abundance in the United States, imTable 11.3
Labor Requirements and Final Demand All Workers
'Qpe of Demand Private consumption Fixed investment Exports Imports ( - ) Federal government: Defense Other State & local: Education Other Addendum:' Exports Imports ( - )
Immigrants
Jobs per $million Expenditure
$/Job
Jobs per $million Expenditure
% Total Jobs
40.7 42.0 26.6 23.0
24,600 23,800 37,600 43.500
3.5 3.2 2.1 2.1
8.5 7.7 8.0 9.2
54.9 61.8
18,200 16,200
3.2 3.5
5.8 5.6
69.4 59.7
14,400 16,700
3.1 3.4
4.6 5.7
38.3 44.2
26,300 22,700
Source: Tabulated by author from I 0 tables, as described in text. a Excludes petroleum, and noncomparable products.
312
Susan M. Collins
ports would be expected to be relatively labor using. However, traditional trade theory has implications for the ratio of capital to labor embodied in trade, not for the absolute amount of any single factor. In fact, Leontief‘s paradox (that capital abundant countries have higher capital-labor ratios embodied in their imports than in their exports) has been a standard result in the empirical trade literature. For a review of this literature, see Deardorff 1984. Early resolutions to the paradox have included disaggregation of labor so as to include human capital as a separate factor and special treatment of natural resource industries. More recently, Learner (1980) has shown that it is not inconsistent with theory for capital abundant countries to have higher capital-labor ratios in imports when they are also running trade surpluses. In the 1980 I 0 tables, further disaggregation of imports and exports generates the more intuitive result that imports are labor intensive relative to exports. Both export and import final demands from the I 0 tables include substantial expenditures on commodities that have little or no domestic labor inputs. For example, 27.8% of exports are classified as commodities from the “rest of the world,” including labor remittances. The relevant “industry” uses no domestic labor. Nine percent of imports are from the rest of the world. In addition, 14% are “noncomparable imports,” also with no domestic labor usage, while 26% are on petroleum-related products that use relatively little labor. The total labor requirement per million dollars spent on imports excluding these special categories is 44. This exceeds the labor requirements for consumption and investment expenditures and the comparable figure for exports, which is only 38. Total labor requirements per dollar of expenditure depend on the type of final demand expenditure. It is also interesting to explore how type of expenditure is likely to influence the availability of jobs for immigrants versus nonimmigrants. To do this requires an additional assumption-that the share of immigrants in total industry employment remains relatively constant. Then, if immigrants account for 8% of agricultural employment, an increase of 100 agricultural jobs will generate approximately 8 jobs for immigrants and 92 for nonimmigrants. Using the data on immigrants as a share of total industry employment from table 11.1 in this way, it is possible to split the total labor requirements from column 1 of table 11.3 into immigrant and nonimmigrant. The third and fourth columns of table 11.3 report the number of “immigrant jobs” per million dollars of expenditure on each demand component together with the jobs “held” by immigrants as a percentage of total labor requirements. Because immigrants are not proportionally distributed across industries, the share of total jobs that are likely to be filled by immigrants changes with the component of demand. The last column of table 11.3 shows that the import commodity basket is the one with the highest immigrant labor concentration. Nine point two percent of the labor required to produce the (1980) import
313
Immigrants, Labor Market Pressures, and the Aggregate Demand
basket was likely to have been immigrant. This is not surprising given the large percentages of immigrant workers in apparel, footwear, and other industries with strong import competition. Private consumption expenditures have the second highest immigrant concentration, followed by exports and fixed investment. Government expenditures, particularly on education, come at the other end of the scale because the labor requirements to satisfy these demands are concentrated in industries with relatively small shares of immigrants. Thus, shifts in the composition of final demand will tend to put different labor market pressures on immigrants and nonimmigrants. In particular, an investment boom will generate a larger rise in total labor requirements than a consumption boom, and fewer of those jobs are likely to go to immigrant workers. Overall, a consumption boom will be relatively less beneficial for native workers. However, these aggregate figures mask differences in the intraindustry labor requirements associated with the demand components and provide an incomplete picture of the likely labor market pressures associated with changes in the composition of aggregate demand. The next step is to consider particular aggregate demand vectors and to compare the implied labor requirements, disaggregating both by immigranthathe and by industry. 11.3.3 Alternative Scenarios The analysis below considers four final demand vectors corresponding to four compositions of aggregate demand. These are shown in table 11.4. The first column shows the base case-the actual composition in the 1980 I 0 tables.8 The second column shows a scenario like the 1983-86 period. The
Table 11.4
Composition of Aggregate Demand (percentages of GNP) % A Alt. 1 Trade Deficit Consumption Boom
From Alt. 2 Trade Deficit Investment Boom
1980 Alt. 3 Investment Slump Consumption Boom
63.1
65.6
63.1
65.6
16.1
18.6
14.1
- .3
- .3
12.6 -11.5
16.6 - .3 11.6 - 14.0
11.6 - 14.0
12.6 - 11.5
4.9 2.4
6.5 2.2
5.4 2.6
4.9 2.2
5.2 7.5
5.0 6.8
5.4 7.6
5.1 7.3
1980 Base Type of Demand % Total Consumption Investment: Fixed Inventory Exports Imports Government, federal: Defense Other Federal State & local: Education Other
- .3
Source: Tabulated by author from I 0 tables, as described in text
314
Susan M. Collins
trade balance shifts from surplus to deficit,while private consumption soars. In addition, government spending shifts toward federal defense spending and away from state and local expenditures. The third column shows an alternative scenario with the same trade deficit as in column 2, but an investment instead of a consumption boom. The final column maintains the consumption boom from column 2 but assumes an investment slump instead of a trade deficit. (Inventories are the same share of output in all scenarios.) Because it is misleading to interpret the labor requirements as employment, the results for the three alternative scenarios are presented as percentage changes from the corresponding 1980 base. Table 11.5 provides an overview of the effect of the compositional shifts in demand on labor requirements. Even though total aggregate demand is held constant, total labor requirements rise in the first two scenarios and fall slightly in the third. The increases for trade deficits combined with either a consumption boom (alternative 1) or an investment boom (alternative 2 ) are not surprising given the relatively small labor requirements per dollar of total imports. The labor requirements rises somewhat more in scenario 2 because scenario 1 includes a shift toward defense and away from other types of government spending and because investment requires relatively more labor than consumption per dollar expenditure. This also explains the decline in the third scenario. It is again interesting to decompose the total labor requirements into immigrant and nonimmigrant segments. This is done in the second and third rows of table 11.5. The figures point out a more striking difference between the two trade-deficit scenarios. In the actual 1983-86 combination described in alternative 1, there is little difference between the growth of “immigrant jobs” and “nonimmigrant jobs.” However, alternative 2 , the historically typical combination of high investment and trade deterioration, implies a substantially larger expansion of “nonimmigrant jobs” than “immigrant” jobs. Similarly, a switch from investment to consumption with no trade change, as in alternative 3, implies an increase in jobs typically held by immigrants and a decline in jobs typically held by natives. To the extent that workers in declining sectors are laid off more quickly than workers are hired in expanding sectors, any demand shift that alters the secTable 11.5
Demand Shifts and Total Labor Requirements % A Alt. 1 Trade Deficit Consumption Boom
From Alt. 2 Trade Deficit Investment Boom
Immigrant Native
1.22 1.27
1.12
.I6
1.81
- .33
Total
1.27
1.75
- .30
1980 Alt. 3 Investment Slump Consumption Boom
315
Immigrants, Labor Market Pressures, and the Aggregate Demand
toral distribution of labor requirements will tend to cause short-term unemployment, even if aggregate labor requirements have increased. A key implication of table 11.5 is that native workers will have a relatively easier adjustment to expanding imports when the trade deficit is associated with an investment boom. The very different distribution of jobs when the trade deficit coincides with a consumption boom places native workers more directly in competition with immigrants for the jobs in expanding sectors. The next step is to look at the distribution of the change in job requirements implied by each scenario across industries. Table 11.6 examines the sectoral decomposition of total labor requirements for all sectors and nondurable manufacturing where immigrants are overrepresented. The first column gives the share of each sector in the total labor requirements in the 1980 base. Columns 2-4 give the percentage change in sectoral labor requirements for each of the three alternative scenarios. The consumption boom-trade deficit alternative leads to a reduction of labor requirements in manufacturing but to increases in service and government sectors. (The decline in manufacturing, with its large concentration of immigrants, is offset by a spurt in household services and other particular industries.) In contrast, the investment boom-trade deficit scenario leads to increased labor requirements in manufacturing but slower growth in the service sectors. An important implication from the top panel of table 1 1.6 is that we should expect a trade deficit cum consumption boom to shift employment from manufacturing to service sectors, but it is incorrect to conclude that such an employment shift is “caused” by an expanding trade deficit. A trade deficit of equal magnitude (alternative 2) cum investment boom will tend to shift employment toward manufacturing and construction and away from services, finance, insurance, and real estate. The analysis has identified two sets of labor market pressures that are likely to arise from a consumption boodtrade imbalance but unlikely to be present in the more standard investment boodtrade imbalance combination. The first is that natives are relatively less concentrated in the sectors and industries with large increases in labor requirements. The second is that labor requirement decreases will be concentrated in the manufacturing sector and therefore more likely to increase pressures for protectionism. The bottom panel of table 11.6 presents further information about the change in labor requirements in (durable and nondurable) manufacturing under each alternative to make a third point. Within manufacturing, many of the industries with large declines under the consumption boom-trade deficit scenario also have high concentrations of immigrants. In particular, leather, footwear, and plastics, three of the industries suffering the largest losses, have immigrants accounting for 9.5%-16.5% of their work forces. Immigrants are extremely “visible” in these industries. Similarly, of the three industries en-
316 Table 11.6
Susan M. Collins Changes in Labor Requirementsin All Sectors and in Nondurable Manufacturing 1980 % A Alt. 1 Base Trade Deficit % Total Consumption Boom
All sectors: Agriculture Mining Construction Manufacturing Transportation, communication Wholesale & retail trade Finance, insurance, real estate Services Government Total Nondurable manufacturing: Food & kindred products Tobacco manufacturers Fabric Textile mill products Apparel & other textile products Miscellaneous textile products Paper & allied products Paperboard containers & boxes Pnnting & publishing Chemical & allied products Plastic materials & synthetics DNgS Paints & allied products Petroleum & coal products Rubber & miscellaneous plastic products Leather products Footwear except rubber products Total
1.95 .80 6.39 20.87
From Alt. 2 Trade Deficit Investment Boom
- .44 - 7.04 - .78 - .53
-2.78 -6.56 8.60 .54
1980 Alt. 3 Investment Slump Consumption Boom 2.53
- .69
-6.72 - 1.85
5.08 16.60
1.32 2.35
1.54
.63 .79
5.12 24.26 18.94 100.00
2.46 2.35 1.28 1.27
.15 .26 4.46 I .75
2.61 2.42 -2.14 - .30
20.60 .84 6.64 1.38
1.59 1.32 -2.01 - 1.02
- 1.80 - 1.94 - 4.40 .90
3.50 3.26 2.25 - 1.74
15.86
- .28
-4.77
4.31
1.89 6.24
.33 -2.36
- 1.62 -2.88
1.94 .92
2.53 15.09
.08 1.18
- .85 1.02
.93 .70
.55
6.36
-3.93
-4.11
- .03
2.68 3.78 .74
- 2.92 .23 - .27
-3.03 - 1.86 3.35
- .08
3.86
- .50
-2.44
8.36 .28
- 1.75 -8.08
- 1.25 - 12.67
- .70
2.86 100.00
-5.70 .I0
- 11.01 -2.20
5.31 1.92
Source: Tabulated by author from I 0 tables, as described in text.
2.64 -3.19 1.43
4.57
317
Immigrants, Labor Market Pressures, and the Aggregate Demand
joying the largest increases, two (tobacco and printing) have relatively small concentrations of immigrants, while the third (food) is about average.
11.4 Concluding Remarks This paper has made two major points. First, changes in the composition of aggregate demand will shift the distribution of labor requirements across industries and sectors. It will also affect total labor requirements. Thus, it is important to identify changes in the components of domestic absorption when analyzing the likely labor market consequences of aggregate demand shifts. By the same token, it is inappropriate to examine changes in imports or exports in isolation. An increase in imports must coincide with an offsetting shift in some other component(s) of demand, and a consumption boom is not equivalent to an investment boom or to a rise in government expenditures. Instead, the total and the sectoral distribution of labor requirements will depend critically on which has occurred. The analysis showed that the 1980s consumption boom with trade deficit has implied that decreased labor requirements were concentrated in the manufacturing and construction sectors. In contrast, an investment boom with trade deficit would tend to raise labor requirements in both these sectors. Therefore, it is incorrect to ask whether trade deficits tend to“deindustria1ize” the economy by shifting employment from manufacturing to services because the answer also depends on the changes in other components of aggregate demand. The second set of points concerns the role of immigrant workers versus native workers. Immigrants are not evenly distributed across sectors but are concentrated in some manufacturing industries and in private household services. Thus, shifts in the distribution of labor market requirements should have quite different short-run implications for immigrants than for native workers. The analysis showed that an investment boom cum trade deficit implies that requirements for jobs typically held by natives will increase nearly twice as quickly as requirements for jobs typically held by immigrants. Immigrants are relatively better off in a consumption boom, in which case there is little difference between the job growth rates for the two groups. Furthermore, some of the industries with the largest declines under the consumption boom are also the ones with the heaviest concentrations of immigrants. Both these factors may tend to make immigrants more “visible” in the labor market and help explain the recent increase in concern over their presence.
Notes 1 . Another possible explanation, an increase in the number of legal and/or illegal immigrants entering the country, is discussed in other papers in this volume. 2. This approach is relevant for short-run analysis only. Over time, both immigrants and native workers will presumably move from contractingto growing sectors. 3. This issue is discussed in Sehgal(l985) and in Borjas (1987).
318
Susan M. Collins
4. For additional discussion of these data, see Sehgal(l985). 5. Net other international transactions are capital grants, net transfers, and interest payments. This term also includes the statistical discrepancy. 6. For further description of these data, see “The Input-Output Structure of the U.S. Economy, 1977” in Survey of Current Business (May 1984). 7. These data are from the Survey of Current Business, November 1985 and May 1986. 8. These data do not correspond exactly to the figures in table 1 1.1 because output is measured in producer prices in the I 0 tables but in consumer prices in the National Income and Product Accounts.
References Abowd, J. M., and R. B. Freeman. 1990. Internationalization of the U.S. Labor Market. NBER Working Paper no. 3321. Cambridge, Mass.: National Bureau of Economic Research. Aho, C. M., and J. A. Orr. 1981. The Growth of Trade-Sensitive Employment. Monthly Labor Review 104; no. 2(February):28-35. Borjas, G. 1987. Self-Selection and the Earnings of Immigrants. American Economic Review 77, no. 4(September):531-53. Branson, W. H . , and J. P. Love. 1986. Dollar Appreciation and Manufacturing Employment and Output. NBER Working Paper no. 1972. Cambridge, Mass.: National Bureau of Economic Research. DeardorfS, A. V. 1984. Testing trade theories and predicting trade flows. In Handbook of International Economics, vol. 1, eds. R. W. Jones and P. B. Kenen. Amsterdam: North-Holland. DeardorE, A. V., and R. M. Stem. 1986. The Michigan Trade Model of World Production and Trade: Theory and Applications. Cambridge, Mass.: MIT Press. Dickens, W. T., P. Shapiro, L. Tyson, and J. Zysman. 1985. The Employment Effects of International Trade: A Review of the Literature. Department of Economics, University of California, Berkeley. Mimeo. Grossman, G. M. 1986. Imports as a Cause of Injury: The Case of the U.S. Steel Industry. Journal of International Economics 20, nos. 3/4(May):201-24. . 1987. The Employment and Wage Effects of Import Competition in the United States. Journal of International Economic Integration 2, no. l(Spring): 1-23. Krueger, A. 0. 1979. The Impact of Foreign Trade on Employment in U.S. Industry. In Current Issues in Commercial Policy and Diplomacy, ed. J. Black and B . Hindley. London: Macmillan. . 1980. Protectionist Pressures, Imports and Employment in the United States. Scandinavian Journal of Economics 82(2): 133-46. Learner, E. 1980. The Leontief paradox reconsidered. Journal of Political Economy 90, no. 3 (June): 495-503. Orr, A. C., and J. A. Orr. 1983. Employment Adjustments in Import Sensitive Manufacturing Industries, 1960-80. Department of Economics, Manhattan College. Mimeo. Schoepfle, C. K. 1982. Imports and Domestic Employment: Identifying Affected Industries. Monthly Labor Review 105, no. 8 (August): 13-26. Sehgal, E. 1985. Foreign Born in the U.S. Labor Market: The Results of a Special Survey. Monthly Labor Review 108, no. 7 (July): 18-24. U.S. Trade Related Employment, 1978-84. 1986. USITC Publication no. 1855, May. Washington, D.C. : United States International Trade Commission.
111
Comparative Experiences: Canada and Australia
This Page Intentionally Left Blank
12
An Analysis of the Earnings of Canadian Immigrants David E. Bloom and Morley Gunderson
The purpose of this paper is to analyze immigrant labor market progress in Canada-a country that, as a matter of official policy, has screened most of its immigrants on the basis of their expected “ability to assimilate.” In particular, we compare earnings profiles for Canadian immigrants and natives and seek to determine whether immigrant earnings profiles reflect any “vintage effects” associated with year of immigration. Over the past ten years, a number of studies of immigrant earnings have focused on these same issues using data for U.S. immigrants. Among the best known is that by Chiswick (1978), which fits a standard wage equation to cross-sectional data on immigrants and natives in the 1970 Public Use Sample of the U.S. Census. Chiswick’s results support the conclusion that, when they first enter the labor market, immigrants earn approximately 25 percent less than natives with comparable years of schooling and experience, marital status, etc. However, Chiswick also finds that immigrants have steeper experience-earnings profiles than “comparable” natives, with immigrant earnings overtaking native earnings within roughly thirteen years of their entry into the United States. A number of other studies have fit the same basic model to similar data and have reached roughly identical conclusions (see, e.g., Carliner 1980; Long 1980; and Borjas 1982). David E. Bloom is professor of economics in the Department of Economics at Columbia University and a research associate of the National Bureau of Economic Research. Morley Gunderson is professor of economics and director of the Centre for Industrial Relations at the University of Toronto. An earlier version of this paper was presented at the NBER Conference on Trade, Immigration, and Labor in September 1987. The authors are indebted to McKinley Blackbum, Anne Hill, Robert Komfeld, David Neumark, Andrew Newman, and Marcus Rebick for assistance in the preparation of this paper and to Blackbum, Rebick, Richard Freeman, Mark Killingsworth, Jacob Mincer, and Glenn Withers for helpful discussions and comments. This research was supported by NIH grant HD18844-02 and by a grant from the Ford Foundation to the National Bureau of Economic Research.
321
322
David E. Bloom and Morley Gunderson
The set of findings based on Chiswick’s approach to measuring immigrant assimilation has been challenged by Borjas (1985), who argues that the steepness of immigrant earnings profiles is inflated by cross-cohort declines in immigrant quality. Evidence supporting this argument is provided by using pooled data from the 1970 and 1980 U.S. Population Censuses to measure earnings growth in the intercensal period for individual entry cohorts of immigrants. On the basis of this analysis, Borjas concludes that “cross-section studies of immigrant earnings provide useless and misleading insights into the process of immigrant assimilation into the labor market” (p. 485). Borjas’s conclusion deserves further examination. A priori theoretical reasoning is perhaps more consistent with Chiswick’s empirical conclusions than with those of Borjas. Low entry wages for immigrants can plausibly be explained as a loss of (origin) country-specific human capital; rapid earnings growth can be viewed as reflecting positive self-selection into immigration (i.e., immigrants are above average in terms of their aggressiveness, ambitiousness, willingness to work hard, etc.). In contrast, sizable cross-cohort declines in immigrant quality are somewhat harder to accept given that it is not overall quality that is hypothesized to have declined but rather that component of overall quality that is unmeasured (i.e., the part of immigrant quality that is not measured by or correlated with variables such as schooling, experience, marital status, country of origin, etc.). Borjas’s results, as he recognizes, may also reflect differential patterns of underenumeration in the successive Censuses or nonrandom intercensal mortality and out-migration. Indeed, out-migration, death, or undercounting of immigrants who are relatively unsuccessful in the labor market are all alternatives to declining immigrant quality as an explanation of Chiswick’s cross-sectional results. We will also use the example of Canada as an opportunity to gain some insight into the importance of intercensal exiting from an immigrant population. Although there is little information on either the covariates of immigrant mortality or on differential Census undercounting of immigrants, there are several established lines of inquiry on the subject of out-migration. According to a group of imperfect information models, out-migration is an event that was unplanned ex ante and that occurs primarily among migrants whose labor market expectations are not satisfied (see Yezer and Thurston 1976; Allen 1979; Blejer and Goldberg 1980; and Lam 1986). These models suggest, at the margin, that out-migrants will tend to be selected from the lower end of the earnings distribution. On the other hand, intertemporal substitution models tend to view out-migration as a planned event among individuals who make short-term moves in order to take maximal advantage of temporarily favorable earnings opportunities (Stark and Bloom 1986; Fox 1987). These models suggest that out-migration will be most prevalent among individuals who are selectively active and successful in the labor market. Although the results are far from definitive, empirical research by Jasso and Rosenzweig (1987, 1989) and by Lam (1987) tends to support this view insofar as out-
323
An Analysis of the Earnings of Canadian Immigrants
migration of U.S. and Canadian immigrants is reported to be most prevalent among those who are relatively successful. Although they are extremely different in spirit, both the imperfect information and the intertemporal substitution models of out-migration share an important empirical implication, namely, that the variance of residuals in a migrant earnings equation will decline with duration of stay (i.e., under the imperfect information models, exit occurs at the lower end of the distribution, while, under the intertemporal substitution models, exit occurs at the upper end). In contrast, job matching or asymmetric information models imply that the residual variance in a wage equation will increase with duration of stay as employers are increasingly able to observe the true productivity of migrants (see Harris and Holmstrom 1982; Katz and Stark 1984). We attempt to infer which set of forces tends to be stronger by examining patterns in the variance and kurtosis of immigrant earnings by duration of stay. For example, we will interpret an increase (decrease) in the variance of earnings with duration of stay as evidence favoring the relative importance of the job-matching models (imperfect information models). Thus, we have four main goals in this paper. First, by fitting the models proposed by both Chiswick and Borjas to data for Canada, we hope to assess the extent to which it is generally true that cross-sectional studies of immigrant earnings are “useless and misleading.” Second, we hope that estimates of these alternative models will lead to clear substantive conclusions regarding the shape of immigrant earnings profiles and the importance of entry-cohort effects on earnings. Third, by comparing corresponding results under different Canadian immigration policies, we hope to shed some light on the significance of a nation’s institutions in determining the economic benefits of immigration. Finally, by analyzing the variance of immigrant earnings by duration of stay in Canada, we hope to assess the relative importance of selective intercensal exiting and job matching/asymmetric information in models of the labor market progress of immigrants. 12.1 Immigration Policy and Immigrants in Canada
In an effort to enrich our interpretation of statistics related to the labor market experience of Canadian immigrants, this section will present a brief review of the history of Canadian immigration policy and of immigration to Canada. 12.1.1 A Brief History of Canadian Immigration Policy From the nineteenth to the twentieth century, international migration to developed countries has been determined less and less by events and decisions of individuals in countries of origin and more and more by regulations established in countries of destination. In this regard, Canada is no exception. Until 1869, Canada’s immigration policy was simply one of free entry. But, begin-
324
David E. Bloom and Morley Gunderson
ning that year, a series of legislative enactments established specific principles of selection and associated regulatory apparatus. Prohibitions were established on the landing of “criminals and other vicious classes” in 1872, paupers and destitute immigrants in 1879, and diseased persons in 1902. In 1904, an exorbitant head tax of $50 (Canadian) was established for Chinese immigrants. During these years, the central government also set up quarantine stations, specified legal responsibilities for companies involved in transporting immigrants, and began to require those companies to make deposits into a fund whose purpose was to cover the expenses of indigent immigrants before they were able to secure employment. The basic structure of Canadian immigration policy during much of the first half of the twentieth century was set forth in the Immigration Act of 1910. This act firmly established the principle of selective immigration by creating a proscribed class of immigrants: those “deemed undesirable because of climatic, industrial, social, educational, labour, or other conditions or requirements of Canada, or deemed undesirable because of their customs, habits, modes of life and methods of holding property and their probable ability to become readily assimilated.” In practical terms, this act led to a distinction between countries in the extent to which they were considered to be “preferred” or “nonpreferred.” The two most preferred countries were the United Kingdom and the United States (and France as of 1947). They were followed by several other countries in northern and western Europe that were “not too different [from Canada] in language and mode of life.” Countries in central and eastern Europe were considered to be nonpreferred, with the most nonpreferred countries being Greece, Italy, Syria, and Turkey. Immigrants from Asian countries were considered so undesirable that their admission was strictly regulated under separate acts. Subject to time-varying restrictions on total immigrant volume, applicants from the most preferred countries were admitted on almost a laissez-faire basis, while the admission of immigrants from other preferred countries depended to varying degrees on whether they possessed training and skills for which there was a need in Canada. Only immigrants in a relatively narrow range of occupations (e.g., agriculture) were admissible from nonpreferred countries, and the range of relatives they could bring with them was quite limited. One of the chief characteristics of twentieth-century immigration policy in Canada is its strong labor market orientation. In a broad statement outlining the principles that have guided Canadian immigration policy throughout the post-World War I1 era, Prime Minister MacKenzie King declared in 1947 that Canada would encourage immigration to meet its need for population. He said further that Canada would accept as many immigrants “as could be advantageously absorbed into the national economy,” with the admissibility of potential immigrants to Canada depending on, among other factors, labor conditions and requirements in Canada and each applicant’s “ability to assimilate.”
325
An Analysis of the Earnings of Canadian Immigrants
King also affirmed the discriminatory features of Canada’s immigration policy, stating that “the people of Canada do not wish, as a result of mass immigration, to make a fundamental alteration in the character of our population. . . . Canada is perfectly within her rights in selecting the persons whom we regard as desirable future citizens. It is not a ‘fundamental human right’ of any alien to enter Canada. It is a privilege. It is a matter of domestic policy.” Because control over the volume of immigrants to Canada and over their national and occupational composition resided in the hands of the Cabinet, immigration policy in Canada has been remarkably responsive to a variety of social, economic, and political situations throughout most of this century. For example, immigration was tightly restricted during the high unemployment years of the 1930s; immigrants were not accepted from Japan, Germany, or Italy during World War 11, although many displaced Europeans were admitted from other countries; and Canada actively assisted and accepted many immigrants from Hungary during 1957. Canadian immigration policy has often been referred to as a “tap-on, tapoff policy” because of its flexibility and its responsiveness to contemporary labor market concerns. For example, the admission of immigrants was increased sharply as a response to labor shortages in the 1950s but was curtailed during the years 1958-62 because of high rates of unemployment. Beginning in the 1950s, immigration officials treated professionals and entrepreneurs with capital quite favorably because of their potential to generate employment opportunities in Canada. Indeed, Canada abandoned its policy of national discrimination in the 1960s partly because it became increasingly clear that Canada would not be able to satisfy its need for skilled manpower via immigration from its list of preferred countries. In 1967, Canada substantially altered the mechanisms by which it administered its immigration policies. First, it eliminated all discrimination on the basis of race or nationality. Second, it defined four classes of immigrant applications: ( I ) sponsored relatives (i.e., dependent relatives); (2) nominated relatives; (3) independent applications; and (4) refugees. Sponsored relatives would be admissible merely if they could demonstrate that they were in good health and of good character. Refugees, a status defined by the United Nations, would be accorded preferential treatment in admission. Finally, nominated relatives and independent applications would be judged on the basis of a point system. The two key features of the point system are that it removed a good deal of subjective authority from the hands of immigration officers and assigned considerable weight in admissions decisions to labor market-related factors. In order to be admitted under the point system, an immigrant needed to receive at least fifty points out of a maximum of one hundred. Points were awarded according to the following nine criteria, with some minor differences in the evaluation of independent applications and applications from nominated relatives:
326
David E. Bloom and Morley Gunderson
1. Education and training: One point for each year of successful formal education or occupational training, up to a maximum of twenty; 2. Personal characteristics: Up to fifteen points awarded at the discretion of immigration officers on the basis of their perception of the applicant’s adaptability, resourcefulness, initiative, and motivation; 3. Occupational demand: Up to fifteen points, for both skilled and unskilled workers; 4. Occupational skill: Ranging from one point for unskilled workers to ten points for professionals; 5 . Age: Ten points for applicants below the age of 35, with one point less for each year above age 35 (with a minimum of zero points); 6. Arranged employment: Ten points for applicants with a definite job in Canada; 7. Knowledge of French and English: Up to ten points depending on an applicant’s fluency in French and English; 8. Relatives: Up to five points for applicants with relatives in Canada that could help them get established; 9. Employment opportunities: Up to five points for applicants moving to areas of strong labor demand.
The point system was amended in 1974, as a response both to the large number of immigrants admitted to Canada in 1972 and 1973 and to increases in the unemployment rate in Canada. A priority system was established for processing immigrant applications that gave preferential treatment to applicants with close relatives in Canada, to applicants with prearranged employment in high-demand occupations, and to entrepreneurs and refugees. A “Canadians-first” policy was also established under which applicants would receive no credit for prearranged employment unless they could show that no equally qualified Canadian citizen or landed immigrant was available to fill the position. In addition, an applicant would lose ten points if there was no evidence of prearranged employment or bona fide demand for their labor. The figures in table 12.1 indicate that a sharp increase occurred in the proportion of Canadian immigrants admitted on the basis of family ties following the 1974 policy changes. 12.1.2 Trends and Patterns in Immigration to Canada The foreign born have constituted a sizable fraction of the Canadian population throughout the twentieth century. In 1901, 13.3 percent of the Canadian population was foreign born. This fraction increased sharply during the first decade of the century and hovered around 22 percent into the 1930s, when difficult economic circumstances led to restrictive immigration policies that caused it to decline. Nonetheless, the foreign-born fraction of the Canadian population had not fallen below 15 percent through the early 1980s (see table 12.2).
327
An Analysis of the Earnings of Canadian Immigrants
Table 12.1
Immigration to Canada by Category of Admission % Admitted as Sponsored or
Assisted Relatives
Year 1954-58 1960-64 1965-69 1970-74 1975-79 1980-84
Total Number of Nonrefugee Immigrants 839,045 456,143 909,882 785,079 593,862 468,731
Total
Sponsored Relatives
Assisted Relatives
% Admitted from Independent Applications (including refugees)
33.0 44.6 37.8 49.3 67.6 64.2
NA NA NA 24.7 45.2 54.2
NA NA NA 24.6 22.4 10.0
67.0 55.4 62.2 50.7 32.4 35.8
Source: Employment and Immigration Canada, Annual Report to Parliament on Immigration Levels, selected years. Note: NA, not available.
Table 12.2
Foreign-born Members of the Canadian Population, Stocks and Flows Total Canadian Population (in millions)
% Foreign Born
5.4 7.2 8.8 10.4 11.5 14.0 18.2 21.6 24.1
13.0 22.0 22.3 22.2 17.5 14.7 15.6 15.3 16.1
Stocks: 1901 1911 1921 1931 1941 1951 1961 1971 1981
Flows: 1901-1 I 1911-21 1921-31 1931-41 1941-51 1951-61 1961-7 1 1971-81
Population Increase (thousands)
Number of Immigrants (thousands)
Ratio of Immigrants to Population Increase
1,836 1,581 1,589 1,130 2,502 4,229 3,330 2,5 15
1,759 1,612 1,203 150 548 1,543 1,429 1,447
.96 1.02 .16 .13 .22 .37 .43 .58
Source: Author calculations based on data reported in Immigration Statistics, 1983(Ottawa: Supply and Services, 1985).
328
David E. Bloom and Morley Gunderson
In order to maintain such a high fraction of foreign born among the Canadian population, immigration flows into Canada have been quite substantial. For example, there were 4.4 million immigrants to Canada from 1951 to 1981, a period during which the population of Canada increased from fourteen to twenty-four million. Although the ratio of new immigrants to the overall increase in the size of Canada’s population has been above 15 percent during every year in the post-World War I1 era, there has been a great deal of year-to-year variation in the number of immigrants. Especially large numbers of immigrants arrived in Canada in 1951 (194,391), 1957 (282,164), 1967 (222,876), and 1974 (218,465); in contrast, relatively few immigrants arrived in 1946-47 (roughly 68,000 immigrants per year), 1961-62 (roughly 73,000 immigrants per year), 1978 (86,300), and 1983 (88,800). From the mid-1950s to the mid-l970s, the percentage of Canadian immigrants intending to enter the labor force was just above 50 percent. That figure dropped to 44 percent starting in the mid-l970s, as the number of admissions from independent applications dropped from nearly 110,000 in 1974 to under 21,000 in 1984. Even more dramatic has been the shift in the distribution of occupations among immigrants expecting to enter the labor force. This shift has been notably in the direction of increased skill and training. Among immigrants entering Canada during 1954-58, only 12 percent listed their intended occupations as managers or professionals. In contrast, 37 percent listed agricultural worker, laborer, or service worker as their intended occupation. During the years 1979-83, the percentage of managers and professionals increased to 28 percent, while only 14 percent of immigrants reported Table 12.3
Distribution of Intended Occupations among Canadian Immigrants Planning to Work, by Year of Immigration (70)
Year of Immigration Occupation Managerial Professional C1erica1 Service Agriculture Construction Manufacturing and mechanical Laborers Other Total
1979-83
1974-78
1969-73
1964-68
1959-63
1954-58
6.3 21.5 11.4 8.2 4.1 4.2
7.5 22.5 14.7 9.1 2.2 7.1
5.1 26.1 15.0 11.1 3. I 6.1
2.4 25.3 13.5 9.6 3.2 8.8
2. I 17.3 11.9 16.5 7.5 7.1
10.2 14.4 9.2 9.8
19.0 1.3 24.0
22.2 1.5 12.6
20.9 2.4 9.0
23.9 7.2 6.1
18.4 12. I 6.5
21.8 13.9 8.5
100.0
100.0
100.0
100.0
100.0
100.0
1.4 10.8
Source: 1954-73: Employment and Immigration Canada, Annual Report to Parliament on Im-
migrurion Levels, 1980, 14. 1974-83: Author calculations based on information reported in annual issues of Immigration Srutisrics (Employment and Immigration Canada). a Includes transportation and communication, commercial and financial, logging, fishing, trapping and hunting, mining and quarrying, and unspecified.
329
An Analysis of the Earnings of Canadian Immigrants
that they intended to work as agricultural workers, laborers, or service workers. While some portion of these changes undoubtedly reflects sectoral shift in the Canadian economy, the bulk of the changes reflect the increased emphasis on skill and training in Canada's immigration policy (see table 12.3). Table 12.4 presents a cross-tabulation of the foreign-born population of Canada by country of origin and year of immigration. The data are taken from the 1981 Canadian Census. The figures clearly show that British and American immigrants dominated the immigration flow to Canada before 1946 (i.e., in 1981, 61 percent of all pre-1946 immigrants in Canada were from the United Kingdom or the United States). That dominance ended immediately following World War I1 as immigration from Europe (excluding the United Kingdom) expanded sharply. From 1946 to 1955,68 percent of all immigrants to Canada were from Europe (excluding the United Kingdom), up from just 37 percent prior to 1946. Germany, Italy, and the Netherlands alone supplied an especially large proportion of immigrants in the ten years following the war (36 percent). Even in absolute terms, no European country increased the number of immigrants it supplied to Canada during the postwar period, whereas the number of immigrants from most European countries actually declined (i.e., based on numbers of immigrants actually in Canada in 1981). Table 12.4
Canadian Immigrants by Country of Origin and Year of Immigration, 1981 Year of Immigration
Country of Origin
1976-80
1971-75
Africa Asia Belgium/ Luxemburg Britain France Germany Greece Ireland Italy Latin America Netherlands Other Europe' Other NonEuropeb Poland Soviet Union United States
233 1,975
390 1,720
20 690 65 73 36 20 86 793 53 399
Total
1966-70
1956-60
255 889
106 245
39 209
16 20 1
14 75
1,053 5,314
12 874 67 92 139 26 182 1,080 46 778
25 1,209 122 213 229 29 703 633 999
10 646 67 176 169 12 718 149 64 456
55 1,104 69 473 176 31 1,068 95 255 869
72 1,833 102 764 84 19 1,030 68 848 848
34 2,324 19 84 18 23 127 35 31 718
228 8,680 511 1,875 851 160 3,914 2,853 1,387 5,067
89 58 80 394
113 59 34 529
95 78 28 466
31 87 21 191
25 121 49 155
29 54 1 614 200
12 473 476 1,007
394 1,417 1,302 2,942
5,064
6,141
6,063
3,148
4,793
7,269
5,470
37,948
90
1946-55 pre-1946
Total
1961-65
Source: 1981 Canadian Census of Population, 1/100 sample. Data include all immigrants (place of birth other than Canada) except inmates, members of the armed forces, and immigrants who arrived in 1981. a Includes Spain, Portugal, Scandinavia, and non-Soviet and Eastern Bloc countries. Includes Australia, Pacific Islands, and other areas not otherwise listed.
330
David E. Bloom and Morley Gunderson
Table 12.4 also shows that the pattern of immigration to Canada changed rather dramatically when Canada stopped discriminating among immigrants on the basis of country of origin. For example, between the first and second half of the 1960s, immigration to Canada from Asia and Latin America increased nearly fourfold. Although these regions of the world supplied only 2 percent of Canada’s immigrants prior to 1946, they supplied 46 percent from 1971 to 1975 and 55 percent from 1976 to 1981. 12.1.3 Immigrants in the Canadian Labor Market We next present a brief descriptive analysis of the employment, unemployment, and earnings experience of male immigrants represented in the 1971 and 1981 Canadian Population Censuses. Table 12.5 reports selected labor market characteristics of immigrants and natives based on data contained in the 1981 Canadian Census of Population. Judging merely on the basis of labor market activity measures, it would not be unreasonable to conclude that Canadian immigrants are well assimilated in the labor market. The labor force participation rate of male immigrants (aged fifteen and over) was 72.4 percent in 1981, just slightly below the rate of 73.6 percent for native Canadians. The closeness of native and immigrant labor supply extends beyond labor force participation rates to hours and weeks worked as well. Ninety-three percent of employed immigrant males worked thirty-five or more hours during the 1981 Census reference week, compared to 92 percent of employed native males. Similarly, 70.6 percent of the male immigrants reported having worked forty-nine to fifty-two weeks in 1980, just .2 percentage points higher than the figure for native males. Despite the closeness in these measures of immigrant and native labor supply, unemployment rates for immigrants were notably lower than for natives in 1981 (i.e., among males, the unemployment rates were 8.5 percent for natives and 5.3 percent for immigrants). These differentials could reflect a variety of factors, including differences in reservation wages, human capital, and demographic composition (for an analysis of immigrant labor supply and unemployment in Table 12.5
Selected Labor Market Characteristicsof Immigrants and Natives Aged Twenty-five to Sixty-four in 1981
Labor force participation rate (ages 15 and over, %) % Who worked 49 or more weeks in 1980 % Working 35 or more hours during the Census reference week % Self-employed Unemployment rate (%) Average wage and salary income in 1980 among those not primarily self-employed (in thousands of dollars) Source: Authors’ tabulations of 1981 Census data
Immigrants
Natives
12.4 70.6 93.3
73.6 70.4 91.9
15.3 5.3 20.4
13.3 8.5 19.0
331
An Analysis of the Earnings of Canadian Immigrants
Canada, see Fox 1987). Presumably, such factors also underlie the explanation of the difference in average income between male immigrants and natives (i.e., the immigrants had a 7.4 percent advantage). It is also worth noting that rates of self-employment are slightly higher among immigrants than among natives, with 15 percent of immigrant males reporting that they were selfemployed in 1981 (compared to 13 percent for native males). Table 12.6 compares labor force participation rates, unemployment rates, and levels of average income among different entry cohorts of Canadian immigrants-using data from both the 1971 and the 1981 Canadian Censuses. The statistics show that the more recent immigrants have relatively low labor force participation rates, relatively low average income, and relatively high unemployment rates. Labor force participation rates are also relatively low in both Censuses for pre- 1946 immigrants, presumably because many immigrants in that cohort had reached retirement age by 1971 and 1981. Although labor force participation rates are quite flat across the cohorts of immigrants that entered Canada between 1946 and 1975, it does appear that unemployment rates are higher for the more recent cohorts. Average immigrant earnings also tend to be lower for immigrants in the more recent entry cohorts. Thus, while there do not seem to be major differences in the employment and earnings experiences of immigrants and natives in Canada, there are notable differences between immigrants in different entry cohorts. Whether these differences represent genuine vintage effects or simply reflect the influence of immigrant labor market characteristics or other variables cannot be determined from these tables. Making such a determination requires that we control for a variety of variables in a multivariate manner, which we turn to in the following section. Table 12.6
Immigrant Labor Force Participation, Unemployment, and Income, by Year of Immigration, for Males in the 1971 and 1981 Censuses Labor Force Participation Rate
Unemployment Rate (%)
Average Income (in thousands of current dollars)
1981
1971
1981
1971
1980
1970
Pre- 1946 1946-55 195MO 1961-65 1966-70 1971-75 1976-80
29 77 84 84 82 81 73
47 89 86 85 81 NA NA
4.4 3.6 3.9 5.3 6.1 6.7 8.0
4.6 4.1 5.0 5.3 6.1 NA NA
23.4 22.9 21.1 19.9 21.1 18.2 15.0
18.8 17.5 17.0 15.9 14.8 NA NA
Total
72
74
5.3
4.9
20.4
16.8
Year of Immigration
Source: Author calculations using 1981 and 1971 Census data. Nore: NA, not applicable.
332
David E. Bloom and Morley Gunderson
12.2 Empirical Analysis of Immigrant Earnings Profiles In this section, we will analyze earnings patterns among Canadian immigrants using data contained in the 1971 and 1981 Canadian Censuses. Our goal is to answer the following three questions. (1) On average, do employed immigrants receive higher wages than employed natives who are comparable in terms of observed productivity-related characteristics? (2) On average, do employed immigrants who have been in Canada for a total of X years receive higher or lower wages than employed immigrants who have been in Canada for X + Y years but who are otherwise comparable in terms of observed characteristics? (3) Does the dispersion of immigrant earnings tend to vary with duration of stay?
12.2.1 Empirical Models and Data Issues The standard model used to compare earnings profiles for immigrants and natives was proposed by Chiswick (1978). The basic regression model, which is fit to cross-sectional data for a pooled sample of both immigrants and natives, is a simple extension of the standard human capital earnings function: (1)
log Y = a,
+ a,(SCH) + a,(EXP) + a,(EXPSQ)
+ a,(IMMIG) + a,(YSM),
where Y is earnings, SCH is years of schooling, EXP is years of labor market experience, EXPSQ is years of labor market experience squared, IMMIG is an indicator variable for immigrants, and YSM is years since migration interacted with the immigrant dummy variable. The estimate of a, measures the average percentage difference between the earnings of natives and newly arrived, but otherwise comparable, immigrants. The estimate of a5measures the average percentage increase in immigrant earnings with each year that immigrants spend in their new home country, beyond the increase in earnings associated with the fact that their human capital stock may have changed during that year (e.g., EXP may have increased). Thus, a positive estimate of a, has been taken to indicate that the average experience-earnings profile of immigrants is steeper than that of natives, which is suggestive of labor market progress and assimilation. Borjas (1985) has recently pointed out that interpreting the coefficients in equation (1) in this manner requires one to assume that there are no omitted variables that are correlated with YSM. This assumption may be difficult to defend because YSM also measures “date of entry into the new country” in a cross-sectional regression. If unmeasured factors relevant to labor market success vary systematically across entry cohorts of immigrants, the coefficient a5 will measure both immigrant labor market progress and the effect of the average difference in unmeasured factors across successive entry cohorts (i.e., it may be a biased measure of the labor market progress experienced by different entry cohorts over time).
333
An Analysis of the Earnings of Canadian Immigrants
The most straightforward way to overcome the fact that YSM is a linear combination of a vector of year-of-immigration dummy variables in crosssectional data is to make use of data that follow cohorts over time. Since such data provide observations on each entry cohort at two or more points in time, it is possible to estimate the effect of time spent in the new country on earnings without the potentially confounding influence of entry cohort effects (i.e., a regression model can be specified with year-of-immigration dummy variables and YSM on the right-hand side because the same individual in a particular entry cohort, with an immutable “year of entry,” will have different values of YSM when he or she is observed at different points in time). To our knowledge, there are no longitudinal data for Canada that are suitable for conducting such an analysis. Thus, following Borjas, we will construct a cohort data set for different entry cohorts of immigrants using data contained in the 1971 and 1981 Canadian Population Censuses. We will fit the following regression model to pooled data from these two Censuses: (2)
+ b,(SCH) + b,(EXP) + b,(EXPSQ) + b,(IMMIG) + b,(YSM) + c,(COH,) + . . . + c,(COH,),
log Y = b,
where COH, through COH, are indicator variables reflecting immigrant membership in different entry cohorts. In principle, fitting this regression provides estimates of cohort-specific effects on earnings as well as an estimate of the average rate of earnings growth that is free of entry-cohort bias (i.e., an estimate of earnings growth within-and not across-entry cohorts). Several features of this econometric approach should be kept in mind. First, unlike longitudinal data, cohort data cannot be used to estimate individualspecific effects on earnings because there is no information to link the same individuals in the different cross sections. Second, a particular entry cohort sample observed in 1971 is not necessarily representative of the same population as the corresponding sample that is observed in 1981. As noted earlier, nonrandom patterns of out-migration, mortality, and differential undercounting-of which we find some evidence in our data as well as in Lam (1987)will tend to undermine the comparability of the samples. Changing patterns of employment and self-employment pose similar problems insofar as our regressions are fit to samples of working individuals who earned their income primarily from wages and salaries. Third, because only two cross sections are available for the present analysis, we will not be able to control for period effects that may affect the earnings of different cohorts differently (e.g., the business cycle). Also deserving mention are two issues raised by the pooling of data from two cross-sectional samples. First, in order meaningfully to compare earnings in the 1971 and 1981 Censuses, it is necessary to make an adjustment for inflation. We do this by using the Canadian Consumer Price Index to transform earnings in the 1971 Census (which refer to the year 1970) into 1980 inflation-adjusted dollars (the multiplication factor is 2.17). Second, intercen-
334
David E. Bloom and Morley Gunderson
sal changes in the real earnings of immigrants may be partly due to changing capital-to-labor ratios, technological change, or business-cycle fluctuations. Since real wage growth due to these factors does not reflect labor market progress that is immigrant specific, we make an adjustment to the real earnings of immigrants in the 1971 Census that transforms those data into “productivityconstant” terms. These adjustments highlight our central interest in this section: measuring the component of intercensal earnings growth for different entry cohorts of immigrants that is independent of human capital accumulation, overall economic growth, business-cycle effects, inflation, etc. In order to explore the robustness of our results, we make two distinctly different types of productivity adjustments. First, we simply multiply immigrant earnings reported in the 197 1 Census by the ratio of real earnings received by native Canadians in the 1981 and 1971 Censuses (the multiplication factor is 1.20). This simple adjustment assumes that immigrants would have experienced the same real wage growth as natives in the absence of any assimilation effects. At a somewhat deeper level, it assumes that average levels of human capital did not change among natives relative to immigrants and that the structure of returns to different types of human capital also did not change between Censuses. To avoid these assumptions, we also employ a slightly more complex productivity adjustment that takes account of changes in the human capital profile of the immigrant and native labor forces and of changes in the returns to different types of human capital. We do this by (a)estimating real wage equations for natives in 1971 and 1981 and (b)using the difference in the estimated coefficients to adjust immigrants’ 197 1 wages for intercensal changes in the returns to different types of human capital. We report results based on both sets of adjustments below. 12.2.2 Empirical Results on the Level of Earnings The immigrants we analyze represent 1/100 samples of individuals born outside Canada, while the natives represent 1/600 samples of individuals reporting Canada as their place of birth. Both the immigrant and the native samples are restricted to individuals aged 20-64 who are not predominantly self-employed and who worked at some point during the year prior to the Census for a wage in excess of fifty cents per hour in the 1971 Census and one dollar per hour in the 1981 Census. The variables IMM46.50-IMM7680 refer to immigrant entry cohorts (e.g., year of entry from 1946 to 1950, etc.). Table 12.7 reports descriptive statistics for the various samples we analyze. Table 12.8 presents ordinary least squares estimates of wage equations using cross-sectional data from 197 1 and 1981 separately. The dependent variable in all equations is the natural logarithm of an individual’s wage and salary earnings in inflation-adjusted (1980) dollars during the year preceding the Census (i.e., 1970 and 1980). These equations were specified with schooling (SCH), experience (EXP and EXPSQ), marital status (MST), and vectors of categorical variables (not reported in the table) measuring hours worked per
335
An Analysis of the Earnings of Canadian Immigrants
Table 12.7
Variable Definitions and Descriptive Statistics for Male Immigrants and Natives in 1971 and 1981 Immigrants
Variables
1971
Definitions
YSM
Log previous year’s wage and salary income, nominal dollars Years since immigration
SCH
Education in years
EXP
Work experience in years
MST
Marital status dummy (1 currently married) Immigrated 197680 Immigrated 1971-75 Immigrated 196670 Immigrated 1961-65 Immigrated 1956-60 Immigrated 194655
LNWAG
IM7680 IM7175 IM6670 IM6165 IM5660 IM4655 Sample Size
=
1981
Natives
1971
1981
8.795 8.710 9.754 9.667 (.622) (.691) ( .689) (.6@) ... ... 15.121 18.383 (8.884) (10.151) 10.277 10.515 11.237 11.088 (3.426) (3.269) (2.963) (2.67 1) 21.320 23.708 19.079 23.730 (13.162) (12.371) (13.380) ( 12.806) ,725 ,762 .802 .806
...
...
.218 ,111 ,187 ,338
,104 ,165 .189 .lo3 ,164 .230
...
... ...
8,290
9,368
5,119
6,295
...
... ... ...
Note: The immigrant samples represent lil00 samples of all immigrants in each Canadian Census aged 20-64 who report working positive weeks and hours, who earned more than 50 cents per hour in the 1971 Census (one dollar per hour in the 1981 Census), and who are not primarily self-employed. Inmates, members of the armed forces, and immigrants arriving during the year in which the Census was taken are excluded from the sample. The native samples represent l i 600 samples of all natives in each Canadian Census. The samples were constructed using the same criteria as those used to construct the immigrant sample, of those criteria that are relevant. The regressions include vectors of categorical variables with information on hours worked per week and weeks worked per year.
week and weeks worked per year as right-hand-side variables. Depending on the specification, the wage equations may also include an indicator variable for immigrants (IMMIG) and a variable measuring the number of years an immigrant has been in Canada (YSM). Although the results are not reported in these tables, we also estimated models with additional right-hand-side variables reflecting an individual’s religion and language ability and the square of YSM. Since these latter variables had little explanatory power either individually or jointly, these specifications are not reported here in the interest of parsimony. The first two columns of table 12.8 report wage equation estimates for the samples of male immigrants in 1971 and 1981. The estimated equations have the basic structure one might expect: a 4-5 percent rate of return to schooling, an earnings-experience profile that increases at a decreasing rate, and a 15-25 percent positive wage differential for married men. In addition, the “years since migration” coefficients are positive and significantly different from zero,
336
David E. Bloom and Morley Gunderson
Table 12.8
Estimated Wage Equations for Immigrants and Natives Using Single Cross Sections: Males Immigrants
Natives
Immigrants and Natives
~~
1971
Variable
YSM
.0044 (.0007) ,0427 (.0022) ,0376 (.0016) - ,068I (.0030) ,1569 (.0137)
IMMIG
(X
IMMIG)
SCH
1981 .0053 (.OO06) ,0447 (.0018)
1971
1981
.0054
(.oow
(.0017) - ,0692 (.0033) ,1820 (.0132)
,0670 (.0025) ,0405 (.0021) - ,0655 (.0042) .1963 (.0169)
,0574 (.OO25) ,0436 (.0019) - ,0701 (.0039) .I920 (.0147)
...
...
...
...
Constant
6.6074 (.0751)
6.6777 (.0683)
6.1279 (.0676)
6.5665 (.0613)
RZ RSS N
,485 1,651.O 8,290
,554 1,089.9 5,119
,543 1,364.6 6,295
EXP EXPSQ(- 100)
MST
.woo
,509 2,004.9 9,368
1971
1981 ,0077
(.OOOS)
,0528 (.0014) ,0385 (.0013) - ,0658 (.0024) .I754 ( .O 107) - ,0693 (.0123) 6.3749
,0486 (.0014) ,0421 (.0013) - ,0705 (.0025) .I875 (.0099) - .I657 (.0120) 6.6566 (.M) (.0399)
,508 2,790.5 13,409
,516 3,433.9 15,663
Note: See note to table 12.7. Estimated standard errors are reported in parentheses below the coefficient estimates.
although the coefficients are rather small in magnitude (i.e., about .5 percent per year). The third and fourth columns of table 12.8 report estimates of similar equations for native Canadians. The pattern of results corresponds quite closely to those for the immigrants, although the rate of return to schooling is slightly higher for natives than for immigrants, probably indicating that schooling has an important country-specific component. Even the residual variances for the immigrant and native wage equations are quite close in magnitude (e.g., .20 for the immigrants and .21 for the natives in the 1971 Census data). Given the closeness of the estimated wage equations for the immigrants and the natives, a simple way to compare the wage profiles is to follow the work of Chiswick (1978) by pooling the data for the two groups and fitting a wage equation that includes a dummy variable for immigrants, both by itself and interacted with the “years since migration” variable. The results of this exercise are presented in the last two columns of table 12.8. Based on the 1971 data, the estimate of the immigrant coefficient indicates that immigrants earn roughly 7 percent less than comparable natives when they first arrive in Canada; the estimate of the coefficient on YSM indicates that immigrants’ wages increase an average of .54percent per year spent in Canada, beyond the increase associated with the acquisition of experience. These estimates imply
337
An Analysis of the Earnings of Canadian Immigrants
that the earnings profiles of comparable immigrants and natives cross at roughly 12.8 years. In contrast, the 1981 data indicate that entering immigrants earn 16.6 percent less than otherwise comparable natives, although their wages increase at the rate of .77 percent per year spent in Canada, suggesting that the immigrant and native earnings profiles do not cross until the immigrants have been in Canada for 2 1.6 years. In order to investigate whether the various cross-sectional estimates of immigrant labor market progress are biased by entry-cohort effects on wages, we now estimate alternative models from pooled 1971 and 1981 data on immigrants. The first column of results in table 12.9 reports estimates of the simple Table 12.9
Wage Equations for Male Immigrants in Canada Based on Pooled 1971 and 1981 Data
Unadjusted Cross-Cohorts Model YSM SCH EXP EXPSQ(/ 100) MST CONSTANT
.0076 (.0005) ,0460 (.0013) .0401 (.0012) - ,0719 (.0022) ,1711 (.0096) 6.6467 ( ,0508)
IM7680 IM7175 IM6670 IM6 165 IM5660 IM4655
R2 RSS N
,494 37.663 17,658
Unadjusted Within-Cohorts Model
,0202
(.W8)
,0437 (.0013) ,0383 (.0012) - ,0675 (.0022) ,1723 (.0095) 6.1493 (.0564) .4713 (.0302) .5062 ( ,0261) ,4548 (.0241) .3684 (.0224) ,2730 (.0189) ,1676 (.0155) .506
36.757 17,658
Within-Cohorts Model Using Sample Average Productivity Adjustment
Within-Cohorts Model Using Weighted Average Productivity Adjustment
,0024
,0042
(.OOW
(.O@W
,0437 (.0013) .0383 (.0012) - .0675 (.0022) .1723 (.0095) 6.8620 (.0564) - ,1874 (.0302) - .0634 (.0261) - ,0259 ( ,0241) - ,0233 (.0224) - .0296 (.0189) - ,0104 (.0155)
.0389 (.0013) .0401 (.0012) - .0703 (.0022) ,1680 (.0095) 6.8828 (.0563) -.1163 ( ,0302) - ,0066 ( ,0260) ,0142 ( ,0240) ,0116 (.0223) - ,0034 (.0188) - ,0026 (.0154)
,496 36.757 17,658
,501 36.612 17,658
Note: See note to table 12.7 Estimated standard errors are reported in parentheses below the coefficient estimates.
338
David E. Bloom and Morley Gunderson
wage equation fit to the pooled data. As one would expect, these estimates are quite similar to the estimates computed for the separate years’ samples (i.e., they are simply a matrix-weighted average of the results in cols. 1 and 2 of table 12.8). In the second column, we include a vector of dummy variables that reflect an immigrant’s entry cohort. The coefficient of YSM in this equation therefore represents a weighted average of within-cohort real earnings growth between 1971 and 1981 for the different entry cohorts, other things equal. The estimate of this coefficient is quite substantial in magnitude (2.02 percent) and is quite well determined (i.e., the standard error is .08 percent). Thus, the within-cohort growth rate of real earnings is three to five times larger than the cross-cohort growth rate of real earnings. In addition, estimates of the cohort fixed effects suggest that the average unmeasured quality of immigrants increased across all entry cohorts until the cohort that entered from 1976 to 1980. As noted earlier, the coefficient on YSM captures both the true “assimilation effect” in which we are interested and any wage growth associated with changes taking place in the economy over time (e.g., increasing capital-tolabor ratios). In order to isolate the assimilation effect, we adjust the 1971 earnings data for changes in labor productivity that occurred among native Canadians from 1971 to 1981. Estimates of the wage equation fit to these productivity-adjusted data are reported in the third and fourth columns ‘of table 12.9. The third column applies the simple adjustment described above based on the ratio of real wages received by native Canadians in 1971 and 1981; the fourth column applies the more complex regression-based adjustment that accounts for human capital changes among the immigrant and native labor forces as well as changes in the returns to different types of human capital. The coefficient of YSM reported in column 3 is .24 (with a standard error of .08). Although it is statistically significant, this estimate is substantially smaller than the estimate of 2.02 computed using non-productivity-adjusted data. Thus, it appears that the effect of assimilation on the earnings of cohorts of immigrants is quite small, indeed, almost negligible. Put another way, within-cohort growth in immigrant earnings is primarily due to economic forces that affect both immigrants and natives. Further confirmation of this result is provided in the fourth column of table 12.9. Although the estimated coefficient of YSM is somewhat larger when we apply the more elaborate productivity adjustment, the assimilation effect of .42 percent per year is still substantially small. Moreover, it is sufficiently close in magnitude to the estimates computed from the individual cross sections to conclude that Borjas’s assertion that estimates of the latter type are “misleading and useless” does not necessarily generalize beyond the data he analyzed. It is also worth examining the estimates of the cohort fixed effects in columns 3 and 4 in table 12.9. There are no statistically significant cohort effects among any of the five pre-1971 entry cohorts of immigrants. However, the
339
An Analysis of the Earnings of Canadian Immigrants
estimates do suggest that average unobserved quality among immigrants arriving from 1976 to 1980 (and perhaps also among the 1971-75 arrivals) was significantly lower than for previous cohorts, the same conclusion suggested by a comparison of the cross-sectional coefficient estimates of IMMIG in table 12.8. This result is not especially surprising given the relatively high fraction of immigrants admitted into Canada as relatives of Canadian citizens and landed immigrants in the late 1970s (see table 12. I). Under the immigration policy in effect in Canada in the 1970s, applications from relatives did not receive the same degree of labor market screening as independent applications did. 12.2.3 The Dispersion of Immigrant Earnings In this subsection, we present and review statistics on the dispersion of income among immigrants. In particular, we are interested in determining whether immigrant incomes tend to become more or less disperse as the duration of their stay in Canada increases. A tendency for dispersion to decline with duration of stay is consistent with the hypothesis that intercensal outmigrants tend to be selected from the tails of the distribution (i.e., immigrants whose earnings expectations were not met selected out of the lower tail and migrants who planned temporary stays to take advantage of favorable earnings opportunities selected out of the upper tail). In contrast, a tendency for dispersion to increase with duration of stay would be consistent with the view that the labor market has more information about the true productivity of immigrants the longer they have stayed in the country. The third and fourth columns of table 12.10 report the raw standard deviations of immigrant incomes by the duration of their stay in Canada as of both 1971 and 1981. Although the standard deviation of income is highest for the oldest entry cohort of immigrants, there is little evidence of a pattern across the more recent entry cohorts in either Census year. There is some tendency for income dispersion to decrease for individual entry cohorts from 1971 to 1981, but this may not be due to increased duration of stay since dispersion among native Canadians also decreased between 1971 and 1981. In addition, it is worth noting that the standard deviation of immigrant incomes is remarkably close to the standard deviation of native incomes in both Census years. Table 12.10 also reports the standard deviation of the estimated residuals for different entry cohorts of immigrants in the 1971 and 1981 Censuses. These statistics are more appropriate measures of dispersion than the raw standard deviations because they do not reflect the influence of variations in observable factors that are associated with earnings. Nonetheless, they tell basically the same story as the unconditional measures of dispersion: there is no substantial evidence of a difference between natives and immigrants, or among immigrants with different durations of stay in Canada, in the dispersion of income. Thus, the statistics in table 12.10 are equally supportive of two conclusions: either selective out-migration and job matching are both empirically unimpor-
340
David E. Bloom and Morley Gunderson
Table 12.10
Dispersion and Kurtosis in Conditional and Unconditional Distributions of Wage and Salary Income, for Male Immigrants, by Census Year
SD of Immigrant Earnings (OOO dollars)
SD of Residuals in Log Earnings Equation
Kurtosis of Residuals in Log Earnings Equation
SamplelEntry Cohort
1971
1981
1971
1981
1971
1981
All pre- 1970 immigrants Native Canadians Pre- 1946
12.2 11.8 15.7
11.0 10.9 12.2
11.3
10.8
.44 .47 .46 .42
2.9 2.3
1946-55 1956-60 196145 1966-70
3.7 2.8 2.4 4.6
11.0
10.4 11.4 10.9
.45 .46 .48 .42 .41
10.3 12.6
.44
.55
.44 .46 .47
3.0 3.6 2.5 2.6 2.1
5.0 3.5
3.3
Note: The equation used to estimate the residuals and to calculate their standard deviation and kurtosis is reported in col. 3 of table 12.9.
tant influences on immigrant earnings, or they are important influences whose effects tend to cancel out. In an attempt to distinguish between these alternative views, we examine higher-order moments in the distribution of residuals from the earnings equation. If the tails of the distribution are thinning as a result of out-migration, and if the variance of earnings is increasing among Canadian immigrants who stay in Canada, we would expect increased kurtosis in the distribution of residuals for particular entry cohorts; that is, the distributions should “thicken” from one Census to the next. This pattern of results is clearly revealed in the last two columns of table 12.10. However, because kurtosis also increases among native Canadians, a finding we had no reason to expect, we are reluctant to view our results for the immigrants as conclusive. It would thus appear that a fuller understanding of the dynamics of immigrants’ labor market outcomes and their out-migration decisions awaits the advent of true, large-scale, longitudinal surveys of immigrants.
12.3 Conclusion This paper has reported estimates of simple wage equations fit to crosssectional and cohort data for Canadian immigrants in the 1971 and 1981 Canadian Censuses. The estimates are used to assess (1) the usefulness of crosssectional analyses for measuring the pace of immigrant earnings growth, (2) the labor market implications of admissions policies that place different weights on the work skills possessed by prospective entrants, and (3) the relative effect of selective out-migration and job matching on the shape of immigrant earnings distributions as duration of stay increases.
341
An Analysis of the Earnings of Canadian Immigrants
The estimates provide evidence of a small to moderate assimilation effect that suggests that immigrants make up for relatively low entry wages, although the wage catch-up is not complete until thirteen to twenty-two years after entry into Canada. These results are revealed clearly in both the pseudolongitudinal and the cross-sectional analyses. The estimates also provide evidence that the unobserved quality of immigrants’ labor market skills declined following changes in Canada’s immigration policies in 1974 that led to a sharp increase in the proportion of immigrants admitted on the basis of family ties. Finally, since there is no evidence that the variance of immigrant earnings increases with their duration of stay in Canada, and since there are no differential immigrant-native changes in higher-order moments of the earnings distribution as duration of stay increases, the results are inconclusive with respect to the importance of selective out-migration and job matching in the evolution of immigrant earnings distributions over time.
References Allen, J. 1979. Information and Subsequent Migration: Further Analysis and Additional Evidence. Southern Economic Journal 45: 1274-84. Blejer, M., and I. Goldberg. 1980. Return Migration-Expectation versus Reality: A Case Study of Western Migrants to Israel. In Research in Population Economics, vol. 2, ed. J. Simon and J. DaVanzo. Greenwich, Conn.: JAI. Borjas, George. 1982. The Earnings of Male Hispanic Immigrants in the United States. Industrial and Labor Relations Review 35:343-53. . 1985. Assimilation, Changes in Cohort Quality, and the Earnings of Immigrants. Journal of Labor Economics 3:463-89. Carliner, Geoffrey. 1980. Wages, Earnings, and Hours of First, Second, and Third Generation American Males. Economic Inquiry 18237-102. Chiswick, Barry. 1978. The Effect of Americanization on the Earnings of Foreignborn Men. Journal of Political Economy 86:897-921. Fox, Marc. 1987. Remittance and Labor Supply Behavior of Immigrants. Ph.D. diss., Harvard University, Department of Economics. Harris, Milton, and Bengt Holmstrom. 1982. A Theory of Wage Dynamics. Review of Economic Studies 49:315-33. Jasso, Guillermina, and Mark R. Rosenzweig. 1987. How Well Do Immigrants Do? Vintage Effects, Emigration Selectivity, and the Occupational Mobility of Immigrants. In Research in Population Economics, vol. 6, ed. T. Paul Schultz. Greenwich, Conn.: JAI. . 1990. Self-selection and the Earnings of Immigrants: Comment. American Economic Review 80:298-304. Katz, Eliakim, and Oded Stark. 1984. Migration and Asymmetric Information: Comment. American Economic Review 74:533-34. Lam, Kitchun. 1986. Imperfect Information, Specificity of Schooling and Rate of Return-Migration. Economics Letters 21 :283-89. . 1987. An Analysis of the Outmigration of Foreign-born Members in a Population. Ph.D. diss., Harvard University, Department of Economics.
342
David E. Bloom and Morley Gunderson
Long, James E. 1980. The Effect of Americanization on Earnings: Some Evidence for Women. Journal of Political Economy 88:620-29. Stark, Oded, and David E. Bloom. 1986. The New Economics of Labor Migration. American Economic Association Papers and Proceedings 75: 173-18. Yezer, Anthony, and L. Thurston. 1976. Migration Patterns and Income Changes: Implications for the Human Capital Approach to Migration. Southern Economic Journal 42:693-702.
13
The Effects of International Competition on Collective Bargaining Outcomes: A Comparison of the United States and Canada John M. Abowd and Thomas Lemieux
The decade of the 1970s heralded the latest era of product and labor market globalization with renewed internationalization of the U.S. economy. Canada, which has long been a very open economy, also experienced substantially increased internationalization during this decade.2 The two countries have very substantial bilateral trading activity. Canada is the largest single destination of U.S. exports and the second largest source of U.S. imports (after Japan). In the 1970s, Canada was the largest destination and origin of U.S. traded goods. The growing importance of internationally traded goods in the U.S. economy and the continuing importance of such goods in the Canadian economy are displayed in figure 13.1. The figure shows exports plus imports as a percentage of gross domestic product in 1960, 1970, and 1980 for both countries. That the Canadian economy is three or four times more open than the U.S. economy is directly evident from examining figure 13.1. 4 The figure also indicates that the Canadian economy experienced very significant increases in its openness over this time period-imports plus exports grew from less than 40 percent of GDP to almost 60 percent. In the United States, imports plus exports grew from about 7 percent of GDP to over 20 percent during this period. The 1960s and 1970s, therefore, represent a period of increasing international economic activity for both the U.S. and the Canadian economies. John M. Abowd is professor of labor economics and management, Cornell University, and a research associate of the National Bureau of Economic Research. Thomas Lemieux is assistant professor of economics, Massachusetts Institute of Technology. The authors acknowledge financial support from the Ford Foundation, the National Science Foundation (grant 88-13847 to Abowd), and the Industrial Relations Section at Princeton University. This research was begun while the authors were at the Industrial Relations Section, Princeton University. They thank David Card for providing a clean version of the Labour Canada wage tape. They thank Charles Beach, David Card, Henry Farber, Richard Freeman, Harry Gilman, Lawrence Katz, W. Craig Riddell, and Gregory Schoepfle for comments on previous drafts.
343
344
r
John M. Abowd and Thomas Lemieux
60% 50%
40%
30%
20%
10%
0%
1960
1980
1970
=
USA
Canada
Fig. 13.1 Openness of the U.S. and Canadian economies Sources: United States, National University Data Base, CANSIM.
Income and Product Accounts,
CITIBASE
1978. Canada,
Both countries experienced changes in economic openness during this period large enough to have detectable effects on the domestic labor market. For the last two decades, the highly unionized Canadian manufacturing sector and the less unionized U.S. manufacturing sector have become more integrated into both the North American and the world market^.^ The share of imported manufactured goods in apparent Canadian consumption has increased from 22 to 3 1 percent over this period. Similarly, exports as a share of Canadian manufacturing production have more than doubled. Much of the increased integration of the North American economy occurred in the transportation equipment industry, which is covered by a 1965 bilateral agreement between the United States and Canada that eliminates most tariffs in both directions. Canadian transportation equipment imports rose from 28 percent of apparent domestic consumption in 1960 to 51 percent in 1968 to 65 percent in 1983. Canadian transportation equipment exports also rose, from 14 percent of domestic production in 1960 to 48 percent in 1968 to 67 percent in 1983. In the United States, transportation equipment imports increased from 2 percent of apparent domestic consumption in 1960 to 18 percent in 1983, while exports increased from 6 percent of domestic production to 14 percent in 1983. The increased bilateral trade in many other U.S. and Canadian industries, although not as substantial as the increased trade in transportation equip-
345
International Competition and Collective Bargaining Outcomes
ment, supports our premise of increased integration of the North American economy. In Canada, the effect of international competition on unionized domestic workers is regarded as a macroeconomic question susceptible to analysis using the tools of open economy international trade theory (see Cousineau 1987; Riddell 1986a, 1986b, 1986~).In the United States, the effect is generally regarded as a microeconomic question susceptible to analysis on an “effected industry” basis6 This paper is an attempt to blend these two views. Like the Canadians, we will estimate equations for the average effect of international competition on unionized wages and employment^.^ Like other U.S. researchers, we will model the microeconomic basis for the trade effects.8 Our study addresses the relations among international product market competition and the outcomes of domestic collective bargains. We have three specific goals. First, we quantify the effects of import and export competition on the wages and employment of unionized domestic workers using comparable data for the United States and Canada. Second, we compare the magnitude of the estimated effects to a reference value-the effects of a comparable change in domestic shipments-to determine if the trade effects are relatively large. Finally, for Canada we compare the estimated effects of value-based measures of import and export activity to the estimated effects of price-based measures, which may be more appropriate from a theoretical viewpoint. Using value-based trade measures, the estimated effect of an increase in import domestic market share, holding constant the rate of growth of the domestic market, is negative for employment in both countries and exceeds the effect of a comparable change in the size of domestic shipments. The import effect on realized real union wage rates is also negative for the United States, but not for Canada. The import effect on real wage rates in the United States is also larger than the effect of a comparable change in the domestic market size. The estimated effect of increased export growth is positive for bargaining unit employment in both countries. The export effect on employment is comparable in magnitude to the effect of a change in the size of the domestic market. The export effect on real wage rates is mixed-weakly positive for the United States and ambiguous for Canada. Increases in world export price indices are associated with increased union employment in Canada. Increases in world import price indices are associated with increased union employment and lower wage settlements in Canada. All the estimated world price effects on the Canadian unionized labor market are consistent with the estimated effects of value-based export and import measures for that economy.
13.1 The Role of International Competition in Collective Bargaining International competition may influence domestic collective bargaining in two ways. First, to the extent that foreign manufactured products are good
346
John M. Abowd and Thomas Lemieux
substitutes for domestically manufactured products, domestic firms must compete for global market share. Within the domestic product market, import competition reduces the effective extent of union organization in an industry and may reduce the quasi rents available to existing bargaining units. Within foreign product markets, exports work in the opposite direction to increase the effective extent of union organization in an industry and possibly to increase the quasi rents available to domestic bargaining units. Second, as substitute foreign manufactured products gain market share within the domestic product market, complementary services-wholesale distribution, retail distribution, and repair-also gain market share. The expansion of service employment opportunities may create quasi rents that could promote the formation of new bargaining units in these industries. This paper deals only with the direct competition effects of imports and exports on unionized workers in domestic labor markets. We do not consider the indirect effects of service sector expansion (see Leonard and McCulloch, in this volume). The major direct effects of international competition on unionized workers occur because either the union wage rate falls (relative to what it would otherwise have been) or there are employment displacements, which may be associated with unusually long unemployment spells andor wage reductions on reemployment. We consider only direct wage costs and direct bargaining unit employment effects. We do not examine unemployment spell length or reemployment wage rates. The major direct benefit to worldwide consumers is lower average product prices in the markets where there is substantial international competition. We do not measure this benefit to the consumer, although this is certainly an essential component of any policy prescription arising from this research. A bargaining unit is an ongoing relationship between a union and a financially viable e m p l ~ y e r The . ~ union represents the interests of the organized employees. The management represents the interests of the shareholders and other ultimate beneficial owners. For simplicity, the claims of other factors of production are ignored. During the negotiations that accompany the expiration of an existing collective bargaining agreement, management and the union use current information to form an estimate of the total value of the productive enterprise for which they represent competing interests. A collective bargaining outcome consists of explicit and implicit rules concerning the allocation of resources (employment) and the division of the resulting quasi rent between union members and shareholders (wage rates) that is expected to remain in force for some fixed term. If international competition is expected to have an adverse effect on the firm’s future profitability, then the current collective bargaining agreement will reflect that expectation. l o If the expected effects of international competition are too severe, the bargaining unit may disappear, and the evidence on surviving bargaining unit settlements will not reflect a complete analysis of either employment or wage effects. If the international competition is ex-
347
International Competition and Collective Bargaining Outcomes
pected to improve the firm's future profitability, current bargaining units should be favorably affected. In this paper, we measure the expected effects of increased foreign competition on the future value of the firm using the relation between future revenues (of organized employers) and current information on domestic shipments, apparent domestic consumption, exports, and imports in the employer's product market. We consider two collective bargaining outcomes-bargaining unit employment and wage settlements. Bargaining unit employment is measured as the ex post growth rate of workers in the bargaining unit over the life of the new collective bargaining agreement, excluding retirees and including members with recall rights." Wage settlements are measured as the realized growth rate of real wage rates over the life of the agreement for the largest group of workers in the bargaining unit. We measure the effects of predictable increases in international competition on bargaining outcomes by relating employment growth and real wage growth during a collective bargaining agreement to the growth of apparent domestic consumption, exports, and imports expected to prevail during the life of the agreement. We measure the effects of unpredicted changes in international competition by relating the same outcomes to unexpected changes in domestic consumption, exports, and imports over the life of the agreement. In an open economy, union employment and wage rates within internationally competitive industries should respond to changes in the world market for manufactured goods. In general, we would expect these changes to depend on the world prices of traded goods. Although the importance of using world prices in modeling the effects of international trade on domestic labor markets has been recognized for some time (see Grossman 1982, 1986, 1987), most empirical analyses of the United States use import penetration ratios and export supply ratios as the main indicators of changes in the international environment.I* In this paper, we also consider the effects of changes in the world price of exports and imports on union employment and wage rates using our Canadian sample. We use Canadian data for two reasons. First, properly constructed price indices exist for a much longer time period in Canada than in the United States. Second, the Canadian economy is substantially more open than the U.S. economy. Our analysis thus permits examination of the consequences of using a variety of measures of international trade-value and price based-on the resulting estimates of employment and wage sensitivity.
13.2 A Model for the Effects of International 'lkade on Union Wages and Employment We begin at the bargaining unit level. Consider the effects of increased international trade on the present value of the quasi rents accruing jointly to the employer firm and union members. The quasi rents are measured as the difference between net revenues and the cost of employment.I3The cost of employ-
348
John M. Abowd and Thomas Lemieux
ment is evaluated at an external or market wage rate, not at the negotiated wage rate. For firmj in year t, let
R,, = net revenue of firm j in year t; Ljr = union employment of firmj in year t; MI, = total employment of firm j in year t; wJr= negotiated wage rate of firmj in year t; ,z, = any exogenous variable for firmj in year t; r, = one-year discount rate in year t; x, = annual opportunity cost of employment in year t; q = length of collective bargaining agreement. If year t is the initial year of a new collective bargaining agreement, then over the next q years the present value of the quasi rents may be expressed as
Since V,, measures the total quasi rent available to the employer firm and union, if increased import or export activity affects V,,, then wage settlements and union employment will be affected by this activity. Alternatively, if increased international trade has no effect on V,,, then neither wage settlements nor union employment should be affected. The present value of the quasi rents accruing to the bargaining unit captures the relevant total value of the enterprise, which may be divided among various claimants, including union members.I4 If the total value increases because of increased export activity, then the potential exists for greater union employment or wage settlements as a consequence of this export activity. Alternatively, if increased import competition lowers Vjr,then there is a presumption that lower employment and/or wage settlements should occur. To make these arguments concrete, consider the effects of a change in real industry shipments. Changes in real industry shipments due to external demand shocks should cause industry employment and total quasi rents to move in the same direction as industry shipments. If existing firms each represent a constant fraction of industry output, then firm level employment and total quasi rents should also change in the same direction as the change in industry output. The effect of a change in industry output on the negotiated wage rate is less clear. An increase in industry output could be associated with an increase or decrease in quasi rents per worker. Since the negotiated settlement divides the quasi rents per worker between the firm and the union members, an increase in quasi rents per worker ought to be associated with higher wage settlements, while a decrease in quasi rents per worker ought to be associated with lower wage settlements, all other things equal. These predictions can be derived explicitly from a simple version of an efficient bargaining relation between the employer firm and the union mem-
349
International Competition and Collective Bargaining Outcomes
bers. Assume that the firm is fully unionized. Suppose the revenue that accrues to the firm is given by the function
0 R(L) = aL - -L*. 2 An efficient bargain chooses L to maximize R(L) - xL and sets w to divide the maximized quasi rents in the proportion y to the union and 1 - y to the owners of the firm. The resulting values for total quasi rents, employment, and the negotiated wage rate are
yo a - x
w = x + - 2 (
0
i.
In this simple model, any increase in demand for the firm’s product would be modeled as an increase in a.As the equations for L, and w show, increases in a are associated with higher total quasi rents, higher employment, and higher negotiated wage rates. Notice that the wage settlement awards each union member the same percentage of the quasi rents per worker (y) as overall negotiation determined. In the quadratic revenue model, an increase in quasi rents is always associated with an increase in quasi rents per worker so that the negotiated wage rate must increase; however, general functional forms for the revenue equation do not imply this particular result. Consider next the effects of a change in the world prices of imports and exports on the output of a particular industry. Assume that the domestically produced goods are Hicks substitutes for the imported goods and that the exported goods are identical to the domestically produced and consumed goods. Then, an increase in the world price of industry imports results in substitution away from the imported goods and into the domestically produced goods within the industry. This should increase domestic output within the industry. Hence, employment and total quasi rents should increase for the existing firms within the industry. Negotiated wage rates will increase or decrease depending on whether quasi rents per worker increase or decrease. An increase in the world price of exports results in expansion of domestic industry output along the industry supply curve. Hence, employment and total quasi rents should increase for the existing firms within the industry. Again, the movement in negotiated wage rates will depend on what happens to quasi rents per worker. IJ
350
John M. Abowd and Thomas Lemieux
13.3 An Empirical Specification for Bargaining Unit Level Data In order to give our model empirical content, we must specify relations connecting the exogenous economic factors (industry output, value-based trade measures, import prices, and export prices) to the total quasi rent, employment, and wage rate outcomes. We will not use comparable quasi-rent data for the United States and Canada; hence, our empirical models will consider only bargaining unit employment and wage rates. We deal with three important practical problems in developing our estimating equations. First, since the model is developed for application to bargaining unit data, the employer firms may have heterogeneous shares of output within the domestic industries and may face heterogeneous industry demand elasticities with respect to import and export price changes. There are insufficient data to attempt estimation of separate elasticities for each major industry. Instead, we formulate the model to permit estimation of the employment-weighted average elasticities across all domestic manufacturing firms in the sample for each country. Second, since nominal wage rates are renegotiated infrequently relative to changes in the economic environment, we distinguish between the effects of expected and unexpected changes in the exogenous variables. Expected changes in the exogenous variables are movements forecast in advance of the current negotiation. The effect of expected changes is captured by including the forecasted value of the exogenous variable, conditional on information available at the time of contract renegotiation, among the explanatory variables in the wage and employment equations. Unexpected changes in exogenous variables are the difference between movements realized over the life of the new agreement and the forecast of these movements formed during the negotiation of the agreement. The effect of unexpected changes is captured by including the forecast error among the explanatory variables. Third, since the important outcomes are not observed at frequent, equally spaced, synchronous intervals, we specify a set of relations that can be estimated using observations on the relevant variables that are measured in contract time. This allows us to estimate the models using vector autoregressions linking the annualized rates of change in the dependent and exogenous variables. In order to distinguish between the expected effects of economic factors on collective bargaining outcomes and the realized effects of these changes over the life of an existing agreement, we decompose exogenous economic variables into expected and unexpected components. For any exogenous variable z,~, we assume that the level follows a discrete martingale, so that for any positive q. Then write the logarithm of union employment and wage settlements as of the end of a contract that begins at date t and expires at date t + qas
351
International Competition and Collective Bargaining Outcomes
In L,r+q = EUnL,r+q I In In
w,r+q
=
z,rI +
ul,t+qt
EUn w,r+q I In L,r?In wjr, z,r+q, z,,I +
u2,r+q.
In w,rT
Z,r+q,
If the conditional expectations are log linear in the levels of the underlying variables, then 1' In
4r+q
w,r+q
=
P,11
+
P,I4',r
=
P,21
+
1' L,r +
P,12
+
+
P,15(z~r+q
PI22
PI13
-
In r j r +
'1)
PI23
1' w,r
+
*l/r+q'
In w,r
+ Pj24jr + P,25(z,r+q + u2,r+q9 where the average of the coefficients P,rs must be estimated and the error vector yr should be vector white noise. Our specification allows for firm-specific factors, summarized by the average elasticities in the log-level equations. We difference this specification across collective bargains for the same firm and correct for the differences in contract length. This produces the following estimating equations:
+
A In L,r = bI2AIn -
(la)
Az,r-q)
+
(P,l5
+
Aul,r +
b13lA
-
b13Aln
w,r-q
(P,,,
ln w,r +
b15)(AZ,r+q
-
CP,,,
+ bI4Az,,-, + b,,(Az,, 1' L,r
b 1 2 ) ~
-
b14)Az,r
-
Prs = employment-weighted average of PjrS; A In L,r = (In L,,,,
- In L,,)lq,
and similarly for A In wjr and Az,. Equations (la) and (1 b) form the basic statistical model used in our analysis. The equations are a vector autoregression relating changes in union employment and wage settlements over the life of the agreement running from t to t q to lagged changes in these variables, lagged changes in exogenous variables, and the innovation in exogenous variables that occurs between t and t q. The last two lines of equations (la) and (lb), respectively, show the complete error structure. We assume that the bargaining unit specific error in each elasticity (Pjrs - fi,J is uncorrelated with all the right-hand-side variables in each equation. Furthermore, we assume that the heteroscedastic error
+
+
352
John M. Abowd and Thomas Lemieux
structure implied by equations (1a) and (1 b) can be adequately corrected by a weighted least squares estimator in which the weight is the size of the bargaining unit at time t . We do not require that the effects of expected changes in exogenous variables and the effects of innovations in those variables be identical. One of the key exogenous variables in the empirical analysis is industry shipments. The rate of growth of real industry shipments can be decomposed into a weighted combination of the rate of growth of real apparent domestic consumption, the rate of growth of real exports, and the rate of change of the import penetration ratio. This decomposition is given by
where SJlis total real shipments for industry j in year t ; X, is real exports, M, is real imports, Djris real apparent domestic consumption ( D = S + M - X ) , IPR, is the import penetration ratio (IPR = M / D ) , and the discrete differencing operation is defined in the notes to equations ( 1 ) . 1 6 When the separate components of this decomposition (multiplied by the indicated weights) are used in the analysis instead of the rate of growth of real industry shipments, the coefficients on the rate of growth of real apparent domestic consumption and real exports should be the same and equal to the coefficient on real shipments. The coefficient on the change in the import penetration ratio should be equal in magnitude and opposite in sign to the coefficients on apparent domestic consumption and exports. The equations including the value-based international trade measures are interpreted as measuring the effects of increased export and import activity for a given level of apparent domestic consumption.
13.4 Description of Data A detailed description of the methods used to form the bargaining outcome, international trade, and other exogenous variables is contained in the Data Appendix and the references therein. Table 13.1 contains a summary of the primary data sources for the United States and Canada. For the United States, we used data from 2,5 15 collective bargaining agreements representing 250 bargaining pairs in the manufacturing sector for the years from 1959 to 1984. The path of the realized nominal wage rate over the life of the collective bargaining agreement and the level of employment at the beginning of the agreement were extracted from the Major Collective Bargaining Agreements file developed by Vroman (1982, 1984, 1986) on the basis of Current Wage Developments. For Canada, we used data from 2,258 collective bargaining agreements in Canadian manufacturing representing 299 bargaining pairs in the manufacturing sector for the years from 1968 to 1983. The path of the realized real wage
353
International Competition and Collective Bargaining Outcomes
Table 13.1
A Comparison of Data Sources for the United States and Canada
Variable
United States
Canada
Contract wage rates
Vroman Agreements File from Current Wage Developments (CWD)
Labour Canada Wage Tape from Collective Bargaining Review
Bargaining unit employment
Report to BLS at settlement in CWD
Report to Labour Canada at settlement
Industry shipments
Annual Survey of Manufactures'
System of National Accounts (SNA)b
Industry prices
BLS shipments deflator
Implicit deflator (SNA)
Industry imports
BLS trade monitoring system'
Input-output tables
Industry exports
BLS trade monitoring system
Input-output tables
Import prices
Not available
CANSIM data
Export prices
Not available
CANSIM
filesd
data files
Revisions in the industrial coding for the Annual Survey of Manufactures have been resolved to 1972 standards. Revisions in the industrial coding for the Canadian System of National Accounts have been resolved to 1971 standards. BLS trade monitoring system data are available only for 1972-81. The same methods have been applied to construct data for the periods 1958-71 and 1982-84. Import and export prices from the Canadian data files are Laspeyres indices (1971 = 100) based primarily on end use prices. Comparable U.S. prices are not available for all manufacturing industries until 1980. a
rate over the life of the agreement and the level of employment at the beginning and end of the contract period were taken from a version of the Labour Canada wage tape based on the Collective Bargaining Review (see Card 1988). Table 13.2 contains a summary of the basic definitions of the dependent variables-bargaining unit employment and real wage growth rates. Table 13.2 also contains summary definitions of the growth rates of real industry shipments, real apparent domestic consumption, and real exports. Finally, the table defines the change in the import penetration ratio. The timing of our measures of collective bargaining outcomes and exogenous variables is based on the effective date of the current collective bargain. To measure a realized change, we must observe bargaining unit employment, real wage rates, and all exogenous variables at two points in time. Of necessity, there is some arbitrariness in the timing conventions used. In selecting intervals over which to calculate the realized changes, we tried to use the available contract and exogenous information in a manner that permitted distinguishing the estimated effects of changes that were known prior to the settlement of the current contract and changes that occurred over the life of the current contract. In order to understand the timing of the various measurements, figure 13.2
354
John M. Abowd and Thomas Lemieux
Table 13.2
Summary of the Definitions of Key Variables
Variable
Detailed Definitiona
Bargaining unit employment growth rate
Annualized rate of growth of bargaining unit employment between the end of the previous contract and the end of the current contract.
Bargaining unit real wage growth rate
Annualized realized rate of nominal wage growth between the end of the previous contract and the end of the current contract minus the annualized growth rate of the Consumer Price Index over the same period.
Real shipments growth rate
Annualized rate of growth of industry shipments over the life of the current contract minus the annualized growth of the industry shipments deflator over the same period. Expectedpart: rate of growth over the previous contract. Unexpected part: difference between the rate of growth over the current contract and the rate of growth over the previous contract.
Real apparent domestic consumption growth rate
Shipments plus imports minus exports, annualized growth rate over the life of the current contract minus the annualized growth rate of the industry shipments deflator over the same period. Expected and unexpected parts: defined in the same manner as real output growth.
Real export growth rate
Exports from all sources to the destination country (United States or Canada, respectively), valued free alongside ship (value at the point of exportation), annualized growth rate over the life of the current collective bargaining agreement. Expected and unexpected parts: defined in the same manner as real output growth.
Import penetration ratio
Imports for consumption (customs value at port of entry, usually foreign port value) divided by shipments plus imports, annualized change over the life of the current collective bargaining agreement. Shipments have been reclassified at the product class code level to conform more closely to the SIC-based imports. Expected part: annualized change over the previous contract (not a growth rate). unexpected part: difference between the change over the current contract and the change over the previous contract (not a difference in growth rates).
All annualized growth rates were constructed using the difference in the logarithms of the appropriate variable at two points in time divided by the length of time intervening.
a
shows a time line for an arbitrary three-year September contract effective in year t . The relative dates of measurement for all variables and the values labeled t , t - 3, and t + 3 are all shown on the figure. The figure illustrates that the change in the real wage rate is taken between the last month of the current contract (dated t q ) and the last month of the previous contract
+
355
International Competition and Collective Bargaining Outcomes
EXOGEVOUS ANb M
il DATA )$(
EXOGEP OUS MON
LY DATA
LJNION EMF
IYMENT I
CONTRACl
(AGE (t)
.$(
A
)
I
JAN --I
PREVIOUS (t-3)
JAN
rffmm
JAN *
CURRENT (t)
JAN TV-rr-7
+3)
Contract time in years
Fig. 13.2 Calendarkontracting dating of variables Nore: The figure illustrates a three-year, September contract.
(dated t ) . Employment changes are measured between the effective month of the next contract (dated t + q ) and the effective month of the current contract (dated t). Monthly exogenous data (average hourly earnings) are measured as of three months before the effective date of the current contract (dated t ) . Annual exogenous data are measured using the year that overlaps with the effective date of the current contract (dated t). Table 13.3 presents comparative summary statistics for the United States and Canada. The period covered is longer for the United States. There are more Canadian bargaining units but fewer Canadian contracts. U.S. bargaining units are larger, reflecting both the prevalence of larger establishments and the difference in the Bureau of Labor Statistics cutoff for inclusion in Current Wage Developments (1,000) versus the Labour Canada cutoff for inclusion in Collective Bargaining Review (500).Real wages grew faster in Canada, but employment fell faster in the United States. Although real exports and import penetration both increased faster in Canada, the time-series variability of the U.S. and Canadian international trade statistics swamps the difference in averages.
13.5 Results of Value and Price Analyses Table 13.4 reports the estimated effects of value-based international trade measures on the bargaining unit employment growth rate in the United States and Canada. The estimated effect of the expected change in the log of real
356
John M. Abowd and Thomas Lemieux
Table 13.3
Comparison of the U.S. and Canadian Collective Bargaining Contract Samples with Selected Summary Statistics (standard deviations in parentheses)
Description Industrial coverage Period covered Bargaining units Total observations Average bargaining unit size Average (%) annual real wage growth rate (size weighted)’ Average (%) annual employment growth rate (size weighted) Average (%) annual industry real shipments growth rate (size weighted) Average (%) annual industry real apparent domestic consumption growth rate (size weighted)b Average (%) annual industry real export growth rate (size weighted)’ Average (9%) annual industry import penetration change (size weighted)d
United States
Canada
All manufacturing 4-digit SICs 1959-84 250 2,515 8,881 1.15 (7.25) - 2.76 (43.5) 3.57 (24.1)
All manufacturing 2- or 3-digit SICs 1968-83 299 1.280
3.17 (23.2)
2.96 (4.43)
.48 (4.21) .55 (3.53)
1,644
2.66 (4.52) -1.91 (16.7) 4.66 (6.55)
1.85 (4.57) .70
(2.11)
Sources: See table 13.1 and the Data Appendix. a Size-weighted statistics were weighted by employment in the bargaining unit. The growth rate of real apparent domestic consumption is multiplied by (shipments - exports)/shipments. T h e growth rate of real exports is multiplied by (exports/shipments). T h e change in import penetration is multiplied by (apparent domestic consumptionkhipments).
shipments on employment growth is much larger for Canada than for the United States, indicating that the basic employment equation is more output sensitive in Canada. For both the United States and Canada, the effect of an expected change in real exports is comparable in magnitude and direction to the same size change in shipments. Employment growth is not affected by the destination of domestic production. The estimated effect of import penetration on employment growth is negative, which is the expected direction from equation (2); however, the magnitude is greater than the magnitude of the effect of a change in real shipments. Employment growth is slowed more severely by import penetration than a comparable change in shipments. The pattern of the estimated effects of unexpected changes in shipments and its components on employment growth is comparable to the effects of expected changes in these variables. In the United States, employment growth is somewhat more sensitive to unexpected changes in exports than to unexpected changes in shipments. Employment growth in the United States is substantially more sensitive to unexpected changes in import penetration than to unexpected changes in shipments generally. Employment growth in Canada re-
357
International Competition and Collective Bargaining Outcomes
Table 13.4
Estimated Effect of Value-based International Rade Measures on the Change in the Logarithm of Bargaining Unit Employment
Independent Variable Expected change in log of industry real shipmentsc Decomposed:d Change in log real apparent domestic consumption Change in log real exports Change in import penetration ratio Unexpected change in log of industry real shipmentsc Decomposedd Change in log real apparent domestic consumption Change in log real exports Change in import penetration ratio Change in log bargaining unit employment during previous contract Change in log real wage rate during previous contract Standard error of equation
R2
Residual degrees of freedom
United States'
Canadab
.311 (.058)
.703 (.124)
.087 (348) ,292 (.289) - 1.408 (.308) ,339 (.038)
,581 (.194) ,733 (.171) -1.176 (.319) ,379
,182 (.034) .645 (.191) - 1.667 (.212) - ,048 (.020) .750 (.136)
,400 (.124) .276 C.099) - ,413 (.214) - ,136 (.034) ,087 (.135)
.403 .I49 2,455
.061 1,007
,160
Sources: See table 13.1 and the Data Appendix. Coefficients (and standard errors in parentheses) were estimated from a sample of 2,515 contracts using weighted least squares with bargaining unit size as the weight. The equation also included an intercept and the expected and unexpected change in both the log of real average hourly earnings and the log of real gross national product. Coefficients (and standard errors in parentheses) were estimated from a sample of 1,019 contracts using the same method and controls as n. a. Coefficients (and standard errors) in this row were estimated from a restricted equation in which this variable replaces its decomposition (the three variables that follow in the table). The components are weighted; see notes to table 13.3.
a
sponds with about the same sensitivity to unexpected changes in real shipments, real exports, and import penetration. Table 13.5 reports the estimated effects of value-based international trade measures on bargaining unit real wage rate growth in the United States and Canada. The estimated effect of the expected change in the log of real shipments on employment growth is positive for the United States but negative for Canada. The finding implies that quasi rents per worker move in opposite directions when expected output growth increases in the manufacturing establishments of the two countries. For the United States, the effect of an expected
358
John M. Abowd and Thomas Lemieux
Table 13.5
Estimated Effect of Value-based International lkade Measures on the Change in the Logarithm of the Bargaining Unit Real Wage Rate
Independent Variable Expected change in log of industry real shipments‘
United States’ .052 (.009)
Decomposed:d Change in log real apparent domestic consumption
Canadab
- ,052 (.025)
- ,179 (.OM)
- .163 (.035)
Change in log real exports Change in import penetration ratio
,010
(.065) ,033
Unexpected change in log of industry real shipmentsc
(.018)
Decomposedd Change in log real apparent domestic consumption Change in log real exports Change in import penetration ratio Change in log bargaining unit employment during previous contract Change in log real wage rate during previous contract Standard error of equation R’ Residual degrees of freedom
- .009 ,058
- ,062 (.025)
(.003) .203 (.021)
.041 (.020) - ,010 (.M) .027 (.007) - ,134 (.028)
,064 ,231 2,455
,033 ,510 1,007
(.030) - .221 (.033) ,004
Sources: See table 13.1 and the Data Appendix. a Coefficients (and standard errors in parentheses) were estimated from a sample of 2,515 contracts using weighted least squares with bargaining unit size as the weight. The equation also included an intercept and the expected and unexpected change in both the log of real average hourly earnings and the log of real gross national product. Coefficients (and standard errors in parentheses) were estimated from a sample of 1,019 contracts using the same method and controls as n. a. Coefficients (and standard errors in parentheses) in this row were estimated from a restricted equation in which this variable replaces its decomposition (the three variables that follow in the table). The components are weighted; see the notes to table 13.3.
change in real exports is comparable in magnitude and direction to the same size change in shipments. For Canada, the direction of the export effect is consistent with the shipments effect, but the magnitude is larger. For the United States, the estimated effect of an expected change in import penetration on real wage growth is negative and very substantially larger in magnitude than the comparable change in real shipments.” For Canada, the estimated effect of import penetration on real wages has sign and magnitude
359
International Competition and Collective Bargaining Outcomes
consistent with the real shipments effect. Increased import competition in Canada is associated with increased, not decreased, real wage rates. The pattern of the estimated effects of unexpected changes in shipments and its components on real wage growth is not generally comparable to the estimated effects of the expected changes. In the United States, real wage growth is less sensitive to unexpected changes in real industry shipments, although the effect is still positive. In Canada, the estimated effect of unexpected changes in real industry shipments on real wage rates is positive. The finding implies that, for both the United States and Canada, quasi rents per worker increase when there is an unexpected increase in output. The size and magnitude of the estimated effect of unexpected changes in real exports are comparable to the estimated effects of expected changes in exports for the United States. For Canada, unexpected increases in real exports are associated with higher real wage rates, which is consistent with the sign and magnitude of the unexpected change in real shipments. For both countries, unexpected increases in real imports are associated with decreased real wage growth. The magnitude of this effect in the United States is substantially larger than the effect of unexpected changes in shipments. Table 13.6 reports the estimated effects of changes in world prices on colTable 13.6
Estimated Effects of World Import and Export Prices on Bargaining Unit Employment Growth and Real Wage Growth in Canada
Independent Variable
Expected change in log real world import price (excluding United States)b Unexpected change in log real world import price (excluding United States)b Expected change in log real world export price (excluding United States)b Unexpected change in log real world export price (excluding United States)b Change in log bargaining unit employment during previous contract Change in log real wage rate during previous contract Standard error of equation
R2
Residual degrees of freedom
Employment"
,186 (.082) ,198 (.046) .249 (.079) ,092 (.050) -.I31 (.031) .I19 (.122) .I62 .063 1,321
Real Wagesa
- ,028
(.018) - ,022
(.010)
- ,022 (.017) - ,045
(.011)
,011 (.007) - ,142 (.026) ,035 .433 1,321
Sources: See table 13.1 and the Data Appendix. a Coefficients (and standard errors in parentheses) were estimated from a sample of 1,331 contracts using weighted least squares with bargaining unit size as the weight. The equation also included an intercept and the expected and unexpected change in both the log of real average hourly earnings and the log of real gross national product. The Laspeyres price index has been reweighted to exclude trade with the United States.
360
John M. Abowd and Thomas Lemieux
lective bargaining outcomes in Canada. The estimated effect of increases in the expected world import and export prices is to increase bargaining unit employment. The same is true of the estimated effect of unexpected increases in the world prices on employment growth. All four of these results are consistent with Canada’s position as a (relatively) small open economy. Increases in the world prices of Canadian exports stimulate Canadian production, which increases employment. Increases in the world prices of Canadian imports also stimulate Canadian production (presumably in response to substitution in consumption away from the imported goods). The estimated effect of world prices (both imports and exports) on real wage rates is negative. For the expected changes in world prices, these estimated effects are consistent with the negative effect of real shipments on real wage rates; however, for the unexpected changes in world prices, the negative estimated effect on wage rates is the opposite of the estimated effect of unexpected changes in real shipments.
13.6 Conclusions We have estimated very comparable bargaining unit level models for the effects of international competition on union employment and wages in the United States and Canada. In the United States and Canada, union employment increases when exports increase by about the same amount as one would predict from the estimated effect of industry shipments on union employment. On the other hand, union employment declines in response to an increase in import competition by substantially more than one would predict from the estimated industry shipment effect and the identity connecting shipments, domestic consumption, imports, and exports. The result suggests that import competition has large employment effects in unionized establishmentslarger than the effects one would predict by mechanically assuming that all imports replace domestic production dollar for dollar. In the United States, real union wage rates are equally sensitive to increased shipments or exports; however, real union wage rates fall more in response to an increase in import penetration than one would predict from the knowledge of the shipment effect on real union wage rates.Ig In Canada, real wage rates move in the opposite direction in response to expected and unexpected changes in shipments, exports, and imports. The estimated export and import effects on real wage rates in Canada are consistent with the shipments finding. The effects of exports on Canadian real wage rates are larger than the effects of imports. The pattern and relative magnitude of the estimated price effects for Canada are very consistent with the estimated value-based effects. The major similarity between the U.S. and Canadian experiences, as summarized by the statistical evidence presented here, is in the employment effects of import competition. For bargaining units in both countries, increased import penetration is associated with very large employment effects. The ma-
361
International Competition and Collective Bargaining Outcomes
jor dissimilarity between the United States and Canada can be found in the effects of expected changes in international competition on real wage rates. For the United States, increased import competition is associated with relatively large decreases in real wage rates, but increased export activity is associated with real wage changes of modest magnitude. For Canada, the import competition effects on real wage rates are modest, but the export effects are relatively large and in the opposite direction of the estimated effects for the United States. Although we have not provided all the pieces required to analyze whether the consumers benefit enough from the increased trade to offset the effects of that trade on the domestic labor market, it is clear from all our analyses that the union employment effects of increased import penetration are substantial. The finding certainly explains the widespread protectionist sentiment within union leadership. Furthermore, for the United States at least, import competition does appear to have created competitive pressures that resulted in lower union real wage rates. In Canada, the openness of the economy makes the domestic labor market susceptible to changes in the world prices of many manufactured goods. The estimated responses of Canadian bargaining units to world price changes are predictable from the small open economy international trade model. Furthermore, the Canadian results using the value-based and pricebased international trade measures are essentially the same. This suggests that, although working with price data is theoretically preferable, the biases associated with the value-based measures may not be severe. Given the similarity of the estimated U.S. and Canadian employment responses to changes in the value-based international trade measures, bargaining units in the United States may now be operating in a more open economy than in previous decades.
Data Appendix United States Data Import and Export Values by Industry
Imports as a percentage of apparent domestic consumption for the period from 1958 to 1984 were derived from the Bureau of Labor Statistics (BLS) Trade Monitoring System and the Bureau of the Census biennial report U S . Commodity Exports and Imports as Related to Output. For the period from 1972 to 198 1, the BLS maintained a collection of times series called the Trade Monitoring System that was designed to provide current information on U.S. imports, the ratio of imports to domestic shipments plus imports, exports, and the ratio of exports to domestic shipments on a detailed industry basis. For a
362
John M. Abowd and Thomas Lemieux
report on the development and uses of these data, see Schoepfle (1982). His appendix contains numerous details of the calculations. For details of the data base construction, see Bennett (1982, available on request from the Department of Labor). We used the NBER Trade and Immigration Data Files (Abowd, in this volume) for the period from 1958 to 1971 and 1982 onward. Output Prices and Quantities by Industry
Industry output was measured at the four-digit SIC level using the value of shipments from the Annual Survey of Manufactures Statistics for Industries and Industry Groups. The basic data file, prepared by Wayne Gray for the NBER, is documented in the discussion of the NBER Trade and Immigration Data Files (Abowd, in this volume). Average Hourly Earnings, Gross National Product, and the Consumer Price Index
Average hourly earnings for private nonagricultural employment, gross national product, and the consumer price index-all urban were all extracted from the CITIBASE machine-readable data file in seasonally unadjusted form. Contract Data
Bargaining unit wage rate and employment information were taken from the data file on 250 major bargaining situations developed by Wayne Vroman from the BLS Current Wage Developments (CWD)printed reports. The realized nominal wage increase over the life of the contract, stated as an annualized percentage rate, was calculated from the periodic wage change reports in CWD and Vroman’s imputation of the average scale wage at the beginning of the contract. For a discussion of the methods, see Vroman (1982, 1984, 1986). The basic situation number list from the BLS was linked to four-digit establishment SICS for the bargaining unit, available from the BLS as unpublished data. The SIC-based trade data were linked to the contract file using the BLS establishment SIC. For a discussion of the link to international trade data, see Vroman and Abowd (1988). The wage rate reports in CWD summarize scheduled fixed increases and realized cost of living adjustments as they occur. The bargaining unit size and average bargaining unit wage rate are reported (by the employer) at the time a new contract settlement was recorded in the basic (confidential) CWD data file, which is used to prepare the summary information on new collective agreements published quarterly. The printed CWD reports the bargaining unit size but not the average wage rate. Canadian Data Import and Export Prices and Values by Industry
The Laspeyres price indices and the trade value measures were derived using CANSIM data supplemented for 1967 and prior years from the Bank of
363
International Competition and Collective Bargaining Outcomes
Canada Review. The system of classification closest to standard industrial classifications (SIC), which is used by Labour Canada to classify the bargaining units, is the System of National Accounts (SNA) classification used to construct the Canadian input-output tables. International trade price and value data are available using the SNA classification; however, all source and destination countries are aggregated. Since we wanted to eliminate trade with the United States from the world price measures, we used the price and value measures available on CANSIM under the import commodity classification (MCC) and the export commodity classification (XCC), which are disaggregated by major countries. Unlike the SIC, the MCC and XCC classifications are systems of classification for products, not for industries. Using the Canadian input-output tables to provide the connection between products and industries, we developed a concordance between two-digit SIC industries and the international trade measures obtained from the MCC and XCC. Data on imports and exports for industrial sectors where the international trade flows are not very substantial are very aggregated. Only an aggregated measure of import (and export) prices and values was available for knitting industries, clothing industries, furniture and fixtures, publishing and printing, and miscellaneous industries. For exports, only an aggregated measure was available for leather industries, fabricated metal, and nonmetallic mineral products. Data for only seven and five aggregated sectors were available for imports and exports before 1968 (in the Bank of Canada Review). The import price index is a combination of transaction prices and unit values. The short description given in CANSIM is as follows: “The Laspeyres price indexes are based on fixed weights derived from shipments 1971 quantities and hence reflect changes in prices alone. Most of the non-end product indexes are based on average prices derived from commodity import value and quantity data. The end product indexes are based on wholesale price indexes from Canadian, U.S. and foreign sources as proxies for import prices. For further details see the September 1976 supplement to the summary of external trade catalogue 65-001” (CANSIM, Statistics Canada 1984a, matrix 003681). The technical documentation may be found in Statistics Canada (1976). The series description for the export prices is as follows: “The Laspeyres price indexes are based on fixed weights derived from shipments 1971 quantities and hence reflect changes in prices alone. Most of the non-end product indexes are based on average prices derived from commodity export value and quantity data. The end product indexes are based on Canadian industry selling price indexes as proxies for export prices. For further information see September 1976 supplement to the summary of external trade catalogue 65-001” (CANSIM, Statistics Canada 1984a; Statistics Canada 1976). Value of imports and exports by country of origirddestination is collected as a part of the System of National Accounts. Combining data from all sources produced the price and value measures for imports and exports for the period from 1961 to 1984.
364
John M. Abowd and Thomas Lemieux
Input and Output Prices and Quantities by Industry Data on output and input price and values by industry were obtained from two publications of Statistics Canada: Real Domestic Product per Sector, 6171 and Gross Domestic Outputper Industry (1978 and 1984 issues). Average Manufacturing Wage Average hourly earnings in manufacturing obtained from the CANSIM University Data Base and the Bank of Canada Review. (See Card 1988). Gross National Product, Unemployment, and the Consumer Price Index Basic monthly, quarterly, and annual time-series data were extracted from the CANSIM University Data Base from 1961 to 1984 and from the Bank of Canada Review thereafter. Contract Data We used 2,258 collective bargaining agreements for 299 bargaining pairs in the manufacturing sector of Labour Canada’s Wage Tape. The wage measure used is a base wage rate. For a description of the data set and of how the wage settlement variable was constructed, see Card (1988). The employment variable provided on the Wage Tape is actually a measure of how many workers were covered by the collective bargaining agreement on the day of the agreement. Inspection of the data suggested that the employment data were substantially contaminated by measurement error. In an effort to reduce those measurement problems, we systematically compared the employment data from the Collective Bargaining Review to the numbers from the Wage Tape for all pairs where the employment was changing by 10 percent or more in absolute value between two agreements at some point of time. In cases of discrepancies between the two numbers, the employment from the Collective Bargaining Review was used. More information is available from the Collective Bargaining Review about the structure of the bargaining pair (e.g., which plants and which union locals are involved). Two levels of correction were performed. The first level consisted of using the number of employees from the Collective Bargaining Review when the only identifiable source of discrepancies between the two data sources was reporting error. Employment data for 175 contracts was corrected on this criterion. Employment from twenty-eight contracts was discarded because of major changes in the definition of the bargaining pair. Second, when the information from the Collective Bargaining Review indicated that the structure of the bargaining pair changed over time, the following rule was applied: if enough information was available from the Collective Bargaining Review to construct a consistent series for a specific pair, such information was used. Otherwise, the changes in employment that could have been explained by changes in the structure of the bargaining pair (e.g., one
365
International Competition and Collective Bargaining Outcomes
local drops off) were eliminated. Employment data for 131 contracts were adjusted, and employment data for 176 contracts were discarded owing to this correction.
Notes 1 . For an overview of the changes in the United States, see Abowd and Freeman (in this volume). 2. For a recent discussion of Canadian labor market responses to increasing global competition, see Riddell (1986b). 3. In 1986, Canada accounted for $45 billion of $2 17 billion total U. S . exports and $68 billion of $370 billion total imports. In 1975, Canada accounted for $22 billion of $108 billion total U.S. exports and $22 billion of $97 billion total U.S. imports (U.S. Department of Commerce 1988,770). 4. The openness measure (imports + exports)/GNP is used by Abowd and Freeman (in this volume) to characterize the growing internationalization of the U.S. labor market. 5. Seventy-five percent of nonoffice employees in manufacturing industries were covered by collective bargaining agreements in 1984 (Kumar 1986). 6. See, in particular, Hufbauer, Berliner, and Elliot (1986). Other examples in the same style as the current research include Heywood (1985) and Kahn (1986). 7. See, e.g., Riddell (1979) and Christofides et al. (1980a, 1980b). These relations are sometimes called “micro Phillips curves” (Hamermesh 1970). 8. For a review of these models, see Farber (1986). 9. When unions and employers engage in industry-wide bargaining, we assume that the multiple employer bargaining unit consists entirely of employers with positive expected quasi rents. When an employer has negative or zero expected quasi rents, we assume that the firm withdraws from the industry-wide bargaining unit. 10. Profitability is measured using quasi rents. It includes the portion of the return on the enterprise that goes to the unionized work force. (See Abowd 1989.) 11. It is difficult to state precisely the definition of number of workers in the bargaining unit. The reason for this is the absence of standards within the Department of Labor’s Office of Wages and Industrial Relations for determining the “number of workers covered’ by a particular collective bargaining agreement. The definition in the text is the usual interpretation of this number. It includes all current employees and those former employees who are on layoff but who could return to work before they lose the right to vote on the contract settlement. 12. See Schoepfle 1982), Vroman and Abowd (1988), and Freeman and Katz (in this volume). Dickens, Tyson, and Zysman (1985) review the employment effects literature. 13. Net revenue is the difference between net sales (gross sales less the value of discounts and other promotions) and the cost of all inputs except labor. 14. For a more complete development of this argument, see Abowd (1989). 15. The world price of Canadian exports stated in terms of the Canadian dollar is equal to the product of the world price stated in terms of a basket of foreign currencies and the exchange rate between those currencies and the Canadian dollar. The Canadian dollar world price can increase either because the world price stated in terms of the trading partner currencies increases at a given exchange rate or because the Canadian dollar depreciates relative to the trading partner currencies at a given trading partner
366
John M. Abowd and Thomas Lemieux
price. In this paper, we do not distinguish between these sources of price change, although we intend to do so in future work. 16. For U.S. results using this decomposition, see Freeman and Katz (in this volume). 17. This finding should be contrasted with the finding in Vroman and Abowd (1988) that nominal wage growth was only weakly affected by import penetration in the same data. 18. The world price effects shown in table 13.6 were estimated using price indices that exclude trade with the United States. The sign and magnitude of the price effects are generally not affected by using a world price index that includes trade with the United States; however, the estimates in table 13.6 are more precise. 19. The finding is consistent with the Freeman and Katz (in this volume) result that the estimated effects of import penetration on average wage rates in the industry is largest for heavily unionized industries in the United States.
References Abowd, John M. 1989. The Effect of Wage Bargains on the Stock Market Value of the Firm. American Economic Review 79, no. 4 (September):774-800. Bank of Canada. Monthly. Bank of Canada Review. Ottawa: Minister of Supply and Services. Bennett, Norman. 1982. Trade Monitoring System, Technical Note, Import Penetration and Export Proportion Data Bases. Washington, D.C.: Bureau of Labor Statistics, Division of Foreign Labor Statistics and Trade, November. Card, David. 1988. Strikes and Wages: A Test of a Signaling Model. NBER Working Paper no. 2550. Cambridge, Mass.: National Bureau of Economic Research, April. Christofides, Louis N., et al. 1980a. A Microeconometric Analysis of Spillovers within the Canadian Wage Determination Process. Review of Economics and Statistics 62(May):213-21. . 1980b. A Microeconometric Analysis of the Canadian Wage Determination Process. Economica 47(May): 165-78. CITIBASE. 1978. Citibank Economic Database [machine-readable magnetic data file, 1946 to present]. New York: Citibank. Cousineau, Jean-Michel. 1987. The Impact of International Trade Shocks on Wage Adjustments in Canada. In Labor Market Adjustments in the Pacific Basin, ed. P. Chinloy and E. Stromsdorfer, 61-78. Boston: Kluwer-Nijhoff. Dickens, William T., with Laura Tyson and John Zysman. 1985. The Employment Effects of International Trade: A Review of the Literature. Report to the Office of Technological Assessment. Berkeley: University of California. Farber, Henry S. 1986. The Analysis of Union Behavior. In Handbook of Labor Economics, vol. 2, ed. 0. Ashenfelter and R. Layard, 1039-90. Amsterdam: NorthHolland. Grossman, Gene. 1982. Import Competition from Developed and Undeveloped Countries. Review of Economics and Statistics 64:271-8 1. . 1986. Imports as a Cause of Injury: The Case of the U.S. Steel Industry. Journal of International Economics 20:201-23. . 1987. The Employment and Wage Effects of Import Competition in the United States. Journal of International Economic Integration, 2, no. 1 (Spring): 1-23.
367
International Competition and Collective Bargaining Outcomes
Hamermesh, Daniel S. 1970. Wage Bargains, Threshold Effects, and the Phillips Curve. Quarterly Journal of Economics 85(August):501-17. Heywood, John S. 1985. Imports and Domestic Wage Levels. University of Michigan, Department of Economics, November. Hufbauer, Gary C., Diane T. Berliner, and Kimberly A. Elliot. 1986. Trade Protection in the United States; 31 Case Studies. Washington, D.C.: Institute for International Economics. Kahn, Shulamit. 1986. Trends in Union Membership in the Postwar Period: The Case of the ILGWU. IRRA Proceedings of the Thirty-Eighth Annual Meeting. Madison, WI: IRRA, 279-86. Kumar, Pradeep, et al. 1986. The Current Industrial Relations Scene in Canada. Kingston, Ont.: Industrial Relations Centre, Queen’s University. Labour Canada. Monthly. Collective Bargaining Review. Ottawa: Minister of Supply and Services. Riddell, W. Craig. 1979. The Empirical Foundations of the Phillips Curve: Evidence from Canadian Contract Data. Econometrica 47(January): 1-24. , coordinator. 1986a. Adapting to Change: Labour Market Adjustment in Canada. Toronto: University of Toronto Press, in cooperation with the Royal Commission on the Economic Union and Development Prospects for Canada. . 1986b. Dealing with InJation and Unemploymentin Canada. Toronto: University of Toronto Press, in cooperation with the Royal Commission on the Economic Union and Development Prospects for Canada. , coordinator. 1986c. Labour Management Cooperation in Canada, Toronto: University of Toronto Press, in cooperation with the Royal Commission on the Economic Union and Development Prospects for Canada. Schoepfle, Gregory. 1982. Imports and Domestic Employment: Identifying Affected Industries. Monthly LaborReview 105, no. 8 (August): 13-26. Statistics Canada. 1976. The 1971-based Price and VolumeIndexes of Canada’s External Trade. Ottawa: Minister of Supply and Services, December. . 1978. Real Domestic Product per Sector: 61-71, Ottawa: Minister of Supply and Services. . 1984a. CANSIM UniversityBase. Ottawa: Minister o f Supply and Service. . 1984b. Gross Domestic Output per Industry. Ottawa: Minister of Supply and Services. . 1984c. The Input-Output Structure of the Canadian Economy in Current Dollars, 1971-1981. Ottawa: Minister of Supply and Services. . 1987. cANsrMMainData Base. Ottawa: Minister of Supply and Services. U.S. Department of Commerce. Bureau of the Census. 1988. Statistical Abstract of the United States. Washington, D.C.: U.S. Government Printing Office. . Annual. Annual Survey of Manufactures Value of Product Shipments. Washington, D.C.: U.S. Government Printing Office. . Biennial. US.Commodity Exports and Imports as Related to Output. Washington, D.C.: U.S. Government Printing Office. U.S. Department of Labor. Bureau of Labor Statistics. Monthly. Current WageDevelopments. Washington, D.C.: U.S. Government Printing Office. Vroman, Wayne. 1982. Union Contracts and Money Wage Changes in U.S. Manufacturing. Quarterly Journal of Economics 97(November):57 1-94. . 1984. Wage Contract Settlements in U.S. Manufacturing. Review of Economics and Statistics 66(November):661-65. . 1986. Union Wage Settlements, Incomes Policy and Indexation. Washington, D.C.: Urban Institute. Vroman, Wayne, and John M. Abowd. 1988. Disaggregated Wage Developments. Brookings Papers on Economic Activity, no. 1:3 13-46.
This Page Intentionally Left Blank
14
Male Immigrant Wage and Unemployment Experience in Australia John J. Beggs and Bruce J. Chapman
This paper analyzes the relative labor market success of immigrants using the 1 percent sample of the 1981 Australian Census, data that have previously been exploited for this purpose. Our major contribution lies in the adoption of flexible estimation techniques. This allows two fundamental insights into the operation of the Australian labor market, both of which are related to the role of education as a determinant of immigrant success. First, it is important to allow the wage effects of labor market experience and ethnicity to differ by education levels. Second, it is clear that the role of schooling in the determination of unemployment can be adequately understood only by estimating relationships in a disaggregated way. The clear and consistent result from our methods is that, relative to similarly educated natives, immigrants with the highest levels of education receive the lowest wages and experience the highest unemployment. We do not explore fully the reasons for these outcomes; we do, however, touch on some possibilities. Apart from the insights allowed through disaggregated estimation of the role of schooling, the paper offers the following technical innovation. For one of the first times, the results of non-parametric estimations of wage functions are presented. The major benefit of this approach is the flexibility afforded, but the method is not unambiguously superior to OLS regression analysis. It is at least clear that useful (graphical) interpretation of casual mechanisms may be highlighted through the use of non-parametric techniques. The data used are cross-sectional; consequently, estimations could be contaminated by important problems associated with the unobserved ability of immigrants. Because this possibility is highly relevant to interpretation of results, some effort is directed to understanding the empirical significance of John Beggs is associate director, Daiwa Securities Australia Ltd., responsible for the management of the Funds Management Division and Research Group. Bruce Chapman is senior fellow, Economics Department, Research School of Social Sciences, Australian National University.
369
370
John J. Beggs and Bruce J. Chapman
this potential. It is to the role of unobserved ability in cross-sectional data that we turn first. 14.1 Examining the Usefulness of Cross-sectional Data
Several difficulties arise in the analysis of relative immigrant labor market outcomes using cross-sectional data. The concern is that immigrant cohorts differing in length of residency are also dissimilar in terms of unobserved ability or motivation. If this is the case, some parameters of major interest, such as the elasticities found between length of residency and both wages and unemployment, cannot be estimated without bias. There are two obvious dimensions to the unobserved ability issue noted above. First, differences in economic conditions or government policy will undoubtedly affect the average quality of the immigrant pool entering in particular years. Second, the act of immigration is reversible, and a sizable (but variable) proportion of immigrants from different countries eventually leaves the new country. If the probability of remigration is correlated with (unobserved) immigrant quality, cross-sectional data will misrepresent underlying relationships, at least as indicators of expected immigrant success. These issues have been examined with U.S. data by both Borjas (1985) and Chiswick (1986). Borjas argues that considerable variation exists in the ability of immigrant cohorts, suggesting bias in the interpretation of the effect of period of residence on wage growth. Chiswick’s study implies that remigration does not markedly affect wage estimates. Since the wage and unemployment analyses reported in this paper use a single cross section of data, it is pertinent to attempt to establish the empirical significance of the unobserved quality issue in the Australian context. Both the quality of entering cohorts and the effect of remigration are examined below using wage data. It is important to stress, however, that the unobserved ability issue applies as much to unemployment as it does to wages, the focus on the latter at this stage being a consequence only of data availability. The methodology suggested by Borjas (1985) to examine immigrant quality may be applied to Australian data. The approach compared predicted wages of different and similar immigrant cohorts to those of natives and can be used to gain insight into both the returns to local experience for a particular cohort of immigrants and the differences between immigrant cohorts in unmeasured wage ability. The data available allowing a similar investigation to that of Borjas’s were drawn from the (Australian National University) 1973 Social Sciences Survey of Australian male residents aged 30-64, which includes a sample of about eighteen hundred wage- or salary-earning individuals, and the 1/100 Census tapes from 1981. The results of the analysis are reported in Beggs and Chapman (1988). They imply that the likelihood of cross-sectional data being
371
Male Immigrant Wage and Unemployment Experience in Australia
markedly contaminated by significant changes over time in unobserved variables is small. The other major possibility rendering cross-sectional analysis questionable concerns the effects of remigration on the average unobserved ability of remaining cohorts. In Australia, this issue is potentially important, particularly if the question concerns the differential relative labor market outcomes of immigrants from English-speaking countries (ESM) and immigrants from nonEnglish-speaking countries (NESM), because marked differences exist in the probability of remigration, as shown in table 14.1. These data, while not ideal as reflections of remigration probabilities over the period 1959-82 (since some of the arrivals may still depart and some of the departures may yet return), are useful as broad indications of the empirical dimensions of the process. They suggest that about 20-30 percent of ESMs are likely to leave but that only 3-6 percent of NESMs do likewise. Two salient points are (1) that the proportion of the former group leaving is high, implying that, if a relationship exists between unobserved ability and remigration, the use of cross-sectional data for this group is suspect and (2) that the problem is much less likely to be the case for NESMs. The biases introduced by remigration could go either way, but they are usually believed to have the following pattern: the least successful immigrants eventually return home or seek economic success elsewhere. Thus, the average ability of identified immigrants could be expected to increase with period of residency, and the coefficient on this variable is thus biased upward in absolute size in both wage and unemployment estimations. Alternatively, successful immigrants are more capable of moving because they have accumulated sufficient wealth. It follows that an unambiguous prediction of the direction of bias is not forthcoming. Immigrants’ decision to remain in or leave new countries cannot be analyzed only on the basis of economic performance in the host country since what will be of importance is expected income in alternative countries. Even if success has been relatively low in the new country, this does not necessarily imply an increased incentive to return. This complexity might render questionable one of Chiswick’s (1986) tests of the return migration proposition. He argues that, if immigrants from counTable 14.1
Arrivals and Departures of Male Immigrants, 1959-82
Country of Birth United Kingdom and Ireland New Zealand Italy Greece
Amvals
Departures
DeparturesiArrivals
1,086,054 118,157 18,736 159,763
231,810 33,878 10,660 5,314
,213 ,287 ,0569 .0336
Source: Obtained from the Department of Immigration and Ethnic Affairs.
312
John J. Beggs and Bruce J. Chapman
tries with relatively high period of residency coefficients are also those immigrants who from other evidence are less likely to return, the self-selection process is empirically unimportant. The indications are that this is the case (Cubans, e.g., have very high rates of return to residency but few returnees), but the noted theoretical ambiguity casts doubt on the result. One empirical way to get at the return migration issue is to examine the residuals of the wage equation. They should exhibit skewness related to job tenure, the direction of which will be determined by whether or not high- or low-ability persons eventually leave. Negative skewness implies that the top part of the intangible ability distribution (the residual) shortens with tenure, that is, that the more able increasingly have left the sample. The opposite is the case if lower-ability persons are more likely to leave as tenure increases. Given the aggregate data, systematic biases from attrition imply the following. If lower-ability immigrants are generally more likely to return, skewness of the residuals with period of residency (PER) should be positive for both ESMs and NESMs and greater for the former group. On the other hand, if higher-ability immigrants are generally more likely to return, skewness of the residuals with PER should be negative for both ESMs and NESMs and absolutely greater for the former group. The finding that skewness does not exist implies that the attrition process is unrelated to unmeasured ability, at least as reflected by the residual of the wage equation. The test took the form of estimating the following equation:
RES3
+ bPER + cPER2 + dYOS + e,,
a
=
where RES3 is the cube of the OLS residuals obtained from the 1981 crosssectional estimations reported in Beggs and Chapman (1986), and YOS is years of schooling, included as a control. The results are presented in table 14.2. The results imply that there is no evidence from the wage data of the 1981 Census of an important relationship between unmeasured ability and the likelihood of remigration of either ESMs or NESMs. The nature of the tests suggests that this cannot be taken as strong confirmation of the lack of correlation between ability and remigration. The appropriate conclusion is that the test
Table 14.2
Skewness Tests of the Return Migration Hypothesis
ESM: ,120 - .00311PER (SO) (.25) NESM:
+
,206 - .0106PER (1.08) (.74)
.0000143PER2 - .00554YOS (.048) (.34)
+
.OOOPERZ - .00345YOS ~27) (.3u
Nore: Absolute r-statistics in parentheses.
373
Male Immigrant Wage and Unemployment Experience in Australia
has not uncovered any information in Australia with respect to remigration that places in question the validity of the usual estimations based on crosssectional data. In summary, the data and tests on wage relationships outlined above reveal that, for Australia, there is no evidence that major differences in the unobserved ability of immigrant cohorts significantly undermine the value of wage and unemployment analyses based on cross-sectional data, a point more obviously true for ESMs than NESMs. If immigrants entering in particular years are of considerably different quality than others, or if persons remigrating come from either end of the ability distribution, we have not been able to uncover powerful evidence for these effects. Given this as background, we now proceed to an examination of immigrant wage and unemployment outcomes as documented in the 1981 Australian Census under the assumption that tests based on such a cross section are at least moderately sound.
14.2 Immigrant Relative Wage Experience in Australia The first multiple regression analysis of immigrant earnings in Australia was conducted by Haig (1980), who used data collected by the Australian Bureau of Statistics for the Henderson Inquiry into Poverty in Australia 1973. The study attempted to determine the role of endowments and discrimination in explaining immigrants’ relative earnings, with a conventional application of the methodology popularized by Oaxaca (1973) and Blinder (1973). Control variables used were, among others, age, hours, education, sex, and country of origin. Because Haig restricted slope coefficients to be identical for men and women, assumed hours worked to be exogenous, and used age as an experience proxy, the results should be treated with caution. Nevertheless, for the purposes of the present investigation, it is of interest to note his finding that immigrants generally, and immigrants from southern Europe in particular, had relatively flat age-earnings profiles in 1973. Period of residency was apparently an insignificant earnings determinant. More disaggregated approaches have been adopted by Stromback (1984), Chiswick and Miller (1985), and Beggs and Chapman (1986) using the 1 percent household sample of the 1981 Australian Census. The analyses used similar specifications and, thus not surprisingly, drew similar conclusions-that ESMs experienced wage structures similar to the Australian born but that other immigrants received relatively low returns to schooling and experience. The data reveal no catch-up for NESMs. These results can be interpreted only imperfectly in the context of the transferability of investment in human capital: while they imply that skills acquired in like countries are rewarded more highly than skills acquired in dissimilar countries (non-English-speaking countries), they also suggest that those immigrants starting with a wage disadvantage relative to the native born remain with a wage disadvantage over their Australian working lives. This finding is
374
John J. Beggs and Bruce J. Chapman
at variance with some U.S. and Canadian conclusions derived from crosssectional analyses. While it is important to acknowledge that cross-sectional analyses implicitly impose the assumption that unobserved ability is uncorrelated with immigrant length of residency, the estimations presented in section 14.1 above suggest that, for Australia, this point is not of great empirical significance. What is, perhaps, of much more importance in an understanding of immigrant wage adjustment processes is the role of education. In particular, previous research has imposed the restriction that relative wage returns and rewards to labor market experience do not vary with education. The major contribution here is to relax these assumptions, an approach revealing quite different insights into immigrant wage outcomes to those reported above. As well as allowing relative wages and experience effects to vary with schooling, an innovation of our approach is the use of nonparametric techniques to estimate the wage functions. The choice of the kernel smoothing (nonparametric) technique was motivated by the following issue. The approach obviates the need to prespecify the precise analytic functional relation between the explanatory variables and the wage rate, and it is this major benefit that we have taken advantage of. Also, the technique is ideally suited to large samples because of its convergence properties (Bierens 1985) and is ill suited to small data sets. The nonparametric approach is not without costs. Specifically, the method lacks the familiar summary of a model as a small number of estimated parameters, and there is no easy access to the usual test statistics, although pointwise confidence intervals may be computed. Also, it is not quite appropriate to estimate the expected wage rate given values of more than just a few of the regressor variables of interest. In short, the approach trades off flexibility and complexity, and in its adoption we have chosen more of the former. The basic model is that hourly income depends on individuals’ measured human capital characteristics. In the analysis reported below, we allow for the effect on income of years of schooling (YOS), potential years in the labor market (LMX), years of Australian schooling (AYOS), and years of potential labor market experience before migrating to Australia (ALMX). As well, there is a separation of the sample according to country of origin. Three male groups are distinguished: Australian born (AUST), ESM, and NESM. The wage equations are (2)
wAUST
= fAIJST
(3)
WE,,
= f,,,
(4)
W,,,,
= f,,,,
>
LMX)
7
(YOS, LMX, AYOS, ALMX), (YOS, LMX, AYOS, ALMX),
and the statistical characteristics of the data are reported in Appendix table 14A. 1. The results of the nonparametric regressions are presented in graphic form, the interpretation of which is straightforward. The computed confidence
375
Male Immigrant Wage and Unemployment Experience in Australia
intervals are not shown for reasons of clarity, their size being such as not to affect conclusions. We consider four education levels, 8, 10, 12, and 14 years of completed schooling, with the results showing the cross-sectional relationship between wage and age (LMX + YOS 5) for natives, ESMs, and NESMs. The immigrants are hypothetically given AYOS = 0 and ALMX = LMX. That is, the relationships for these groups should be interpreted as representations of the experience of male individuals entering Australia immediately after completing their schooling abroad. They are shown in figures 14.1. The data of figure 14.la reveal the following. First, natives with very low schooling earn, overall, lower hourly incomes than all immigrants, although there is no obvious difference between natives and NESMs until about age 36, after which NESMs have higher incomes. Second, ESMs experience higher incomes at all ages than the other two groups, with the difference being maximized for those in their mid-forties. Third, and related to the above, at relatively young ages the age-earnings profiles are steeper for immigrants than natives and steeper for ESMs than NESMs. For males with 10 years of schooling (fig. 14.lb), the following points are pertinent. (1)Unlike the situation for the lowest level of schooling, natives earn higher incomes than NESMs at all ages. (2)Similar to the results for the lowest level of schooling, ESMs earn higher incomes than natives at all ages, with the difference being the greatest for persons in their late forties. From figure 14.lc, it is apparent that, for those with 12 years of schooling, native incomes are higher than those of NESMs and that this advantage is greater than is the case of persons with 10 years of formal education. Also, apart from those younger than 30 years-where income per hour is about the same-natives earn more than ESMs. As is the case with other schooling levels, ESMs earn higher incomes than NESMs, and the profile of the former is relatively steep, at least until about age 50. Figure 14.ld reveals that, at high levels of schooling (14years), natives earn substantially higher incomes than immigrants. It is important to note that the relative income advantage of the former group is apparently greater than was the case for those with 12 years of schooling. As is the case with the groups having 10 and 12 years of schooling, NESMs earn less than ESMs and natives at all ages and have flatter profiles. The figures, considered sequentially, reveal a striking pattern: as education increases, so too does the relative income advantage of natives. At lower levels of schooling, immigrants earn the same as, or more than, natives, but, at the highest levels of schooling, this situation is reversed. Also, NESMs earn lower incomes than ESMs for all levels of schooling and generally have flatter age-earnings profiles. These findings highlight the importance of disaggregated analysis of immigrant relative wage outcomes, in particular as regards education. Most important for the analysis, the nonparametric approach reveals the
+
376
John J. Beggs and Bruce J. Chapman 8 Years
G
5
L
+
,
.
:
:
:
:
~
01
Schooling
:
:
.
, ’
.
:
.
,
*
2 4 2 6 28 3 0 3 2 3 4 3 6 3 8 4 0 4 2 4 4 4 6 4 8 5 0 5 2 5 4 56 58 6 0 62
Age (years)
10 Years 01 Schooling
8 2 7 8 7 8 76 Wage 74 (S/hrl 72
7 6 8 6 6 - : : : , i . i . : . i i . , 24 2 6 2 8 30 3 2 3 4 3 6 3 8 4 0 4 2 4 4 4 6 4 8 5 0 5 2 5 4 5 6 5 8 6 0 6 2 ~
Age (years)
1 2 Years 01 Schoolhg
Wage (Whr)
-.-0-
NATIVES ESM
1511
JO’
/
10 --
9.-
1 0
‘0
0 5 ..
9 5 --
.’ ./.-.-.-.\,‘.
o/o-o-o-o-o-~-o-o
--
/
/
a ’,.
/’
a ’,.
&’
.L.-
Fig. 14.1 Nonparametric regression estimates of average hourly wage rate
377
Male Immigrant Wage and Unemployment Experience in Australia
importance of flexible estimation, two salient points being clear. (1) It is obviously not the case that returns to Australian labor market experience are identical for different schooling levels, the assumption usually imposed in OLS wage estimations. (2) Most important, interpretation of which immigrant groups are apparently relatively disadvantaged in wage outcomes is inadequate without considering the effect of schooling. Interestingly, the essence of this story is replicated with respect to unemployment, an investigation to which we now turn.
14.3 Immigrant Relative Unemployment Experience in Australia Analyses of the relative unemployment experience of immigrants in Australia (Miller 1986; Inglis and Stromback 1986) have, as with wage determination research, imposed parsimonious functional forms. In particular, the approaches have not allowed the effect of unemployment determinants to vary between groups, a method that restricts the effects of schooling to be the same irrespective of ethnicity. In the analysis reported below, we allow the effects of schooling (and other variables typically associated with unemployment) to vary between AUST, ESM, and NESM. Although the estimations constrain Australian labor market experience to have the same effect on unemployment for different schooling levels, they nevertheless offer compelling evidence on the effect on relative immigrant labor market outcomes of education entirely consistent with the wage analysis reported above. Search theory is used to motivate the empirical approach adopted, and we have described the advantages and problems of this framework elsewhere (Beggs and Chapman 1990). The model allows only one prediction unambiguously (in essence because the distribution of the wage offer curve will not be the same between groups). This is that the relative unemployment rate is lower for unskilled immigrants than unskilled natives. The prediction follows from our critique of the theory but is not crucial to understanding the results following. The simple reduced form derived from search theory is (5)
+ G,LMX, + 6,LMSf + 6,YOS, + S,YOS, + G,ALMX, + 6,ALMXf + G,MAR, + G,LANGD, +
P(u), = 6,
E&,
where P(u),equals one if the individual is unemployed and zero if employed, and the explanatory variables are potential labor market experience (LMX); potential Australian labor market experience (ALMX); a dummy variable equal to one if currently married, spouse present, and zero otherwise (MAR); and a language dummy variable equal to one if the respondent has Englishlanguage problems and zero otherwise (LANGD). The probit estimations were run separately for natives, ESMs, and NESMs, a procedure allowing important insights into the role played by education. Statistical characteristics of the data, random samples from the 1 percent of
378
John J. Beggs and Bruce J. Chapman
males in the 1981 Census, are shown in Appendix table 14B. 1, and the results are reported in table 14.3. From table 14.3, the strongest single effects on unemployment are from schooling and marital status. The potential labor market experience variables exhibit low individual t-statistics. This reflects the fact that LMX, LMX2, and ALMX2 are highly correlated in the sample, implying that it is difficult to distinguish their individual contribution to unemployment. Consistent with this interpretation, the log-likelihood ratio statistics indicate that only the joint test of the null hypothesis,,p , =’,,p , = 0 for NESM fails to be rejected by the data. The highly nonlinear form of the probit probability model hampers straightforward comparison of immigrant and native coefficients. Interpretation of the main relationships is facilitated through consideration of figure 14.2. To allow comparisons with the results reported in section 14.2, each section of the figures corresponds to a different level of total schooling, with the levels chosen being 8, 10, 12, and 14 years. AYOS and LANGD are set equal to zero, and MAR is set equal to one. Three categories of workers are considered: AUST = Australian born; ESM = immigrants from English-speaking countries with no labor market experience before entering Australia; NESM = immigrants from non-English-speaking countries with no labor market experience before entering Australia. Table 14.3 Variable Intercept LMX LMX2 YOS AYOS
Probit Estimates of Probability of Unemployment Australian Born
ESM
NESM
,359 (.95) - .0213 (1.15) .OOO254 (. 76) - ,129 (5.81)
,397 (1.23) - ,0533 (2.96) .00104 (3.30) - ,0773 (4.20)
- .568 (6.38)
(.OO932) - ,0252 (1.92) .OOO299 (.96) - ,559 (6.67)
- ,630 (2.2) ,00137 (.075) .ooOo770 ~25) - ,0335 (2.72) - ,0134 (1.16) - ,0250 (1.71) .oOo240 (.61) - ,468 (5.18) ,123 (1.62)
ALMX ALMX2 MAR LANGD
Note: Absolute t-statistics in parentheses.
- .OOO104
379
Male Immigrant Wage and Unemployment Experience in Australia
The figures reveal some important similarities. First, ESMs have lower probabilities of unemployment than do NESMs, for all levels of schooling and preimmigration work experience. From theory, this implies that the greater search costs of the latter group are outweighed by other job search factors related to ethnicity. Second, as period of residence increases, the gap between immigrant and native unemployment changes only very slowly. In general, over the range of normal life-cycle labor market experience, there is no crossover. Most important for the theme of this paper, it is clear that, as the level of schooling increases, so too does the relative immigrant unemployment rate. From figure 14.2a, immigrants with low levels of schooling (8 years) have lower probabilities of unemployment than natives over the life-cycle range of labor market experience. On the other hand, at the highest level of schooling considered (fig. 14.2d), immigrants have higher probabilities of unemployment than like natives over the life-cycle range of labor market experience. Clearly, and interestingly, the result is the same as that revealed for wages: as education increases, the labor market position of immigrants relative to natives systematically deteriorates. The important upshot of these results is that it is not possible to understand influences on relative unemployment status without distinguishing the differential effects of schooling. As with the wage analysis, existing research has missed an important point by the restriction of education effects. We offer some tentative conjectures on the results below.
14.4 Interpretation Through a disaggregated analysis of immigrant relative wages and unemployment, a distinct phenomenon has been revealed. For these labor market outcomes, it is apparent that, relative to like natives, immigrants with low levels of education fare well. However, as years of schooling increase, immigrants’ relative labor market success decreases. At the highest level of education considered, immigrants both earn less and have a higher probability of unemployment than is the case for natives. For both Australian and other research, it is important to note that our estimation techniques have highlighted the potential for misinterpretation inherent in existing approaches. The interesting and difficult challenge is to explain satisfactorily the consistent relationships uncovered between relative educational status and immigrant labor market outcomes. Tentatively, we offer four possibilities, perhaps in order of importance. First, and most obvious, is the issue of accreditation of educational qualifications. This could result from two possibilities. (1) Australian employers, if risk averse and with less than full information about the value of overseas schooling, systematically devalue immigrant formal training. (2) Local
P
m
I
m
,
, h
,
‘D
,
, “Y
m N
-
R
L
9 0 _1
382
John J. Beggs and Bruce J. Chapman
employers, trade unions, or professional agencies act in such a way as to protect domestic special interest groups. There is considerable anecdotal (Iredale 1988) and other (Chapman and Iredale 1990) evidence for this perspective. Second, a lower valuation of overseas schooling at high levels of schooling may result because education is less transferable internationally at higher levels. In other words, education is positively correlated with the acquisition of country-specific skills. To take an extreme example, consider the transferability of accountancy qualifications relative to the transferability of streetsweeping skills. In the former case, a high level of country-specific knowledge, such as the understanding of tax and company law, is presumably required. For the latter, sweeping a street in Rome is probably very similar to sweeping a street in Melbourne, and such work would require very little understanding of Australian institutions or the legal system. A third possibility is that of an unobserved variable, motivation or ability, and its relation to immigrant selection procedures. If low-education immigrants are more likely to be selected by immigration authorities if they are particularly work oriented, it follows that relative to the native born their Australian labor market outcomes would be favorable. In this view, immigration authorities are trading off work orientation or talent for formal qualifications in selection of applicants. This possibility implies a shortcoming of the estimation techniques in that the model may be misspecified because of omitted variables and their correlation with education. A similar econometric issue motivates the last explanation of results. This is that the quality of schooling is actually higher in Australia than overseas, or at least is perceived to be so by local employers, a point obviously related to the first possibility offered. Thus, as schooling increases and the quality issue becomes more important, more highly educated immigrants will have their overseas credentials increasingly unrewarded. As noted, we do not have strong priors as to which possible explanation of results is most compelling, the goal having been to demonstrate the importance of more flexible estimation of immigrant relative labor market outcomes than has so far characterized the literature. The very important finding, unrecognized in other research, is that immigrant labor market outcomes (at least in Australia) become more adverse-relative to like natives-as schooling increases. One challenge highlighted is the explanation of the consistency of results found for the role of education. Another is to determine whether the systematic relationships discovered also exist for other countries. If they do, that surely implies the usefulness of a reorientation of research generally.
383
Male Immigrant Wage and Unemployment Experience in Australia
Appendix A Table 14A.l
Statistical Characteristics of the Data for Wage Estimation
Variable
AUST
ESM
NESM
Years of schooling
22.78 (2.45) 25.61 (12.19)
11.13 (3.30) 27.36 (1 1.68) 17.80 (9.52) 1.82 (4.06) 262.95 3,083
Weekly income (dollars)
301.90
12.12 (2.47) 25.91 (1 1.93) 15.67 (10.28) 1.73 (4.07) 314.39
Number of observation
10,532
2,917
Years of labor market experience Years of Australian labor market experience Years of Australian schooling
Note: Means with standard deviations in parentheses.
Appendix B Table 14B.l
Statistical Characteristics of the Data for Unemployment Estimation
Variable YOS (years) LMX (years) MAR ALMX (years)
English-SpeakingCountry Born
Non-English-SpeakingCountry Born
11.65 (2.42) 25.54 (12.22) .78
12.12 (2.46) 25.87 ( 1 1.94) .79 15.68 (10.28) 1.72 (4.06)
3.44
4.36
11.15 (3.33) 27.33 ( I 1.63) .84 17.80 (9.53) 1.78 (4.17) .44 4.60
3,634
3,668
3,607
Australian Born
AYOS (years) LANGD Aggregate unemployment rate (%) Number of observations
Note: Means, standard deviations in parentheses.
384
John J. Beggs and Bruce J. Chapman
References Beggs, J. J., and B. J. Chapman. 1986. Immigrant Wage Adjustment in Australia. Paper presented to the NBER Labor Studies Summer Institute, Cambridge, Mass., August. . 1988. Immigrant Wage Adjustment in Australia: Cross-Section and Time Series Estimates. Economic Record 64(September): 161-68. . 1990. Search Efficiency, Skill Transferability and Immigrant Relative Unemployment Rates in Australia. Applied Economics 22(2):249-60. Bierens, J. 1985. Kernel Estimators of Regression Functions. Research Memorandum no. 8518. University of Amsterdam, Department of Economics. Blinder, A. S. 1973. Wage Discrimination: Reduced Form and Structural Estimates. Journal of Human Resources 8(4):436-55. Borjas, G. J. 1985. Assimilation, Change in Cohort Quality, and the Earnings of Immigrants. Journal o f h b o r Economics 3(0ctobe.r):463-89. Chapman, B. J., and Robyn Iredale. 1990. Immigrant Qualifications and Relative Wage Outcomes in Australia. Centre for Economic Policy Research Discussion Paper No. 240, Australian National University (August). Chiswick, B. R. 1986. Human Capital and the Labor Market Adjustment of Immigrants: Testing Alternative Hypotheses. Research in Human Capital and Development 4: 1-26. Chiswick, B. R., and P. W. Miller. 1985. Immigrant Generation and Income in Australia. Economic Record 61(June):540-53. Haig, B. D. 1980. Earnings of Immigrants in Australia. Journal of Industrial Relations 22(3):254-74. Inglis, P., and T. Stromback. 1986. Migrants’ Unemployment: The Determinants of Employment Success. Economic Record 62( 178):310-24. Iredale, R. 1988. The Recognition of Overseas Qualifications and Skills. Canberra: Office of Multicultural Affairs, Department of Prime Minister and Cabinet, AGES, July. Miller, P. W. 1986. Immigrant Unemployment in the First Year of Australian Labour Market. Economic Record 62(March):82-87. Oaxaca, R. 1973. Male-Female Wage Differentials in Urban Labor Markets. International Economic Review 14(3):693-709. Stromback, T. 1984. The Earnings of Migrants in Australia. Conference Paper no. 46. Canberra: Bureau of Labor Market Research.
15
Why Are Low-skilled Immigrants in the United States Poorly Paid Relative to Their Australian Counterparts? Some of the Issues Illustrated in the Context of the Footwear, Clothing, and Textile Industries R. G. Gregory, R. Anstie, and E. Klug
Many low-skilled immigrant workers in the United States are poorly paid relative to their Australian counterparts. Consider, for example, average earnings of foreign-born male workers in the textile, clothing, and footwear (TCF) industries relative to average earnings of all males in full-time employment. On this relative basis, foreign-born male workers in TCF in the United States earn 40 percent less than they would earn in Australia. Foreign-born females earn 30 percent less. Can these low earnings in the United States be explained by economic factors, such as different levels of human capital or the relative demand and supplies of low-skilled labor in each country? Or are they, as we argue, the outcome of different labor market institutions?' The Australian economy has three institutional features that may increase earnings of low-skilled immigrants. First, there is a high degree of trade union membership. Approximately 49 percent of all employees in Australia belong to trade unions, and among immigrants the proportion is even higher. For example, 75 percent of male employees born in Yugoslavia belong to a trade union, as do 73 percent of those born in Greece. Among female employees from these countries, 75 and 69 percent, respectively, belong to a union. The union movement in also strong in TCF industries, with membership well over 50 percent. In the United States, trade union representation is much lower. In 1984, trade union membership in manufacturing was 27.8 percent (51.2 percent in Australia) and for women, in aggregate, as low as 14.6 percent (43.0 percent in Australia). Second, the Australian labor market is heavily regulated by a system of federal and state tribunals that set minimum wages for each occupation: the pay of university professors is fixed, as is that of sewing machine operators, R. G. Gregory is professor of economics at the Australian National University. R. Anstie is a research fellow at the Centre of Policy Studies at Monash University. E. Klug is a research assistant in economics at the Australian National University.
385
386
R. G. Gregory/R. Anstiem. Klug
laborers, cutters in a clothing factory, and so on. These rates of pay are called awards and are legally enforced minimums. Where awards are determined by the federal tribunal, they are set on a nationwide basis. For example, all clothing cutters, irrespective of the factory in which they work, will be covered by the same award rate of pay. Employers may pay rates over the award rates, and often do so, but most workers receive the award rate of pay for the job.* It is widely believed that these two institutional features exercise a considerable effect on pay relativities. It is often argued, for example, that tribunal judgments, reflecting trade union views, have compressed the pay structure, particularly with regard to the low paid and low skilled, who have had their pay lifted relative to the average (Hughes 1973; Norris 1980; Gregory, Daly, and Ho 1986). The extensive trade union membership among immigrants and the low skilled effectively monitors compliance so that most of the labor market is directly affected by tribunal decisions. In May 1983, for example, 83.6 percent of male employees and 89.7 percent of female employees were covered by award rates of pay, Widespread trade union membership has meant that there is little opportunity for the development of an uncovered sector where wages and conditions are less attractive. The third institutional feature that may affect earnings of the low skilled is that, throughout this century, Australian governments have levied tariffs and quotas on imports to protect manufacturing jobs. It is well known that trade barriers can affect income distribution. Early developments of the StolperSamuelson theorem-tariffs that protect the output of labor-intensive industries can increase the absolute rewards to labor-can be found in the Australian literature just after the turn of the century. Indeed, Stolper and Samuelson (1941) refer to their theorem as the Australian case for tariff protection. The United States also imposes tariffs and trade interventions that do reduce TCF imports and may protect low-skilled labor but that are not as high as those in Australia. These three institutional features of the Australian economy extend to all industries. In this paper, however, we focus on the TCF sector because it illustrates, in the clearest manner, the effect of institutional differences in each country. Our comparison with the United States shows that Australian labor market institutions, supported by trade policy, have increased the pay of lowskilled TCF workers relative to the community average. The wage tribunals have encouraged “comparative wage justice” whereby workers of similar skills and responsibility are paid the same rate of pay regardless of where they work. 15.1 Background
Just before World War 11, Australia had a population of 6.9 million; 13.6 percent were born overseas, and, of these, 80 percent were of British origin. The threat to Australia during World War I1 quickly led to the realization that
387
Low-skilled Immigrants in the U.S. and Australia
we could no longer depend on the United Kingdom for military protection and that there was a need to be more self-reliant. To achieve this end, Australia needed more people-hence the adoption of a vigorous postwar immigration program. It was believed that there might be two important constraints on a policy of seeking more immigrants. First, in ethnic, religious, and racial terms, Australia was a very homogeneous society, so there was some doubt as to the ease with which strangers could be absorbed. As a result, emphasis was placed on the ability of immigrants to become assimilated. Given the origins of the Australian population, the obvious preference was for immigrants from the United Kingdom and northwestern Europe, but, in response to changing conditions in European labor markets, the source of immigrants gradually moved south and east. Italians and Greeks arrived in large numbers during the 1950s and early 1960s, to be followed subsequently by immigrants from the Middle and Near East. During the 1980s, Asian immigration became important (table 15.1). The second possible constraint was jobs. Where were immigrants to work? It was not expected that they would be farmers or service-sector workers. Nor was it expected that they would create their own jobs. It was natural, at the time, that immigration and manufacturing development should be seen as interrelated: immigrants needed a growing manufacturing sector as a source of employment, and Australia needed both immigrants and a larger manufacturing sector to be more self-reliant. The desire to develop manufacturing was also encouraged by the international trading environment of the time. Both the depression of the 1930s and the boom and bust of primary product markets Table 15.1
Australian Population by Birthplace (%)
Country of Birth
1933
1947
1961
1981
1984
Australian born United Kingdom and Ireland
86.4 10.8
90.2 7.1
83.1 7.2
78.2 7.8
78.9 7.7
97.2
97.3
90.3
86.0
86.6
.4
.4
2.2 1.o .7
1.8 .8 1.o .7 .5 3.1 5.5 100.0
Subtotal Italy Germany Greece The Netherlands Poland Asia Others Total
.4 2.0
.3 2.0
.6 .8 3.4
1.9 .8 1.o .7 .4 2.2 7.0
100.0
100.0
100.0
100.0
I .o
Source: 1933, 1947, 1961, 1981 Australian Census. “1984, Resident Population by Birthplace,” Australian Bureau of Statistics, Australian Demographic Statistics, Catalogue no. 3 101.O (Canberra: June 1985).
388
R. G. GregoryIR. AnstieiE. Mug
in response to the Korean War had illustrated the dangers of an economy organized around one or two primary product exports. To facilitate development of manufacturing, the government pursued an active policy of industry protection by subjecting manufactured imports to tariffs and quotas. In terms of the original objectives, these policies were largely successful. During the first twenty years, the immigration flow added .78 percent per annum to the population, and, by 1981, 21.8 percent of the work force had been born overseas, 36 percent being of British origin. Australia had become a multiracial society, and assimilation had proceeded smoothly. Manufacturing also developed quickly, with immigrants providing additional labor and protection providing manufacturing jobs. By the late 1960s, however, the Australian government had become more aware of the high cost of tariffs and began to reduce trade barriers. At the margin of policy adjustments, manufacturing jobs were not encouraged. Since 1968-69, the average effective tariff rate has been reduced from 36 to 19 percent (table 15.2), and the import share of domestic market supplies has increased. Manufacturing declined as protection was reduced, and the employment share of full- and part-time workers fell from around 29 to 16 percent. Despite this decline, the proportion of immigrants that work in manufacturing remained virtually unchanged. Among full-time employees at the 1981 Census, 38.8 percent of foreign-born males and 33.1 percent of foreign-born females were employed in manufacturing; for particular ethnic groups, the concentration was particularly high: 49 percent for males born in southern Europe and 60.6 percent for females (the shares of the Australian work force employed in manufacturing were 24.1 percent for males and 13.5 percent for females; table 15.3). The concentration of immigrants in TCF was even greater. Among full-time female employees born in southern Europe, 27.1 percent worked in TCF; for those born elsewhere in Europe, the proportion was 14 percent (the proportion of native born was 3.1 percent).3 Although tariffs for most Australian industries have been reduced substantially since the early 1970s, immigrant-intensive industries such as motor vehicles and TCF have received increased protection from quotas; as a result, there remains a strong positive association between the proportion of the work force born overseas and industry protection. Those employed in immigrantintensive industries have received increased levels of community assistance in terms of either job maintenance or higher wages. In the United States, the distribution of native and foreign-born male employment is similar; 30.9 percent of native-born and 31.9 percent of foreignborn full-time employees work in manufacturing. The industrial distribution of female immigrants, however, is similar to that in Australia; 26.9 percent of foreign-born, 18.3 percent of native-born, and 42.1 percent of southern European-born females work full time in manufacturing. Furthermore, more than 20 percent of women born in southern Europe and employed full time work in TCF. For native-born females, the proportion is similar to that in Australia, between 2 and 3 percent.
389
Low-skilled Immigrants in the U.S. and Australia
Table 15.2
Average Effective Rates of Assistance Manufacturing Subdivisions from 1968-69 to 1987-88 (%)
ASIC Subdivision 21 Food, beverages, & tobacco 23 Textiles 24 Clothing and footwear 25 Wood, wood products, & furniture 26 Paper & paper products, printing & publishing 27 Chemical, petroleum, & coal products 28 Nonmetallic mineral products 29 Basic metal products 31 Fabricated metal products 32 Transport 33 Other machinery & equipment 34 Miscellaneous manufacturing Total manufacturing
1968-69
1974-75
1975-76
1977-78
1982-83
1987-88
16
21
20
10
9
5
43 97
39 87
50 99
47 141
54 220
68 183
26
18
19
18
13
18
52
31
30
24
24
16
31
23
26
19
14
12
15
11
10
5
4
4
31 61
16 39
16 38
10 30
11 27
9 23
50 43
45 24
59 25
48 20
72 18
44 23
34
27
26
30
25
28
36
27
28
24
25
19
Source: Assistance to Manufacturing Industries in Australia, 1968-69 to 1973-74 Industries Assistance Commission (Canberra: Australian Government Publishing Service, 1976). Assistance to Manufacturing Industry, 1977-78 to 1982-83 Industries Assistance Commission (Canberra: Australian Government Publishing Service, 1985). Industries Assistance Commission,Annual Report, 1980-81 (Canberra: Australian Government Publishing Service, 1981). Industries Assistance Commission, Annual Report, 1987-88 (Canberra: Australian Government Publishing Service, 1988). Note: The estimates from 1968-69 to 1987-88 are in four series: from 1968-69 to 1972-73; from 1974-75 to 1975-76; from 1977-78 to 1982-83; and from 1983-84 to 1987-88. The first series is based on 1971-72 production weights; the second series uses 1974-75 production weights; the third series employs 1977-78 production weights and also incorporates forms of assistance not included in previous series estimates; and the fourth series employs 1983-84 production weights.
15.2 Immigrant Earnings in TCF Industries 15.2.1 Background Table 15.4 presents average weekly earnings of full-time workers in TCF expressed as a ratio of male average weekly earnings in all industries. The data are taken from the Census of each country: 1981 for Australia and 1980 for the United state^.^ Adult male workers in U.S. TCF earn 21.3 percentage points less than the average of all male workers. In Australia, the shortfall is 9.4 percentage points. For females, the earnings gap is greater. The average
Table 15.3
Distribution of Ethnic Groups over Industry Categories (%) Australia I98 1 Footwear
Birthplace Males, full-time employees, 15-64 years: Native born United Kingdodreland Southern Europe Rest of Europe Asia Central and South America Other Total foreign born Total Females, full-time employees. 15-64 years: Native born United KingdodIreland Southern Europe Rest of Europe Asia Central and South America Other Total foreign born Total
& Clothing
.3 .6 1.4 1.1 1.5
Textiles
.5
.8
1.7 1.1 2.2
United States 1980
All Manufacturing
All Other Industries
24.1 34.0 49.3 40.3 43.0
75.9 66.0 50.7 59.7 57.0
Footwear Total
100 100 100 100
& Clothing
.6
.o 2.0
.o
All Manufacturing
All Other Industries
Total
1.6 .9
30.9 32.5 35.6 33.7 25.8 32.5 37.3 31.9
69.1 67.5 64.4 66.3 74.2 61.5 62.7 68.1
100 100 100 100 100 100 100 100
Textiles
.8 .8 2.3 .3
.o
.5 .9
1.4 1.3
30.8 38.8
69.2 61.2
100 100
1.1 2.6 2.8 1.7
.5
.7
28.1
71.9
100
.7
.8
31.1
68.9
100
2.4 3.4 22.4 10.7 10.6
.7 1.0 4.7 3.3 1.9
13.5 19.7 60.6 34.5 38.6
86.5 80.3 39.2 65.5 62.4
100
2.0
.9
100 100 100 100
1 .o
.o 1.8
1.8 2.4
17.7 33.1
82.3 66.9
100 100
.2 .7 I .4 .6
81.7 86.3 57.9 81.4 73.6 68.6 72.8 73.1
100 100 100 100
2.7 9.5
18.7 1.9 8.6 12.8 8.0 8.7
18.3 13.7 42.1 18.6 26.4 31.4 21.2 26.9
.5
1.1
18.4
81.6
100
3.6
.9
20.3
79.7
100
Source: Australia, 1981 Census. United States, 1980 Census.
100
1 .o
.o
100
100 100 100
Table 15.4
Earnings of Textiles, Clothing, and Footwear, Full-Time Workers, Average Weekly Earnings as a Proportion of Male Average Weekly Earnings in All Industries United States 1980
Australia 1981
(1)
Native Born (2)
Foreign Born (3)
Central or South American Born (4)
82.2 74.4 78.2
79.6 77.9 78.7
83.0 79.8 81.2
57.4 46.1 53.7
52.3 47.9 50.8
100.6
84.3
100.0
100.7
87.4
65.1
58. I 64.6 59.4
56.5 62.0 57.6
53.7 58.3 54.4
40.1 48.4 42.5
40.7 49.0 43.5
40.3 38.5 40.2
38.4 35.1 38.3
75.3
76.2
72.6
61.2
61.2
56.2
47.0
64.4 78.8 69.3
64.9 83.2 71.8
64.0 74.4 67.2
57.7 65.6 59.5
44.0 44.0 44.0
41.2 44.2 41.6
92.4
92.5
92.4
77.4
Native Born (2)
Foreign Born (3)
Southern European Born (4)
Total
92.0 89.7 90.6
87.4 94.0 91.5
96.2 84.6 89.6
100.0
99.8
57.2 63.2 58.4
Total (1)
Males: Footwear & clothing Textiles Combined TCF Industries Females: Footwear & clothing Textiles Combined TCF Industries Total: Footwear & clothing Textiles Combined TCF Industries
58.2 70.1 63.5 92.0
Sources: Australia, 1981 Australian Census. United States, Household Sample File, 1980 Census. Nore: All full-time workers fifteen to sixty-four years. Full-time workers, 36 hours or more a week
61.9 63.8 72.4 68.0 93.2
76.0
58.9
392
R. G. GregoryIR. AnstieiE. Mug
adult female TCF full-time worker in the United States earns 42.5 percent of male average weekly earnings. In Australia, the ratio is 58.4 percent. Men who work in TCF in Australia are 15 percent better off than their U.S. counterparts, and women are 37 percent better off. It is also apparent from table 15.4 that foreign-born TCF workers earn less than native In Australia, the gap between native and foreign-born males is slight, 1.9 percentage points. In the United States, it is larger, 27.5 percentage points. Foreign-born females earn 1.8 percentage points less in Australia and 3.3 percentage points less in the United States. There are three important questions that emerge from table 15.4. Why do TCF workers earn less than other workers? Why do immigrants in TCF earn less than the native born? Why are these earnings gaps larger in the United States? As we search for answers, we will be moving toward judgments as to the role of different institutions in each country. 15.2.2 A Human Capital Model for All Workers Over the last decade or so, the dominant paradigm that economists have used to explain the distribution of individual earnings has been the human capital model. We also adopt this framework to address our three questions. We begin by fitting the usual human capital equations to full-time workers in the economy as a whole. For simplicity, we add together the male and female earnings equations and form one equation, which can be written as
E,
BjX,
= j= 1
+
B,"X:
+ U,,
11
where Eiis the log of earnings of the ith person, and X , are human capital and experience variables. The superscript, E refers to whether the individual is female. U, is an error term. The results are listed in table 15.5. The regression equations are as in equation (l), with the addition of a constant term. We use the natural log of weekly earnings as the dependent variable because Australian data do not provide good estimates of hourly earnings. In each country, a full-time worker is employed 35 hours or more per week. The coefficients of equation (1) are interpreted as percentage changes in earnings in response to a one-unit increase in the value of an independent variable. The constant term measures the average log of weekly earnings of a male high school graduate, of urban residence, never married, working full-time, and during his first year in the labor market. The first set of coefficients estimates the additional payoff for men over and above the constant term. Thus, an estimate of the average earnings of a male university graduate, with all other attributes included in the constant term, is given by the addition of the constant term and the estimated coefficient attached to the graduate dummy
Low-skilled Immigrants in the U.S.and Australia
393
lsble 15.5
Earnings Equations for Australia and the United States Fifteen- to Sixty-four-Year-Old Full-Time Workers
R2 Dependent variable Constant Education:' Dropout High school Postsecondary qualificationsb University degree Postgraduate degree Female X dropout Female x high school Female x postsecondary qualificationsb Female X university Female X postgraduate Experience? Experience Experience2 Female x experience Female x experience2 Area: Rural Female X rural Marital status: Spouse present Other marital status Female X spouse present Female x othermarital status Children under 18 Female x childrenunder 18
- ,1468
... .0527 ,2148 ,4672 ,6127 - .0834 - .0979
- .I945 - ,1612 - .0939 - ,0727 .0368 - ,0007 ,0034 - .0002
Australia
United States
0.44 In W 5.0456
.25 In W 5.1635
( - 17.80)
...
- ,1350 ...
( - 9.22)
...
(5.96) (17.04) (39.56) (28.55) (-5.37) (-6.45)
(.1523)
(10.1)
,2639 ,3890 - ,0844 - ,1412
(12.48) (22.00) ( - 3.27) (-3.88)
(-5.48)
- .I093
( - 4.47)
- ,0371 - .0457
( - .94) (-1.38)
( - 8.33)
(-4.74) (-1.83) (35.5I)
,0419
(37.03)
( - 30.83)
- ,0007
(1.87)
- ,0069 ,0003
( - 29.57) ( - 2.85)
( - 3.66)
(.61)
-.I187 .0204
( - 11.94)
- .0648 - ,0038
( - 6.57)
(1.03)
.I789 .1174 - .I279
(17.46) (7.88) (-7.72)
,1720 ,0730 - ,2622
( 12.97) ( - 10.43)
- .0183 - ,0052
( - .79) ( - .63)
- .I296 - ,0182
(-5.17) ( - 2.23)
- .I361
(-7.59)
- ,1694
(-11.20)
(-.12)
(4.30)
Note: t-statistics are in parentheses. 'Education is defined in the following ways. Dropout: Australia, left school before age fifteen, no postsecondary degree; United States, less than four years of high school completed. High school: Australia, left school after age sixteen, no postsecondary degree; United States, completed four years of high school. Postsecondmy qunl$carions: Australia, trade certificate or other postsecondary degree; United States, completed one to three years of college. Posrgruduare, Australia, higher degree level; United States, completed five or more years of college. bFor Australia, this group has been divided into two parts, the first coefficient related to those who have completed trade qualifications and the second to those with other postsecondary qualifications. 'Australia: age minus number of years of schooling minus 5 years. United States: age minus number of years of schooling minus 6 years.
394
R. G. GregoryIR. AnstieIE. Klug
variable. The estimated earnings of a female university graduate, with all other attributes of the constant term, is given by the addition of the constant term to the sum of male and female university graduate coefficients. The equations produce the expected outcomes; average weekly earnings are positively associated with more schooling and more experience, and human capital coefficients are lower for women. Being married and having children under 18 years of age also depresses women’s wages. Since the data are taken from Census tapes, we are unable to measure work force experience accurately. Like many others, we use potential experience as a proxy variable for actual experience. For Australia, potential experience is measured as age minus years of schooling minus five and, for the United States, as age minus years of schooling minus six. Potential experience is not a good proxy for work force experience of women, who are less likely than men to have been continuously in the labor force. However, this inadequacy should not be a major source of difficulty. Most of the analysis involves comparisons across different sets of women where the relative bias should be much less than that which arises from comparisons between men and women. The equations seem to work well within each country and on the surface are remarkably similar, despite different institutions. It appears that the human capital model does reasonably well, at this level of aggregation, and can explain some of the earnings distribution of full-time workers in the economy as a whole. With the aid of these equations, we now return to address our three questions. 15.2.3 Allocating the Earnings Gaps Table 15.6 provides data to answer our questions; the earnings gaps are allocated to differences in the general pay structure, human capital variables, industry-specific factors, and women’s pay relative to men. Row 1 of table 15.6 lists the earnings ratios of TCF workers to average male weekly earnings in each country and by categories of workers. The difference between these ratios and 100 are the earnings gaps to be explained (row 5). In every instance, the gap is positive, indicating that TCF workers in all categories are paid less than average male weekly earnings. The gaps range from 8.5 percentage points for native-born Australian males to 57.5 percentage points for U.S. females. Male immigrants earn less than native-born workers, particularly in the United States, where the earnings gap for foreign-born males is 46.3 percentage points. Row 2 lists the hypothetical earnings ratio when average weekly earnings are estimated from human capital equations for each country on the assumption that endowments of TCF workers are rewarded at the same rate as all workers. The differences between row 2 and row 1, which are listed in row 6, indicate that TCF workers earn less for their endowments than they would in other industries. We refer to this as an industry effect. The industry effect for women is similar in both countries, substantially depressing earnings by 13.6
Table 15.6
Allocations of Earnings Gap for TCF Workers, 1980, 1981 U.S. Males
I . Earnings ratio 2. Hypothetical ratio, own country coefficients 3. Hypothetical ratio, Australian coefficients 4. Female average earnings economy-wide 5. Earnings gap to explain 6 . Industry effects, row 2 - row I 7 . General pay structure, row 3 - row 2 8. Human capital contribution, 100 - row 3
Australian Males
Total
Native Born
Foreign Born
Total
Native Born
Foreign Born
U.S. Females
Australian Females
78.7
81.2
53.7
90.6
91.5
89.6
42.5
58.5
91.7
92.0
89.9
97.3
97.4
103.9
57.6
72.0
94.9
94.9
94.6
97.3
97.4
103.9
72.2
72.0
21.3
18.8
46.3
9.4
8.5
10.4
64.4 57.5
76.5 41.6
13.0
10.8
36.2
6.7
5.9
14.3
15.1
13.6
3.2
2.9
4.7
NA
NA
NA
35.6
23.5
5.1
5.1
5.4
2.7
2.6
-3.9
6.8
4.5
Source: Table 15.4 and calculations from table 15.5
396
R. G. Gregory/R. Anstie/E. Klug
and 15.1 percentage points in Australia and the United States, respectively. For males, the industry effect varies across countries and categories of workers, but in every instance it too depresses earnings. The smallest effect occurs in Australia, where industry depresses earnings of male native-born workers by as little as 5.9 percentage points. The largest industry effect occurs in the United States, where earnings of foreign-born male workers are depressed by 36.2 percentage points. At the aggregate level, the industry effect in the United States is similar for men and women, depressing earnings between 13.0 and 15.1 percentage points, respectively, but in Australia it is greater for women (13.6 percentage points) and less for men (6.7 percentage points). The industry effect for U.S. male workers is twice that in Australia. Row 3 lists the hypothetical earnings ratio derived by placing the average endowments of U.S. TCF workers, and all U.S. workers in aggregate, in the Australian earnings equation of table 15.5. In this way, we estimate earnings ratios for U.S. workers under the assumption that they are paid according to the Australian general pay structure in the same way as all Australian workers. A comparison of row 3 and row 2 will indicate the similarity of general pay structures. For example, if the two hypothetical pay ratios for U.S. TCF workers-row 2 (calculated from the U.S. equation) and row 3 (calculated from the Australian equation)-are approximately the same, then the reward structure for the average bundle of TCF worker attributes in the economy as a whole is not that different in each country. Alternatively, the general U.S. pay structure may reward the typical bundle of attributes possessed by U.S. TCF workers less than the Australian pay structure and thus contribute to an explanation of the earnings gaps across countries. The gap between row 3 and row 2 is listed as row 7 and indicates that the U.S. pay structure does provide lower rewards for human capital attributes possessed by TCF workers. For males, the U.S. general pay structure depresses earnings by between 2.9 and 4.7 percentage points relative to Australia. For U.S. women, the gap is larger-35.6 relative to 23.5 percentage points in Australia. Finally, the gap between 100 and row 3 is listed as row 8 and provides a measure of the degree to which fewer human capital endowments can explain low pay among TCF workers (using Australian human capital rewards as weights). For example, the calculations show that U.S. males are less well educated than their Australian counterparts, which depresses earnings by 5.1 rather than 2.7 percentage points, but the education mix of immigrants and U.S. native-born workers is similar, U.S. women are also less well endowed than their Australian counterparts. We can now utilize the data of table 15.6 and provide answers to our three questions. 15.2.4 Why Do TCF Workers Earn Less than Other Workers? We begin with male workers. First, in both countries, male workers possess less human capital than average, which accounts for 5.1 of the 21.3 percent-
397
Low-skilled Immigrants in the U.S. and Australia
age point earnings gap in the United States and 2.7 of the 9.4 percentage point gap in Australia (row 5 and row 8). Second, the general pay structure rewards the average human capital endowments of TCF workers less in the United States than in Australia, and this accounts for 3.2 percentage points of the U.S. gap (row 7). Third, the largest effect is the industry effect, which accounts for about two-thirds of the low earnings of TCF male workers in both countries (row 6). The determinants of industry and general pay structure effects are not known, but these effects are substantial. It is interesting to note that the industry effect seems to be similar across matched industries in Australia and the United States (Krueger and Summers 1988; Gregory and Daly 1990). To conclude, therefore, the human capital model, with the usual list of variables, is not a useful tool for providing a detailed answer to our question. It can explain one-third of the aggregate male earnings gaps at most (row 8). For women, human capital variables as measured are even less important. They account for about 10 percent of the earnings gap in both countries (4.5 and 6.8 of 41.6 and 57.5 percentage points, respectively, for the United States and Australia). Once again, the industry effect is large and depresses earnings between 15.1 and 13.6 percentage points. Finally, the largest effect arises from the general pay structure, which rewards women less than men. This effect is particularly important in the United States, where it accounts for 35.6 percentage points of the gap between the earnings of TCF female workers and average male earnings in the economy.
15.2.5 Why Do Immigrants in TCF Earn Less than the Native Born? It is unlikely that human capital equations are sufficiently precise to explain the small differences in earnings between native and foreign-bornwomen with an acceptable level of confidence. Consequently, attention is directed to male earnings gaps. For male workers, the human capital model as specified cannot provide an answer to this question. In the United States, immigrants seem to possess as much human capital as the native born, yet they earn much less, and, in Australia, they possess considerably more human capital but earn marginally less. Most of the lower pay of male immigrants is attributed to industry effects, which in the United States are quite substantial. To explain low immigrant earnings, therefore, either the list of human capital variables needs to be extended-to capture attributes that are particularly concentrated among immigrants, perhaps lack of spoken English-or a different theory of earnings determination is needed, a theory that focuses on the way in which labor market institutions affect earnings in these industries. We offer further observations along these lines later. 15.2.6 Why Are Earnings Gaps Larger in the United States? Table 15.7 presents data drawn from table 15.6 to isolate where the differences lie between the two countries. In the United States, male TCF workers are paid 11.9 percentage points less than their Australian counterparts, and
398
R. G. Gregory/R. Anstie/E. Mug
Table 15.7
Allocation of the Larger Earnings Gap in the United States (percentage points) ~
~
Males
1. Larger U.S. gap to be explained 2. Industry effects
3. General pay structure 4. Human capital
Foreign Born
Total Females
10.3
35.9
16.0
4.9
21.9
I .5
2.5 2.9
4.7
12.2 2.3
Total
Native Born
11.9 6.3 2.4 3.2
9.3
Source; Table 15.6.
female TCF workers are paid 16.0 percentage points less. For men, more than half the gap occurs because the industry effect is stronger in the United States (row 2). The other half of the gap is allocated between lower human capital endowments in the Untied States and the U.S. general pay structure, which pays average attributes of TCF workers less than they are paid in Australia. The human capital model, therefore, accounts for approximately 20 percent of the male earnings gap across the countries (row 1 and row 4). For women, about 80 percent of the earnings gap between Australia and the United States is accounted for by the general pay structure. Australia pays all women relatively more. Differences in industry effects and human capital endowments are less than for males. For foreign-born male workers, the gap between the two countries is considerable, 35.9 percentage points, of which 21.9, or two-thirds, arise from industry effects. There is a large difference in human capital, 9.3 percentage points, and a relatively small effect for differences in general pay structures, 4.7 percentage points. Industry effects are the major contribution to the eamings gap differences across countries. To summarize; the important questions that arise, in order of quantitative significance, are why industry effects are so large for foreign-born men in the United States (a difference of 21.9 percentage points between countries) and why the general pay structure rewards Australian women better than their U.S. counterparts (a difference of 12.2 percentage points). The answers to both these questions can be found in the way in which Australian tribunals have determined male and female pay. We begin by analyzing the reasons for the high pay ratio of women. Once this is done, much of the explanation for the pattern of the earnings gap for males and immigrants will fall into place. The difference in the male-female earnings gap between the United States and Australia can be attributed to the different labor market institutions in each country. Australian labor market institutions have been very effective at implementing changes in women’s pay. Before 1975, the Australian system of wage tribunals had always set wages for males and females using different criteria. Between 1950 and 1969, most female wages were set at levels that were ap-
399
Low-skilled Immigrants in the U.S. and Australia
proximately 75 percent of that which would be paid to a male doing a comparable job. This practice of explicit discrimination against women produced an earnings gap between men and women, for the labor market as a whole, that is similar to that prevailing in the United States today (Gregory and Ho 1985). Then, in 1969, the federal tribunal introduced “equal pay for equal work’ over a three-year period. From 1972, gender was not to be used as a wage criteria in those jobs that were not predominantly male or female. Before this decision, unskilled female workers in textile factories were paid less than unskilled male workers who may have been doing the same job. The “equal pay for equal work” decision might be thought of as the Australian equivalent of the Equal Pay Act in the United States. Then, in 1972, the federal tribunal decided that the concept of “equal pay for equal work” should be widened to “equal pay for work of equal value.” This concept might be thought of as the approximate equivalent to the “comparable worth” concept developed in the United States. This wider concept was introduced into the award wage structure in three uniform steps over the period to June 1975. After 1975, the tribunals would make wage judgments on the principle that award rates for all work should be considered without regard to the sex of the employee. The result of these two equal pay decisions was to increase female pay by 30 percent relative to male pay. The TCF workers shared in this pay increase, and this is the prime reason why female Australian TCF workers, compared to those in the United States, are 37 percent better off relative to average male earnings. As indicated earlier, the industry and human capital effect for women is approximately the same in both countries. For women, it is the treatment of all women by the general pay structure that matters. Figure 15.1 documents the changing female-to-male pay ratio for TCF workers in Australia relative to the wages of all workers. The large increase in earnings is apparent between 1969 and 1975, and it is interesting that there is no evidence of TCF workers slipping behind in achieving equal pay increases. The Australian institutional structure delivered the pay increase to all female workers regardless of the industry in which they worked, and, if there are large market adjustments to follow, they must be effected through employment falls or policy adjustments elsewhere rather than through wage adjustments. As indicated earlier, the combination of wage tribunals and extensive trade union membership effectively prevented the development of a secondary labor market and ensured that all workers were treated equally with regard to access to award wages and conditions. How do these tribunal decisions explain the higher pay of foreign-born men in Australia? The tribunals always attempted to set the pay of men without discriminating between native and foreign born, without regard to the industry in which they work, and without regard to whether the occupation is closely associated with female labor. Male pay in occupations where the predominant group of workers is female was always set on the same criteria as other male
400
R. G. Gregory/R. AnstieE. Klug 0.90
1966
1
1968
1970
1972
1974
1976
1978
1980
1982
1984
Year
Fig. 15.1 The ratio of female to male average hourly earnings, Australia Source: Australia Bureau of Statistics, Distribution and Composition of Employees, Earnings and Hours, Australia, Catalogue no. 6306.0 (Canberra, various issues).
wages (the principle of comparative wage justice). As a result of this procedure, the pay of males in female occupations is much higher, relative to the economy average, than is usual in other countries. For example, if workers are grouped into male-intensive and female-intensive occupations, classified according to the sex of the predominant group of workers, then men who work in female-intensive occupations in Australia earn marginally less than other male workers, 96.2 percent, but in the United States earnings for these workers is much lower, 82.6 percent (Gregory, Daly, and Ho 1986). An important part of the story, therefore, is not so much the relative demand and supply of low-skilled labor in each country or the different endowments of human capital but the way in which the labor market in general generates men's and women's pay for workers in female occupations or industries. Australian labor market institutions, like those of most European countries, have directed their attention toward increasing women's pay, with a fair degree of success. As a result, immigrant male labor and especially female labor has gained enormously in terms of average earnings from full-time work.
15.3 Immigrants, Earnings, and 'lkade Policy in Australia Earlier, we discussed the historical link between industry protection and the labor market. A 30 percent increase in female pay in TCF, ceteris paribus, increased the cost structure of these industries, relative to others, by about 78 percent. What were the trade policy reactions to such an increase in costs? By the standards of the last decade or so, a 7-8 percent change in relative
401
Low-skilled Immigrants in the U.S. and Australia
costs against imports is not large. By far the most important influence on relative costs over this period, and certainly in the short term, has been exchange rate changes. Between December 1972 and September 1974, for example, the Australian exchange rate appreciated by 20 percent. This is about three times larger than the increase in the TCF cost level flowing from the equal pay decisions. In addition, the Australian government reduced all tariffs by 25 percent in July 1973, and this reduced TCF competitiveness by about the same magnitude as the pay increase. As a result of all these influences and a higher rate of inflation of wages and prices than our trading partners, there was a large deterioration in the fortunes of TCF, which began to lay off workers at a fast rate. Initially, the government was slow to react, but, when the general unemployment situation deteriorated, it moved to introduce import quotas more or less across the board for TCF. Import quotas, which were to be a temporary measure, are still in place today, a decade and a half later. As the competitive situation of TCF industries continued to deteriorate, the protection offered by import quotas increased. When import quotas were introduced, their protective effect for clothing and footwear was equivalent to an effective tariff rate of 99 percent. By 1987-88, the effective tariff rate had increased to 183 percent. Over the last five years, the Australian exchange rate has depreciated by approximately 30 percent; as a result, the tariff equivalenceof the quotas has fallen back toward earlier levels. After the large pay increases in the early 1970s, the earnings of full-time TCF workers have not increased further relative to community averages. The wage relativities between industries have remained more or less rigid. The main effect of the quotas therefore has been not to influence the pay of TCF workers, including immigrants, but to protect the number of jobs that are available. From the viewpoint of TCF workers, there has been a consistency in Australian policy settings. The wage tribunals have stressed equality of pay outcomes, with respect to occupation and industry comparisons and with respect to lifting pay of employed workers at the bottom of the pay distribution. Tariffs have complemented this policy by protecting those parts of manufacturing that are vulnerable to import competition from low-wage countries. In parts of manufacturing, and certainly in the short term, the tariff policy has increased the number of jobs for low-skilled immigrants. The long-run effects of these policies, however, are much more difficult to judge. We would need to know the industry structure that would have evolved in the absence of the accumulated and fairly consistent tariff and wage decisions that have been made over the last few decades. It is obvious, of course, that the combination of quotas and higher women’s pay must increase the relative price of products from these female-intensive industries. This is the principal way in which the community has chosen to pay for increased earnings and job protection of the low paid. Figure 15.2 illustrates the price movements of clothing and footwear prod-
402
R. G. Gregory/R. Anstie/E. Klug 1.2
,
*'*.
quotas introduced in Australia 0.8
*.
**.
-
*+. '0.
0.7
1
.
1
.
1
.
1
'
1
.
1
.
1
*. .
*.......*. 1
ucts relative to consumer prices generally in both countries over the last two decades. Since 1966-67, the relative price of clothing and footwear products has fallen fairly steadily in the United States and is now approximately 40 percent lower than two decades ago. In Australia, the relative price fall is approximately 6 percentage points. Furthermore, the large increase in the relative price of footwear and clothing after imposition of import quotas is clearly apparent. Import quotas allowed the industry to increase prices, reduce the rate of layoffs, and pay the new wage scales. A similar story is evident for the relative price of textiles in each country (fig. 15.3), where relative prices have fallen by 30 percent in the United States and 12 percent in Australia. It is important to realize that import quotas have not prevented the number of jobs from falling. TCF industries are still subject to the fortunes of the domestic market, and jobs are still affected by growth rates of technological change and output. To measure properly the efficiency-welfare trade-off involved in the Australian regulatory system would require a general equilibrium model with welldefined demand and supply elasticities for factor and product markets. This is a very large task. However, on the basis of a number of simplifying assumptions, we can approximate the extent of community subsidies to TCF workers to gain some idea of the importance of trade policy. In columns 1 and 2 of table 15.8, we measure the ratio of the subsidy equivalent of industry
403
0
$
-
Low-skilled Immigrants in the U.S. and Australia
60
..............
-
._ I
-
-
40-
20
........... I
.
I
.
I
.
I
.
I
.
Australia United States 1
.
Year
Fig. 15.3 Relative price movements; Wholesale Price Index of textiles divided by the economy-wide Wholesale Price Index Source: The Textile, Clothing and Footwear Industries, Industries Assistance Commission Report, vol. 2 (Canberra: Australian Government Publishing Service, May 1986).
Table 15.8
Industry Assistance Relative to the Wage Bill Ratio of Subsidy Equivalent of Industry Protection to the Wage Bill
Year
196&69 1969-70 1970-7 1 1971-72 1972-73 1973-74 197r175 1975-76 197677 1977-78 1978-79 197!%80 1980-8 1 1981-82
Ratio of the Wage Bill to Value Added (free trade prices)
(1) Footwear & Clothing
(2) Textiles
(3) Footwear & Clothing
(4) Textiles
.76 .77 .77 .74 .74 .62 .71 .76 .90 .92 .94 .92 .97 1
.56 .54 .54 .54 .55 .48 .43 .56 .58 .55 .58 .59 .61 .60
1.27 1.23 1.23 1.16 1.19 1.03 1.22 1.30 1.56 1.59 1.52 1.47 1.45 1.85
.I7 .78 .78 .83 .81 .73 .90 .89 .88 .85 .81 .87 .90 .90
.w
~
Sources: Australian Trade Classijied by Industry: 1968-69 to 1981-82. Working paper (Canberra: Industries Assistance Commission, March 1985).Assistance to Manufacturing Industries: 1977-78 to 1982-83, Information paper (Canberra: Industries Assistance Commission, 1985).
404
R. G . Gregory/R. Anstie/E. Klug
protection to the wage bill. The subsidy equivalent of tariffs and quotas is defined as the subsidy that would be necessary for local industry to produce the same domestic output in the absence of tariffs or quotas. For footwear and clothing in 1968-69, the subsidy equivalent was equal to two-thirds of the wage bill. By 1981-82, it exceeded the wage bill. A great deal is being paid to keep workers in these industries. Columns 3 and 4 list the ratio of the wage bill to value added at free trade prices. Value added at free trade prices can be thought of as the alternative to producing value added in Australia. These calculations show that, before the equal pay decisions and under the tariff regime before 1975, the wage bill typically exceeded free trade value added by about 20 percent. After 1975, the economic situation of the industry deteriorated considerably so that the wage bill typically exceeds the free trade value added by around 50 percent. 15.4
Concluding Remarks
The earnings of low-paid immigrant labor, relative to average weekly earnings of an adult male full-time worker, seem to be considerably higher in Australia. The average male immigrant in the United States earns 13.2 percent less than his Australian counterpart. The average female immigrant earns 22.6 percent less. For those who work in the TCF industries, the gaps are even larger, 40.0 percent for foreign-born males and 30.2 percent for foreign-born females. The human capital model, with its usual list of variables, can explain only a small fraction of these earnings gaps. The differences in earnings for male immigrants seem to arise primarily from an industry effect, and for women it is primarily a reflection of the general -pay distribution within each country. We explained earlier how tribunal wage criteria in Australia link these two influences together. Low-paid immigrant workers seem to do relatively well “down under.” To explore the determinants of pay distribution within each country thoroughly is a very large job. Nevertheless, on the basis of the evidence offered here, we have argued that the principal determinant of the relative pay of TCF workers is the different institutional structure of each labor market. We have also argued that women’s pay plays a very special role in determining earnings of low-paid immigrants and that the level of women’s pay is very sensitive to the degree of outside intervention in the labor market. Almost all our analysis has been applied to TCF, but it can be generalized to include all low paid workers, and we suspect that the key result for TCF, that the low paid do relatively poorly in the United States, will hold for all low-paying industries. Our results have important implications for policy discussion in Australia. Recently, there has been extensive questioning of the efficiency of Australian labor market institutions. Income distribution questions have not received much coverage. The results reported here suggest that income distribution
405
Low-skilled Immigrants in the U.S. and Australia
questions may be important and should not be put aside. With respect to relative earnings, women and low-paid workers have gained enormously from the Australian wage system, but some of these gains have been paid for by some efficiency loss from the higher relative prices of imports and lower relative prices of exports.
Notes 1 . This paper is one of a series that explores the effect of institutions on the labor market by comparing Australian labor market outcomes with those of the U.S. labor market. Other papers include an analysis of women’s wages (Gregory and Ho 1985) and the response of the labor markets to the depression of the 1930s (Gregory et al. 1987). 2. A fuller discussion of the Australian institutional framework can be found in Niland (1986). 3. A seminal paper by Hughes (1973) explores the proposition that the Australian system of wage tribunals has compressed the distribution of industry wages in Australia relative to the United States. 4. The Australian Census records weekly income rather than earnings. A crosscheck with earnings data from other sources suggest that this is not a serious problem. For 1978-79, the earnings of full-year, full-time male workers were 98 percent of total income. For women, the ratio was 94 percent (Australian Bureau of Statistics, Income Distribution, Australia 1978/79, Individuals Cat. no. 6502.0, tables 17, 18 [Canberra: August 19821). 5. It should be emphasized that foreign-born TCF workers are not typical of immigrants in Australia. On average, immigrants are better educated than Australians and, if employed full-time, earn weekly incomes that are similar to those earned by the Australian born. But the immigrant group is very diverse, being disproportionately represented among the high- and low-income earners and among the well and poorly qualified. The central concerns of this paper therefore are not with the representative immigrant but with those at one extreme of the earnings and human capital distributions.
References Gregory, R. G., and A. Daly. 1990. Can Economic Theory Explain Why Australian Women Are so Well Paid Relative to Their US Counterparts? Discussion Paper no. 226. Australian National University, Centre for Economic Policy Research, November. Gregory, R. G., A. Daly, and V. Ho. 1986. A Tale of Two Countries: Equal Pay for Women in Australia and Britain. Discussion Paper no. 147. Australian National University, Centre for Economic Policy Research, August. Gregory, R. G., and V. Ho. 1985. Equal Pay and Comparable Worth: What Can the U.S. Learn from the Australian Experience? Discussion Paper no. 123. Australian National University, Centre for Economic Policy Research, July.
406
R. G . GregoryIR. AnstielE. Klug
Gregory, R. G., V. Ho, L. McDermott, and J. Hagan. 1987. The Australian and U.S. Labour Market during the Thirties. In Interwar Unemployment in International Perspective, ed. B. J. Eichengreen and T. J. Hatton, NATO AS1 Series (series DBehavioural and Social Sciences-vol. 43). London: Kluwer Academic Publishers. Hughes, B. 1973. The Wages of the Strong and the Weak. Journal of Industrial Relations 15(March): 1-24. Krueger, A. B., and L. H. Summers. 1988. Efficiency Wages and the Inter-industry Wage Structure. Econornetrica 56, no. 2(March):259-93. Niland, J. R. 1986. Wage Fixation in Australia. Sydney: Allen & Unwin. Noms, K. 1980. Compulsory Arbitration and the Wage Structure in Australia. Journal of Industrial Relations 23, no. 3(September):249-63. Stolper, W. F., and P. A. Samuelson. 1941. Protection and Real Wages. Review of Economic Studies 9(November):58-73.
Appendix: The NBER Immigration, Trade, and Labor Markets Data Files John M. Abowd
The National Bureau of Economic Research Immigration, Trade, and Labor Markets Data Files were developed from public data sources to facilitate industry-based research on the effects of international trade and immigration on labor markets in the United States. Many of the papers in this volume make use of data derived from these files.’ The purpose of this paper is to document the sources and methods used to prepare the data files. There are three basic types of files. Trade data files contain information organized on an industry basis for U.S. manufacturing industries. Industry immigration files contain information on work force characteristics, including immigrants, organized on an industry basis for all U.S. industries. Area immigration files contain information on work force characteristics, including immigrants, organized on a state and standard metropolitan statistical area (SMSA) basis. The Trade Data File contains shipments, a shipments deflator, value added, employment, payroll, real capital stock, imports, exports, unionization, and immigrant ratios for 450 four-digit ( 1972 Standard Industrial Classification) industries. The files provide annual data covering the period from 1958 to 1986. The Industry Immigration File contains information on the education, demographics, and immigrant proportions in the labor force of 292 Census Industrial Classification (CIC) industries. The area immigrant files contain information on immigrant proportions in the fifty states (State Immigrant File) and SMSAs (SMSA Immigrant File). John M. Abowd is professor of labor economics and management, Cornell University, and a research associate of the National Bureau of Economic Research. The author acknowledges financial support from the Ford Foundation and the National Science Foundation (grant 88-13847). George Borjas, Richard Freeman, Wayne Gray, and Lawrence Katz contributed substantially to the development of these data. Daniel Kessler, Laura Leete, and Ana Revenga served as research assistants.
407
408
John M. Abowd
Overall Organization of the Data Table A. 1 contains a list of the variable names, short definitions, and basic sources for the variables in the Trade Data File. Table A.2 contains a comparable list for the variables in the Industry Immigration Data File. Table A.3 contains a list for the variables in the Area Immigration Data file. Variable names shown in tables A . l , A . 2 and A.3 are used throughout this paper to refer to the variables in the files.
Annual Survey of Manufactures Data The industry level production, factor use, and price index data were prepared by Wayne B. Gray as a part of the NBER’s Productivity Project. The industry definitions conform to the 1972 Standard Industrial Classification. These definitions are used throughout the Trade and Immigration Data Files. The initial version of the industry level production, factor use, and price index data, covering the period from 1958 to 1976, was developed as a joint Table A.1 Acronym SIC YEAR VSHIPPED SHIPDEFL VADDED ALL-WRK PROD-WRK PROD-HRS ALLJ’AY PROD-WAG ALL-UN PROD-UN RCAPSTCK IMPORTS EXPORTS IMMRAT
Names, Short Definitions, and Sources for Variables in the NBER ’lkade Data File Short Definition Four-digit industry identifier Yea? Value of product shipments Shipments deflator Value added by manufacture All employees, number Production workers, number Production workers, hours All employees, payroll Production workers, payroll Percentage of all employees unionized Percentage of production workers unionized Real capital stock in 1972 dollars Customs value of imports Free alongside ship value of exports Ratio of immigrants to labor force in SIC ratio
Units
Source 1972 SIC’
$millions 1972 = 1.00 $millions thousands thousands millions $millions $millions % %
$millions $millions $millions
ASM‘ BLSd ASM ASM ASM ASM ASM ASM CPS’ CPS SRI-Penn‘ TMSr TMS NBER I m m h
~
* Standard Industrial Classification, manufacturing only.
The year includes the century (e.g., 1971). Annual Survey of Manufactures Statisticsfor Industry Groups and Industries.
Bureau of Labor Statistics (unpublished). Current Population Survey, May public use data. University of Pennsylvania, Bureau of the Census, SRI, Incorporated, 1958-76. Bureau of Industrial Economics, Commerce Department, 1977-86. g Bureau of Labor Statistics Trade Monitoring System. NBER Immigration Data Files, see table A.2.
409
NBER Immigration, Trade, and Labor Market Data Files
Table A.2 Acronym SIC
Names, Short Definitions, and Sources for Variables in the NBER Industry Immigration Data Files Short Definition
Four-digit industry identifier Immigrantsitotal employment 1960 Immigrants/total employment 1970 Asian immigrantsitotal employment 1970 ARAT70 Black immigrantsltotal employment BRAT~O 1970 White immigrantsitotal employment I970 WRAT'IO Mexican immigrantsitotal employment 1970 MRAT70 Other hispanic immigrantsitotal 0 ~ ~ ~ 7 0 employment 1970 Recent (last 2 years) immigrantsitotal employment 1970 RRAT~O Immigrantsitotal employment 1980 IMRAT~~ Asian immigrants1total employment 1980 ARAT~O Black immigrantskotal employment 1980 BRAT~O White immigrantsitotal employment 1980 WRAT80 Mexican immigrantsitotal employment 1980 MRATSO Other Hispanic immigrantsitotal employment 1980 ORAT~O Recent (last 2 years) immigrantsitotal employment 1980 RRAT~O Census Industrial Code CIC Year of Census YEAR PR 1624 Labor force age 16-24 Labor force with at least 2 years college PR2COLL Females in labor force PRFEMALE Blacks in labor force PRBLACK Production worker unionization rate PROD-U N IMRAT60 IMRAT70
a
Units
Source
Proportion Proportion
1972 SIC' CP60 l/lWb CWO 1 / 1 w
Proportion
CWO 11100
Proportion
CP 70 11100
Proportion
CP70 lil00
Proportion
CP70 lil00
Proportion
CP70 11100
Proportion Proportion
CP70 Ii100 CP80 Ad
Proportion
CP80 A
Proportion
CP80 A
Proportion
CPSO A
Proportion
CP80 A
Proportion
CP80 A
Proportion
CPSO A 1970 or 1980
Proportion
CP80 A
Proportion Proportion Proportion Proportion
CP80 A CP80 A CP80 A See table A. 1
Standard Industrial Classification, manufacturing only. Census of Population 1960 1/100 Public Use Sample. Census of Population 1970 11100 Public Use County Group Sample. Census of Population 1980 extracts from the A Sample.
project by the University of Pennsylvania, the Bureau of the Census, and SRI, Incorporated (called SRI-Penn hereafter). The SRI-Penn data are documented in Andrews and Zabala (1984) and the references therein.2 SRI-Penn variables from the Annual Survey of Manufactures (ASM) Statistics for Industry Groups and Industries have not been changed. The SRI-Penn data were extended to
410
John M. Abowd
Table A.3 Acronym STATE SMSA IMRAT70 IMRAT80
Names, Short Definitions, and Sources for Variables in the NBER Area Immigration Data Files Short Definition Census state code Census metropolitan area codesb Immigrantsitotal employment 1970 Immigrants/total employment 1980
Units
Proportion Proportion
Source CP70/CP8@ CP70/CP80 CP70 1/1w CP80 Ad
Census of Population 1970 1/100 Public Use County Group Sample or Census of Population 1980 extracts from the A Sample. Codes for metropolitan areas vary between 1970 and 1980; see the Census of Population documentation cited in the references for appropriate codes. Census of Population 1970 1/100 Public Use County Group Sample. Census of Population 1980 extracts from the A Sample.
a
1986 using ASM Statistics for Industry Groups and Industries, table 2 (1986 and earlier years). The implementation of consistent bridges to recode all industry data to a 1972 SIC basis was a major problem in the creation of the four-digit industrybased data. The SRI-Penn data (1958-76) were based on 1967 SIC definitions from the defunct Industry Projiles data base (U.S. Department of Commerce 1971, 1978). The SRI-Penn project recoded the 1967 SIC industry data to a 1972 SIC basis using the many-to-many bridge based on the industry translation table published as a part of the Census of Manufactures (U.S. Department of Commerce 1976). This bridge assigned to the 1972 SIC the proportion of the 1967 SIC-based variable that was appropriate based on the proportion of 1972 value added represented by the 1967 SIC-based industry. The data from 1977 to 1986 were restated on a 1972 basis as follows. Fourdigit 1977 SIC-based data were recoded to a 1972 basis using the manyto-many bridge that accompanied the 1977 Census of Manufactures (U.S. Department of Commerce 1981). This bridge assigns to the 1972 SIC the proportion of the 1977 SIC that was appropriate based on the proportion of 1972 value added represented by the 1977 SIC-based industry. (Although there are only 448 1977-based four-digit SIC industries, the NBER files continue to use the 450 1972-based four-digit SIC industries.) Some missing data were imputed because the values for a particular fourdigit SIC were suppressed from the original published ASM tables in order to avoid the breach of establishment confidentiality. For the period from 1958 to 1976, the SRI-Penn data estimated the value of missing data by a combination of remainder assignment and interpolation between the Census of Manufactures years (see Andrews and Zabala 1984, 7). For the period from 1977 to 1986 in the NBER Trade Data File, the following procedure was used. The complete two-digit industry group data and the available three-digit industry group data were used to impute the missing three-digit data by calculating the remainder of the two-digit classification after subtracting all available three-digit data. The two-digit remainder was
411
NBER Immigration, Trade, and Labor Market Data Files
allocated to missing three-digit industry groups by imputation based on available three-digit data in surrounding years. The complete three-digit industry group data (including imputations) and the available four-digit industry data were used to impute missing four-digit data following the same remainder and allocation process as used for the missing three-digit data. The detailed definitions of ASM variables can be found in Appendix A- 1 of ASM Statistics for Industry Groups and Industries (1986 and earlier years). The ASM is a survey of manufacturing establishments with a sampling frame derived from the Census of Manufactures conducted every five years. The following variables were used directly from ASM Table 2. WHIPPED is the total annual value of industry shipments in millions of current dollars. VADDED is value added by manufacture in millions of current dollars. ALL-WRK is the annual average employment of full- or part-time persons in thousands. PROD-WRK is the annual average full- and part-time employment of production workers (through the line supervisor level) in thousands. PROD-HRS is the total annual hours of production workers in millions. ALLJAY is the total annual value of gross earnings paid to all employees in millions of current dollars (includes salaries, wages, commissions, bonuses, vacation pay, sick pay, and compensation in kind as used in calculating federal withholding taxes). PROD-WAG is the total annual value of gross earnings paid to production workers in millions of current dollars. The estimated real value of the capital stock (in 1972 dollars) is based on the SRI-Penn data for the period 1958-76. The methods used to develop this series are described in Andrews and Zabala (1984, 10-15). The variable RCAPSTCK for the period 1958-76 is an accumulation of the depreciated, constant dollar plant and equipment investment series for the industry. For the period from 1977 to 1986, the real capital stock was estimated using unpublished information from the Industry Capital Stocks data base, which provides deflators for new investment and separate measures of real plant and equipment capital stocks for three-digit ind~stries.~ These data were used to estimate implicit depreciation rates for plant and equipment investments in the three-digit SIC industries. The price deflators and depreciation rates for each three-digit industry were assigned to all four-digit industries within that SIC. The published ASM new capital expenditures were divided into plant and equipment using unpublished data from the Industry Capital Stocks data base. The four-digit SIC real plant and equipment stocks were estimated separately using the formula real plant stock = (1 - depreciation rate) (previous year plant stock) + (previous year new plant spending)/ (previous year plant deflator), and similarly for the real equipment stock. The variable RCAPSTCK for the period from 1977 to 1986 is the sum of the real plant and real equipment stocks. Because the Industry Capital Stocks data base ends in 1982, deflators for
412
John M. Abowd
new plant and equipment were derived from the Detailed Investment by Industry data base. These data are available only at the two-digit level. The fourdigit industry was assigned the new plant and equipment deflator from its twodigit industry group for 1983 forward. The shipments deflator for the period 1958-76 comes from the SRI-Penn data. For the period from 1977 to 1986, the deflator is based on unpublished data from the Bureau of Economic Analysis (BEA) in the Department of Commerce. The BEA four-digit SIC shipment deflators are based on the Bureau of Labor Statistics product price indices. The BLS product price indices are based on the seven-digit detailed product class codes, which are aggregated to a four-digit SIC basis. There have been many changes in the product definitions and available price indices throughout the years. There are apparently no comprehensive descriptions of these series within the BLS or the BEA. In recent years (varying depending on the industry), the BLS has created industry price indices (as opposed to product price indices) at the fourdigit SIC level. The BLS industry price indices were used whenever possible. The shipments deflator is the variable SHIPDEFL.~
International Bade Data Since most domestic U.S. data are maintained on an industrial classification basis, primarily the SIC, the NBER Trade Data File was designed to provide import and export data on an industrial basis. The difficulty with this approach is that the SIC system requires information about the product type and the manufacturing method while the basic classification system used for imports and exports identifies only the product type. As a consequence, the most detailed SIC-based classifications of imports and exports group together products that span several four-digit SIC categories. The noncomparability of SIC-based import, export, and production data, as published, required the development of alternative industry-based estimates of international trade values. The annual industry measures of import and export value that appear in the NBER Trade Data File were derived from the Bureau of Labor Statistics Trade Monitoring System (1979, 1983a, 1983b), the Bureau of the Census US. Commodity Exports and Imports as Related to Output ( 1965164 and 1972171) , the ASM data described above, and the ASM Value of Product Shipments (1982-85). For the period from 1972 to 1981, the Bureau of Labor Statistics maintained a collection of time series called the Trade Monitoring System (TMS) that provided annual information on U.S. imports, the ratio of imports to new supply, exports, and the ratio of exports to domestic shipments on a modified four-digit SIC-based industry classification. Schoepfle ( 1982) reports the development and uses of the TMS data. His appendix contains numerous details of the calculations. Bennett (1982, available on request from the Department of Labor) reports the details of the TMS data base construction.
413
NBER Immigration, Trade, and Labor Market Data Files
The TMS was based on a very careful attempt to construct domestic shipment data that were as comparable as possible to the most detailed industrial classification-based import and export data. The NBER Trade Data File follows the TMS methodology to extend the series backward to 1958 and forward to 1985. The TMS methods were approximated for the period from 1958 to 1972, using 1972 as a splice year. The TMS methods were used exactly for 1982-85. For imports, the TMS defined 317 mutually exclusive manufacturing groups in a classification system called the MSIC (for import SIC). The MSIC covers all four-digit 1972 SIC-based manufacturing industries that can be distinguished from the automatic application of the concordance relating U.S. Department of Commerce Tariff Schedules of the United States, Annotated (TSUSA), to Product Class code (five-digit 1977-based SIC), which forms the basis for the tables in the publication US.Commodity Exports and Imports as Related to Output and the Department of Commerce online data base of official international trade statistic^.^ The basic import data are collected continuously at the port of entry to the United States by the Bureau of the Census and are classified according to the TSUSA as a part of the Customs process. The concordance between TSUSA and Product Class codes is updated monthly as TSUSA codes are created or eliminated. Annual import data aggregated to a four-digit SIC basis are published regularly by the Census Bureau on the basis of the current TSUSA-to-Product Class concordance and summarized biennially in the publication US.Commodity Exports and Imports as Related to Output. In this publication, a modified four-digit SIC, similar to the MSIC, is used to distinguish the industries.6 For exports, the TMS defined 370 mutually exclusive four-digit SIC manufacturing industries in a classification system called the XSIC (for export SIC). The classification is based on all 1972-based SIC industries that can be distinguished from an automatic application of the U.S. Department of Commerce concordance between Schedule B commodity numbers and five-digit SIC-based Product Class codes. The basic export data are collected at the port of departure from the United States by the Bureau of the Census and are classified according to the Schedule B number as a part of the coding process for international trade statistics. The concordance between Schedule B number and Product Class codes is updated monthly as Schedule B numbers are created or eliminated. Annual export data aggregated to a four-digit SIC basis are published regularly by the Census Bureau on the basis of the current Schedule B-to-Product Class concordance and summarized biennially in the publication U S . Commodity Exports and Imports as Related to Output. In this publication, a modified four-digit SIC, similar to the XSIC, is used to distinguish the industries.’ The import data in the NBER Trade Data File were assembled for the 317 basic MSIC industries. For each such industry, imports are defined as the customs value of imports in millions of current dollars. The customs value generally excludes transportation and insurance costs from the foreign port. For
414
John M. Abowd
the period from 1958 to 1971, imports were reported in U.S. Commodity Exports and Imports as Related to Output using an old definition. For 1972 to 1981, imports were used directly from the TMS. For 1982 to 1985, imports were reported in the U.S. Department of Commerce online data base of official statistics from the table All Items in US.Imports for Consumption from World (1986b) using a consistent definition but a revised concordance. The year 1972 was used as a splice to make the 1958-71 values conform to the 1972 definitions and concordance.8 The year 1981 was used as a splice to make the 1982-85 values conform to the 1981 conc~rdance.~ The MSIC-based value of domestic shipments for the industry or industry group corresponding to the MSIC was taken from the TMS for the years 1972-81. For the years 1958-71, the MSIC-based value of domestic shipments was imputed using the following technique. The value of industry shipments for the 1972 SIC-based four-digit industries was obtained from the ASM data discussed above. Using the concordance between MSIC and SIC developed at the NBER, the ASM value of shipments data were merged with the pre-1972 MSIC-based import data. Then the SIC-based shipments data for the years 1958-71 were multiplied by the ratio of MSIC-based value of shipments in 1972 to SIC-based value of shipments in 1972. For the years 198285, the TMS definitions of MSIC-based value of domestic shipments, which are based on comparable five-digit Product Class codes, were constructed directly from the table 1 values in ASM Value of Product Shipments. Data from the five-digit value of product shipments table represent sampling estimates of the net sales value freight-on-board at the point of manufacture in millions of current dollars (excludes discounts, transportation costs, and excise taxes). For each MSIC-based industry, the import penetration ratio (IPR) was defined as the ratio of imports to the sum of imports and (MSIC-based) domestic shipments. Using the MSIC-to-SIC concordance, the import penetration ratio was merged to the data for the 450 1972 SIC-based industries for the years 1958-85. The value of imports for the SIC-based industry was set to zero if there was no concordant MSIC. Otherwise, IMPORTS in the Trade Data File was calculated as the product of ASM shipments (the variable VSHIPPED) and the ratio IPW(1 - IPR). In the Trade Data File, the basic import penetration ratio can be reconstructed exactly by forming the ratio IMPORTS/(VSHIPPED + IMPORTS). Using 1972 as the reference year, the quality of the MSIC-to-SIC concordance at the four-digit level can be assessed by considering the summary statistics in table A.4. Slightly more than 50 percent of the industry value of shipments can be matched exactly. An additional 48 percent of industry shipments can be match by combining SICS, using special MSICs, and other concordance arrangements. Overall, import statistics are available for 98.3 percent of all shipments. The export data in the NBER Trade Data File were assembled for the 370 basic XSIC industries. For each such industry, exports are defined as the free
415
NBER Immigration, Trade, and Labor Market Data Files
Table A.4
Quality of the Concordance between Import-basedMSICs and Conventional 1972-based SICs
Category
Percentage of 1972 Shipments
Percentage of 1972 Industries
50.9 30.7 15.0
55.8 28.7 10.7 1.7 3.1
Exact match Combined 2 or more SICs into MSIC Special code required Other basis for concordance Excluded from TMS
1.6
I .7
Source: NBER Trade Data File and author's calculations
alongside ship value of exports in millions of current dollars. The free alongside ship value generally includes transportation and insurance costs from the point of manufacture to the port of departure from the United States. For the period from 1958 to 1971, the export value was reported in U S . Commodity Exports and Imports as Related to Output. For 1972-8 1, the export value was taken directly from the TMS. For 1982-85, the export value was reported in the U.S. Department of Commerce online data base of official statistics from the table All Items in U S . Domestic Exports to World (1986a) using a consistent definition but a revised concordance. The year 1972 was used as a splice to make the 1958-71 values conform to the 1972 definitions and concordance.l0 The year 1981 was used as a splice to make the 1982-85 values conform to the 1981 concordance. I I The XSIC-based value of domestic shipments for the industry or industry group corresponding to the XSIC was taken from the TMS for the years 197281. For the years 1958-71, the XSIC-based value of domestic shipments was imputed using the following technique. The value of industry shipments for the 1972 SIC-based industries was obtained for the four-digit industry from the ASM data discussed above. Using the concordance between XSIC and SIC developed at the NBER, the ASM value of shipments data were merged with the pre- 1972 XSIC-based export data. Then the SIC-based shipments data for the years 1958-71 were multiplied by the ratio of XSIC-based value of shipments in 1972 to SIC-based value of shipments in 1972. For the years 1982-85, the TMS definitions of XSIC-based value of domestic shipments, which are based on comparable five-digit Product Class codes, were constructed directly from the table 1 values in ASM Value of Product Shipments. For each XSIC-based industry, the export supply ratio (XS) was defined as the ratio of exports to (XSIC-based) domestic shipments. Using the XSIC-toSIC concordance, the export supply ratio was merged to the data for the 450 1972 SIC-based industries for the years 1958-85. The value of exports for the SIC-based industry was set to zero if there was no concordant XSIC. Otherwise, EXPORTS in the Trade Data File was calculated as the product of ASM shipments (the variable WHIPPED) and the variable xs. In the Trade Data File,
416
John M. Abowd
Table A S
Quality of the Concordance between Export-based XSICs and Conventional 1972-basedSICs
Category Exact match Combined 2 or more SICs into XSIC Special code required Other basis for concordance Excluded from TMS
Percentage of 1972 Shipments
Percentage of 1972 Industries
61.9 33.8 1.7 .7 1.8
74.2 18.7 2.0 1.1 4.0
Source: NBER Trade Data File and author’s calculations.
the basic export supply ratio can be reconstructed exactly by forming the ratio EXPORTS/VSHIPPED.
Using 1972 as the reference year, the quality of the XSIC-to-SIC concordance at the four-digit level can be assessed by considering the summary statistics in table A.5. Almost 62 percent of the industry value of shipments can be matched exactly. An additional 36 percent of industry shipments can be matched by combining SICs, using special XSICs, and other concordance arrangements. Overall, export statistics are available for 98.2 percent of all shipments. l2
Current Population Survey Data The industry unionization data were derived using the methods of Freeman and Medoff (1979). The Freeman-Medoff estimates were updated into the 1980s but cannot be used for the period from 1958 to 1972 because the estimates rely on the unionization questions from the May Current Population Survey (CPS). Union membership percentages were calculated as
where Uj is percentage of workers in Census industry j who are unionized; A, = 1 if worker i is employed and in a union and is zero otherwise; and W, is the CPS sampling weight attached to worker i. Separate unionization rates were calculated for all workers and for production workers. Production workers were defined as employed individuals in the following occupations: craftsmen and kindred, operatives except transport, transport operatives, nonfarm laborers, private household, all other service, farm laborers and foremen. The values for 1974 were based on the 1973, 1974, and 1975 May surveys. The values in the Trade Data File for 1958-73 are identical to the 1974 values. The values for 1980 were based on the 1979, 1980, and 1981 May surveys.
417
NBER Immigration, Trade, and Labor Market Data Files
The values for 1984 were based on the 1984 May survey. All other years were linearly interpolated. The unionization data for detailed CIC from the May Current Population Survey were matched to the 1972 SIC-based industry data using a one-tomany concordance between 1970 CICs and 1972 SICS. The variable ALLUN contains the estimated overall unionization rate from the concordant CIC. The variable PROD-UN contains the estimated production unionization rate from the concordant CIC.
Census of Population and Housing Data Public use microdata samples from the 1960, 1970, and 1980 decennial Census of Population and Housing were used to derive estimates of the ratio of immigrant employees to total employees by industry for the Census years. In 1970 and 1980, these immigrant ratios are also available by racial and ethnic groups. The 1970 and 1980 Censuses were also used to create industry and area data on labor force characteristics. The 1960 immigrant ratio was created from the 1960 Census of Population and Housing using the 1/100 Public Use Sample (Bureau of the Census 1975). The numerator of the ratio is the count of all immigrants employed in the detailed 1960 CIC industry. The denominator is the count of all individuals employed in the detailed CIC. Using a one-to-many concordance between the 1960 CIC and the 1972 four-digit SIC (for the 450 manufacturing industries only), the variable IMMRAT60 was created on an SIC basis from the concordant CIC. The 1970 immigrant ratios were created from the 1970 Census of Population and Housing using the 1970 5% Data County Group Sample, which is a 11100 sample of the U.S. population (Bureau of the Census 1972). The numerator of the ratio is the count of all immigrants (or all immigrants in the particular racial/ethnic group) employed in the detailed 1970 CIC industry. The denominator is the count of all individuals employed in the detailed CIC industry. Using a one-to-many concordance between the 1970 CIC and the 1972 four-digit SIC (for the 450 manufacturing industries only), the immigrant ratios for 1970 were created on an SIC basis from the concordant CIC. The immigrant groups used were all immigrants (IMMRAT~O), Asian immigrants (ARAT~O), black immigrants (BRAT~O), white immigrants (WRAT~O), Mexican immigrants (MRAT~O), other Hispanic immigrants (ORAT~O), and all immigrants who arrived within the last two years (RRAT~O). The Asian, black, and white racial groups are mutually exclusive and include Hispanics as appropriate. The two Hispanic ratios are mutually exclusive. The 1980 immigrant ratios were created from the 1980 Census of Population and Housing using the A Sample, which is a 5/100 sample of the U.S. population (Bureau of the Census 1983a, 1983b). The numerator of the ratio is the count of all immigrants (or all immigrants in the particular racial/ethnic
418
John M. Abowd
group) employed in the detailed 1 9 8 0 CIC industry. The denominator is the count of all individuals (from a 1 / 1 , 0 0 0 random subsample of the A Sample) employed in the detailed CIC industry (multiplied by fifty to reflect the different sampling rates). Using a one-to-many concordance between the 1 9 8 0 CIC and the 1 9 7 2 four-digit SIC (for the 450 manufacturing industries only), the immigrant ratios for 1 9 8 0 were created on an SIC basis from the concordant CIC. The immigrant groups used were all immigrants ( I M M R A T ~ O ) ,Asian immigrants (ARATSO), black immigrants (BRATSO),white immigrants (WRATSO), Mexican immigrants (MRATSO),other Hispanic immigrants (ORATSO), and all immigrants who arrived within the last two years ( R R A T ~ O ) . The other variables in the Industry Immigration Data Files are labor force characteristics by industry in 1 9 8 0 . The variable ~ ~ 1 6 is2 the 4 proportion of persons age 1 6 - 2 4 employed in the detailed CIC industry. The variable PR2COLL is the proportion of employed persons with two or more years of college in the detailed CIC industry. The variable PRFEMALE is the proportion of females employed in the detailed CIC industry. The variable PRBLACK is the proportion of blacks employed in the detailed CIC industry. There are 292 detailed CIC industries represented. For manufacturing industries, only a oneto-many concordance between the 1 9 8 0 CIC and the 1 9 7 2 four-digit SIC was used to assign the labor force characteristics of the CIC to its concordant SIC. The Area Immigration Data Files contain the variables I M M R A T ~ O and IMMRATSO, as defined in the Industry Immigration Data Files, except that the numerator and denominator of the ratios were calculated for immigrants and all employed individuals in states and SMSAs. The states were defined using the FIPS State Code for both Census years. The SMSAs were defined by the SMSA code in the 1 9 8 0 Census and by the area and subarea codes in the 1 9 7 0 Census. See the technical documentation in the references for code lists.
Availability The NBER Immigration, Trade, and Labor Markets Data Files are available on high density floppy disk (StrataTMformat or ASCII format) or computer tape (SASTMformat) from the author (address requests to NBER, Labor Studies Program, 1 0 5 0 Massachusetts Avenue, Cambridge, MA 0 2 1 3 8 ) . The files are also available on Internet from the author (contact [email protected] on Internet or JMA@CORNELLA on Bitnet).
Notes 1. These data are used in the following papers in this volume: Abowd and Freeman, Abowd and Lemieux, Collins, Freeman and Katz, Kuhn and Wooton, and Leonard and McCulloch.
419
NBER Immigration, Trade, and Labor Market Data Files
2. The SRI-Penn data files and documentation are available on request from Stephen Andrews at the Center for Economic Studies in the Bureau of the Census. 3. Information regarding the Industry Capital Stock data base is available from the Bureau of Industrial Economics in the U.S. Department of Commerce. The data base contains information on a three-digit SIC basis. 4. Only a subset of the annual industry data developed for the NBER Productivity Project are available in the NBER Trade Data File. For additional information about the ASM and related industry data, contact Wayne B. Gray at the National Bureau of Economic Research. 5. The basic classification systems for imports and exports changed in 1988. The Department of Commerce online data base currently produces tables of imports and exports by industry that are created using a different set of concordances than described in this paper. The tables from the Commerce Department online data base used in the NBER Trade Data File were based on the old concordances, described herein. 6. The Trade Monitoring System (1983b) describes the concordance procedure as follows: “For the purpose of relating imports to output, individual TSUSA commodity numbers are assigned to the five-digit SIC-based Product Class from the numerical list of manufactured products, 1977 Census of Manufactures, which contains the same products as the TSUSA number. In cases where the TSUSA numbers include items which should be classified in two or more SIC-based output codes, an assignment is made to the SIC-based output code to which the principal content of the TSUSA appears to belong, where such an assignment will not significantly overcount the classification to which the TSUSA number belongs. In cases where it appears that distortions will result from an assignment of an entire TSUSA number to a single SIC-based output code, the principal SIC-based output product classes are combined to form an SIC-based import code and TSUSA numbers are assigned to the combination.” 7. The Trade Monitoring System (l983a) describes the concordance between Schedule B numbers and Product Class codes as follows: For the purpose of relating exports to output, individual Schedule B commodity numbers are assigned to the five-digit SIC-based Product Class from the numerical list of manufactured products, 1977 Census of Manufactures, which contains the same products as the Schedule B number. In cases where the Schedule B numbers include items which should be classified in two or more SIC-based output codes, an assignment is made to the SIC-base output code to which the principal content of the Schedule B number appears to belong, where such an assignment will not significantly overcount the classification to which the Schedule B number is assigned or undercount the other classification to which it partially belongs. In cases where it appears that distortions will result from an assignment of an entire Schedule B number to a single SIC-based output code, the principal SIC-based output product classes are combined to form an SIC-based export code and the schedule B numbers are assigned to the combination.
8. That is, the values from 1958 to 1971 were multiplied by the ratio of the 1972 TMS import value to the 1972 US.Commodity Exports and Imports as Related to Output value. 9. That is, the values from 1982 to 1985 were multiplied by the ratio of the 1981 TMS import value to the 1981 U.S. Department of Commerce official statistic. 10. That is, the values from 1958 to 1971 were multiplied by the ratio of the 1972 TMS export value to the 1972 U.S. Commodity Exports and Imports as Related to Output value. 11. That is, the values from 1982 to 1985 were multiplied by the ratio of the 1981 TMS export value to the 1981 U.S. Department of Commerce official statistic.
420
John M. Abowd
12. The international trade data were constructed by the author. For further information. contact the author at the National Bureau of Economic Research.
References Andrews, Stephen, and Craig Zabala. 1984. Documentation of the SRI-Penn Manufacturing Industry Dataset Developed by David L. Crawford, Gary Fromm, Lawrence R. Klein, and Frank C. Ripley. Technical Notes, Center for Economic Studies. Washington, D.C.: Bureau of the Census, January. Bennett, Norman. 1982. Trade Monitoring System, Technical Note, Import Penetration and Export Proportion Data Bases. Washington, D.C.: Bureau of Labor Statistics, Division of Foreign Labor Statistics and Trade, November. Freeman, Richard B., and James L. Medoff. 1979. New Estimates of Private Sector Unionism in the U.S. Industrial and Labor Relations Review 32, no. 2 (January):143-74. Gray, Wayne. 1986. Productivity vs. OSHA and EPA regulation. Ann Arbor, Mich.: UMI Research Press. . 1987. The Cost of Regulation: OSHA, EPA and the Productivity Slowdown. American Economic Review 77, no. 5 (December):998-1006. Schoepfle, Gregory. 1982. Imports and Domestic Employment: Identifying Affected Industries. Monthly Labor Review 105, no. 8 (August): 13-26. U.S. Department of Commerce. 1986a. All Items in US.Domestic Exports to World, 1981-85 [online data base of official statistics]. Washington, D.C.: U.S. Department of Commerce. . 1986b. All items in U S . Imports for Consumptionfrom World, 1981-85 [online data base of official statistics]. Washington, D.C.: U.S. Department of Commerce. U.S. Department of Commerce. Bureau of the Census. 1971. Industry Projles, 19581969. Washington, D.C.: U.S. Government Printing Office, October. . 1972. Public Use Samples of Basic Records from the 1970 Census: Description and Technical Documentation. ICPSR Study no. 0018. Washington, D.C.: Bureau of the Census. . 1975. Public Use Sample of Basic Recordsfrom the 1960 Census: Description and Technical Documentation. ICPSR Study no. 7756. Washington, D.C.: Bureau of the Census, April. . 1976. 1972 Census of Manufactures. Washington, D.C.: U . S . Government Printing Office. . 1978. Industry Profiles, Annual Survey of Manufactures. Washington, D.C.: U.S. Government Printing Office, June. . 1973, 1974, 1975, 1979, 1984. Current Population Survey May 1979 Technical Documentation. ICPSR Study nos. 7936, 7937, 7938, 7974, 8461. Washington, D.C.: Bureau of the Census. . 1981. 1977 Census of Manufactures. Washington, D.C.: U . S . Government Printing Office, August. . 1983a. Census of Population and Housing, 1980: Public-Use Microdata Sample ( ASample) [machine-readable data file]. ICPSR Study no. 8101. Washington, D.C.: Bureau of the Census. . 1983b. Census of Population and Housing, 1980: Public-Use Microdata Samples Technical Documentation. Washington, D.C.: Bureau of the Census, Data User Services Division.
421
NBER Immigration, Trade, and Labor Market Data Files
. Annual. Annual Survey of Manufactures Statistics for Industry Groups and Industries. Washington, D.C.: U.S. Government Printing Office. . Annual. Annual Survey of Manufactures Value of Product Shipments. Washington, D.C.: U.S. Government Printing Office. . Biennial. U S . CommodityExports and Imports as Related to Output. Washington, D.C.: U.S. Government Printing Office. U.S. Department of Commerce. Bureau of Economic Analysis. Monthly. Survey of Current Business. U.S. Department of Labor. Bureau of Labor Statistics. Office of Productivity and Technology. Division of Foreign Labor Statistics and Trade. 1979. Trade Monitoring System Import Penetration by Four-Digit SIC-based Manufacturing Commodiry Group [unpublished computer listing]. March. . 1983a. Trade Monitoring System US.Exports and Related Output by FourDigit SIC-based Commodity Group Manufactures Exports, Product Shipments, and Exports to Shipments Ratio, 1972-1981 [unpublished computer listing]. November. . 1983b. Trade Monitoring System U S . Imports and Related Output by FourDigit SIC-based Commodity Group Manufactures Imports, Product Shipments, New Supply, and Import Penetration, 1972-1 981 [unpublished computer listing]. November.
This Page Intentionally Left Blank
Contributors
John M. Abowd NYSSILR 264 Ives Hall Cornell University Ithaca, NY 14851-0952
David E. Bloom Department of Economics Columbia University 420 W. 118th Street, 10th floor New York, NY 10027
Joseph G. Altonji Department of Economics Research School of Social Sciences Northwestern University 2003 Sheridan Road Evanston, IL 60208
George J. Borjas Department of Economics, D-008 University of California, San Diego La Jolla, CA 92093
R. Anstie Centre of Policy Studies Monash University Clayton, VIC 3168 Australia
David Card Industrial Relations Section Firestone Library Princeton University Princeton, NJ 08544
Ann P. Bartel Graduate School of Business 710 Uris Hall Columbia University New York, NY 10027
Bruce J. Chapman Senior Fellow Centre for Economic Policy Research The Australian National University P.O. Box 4 Canberra, Australia 2601
John J. Beggs Associate Director Daiwa Securities Australia, Limited Level 48, 80 Collins Street Melbourne, VIC 3000 Australia
Susan M.Collins Department of Economics Littauer M-7 Harvard University Cambridge, MA 02138
423
45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63
64
65 66 67 68 69 70 71 72 73 74
424
Contributors
Juan Diez-Canedo R. Deputy Director General Investment Banking Banco Intemacional Reforma 156 Piso 13 Col. Juarez 06600 Mexico, D.F. Richard B. Freeman National Bureau of Economic Research 1050 Massachusetts Avenue Cambridge, MA 02138 R. G. Gregory Research School of Social Sciences Department of Economics The Australian National University P.O. Box 4 Canberra, Australia 2601 Morley Gunderson Director, Centre for Industrial Relations University of Toronto 123 St. George Street Toronto, Ontario M5S 1Al Canada Lawrence F. Katz Department of Economics Littauer Center 107 Harvard University Cambridge, MA 02138 E. Klug Research School of Social Sciences Department of Economics The Australian National University P.O. Box 4 Canberra. Australia 2601 Marianne J. Koch College of Business Administration University of Oregon Eugene, OR 97403 Peter Kuhn Department of Economics McMaster University Hamilton, Ontario L8S 4M4 Canada
Robert J. LaLonde Graduate School of Business University of Chicago 1101 East 58th Street Chicago, IL 60637 Kevin Lang Department of Economics Boston University 270 Bay State Road Boston, MA 02215 Thomas Lemieux Department of Economics, E52-262B Massachusetts Institute of Technology Cambridge, MA 02139 Jonathan S. Leonard Walter A. Haas Graduate School of Business 350 Barrows Hall University of California Berkeley, CA 94270 Rachel McCulloch Department of Economics Brandeis University Waltham, MA 02254 Marta Tienda Department of Sociology University of Chicago 5848 South University Chicago, IL 60637 Robert H.Topel Graduate School of Business University of Chicago 1101 East 58th Street Chicago, IL 60637 Franklin D. Wilson Center for Demography University of Wisconsin 1180 Observatory Drive Madison, WI 53706 Ian Wooton Department of Economics Social Science Centre, Room 4028 University of Western Ontario London, Ontario N6A 5C2 Canada
Author Index
Abowd, John M., 2571115, 362, 365nnl ,lo, 12.14, 366n17 Allen, I., 322 Altonji, Joseph, 184 Andrews, Stephen, 409-1 1
Bach, R. L., 136, 140-41, 147, 1621112 Bachu, Amara, 82,97n6 Baker, R. P., 121 Baldwin, Robert E., 298, 300 Barker, Betty L., 281n4 Bartel, Ann P., 127-28, 133118,219 Batra, R. N., 285 Bean, Frank D., 97n2,98n17, 135-36, 144, 147 Beggs, John J., 370, 373, 377 Bennett, Norman, 362,412 Bergsten, C. Fred, 262, 282n8 Berliner, Diane T., 365n6 Bhagwati, Jagdish, 268 Bierens, J., 374 Birrell, R., 61 Blejer, M., 322 Blinder, Alan, 373 Bloom, David, 198n4, 322 Bolin, R., 106, 108 Bojas, George, 29-30, 35, 51, 74nn8.1011, 171, 197n1, 198n5,201,203,230, 232111, 2331127,246, 317113, 321-22, 332-33.370 Boyd, Monica, 741114 Branson, William, 257n17 Brecher, R. A,, 285
425
Brown, Peter, 97n2 Browning, Harley L., 98n17 Card, David, 184,233n28, 364 Carliner, Geoffrey, 29, 321 Casas, F. R., 285 Caves, Richard, 282116 Chang, W. W., 286 Chapman, Bruce J., 370, 373, 377, 382 Chiswick, Barry R., 29, 741113, 95, 171, 197n1, 201, 230,246, 321, 332, 336, 370-73 Choudri, E. U., 285 Christofides, Louis N., 365n7 Cornelius, Wayne A., 86 Corwin, Arthur F., 97n2 Cousineau, Jean-Michel, 345 Cuthbert, Richard W., 97112 Daly, A , , 386, 397,400 Deardorff, A., 303n7, 312 Dewald, William, 262 Dickens, William T., 255nn1,3, 3651112 Diez-Canedo, J., 104, 118nl Dobson, Annette J., 111 Douglas, Paul H.,33, 197nl Dunlevy, J. A., 128 Dunning, John H., 282n7 Easton, S. T., 285-86 Eichengreen, Barry, 257n17 Eliason, Alan, 89 Elliot, Kimbedey, 365n6
426
Author Index
Elwell, Patricia J., 74n14 Espenshade, Thomas, 232111 Fallon, P. R., 285 Farber, Henry S., 36518 Fields, G., 127 Filer, Randall, 232nI7 Flamm, K., 106, 109 Fogel, Walter, 97n2 Fox, Marc, 322, 331 Frank, Robert H., 262 Freeman, Richard, 198114, 255113, 262, 365n1,366nn16,19,416 Frisbie, W. Farker, 981117 Garcia y Griego, M., 102, 104, 110, 119n2 General Accounting Office, 232111 Gerking, S . , 300 Goldberg, I., 322 Gorman, W., 302111 Graham,Edward M., 266 Grant, J., 300 Gray, Wayne, 408,419n4 Greenwood, Michael J., 33, 73n1, 104, 127, 133118, 198112, 219, 232111 Gregory, R. G., 386, 397, 399-400,405nl Grossman, Gene, 237, 2571117, 347 Grossman, Jean, 232111 Grunwald, J., 106, 109 Haig, B. D., 373 Hamermesh, Daniel, 176, 206, 232n4,300, 365n7 Hansen, Niles, 106 Harris, Milton, 323 Hechter, M., 137 Heckman, James, 30 Heer, David M., 97n2 Heywood, John S., 365116 Hicks, John R., 32, 285 Ho, V., 386,399-400,405nl Holmstrom, Bengt, 323 Horst, Thomas, 262, 282118 Houstoun, M. F., 97n2 Hufbauer, Gary, 365n6 Hughes, B., 386,405113 Huizinga, Hmy, 238 Hymer, Stephen, 282n6 Immigration and Naturalization Service, 83 Inglis, P., 377 Iredale, R., 382 Isoard, C., 104
Jackson, J. A., 29 Jasso, Guillermina, 73n2, 322 Jensen, L., 135, I51 Johnson, George, 198112, 201, 206, 2321110 Jones, R. W., 285-86, 302n2 Jones, Richard C., 97n2 Kahn, Shulamit, 36.5116 Katz, Eliakim, 323 Katz, Lawrence F., 255113, 366nn16,19 Keely, Charles B., 74n14 Kemp, M. C., 285 Kindleberger, Charles, 282n6 King, Allan G., 97112 Kogut, Bruce, 266 Krueger, Alan, 255113, 397 Kruse, Douglas, 22 Kubat, Daniel, 741114 Kuhn, Peter, 232nn2,5 Kumar, Pradeep, 365n5 Ladman, J., 104 Lam, Kitchun, 322, 333 Lancaster, Clarice, 97n2 Lawrence, Colin, 237, 2581127 Lawrence, Robert Z., 237, 255111, 2581127 Layard, P. R. G., 285 Leamer, Edward, 312 Leonard, I., 160111, 346 Leontief, Wassily, 303n10 Lewis, H. G., 255113, 2581126 Lieberson, S., 137, 141 Lii, D. T., 139, 141 Lipsey, Robert E., 261, 264 Long, James E., 321 Lore, Dave, 282n11 Love, James P., 257n17 McCulloch, Rachel, 281115, 346 McDonald, Ian, 238,256n6 McDowell, John M., 73111, 198112, 219, 232111 Mankin, Eric, 268 Markusen, J., 285 Maskus, K. E., 300, 303n7 Massey, D. S., 97n2, 123 Medoff, James L., 255113,416 Melvin, J., 285 Miller, Paul W., 74n13, 373, 377 Moran, Theodore H., 262, 282118 Muller, Thomas, 232111 Murphy, Kevin M., 255113
427
Author Index
Musgrave, Peggy B., 262 Mutti, J., 300 Nelder, J. A,, 11 1 Nelson, C., 142 Niland, J. R., 405112 Noms, K., 386 North, D. S., 97n2, 121 Oaxaca, R., 373 O’Connell, Martin, 82, 97n6 Office of Management and Budget, 191 Papademetriou, Demetrios, 232nnl,2, 2331118 Passel, Jeffrey S., 5, 1 2 , 7 8 4 0 , 83,91, 97n2 Petersen, T., 142 Piore, M. J., 106, 108 Portes, A., 136, 140-41, 147, 162n12 Price, Charles, 741114 Psacharopoulos, George, 54 Reich, Robert, 268 Reichert, J. S., 97n2 Revenga, Ana L., 256n12 Richardson, J. David, 281n5 Riddell, W. Craig, 345, 365nn2,7 Rivera-Batiz, F., 285-86, 300-301 Robb, Richard, 30 Robinson, J. G., 97nn2,3 Rose, Nancy L., 237 Rosenzweig, Mark R., 73n2, 322 ROY,A. D., 31-36 Ruffin, R. J., 285-87 Rugman, Alan M., 282117 Samuelson, Paul A,, 285, 386 Sandefur, G., 135, 142, 145-46 Sceuren, Frederick J., 97n2 Scheinkman, J. A., 286, 302n2 Schoepfle, Gregory, 362, 3651112,412 Schwartz, Aba, 47 Sehgal, E., 317113, 318114
Shue, Henry, 97n2 Siegal, Jacob S., 97n2 Sjaastad, Larry, 32, 126 Smith, J. P., 160nll Snipp, C. M., 145 Solow, Robert, 238, 256n6 Stark,Oded, 322-23 Sterens, Joe B., 97112 Stern, R. M., 300, 303117 Stolper, W. F., 386 Stromback, T., 373, 377 Summers, Lawrence, 255113, 397 Tandon, B. B., 741113 Teitelbaum, Michael S., 77 Thurston, L., 322 Tienda, Marta, 135-36, 139, 141-42, 14445, 147, 151 Topel, Robert H., 198117, 255n3 Tyson, Laura, 365n12 Unesco, 74119 United Nations, 741112 U.S. Bureau of the Census, 79, 191, 417 U.S. Department of Commerce, 14,301, 365113,410 Vernon, Raymond, 262 Vroman, Wayne, 352, 362, 365n12, 3661117 Warren, Robert, 5,78-80,91,97n2 Welch, F., 160n1, 198n4 White, Halbert, 2571118 Wilson, William J., 135, 142 Woodrow, Karen A., 12, 97n2 Wooton, Ian, 232nn2,5 World Bank, 73n4 Yezer, Anthony, 322 Zabala, Craig, 409-1 1 Zazueta, C., 102, 104, 110, 119n2 Zysman, John, 3651112
This Page Intentionally Left Blank
Subject Index
Apprehensions, illegal immigrant, 84-91,97 Assimilation, immigrant, 30; in Canada, 338, 341; geographic dispersion as indicator of, 123 Assimilation effect, 40-41, 69, 338, 341 Balance of payments, US.,6 Bargaining unit, 346-52, 365n11 Border Industrialization Program, Mexico, 106 Bracero Program, 85-87, 90, 106, 116 Capital flows: deregulation of, 263-64; types of, 8-9 Collective bargaining: effect of international competition on domestic, 346-47; in U.S.foreign-owned businesses, 278-79. See also Bargaining unit; Wage settlements Commercial zone, border, 106-10. See also Border Industrialization Program, Mexico; Manufacturing sector; Maquila Program, Mexico; Immigrants, Mexicanborn; Migration; Trade policy Comparable worth concept, 399 Data sources: for collective bargaining, 347, 352-55, 361-65; for earnings of immigrants, 43; for ethnic job queues, 15556; for foreign-owned business, 269-70; for illegal Mexican workers, 77-78, 8082, 84, 89-90, 96; for immigration’s effect on labor markets, 168, 178-79, 188, 191-92, 197; for immigration’s effect on
429
less-skilled natives, 217, 226-27; for immigration in Australia, 370-73; for internal migration, 128, 137, 139; NBER Immigration, Trade, and Labor Markets Data Files, 11,407-20; for open economy wage and employment, 242-45, 248; for self-selectionanalysis, 61, 63 Demand, aggregate, 308-17 Earnings: Australia, of immigrants in, 67, 69, 373-77; Australia, of low-skilled immigrants, 385; Canada, analysis of pattern of immigrant, 332-40; differentials between immigrants and natives, 40-43; effect of changes in sales on, 246-48; effect of immigration on, 174, 178-90, effect of mobility and job segmentation on, 151-53; ranking of minority men for, 145-46; trend for immigrants, 63-66. See also Assimilation effect; Wage rates; Wages; Wage structure, industry Education: and immigrant unemployment, 379-82; level of, for Mexican immigrants, 109-10; level of, for minority men, 142; role in immigrant wage adjustment, 374 Employment: displacement, 212-16; in foreign-owned firms, 1-2, 22-23; and import competition, 360-61; model for effect of international competition on, 347-49; opportunities with labor shortage, 161113;patterns with affirmative action, 160111; sectoral distribution of, 306-8; shift in composition of, 1
430
Subject Index
Encuesta Nacional de Emigracih a la Frontera Norte del Pais y a 10s Estados Unidos, 89, 101-2 Equal pay concepts, 399 Export-intensive sector, 2, 16-17 Factor intensity matrix, 301-2 Factors of production: effect of (im)migration on supply of, 288-91; and endowment changes, 287-91; intensities for traded and nontraded sectors, 287, 296-300; and prices, 286-88, 291-96, 300-301 FDI. See Investment, foreign direct Final demand. See Demand, aggregate Financial markets, 263-64 Geographic concentration of immigrants, 12, 21-22, 170-71, 187, 136 Geographic dispersion: effect on ethnic groups of, 136; evidence of, 132; as indicator of immigrant assimilation, 123; of minority men, 143-44. See also Herfindahl index; Mobility Herfindahl index (for geographic distribution), 122-25 Human capital: characteristics in Australian analysis of, 374; earning levels in TCF industries with low, 396-400. See also Education Human capital model, 392,394-400,404 Immigrant-intensivesector: characteristics of workers in, 3, 17; protection in Australia for, 23 Immigrant pool, 270. See also Migrant pool Immigrant quality, 66-7 1. See also Assimilation effect; Present value differentials; Self-selection; Worker characteristics Immigrants: clustering of, 121; differences in observed and unobserved skills of, 48; flow to U.S. of, 5 , 21; geographic concentration of, 12, 21-22, 136, 170-71, 187; industry distribution of, 208-16; influences on characteristics of, 21; national origin of, 1; ratio of employment in industries of, 210; sectoral employment distribution for, 306-8; selfselection trend for, 63-66, 70, 73; unobserved ability in cross-sectional data of, 369-70; in U.S. labor force, 3, 5. See nlso Labor market; Workers, immigrant
Immigrants, illegal Mexican, 5, 77; amnesty for, 83-84, 981119; characteristics of, 91-95; estimates of, 78-81, 83, 85,9697, 97nn1,2; flow of, 21; increased apprehension of, 84-88, 97; relation of apprehensions to stock of, 88-91; seasonal pattern of, 90-91 Immigrants, Mexican-born: characteristicsof, 108-10; mortality rates of, 81-84, 97nn3,5,7; pattern of migration, 102-6, 112-17 Immigration: to Canada, 326-30; crowding, 184-90; effect on employment, wages, and earnings, 169-70, 178-90; effect on less-skilled natives, 216-25; to U S . , 167, 187. See also Migration; Refugees, Indochinese; Return migration Immigration Acts (1910, 1968), Canada, 60, 324 Immigration and Nationalization Act (1952), US.:effect of 1965 amendments to, 35, 53, 60. See also Immigration Reform and Control Act Immigration and Naturalization Act (1923) U S . , 197nl Immigration policy: Australia, 23, 386-88; Canada, 23, 60, 71-72, 323-30,341; of host countries, 60-61; U.S., 35, 53, 68, 71, 73. See also Point system; White Australia Policy Immigration Reform and Control Act, or IRCA (1986), U.S., 77, 83, 197111 Immigration theories, 3 1-40 Import competition, 360-61 Import-intensivesector, 2, 16-17, 22 Income dispersion, Canada, 339-41. See also Out-migration, Canada Income-maximizationhypothesis, 32, 35 Index of labor market competition, 212 Industrial sector: distribution of immigrant workers in, 306-8, 317; employment differences in, 2; and foreign-owned businesses in U.S., 273-74; ratio of immigrant and native employment in, 20816 Industries, foreign-owned (in U.S.): research and development in, 276-78, 281; trade unions in, 278-79 Input-output analysis, 305-6 Investment, foreign direct (FDI): conditions and motivation for, 264-69; role of protection in, 267-68; in U.S., 1-2, 264,
431
Subject Index 270-79; U.S. inward and outward, 26162,264, 279, 281
Job segmentation, ethnic: analysis of effect of, 155-57; definition of, 139-40; representation of minority men in, 144-45 Labor force: Australia, wage policy for, 398401; composition and growth of, 1-2; effect of immigration on, 1; in foreignowned businesses, 275-76; immigrant ratio characteristicsin, 17; immigration as source of growth in, 167-68; participation of minority men in, 146-47, 150-51; U.S., immigrants in, 3, 5; workers in foreign- and U.S.-owned businesses, 18; workers in traded- and nontraded goods sectors, 16-17. See also Employment; Unemployment rates; Workers, immigrant; Workers, nativeborn; Workers, unionized Labor market: Canada, immigrants and native-born workers in, 330-31; Canada, response of immigration policy to, 32425; competition between immigrants and native workers in, 167-70,202-8, 21213, 226; competition in immigrant, 58; effect of aggregate demand on requirements of, 309-13; effect of immigration on domestic, 201-2; effect of migration on ethnic composition of, 161117; effect of shift in requirements on workers, 309-17; high-density areas of, 154-55; immigrant importance in local, 168-69, 171-74; minority men in, 136-37, 1425 1;response to trade-induced changes of, 236-42; skilled and unskilled sector in, 203; substitution effects in, 174-78; U.S., changing content of, 6-8; US., immigrant performance in, 55-60; US., internationalization of, 1-2; U.S., level of immigrants in, 167 Labor market areas: concentration of ethnic groups in, 154-55; determination of, 162n18 Manufacturing sector: composition of employment in, 1-2; Mexico, role of border commercial zone in, 106-10 Maquila Program, Mexico, 106 Migrant pool: determinant of quality of, 5253; Roy model categories for, 33-35
Migration: costs of, 37; determinants of, 110-17; effect on minority men of, 146; patterns of Mexican workers, 102-6; rate of, 32-33, 37; seasonal pattern of illegal, 90-91; theory of, using Roy model, 31-36,51, 54-55. See also Outmigration; Return migration Migration, internal, 121; determinants of, 126-30, 132; effect on wages of, 131; geographic dispersion as outcome of, 136; and minority men, 137-53; study of, 121 Mobility: costs, 36-37; for minority men, 143-44; patterns for recent immigrants, 122-25, 132. See also Migration Multinational corporations, U.S., 262, 26566,279,281n4 Nonparametric technique, to estimate wage functions, 374-75, 377 Out-migration, Canada, 322-23, 340-41 Point system, Canada, 71-72,325-26 Present value differentials, 41-43,55,66-70. See also Earnings; Wage rates; Wages; Wage structure, industry Present value differentials, 41-43, 55,66-70 Protection levels: Australia, 401; role in foreign direct investment of, 267-68 Refugees, Indochinese, 121 Remigration. See Return migration Return migration, 31, 370-73 Roy model, 31-36, 51, 54-55 Rybczynski effect, 289-90, 292-95, 302115 SAW. See Special Agricultural Worker (SAW) PWSelection mechanism: determinants for education, 53-54; determinants for unobserved characteristics, 33-36, 38, 51-53; of migrants, 31-33; in observed characteristics, 37-40. See also Self-selection Selectivity hypothesis, 29-31 Self-selection, 55, 63-66, 70, 73. See also Present value differentials SMSAs. See Standard Metropolitan Statistical Areas Special Agricultural Worker (SAW) program, 83-84
432
Subject Index
Standard Metropolitan Statistical Areas (SMSAs): on border with Mexico, 105; definition of, 191-92; in determination of labor market areas, 1621118;redefinition for analysis of, 228-29 Stolper-Samuelson theorem, 292-93, 295, 386 Textile, clothing, and footwear industries (TCF): Australia, 38546,401-2; immigrant distribution and earnings in, 38889, 392, 396-400 Traded-goods sector: characteristics of workers in, 6-7, 16; estimates by state of, 12-14. See also Export-intensive sector; Immigrant-intensive sector; Importintensive sector; Worker characteristics Trade flows, U.S., 6 Trade policy: Australia, to protect TCF jobs, 400-4; Mexico, Maquila Program of, 106-10; U.S., for products of border commercial zone, 106 Trade unions: effect of high representation, 23, 385-86; effect on wage responsiveness, 25 1-52; in foreign-owned industries, 278-79; model of behavior of, 238-39. See also Bargaining unit; Collective bargaining; Wage settlements Unemployment rates: for Australian immigrants, 377-79; for migrating minority men, 145, 147, 150 U.S. Bureau of Labor Statistics (BLS), Trade Monitoring System, 412-13, 419nn6,7 Wage rates: divergence of, from market clearing, 237; effect of immigrant labor on, 188-90,241-42; effect of internal migration on, 131; for minority men, 143;
in U.S. foreign-owned businesses, 27576 Wages: effect of changes in sales on, 251-55; effect of immigration on, 178-90; model for effect of international competition on, 347-49 Wage settlements, 347 Wage structure, industry: competitive decentralized setting of, 236, 255n2; effect of immigrant labor on, 22-23, 241-42, 246, 255, 373-77; effect of unionism on, 25 1-52, 255; with foreign'ownership, 275-76; minimum wage in, 385-86; response to product demand and trade flows, 235-36, 246-48,250-55; wages for men and women, 398-99 White Australia Policy, 60-61 Worker characteristics: export- and importintensive sectors, 2, 16-17, 22; observed and unobserved, 33-40,48, 51-53, 7172,369-70 Workers, immigrant: Canada, analysis of earnings patterns of, 332-40, Canada, labor market experiences of, 330-31; competition with native-born workers of, 167-70,202-8, 226; determinant of performance of, 31; distribution in industries of, 306-8 Workers, Mexican-born, 102-8, 112-17 Workers, native-born: analysis of effect of immigration on, 216-26; Canada, labor market experience of, 330-31; competition with immigrants of, 167-70,202-8, 226; employment displacement for lessskilled, 212-16; industry distribution of, 208-16; ratio to immigrants, 208-16; sectoral employment distribution for, 306-8 Workers, unionized, 345-47
This Page Intentionally Left Blank