Studies of Supply and Demand in Higher Education (National Bureau of Economic Research Project Report)

This Page Intentionally Left Blank

Studies of Supply and Demand in Higher Education

m[

A National Bureau of Economic Research Project Report

Studies of Supply and Demand in Higher Education

Edited by

Charles T. Clotfelter and Michael Rothschild

The University of Chicago Press Chicago and London

CHARLES T. CLOTFELTER is professor of public policy studies and economics at Duke University and a research associate of the National Bureau of Economic Research. MICHAEL ROTHSCHILD is professor of economics and dean of the Division of Social Sciences at the University of California at San Diego and a research associate of the National Bureau of Economic Research.

The University of Chicago Press, Chicago 60637 The University of Chicago Press, Ltd., London 0 1993 by the National Bureau of Economic Research All rights reserved. Published 1993 Printed in the United States of America

02010099989796959493 ISBN: 0-226-1 1054-0 (cloth)

12345

Library of Congress Cataloging-in-PublicationData Studies of supply and demand in higher education / edited by Charles T. Clotfelter and Michael Rothschild. p. cm. - (A National Bureau of Economic Research project re-

Po*) Papers presented at a conference held in May 1991 in Williamsburg, Virginia. Includes bibliographical references and index. 1. Education, Higher-Economic aspects-United StatesCongresses. 2. College attendance-United States-Congresses. 3, Education, Higher-United States-Finance-Congresses. I. Clotfelter, Charles T. 11. Rothschild, Michael. 111. Series. LC67.68. U6S78 1993 338.4'737873-dc20 92-37932 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 offinance and administration

Directors at Large John H. Biggs Andrew Brimrner Carl F. Christ George T. Conklin, Jr. Don R. Conlan Kathleen B . Cooper Jean A. Crockett George C. Eads

Martin Feldstein George Hatsopoulos Lawrence R. Klein Franklin A. Lindsay Paul W. McCracken Leo Melamed Robert T. Parry Peter G. Peterson

Douglas D. Purvis Robert V. Roosa Richard N. Rosett Bert Seidman Eli Shapiro Donald S. Wasserman

Directors by University Appointment Jagdish Bhagwati, Columbia William C. Brainard, Yale Glen C. Cain, Wisconsin Franklin Fisher, Massachusetts Institute of Technology Saul H. Hymans, Michigan Majorie B. McElroy, Duke

James L. Pierce, California, Berkeley Andrew Postlewaite, Pennsylvania Nathan Rosenberg, Stanford Harold T. Shapiro, Princeton Craig Swan, Minnesota Michael Yoshino, Harvard Arnold Zellner, Chicago

Directors by Appointment of Other Organizations Marcel Boyer, Canadian Economics Association Rueben C. Buse, American Agricultural Economics Association Richard A. Easterlin, Economic History Association Gail Fosler, The Conference Board A. Ronald Gallant, American Statistical Association Robert S . Hamada, American Finance Association

Charles Lave, American Economic Association Rudolph A. Oswald, American Federation of Lnbor and Congress of Industrial Organizations Dean P. Phypers, Commifteefor Economic Development James F. Smith, National Association for Business Economists Charles A. Walworth, American Institute of Certified Public Accountants

Directors Emeriti Moses Abramovitz Emilio G. Collado Thomas D. Flynn

Gottfried Haberler Geoffrey H. Moore James J. O’Leary

George B. Roberts 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 the 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 manuscript 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 manuscripts 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 general or 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 September 30, 1974)

Contents

Preface

Introduction Charles T. Clotfelter and Michael Rothschild 1.

2.

The University in the Marketplace: Some Insights and Some Puzzles Michael Rothschild and Lawrence J. White Comment: Martin Feldstein Adolescent Econometricians: How Do Youth Infer the Returns to Schooling? Charles F. Manski Comment: Eric A. Hanushek

ix 1

11

43

3.

Rends in College Entry among Whites, Blacks, and Hispanics 61 Robert M. Hauser Comment: Stephen V. Cameron and James J. Heckman

4.

The Growing Concentration of Top Students at Elite Schools Philip J. Cook and Robert H. Frank Comment: Malcolm Getz

5.

vii

Future Graduate Study and Academic Careers Jerry R. Green Comment: Charlotte V. Kuh

121

145

viii 6.

Contents

How Would Universities Respond to Increased Federal Support for Graduate Students?

183

Ronald G . Ehrenberg, Daniel I. Rees, and Dominic J. Brewer Comment: Michael S. McPherson 7.

8.

Optimal Investment Strategies for University Endowment Funds Robert C. Merton Comment: George M. Constantinides

21 1

Public Choices in Public Higher Education John M. Quigley and Daniel L. Rubinfeld Comment: Helen F. Ladd

243

Contributors

285

Author Index

287

Subject Index

29 1

Preface

In 1989 the National Bureau of Economic Research launched a program on the economics of higher education. Although numerous NBER studies have touched on economic aspects of education, previous to this effort the National Bureau had sponsored only a limited number of studies focused solely on education.’ Similar to those operating in other areas of applied economic analysis, such as labor and public economics, the NBER program in higher education has sponsored research projects and periodic meetings at which scholars discuss ongoing and completed research. The program’s first research project focused on three topics of current policy significance: the demand for undergraduate education, the supply of faculty, and the rise in the cost of higher education. This investigation culminated in the publication of Economic ChalEenges in Higher Education (1991, University of Chicago Press), by Charles T. Clotfelter, Ronald G. Ehrenberg, Malcolm Getz, and John J. Siegfried. The aim of this volume was to frame each of these questions in economic terms and to discuss data and empirical research that would be useful in answering them. The program’s working group on higher education met three times in 1989 and 1990 in Cambridge, Massachusetts, with some 35 NBER research associates and other economists participating. Unlike most groups of economists who study an industry, all of the participants had first-hand experience with “firms” in this industry, and a few were current or past administrators, includ1 . Among these sponsored studies are Education, Income, and Human Capital (1970). edited by W. Lee Hansen; Schooling, Experience, and Earnings (1974), by Jacob Mincer; Higher Education and Earnings: College as an Investment and Screening Device (1974), by Paul J. Taubman and Terence Wales; The Effect of School Quality on Achievement, Attainment Levels, andLifetime Earnings (1973, by Paul Wachtel; The Dejnition of College Quality and Its Impact on Earnings (1975), by Lewis C. Solomon; and Education as an Industry (1976), edited by Joseph T. Froomkin, Dean T. Jamison, and Roy Radner.

ix

x

Preface

ing several deans and provosts and one former president. While some of the participants brought with them considerable experience in research on higher education, others were experts in fields such as labor economics and industrial organization who were applying economic models to higher education for the first time. It became clear from the meetings of this group that higher education offered numerous questions to which economic analysis might be fruitfully applied. In light of these possibilities for applying economic approaches, the editors in 1990 invited a number of scholars to write papers for a conference devoted specifically to research on the economics of higher education. Authors were encouraged to present new empirical findings and to discuss issues of methodology that arise in the study of higher education. The authors could also consider the effects of public policy, but in accordance with the practice followed in Bureau studies, published papers can offer no policy recommendations. In commissioning these studies, we did not attempt to achieve comprehensive coverage of all aspects of the economics of higher education. Rather we sought to have solid research done on important topics, particularly topics that had previously not received a great deal of careful analysis by economists. The conference was held in May 1991 in Williamsburg, Virginia, with drafts presented and discussed that corresponded to the chapters in the current volume. Following the conference, the authors revised their papers, and the assigned discussants put their comments into written form. The objective of a conference such as this one is to foster research on an important topic of study in two ways-directly, through the published papers themselves, and indirectly, through the subsequent research that this work will stimulate. We believe not only that the first objective has been achieved but that there is every propect for success on the second as well. The chapters contained here provide new insights on important issues and raise a host of questions that should motivate future research. We are grateful to the Andrew W. Mellon Foundation for its support of this project. Charles 7: Clotfelter and Michael Rothschild

Introduction Charles T. Clotfelter and Michael Rothschild

Higher education in the United States is an enterprise of large proportions and far-reaching effects. There are today some 3,400 separately governed colleges and universities, ranging in size from colleges with a few hundred students to giant state universities enrolling tens of thousands. In 1990 these institutions together enrolled 14.2 million students and made expenditures amounting to 2.8 percent of gross national product (Clotfelter et al. 1991, 3; U.S. Department of Education 1991a, table 29; 1991b, table 1). At current rates of educational attainment, more than a quarter of all adults will have completed four years of college by middle age (U.S. Bureau of the Census 1990, table 217). Higher education not only affects the overall level of productivity in the economy but also is a major factor in determining the distribution of income.’ It also has widespread effects through the research undertaken at universities, although those effects are difficult to quantify. One distinctive feature of higher education in this country is the existence of a sizable private nonprofit sector. In this sector are some of the country’s most prominent colleges and universities. But the public sector, in the form of community colleges and state-supported colleges and universities, remains larger than the private and in fact constitutes one of the most important activities of state government, providing service functions to agriculture, industry, and local governments, in addition to research and teaching functions. Relative to other industries of comparable size, higher education has distinctive Charles T. Clotfelter is professor of public policy studies and economics at Duke University and a research associate of the National Bureau of Economic Research. Michael Rothschild is professor of economics and dean of the Division of Social Sciences at the University of California at San Diego and a research associate of the National Bureau of Economics. 1. For an analysis of the relationship between college enrollment trends and recent changes in earnings differentials, see Murphy and Welch (1989).

1

2

Charles T. Clotfelter and Michael Rothschild

features. Among them are the high degree of autonomy accorded to one group of employees-the faculty-and their nonhierarchical organization. As a subject of economic analysis, higher education is certainly not unexplored territory. Despite the inherent difficulties in attempting to undertake objective analysis of one’s own industry, a significant number of economists have devoted scholarly attention to colleges and universities. For example, economists have used the human capital model to examine such topics as the decisions of individuals to undertake the investment in college and the impact of college training on their subsequent earnings. There has also been considerable attention to policy issues related to student demand, such as the impact of tuition levels and scholarship programs on the number and composition of students entering college. A second major component of the literature on higher education has been supplied by labor economists, who have applied their tools of analysis to the academic job market. In addition, economists have examined a variety of other topics, such as research expenditures, productivity, and implications of public subsidies to state institution^.^ Despite the advances made by this research, the trends and debates of the last decade make it clear that there is much about the higher education industry that we still do not know. Although a decade of growth for higher education, the 1980s were also a period in which problems and criticisms became more prominent. Colleges and universities consistently raised their tuitions faster than inflation, prompting critics to call them “greedy” and inefficient (see, e.g., Bennett 1987, A31; Washington Post Weekly 1989; Finn 1984,2933, 47-51). Between 1979 and 1987, for example, tuition and fees increased in real terms at an average rate of 3.0 percent a year in public institutions and 4.9 percent a year in private institution^.^ Combined with reductions in federal funding for student aid grants, these increases raised concerns about the ability of low- and middle-income students to afford to attend college. One aspect of the existing financial aid system that came under special scrutiny was the practice, by several groups of selective private institutions, of comparing and adjusting the financial aid offers made to individual students. Defended as a means to take financial considerations out of college choice, this practice was investigated by the Justice Department as a possible antitrust violation. More generally, there were increasing signs that colleges were using marketing techniques and non-need-based scholarships often to attract top applicants. The academic job market also presented new challenges. One study (Bowen and Sosa 1989) predicted that during the period 1997 to 2002, shortages of arts and sciences faculty could develop on the order of 40 percent. After a decade of slack demand for Ph.D.’s in many fields, some questioned whether gradu2. For a discussion of the internal organization of universities, see Coleman (1973). 3. For references to the economic research on higher education, see, for example, Bowen (1968), Radner and Miller (1973, Froomkin et al. (1976), Hoenack and Collins (1990), and Clotfelteret al. (1991). 4.Calculations based on figures in Clotfelter et al. (1991), 125.

3

Introduction

ate programs would attract the number and quality of applicants necessary to sustain research and graduate education at previous levels. When confronted with issues such as these, economists see questions not unlike those that arise in the study of other industries in the economy. In particular, economists have tended to ask two basic questions about higher education: What mix of products does the higher education industry produce, and at what cost? Who pays for these products, and who benefits from them? The eight studies in this volume are no exception. Although most of them examine topics that have received relatively little attention to date, the studies presented here are concerned with these two basic issues. Two of the studies focus primarily on the first question. Rothschild and White examine the nature of competition in the higher education industry. Merton’s chapter is an analysis of how universities ought to manage their endowments, which is nothing more than using resources efficiently. Successful endowment management should reduce costs. Five studies (those by Manski, Hauser, Cook and Frank, Green, and Ehrenberg, Brewer, and Rees) address different aspects of who chooses to purchase what kind of education and why. The chapter by Quigley and Rubinfeld addresses both questions as it attempts to explain why in some states taxpayers pay for extensive (and expensive) systems of higher education while in others the variety and quality of state-subsidized higher education is more limited. A brief overview of these studies reveals the economic content in the issues they raise. In the first chapter of this volume, Michael Rothschild and Lawrence White look at how individual institutions operate within a larger marketplace. Although it should not be surprising that economists looking at higher education might view colleges and universities in much the same way as they view firms in other industries, there has in fact been little research with this kind of market orientation. But colleges and universities clearly do compete in a marketplace, as was recognized many years ago by University of Chicago president Robert Maynard Hutchins, who commented on the emphasis in college advertising on the beauty of campuses and the availability of recreational opportunities (1936, 29). And the issue of competition has, of course, recently taken on added policy importance in light of the Justice Department’s investigation into the financial aid practices of several groups of private institutions, noted above. Considered from a global perspective, one of the most unusual features of higher education in the United States is the amount of competition between different institutions. A strong and variegated private sector exists alongside many different public systems. These institutions compete for faculty, students, research grants, contributions, and access to the public purse. Some observers, such as Rosovsky (1990), attribute the vitality of U.S. higher education to its competitive nature. Yet very little work has been done analyzing how competition works in the higher education industry. White and Rothschild note several puzzles regarding the behavior of U.S. colleges and universities. One is that institutions with small endowments

4

Charles T. Clotfelter and Michael Rothschild

compete successfully (at least in the sense of survival) with institutions whose per student endowments exceed theirs by a hundredfold and more. This observation raises obvious and interesting questions about the nature of competition in the higher education industry. Other puzzles include the near-uniformity of tuition charges despite clear distinctions in prestige and the apparent resistance to charging revenue-maximizing rates. Questions such as these motivate the Rothschild and White paper, which looks at how individual institutions operate within a larger labor market. One explanation for below-market tuition charges is that it enables institutions to be choosy about what kind of students to accept; where there are externalities among different kinds of students, this may be efficient. The authors also consider the argument that universities use undergraduate education to subsidize research. Using a “standalone test,” they reject this notion because of the clear fact that universities are able to compete in the market for undergraduate education with colleges that produce only that service. In his comment on the paper, Martin Feldstein suggests that some of the puzzles raised by Rothschild and White can be explained by the set of incentives facing university administrators. Because of the nonprofit nature of these institutions, he argues, there is little to gain and much to suffer from undertaking unpopular but potentially beneficial changes. At the heart of the economic model of enrollment demand is the assumption that potential college students have a way of assessing the future increase in earnings that would result from attending college. Indeed, one likely explanation for the continued growth in college enrollments during the 1980s, a period in which the number of 18-year-olds was falling, is the strong rebound during the decade in the earnings advantage enjoyed by college graduates over high school graduates. Although this differential fell during the 1970s, it rose smartly after 1979. For example, the gap in average full-time earnings of male high school and college graduates fell from 42 percent in 1970 to 29 percent in 1979 but then jumped back to 50 percent by 1987 (Clotfelter et al. 1991, 65). How are young people-especially those from lower incomes who know fewer college students and graduates-expected to gather and evaluate information on the economic return to college? This is the beginning point for Charles Manski’s chapter. Manski works through a simple model of educational choice to illustrate the difficulty of processing information of this sort. He supposes that youths differ both in their abilities-the extent to which their incomes will increase if they go to school-and in their taste for schooling. He supposes that youths choose to attend college if they believe that college attendance will increase lifetime well-being, including both the effect of college on income and the actual utility (or disutility) of attending college. Expectations of the effect of college attendance on future income play a key role in this model. Manski considers two alternatives. In each case, youth look to the actual experience of the generation that attended college before them. In one, youth are presumed to have information about the ability of

5

Introduction

their elders and thus to know the relationship between ability, education, and income. In the other case, youth do not know the abilities of their elders and assume that all who go to college have the same expected income. The two assumptions about expectations yield two different equilibria; the differences are illuminating. For example, the average ability of the college-bound is higher in the first model than in the second. Manski goes on to consider how an econometrician, ignorant of the actual mechanism used to generate expectations but armed with the conventional faith in rational expectations and the conventional lack of concern for unmeasured variables (in this case, a taste for education)-would analyze the data produced by such models. Not surprisingly, the hypothetical econometrician would fail to grasp the mechanism which generates the returns to schooling and the determination of the decision to enroll in college. Manski concludes that the only way out of this kind of dilemma is to study explicitly the mechanism youth use to form expectations about schooling. He argues that economists must measure and use subjective variables in their studies of the enrollment decisions and returns to schooling. In his comment on this paper, Eric Hanushek expresses skepticism about the likelihood of economists being able to do this and argues that economists using and refining conventional tools can make progress in analyzing the role and effects of expectations. Recent large changes in the apparent returns to education provide a rich opportunity for this kind of analysis. Despite the overall increase in rates of college enrollment in the United States in recent decades, considerable concern has been expressed about the rates for minority groups. One statistic that has gained attention is the decline in one measure of the college enrollment rate for black and Hispanic high school graduates since the mid-1970s. There is also evidence that college completion rates among blacks have declined markedly from the early 1970s to the mid-1980s. A related trend is a growing gap in enrollment rates between children in families in the top quintile of incomes and other college-age youth. Robert Hauser uses data from the annual Current Population Surveys from 1972 to 1988 to examine trends in college enrollment of young people, with special attention to differences by race and ethnicity. He estimated equations explaining college entry over this period and found that the difference in rates between blacks and whites can be explained by differences in social background. Compared to those of other racial and ethnic groups, white high school graduates come from families that have fewer children, are more likely to own their own house, and have parents who are better educated and have higher status jobs. Holding social background constant, Hauser shows that college entry rates of blacks actually have remained above those of whites. In their comment on this paper, Steven Cameron and James Heckman question some of Hauser’s conclusions. They argue that the census data Hauser uses are not sufficiently rich to permit a complete analysis of the determinants of college entry. Their own work (Cameron and Heckman 1992) using a data set which has richer and better information about individual characteristics-in

6

Charles T. Clotfelter and Michael Rothschild

particular, family income-suggests that differences in college attendance among racial groups cannot be attributed solely to differences in backgrounds. In their chapter, Philip Cook and Robert Frank focus on another aspect of undergraduate enrollment patterns: where do the best students go? Among the vast number of American undergraduate colleges, an acknowledged few certainly stand out as “elite” institutions. These have contributed far more than their numeric share of leaders in various occupations. Yet the hold these institutions have on the leadership in this country is less than comparable institutions have in some other countries. In Japan, for example, a third of presidents of large companies, over 60 percent of senior government officials, and virtually all postwar prime ministers are graduates of Tokyo University, which produces only about 1 percent of the country’s university graduates (Rohlen 1983, 88, 91; Fallows 1990, 17-18). In the United States, the influence of elite institutions appears to be quite a bit less. Of the eight presidents since 1960, for example, only two, John Kennedy and George Bush, received bachelor’s degrees from elite private institutions. But is the concentration of talented students in such colleges increasing in this country? Cook and Frank present evidence that it is. They show, for example, that the percentage of a well-known national science competition’s finalists going to Harvard increased between the 1960s and the 1980s from 18 to 22 percent, and the percentage going to one of the top seven colleges increased from 47 to 59 percent. Similarly, the odds that a top-scoring freshman at the University of California will attend the flagship Berkeley campus rose from 2.8 times the odds that any freshman would attend Berkeley in 1980 to 14.7 times the average odds in 1988. If it is indicative of a more general concentration of influence among the set of American colleges and universities, this trend would represent a moving away from a system that offers a relatively large number of independent avenues to positions of power. In his comments, Malcolm Getz questions the definition of elite used by Cook and Frank, considers the possibility that the trends they uncover may be part of a very long-term change, and points out that there are still enough elite institutions to ensure some level of choice and competition among them. Because of their role in determining the quality of academic research and graduate education, undergraduates who choose academic careers are obviously an important input in the economics of higher education. We have a good idea of the numbers of undergraduates who enter graduate programs but lack good information on their quality. Using a unique data set composed of questionnaires completed by virtually all of the graduates of Harvard College from 1985 to 1990, Jerry Green asks whether the quality of college students who intend to become academics has been changing. Overall, he finds little evidence of increased or decreased interest in academic careers during the period, although the data for 1990 may indicate the beginning of a trend toward academic careers. Among humanities majors, however, an increase in interest over the period is evident. Probably the most noticeable change in the

7

Introduction

pattern of career choice over this period is a decline in interest in medicine and a corresponding increase in interest in life sciences. In her comment on Green’s paper, Charlotte Kuh presents a comparison of the distribution by field of Harvard B.A.’s with the overall distribution and concludes that the Harvard distribution is roughly representative. She draws attention to the decline in the percentage of high honors students selecting doctoral work in the physical sciences as a potentially worrisome trend among the findings presented by Green. Another element in determining the flow of talented students into Ph.D. programs is the amount of financial support available. The federal government has been a major source of such support, but this source has also been subject to fluctuation. Ehrenberg, Rees, and Brewer examine the effect that changes in federally financed fellowships have had on universities. Looking at the pattern of adjustments made by institutions in reaction to changes in outside funding of graduate fellowships, they find that institutions tend to compensate for such changes by substituting internally financed fellowship support for federal funds. Although such response is quick, it is not one-for-one; on average, institutions decrease their own funding so as to drop about one student for each four additional students supported by outside funds. Examining behavior by field, the authors again find that this general story of internal fungibility applies in most areas. They caution, however, that institutions may redirect or save internal funds in ways that may further mitigate the effects of changes in outside support. In his discussion of the paper, Michael McPherson notes that the subject of this research can be seen as a special case of the more general question of how outside funding affects the behavior of universities. Do they view the funding as temporary and hoard it, or do they view it as permanent and adjust their long-run behavior accordingly? McPherson echoes the authors’ conclusion that answering such questions requires a fuller model of the university than we have at present. In his chapter, Robert Merton focuses on an important topic that has received surprisingly little attention from economists: How should universities manage their endowments? Although only a few universities have endowments ranking them in the billion-dollar club, some 67 universities had endowments as large as $200 million in 1990, and many more had holdings whose income represents a significant share of their budgets. Merton begins with the standard model of portfolio management. Applying this model to universities requires care because university wealth includes much more than the financial assets in its portfolio. Other major assets include such obvious ones as the institution’s land and physical plant as well as less easily measured ones, such as the future stream of expected gifts from alumni and other donors. Because of the variety of this asset mix, the principle of diversification that underlies the theory of optimal portfolio allocation cannot be applied only to financial assets; rather, all assets must be taken into account. Since it is relatively easy to manage, the university’s endowment can be used to offset

8

Charles T. Clotfelter and Michael Rothschild

changes in the larger set of assets. It can be used, for example, as a hedge against anticipated changes in costs. If the local cost of housing affects faculty salaries, an institution can hold local real estate in its portfolio as a hedge against future inflation. This approach may also have implications for the kinds of financial assets a university holds. For example, a technical institution, many of whose alumni work in high-tech industries, should probably invest less of its endowment in those industries, since contributions it receives in the future will most likely be positively correlated with the fortunes of those industries. In his comment on this paper, George Constantinides raises questions about the basic assumption of Merton’s model-that the university should, like a consumer, maximize a discounted sum of future utility. Universities are not, Constantinides notes, individuals. Instead they are more like business firmseconomic institutions which exist to serve individual needs. Heterogeneous investors benefit by choosing from the diversified offerings of specialized investment vehicles that are harmed if firms force diversification by forming inefficient conglomerates. In the same way, Constantinides argues, society may be the poorer if some universities diversify as Merton suggests they should. Constantinides’ comment also shows how Merton’s analysis, which uses the technology of continuous time stochastic processes common to the finance literature, may be recast in more familiar terms. As noted above, one distinguishing feature of the American system of higher education is the coexistence of strong private and public sectors. The public sector, consisting of institutions operated by state and local governments, is by far the larger, with public institutions enrolling more than twothirds of all four-year college students and virtually all two-year college students. For state governments, higher education is a major function, accounting for a fifth of all direct expenditures. In the final chapter of the volume, John Quigley and Daniel Rubinfeld examine the provision of public higher education by the states. Historically, they explain, public colleges and universities arose in the shadow of largely preexisting private institutions. In the East, where private colleges were established early, public institutions tend to be less important. By contrast, the public sector is dominant in the newer states of the West and Midwest. One interesting fact consistent with the view that public and private institutions act as close substitutes is the finding that public and private tuitions in a state tend to be positively correlated. Not only does there appear to be a trade-off between states’ public and private enrollments, but there is also an apparent trade-off between two-year and fouryear colleges. There is substantial variation among states in the amount of public higher education provided and also in the degree to which attendance is subsidized. Perhaps the biggest question concerning the public provision of higher education is why state governments do it at all. Quigley and Rubinfeld provide several alternative explanations, including the possibility that in this country higher education approaches the status of secular religion, receiving

9

Introduction

almost unquestioned support. In her comment on the paper, Helen Ladd draws attention to the heavy hand of history in determining the different publicprivate splits observed by region, and she suggests alternative ways to model the public supply of higher education. In the end, we are left with a number of unresolved puzzles, among them why states differ in the degree to which they subsidize higher education, and particularly the education of out-of-state students. The purpose of this volume is to bring economic analysis to bear upon issues affecting higher education. We believe that the importance of the studies contained herein can be judged using more than one yardstick. Not only do they reveal new empirical findings and provide methodological insights, they also raise questions for future research. They reveal ways in which our understanding of supply and demand for higher education remains undeveloped. More work needs to be done, for example, on the effects of tuition and financial aid programs on the level and composition of demand for college. The labor market for academics, including the supply of women and minorities, remains inadequately understood. And the technology of production in the industry-in both the teaching and the research aspects-continues to be difficult to model, even for those who work there. We believe, however, that the studies presented here represent a useful step in the direction of a fuller understanding of the workings of this important industry.

References Bennett, William. 1987. Our greedy colleges. New York Times, February 18:A31. Bowen, William G. 1968. The economics ofmajor private universities. Berkeley: University of California Press. Bowen, William G . , and Julie Ann Sosa. 1989. Prospects for faculty in the arts and sciences. Princeton, N .J.: Princeton University Press. Clotfelter, Charles T., Ronald G. Ehrenberg, Malcolm Getz, and John J. Siegfried. 1991. Economic challenges in higher education. Chicago: University of Chicago Press. Coleman, James S. 1973. The university and society’s new demands upon it. In Content and context, ed. Carl Kaysen. New York, McGraw-Hill. Fallows, James. 1990. Wake up, America! New York Review of Books, March 1:17-1 8. Finn, Chester. 1984. Trying higher education: An eight count indictment. Change 16(May-June): 29-33,47-51. Froomkin, Joseph T., Dean T. Jamison, and Roy Radner, eds. 1976. Education as an industry. Cambridge, Mass.: Ballinger. Hoenack, Stephen A., and Eileen L. Collins, eds. 1990. The economics ofAmerican universities. Albany: State University of New York Press.

10

Charles T. Clotfelter and Michael Rothschild

Hutchins, Robert Maynard. 1936. The higher learning in America. New Haven, Conn.: Yale University Press. Murphy, Kevin, and Finis Welch. 1989. Wage premiums for college graduates: Recent growth and possible explanations. Educational Researcher 18(May): 17-26. Radner, Roy, and Leonard S. Miller. 1975. Demand and supply in U S . higher education. New York: McGraw-Hill. Rohlen, Thomas P. 1983. Japan’s high schools. Berkeley: University of California Press. Rosovsky, Henry. 1990. The university: an owner’s manual. New York: W. W. Norton. U.S. Bureau of the Census. 1990. Statistical abstract of the United States. U.S. Department of Education. 1991a. Digest of education statistics 1991. Washington, D.C.: Government Printing Office. . 1991b. National higher education statistics: Fall 1991. Washington, D.C.: Department of Education. Washington Post Weekly. 1989. Colleges: A machine with no brakes. August 21-27.

1

The University in the Marketplace: Some Insights and Some Puzzles Michael Rothschild and Lawrence J. White

1.1 Introduction The application of economics principles to the behavior of colleges and universities is a topic of substantial interest and importance. The literature on various aspects of the economics of higher education is large and growing rapidly. The resources commanded by all institutions of higher learning are large. In 1989 the aggregate expenditures of all two- and four-year undergraduate colleges and postgraduate institutions came to $131.4 billion. For purposes of comparison, this sum exceeded the sales of any three-digit manufacturing industry except petroleum refining and motor vehicles and of any threedigit service industry except hospitals. Much of the application of economics principles to university behavior has focused on cost measurements and allocation issues. Surprisingly, there has been little attention given to the questions concerning the marketplace context of universities: how they compete for faculty (“inputs”), how they “position” themselves in the marketplace, how they decide on “prices” (tuition, room and board charges for resident students, etc.), how they decide on production levels (the number of students to admit), when to enter new markets (e.g., offering new programs or degrees, establishing new professional schools), and so on. Though a few authors briefly mention “competition” among uniMichael Rothschild is professor of economics and dean of social sciences at the University of California at San Diego and a research associate of the National Bureau of Economic Research. Lawrence J. White is the Arthur E. Imperatore Professor of Economics and chairman of the Department of Economics at the Stern School of Business, New York University. The authors would like to thank Vince Crawford, Dermot Gately, Julianne Nelson, Sherwin Rosen, and the participants in the NBER Conference on Higher Education, especially Charles Clotfelter, for useful comments and suggestions on an earlier draft; they also thank Peter Rousseau for research assistance on this project. 1, See, for example, the recent collection of surveys edited by Hoenack and Collins (1990).

11

12

Michael Rothschild and Lawrence J. White

versities (e.g., Bok 1990, 104; Bowen 1980; Garvin 1980), none has analyzed this competitive process or made serious estimates of relevant parameters. We believe that this absence of apparent interest in the market context may undermine-or, at a minimum, mask some crucial assumptions in-the cost or allocation analyses undertaken by some authors. For example, the “autonomous cost increase” model of Massy (1989) has embedded in it an implicit assumption that every university is a separate monopoly that faces an inelastic demand and that can raise its prices at will to cover all cost increases. Perhaps this is indeed the case; but if so, an unaddressed issue in the Massy analysis is the question of why universities have been so slow to raise their prices and revenues and thereby raise their expenditures. In any event, an explicit statement of this assumption would make clearer the basis for the Massy analysis. At the opposite extreme, the allocation analyses of James (1978, 1986) and James and Neuberger (1981) assume that tuition prices are predetermined and beyond the control of the individual institution. Should we be comfortable with that basis for analysis? As yet a third example, we note that a number of the authors providing estimates of the price elasticity of demand for higher education seem uninterested in whether they are measuring the price elasticity for higher education in aggregate or the cross-elasticity of some institutions vis-h-vis others.2 Only researchers who were uninterested in market contexts would fail to be interested in the distinction. We believe that the market context of higher education-whether universities compete, how they compete, and the consequences of that competition for university input, production, pricing, and output decisions-is interesting in its own right and important for understanding the cost and allocation issues that have concerned most researchers. This paper cannot possibly answer all of the relevant questions concerning competition among universities. But we hope to deal with some of them and raise important questions and puzzles for others to pursue. Indeed we hope to provoke and challenge at least as much as we analyze and explain. In section 1.2 we address an allocation issue that has been raised by others (e.g., James 1978, 1986): Is there “cross-subsidy” among a typical university’s activities and specifically between undergraduate and graduate teaching? We introduce the “stand-alone test” of Faulhaber (1975) to show that the previous claims of substantial cross-subsidy do not rest on a solid analytical foundation. Section 1.3 analyzes issues of student quality and diversity in a university and their consequences for output and pricing. Suppose the mix of students affects the efficiency of teaching (e.g., for any average level of learning ability by students, the necessity for repetition or remedial effort will be less when the variance of learning abilities is less) or the quality of output (e.g., stu2. See the surveys in Leslie and Brinkman (1987) and Becker (1990).

13

The University in the Marketplace

dents’ learning is enhanced by having fellow students with a diversity of backgrounds). An efficient university admissions policy would be to encourage a large pool of applicants (e.g., through apparent “underpricing”) and then to accept students selectively and to price selectively (i.e., to practice price discrimination through selective scholarships) so as to achieve the efficiencies that accompany the mix and diversity characteristics. Simple market-clearing prices would not be as likely to achieve these efficiencies. In essence, the incoming students themselves are an important input to (and affect the efficiency of production of) the educational services output of the university, and the university’s admission and pricing policies are likely to reflect this special relationship between input and o u t p ~ t . ~ Section 1.4 addresses some of the broader pricing, market, and competition questions. After noting that the preferences of the providers of nontuition funds must be a part of the analysis of university behavior, we first examine some questions concerning input behavior (e.g., why are research universities reluctant to reward teaching performance?) and then largely focus on output pricing and market behavior: For example, why do universities pass up apparent opportunities to practice revenue-increasing price discrimination? Why do universities generally change uniform tuition levels across different fields that appear to have substantially different marginal costs? Why do universities fail to price so as to capture the rents that attach to their brand-name reputations? What motivates entry and exit among universities? For most of these (and other) questions, we can offer insights and clarifications, but many basic puzzles remain. Section 1.5 offers a brief conclusion.

1.2 The Criteria for Cross-Subsidization The observation that in the modem research university, undergraduate education subsidizes graduate education and research is commonplace. Estelle James (1978, 1986, 1990) has been the most prominent and consistent proponent of this view. It is based on an analytic model of the goals of the university set down most precisely in a joint article with Egon Neuberger (James and Neuberger 1981). The simplest version of their argument runs as follows: Suppose that a university department’s only revenue comes from teaching undergraduates and that its only expense is buying (at the market price) faculty time. Faculty time may be allocated either to teaching undergraduates or to research. The department maximizes a utility function in which research is an argument subject to the constraint that expenditures may not exceed revenue. If any faculty time is spent doing research, then James would argue that undergraduate education 3. We can also draw the parallel here with the hiring policies of some companies that offer apparently “above-market” wages in order to obtain a selective and stable work force.

14

Michael Rothschild and Lawrence J. White

is subsidizing research because undergraduate education produces revenues while research does not. A more sophisticated version of the argument allows the faculty to care about the quality (but not the quantity) of undergraduate education and allows for revenue from graduate students and research, but reaches the same conclusion because large undergraduate courses can be used to bring in revenue so that faculty can do more research. It is this sophisticated version of the argument that James takes as embodying cross-subsidization; “profits” from undergraduates subsidize graduate education and research that “usually do not bring in enough revenues to cover their costs” (1986, 237). One of the predictions of James’s model is that universities will produce undergraduate education with a different technology than institutions that do not have graduate students and that do not do research. Liberal arts colleges and community colleges have smaller classes on average than research universities and do not use graduate students as teachers. Because the former technology is cheaper than the latter, undergraduates subsidize graduate education. James (1986) argues that consideration of the cross-subsidy issue should make one reconsider arguments about the effects of state educational policy on the distribution of income. Hansen and Weisbrod (1969) argued that the California system of public education subsidizes higher-income residents of that state, since the latter’s children tend to go to institutions (universities) at which the cost per pupil is higher than at the institutions (state and community colleges) to which the poor send their children. By James’s reckoning, this is incorrect because the real cost of providing undergraduate education in a university is rather low. Intuitively, there seems to be something wrong with the James argument because, unlike the privilege of sleeping under the bridges of Pans, admission to the University of California is restricted. Because there is evidence that consumers will pay more to attend universities than to attend community colleges, attendance at the former institutions must be worth more than enrollment at the latter. This observation is, in essence, the basis for our belief that undergraduate education does not subsidize graduate education and research. Undergraduate education is produced as both a joint product with graduate education (in research universities) and, at the same time, the only product of some firms in the education business-particularly liberal arts colleges and community colleges. Thus, undergraduate education produced as a joint product survives in a competitive market with undergraduate education produced as a sole product. The modem definition of cross-subsidization takes this fact as evidence that graduate education and research are not subsidized by undergraduate education. A concise statement of the modern definition of cross-subsidization for the multiproduct firm is due to Faulhaber (1975). It is a sophisticated version of the stand-alone test. Suppose a firm produces goods that serve N different classes of customers. We want to ask whether or not the existing prices have

15

The University in the Marketplace

an element of cross-subsidy. According to Faulhaber, they do if it is possible for another firm to serve a subset of these customers and make a profit. The entering firm, of course, can only serve this subset of customers if these customers choose to be customers of the entrant rather than the incumbent-that is, if the entrant offers a more attractive price and quality combination. In symbols, let S = {1,2, . . . , N } ; then if y C RNproduces revenue ~ ( y ) , there is no cross-subsidization if for all subsets TcS,C ( y 3 2 r ( y 3 , where

and C(y’) is the cost of producing yT. The application to education is immediate. Suppose that the three classes of consumers are undergraduates, graduates, and consumers of research. Then the fact that firms in the education industry that serve all three kinds of consumers survive in competition with firms that serve only one kind of customer is a demonstration that undergraduate education does not crosssubsidize graduate education and research. Two arguments against this position must be considered. The first rests on the observation that the undergraduate education students get from a liberal arts college is different from the undergraduate education that students at a “multiversity” (Kerr 1964) receive. As we have observed, James states that a confirmed prediction of her theory is the fact that undergraduate colleges will have smaller class sizes than research universities. However, this seems to miss the point. Harvard and Swarthmore compete for the same students; so do UCLA and the Claremont colleges. Large research universities have larger classes than liberal arts colleges, but the different variants of the product survive in competition; Fords are different from Chevys, but both brands compete for the same customers. Another objection is the observation that the zero-profit constraint is inappropriate for institutions of higher education. This has considerable force. Certainly it is a bit difficult to state precisely the yearly budget constraint of a private nonprofit institution with a large endowment that receives many charitable contributions (some from alumni, which might be considered as deferred payments of tuition) and sells research to many governmental agencies. Equally murky is the budget constraint of a public university that receives capitation fees for some students, sells research to governments, has an endowment, and can call on the state to fund its buildings with various kinds of bonds. However, it remains true that institutions of higher education do face some kind of long-run budget constraint. These constraints clearly involve subsidies. Within this complex system of subsidies, institutions that sell both graduate and undergraduate education survive in competition with institutions that sell only undergraduate education. While colleges that produce only undergraduate education are common,

16

Michael Rothschild and Lawrence J. White

institutions that produce only graduate education and/or research are rare . 4 Even if they did not exist at all, it would, we think, be incorrect to conclude that undergraduate education in the large research university subsidizes graduate education. Although we can only offer partial e ~ i d e n c e it, ~is clear that there are economies of scope in higher education. Being part of a research university confers considerable benefits to undergraduates, benefits for which they are willing to pay both in money and in the acceptance of what some deem a poorer educational technology-larger classes and graduate student instructors. Some of the sources of these economies are obvious: library and computer facilities, the possibility of contact with the latest research, sheer size, and diversity; doubtless there are many others. Undergraduates and their parents value these things.

1.3 Admissions Policies: Selectivity and Pricing Prices ration access to many goods in our society. A conspicuous exception is the right to attend the best institutions of higher education. Cost considerations do affect where and whether people go to college. However, by definition, select colleges and universities receive more applications than they can accept; many public colleges and universities will only admit students who achieve a particular academic standard. Why should this be so? The obvious answer is that it is not fair (in some sense) to let people buy their way into the best universities; access ought to be based on merit.‘j However, since in most other arenas the price system is an efficient way of allocating resources, it is interesting to examine whether or not the price system could in principle lead to an efficient allocation of students to different institutions of higher education. 1.3.1 The General Problem Suppose there are sets S = {1,2, . . . , N } of students and C = {1,2, . . . ,T } of colleges. An allocation A is an assignment of students to colleges, a mapping from S to C. Since one of the “colleges” in C can represent not going to college at all, the formulation is general. Under the allocation A , A(S) is the college that student S attends. We summarize the benefit a student gets from a college in a single number W,[A(S)]. W,[A(S)] is a net benefit; the real costs of attending college (mostly forgone earnings) are included; the price, or tuition, that the college charges is excluded. The subscript A indicates that 4.RAND, the Salk Institute, the Institute for Advanced Study, and Rockefeller University are examples. It is our casual impression that these institutions have somewhat more difficulty providing a steady flow of funds for their researchers than do institutions that produce undergraduate education as well as research and graduate education. 5 . The cost functions estimated by Cohn, Rhine, and Santos (1989) indicate significant economies of scope in universities. 6. Rosovsky (1990) makes this argument eloquently.

17

The University in the Marketplace

benefits depend not only on the college attended but also on the complete allocation of students to colleges. The total surplus of an allocation A is just B(A) = C;=, W,[A(S)].An allocation is efficient if B(A) 2 B(A’) for all allocations A ’. Allocations differ in efficiency only if there is some synergy. If the costs that a college incurs are the same for all students and if attendance at that college increases a student’s human capital by the same amount regardless of the ability or composition of the student body, then the specific identities of the students who attend that college are irrelevant. If all colleges are like this, then all allocations are efficient. For allocations to have different efficiencies, it must be the case that students get different benefits from attending different colleges and that colleges’ net contribution to their students depends on the students themselves. We can only have a concern for the efficiency of different allocations if students themselves are an important input into the educational process. If students are inputs in this sense, then there are externalities in the higher education industry. A large and beautiful literature focuses on matching problems of this sort. The phenomena of college admissions motivated much of the work in this area. In fact, the title of the seminal paper is “College Admissions and the Stability of Marriage” (Gale and Shapley 1962). Unfortunately the line of research that Gale and Shapley initiated can deal easily with only a restricted set of externalities. Perhaps the most convincing demonstration of the relevance of sophisticated game theory to real economic decisions is Roth’s (1984) study of the matching problem for medical interns. Briefly, Roth showed that the procedure (called the National Intern Matching Program, or NIMP) used since the early 1950s to assign interns to hospitals worked because it produced stable allocations in the following sense: given the allocation produced by the NIMP, there did not exist a hospital or an intern not matched by the NIMP such that the intern preferred to be matched to the hospital and the hospital preferred the intern to an intern assigned to the hospital by the NIMP. The NIMP was stable because no hospital and intern could both improve their situation by defecting from the NIMP. This abstract result goes a long way to explain the remarkable success and stability of the entirely voluntary NIMP. The proof that the NIMP produces stable allocations assumes that the preferences of hospitals for groups of interns and of interns for hospitals are very simple. Interns are assumed to have preferences only about hospitals and not to care about who their fellow interns are. Hospital preferences concerning groups of interns have a property called responsiveness, which means that they could be derived from a simple ranking of interns and are essentially free of compositional effects. Absent these restrictions, the NIMP may not be stable; worse still, stable matchings may not exist. Again, this abstract result has empirical bite. If interns may marry one another and if they want to work near one another (so that an assignment is not acceptable unless it allocated

18

Michael Rothschild and Lawrence J. White

married interns to the same hospital, or at least to hospitals in the same city), then stable allocations may not exist. The medical community noted that defections from NIMP started occurring in large numbers when increasing numbers of interns were married to one another. The matching literature has generally produced results that state that its most powerful and positive conclusions may not apply when people care about whom they are matched with. Those who have studied the college admissions problem have relatively little to say when students care who their classmates are and when colleges explicitly desire some sort of diversity. Roth’s analysis of the intern market had no place for prices. Kelso and Crawford (1982) showed that this was not an inherent limitation of the matching literature. Their very general model shows how competitive prices can be made an important part of the matching process. However, their analysis does not apply if colleges have explicit preferences about the composition of their student body or if students care about the identities of their fellow classmates.’ Roth and Sotomayor (1990) provide a lucid review of this research. Our general question is whether or not the price system will lead to an efficient allocation. We start with a trivial observation. Any allocation can be supported by a price system. A price system is just a listing of the prices that colleges charge students; that is, a price system is given by specifying p ( c , s ) , the price that college c charges student s. If =

A(s) [xs ifif A(s) ~0

= c

# c’

where xs 5 m, then pA(c,s) implements A. This price system may seem strange; yet it is in some respects close to the system that some colleges use. A denial of admission is the same as a price of a.We do not generally think of price systems as being so personalized. However, scholarships determine the net prices that students pay for colleges, and these scholarships depend on a great many personal characteristics. What is perhaps most strange about the price system equation (1) is that it is not competitive; colleges must collude to implement it. 1.3.2 Benefits of Homogeneity It is hard to teach a class when the students differ greatly in ability and background. We might surmise that an efficient allocation of students to colleges would group students of the same ability. Suppose the benefits that a 7. If the preferences reduce to money, then the general results apply. That is, if students care only about how much human capital their school gives them and if they recognize that this is affected by whom they meet in school, everything works. If, however, students care about both money and about who their classmates are, then stable allocations may not exist. This is true even if students can put a price on good fellowship. Similarly, if colleges care about the composition of the student body-a preference for diversity-as well as its efficiency in producing human capital, stable allocations may not exist.

19

The University in the Marketplace

college conferred on its students were an increasing function of average ability of the student body and a decreasing function of the variance of ability. An efficient price system would necessarily group students of similar ability together. It is natural to ask whether a price system can accomplish this. If prices at each college are based on a student’s ability, the signaling models of Spence (1974) and others can be brought to bear. In those models, people differ in some characteristic t . People can purchase differing amounts of a commodity g; here g denotes the amount of the commodity that people buy. The surplus from a t person’s consuming g is W ( g , t ) .However, the person must pay a price p ( g , t ) , so the net benefit that accrues is just w ( g , t ) = W(g,r) - p ( g , t ) . In such a situation, a person of ability t will choose g ( t ) to maximize w(g,t). Under mild conditions on w(g , t ) , g(t) will be an increasing function of t; people with different skills consume different amounts of g. In the original signaling literature, g was taken to be years of schooling, but the structure of the argument does not depend on this interpretation. It is easy to devise a price system that will segregate people of different ability levels. Two problems with such a price system must be mentioned. First, people differ in many characteristics, perhaps most importantly in liquidity or wealth. If capital markets are imperfect, then potential students may not be able to purchase the education that maximizes the present discounted value of future consumption. Second, if the benefit that a college education confers on its graduates depends on the mix of abilities of those graduates and if each college sets p ( c , s ) to compete for students, then the structure of the model is the same as the model of competition among insurance companies for customers that Rothschild and Stiglitz (1976) and Wilson (1977) have analyzed. Such markets may lack equilibria-or at least the most obvious kinds of equilibria do not exist.

1.3.3 Benefits of Diversity It is sometimes argued that a diverse student body is desirable. A competitive price system will achieve diversity only with difficulty or by accident. The prices that companies charge can in some cases depend on the observable characteristics of customers. They cannot depend simply on the identification of customers. However, without such prices it is not possible to achieve diversity. Consider a very simple model. Suppose that four people are to be allocated among two colleges. Each college has a capacity of two students. We model the desire for diversity by presuming that students are risk averse and that the students of a given college share equally in the college’s output, which is simply the sum of the random inputs of its students. There are two kinds of students. The input of students x , and x2 is the random variable 2, and the input of students y, and y2 is the random variable j ; f and 9 are independent, identically distributed random variables.

20

Michael Rothschild and Lawrence J. White

Clearly, the optimal allocation sends one xi and one y i to each college. An anonymous price system-that is, one that ignores a person’s observable characteristics-could not accomplish this result. This is clearly a smallnumbers problem; if a college admits a large number of students, then the law of large numbers indicates that each college should be able to achieve approximately the right mix of students. Similarly, an insurance company expects that its customers’ risks are uncorrelated. Since the law of large numbers does require large numbers to work, colleges may feel that using a competitive price system (one in which supply equals demand) would leave them with less control over the composition of the student body than they would like. Do colleges use the excess demand, which their less-than-market clearing prices generate, to make efficient allocations? We do not know; and given the difficulty of assessing the effects of matching in our economy,* we doubt that it is easily knowable. Still, it is important to understand the weakness in the a priori argument that competition will allocate students to colleges efficiently.

1.4 Markets and Competition In this section, we address directly the question concerning the markets within which universities operate and the nature of the competition among universities. More often than not, as will be clear, we can only offer insights and raise questions and puzzles.

1.4.1 The Nature of the Enterprise The standard economic model of anything is to assume it maximizes something subject to a resource constraint. This paradigm is hard to apply to higher education because it is difficult to state what is being maximized or what the resource constraint is. It is unclear who “the university” is, so it is not obvious who (or what) is doing the maximizing. This makes it difficult to state what is being maximized. The theory of the firm (in the absence of complete markets or perfect certainty) faces the same difficulty, but it is not difficult (in principle) to describe the different interests and prerogatives of the important actors (management, shareholders, and employees). The goals of some members of the university community (faculty and students) are perhaps not too difficult to model, but the motivations of others (in particular, senior administrators, regents, and trustees) resist easy characterization. It is even harder to specify the prerogatives and bargaining power of the different constituents of the university. Faculty like to say (and to hear administrators say) that the faculty is the university. However, faculty often disagree among themselves. Biologists and historians may have very different 8. See Hartigan and Wigdor (1989, chap. 9) for a discussion of this issue in the context of job matching.

21

The University in the Marketplace

views of the nature of the university and its goals and problems. Administrators and trustees make important decisions about how the university is to run (and who is going to run it). Although institutions of higher education do face resource constraints (and as we note below, some actually go bankrupt and leave the business), it is (as we observed above) hard to state this budget constraint very easily. Two important simple observations are that almost all institutions of higher education are nonprofit organizations9 and that most rely significantly on other resources of revenue (e.g., governmental appropriations, alumni and corporate donations, research contracts and grants) to supplement tuition. lo There are immediate implications: (a) the standard paradigm of profitmaximizing behavior as a motive for pricing, output, and/or entry decisions has only limited explanatory power; (b) the survivorship paradigm (Alchian 1950; Winter 1971; Nelson, Winter, and Schuette 1976), as a backstop to profit maximization, loses much of its force in explaining these decisions, since nontuition contributors’ goals will be important in determining a university’s survival. In short, market pressures impose less discipline on the university than they do on the firm. Senior administrators, or more generally the decision processes of the university, operate under conditions of considerable slack, This freedom leaves the university room to live its version of the quiet life or to pursue the funds (and thus necessarily the goals) of nontuition contributors to the university. The absence of profit-maximizing enterprises among universities is worthy of further consideration. Why should this be so? A simple claim that there are substantial asymmetric information (agent-principal) problems surrounding the instructor-student relationship-which might make student “customers” suspicious of the motives of the instructors in a profit-seeking enterprise-is not sufficient by itself. I t Our society tolerates and supports profit-seeking trade schools, law firms, and medical practices, where agent-principal problems are substantial. The hospital sector has a mix of private nonprofit, religious, and government-operated enterprises (as is true of universities); but the hospital sector also includes for-profit enterprises. A better explanation than information asymmetry is the absence of good (human) capital markets. For most people, higher education is a good investment; it would remain a good 9. For 1985-86, only 220 (6.6 percent) out of 3,340 institutions of higher learning listed by the

U.S.Department of Education were in the category of “organized as profit making” (U.S. Department of Education 1991, 229). Of the 220, over 86 percent (190) offered a program that extended for less than four years. It appears that a large fraction of this “for-profit’’ group was trade and technical schools (ibid., 228). 10. For public universities in 1986-87, tuition accounted for 14.7 percent of total current-fund revenues, and sales and services accounted for another 21.2 percent, leaving 64.1 percent to be covered from nonfee sources. For private universities, tuition accounted for 39.6 percent of revenues, and sales and services accounted for another 21.7 percent, leaving 38.7 percent to be covered from nonfee sources (ibid., 295-96). 11. A related, and more insidious, possibility is raised by Spencer (1991a. 1991b), who claims that college accreditation bodies are hostile toward for-profit educational enterprises.

22

Michael Rothschild and Lawrence J. White

investment even if tuition were set equal to cost.I2 However, most young people cannot pay the full cost of an education; they cannot borrow the funds, since they have no collateral. An interesting consequence of this shortfall between tuition and the costs of education is an attenuation of the ability of students (as customers) to influence the ways in which universities behave.

1.4.2 Inputs We start with input markets,I3 primarily because the analysis seems clearest there. With the exception of the teacher (professor) inputs, universities are just one among many input users, and the markets are basically competitive. Further, with respect to professor inputs, universities clearly do compete among themselves to fill positions. The individual university demand curves for professors, though, warrant some further consideration. Those demand curves are, arguably, derived demand curves-derived from the demand for the university’s outputs. To some extent those demand curves do reflect the nature of universities’ outputs: for example, teaching colleges are more likely to look for good teachers; research universities are more likely to demand productive researchers. Still, research universities “sell” large amounts of undergraduate education; the marginal revenue product of outstanding teachers would seem to be quite high. Why do good teachers command such small monetary (and other) rewards in large research universities? Why have research universities been so reluctant to establish job categories for outstanding teachers? Why has competition not operated in this dimension? On this last point, we note that professional schools have been more responsive with respect to teaching. Many schools, even those that pride themselves as research institutions, have established “clinical professor” positions that often emphasize teaching or other nonresearch contributions of a faculty member. We suspect a reason for this is that in some professional schools, particularly law and business schools, a high proportion of gross receipts comes in the form of tuition and deferred tuition (alumni gifts). For such schools, student satisfaction impinges more immediately on the school’s budget constraint. (This argument, however, cannot explain the existence of clinical professors in medical schools.) Finally, our casual impression is that university teaching has been resistant to technological change. Why is this so? Surely it is not the case in the age of the computer and the VCR that the technology of teaching is inherently incapable of significant technological improvement. It is also our casual impression that the education that takes place outside the higher education industry 12. Human capital is even more of a bargain than it usually appears, if one assumes that education not only improves productivity on the job (an effect that shows up in wages) but also increases the ability to use leisure time. See Jorgenson and Fraumeni (1991) for some astonishing calculations. 13. We exclude the analysis of incoming students as inputs, which was covered in section I .3.

23

The University in the Marketplace

(businesses and the military, for example) has embraced technical change more rapidly than have colleges and universities.l4 1.4.3 outputs Universities are clearly multiproduct enterprises that operate in many markets. Among their outputs are educational services for undergraduate students, educational services for graduate students (arts and sciences, as well as professional education), research, room and board services for resident students, and athletic entertainment services. We will focus primarily on the market for educational services. Do universities compete with each other in the market for educational services? Casual empiricism suggests that they do compete for students. The terms of this competition include the quality (somehow measured) of the university, the quality of the student body that the university attracts, the location and physical surroundings of the university, and the price (tuition) charged. To support our claim that universities do compete on price, we offer the following: University deans (including heads of programs and heads of professional schools) do pay attention to the tuition levels of universities of similar quality and/or in similar locations and are concerned that their own tuition levels not diverge appreciably from those of their rivals. Further, the U.S. Department of Justice’s Antitrust Division recently investigated alleged meetings by administrators from at least 23 prestigious East Coast colleges and universities, who met annually to agree on the scholarship levels that would be offered to prospective freshmen (Jaschik 1989; Putka 1989; Salop and White 1991).Is One participant apparently feared that without these meetings, the universities “might be dragged into a kind of ‘bidding war’ for the best students” (Cotter 1989). It is interesting to note that another 33 universities that were under investigation for sharing information on scholarship aid included the Great Lakes Colleges Association (a group of 12 liberal arts colleges in Indiana, Michigan, and Ohio), a group of 8 women’s colleges (6 of which are located in the South), and an additional group of 12 private universities that had very high tuition fees (Jaschik 1989). Within each of these three groups, the schools would likely have perceived one another as direct competitors and would have been interested in restraining price competition. This evidence is, at best, only indirect support for the claim that price competition among universities is a significant phenomenon. It is supported, however, by many of the studies of student enrollment choices among universities. 14. For brief discussions of efforts to provide higher education that is more responsive to “customers’” demands, see Spencer (1991a, 1991b) and Charlier (1991). 15. In May 1991 the Justice Department formally charged eight Ivy League schools and Massachusetts Institute of Technology with price fixing. The eight Ivies immediately settled the case with a consent decree (in which they did not admit any guilt but agreed to discontinue the meetings), but MIT declined to join the settlement and was subsequently found guilty at trial.

24

Michael Rothschild and Lawrence J. White

These studies often include as explanatory variables the tuition (and other fees) and scholarship amounts of both the selected university and those that were rejected. The coefficients on the tuition and scholarship amounts (or, in some studies, the net cost) offered by the rejected schools are usually significant and have the expected signs (e.g., the coefficient on the tuition level of a rejected university has a positive sign) (Radner and Miller 1975; Miller and Radner 1975; Hight 1976; Fuller, Manski, and Wise 1982; Corman 1983; Manski and Wise 1983; Ehrenberg and Sherman 1984). Thus, students do seem to be sensitive to the prices of the alternatives open to them. (Unfortunately, we have not been able to uncover studies that examine the price crosselasticity of demand among specific universities, which would provide us with a greater understanding of the specific nature of the competition among the universities.)16 It seems unlikely that price competition among universities approaches the textbook model of the perfect competition among wheat farmers. Individual universities have perceived quality differences and ‘brand-name’’ reputations that surely influence student choice. Also, locational differences among universities imply transportation cost differences (as well as psychic “away from home” differences, which can be a plus or a minus for a university’s attraction) for many students. l7 Competition among universities appears to have both geographic-space and product-space dimensions. High-prestige schools probably compete in a nationwide market. For example, in the market for freshman applicants, Har16. These investigations would require time-series cross-section panels that would either use individual university applications as the dependent variables (and include university tuition levels as a right-hand side variable) or use individual student applications and acceptance choices as the dependent variables. Spies (1990) has studied how family income affects the probability of applying to an expensive and selective private college or university. Spies found that the relationship between the probability of applying and income had a gentler slope for those who applied for financial aid than for those who did not. Without criticizing Spies’s work (which is careful and involves the replication of the basic results over three different cohorts of applicants), we note that he did not pose his question (what determines the probability of applying to a particular kind of college?) as that of estimating a demand function. Price (gross or net tuition) is not included as a variable. McPherson and Winston (1991) develop a model in which universities compete but in which information asymmetries between sellers (universities) and buyers (students) cause the terms of competition to focus on costly symbols of quality and also cause buyers to judge quality on the basis of price (tuition); this latter effect would deter the sellers from cutting prices in order to compete and could even impart a price-raising bias to their behavior. McPherson and Winston offer no evidence to support their model. We believe that the evidence from the enrollment choice studies cited in the text, in which the coefficients on the tuition levels of rejected universities have positive signs, casts doubt on the validity of the McPherson-Winston hypothesis. 17. In fall 1988 over 80 percent of freshmen enrolled in a college or university in the same state in which they had previously resided (U.S. Department of Education 1991, 196); this percentage has been remarkably stable over the past two decades (Harris 1972). In most of the demand studies, distance from home is a negative factor in a potential student’s choice; see Hoenack and Weiler (1976); Fuller, Manski, and Wise (1982); Manski and Wise (1983); Ehrenberg and Sherman (1984); and McClain, Vance, and Wood (1984). 18. Garvin (1980, chap. 2) makes some of these same arguments.

25

The University in the Marketplace

vard and Stanford probably compete for roughly the same pool of students (and probably also compete for a common pool of applicants to their medical, business, and law schools and to most of their graduate programs in arts and sciences). Schools with lesser prestige are likely to compete among themselves on a regional basis; the lure of a specific national “brand name” is likely to be less important for students in this market segment, and the costs associated with regional location are likely to loom relatively larger. Finally, universities are likely to compete most intensively with universities in their same quality segment. For instance, Yale and Harvard are likely to consider each other as competitors, while neither is likely to think of the University of Bridgeport as a competitor. We can now discuss a number of important topics related to competition among universities.

Tuition Levels and Scholarship Levels Suppose a university charges a tuition level of X to all its i students and offers a vector of scholarships Y to those same i students (05 Y, 5 X). As a first approximation, if the university instead charged a tuition of X $10,000 and offered a new vector of scholarships of Y + $10,000, nothing should change;I9 if the university-because it asks for family financial information from all its applicants-could selectively offer scholarship increments that were less than $lO,OOO to some students and still not lose those students, then the university’s net revenues would increase.2o In principle, the university’s net revenues would continue to increase as it raised tuition levels and selectively increased scholarship amounts until all but one of its students were on partial or complete scholarship; in essence, the university would be practicing first-degree price discrimination. Universities clearly do engage in price discrimination to some extent. Scholarship aid (including Pel1 Grants) amounted to 24 percent of aggregate tuition receipts by private universities and to 35 percent by public universities in 1986-87 (U.S. Department of Education 1991, 291-92). Still, one can ask why universities do not engage in more of it and why they do not make a greater effort to achieve the first-degree price discrimination ideal described in the previous paragraph. There are a number of possible answers to this question, but one of them, we believe, can immediately be discarded. It might be claimed that students would somehow perceive tuition increases matched by identical scholarship funding increases as not being neutral and that they would thereby be deterred

+

19. This is equivalent to an auto dealer’s adding $10,000 to all list prices but also offering $IO,OOO “discounts.” 20. We abstract from any added administrative costs. Also, it is worth noting that the auto dealer would be unlikely to succeed with a similar price discrimination scheme, because of competition among auto dealers and because auto dealers typically do not know a prospective buyer’s income or other characteristics (though the dealer may learn them after the sale, while arranging for the financing of a purchase).

26

Michael Rothschild and Lawrence J. White

by the tuition increase (Hearn and Longanecker 1985). The available evidence, however, points strongly toward our equivalence hypothesis. Studies of student enrollment choices among types of universities sometimes include both tuition levels and scholarship amounts (offered by the chosen and rejected universities) as explanatory variables. These studies show that tuition levels and scholarship amounts have virtually identical coefficients (with opposite signs) in explaining student enrollment choices (Fuller, Manski, and Wise 1982; Manski and Wise 1983; Ehrenberg and Sherman 1984). Thus, students who are offered scholarship aid do not seem to suffer from “tuition illusion,” and claims of nonneutrality are unlikely to be adequate explanations for why universities do not practice price discrimination to a greater extent. We are left with two possible answers to this question. The first is that price competition among universities would undercut and unravel this extreme form of price discrimination. The second is that the nontuition funds providers would be offended by this apparent gouging by the university (i.e., the increases would not be neutral from their perspective), and their contributions would decrease, thereby reducing (or eliminating) the net revenue gain to the university from the price discrimination scheme. Among the most important contributors are future alumni, whose generosity toward their alma mater could possibly be severely tempered by the memory that she had charged all that the traffic would bear. We currently do not have enough information about price competition among universities or about the behavior of nontuition funds providers to assess the relative importance of these explanations. Scholarships and Price Competition

In section 1.3 we suggested that a price discrimination scheme (i.e., selective scholarships) could allow the university to achieve a desired mix of students, which would enhance the efficiency and productivity of the university’s educational output. Is this form of price discrimination compatible with competition among universities? Or does the university’s desire for an optimal mix create a potential market failure that would argue for limits on competition and that could justify the alleged agreements on scholarship levels that the Justice Department investigated? The case for a market failure does not appear to be strong. The externality of the “desirable” students is wholly internalized within the university. If, say, a “desirable” student enhances the educational experience of other students, then those other students should be willing to pay higher tuition to a university that offers this diversity; the externality is internalized. Though competition for desirable students, through larger price discounts (i.e., larger scholarships), reduces university net revenues, this is true of competition for all of the university’s outputs.*‘ Further, the experience of the past decade in the 21. We see only one special problem that suggests special treatment for this industry. If, as we argued in the last part of section 1.3, diversity is a small-numbers problem, then coordination among universities in allocating students may be desirable.

27

The University in the Marketplace

airline industry suggests that modest levels of price discrimination can survive in markets that are workably competitive. Pricing within the University

Casual empiricism suggests that the marginal costs of educating an undergraduate in the sciences are substantially higher than the marginal costs of educating an undergraduate in the humanities. Nevertheless, we generally see uniform tuition levels within a university across most majors (though different schools or programs within a university may charge modest fee differentials). Why is this so? We have already (in section 1.2) dealt with the normative issue of whether such uniform pricing generates cross-subsidies among areas. There is still the positive question of why this uniformity occurs and persists. In a multioutput (profit-maximizing) enterprise with common costs (economies of scope) and with differing marginal costs among the separate outputs, pricing is a complex phenomenon. A monopolist will look to the demand elasticities of its separate products, as well as their marginal costs, to determine its prices. A firm in competitive markets will seek a combination of prices and products that yields an aggregate surplus over its separate marginal costs that is adequate to cover its common costs. Though neither market structure necessarily generates an outcome in which the firm’s prices correlate positively with its marginal costs, uniformity of prices for outputs with substantially different marginal costs would occur purely by chance (and would be highly unlikely to replicate itself in thousands of separate enterprises).22And with marginal costs as the starting point for pricing under either form of market structure, there is a mild presumption that a positive correlation between prices and marginal costs should emerge.23 At first glance, then, tuition uniformity seems to be an oddity that is inconsistent with profit-maximizing behavior in any market structure. One explanation might be as follows: Many undergraduate institutions do not charge per course or per credit but rather per semester or quarter. In principle all students can take all courses (or could if they so planned their programs). What is being sold is the ability to pick from a menu, and this is no more strange than the observation that many salad bars charge per trip rather than per nutrient. The salad bar analogy is strongest, however, where monitoring costs are high relative to the price of the items. This does not seem to be the case for student course enrollments. An alternative model would be that of two-part 22. Where marginal cost differences are small and the transactions costs of enforcing marginal cost pricing are high, we are likely to see uniform pricing. For example, restaurants typically charge a uniform price for coffee, regardless of whether a customer adds cream and/or sugar. On the other hand, delicatessens often charge extra for extra materials that can be ordered with a sandwich (e.g., lettuce and/or tomato), presumably because the marginal costs are higher and the monitoring costs are small. 23. Restaurants generally charge higher prices for their steaks than for their hamburgers and higher prices for their strawberry shortcake than for their donuts.

28

Michael Rothschild and Lawrence J. White

tariffs, in which customers are charged a lump-sum entry fee and are then charged prices for individual services that approximate marginal costs (Oi 1971). In this framework, then, we would expect to see all students pay a common enrollment fee (subject to the price discrimination possibilities discussed above) and then be charged specific course fees that were roughly commensurate with the marginal costs of those courses. We are thus left with the puzzle of uniform or near-uniform tuition levels in the presence of substantial marginal cost levels. Perhaps this is another area where the preferences and prejudices of nontuition funds providers are important. Again, we believe that this is an area that warrants further research. Pricing and Prestige

Mercedes automobiles sell for appreciably more than Chevrolets; Rolex watches sell for appreciably more than Timexes. But even among private universities, high-prestige institutions often do not charge tuition levels substantially above those of lower-prestige institutions. Why is this so? Why do high-prestige institutions decline to try to capture most of the rents that are associated with their “brand names”? A recent survey of graduate professional schools provides striking evidence to support this picture of relative uni f~rmi ty.~~ In tables 1.1 and 1.2 we present the tuition levels and expected starting salaries for graduates of top-ranked business schools and law schools. If we focus on the private universities in the we find a picture of relative uniformity of tuitions among the leading schools. There is a mild positive correlation between a school’s tuition and its rank: for business schools the rank correlation is 0.58; for law schools it is 0.46. When we look at the correlation between tuition and expected annual starting salaries, there are again positive rank correlations: 0.56 for business schools and 0.71 for law schools. Simple ordinary least squares regressions of tuition levels (TL) on expected salaries (ES), however, yield the following (with t-statistics in parentheses):

+

Business schools: TL = 11.60 (6.35) Law schools: TL = 9.05 + (5.01)

0.085 ES; r = 0.55; n (2.48) 0.095 ES; r = 0.69; n (3.48)

=

16 .

=

15 .

These results indicate that students at business and law schools where expected starting salaries are higher do pay higher tuitions, but those higher annual tuitions are less than 10 percent of the higher expected annual starting salary.26 24. We do not have any immediate evidence concerning undergraduate institutions, but we are reasonably confident that a similar picture would emerge. 25. State universities, with the exception of the University of Michigan, are charging tuitionseven to out-of-state students-that have more to do with state legislatures’ policies than with any notions of market pricing. 26. These results are consistent with those found by Ehrenberg (1989).

29

The University in the Marketplace

Table 1.1

RanWSchool

Rankings of Leading Business Schools I990 Out-of-State Tuitiona

Average Starting Salary'

$16.4 16.6 16.5 16.6 17.2 16.7 16.2 16.5 11.7 15.7 16.3 16. I 16.5 5.6 7.8 8.1 3.6 8.2 15.5 16.8 14.4 16.9 14.5 7.1 14.7

$63.0 60.5 55.0 54.0 59.0 54.5 51.0 57.0 55.3 53.3 52.0 50.7 52.0 50.8 50.0 51.5 44.0 44.1 53.2 43.5 49.1 43.5 45.2 42.9 44.5

Rank, Excluding State Universities Overall

'hition

Salary

~

1. Harvard 2. Stanford 3. Penn 4. Northwestern 5. MIT 6. Chicago 7. Duke 8. Dartmouth 9. Virginia 10. Michigan 11. Columbia 12. Cornell 13. Camegie 14. N Carolina 15. UC Berkeley 16. UCLA 17. Texas 18. Indiana 19. NYU 20. Purdue 21. USC 22. Pittsburgh 23. Georgetown 24. Maryland 25. Rochester

9 10

11 -

9 4.5 7 4.5 1 3 11 7 10 12 7 -

10 12 9

-

12 13 14 15 16

13 16 2 15 14

8 13 16 14

15

Source: U S . News & World Report, April 29, 1991, p. 68. "In thousands.

Finally, the more limited data in table 1.3, for medical schools, show even less correlation (rank correlation = 0.09) between rank and tuition than for the business and law schools. Again we have a puzzle. The students, rather than the schools, are capturing the rents.27Even if schools provide only signals (Spence 1974) or filters, is the filter worth this little? Are the preferences of nontuition funds providers important here? Again, we suggest that this is a fruitful area for future research. 27. It has been suggested to us that the higher starting salaries offered to the graduates of the leading law and business schools may be just a cost-of-living compensation adjustment; that is, the leading professional schools tend to be located in metropolitan areas with above-average living costs and their graduates tend to work in these same pricey areas. If this were so, the students' net rents would be much smaller than the gross differentials in starting salaries indicate. Our casual impression from the cost-of-living comparison data gathered by Kramer (1989) for law school graduates is that the net rents accruing to the graduates of leading professional schools are still substantially positive. Without more complete data on the location choices of the graduates of the leading and lesser schools and of the cost-of-living differentials among these locations, however, we are unable to pursue this net rent hypothesis any further.

30

Michael Rothschild and Lawrence J. White

Table 1.2

RanWSchool 1. Yale 2. Harvard 3. Chicago 4. Stanford 5. Columbia 6. Michigan 7. NYU 8. Virginia 9. Duke 10. Penn 11. Georgetown 12. UC Berkeley 13. Cornell 14. Northwestern 15. Texas 16. USC 17. Vanderbilt 18. UCLA 19. Iowa 29. UC Hastings 21. Wisconsin 22. G Washington 23. Minnesota 24. Notre Dame 25. N Carolina

Rankings of Leading Law Schools 1990 Out-of-State Tuition‘

Average Starting salarya

$15.4 14.5 15.7 14.9 16.1 15.7 16.6 10.1 15.3 15.1 15.4 8.8 15.9 15.5 6.0 16.4 14.8 9.0 7.7 8.7 9.1 15.2 8.7 13.0 7.0

$66.1 67.2 71.0 65.0 78.3 59.6 76.7 63.0 60.2 64.6 66.0 58.0 66.2 65. I 52.6 66.7

55.0 62.7 50.0 62.7 41.5 61 .O 45.7 56.9 40.5

Rank, Excluding State Universities Overall 1 2 3 4 5

-

Tuition

Salary

7.5 14 5 12 3

7 4 3 10

-

-

1

6

1

2

-

-

-

9 11 7.5

13 11 8

-

-

-

10 11

4 6

6 9

-

-

-

12 13

2 13

5 15

-

-

-

7 8 9

-

-

-

14

10

12

15

-

-

-

15

-

14

-

Source: US.News & World Report, April 29, 1991, p. 74. thousands.

Entry and Exit

Entry and exit play important roles in the standard competitive model, helping to expand or contract supply and thereby hastening the elimination of short-run rents or losses. Entry can occur de novo (by start-up firms) or through “product extensions” by existing firms. Table 1.4 shows the number of two-year and four-year colleges and universities that have been in the market over the past 40 years. There has been substantial growth in these numbers; that is, net entry has been considerable. (It should be noted that over time some two-year schools have converted to four-year schools and some schools in both categories have exited the market entirely, so gross entry in all categories has been larger than any net calculation would indicate.) Table 1.5, covering professional schools, tells the same story of substantial net entry. What motivated these entry decisions? It is clear that the expanding population and rising incomes of the U.S. economy created an increased demand for university education in the United States; the rising international repu-

31

The University in the Marketplace Rankings of Leading Medical Schools

lsble 1.3

Rank, Excluding State Universities

1990

Out-of-State Tuitiona

RanWSchool

$18.0 16.5 14.2 5.9 17.0 14.9 18.3 17.9 8.0 19.2 20.4 11.9 12.5 16.1 14.6

1. Harvard 2. Johns Hopkins 3. Duke 4. UC San Francisco 5. Yale 6. Washington University 7. Penn 8. Stanford 9. UCLA 10. Cornell 11. Michigan 12. Columbia 13. U Washington 14. Chicago 15. Vanderhilt

Overall

Tuition

1 2 3

3 7 11

-

-

4 5.5 5.5 7

6 9 2 4

-

-

8

1

-

-

9

5

-

-

10 11

10

8

Source: U.S. News & World Report, April 29, 1991, p. 68.

'In thousands.

Number of Institutions of Higher Education

Table 1.4

Excluding Branch Campuses Publicly Controlled

1949-50 1954-55 195940 1964-65 1969-70 1974-75 1979-80 1984-85 1989-9@

Privately Controlled

Including Branch Campuses Publicly Controlled

Privately Controlled

4-year

2-year

4-year

2-year

4-year

2-year

4-year

2-year

344 353 367 393 426 447 464 46 I

297 295 328 406 634 767 846 868

983 980 1,055 1,128 1,213 1,297 1,399 1,450

221 22 1 254 248 252 236 266 367b

n.a. n.a. n.a. n.a. n.a.

n.a. n.a. n.a. n.a. n.a.

n.a. n.a. n.a. n.a. n.a.

n.a. n.a. n.a. n.a. n.a.

n.a.

n.a.

n.a.

537 549 566 595

896 926 935 968

1,329 1,408 1,459 1,532

242 269 37Ib 440

n.a.

Source: U.S. Department of Education (1991), 228.

'Data for this year are not entirely comparable with earlier years because of revised survey procedures. bLarge increases are due to the inclusion of trade and technical schools.

tation of U.S. universities also added to demand. Total student enrollment (the intersection of demand and supply) rose from 2.3 million in 1947 to 13.0 millionzs in 1988. Still, this increase in output might have been accom28. This includes part-time students.

32

Michael Rothschild and Lawrence J. White

Table 1.5

Number of Institutions Conferring Professional Degrees Year

1949-50 1959-60 1969-70 1974-75 1979-80 1984-85 1987-88

Dentistry

Medicine

Law

40 45 48 52 58 59 55

72 79 86 I04 I12 I20 I20

n.a. 134 145 154 I79

181 I80

Source: U.S. Department of Education (1991),248

modated solely through internal expansion of the 1,85 1 institutions that existed in 1949-50. Why did entry occur alongside internal e x p a n ~ i o n ?Even ~~ if we exclude the growth in the number of publicly controlled institutions (the causes of which might be harder to model), there were still increases of over 50 percent in the numbers of two-year30 and four-year privately controlled institutions. Why did this entry occur? We would guess that the availability of private donations and endowments to provide the start-up capital for new private institutions (the equivalent of the owners’ initial investments in any forprofit enterprise) was often an instrumental factor, but there were surely other factors as well. Research on university entry behavior (including “product extensions”-new programs or schools begun by existing universities) would appear to be worthwhile. One other feature of table 1.4 is worthy of notice: the data indicate that publicly controlled universities are much more likely to establish branch campuses than are privately controlled universities. It is unclear to us why these private institutions believe that their brand names cannot be extended to multiple 10cations.~’This too appears to be an area that warrants research. Finally, table 1.6 shows the number of colleges and universities that have shut their doors in the past three decades-that is, they have exited the education market.32The exit decision by for-profit firms in the private sector is 29.Enrollments at publicly controlled universities expanded by over 780 percent between 1947 and 1988,while enrollments in privately controlled institutions expanded by over 240 percent. Both of these expansions greatly exceeded the percentage increases in the numbers of institutions, so internal expansion clearly did accompany entry. 30. Some of the increase occurred through entry by for-profit trade and technical schools. 3 1. State chartering restrictions appear to prevent universities from branching across state lines (much as is true for commercial banks). But the near-absence of intrastate branching by private universities remains a puzzle. Why does the University of California have eight branch locations, while Stanford only has its “home office”? A few universities have established locations abroad and in Washington, D.C., but these branch locations are usually designed for special programs of their students based at the home campuses, rather than as freestanding (full-service) branches. 32. In some instances, private universities have in essence exited, but they have been superseded by public institutions.

33

The University in the Marketplace

Table 1.6

Number of Institutions of Higher Education that Have Closed Their Doors Publicly Controlled 4-year

Total, excluding branch campuses, 1960-61 to 1989-90 Total, including branch campuses, 1969-70 to 1989-90

2-year

Pnvately Controlled 4-year

2-year

1

37

167

118

4

29

152

90

Source: U.S. Department of Education (1991), 231.

not a well-researched area, so we have even less here to serve as a basis for explaining university behavior. Again, research would be worthwhile.)’

1.4.4 Positioning in the Market How do universities position themselves in the market? Why do Harvard, Northeastern, Antioch, and Grinnell attract the specific groups of students that they do? How can they change their positioning (e.g., improve their perceived quality and prestige)? How often (and why) do universities attempt to change their positioning? When (and why) do they succeed (or fail)? As was true for entry, we suspect that availability of private and public contributions and endowments are important (this especially seems to be true for professional schools in the past two decades). Still, further research could surely shed useful light here.

1.4.5 What about a Monopoly Model? As noted earlier, the autonomous cost increase model advocated by Massy (1989) assumes that most (if not all) universities are separate monopolies that face inelastic demands and thus can raise their prices at will to accommodate rising We believe that the empirical evidence, scanty though it may be, throws substantial doubt on this basis for Massy’s analysis. Still, let us suppose that universities truly were monopolies. The theory of monopoly, of course, yields a prediction about the level of prices of a monopoly relative to those of an otherwise similar competitive industry. It says nothing about mtes of price increases. If universities really were separate monop-

33. It has been suggested to us that the cloudy property rights that accompany the nonprofit status of private universities may impede their ability to shut their doors and liquidate assets. 34. As we noted in footnote 16 above, McPherson and Winston (1991) offer an alternative model that might explain a pattern of secular cost increases: asymmetric information problems cause universities to compete through costly symbols of quality. As we explained there, however, we believe that the available evidence casts serious doubt on the McPherson-Winston hypothesis.

34

Michael Rothschild and Lawrence J. White

olies and could raise their prices at will, then the important question would be: Why have universities not raised their tuition earlier and faster? We find it hard to believe that over 3,000 monopoly university administrators, year after year, would have consistently passed up opportunities to increase revenues substantially by raising tuition. Though it is possible that perceptions of gouging by nontuition funds providers might have stayed the tuition-raising hands of some university administrators during some periods, we doubt that the gouging perceptions could have been a complete restraint at all times. Could it be that universities are already pricing at monopoly levels and that it is these elevated prices that generate substantial cross-elasticities of demand and thus bring the universities into competition with each other? If this proposition were true, it would mean that universities’ prices are currently generating explicit or implicit rents and that there is a lower set of pices that would eliminate the rents and at which there would be low or zero demand crosselasticities among the u n i v e r ~ i t i e s . ~ ~ The proper test of this proposition would require the measurement of universities’ rents at current prices. Since universities currently charge tuition and other fees that cover only a fraction of their costs and since universities’ input prices are largely determined in competitive markets, the existence of explicit rents seems unlikely. Also, as we noted above, it appears that many high-prestige universities are not even exploiting the rents associated with their brand names. It is possible that universities are absorbing potential explicit rents in the form of production inefficiencies-Leibenstein’s (1966) X-inefficiency. With the presence of over 3,000 universities in the market, we consider it unlikely that X-inefficiency would uniformly hide the rents that would otherwise be accruing to these monopolies. Still, in the absence of a comparison model of an X-efficient university, we must remain somewhat agnostic on this point.

1.5 Conclusion The analysis of university behavior in a market context has been an underresearched area in economics. In this paper we have argued that a competitive framework for analysis appears reasonable but that the nonprofit status of universities and the major role of nontuition funds providers introduce special 35. For antitrust purposes, this is the proper test of a monopoly. In a major antitrust case that tried to determine whether Du Pont had monopoly power in the sale of cellophane (US.v. E . I . DuPont de Nemours and Co., 351 U.S. 377 [I956J)the U.S. Supreme Court made the mistake of looking only at the cross-elasticities of the demand at the prevailing prices for cellophane and not asking about the rents that were accruing and about what the cross-elasticities and rents might have been at lower prices. As many commentators noted, if Du Pont did have a monopoly in cellophane, profit-maximizing behavior would call for the company to raise its price to the point where significant cross-elasticities with other flexible wrapping materials would have developed (Stocking and Mueller 1955; Posner 1976, 127-128).

35

The University in the Marketplace

features into any competitive structure. We have offered some insights into university behavior and raised a number of interesting questions and puzzles. We suggest that these questions and puzzles provide a rich agenda for future research that will help us better understand market behavior in this important sector of the U.S. economy.

References Alchian, A m e n A. 1950. Uncertainty, evolution, and economic theory. Journal of Political Economy 58:211-21. Becker, William B. 1990. The demand for higher education. In The economics of American universities, ed. Stephen A. Hoenack and Eileen L. Collins, 155-88. Albany: State University of New York Press. Bok, Derek. 1990. Universities and the future of America. Durham, N.C.: Duke University Press. Bowen, Howard R. 1980. The costs of higher education. San Francisco: Jossey-Bass. Charlier, Marj. 1991. First principles: Ailing college treats student as customer, soon is thriving. The Wall Street Journal, July 17, A l . Cohn, Elchanan, Sherrie L. W. Rhine, and Maria C. Santos. 1989. Institutions of higher education as multi-product firms: Economies of scale. Review of Economics and Statistics 7 1:284-90. Corman, Hope. 1983. Postsecondary education enrollment responses by recent high school graduates and older adults. Journal of Human Resources 17:247-67. Cotter, William R. 1989. Colleges’ efforts to rationalize the financial-aid system should not be treated as violations of antitrust laws. Chronicle of Higher Education, October 4, B 1. Ehrenberg, Ronald G. 1989. An economic analysis of the market for law school students. Journal of Legal Education 39:627-54. Ehrenberg, Ronald G., and Daniel R. Sherman. 1984. Optimal financial aid policies for a selective university. Journal of Human Resources 19:202-30. Faulhaber, Gerald R. 1975. Cross-subsidization: Pricing in public enterprise. American Economic Review 65:966-77. Fuller, Winship C., Charles F. Manski, and David A. Wise. 1982. New evidence on the economic determinants of postsecondary schooling choices. Journal of Human Resources 17:477-98. Gale, David, and Lloyd S. Shapley. 1962. College admissions and the stability of marriage. American Mathematical Monthly 69:9-15. Garvin, David A. 1980. The economics of university behavior. New York: Academic Press. Hansen, W. Lee, and Burton A. Weisbrod. 1969. Benefits, costs, andjnance ofpublic higher education. Chicago: Markham. Harris, Seymour E. 1972. A statistical portrait of higher education. New York: McGraw-Hill. Hartigan, John A., and Alexandra K. Wigdor, eds. 1989. Fairness in employment testing: Validity generalization, minority issues, and the general aptitude test battery. Washington, D.C.: National Academy Press. Hearn, James C., and David Longanecker. 1985. Enrollment effects of alternative postsecondary pricing policies. Journal of Higher Education 56:485-508.

36

Michael Rothschild and Lawrence J. White

Hight, Joseph E. 1976. The demand for higher education in the U.S., 1927-72: The public and private institutions. Journal of Human Resources 1 0 512-19. Hoenack, Stephen A,, and Eileen L. Collins, eds. 1990. The economics of American universities. Albany: State University of New York Press. Hoenack, Stephen A,, and William C. Weiler. 1976. Cost related tuition policies and university enrollments. Journal of Human Resources 10:332-60. James, Estelle. 1978. Product mix and cost disaggregation: A reinterpretation of the economics of higher education. Journal of Human Resources 12:157-86. . 1986. Cross-subsidization in higher education: Does it prevent private choice and public policy? In Private education: Studies in choice and public policy, ed. Daniel Levy, 237-57. New York: Oxford University Press. . 1990. Decision processes and priorities in higher education. In The economics of American universities, ed. Stephen A. Hoenack and Eileen L. Collins, 77-106. Albany: State University of New York Press. James, Estelle, and Egon Neuberger. 1981. The university department as a nonprofit labor cooperative. Public Choice 36585-612. Jaschik, Scott. 1989. Investigation into tuition fixing spreads. Chronicle of Higher Education, October 4 , A 1. Jorgenson, Dale W., and Barbara M. Fraumeni. 1991. The output of the education sector. Harvard Institute of Economic Research Discussion Paper no. 1542. Cambridge, Mass.: Harvard University. Kelso, Alexander S . , Jr., and Vincent P. Crawford. 1982. Job matching, coalition formation, and gross substitutes. Econometrica 50: 1483-1504. Kerr, Clark. 1964. The uses of the university. Cambridge, Mass.: Harvard University Press. Kramer, John R. 1989. Who will pay the piper or leave the check on the table for the other guy? Journal of Legal Education 39:655-95. Leibenstein, Harvey. 1966. Allocative efficiency vs. X-inefficiency. American Economic Review 56(June): 392-415. Leslie, Larry L., and Paul T. Brinkman. 1987. Student price response in higher education. Journal of Higher Education 58: 181-203. Manski, Charles F., and David A. Wise. 1983. College choice in America. Cambridge, Mass.: Harvard University Press. Massy, William F. 1989. A strategy for productivity improvements in college and university academic departments. Stanford, Calif.: Stanford University. McClain, David, Bradley Vance, and Elizabeth Wood. 1984. Understanding and predicting the yield in the MBA admissions process. Research in Higher Education 20:55-76. McPherson, Michael S . , and Gordon C. Winston. 1991. The economics of cost, price, and quality in U.S. higher education. Williams College Working Paper no. DP-13. Williamstown, Mass. April. Miller, Leonard S . , and Roy Radner. 1975. Demand and supply in United States higher education: A technical supplement. Berkeley, Calif. : Camegie Commission on Higher Education. Nelson, Richard R., Sidney Winter, and H. L. Schuette. 1976. Technical changes in an evolutionary economy. Quarterly Journal of Economics 90:90-118. Oi, Walter Y. 1971. A Disneyland dilemma: Two-part tariffs for a Mickey Mouse monopoly. Quarterly Journal of Economics 85:77-96. Posner, Richard A. 1976. Antitrust law: An economic perspective. Chicago: University of Chicago Press. Putka, Gary. 1989. Do colleges collude on financial aid? The Wall Street Journal, May 2, B1.

37

The University in the Marketplace

Radner, Roy, and Leonard S. Miller. 1975. Demand and supply in United States higher education. New York: McGraw-Hill. Rosovsky, Henry. 1990. The university: An owner’s manual. New York: W. W. Norton. Roth, Alvin E. 1984. The evolution of the labor market for medical interns and residents: A case study in game theory. Journal ofPolitica1 Economy 92:991-1016. Roth, Alvin E., and Marilda A. Oliveira Sotomayor. 1990. Two-sided matching: A study in game theoretic modeling and analysis. Cambridge: Cambridge University Press. Rothschild, Michael, and Joseph E. Stiglitz. 1976. Equilibrium in competitive insurance markets: An essay on the economics of imperfect information. Quarterly Journal of Economics 90:629-50. Salop, Steven C., and Lawrence J. White. 1991. Antitrust goes to college. Journal of Economic Perspectives 5: 193-202. Spence, A. Michael. 1974. Market signaling: Information transfer in hiring and related screening processes. Cambridge, Mass: Harvard University Press. Spencer, Leslie. 1991a. College education without the frills. Forbes, May 27,290-94. . 1991b. The perils of socialized higher education. Forbes, May 27,294-304. . 1991c. Good school story. Forbes, May 27,304-6. Spies, Richard R. 1990. The effect of rising costs on college choice: The third in a series of studies on this subject. Princeton, N.J.: Princeton University. Stocking, George W., and Willard F. Mueller. 1955. The cellophane case and the new competition. American Economic Review 4529-63. U.S. Department of Education, Office of Educational Research and Improvement, National Center for Education Statistics. 1991. Digest of education statistics, 1990. Washington D.C. : National Center for Education Statistics. Wilson, Charles, 1977. A model of insurance markets with incomplete information. Journal of Economic Theory 16: 167-207. Winter, Sidney G. 1971. Satisficing, selection, and the innovating remnant. Quarterly Journal of Economics 85:237-61.

Comment

Martin Feldstein

This is an excellent paper, interesting both for the answers that it provides and for the additional questions that it raises but leaves unanswered. It is an important paper because it looks beyond the previous studies of demand and costs to try to understand the structure of the market within which institutions of higher education operate. The authors recognize that almost all colleges and universities are nonprofit institutions and then proceed to ask why in so many cases these institutions do not behave as w e might expect for-profit institutions to behave. Before discussing some of the specific topics raised by Rothschild and White, I will offer my own general point of view on this subject. Martin Feldstein is the George F. Baker Professor of Economics at Harvard University and president of the National Bureau of Economic Research.

38

Michael Rothschild and Lawrence J. White

I believe that two facts provide the key to understanding the behavior of institutions of higher education and of the higher education marketplace in general. First, private colleges and universities and prospective entrants into the market must compete with state institutions. State universities, colleges, junior colleges, and specialized institutions are subsidized by state governments in a way that permits them to offer every type of education at much lower prices than even well-endowed private institutions can. Why states choose to act in this way rather than to provide funds to students and allow them to purchase services in the market (as states do for health care through the Medicaid program) is an interesting question in itself but one that I will not discuss here. The second principal fact is that, because private colleges and universities are nonprofit institutions, those persons in positions of authority generally have little incentive to make the kinds of unpleasant decisions and unpopular changes that would be required in a for-profit context. This lack of incentive is reinforced by a traditional lack of power of university administrators. Colleges and universities are not hierarchical institutions like business corporations, in which the chief executive officer can make major decisions on business policy, personnel, and the like. Instead there is a tradition that requires the president and other key university officials to consult faculties and alumni representatives before making major changes in the structure of the university or its operating policies. This lack of power and lack of incentive reinforce each other. Corporate chief executive officers could decide to sell a major portion of the company, to change the product mix, to change the pricing policy, or to make other such fundamental shifts. They might discuss these plans with key senior corporate officers or with the board of directors, but in the end everyone recognizes that the CEO has the authority to make the decision. The president of a university or the dean of a faculty does not have the same authority. It is hard to imagine a university president announcing a unilateral decision to eliminate the biology department, to acquire another college to be operated as a branch, or to double tuition. Any major decision within a university can only be reached after long and often painful confrontation and negotiation. Such tough decisions may be made when the institution faces very serious financial problems and is threatened with the possibility of bankruptcy. But as a general matter, university officials lack the incentive to make such tough and confrontational decisions in order to reduce costs or increase surplus. There is an interesting analogy to managerial behavior after leveraged buyouts in private shareholder-owned corporations. Although the management of a large for-profit company is supposed to be motivated to make decisions that will increase long-term profits, it is often reported that management behavior changes substantially after a leveraged buyout puts managers in the position

39

The University in the Marketplace

of owners. They are then much more aggressive about cost reductions, including eliminating levels of middle management and making other unpopular changes. If corporate manager-owners are at one extreme of the power-incentive spectrum, university administrators are at the opposite extreme. Lacking the power to make changes without painful confrontation and lacking the personal incentive to overcome that obstacle, administrators are likely to prefer the status quo and to avoid initiatives that would make their institution different from others. Many specific features of college and university behavior can therefore only be explained in terms of the history of higher education in the United States rather than with a model of profit maximization or cost minimization. To those who would insist that the only satisfactory form of explanation is a model of maximizing behavior, I offer the following formal reinterpretation of what I have been saying: Decision makers in universities and colleges are (of course) utility maximizers whose personal utility is a function of such things as compensation, the pleasantness of their day-to-day work experience, the satisfaction of doing their job well, and the prestige of their positions. They know moreover that their future employment prospects (salary, position, etc.) depend on their current performance and reputation. Seeking to achieve in the institution a major change that runs counter to existing practice at that and other institutions might increase the “satisfaction of doing the job well,” but it would not increase salary. It would create confrontations that reduce the pleasantness of the daily work experience, and it might create a reputation for being disruptive that would hurt the individual’s future job prospects at that or other institutions. In such a situation, the utility maximizer generally does not make major changes in the status quo or seek to depart from general practice among similar institutions. The competition from heavily subsidized state institutions prevents the entry of for-profit institutions that could create a different style of management based on different incentives and different authority. Consider now how this perspective helps to answer some of the apparent puzzles raised by Rothschild and White.

The Lack of For-Profit Institutions of Higher Education Rothschild and White suggest that for-profit educational institutions do not exist because students cannot borrow adequately against the human capital that will be created. That is not convincing, since parents now pay as much as $80,000 for four years of undergraduate education at private institutions. A more plausible answer is that they are willing to pay those fees because of the reputation and presumed exclusiveness of the private colleges and universities. A new for-profit institution would be unlikely to develop the reputation required to overcome the very subsidized tuition at public institutions.

40

Michael Rothschild and Lawrence J. White

Input Policy In looking at inputs, Rothschild and White ask why research universities have been so reluctant to establish job categories for outstanding teachers or to use new video technology to increase the efficiency of teaching. I would add another “puzzle.” Colleges and universities do not permit faculty members to teach regularly at other institutions-even during the hours that they are permitted to engage in outside activities, even for “noncompeting” institutions. Why, for example, does Yale not permit a faculty member to spend a few hours per week teaching a regular course at the University of Bridgeport (to use the institution that Rothschild and White cite as one that does not compete with Yale for students)? The professor might augment his or her income by 20 percent or more, students (and possibly faculty) at Bridgeport would benefit, and the professor would be diverting no more time from Yale duties than would be spent in consulting, editing, or textbook writing. There is nothing inherently unprofessional about such behavior, since physicians frequently work at more than one hospital. A for-profit university might permit such outside activities as a way of increasing a faculty member’s income with little or no extra effort or might even organize such an outside market for its faculty members’ services. Any such change would antagonize a considerable number of faculty members, who might worry that this would eventually lead to lower salaries as it becomes expected that faculty members will do such outside teaching. The academic profession as a whole would frown on such an innovation as potentially reducing the total demand for faculty members. Students and alumni of Yale would fear that the Yale education would no longer be seen as unique. The same considerations relate to the increased use of video recording that Rothschild and White mention. A dean or provost who contemplated organizing the Yale “faculty timesharing service” to offer Yale faculty services to neighboring institutions would probably be more impressed by the confrontations that lie ahead than by the potential gains if he succeeded. With no market competition to force the change and no incentive for personal gain to make the university administrator accept the pain of making the change, the status quo continues. Pricing and Output Mix

Or consider the Rothschild-White puzzle that universities charge the same amount per course (or at least per point of academic credit) regardless of the marginal cost of producing that bit of educational service and of the pattern of demand elasticities. There is of course a problem of defining the relevant marginal cost. An additional student can enter a large lecture at no extra cost in terms of instruction and without imposing any adverse externalities on other students in the course. The only additional resource requirement may be the cost of grading or perhaps of a small fraction of a graduate student teaching

41

The University in the Marketplace

assistant. Taken literally, the marginal cost is so close to zero that a two-part tariff pricing system of the type suggested by Rothschild-White would degenerate into the existing flat fee. A looser and perhaps better definition of marginal cost would regard the course rather than the student as the unit to be evaluated. Thus a small class of 10 taught by a professor would be deemed to have 20 times the marginal cost per student in a class with 200 students, at least if the professor’s salary does not have to be increased for teaching large classes. But assuming that this problem of defining marginal cost is overcome, consider the effect of introducing a new schedule of tuition charges that reflects the fact that the marginal cost of a large lecture course in economics or history is lower than a small class in French drama or Irish poetry. Many students might decide that the extra cost of the more obscure courses was not worth paying. They would flock to the large low-cost lectures. As the specialized courses shrink, their price would rise, accelerating this adjustment. This move to take advantage of economies of scale while still providing the specialized products when there is sufficient market demand is just what we as economists like to see happen in other industries. We might have certain reservations about the narrowing of undergraduate education or the lack of in-depth specialization and of faculty-student contact, but even this might be overcome by requiring students to take a certain number of small specialized courses in order to receive a degree. Yet think of the transition problem from the point of view of the dean or university president. The faculty members whose courses are no longer wanted cannot be discharged because of tenure commitments. Earlyretirement incentives and other policies might help to eliminate these quasifixed costs, but the faculty would be unhappy, other educational institutions would be critical, some students would object to the higher cost of the courses that they had planned to take, and so on. Even if a new variable-price tuition system with adequate educational safeguards could be designed that would make the university more efficient, the time and pain of the transition make it easy to understand why an administrator with no personal financial incentive would be loath to try.

Market Failures Rothschild and White discuss (section 1.3) whether a competitive allocation is efficient. They reach the conclusion that, although one cannot be certain, there are “weaknesses in the a priori argument that competition will allocate students to colleges efficiently.” Nevertheless, when they discuss the specific issue of collusion in the setting of scholarships and tuitions (section 1.4), they conclude that the “case for a market failure does not appear to be strong.” I agree with that conclusion. More generally, while I have no doubt that there are market failures that would cause a theoretical purist to reject a decen-

42

Michael Rothschild and Lawrence J. White

tralized system of education in favor of government regulation or private collusion, 1 think it is important to recognize the imperfections of the government system and the failure of nonprofit institutions to act optimally. Certainly, the experience around the world in a variety of other fields is causing governments everywhere to reduce regulations and to privatize previously state-owned or state-subsidized institutions.

Future Research Making the best use of our higher education resources is important not only because of the volume of inputs in this industry but, even more significantly, because of the contribution of higher education to aggregate economic growth and the level of individual economic success. Research on the economics of the higher education industry is also something that we as university-based economists are particularly well suited to do. We start with a much better understanding of the institutions of this industry than of other manufacturing and service industries. I hope that the fascinating paper by Rothschild and White and, more generally, this volume will stimulate substantial research on the important issues in the economics of higher education.

2

Adolescent Econometricians: How Do Youth Infer the Returns to Schooling? Charles F. Manski

2.1 Basic Ideas Economists analyzing schooling decisions assume that youth, having compared the expected outcomes from schooling and other activities, choose the best feasible option. Viewing education as an investment in human capital, we use the term returns to schooling to refer to the outcomes from schooling relative to nonschooling. Given the centrality of the expected returns to schooling in economic thinking on educational behavior, it might be anticipated that economists would make substantial efforts to learn how youth form their expectations. But the profession has traditionally been skeptical of subjective data; so much so that we have generally been unwilling to collect data on expectations. Instead, the norm has been to make assumptions about expectations formation. 2.1.1

Prevailing Expectations Assumptions

Economic studies of schooling behavior have universally assumed that expectations formation is homogeneous; all youth condition their beliefs on the same variables and process their information in the same way. On the other hand, the hypothesized conditioning variables and information processing rule have varied considerably across studies. In his analysis of the major field decisions of male college students, Freeman (1971) assumed that these youth condition their expectations on their sex Charles F. Manski is Wolfowitz Professor of Economics at the University of Wisconsin-Madison and a research associate of the National Bureau of Economic Research. The research reported here was partially supported by grant 87ASPE041A from the Office of the Assistant Secretary for Planning and Evaluation, U.S. Department of Health and Human Services, and by grant lROl HD25842 HLB from the National Institute of Child Health and Human Development. The author is grateful to the NBER Conference on Higher Education participants for their comments.

43

44

Charles F. Manski

and on their common knowledge of the incomes realized by earlier cohorts. He assumed that expectations formation is myopic. Each youth believes that by selecting a given college major, he will obtain the mean income realized by the members of a specified earlier cohort who made that choice.' Willis and Rosen (1979), in their study of college enrollment, took the personal conditioning variables to be sex, armed forces status, and ability. They assumed that youth have common knowledge of the actual process generating life-cycle incomes conditional on these personal variables and on schooling. They hypothesized that expectations are rational, each youth applying knowledge of the true income-generating process to forecast future personal income should he or she enroll or not enroll in college. In the Manski and Wise (1983, chap. 6 ) analysis of college choice, youth condition their expectations for the utility of enrolling in a given college on their own SAT score and on the average SAT score of students enrolled at the college. They do not necessarily know either the outcomes realized by earlier cohorts or the actual process generating outcomes. Rather, they believe the returns from enrolling to be a function of the difference between their own SAT score and the average at the college. The three studies just cited are noteworthy because they make explicit assumptions about expectations formation. In most economic analyses of schooling behavior, the expectations assumptions are implicit in the specification of the decision model. The recent literature shows little concern with expectations formation. The prevailing sentiment seems to be complacency. Either researchers are confident that their expectations assumptions are correct, or they believe that misspecifying expectations is innocuous. 2.1.2 Two Identification Problems In fact, there is no evidence that prevailing expectations assumptions are correct nor reason to think that misspecifying expectations is innocuous. To the contrary, rudimentary treatment of expectations has placed the economics of education at an impasse, caught in a pair of basic identification problems that plague attempts to understand schooling behavior and to measure educational productivity. The first problem is that, not knowing how youth perceive the returns to schooling, one cannot infer their decision processes from their schooling choices. The point can easily be made with a few symbols. The standard economic model assumes that a youth's schooling choice c is a functionfl.) of his or her expected returns to schooling r; that is, c = f i r ) . Suppose that one wishes to learn the decision rulef( .) mapping expectations into choices. If one observes the choices and expectations of a sample of youth, then one can infer 1. In the final chapters of his book, Freeman reported findings from a one-time survey of college students regarding their income expectations in various occupations. But his analysis of these data sheds no light on the realism of the myopic expectations assumption made earlier on.

45

Adolescent Econometricians

the decision rule. But if one observes only the choices of these youth, then clearly one cannot inferfl.). The most that one can do is infer the decision rule conditional on maintained assumptions on expectations. The second problem is that, not knowing youth’s decision processes, one cannot infer the objective returns to schooling from data on realized outcomes. As is well known, any attempt to learn the objective returns to schooling involves facing the selection problem. The problem arises because the youth who choose to enroll in school are those who expect schooling to have favorable outcomes for them. If expected outcomes are related to objective ones, then the outcomes experienced by youth who choose to enroll in school differ from those that nonenrollees would experience if they were to enroll. Likewise, the outcomes experienced by nonenrollees differ from those that enrollees would experience if they were not to enroll. See, for example, Griliches (1977), Heckman and Robb (1985), and Manski (1989). The selection problem implies that any effort to infer the objective returns to schooling from observations of realized outcomes requires at least some knowledge of the way youth make their schooling decisions. But we have already observed that, lacking data on the expectations of youth, one can only learn youth’s decision rules conditional on maintained assumptions on expectations. Hence, one can only infer the objective returns to schooling conditional on the validity of expectations assumptions. It is important to understand that these identification problems arise even in a stationary world, where the objective returns to schooling are constant over time. This will be illustrated through an example in section 2.3. Further identification problems may arise in a world with aggregate productivity shocks, where the objective returns to schooling change with time. 2.1.3 The Econometrics of Expectations Formation

The two identification problems just described would not be of concern if there were reason to think that prevailing expectations assumptions are correct. Logic and some indirect empirical evidence suggest otherwise. In particular, there is little reason to think that all youth form their expectations in the same way. The logical point is that youth forming expectations face the same kind of inferential problem as do econometricians measuring educational productivity. Youth and econometricians may possess different data on realized outcomes, may have different knowledge of the economy, and may process their information in different ways. But both want to use their data and knowledge to learn the objective returns to schooling conditional on the available information. It follows that youth, like econometricians, face the selection problem. If youth use data on realized outcomes to form their expectations, then their interpretation of these data must depend on how they think other youth make schooling decisions. Expectations formation will be homogeneous only if all youth make the same assumptions about the behavior of their peers.

46

Charles F. Manski

The empirical evidence is indirect but, I believe, compelling. Although we lack data on the expectations of youth, we have extensive data on the practices of econometricians studying educational productivity. For 30 years, in perhaps hundreds of published studies, econometricians have sought to learn the objective returns to schooling. Reading this literature reveals that econometric studies of the returns to schooling vary greatly in the conditioning variables used, in the outcome data analyzed, and in their handling of the selection problem. Compare, for example, Willis and Rosen (1979) and Murphy and Welch (1989). The former study analyzes data from the NBER-Thorndike Survey, estimates returns to schooling conditional on measured ability, and is explicitly concerned with the effect of unmeasured ability on the selection of students into schooling. The latter piece analyzes data from the Current Population Surveys, which contain no ability measures, and implicitly assumes that the selection of students into schooling is unrelated to ability. If experts can vary so widely in the way they infer the returns to schooling, it is reasonable to suspect that youth do, as well. 2.1.4 Elaboration on the Basic Ideas The remaining sections of this paper elaborate on the foregoing basic ideas. Section 2.2 indicates that, if economists want to learn how youth perceive the returns to schooling, we cannot rely on the expectations research performed by other social scientists. Section 2.3 uses a simple formal model to show the different patterns of choices and outcomes that can result if youth do or do not condition their expectations on ability. Section 2.4 makes concluding comments on expectations research in economics. 2.2 Expectations Research in Psychology and Sociology In contrast to economists, psychologists and sociologists routinely collect and analyze subjective data of many kinds, including expectations data from youth. I have sought to determine whether useful lessons can be extracted from these literatures. Unfortunately, my findings have been largely negative. 2.2.1 Measurement of Expectations The prevailing measurement practice is to interpret responses to loosely worded questionnaire items as indicators of youths’ expectations. Berndt and Miller (1990), for example, ask their sample of junior high school students to respond, on a five-point scale, to the question “How valuable do you think your education will be in getting the job you want?” Mickelson (1990) asks her sample of high school seniors to express their degree of agreement with the statement “Studying in school rarely pays off later with good jobs.” Most of the literature poses such vague questions. An exception is a recent study of the income expectations of college seniors, by Smith and Powell (1990). These authors ask respondents to make unconditional forecasts of their “anticipated annual income in 10 years” and their “expected earnings” in the first

47

Adolescent Econometricians

year of their first job. They also ask respondents to provide similar forecasts for the average member of their class.

2.2.2 Theories of Expectation Formation The looseness with which psychologists and sociologists measure youth’s expectations is matched by looseness in their thinking about expectations formation. Researchers in these fields theorize verbally rather than mathematically. As a consequence, it is even difficult to determine whether different researchers interpret the term expectations in a common, coherent fashion.* The central social psychological idea is that expectations formation is a social phenomenon, each person learning about his prospects by observing the experiences of others. Bandura (1986,47) writes: If knowledge could be acquired only through the effects of one’s own actions, the process of cognitive and social development would be greatly retarded. . . . Fortunately, most human behavior is learned by observation through modeling. By observing others, one forms rules of behavior, and on future occasions this coded information serves as a guide for action. . . . Much social learning is fostered by observing the actual performances of others and the consequences for them. This statement seems sensible; indeed I could interpret it as endorsing the idea that youth learning the returns to schooling are implicit econometricians. Unfortunately, the social psychological literature does not go much beyond the generalities expressed by Bandura. A long line of research, beginning with Hyman (1942), has sought to operationalize the idea that individuals learn from their “reference groups”; Bank, Slavings, and Biddle (1990) give an interesting historical account. But the idea of a reference group seems as amorphous today as it was 50 years ago. It appears to me that if social psychologists are to make progress in understanding expectations formation, they must end their dependence on verbal reasoning, which invites conceptual ambiguity and logical inconsistency. Coherent analysis of complex social processes demands the discipline of formal modeling.

2.3 A Model of Information, Schooling Choices, and Outcomes I observed in section 2.1 that some econometric studies (e.g., Willis and Rosen 1979; Manski and Wise 1983) assume that youth condition their expectations on their ability, while other studies (e.g., Freeman 1971; Murphy and Welch 1989) assume that they do not. Given the variation in econometric 2. There are mathematical psychologists who interpret expectations in the same subjective probabilistic way as economists do. See, for example, Kahneman and Tversky (1979) or Camerer and Kunreuther (1989). Their work, however, seems to have had no impact on psychologists or sociologists concerned with schooling behavior.

48

Charles F. Manski

practice, it is of interest to determine how observed patterns of schooling choices and outcomes may depend on this aspect of expectations. To address the question, I pose a simple stationary human capital model (section 2.3.1) and consider two alternative assumptions on expectations: myopic youth either condition expectations on ability (assumption A) or they do not (assumption B). I then derive the schooling choices and outcomes that result in the two cases (section 2.3.2). It turns out that in both cases there is a unique equilibrium in which expectations, although myopic, are fulfilled. But the characteristics of these equilibria differ. The main findings are: Assumption A yields a rational expectation equilibrium. Assumption B yields equilibrium expectations which are fulfilled, yet systematically incorrect. Fewer low-ability and more high-ability youth enroll under expectations assumption A than under B. The gross enrollment rate under A may be less or greater than under B, depending on the values of the model parameters. For some parameter values, the mean income realized by enrollees is known to be higher under assumption A than under B. Having compared the two patterns of choices and outcomes, I consider the implications of misspecifying expectations for econometric analysis of schooling behavior (section 2.3.3). It is found that if youth do not condition their expectations on ability, then an econometrician who assumes they do so may mistakenly conclude from observed schooling behavior that youth are unconcerned with the returns to schooling. 2.3.1 The Model Maintained Assumptions

Assume an overlapping-generations world in which each person lives for two periods. In the first period, a youth can choose to work (c = w) or to enroll in school (c=s); in the second period, all adults work. At the time of the schooling decision, youth know their real-valued ability z, their real-valued taste for schooling v, and the present discounted life-cycle log-income q that they would receive if they were to work immediately; for simplicity, assume that q is constant across the population and normalize the income scale by setting q = 0. Youth do not know the discounted log-income y they would receive if they were to enroll in school; y is a random variable whose realization becomes known after schooling is completed. Each youth’s value of (y, v, z ) is independently drawn according to the following time-stationary process: y = a, v = a*

+ p,z + El, + p,z + E,

p, 2 0

49

Adolescent Econometricians

(el, E,, z )

-N

1 I: :

0, 0 a: 0

Thus, the objective probability distribution of ( y , v, z ) is trivariate normal. Letting z be a standard normal random variable and assuming that PI 2 0 are normalizations that make ability a well-defined concept. Assuming the variance of (el, e2, z ) to be diagonal is a real restriction; conditional on ability z, a youth’s postschool income y and taste for schooling v are statistically independent. The youth in a given generation share certain information about the schooling choices and realized incomes of the preceding generation. Let E*(y I z, v) be a youth’s subjective expected value of y conditional on (z, v) and the common information. The decision rule is c = s if E*(y I z , v) = w otherwise

+ v >O

.

Expectations Assumptions

The model is complete when the subjective expected income E*( y 1 z, v) is specified. Although I have earlier criticized the prevailing assumption that expectations formation is homogeneous, I retain that assumption here. The recent fashion in economics has been to assume that expectations are rational; youth a priori know that equation (1) holds and so set

(3)

E * ( y I z , v) = a,

+ plz.

The realism of this assumption is most questionable. Having witnessed the struggles of econometricians to learn the returns to schooling, I find it difficult to accept the proposition that adolescents are endowed with this knowledge. I instead assume that youth form their expectations in the manner of practicing econometricians: youth observe the incomes realized by members of the preceding generation who chose schooling, and they make inferences from these observations. But what information do they possess about the experiences of the preceding generation, and how do they use this information to form their expectations? I shall consider two cases of myopic expectations. In each case a youth, having observed the mean income E,( y I R, c = s) realized by those members of the preceding generation who chose schooling and who had specified characteristics R, believes that he or she will receive the same mean i n ~ o m eThe . ~ two cases differ in the characteristics R on which youth condition their expectations. They are

3. The mean income E,, (y I R, c = s) is well defined only if there exist members of the preceding generation who chose schooling and who had characteristics R. The assumptions made in this section guarantee that this condition is satisfied (see Manski 1991).

50

Charles F. Manski

Assumption A: E*,(y I z , v ) = E,(y 1 z , c=s) Assumption B: E*B(y 1 z, v) = E,(y 1 c=s). Youth might form expectations as in assumption A if they observe the abilities and realized incomes of those members of the preceding generation who chose schooling. Suppose, however, that youth cannot observe the abilities of their elders. Unaware that income varies with ability, they might then form expectations as in assumption B.4

2.3.2 Schooling Choices and Realized Incomes The two expectations assumptions imply systematically different patterns of schooling choices and realized incomes. To see this, I first derive the choice and income patterns that emerge under the two assumptions.

Expectations Conditioned on Ability and Schooling By (2), a youth’s schooling choice c is a function of his or her ability-taste pair (z, v). By ( l ) , income y is statistically independent of v, conditional on z. Hence, (4)

Eo(y 1 z, c=s)

=

Eo(y I z )

=

a]

+ plz.

Thus, in the time-stationary environment ( l ) , the myopic expectations (A) turn out to be rational. (These expectations would not generally be rational if the process generating (y, v, z) were not time-stationary.) By (l), ( 2 ) ,and (4), the decision rule is c = s if a,

(5)

+ a2 + (PI + P,) z + e2 > 0

= w otherwise.

So the probability that a youth with ability z selects school is P,(c=s

1 z)

[*, +

= @

a2

+ (PI +

P2)

m2

1.

where a(.)is the standard normal distribution function. The unconditional probability of schooling is (7)

=

@(yA),

+

wherey, = (a,+ a,)[@, + P,)’ U;I-I’~. The mean income realized by youth with ability z who choose schooling is 4. Other specifications for Cl may be of interest. For example, Streufert (1991) assumes that youth observe the abilities, choices, and incomes of residents of their neighborhoods. He also supposes that neighborhoods are segregated by income classes. These assumptions suggest the expectations model

E*(y I z, v) = E J y I z, ye@, b), c=sl , where [a,b] is the interval of incomes found in a youth’s neighborhood.

Adolescent Econometricians

51

a, + p, z. Thus, income expectations are fulfilled. The mean income realized by all youth who choose schooling is E,(y

I c=s)

I

+ a, + (PI + P,) z +

= E [ y a,

€2

> 01

Expectations Conditioned on Schooling Only

Suppose that assumption B holds. Then the decision rule is

1

c = s if E, ( y c=s)

(9)

+ a, + p,z + e2 > O

= w otherwise.

So the probability that a youth with ability z selects school is

and the unconditional probability of schooling is P,(c

(11)

= s, = @(yB)

7

wherey, = [E,(y I c=s) + a,l(p: + u;)-’”. The mean income realized by youth with ability z who choose schooling remains a, p , z as before. The mean income realized by all youth who choose school is

+

E,(y

I E,

0, I c=s)

= =

(12)

E [ y I EoO, I c=s) + 01, + P,z + 6, > 01 a, P,E[z I E, ( y I C=S) + a, + P,Z + E, > 01

+

=a,+-

~BWY,)

@(%)

+

where6, 5 p, p,(p: U;)-I’~. Suppose, as seems reasonable, that the taste for schooling does not decrease with ability; that is, let p, 2 0. Then there is a unique E,( y I c = s) 2 a,such that expectations are fulfilled. To see this, observe that expectations are fulfilled if (12) holds with EB(y I c = s) = E, ( y 1 c = s); that is, if (13) E,(y I c=s)

=

a,

+ P,E[z I E,(y I

C=S)

+ a, + P,z + ez > 01.

If p, = 0, (13) is solved at E,(y 1 c=s) = a,.If p, > 0, (13) is solved at some E,(y I c=s) > a,;this is so because E[z I E, ( y I c=s) + a, + p, z E, > 01 is a differentiable, strictly decreasing function of E, ( y I c = s) whose value falls to 0 as E, (y I c = s) rises.

+

52

Charles F. Manski

Observe that equilibrium expectations under assumption B , even though fulfilled, are systematically incorrect except in the special case P, = 0. Unconditional on ability, a youth’s objective expected income following schooling is ai.But it has just been shown that, in equilibrium, youth’s common subjective expected income exceeds a,whenever P, > 0. The fulfilled-expectations equilibrium (1 3) is globally stable when Pi < P,; I do not know the stability properties when P, 2 P,. To show that P, < P, implies global stability, observe that global stability is guaranteed if the derivative of the right-hand side of (13) with respect to E, ( y 1 c=s) is always less than one in absolute value. It is shown in Goldberger (1983) that 0 < dE(z 1 z - t)/dt < 0. It follows that, for all [E, ( y I c = s), E ~ ] ,

Taking the expectation over E, of the derivative in (14) yields

So the derivative is less than one in absolute value if

PI < P, .

Comparative Schooling Choices

The remainder of this section compares the patterns of schooling choices and realized incomes that emerge under the two expectations assumptions. In this discussion, I assume that the taste for schooling does not decrease with ability; that is, P, 2 0. In discussing expectations assumption B, I restrict attention to the fulfilled-expectations equilibrium (1 3).5 Let us first compare the ability-conditioned enrollment probabilities P,(c = s I z ) and P,(c = s I z ) , given in (6) and (10). Recall that the solution to (13) is E&y I c=s) = a, if P, = 0, and satisfies E,(y I c=s) > a I if P, > 0. Hence, evaluated at z = 0, P,(C=S

(16)

P,(c=s

I z=O) I z=O)

This and the fact that (PI (17)

+ P,)

= P,(C=S

< P,(c=s

> P, imply that

P,(c=s I z ) < P,(c=s

On the other hand, (16) and the fact that exists a zo 2 0 such that (18)

P,(C=S

I z=O) if P, = 0 I z=O) if P, > 0 . I z ) , all z < 0 . (PI + p,) > P,

imply that there

1 z ) > P , ( C = S I z ) , all z > z, .

5 . Thus, this discussion is not concerned with the dynamic adjustment questions studied by Freeman (1971).

Adolescent Econometricians

53

Thus, fewer low-ability youth and more high-ability youth enroll under expectations assumption A than under B. Overall, enrollments under assumption A may be less or greater than under B, depending on whether yA is less or greater than yB(see equations 7 and 11). We find that Y B YA

(19)

if a, +
yA = y, = 0 yA = 0 < ye O
a, < 0 and

p, = 0 a2< 0 and p, >> p, a, = 0 and p, = 0 a, = 0 and p, > 0

if a, + if a, + ifa, +a,> 0 .

Hence,

P,(c=s) < PA(c=s) < 1/2 PA(c=s) < min [1/2,P,(c=s)] (20) PA(c=s) = P,(c=s) = 1/2 PA(c=s) = 1/2 < P,(c=s) < PA(c=s) < P,(c=s) 1/2

if if if if if

a, a,

+ a, < 0 and (3, = 0 + a, < 0 and p, >> p,

+ a, = 0 and p, = 0 a, + a, = 0 and p, > 0 a, + a, > 0 . a,

If a, + a, < 0 and if p, and p, are the same order of magnitude, then the ordering of P,(c = s) and PA(c= s) appears to depend on the specific values of the model parameters. Comparative Realized Incomes

The mean income realized by a youth of ability z who enrolls in school is + p , z , whether expectations assumption A or B holds. The mean income of all enrollees depends on the ability distribution of enrollees and so varies with the expectations assumption, as follows. By (8) and (12), a,

(21)

E,(y

I c=s)

- E,(y

I c=s)

=

~A+(Y,J ~

@(‘YA)

-

~,+(YJ ~

.

‘(YE)

It can be shown that 6, > 6, for all values of the model parameters; moreover, 6, = 0 if p, = 0.6The Mills ratio +( .)/@(.) is strictly decreasing in its argument, so 6. To prove that 6, > 6,, observe that

54

Charles F. Manski

Equation (23) shows that, for some values of the model parameters, the mean realized income of school enrollees is higher under expectations assumption A than under B. I have not been able to determine the relationship between EA(y I c = s) and E,( y 1 c = s) for other parameter values. 2.3.3 Econometric Analysis with Misspecified Expectations Analysis of Behavior It remains to inquire into the consequences for econometric analysis of misspecifying expectations. Consider the following idealized description of an econometric analysis of schooling choices: For each member of a random sample of youth, an econometrician observes (c, z ) and observes y when c = s; he does not observe v. The econometrician assumes that (1) describes the objective probability distribution of ( y , v , z ) and that (2) is the decision rule youth use to make their schooling choices. As is common in the literature, he assumes that tastes for schooling are independent of ability; that is, p, = 0. Moreover, he makes the conventional assumption that expectations are rational. Believing that (6) describes choice behavior and that p, = 0, the econometrician would form the probit model (24)

P[c=s I E * ( y I z , v )

=

a,

+ p,z]

+

= @'(To T Iz )

and estimate (T,,, T,) by maximum likelihood. He would interpret T~to be (a,+ a,)/u, and T ,to be PI/'+,. Suppose that the econometrician is correct in assuming (1) and (2) but incorrect otherwise; in fact, p, may be positive and assumption B holds. Then (10) describes actual choice behavior, and the econometrician's interpretation of (T,,, T,)is incorrect. In reality, T~ = [E&y 1 c=s) + a,]/u, and IT^ = Pz/'T,-

The misinterpretation of T ,is of particular interest. The econometrician believes IT, to measure educational decisions' sensitivity to changes in the income returns to schooling. In fact, T ,measures the degree to which tastes for schooling vary with ability. Suppose, for example, that p, = 0 as assumed. Then IT, = 0. Finding this, the econometrician would conclude that, in making their schooling choices, youth are unconcerned with the income returns to schooling. This conclusion would, of course, be incorrect. If the returns to schooling were to shift through a change in a ] ,then the intercept no would change and so would the probability of enrolling.

55

Adolescent Econometricians

Analysis of the Returns to Schooling

I have made the idealized assumption that the econometrician observes ability z without error. Given this, data on enrolled youths’ abilities and realized incomes can be used to obtain a consistent least-squares estimate for the parameters (a,, p,). It might therefore seem that, if z is observed, analysis of the objective returns to schooling requires no knowledge of how youth make their schooling choices. But there is an implicit expectational assumption, namely that youth do not know E, at the time of their schooling decisions. If this assumption fails, then the econometrician’s estimate of (a,, p,) is not consistent. The selection problem implies that an econometrician analyzing the returns to schooling must take a stand on the information youth use in forming their expectations.

2.4 Conclusion: Expectations Research in Economics The question posed in the title of this paper cannot be answered at this time. Having chosen to make assumptions rather than to investigate expectations formation, economists do not know how youth infer the returns to schooling. If youth form their expectations in anything like the manner that econometricians study the returns to schooling, then prevailing expectations assumptions cannot be correct. Without an understanding of expectations, it is not possible to interpret schooling behavior nor to measure the objective returns to schooling. As a consequence, the economics of education is at an impasse. As I see it, progress is possible only if economists become more willing to entertain the use of subjective data in empirical analysis. Decisions under uncertainty reflect the interplay of preferences, expectations, and opportunities. Choice data alone cannot disentangle these factors. The identification problem can be solved if choice data are combined with interpretable subjective data on expectations and/or preferences. The question, of course, is whether interpretable subjective data can be obtained. The dominant view expressed by economists today is negative. In particular, economists often assert that respondents to surveys have no incentive to answer questions carefully or honestly; hence, they conclude, there is no reason to think that subjective responses reliably reflect respondents’ thinking. But this reasoning is not applied consistently. Empirical economic analyses of schooling behavior routinely use respondents’ self-reports of their backgrounds, choices, and outcomes. Many analyses use scores on tests administered with surveys to measure respondents’ ability. Thus, ironically, economists’ own revealed preferences in empirical analysis are somewhat at variance with their expressed views about the interpretability of survey data. It should be noted that economists’ views on the use of subjective data have not always been so negative. In the 1940s it was common to interview businessmen about their expectations and decision rules. In an influential article,

56

Charles F. Manski

Machlup (1946) sharply attacked then-existing survey practices as not yielding credible information. This article apparently played an important role in dampening the enthusiasm of economists for subjective data. But Machlup only sought to criticize the collection of subjective data through standardized questionnaires. He stressed that cost and revenue expectations are subjective. He advocated research in which the economist learns the institutional peculiarities of a firm and then questions its managers in language they understand. From the mid-1950s through the mid-l960s, economists analyzed data on consumers’ buying intentions (see, e.g., Juster 1966). Although this practice has since almost ceased among economists, it remains firmly entrenched among demographers and market researchers. I have recently reviewed and reinterpreted this literature in Manski (1990). The early literatures on businessmen’s expectations and on consumers’ intentions may hold lessons for efforts to learn youth’s expectations. The present problem, however, seems more difficult than those treated previously. Whereas past efforts have sought to elicit unconditional forecasts from adult respondents, here we need to elicit choice-conditioned forecasts from adolescent respondents. We shall not know whether this is feasible until we try.

References Bandura, A. 1986. Social foundations of thought and action. Englewood Cliffs, N.J.: Prentice-Hall. Bank, B., R. Slavings, and B. Biddle. 1990. Effects of peer, faculty, and parental influences on students’ persistence. Sociology of Education 63:208-25. Berndt, T., and K. Miller. 1990. Expectancies, values, and achievement in junior high school. Journal of Educational Psychology 82:319-26. Camerer, C., and H. Kunreuther. 1989. Decision processes for low probability events: Policy implications. Journal of Policy Analysis and Management 8565-92. Freeman, R. 1971. The marketfor college-trained manpower. Cambridge, Mass.: Harvard University Press. Goldberger, A. 1983. Abnormal selection bias. In Studies in econometrics, time series, and multivariate statistics, ed. S . Karlin, T. Amemiya, and L. Goodman. Orlando, Fla.: Academic Press. Griliches, Z. 1977. Estimating the return to schooling: A progress report. Econometrica 45:1-22. Heckman, J., and R. Robb. 1985. Alternative methods for evaluating the impact of interventions. In Longitudinal analysis of labor market data, ed. J. Heckman and B. Singer. Cambridge: Cambridge University Press. Hyman, H. 1942. The psychology of status. Archives of Psychology, no. 269. Juster, T. 1966. Consumer buying intentions and purchase probability: An experiment in survey design. Journal of the American Statistical Association 61:658-96. Kahneman, D., and A. Tversky. 1979. Prospect theory. Econometrica 47:263-92. Machlup, F. 1946. Marginal analysis and empirical research. American Economic Review 36519-54.

57

Adolescent Econometricians

Manski, C. 1989. Anatomy of the selection problem. Journal of Human Resources 24:343-60. . 1990. The use of intentions data to predict behavior: A best case analysis. Journal of the American Statistical Association 85:934-40. . 1991. Nonparametric estimation of expectations in the analysis of discrete choice under uncertainty. In Nonparametric and semiparametric methods in econometrics and statistics, ed. W. Bamett, J. Powell, and G. Tauchen. New York: Cambridge University Press. Manski, C., and D. Wise. 1983. College choice in America. Cambridge, Mass.: Harvard University Press. Mickelson, R. 1990. The attitude-achievementparadox among black adolescents. Sociology of Education 63:44-61. Murphy, K . , and F. Welch. 1989. Wage premiums for college graduates: Recent growth and possible explanations. Educational Researcher 18(4):17-26. Smith, H., and B. Powell. 1990. Great expectations: Variations in income expectations among college seniors. Sociology of Education 63:194-207. Streufert, P. 1991. The effect of underclass social isolation on schooling choice. Institute for Research on Poverty Discussion Paper no. 954-91. University of Wisconsin-Madison. Willis, R., and S. Rosen. 1979. Education and self-selection. Journal of Political Economy 87:S7-S36.

Comment

Eric A. Hanushek

Charles F. Manski has a history of contributions to the understanding of higher education and, particularly, college choice that goes back farther than that of virtually everybody participating in this conference. Moreover, his contributions have been especially important, bringing serious analytical effort to bear on an area that tends to be punctuated more by fuzziness. Thus, it is good to have him return to the general area. His paper pursues an extraordinarily important set of issues: How do prospective students form expectations about the advantages of higher education? How do expectations condition reality and the outcomes that are observed? And how do the efforts of analysts interact with the actual choice process of students? The overall idea is quite straightforward. If students’ expectations determine the pattern of college enrollment, ignoring expectations could lead to selection problems that imply biased estimates of the value of schooling. Unfortunately, however, economists have directed virtually no attention to understanding the expectations problem in higher education. Manski is led to conclude that the only hope for understanding not only college choice but also labor market returns on schooling is the collection and analysis of subjective data on students’ expectations. Eric A. Hanushek is professor of economics and political science at the University of Rochester.

58

Charles F. Manski

I am sympathetic with his concerns about the need to understand expectations of students. I think that this is an important area of research. Investment in human capital is really an exercise in decision making under uncertainty, and the character of individual perceptions and expectations must be important. Moreover, having just participated in the college decision-making process as a parent, I am struck by how far assumptions about complete information appear to be from reality. Finally, I am very supportive of more refinements in modeling college choice in general. My caricature of the current state of modeling is that the analysis begins with a standard human capital investment model which compares expected benefits to the costs of schooling but then, in the empirical work, turns to a simple regression of college attendance on tuition. While there are clear exceptions, much of the work in this general area is simply very primitive. Therefore, the systematic study of this by Manski is most welcome. Of course, as critic, I must also be clear that I am less persuaded that nothing can be done about understanding the returns to schooling or the choice of colleges without delving into subjective views of students. And I am currently unsure of exactly how Manski would have us proceed. The paper has two distinct sections. The first argues the general principle that understanding expectations is important, while the second works through a specific model. I will consider each in turn. The overall motivation for considering student expectations follows as a logical extension of much of the current work in labor economics. The discussion of income determination has been dominated for some 30 years by consideration of unmeasured individual characteristics and how neglect of these might bias statistical results. Manski simply takes this argument a step further: if trained econometricians have such difficulties, surely high school seniors also have problems. A central feature is then to understand how different expectations of students affect actual outcomes. The general discussion of expectations makes two points. First, expectation formation is central to much of current economics, both micro and macro, but the underlying basis for individual expectations is often not even discussed, let alone analyzed. Manski highlights the importance of individual expectations about future earnings, an appropriate starting point; but the issues are clearly much larger, including such things as expected schooling costs and attrition probabilities. Second, other disciplines which purportedly consider expectations-psychology and sociology-do not do very well at it. Frankly, this motivation does not lead me to great optimism about our ability to generate or use subjective expectations information in refining our ideas of income determination or college choice. For some time, economists have flirted with the idea of using subjective information for predictive or interpretative purposes. I think, for example, of debates in the 1950s about the importance of expectations in determining investment decisions in productive capacity. While there continue to be periodic surveys of the sentiments of

59

Adolescent Econometricians

purchasing agents and the like, I see little evidence that such subjective expectations information has made great inroads into aggregate econometric models or that it has helped in producing improved forecasts of investment activity. This is matched by specialists in other disciplines who purport to be able to measure expectations but do not appear able to do so. The implicit argument in Manski’s paper, moving to the second general section, is that developing an explicit model of behavior will inform us on what data to collect and what subjective information to gather. It will also inform us on how subjective information affects specification and estimation of the choice model. Unfortunately, I think his formal models tend to confuse the issues and to distort the analysis. The basic model has three distinct features: formation of expectations about future earnings, unmeasured individual heterogeneity (which, in the timehonored labor economics tradition, is simply labeled “ability”), and heterogeneity of individual taste for schooling. The unfortunate part of his specific model is that the results depend crucially on the full structure of the model. While similar results might come from other models, much of the leverage of this structure relates directly to individual tastes for schooling-something that is quite independent of student expectations or how they are formed. To be clear, allowing for heterogeneity of tastes is not inherently peculiar, but it does make the expectations story very hard to parse out. Moreover, many of the central results in the theoretical section appear to evaporate if tastes do not enter the model systematically. In other words, as I work through the formal models, the role of expectations does not seem to be central to the results, even though the paper starts and ends with a plea for better understanding of expectations. At the outset, I had hoped that the Manski model and foray into expectation formation might shed light on some currently perplexing issues about college attendance. Specifically, I was hoping that the explicit consideration of expectations might aid in untangling the effects of the substantial changes in relative wages on college-going behavior. From the mid-1970s to the mid-l980s, the wage premium for college education (as opposed to high school education) appeared to explode. By the estimates of Murphy and Welch (1989)-which Manski discounts because of lack of consideration of expectations-the college premium went from roughly 30 percent to 70 percent for new labor market entrants. How have students taken this information into account, and does expectations formation explain the apparently sluggish response of students to what appear to be extremely strong market forces? Or, on a slightly different front, is there something about income expectations that fits into the apparently peculiar pattern of relative black and white college attendance in the 1980s? These issues, while alluded to in Manski’s discussion, are entirely different from the ones he considers-the role of ability and tastes in a simple selection model where all individuals of the same ability face the same income stream. When I look at the aggregate raw data, I suspect that the effects of

60

Charles F. Manski

overall college-high school earnings differences and changes in the differential are much more important than pure ability rents and misperceptions of ability rents. Let me return to the starting point. I could not agree more with Manski that student expectations must be central to the college choice problem. I also believe that we possess a very primitive understanding of expectations, even though we have elevated the role of expectations in virtually every area of economics. Finally, I believe that economists have much to offer in measuring and understanding expectations, because the empirical force of expectations can only be understood from an underlying decision theoretic perspective. On the other hand, my comments should suggest that it might be some time before investments in understanding expectations from subjective data pay off. I conclude that we should pursue better measurement and analysis of expectations. I also conclude that, short of this, there are useful things that economists can do to understand better college choice, income determination, and the like. Reference Murphy, Kevin, and Finis Welch. 1989. Wage premiums for college graduates: Recent growth and possible explanations. EducationalResearcher 18(4):17-26.

3

Trends in College Entry among Whites, Blacks, and Hispanics Robert M. Hauser

For the past two decades and in the foreseeable future, the key educational transitions among American youth have occurred and will occur during middle to late adolescence. These include, but are not limited to, high school dropout or completion and entry into colleges, universities, or other postsecondary schools. These transitions are keys to the quality and productivity of the future work force because they are the main points at which youth now leave the educational system for work, military service, family formationand, in some cases, street or prison life. For the past several years, public attention in the United States has focused mainly on the first of these transitions-high school dropout. For example, the highly publicized National Goals for Education (U.S. Department of Education 1990) proclaim 90 percent high school completion as one of six primary goals, but they focus less attention on the transition from secondary to postsecondary schooling, which is mentioned as one among several objectives subsidiary to the goal of “adult literacy and lifelong learning.” I Robert M. Hauser is Vilas Research Professor of Sociology and director of the Institute for Research on Poverty at the University of Wisconsin-Madison. This research was supported by grants from the National Institute on Aging, the Spencer Foundation, the Kenneth and Carolyn Brody Foundation, and the William F. Vilas Trust Estate. It was carried out at the University of Wisconsin-Madison, using facilities of the Center for Demography and Ecology, for which core support comes from the National Institute of Child Health and Human Development and the William and Flora Hewlett Foundation, and facilities of the Institute for Research on Poverty, which is supported by a grant from the Office of Assistant Secretary for Planning and Evaluation, U.S. Department of Health and Human Services. The author thanks Linda Jordan, Taissa S. Hauser, Julia Gray, and Yu Xie for assistance in the preparation and documentation of the Uniform October Current Population Survey file, 1968-1988, and he thanks Hanam Samuel Phang for assistance in research. The opinions expressed herein are those of the author. 1. Goal 2 says, “By the year 2000, the high school graduation rate will increase to at least 90 percent,” and adds the objective, “The gap in high school graduation rates between American students from minority backgrounds and their nonminority counterparts will be eliminated.” By

61

62

Robert M. Hauser

The transition from high school completion to whatever may follow is and will be the most important decision point in the American educational system. High school completion is the single point at which the most Americans leave schooling.2 It is the point at which the largest share of the cost of schooling shifts from public to private hands-even though there is massive public funding for postsecondary schooling. It is the point that determines access to the kinds of jobs that are and will be most in demand in the American economy of the twenty-first century. Wage differentials are growing between the college educated and persons with some college or a high school diploma or who are high school dropouts (Murphy and Welch 1989). For example, figures 3.1 and 3.2 show trends since the 1960s in the earnings of black and white male high school graduates and in the earnings of high school graduates relative to men with other levels of completed schooling. After increasing from the middle 1960s to the middle 1970s, the real earnings of male high school graduates declined through the middle 1980s. The earnings of high school dropouts relative to high school graduates also declined. After the middle 1970s, the relative earnings of men with college experience increased. Those for college graduates rose most rapidly, from about 20 percent more than the earnings of high school graduates to 40 or 50 percent more. There is every reason to believe that these differentials are a valid reflection of the growing demand for a highly educated work force, that they will continue (Bishop and Carter 1990), and that they provide sound and compelling evidence of the need to monitor and foster the transition from high school completion to further schooling and the labor market. Were no other factors at work, one might expect the chances (i-e., likelihood) of college entry to follow the trends in the relative earnings of college and high school graduates. In fact, this has roughly been the case for white men, but factors other than wages in the civilian labor market have also influenced trends in college entry. These include changes in social and economic background, in rates of high school completion and the academic performance of graduates, in the size and composition of the armed forces, in the cost of going to college, in the amount and composition of financial aid for college education, and in the social and economic opportunities of minorities and women (Kane 1991b; Hauser, in press). Unfortunately, limits on the coverage

contrast, goal 6 says, “By the year 2000, every adult American will be literate and will possess the knowledge and skills necessary to compete in a global economy and exercise the rights and responsibilities of citizenship.” Among the objectives subsidiary to goal 6 is “the proportion of those qualified students, especially minorities, who enter college; who complete at least two years; and who complete their degree programs will increase substantially.” 2. To be sure, college dropout is also large. Slightly more than half of white college entrants complete 16 years of school by the time they reach ages 25 to 29, and only about one-third of minority entrants complete 16 years of school by ages 25 to 29. However, college dropout occurs over a prolonged period, and it affects only the survivors of the transition from high school to college.

Trends in College Entry

63

21

1.6

1.4 0 .c (d

0: 1.2

-

.*

0

1

.a.

,, - ...', ......

Ii 0

I.

b

......

---.**--

'L _,'

v)

.-Cr

.......*.

........ . . -. 20

-

-

zl

-

\

0

W

16

0.8

0.6

+

15 Q

i

I

I

I

I

I

I

I

l

l

I

I

14

1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 Some Coll~e_v,s.. 12 Years

Less than 1 2 E v s . 12 Years

College, 16 or Mo'eYears, vs. 12 Years Earnings of His_hgtoolGraduates

Fig. 3.1 Earnings of black males age 25-34: Ratios at various education levels, compared to high school graduates, 1964-1988 Note: Data are three-year moving averages from March CPS.

23

1.6

1.4

.-0

- ........

*.*

4-9

(d

[r 1.2

0

-

.-c

pp0-OS~b 'sQ

'*../ d"',.............

...... .......................Y.""

--

.......

........'- 22 m m u 0 c

1- - - - - - * .,__-,-----,&------" - - - - - . -

*.---

u)

E

-..

d

0

c(d

21

'5 2

/

l -

0

(d

W

0.8

- 7---a

0.6

I

I

I

I

I

I

I

I

I

I

I

I

I

19

and detail of federal educational statistics make it impossible to produce a comprehensive account of the influence of these factors; in fact, the data have not been good enough to provide a definitive and timely account of trends in college access of major population groups, such as those defined by race, gender, and family income (Hauser 1991a, 1991b). For example, accounts of

64

Robert M. Hauser

post-1970 trends in the college attendance of blacks and whites range widely: there have been no adverse trends for blacks, or none that could not be explained by the decreasing selectivity of high school graduation (Hu 1991); there was an anomalous upward bubble in black college attendance in the middle 1970s, after which things returned to the level of the late 1960s (Pelavin and Kane 1990; Koretz 1990); after the middle 1970s there was a decline in college enrollment among 18- and 19-year-old blacks, but this was offset by delayed entry into college (Kane 1991a, 1991b); and there was a decline in black college entry after the middle 1970s, from which there has been little or no recovery (Jaynes and Williams 1989; Carter and Wilson 1990; Mortenson 1991; Hauser and Anderson 1991 ; Hauser, in press). In this essay, I take a fresh look at trends since 1972 in college entry by gender among black, Hispanic, and white high school graduates. The analysis is based on a new time series of cross sections from October Current Population Surveys (CPS), 1972 to 1988 (Hauser 1991c), in which the records of high school graduates have been linked to the characteristics of their households and parents. Using these data, I ask to what degree the observed differences and trends in college entry among white, black, and Hispanic men and women can be explained by group differences and trends in social and economic background. I first estimate the basic trends in college entry from the October CPS and compare them with an independent, alternative series from the March CPS. Second, I describe trends in social background and in household residence among whites, blacks, and Hispanics. Third, I estimate levels and trends in college entry, controlling social background, within each racialethnic group, and I assess the importance of social background in the observed trends. Fourth, I estimate a pooled equation in social background across all the groups and use it to compare levels and trends in college entry among the groups. My analyses lead to four major findings and a caution about our ability to monitor future trends in college entry. First, there has been an almost continuous increase in women’s chances of college entry relative to those of men from the early 1970s to the late 1980s. This gain cuts across racial and ethnic lines; it is virtually the same among whites, blacks, and Hispanics. Among dependent high school graduates, women’s chances of college entry have exceeded those of men in every year since 1975. Second, the chances of college entry among white men declined from the early 1970s through 1980, and they subsequently recovered to match the high levels of the Vietnam War era. Thus, the chances of college entry among white men and women have grown to unprecedented levels. Third, blacks’ chances of college entry relative to those of whites rose from 1973 to 1978 and declined thereafter to levels at or below those of the early 1970s. The growth of the 1970s was accelerated by steady improvements in the educational attainmerlts of black parents and by a decline in the size of black families; the decline of blacks’ chances of college entry in the 1980s was moderated by continuing improvements in social back-

65

Trends in College Entry

ground. Together with other available evidence, these findings suggest that net increases in the cost of college attendance are the major factor in the decline of blacks’ chances of college entry. Fourth, when social background is controlled, the chances of college entry among Hispanics exceed those among whites from the early 1970s through the late 1980s, while the chances of college entry among blacks exceed those among whites from the early 1970s through the early 1980s. The decline in college entry among blacks has brought their chances just below those of socially and economically similar whites, while Hispanics’ chances of college entry are consistently much higher than those of comparable whites or black^.^ These findings raise difficult questions about efforts to increase the chances for educational and economic advancement among blacks and Hispanics. Should efforts to increase minority opportunities rest with the achievement of statistical parity? Or should public policy, recognizing disparities of social background as well as uneven rates of advancement in educational, economic, social, and political status, tolerate or even encourage minority advantage in school transitions? Finally, while it is important to continue to monitor trends and differentials in the chances of college entry among men and women and across racial and ethnic groups, a recent change in the content of the October Current Population Survey will substantially reduce the statistical reliability of the data series after 1988. Unless the previous content of the CPS is restored, or alternative data series become available, we will be less able in the future than in the past to monitor year-to-year changes in the transition from high school to college (Hauser 1991b). Racial and ethnic differences are important, both because of their obvious relevance to issues of equity and equality of opportunity and because of their implications for the future American economy. The demographer’s stock in trade is the explanation of differences by population composition. If minorities are less successful in educational transitions than whites, or even if improvements in the status of minorities occur slowly, the growing share of minorities in the American population will itself reduce the educational attainment of the future work force. Much of my analysis focuses on trends in college entry since 1972, the first year in which it was possible to identify Hispanics consistently in the October CPS. As shown in figure 3.3, the share of minorities among high school graduates has grown steadily. From 1972 through 1988, Hispanics grew from 4.3 to 6.5 percent of high school graduates, and blacks grew from 10.3 to 14.3

3. Cameron and Heckman’s discussion (in this book) is irrelevant to any but the last of my conclusions, and it ignores many of the analyses reported herein. Selection into high school graduation obviously affects ethnic differentials in college entry, but it is not a likely source of trends in ethnic differentials during the 1970s and 1980s. Also, Cameron and Heckman offer little evidence that effects of social background on college entry differ between dependent and nondependent graduates.

Robert M. Hauser

66 100

80 Q)

3 60 c

c

Q)

$

40

Q

20

0 1972

1974

1976

1978

1980

White E l Black

1982

1984

1986

1988

f3 Hispanic

Fig. 3.3 Race and ethnicity of recent high school graduates: 1972-1988 Note: Data are three-year moving averages for youth in October CPS.

percent of high school graduates. Obversely, the share of whites (and others) has declined from 85.5 percent to 79.2 p e r ~ e n tThis . ~ change in composition stems in part from increasing rates of high school completion by minority youth, but it is also fed by changes in the racial and ethnic composition of birth cohorts. The shift in population composition will continue. In 1972, Hispanics were 6.7 percent of five- and six-year-olds, and blacks were 14.3 percent of five- and six-year-olds (U.S. Bureau of the Census 1974, table 1). In 1988, Hispanics were 11.2 percent of five- and six-year-olds, and blacks were 15.3 percent of five- and six-year-olds (U.S. Bureau of the Census 1990, table 1). Thus, before the turn of the century, these disadvantaged minorities will constitute about one-quarter of persons reaching adulthood in the United States.

3.1 Data on College Entry The Current Population Survey of the Bureau of the Census is a large national survey of the civilian, noninstitutional population, currently covering 4. Unless otherwise noted, all data reported herein are based on tabulations from public-use versions of the March or October CPS. 5. The percentages of Hispanics and blacks are not additive in published tabulations of the U.S. Bureau of the Census because persons of Hispanic origin may be of any race. However, in the independent analyses reported here, I have given precedence to blacks in order to construct a mutually exclusive and exhaustive racial-ethnic classification. That is, all blacks are classified as black, and Hispanics are all classified as nonblack. Only about 5 percent of Hispanics identify themselves as black in the CPS (U.S. Bureau of the Census 1988). “White” is used throughout to refer to persons who are neither black nor Hispanic.

67

Trends in College Entry

about 55,000 households each month. Each October it fields an educational supplement that ascertains the school enrollment status of persons aged three to thirty-four. Aside from standard labor force and employment variables, the educational supplement ascertains race-ethnicity, sex, age, highest grade of school completed, current grade or year in school, year of high school completion, and school enrollment status in the previous October. The CPS treats children who are living in group quarters while away at school as if they were living in their parents’ households, and it is thus feasible for us to attach the social and economic characteristics of parents and parental households to children who are living in their parents’ households or away at school.6 There are some problems in using the CPS data to measure adolescent educational transitions. The samples become excessively small and statistically unreliable when we try to focus on key transitions, especially among minority groups. Family income is not measured well,’ and academic ability is not measured at all. We lose the link with parents when children leave their parents’ household and do not live in group quarters at school. The CPS does not cover persons in the military or in institutions, such as prisons and jails, that now house a substantial minority of young adults. The CPS tells us little about the schools or colleges in which students are enrolled; we learn only whether enrollment is at a two-year or four-year public or private institution. Other recent content changes have further reduced the usefulness of the October data (Hauser 1991b). At the same time, unlike the institutional or longitudinal surveys of the National Center for Education Statistics, the October CPS does provide annual data on college entry and enrollment. For many years, the October CPS has included a question about the year of high school graduation of persons aged 14 to 24; together with current college enrollment data, this permits a highly focused look at the transition from high school to college. We can ask what share of each year’s high school graduates was enrolled in college in the following October. Most of these graduates are young enough to be dependents at the time of the survey, so their records can be linked with those of their parents. Unlike age-specific rates of college par6. I have created uniformly formatted versions of this file for the years 1968 to 1989 (Hauser 1991~).However, the present analysis covers only the years 1972 to 1988 because Hispanics were not identifiable before 1972, and the 1989 data became available too late to be included. 7. The October CPS family income variable is probably the worst income measure obtained in any major federal statistical program; yet it is the main economic measure used in the measurement of access to postsecondary education. It is a CPS control card item, which means that it is asked of anyone entering the sample for the first time in a calendar year. The item is a single question about family income in the 12 months preceding sample entry, not in a calendar year, and the responses are coded in broad groups. By contrast, the March CPS now ascertains about a dozen specific sources of income in the preceding calendar year, and the Survey of Income and Program Participation (SIPP) ascertains more than 50 sources of income. For this reason, among others, I have given no precedence to family income among the several socioeconomic background variables used in the analysis. I have also introduced two more reliable, long-term measures of economic standing: housing tenure (own versus rent) and household head’s occupational status. The former is a proxy for wealth, while the latter is a proxy for permanent income; neither of these variables was used in Cameron and Heckman’s analysis of the NLSY data.

68

Robert M. Hauser

ticipation, enrollment, or attendance, college entry rates are both timely and specific (Hauser 1991b). One problem with this series is that it is ordinarily based on the experience of a single cohort of high school graduates as reported in a single October CPS (Jaynes and Williams 1989; Mortenson 1990); thus, the number of observations and their statistical reliability are limited. There are typically about 2,100 recent high school graduates in an October CPS, of whom about 200 are black and one hundred are Hispanic. While it is possible to draw valid conclusions when the data are cumulated over a period of years, the data are not reliable in any one year for minority groups or for other similarly small subpopulations. There is a trade-off between timeliness and specificity on one hand and reliability on the other. To increase the statistical reliability of the college entry series, I used a feature of the October design that has recently been dropped. Until 1988, the CPS identified the calendar year of high school graduation for several years preceding the calendar year of the survey. Using this question, plus other questions on highest grade attended and college enrollment in the preceding year, I pooled reports from each year’s CPS about college enrollment in the previous October by the high school graduating class of the preceding year together with the contemporaneous reports about the college enrollment of that year’s class. There are changes in population coverage between the first and second year after high school graduation because some youths leave their parents’ home to join the military, form independent households, or for other reasons. Thus, I estimated trends in college entry using a statistical model that takes the effect of the coverage difference into account.* In 1988 the Census Bureau dropped the detailed responses to the question about year of high school graduation, retaining only the distinction between graduates in the current year and in any previous year. Thus, in future years, it will not be possible to pool observations across years as I have done in the present analysis.

3.2 Rends in College Entry Figure 3.4 shows the college entry series for black, Hispanic, and white men and women from 1972 through 1988.9The estimated proportions enter8. Before pooling the contemporaneous and retrospective reports, I tested for interaction effects between the effects of graduation year and year of report within each racial-ethnic group; there were no statistically significant interactions. Because there is 50 percent overlap in CPS households in the same month from one year to the next, this procedure does not double the precision of the estimates, but it is a substantial improvement. Effects of year of report have been included in all models used in the present analysis, but all reported estimates are normed on contemporaneous reports. 9. The estimates are based upon samples of 6,102 blacks, 2,801 Hispanics, and 50,348 whites (and others, e.g., Asians and American Indians) from the October Current Population Surveys, 1972 to 1988. All of the reported analyses are based upon the logit model for individual observations. Graphical displays of time series were constructed by taking three-year moving averages of predicted logits or of predicted contrasts between logits. The analyses were carried out without

Trends in College Entry

69

0.65 0.6

-W

6 0.55 0

c

'c 0.5 a,

c.

c

w

0.45

C

.-

0

0.4

P

0.35 03

1972

1974

1976

1978

1980

1982

W h E e n Whi!e_tomen Bkck.men Blatk W o p n Hiqx-n.k-men

1984

1986

1988

Hispa.n&-Eomen

Fig. 3.4 College entry by sex and race-ethnicity:Recent high school graduates, 1972-1988 Note: Data are three-year moving averages of model estimates from October CPS.

ing college are based on a logit model that specifies effects of sex, raceethnicity, sex by race-ethnicity, the interactions between year of graduation and race-ethnicity, and the interactions between year of graduation and sex. Although the model includes the two-way interactions among sex and raceethnicity, these terms were not statistically significant (L2 = 3 with 2 dfi. That is, gender differences in college entry were essentially the same among whites, blacks, and Hispanics. Furthermore, the three-way interaction effects among sex, race-ethnicity, and year of graduation were of borderline statistical significance, and these were not included in the model.Io That is, although some trends in college entry differed by gender and some by race-ethnicity, gender differences in the trends were similar within each racial-ethnic group. The absence of interactions or trends in interactions between race-ethnicity

any correction for a sampling design factor. For this reason, the analysis may give undue importance to small differences among time periods or among racial-ethnic groups. Among graduates for whom there were contemporaneous reports of October enrollment, family income refers to the previous 12 to 15 months-that is, the period during which college entry decisions were most likely to have been made. However, among graduates with retrospective enrollment reports, family income pertains to the year after high school graduation. 10. While these effects are nominally statistically significant withp = 0.01 (L2 = 44 with 32 d n , they are not large enough to reject null under a Bayesian information criterion, bic = Lz - df x In(M = - 308, where negative values of bic suggest that the model is acceptable (Raftery 1986). Furthermore, since the data include two observations for each household covered in both the contemporaneous and retrospective reports of college attendance, the chisquare test statistic is too large by a factor of about one-third, even without an adjustment for the sample design factor.

70

Robert M. Hauser

and gender is an important finding, for it disconfirms some highly publicized claims about the distinctive problems of black men. If black men are at a disadvantage with respect to college entry, it is because, like black women, they are black and, like white men, they are men; there is no unique effect on college entry nor a unique trend associated with being both black and male.ll Figure 3.4 shows distinct trends for blacks, whites, and Hispanics, along with distinct differences from those trends between men and women of each racial-ethnic group. Whites of both sexes enjoyed consistently higher chances of college entry than any other group, except that college entry chances of white women were less than those of Hispanic men early in the 1970s. College entry chances of white men declined throughout the 1970s but recovered dramatically after 1980, when the chances of college entry among white men and women rose in parallel. College entry chances of Hispanics peaked in the middle 1970s and have been essentially stable since then, possibly excepting a recent increase. College entry among blacks rose during the 1970s but declined from 1979 to 1983, after which they may have recovered some of the earlier loss; however, through most of the period, the college entry chances of blacks have been less than those of whites or Hispanics. Even though the series in figure 3.4 are based on a constrained model, it is difficult to follow as many as six trend lines. Figure 3.5 shows the college entry series for white men on a logarithmic scale, along with three other series that document major effects on the trends: women versus men, blacks versus whites, and Hispanics versus whites. I 2 That is, the male-female comparison holds for whites, blacks, and Hispanics, while the black-white and Hispanicwhite comparisons hold for men and women. The college entry chances of white men declined in the last half of the 1970s and then rose by 1988 to a peak above that of the 1970s. In the mid-l970s, about 53 percent of white men entered college; the college entry rate dropped to 50 percent by 1980 but increased to 60 percent in 1988. The series shows growth in black college entry chances relative to those of whites during the 1970s, with a peak late in the decade. At the peak, the college-going chances of blacks were almost equal to those of whites. But the peak was followed by an equally rapid decline that lasted through the first half of the 1980s. Hispanic enrollment chances follow those of whites more closely than do those of blacks. Hispanic chances of college entry converged upward toward parity with those of whites by the middle 1970s. After this peak, they declined to a level about 5 percentage points less than whites until the middle 198Os, after which the series appears to fluctuate unreliably. Women’s chances of college entry grew steadily relative to those of men from the early 1970s through 1983; in the early 1980s, women enjoyed greater chances of college entry than men. After 1983, wom11. Similarly, there is no unique advantage in being both white and female; compare Koretz (1990). 12. Although figure 3.5 shows series on two different scales, the metric and range of each scale are the same.

Trends in College Entry

71

0

2.

c

; w

-0.2

9

8

J

-0.4

0

v

w Y=

-0.6

1972

1

I

I

I

I

I

1

I

1974

1976

1978

1980

1982

1984

1986

1988

-

Black vs. white Hispanic vs. white Women vs. men Whit men

.---

.......

-t+,

Fig. 3.5 Effects of sex and race-ethnicity on college entry: Recent high school graduates, 1972-1988 Note: Data are three-year moving averages of model estimates from October CPS.

en's college entry chances declined to a level slightly below those of men, but they reached a new peak later in the decade. These series do not measure whether individuals ever entered college but only whether they entered in the fall after high school graduation; this is a weakness of the CPS since, as Kane (1991a, 1991b) and Hauser (1991b) have shown, the black-white gap in college entry closes to some degree through delayed enrollment. If disadvantaged groups delay college entry, then a decline in the initial transition from high school to college need not lead to a decline in the chance of ever entering college. On the other hand, the costs of delayed or prolonged schooling are real and should not be ignored. It is also difficult to come up with a later measure of college entry that does not confound the effects of delayed entry with those of prolonged or part-time attendance (Hauser 1991b), and it is impossible to maintain the CPS linkage between youth and their parents' households much beyond the completion of high school. Fortunately, we can check the findings in figure 3.5 against an entirely independent and cumulative measure of college attendance. Figure 3.6 shows time series from the March CPS, corresponding to those in figure 3.5 but based on the share of high school graduates who ever enrolled in college by ages 21 to 24. The share of men entering college is less here than in the series in figure 3.5, and the college-going chances of blacks and Hispanics relative to whites are also less. These differences may be attributable to the inclusion of reported high school graduates who completed the 12th grade late or earned a high school equivalency. Such persons are less likely than on-time graduates

72

Robert M. Hauser 0.20

,.....

* 0.00

***'

c

0

- 0.6 $

rn

w

-

'E

0.4

-

g H -0.60 -0.80

1

I

I

I

I

5 C

g. -0- 8 -0-

5al

z . ?'Yo *iy 0.2

I

-0.2

'

I

to enter college, and there are relatively more of them among blacks and Hispanics than among whites. Aside from these differences, there is substantial similarity between the shapes of the trends shown in figures 3.5 and 3.6, although those shown in figure 3.6 are smoother because of the four-year width of each age cohort. In figure 3.6 as in figure 3.5, there is a fall and rise of college attendance among white men, a rise and fall among blacks relative to whites, and sustained growth among women relative to men. There is less similarity between the immediate and cumulative series for Hispanics, whose college chances relative to whites follow an irregular downward trend. It is not clear whether the Hispanic series are simply unreliable or whether there are real differences in trend between immediate and cumulative college entry.

3.3 lkends in Social Background Social background composition is a highly significant source of group differences and of historic changes in educational attainment (Hauser and Featherman 1976; Mare 1979; Kane 1991b). In this section of the analysis, I describe differentials and trends in the social background characteristics of high school graduates that are obtained in the October CPS; in the next section, I present estimates of the effects of these characteristics on college entry. These variables include geographic location, age, household structure, and parental education, occupation, income, and housing tenure. Geographic location and age have been measured for all of the graduates, without regard to their residence in parental households (hereafter, dependency). I have ignored nonde-

73

Trends in College Entry

pendent graduates in describing the other background characteristics. Note that the population is high school graduates, not all youth, and that high school graduation is differentially selective among whites, blacks, and Hispanics; that is, in recent years, graduates represent about 90 percent of whites, 80 percent of blacks, and 60 percent of Hispanics (Frase 1989). When I present trends in social background (in figures 3.7 through 3.12), I present time series for racial-ethnic groups only; but when I show the effect of these trends on college entry (in figures 3.13 through 3.15), I present time series for racialethnic groups by gender. One can assume that, with minor exceptions, the social background of men and women in any racial or ethnic group is the same, and that is why I have not shown trends in social background by sex. We cannot expect gender differences in social background composition to explain gender differences in entry to college because there has been very little difference in the selectivity of high school graduation by gender. 3.3.1 Metropolitan Location Metropolitan location may indicate proximity to institutions of higher education, differences in the quality of schooling, and access to labor market opportunities that compete with college entry. Figure 3.7 shows the distribution of white, black, and Hispanic graduates by metropolitan location. For this analysis, the 17 largest Standard Metropolitan Statistical Areas in 1970 were designated as metropolitan. Graduates were classified as residents of the central cities or suburban rings of those areas or as “other,” even though they may have lived in smaller metropolitan areas. White graduates became less likely to live in the large metropolitan areas between 1972 and 1988, but there appear to have been no reliable changes in the metropolitan location of blacks or Hispanics. l 3 Blacks and Hispanics were far more likely than whites to live in the large metropolitan areas, and within those areas, they were more likely to live in central cities than in suburban rings. 3.3.2 Regional Location Regional location may affect access to higher education through differential access to low-cost public colleges or community colleges, and regional differences in access may also differ among racial-ethnic groups, as in the case of the traditionally black colleges in the South. Figure 3.8 shows no substantial shifts in the regional location of white, black, or Hispanic graduates from the 1970s through the 1980s, but there are consistent regional differences in the location of the three groups. Whites are more evenly distributed across the four major census regions, while blacks remain highly concentrated in the South and underrepresented in the West. Hispanics are highly concentrated in the West and South and are underrepresented in the North. 13. There is anomalous variation in several of the series for Hispanics, and I have ignored it throughout as a result of the small numbers of Hispanic graduates covered in the CPS.

74

Robert M.Hauser

Whites 100

80

8,

d c

a,

g a

n

20 n

1972

1974

1976

1978

1880

1882

1QM

1988

1-

1982

19&(

1986

1988

1982

1964

1988

1888

Blacks

'is72

1974

1976

1978

lQ80

HisDanics 100

80

a,

p" a,

ea , Q

n

20

7972

1974

1978

1978

lee0

ElRing EEentral City EWther Fig. 3.7 'Ikends in metropolitan location by race-ethnicity: Recent high school graduates, 1972-1988 Nore: Data are three-year moving averages of residence in 17 largest metropolitan areas among youth in October CPS.

75

Trends in College Entry

Whites 100

80

a

P" 0 0

b*o

a

20

0

is72

1974

1076

1978

1080

1982

1984

1986

1988

1082

1984

1983

1088

Blacks 100

80

p" 0)

0)

$40

a

20

0 1072

1074

1978

1978

1980

Hispanics

@EastHNorth aSouth m e s t Fig. 3.8 Rends in regional location by race-ethnicity:Recent high school graduates, 1972-1 988 Note: Data are three-year moving averages for youth in October CPS.

76

Robert M. Hauser

3.3.3 Age of Graduates

Among whites the age at high school completion increased slightly from 1972 to 1988. Later graduation may lower the chances of college entry because it is a proxy for previous academic problems or for greater family responsibility, economic independence, or labor market opportunity. As shown in figure 3.9, the share of white graduates less than 18 years old has declined steadily, while the share aged 19 or more has increased. This may be a result both of increasing school retention and of increased reporting of high school equivalency as graduation. There are fewer signs of reliable trends among blacks or Hispanics, but the age at high school completion is higher among the minority groups. Even in 1988 about 20 percent of white high schi-d graduates were 19 years or older, but during the 1970s and 1980s roughly 25 percent of black graduates and 30 percent of Hispanic graduates were that old. 3.3.4 Household Structure For consistency in the analysis, I arranged the household data for dependent children so there was always a record for a household head. If there were a householder and spouse present, I defined the male as the household head and the female as the spouse of head. In the few cases where there was a male householder but no spouse, the male was also defined as the head; if no male householder or spouse was present, I defined the female as the head. Thus, by construction, there were no data for the spouse of head in single-parent households. Figure 3.10 shows time series in three measures of household structure: the share of female-headed households, the share of household heads without occupations, and the mean number of children in the household. The share of white graduates living in female-headed households grew from about 15 to 20 percent from 1972 to 1988, while that of Hispanics grew from about 20 percent to more than 30 percent.I4 Black graduates were far more likely than whites or Hispanics to live in single-parent households; their share of femaleheaded households grew from 40 percent to more than 50 percent between 1972 and 1980, and it appears to have been stable thereafter. Only about 7.5 percent of the household heads of white graduates were without occupations as reported in the CPS, while nearly one-quarter of black household heads were without occupations. The share of Hispanic heads without occupations was between that of whites and blacks. The share of white heads without occupations was essentially stable from the 1970s through the 1980s, but that among blacks was appreciably higher from the late 1970s through the early 1980s. The number of children in the household included persons 18 years or less in age, plus the reference person if he or she was 19 or older. This is a proxy 14. I doubt the reliability of the rapid increase in female-headed households among Hispanic graduates in the late 1980s.

77

Trends in College Entry

Whites

1972

1074

1976

1978

1 W

1982

1084

1086

1988

1882

1984

1986

1988

Blacks

Hispanics 100

80 0

2" Q)

g a a 20

1972

1974

1976

1978

1 W

HLess than 18 Ell 8 El1 9 B O or more Fig. 3.9 'kends in age of graduates by race-ethnicity: Recent high school graduates, 1972-1988 Note: Data are three-year moving averages for youth in October CPS.

78

Robert M. Hauser

0.6 v)

U

2a 0.5 -aa

$ 0.4

**----*--*.*.---*

*/---.a/.-*-*

LL

s3 0.3 c 0 .-

g 0.2 -........".

r

.'

..........................................

e

a 0.1

I

1

I

I

--

0.25

3

3

0.2

z! 0

c 0)

w

0.15

3

P6

0.1

E

.-

5n 0.05

-

**

) - - o m - - -

I

I

.. .. .... ...

........

- .'....*. ............ *.

.....

*

/

0.

-

-

I &

- - '*

*. ' * - - O M

)----#

. -

1

.... *

2 n

o

1

1

I

I

1

I

I

1

I

3.5 .o--

-*

# #

c

F

3

9 .-

6 5 t n

2 . : .

*

. 2.5

......,.. - *.

*

*.

-

*.

.'.

...............:+.- ..:: *".. '- 2 . 9.

g.4.

,

..

\

\

E

3

2 2 -

15

I

I

I

I

I

I

I

I

Fig. 3.10 Trends in household structure by race-ethnicity: Recent high school graduates, 1972-1988 Note: Data are three-year moving averages for dependent youth in October CPS.

79

Trends in College Entry

measure of the size of the sibship, which is known to lower educational chances, but it excludes older siblings and those who have left the parental household.I5 Excepting a brief rise among blacks from the early to middle 1 9 7 0 ~this ~ measure of demand for household resources decreased steadily among all three racial-ethnic groups from the 1970s through the 1980s. The number of children was consistently lower among whites than among blacks or Hispanics. The number of children was initially greater among blacks than among Hispanics, but those two series converged early in the 1980s.

3.3.5 Parental Schooling Figure 3.11 shows trends in the completed schooling of parents of high school graduates. Consistent with the long-term secular growth of schooling, there are steady increases in the average levels of school completion among the fathers and mothers of white and black high school graduates.I6 There is much less evidence of a trend among the parents of graduates of Hispanic origin, and the educational attainments of Hispanic parents are consistently less than those of black or white parents. One reason for the lesser growth in schooling among parents of Hispanic graduates, and for some of the other distinctive characteristics of the Hispanics in this population, may be the contribution of new immigrants. Unfortunately, the CPS data do not include a measure of nativity or of date of immigration, so it is not possible to distinguish trends in the characteristics of the native population from those of immigrant groups.

3.3.6 Parental Socioeconomic Status Figure 3.12 shows trends among the parents of high school graduates in four measures of socioeconomic status: occupational status (on the Duncan [ 19611 socioeconomic index [SEI] for occupation^),^^ heads with farm occupations, annual household income, and home ownership. Father's occupational status affects children's educational and occupational chances, and among whites its influence on postsecondary schooling is about as large as that of parents' income or education (Sewell and Hauser 1975; Featherman and Hauser 1978). Mean occupational status of the household head was consistently higher among whites than among blacks or Hispanics. The mean for whites was around 40 to 45 points-roughly the level of self-employed cabdrivers, electricians, construction supervisors, and policemen-while that for 15. Thus, some dependent college students in the household were not counted. 16. Also, see Hauser and Featherman (1976) and Hauser and Anderson (1991). 17. The SEI is a weighted average of occupational education and income in which the weights have been chosen to predict a survey-based measure of occupational social standing. The version of the scale used here was constructed for 1970-basis census occupational titles by Stevens and Featherman (l981), using characteristicsof occupations in the 1970 census and prestige measures obtained by Siege1 (1971) from surveys by the National Opinion Research Center in the mid1960s. It was updated for 1980-basis census occupation titles by Stevens and Cho (1985).

Robert M. Hauser

80

Mean Years of Schooling of Household Heads 14

I

13 0,

.c -

12

5

11

0 0

-cn

0 10

r (d

a

>

9 8

--- .--_.

I-

.......

-.

1972

1

I

I

I

I

I

I

I

1974

1976

1978

1980

1982

1984

1986

1988

Mean Years of Schooling of Mothers 14

I

L

-

-

c - 0 - O

.

-

___---.-----

- 0

_ _ _ _ _ _ L - - - - - - - _ _ _ c _ - - - -

...... ............................................ - . ... .....................

....

1972

1974

1976

1978

1980

1982

1984

1986

1988

White - Black Hispanic .... Fig. 3.11 'lkends in parental schooling by race-ethnicity: Recent high school graduates, 1972-1988 Note: Data are three-year moving averages for dependent youth in October CPS.

blacks and Hispanics was about 25 to 30 points-roughly the level of office messengers, brickmasons, plasterers, ushers, and oil refinery laborers. There was no consistent evidence of a trend in status among Hispanics, but that of blacks and of whites grew from the 1970s through the 1980s. The prevalence of farm occupations declined among white and black household heads, but it fluctuated wildly among Hispanics. Historically, farm youth obtain less schooling than persons with otherwise similar social and economic background. However, the negative effects of farm background have gradually declined among blacks and whites, and the effect of farm background on edu-

81

Trends in College Entry

Mean Occupational Status of Household Head

Farm Oocupalion of Househdd Head 0.08

0 07

0'01

I

1

t 1972 1974 1978 1978 lae0 1882 1-

1 9 0

HouseholdsIn Owner-OcMlpied Housing

' I

1972 1974 1976 1078 I s 0 1882 1-

Whhte Mask Hispanic

1 M 18(

WhJe B&k HisEnic

Fig. 3.12 *ends in socioeconomic status by race-ethnicity: Recent high school graduates, 1972-1988 Note: Data are three-year moving averages for dependent youth in October CPS.

cational attainment is positive for white men born after the middle 1930s (Hauser and Featherman ,1976, 109, 114).18Thus, in recent cohorts, one might expect the decline in farm origins to reduce chances of college entry. 18. The effects of farm background depend on the position of farm occupations in the Duncan scale, which reflects the low levels of schooling and income among farmers. Farm occupations rate lower in socioeconomic status than they do in prestige, as determined by popular ratings of occupational standing. However, the changing influence of farm background on schooling is not an artifact of changes in the position of farmers in the Duncan scale. Farm youth used to fare even

82

Robert M. Hauser

Household income is a key variable in economic analyses of college entry, as well as a trigger of eligibility for financial aid. However, previous research has shown that parents’ incomes are by no means dominant among the socioeconomic variables affecting chances of postsecondary schooling (Sewell and Hauser 1975; Hauser, Tsai, and Sewell 1983). The CPS household income item is not of high quality (Hauser 1991b), and this adds to my doubts about its significance in the present analysis. There were great differences in household incomes among whites, blacks, and Hispanics; figure 3.12 shows that in constant 1988 dollars white families earned about $30,000, Hispanic families about $16,000, and black families about $13,000. White household income declined slightly from 1972 to 1975, rose through 1978, and declined again through 1983, after which it rose almost to the level of the early 1970s. Black household income declined from the mid-1970s through 1983, after which it appears to have grown sharply back to the levels of the early 1970s. Home ownership is a crude measure of wealth; thus, we would expect it to increase the chances of college entry. The racial-ethnic differentials in home ownership are similar to those in household income. Nearly 90 percent of white high school graduates came from families in owner-occupied housing, compared to about 60 percent of blacks and 70 percent of Hispanics. There is no evidence of a trend in home ownership among the families of white graduates. It may have declined among blacks and Hispanics from the middle 1970s through the middle 1980s, but some of the year-to-year fluctuations appear too large to be credible. It may be useful to summarize the major differentials and trends in social background. As expected, white graduates are better off than black or Hispanic graduates in almost every way: they are younger; they are less likely to come from single-parent households or households without an employed head; they come from households with fewer children; and their parents are better educated, hold higher-status jobs, make more money, and are more likely to own their own homes. It is less easy to characterize differentials between blacks and Hispanics. Blacks are more likely to come from singleparent households or households without an employed head; their household incomes are lower; and their rates of home ownership are less than those of Hispanics (except in 1987 and 1988). On the other hand, black parents have more schooling than the parents of Hispanic graduates. Among all three racial-ethnic groups, the prevalence of female-headed households increased during the 1970s and 1980s, while the number of children in the household decreased. Among white and black graduates, parental schooling and occupational status increased, while farm background decreased; but these trends appear not to have occurred among Hispanics. There is no uniform trend to-

worse than one would expect from their fathers’ low level of occupational status, and they now fare better.

83

Trends in College Entry

ward improvement or deterioration in social background among any of the racial-ethnic groups, but one might expect, given the importance of parental schooling and family size, that there was a general improvement in the predisposing conditions of family background for college entry among whites and blacks.

3.4 Social Background and College Entry In order to measure the influence of social background on college entry, I first carried out separate analyses for white, black, and Hispanic graduates. The estimates are based on a logistic regression equation that also includes effects of year of graduation and of the year of the survey report but no interactions between gender and the year of the survey report. Recall that characteristics of the household and its members, other than the reference person, were treated as missing for all graduates who were not classified as dependents. Estimated effects of variables other than age, sex, and regional and metropolitan location pertain only to dependent graduates, and those effects could be somewhat different among all graduates. Within each racial-ethnic group, I recoded the characteristics of nondependents at the mean values of the variables for dependent graduates. Thus, the estimated effects of nondependency contrast the college entry of nondependent graduates with the college entry of the average dependent graduate in that racial-ethnic group. l9 There were also some households for which income was not reported, some heads without occupations, and a large number of female-headed households where, by construction, there were no data for spouse’s education. In these cases, I recoded the missing cases at the mean values for nonmissing cases in the racial-ethnic group and introduced a dummy variable for the cases with missing data. Thus, within each racial-ethnic group, the dummy variable for female-headed household contrasts the college-going chances of graduates from female-headed households with those of graduates from two-parent 19. Obviously, the data for nondependents add no information to the models about the slopes of variables for which those data were missing, but they do add information about the effects of sex, race, age, dependency status, regional and metropolitan location, and year of high school graduation. This use of data for nondependent graduates is, unfortunately, ignored in Cameron and Heckman’s commentary. I looked for interactions between year of report and the effect of each social background variable in each racial-ethnic group. Different effects could result from differences in dependency status between the year of high school graduation and the following year, when dependency is less prevalent, or they could result from differences in the temporal referent of the background variable (e.g., to the year of high school graduation or the following year). In a global test, none of the slope differences is statistically significant. The effect of family income is significantly less in the second than in the first year among blacks but not among whites, nor are slope differences of housing tenure or occupational status significant among blacks or whites. These tests give very little support to Cameron and Heckman’s findings about dependency status and the effects of family income in the NLSY.As in my analysis, Cameron and Heckman find no appearance of slope differences for any variable other than family income and they do not report a test of the significance of the observed differences.

84

Robert M. Hauser

households whose mothers had completed the average level of schooling among mothers in such households.20 Table 3.1 shows estimated effects of social background and dependency status on college entry in each racial-ethnic group. Dependent women are more likely than men to enter college in all three groups, and the effect is largest among blacks and smallest among Hispanics; however, the interaction effect of sex with race-ethnicity is not statistically significant. Nondependent youth are much less likely to enter college than dependent youth, and in every group this effect is even larger among women than among men.21The interpretation of these effects is problematic because nondependency may be an effect, rather than a cause, of college entry. Graduates may be living in independent households because they have not entered college, rather than not attending college because they are no longer dependents. This does not affect other estimates in table 3.1, except in the cases of age and of regional and metropolitan location, because the other estimates pertain only to dependent graduates. However, dependency status does affect the overall comparison of male and female college entry chances, and it could affect the comparisons among whites, blacks, and Hispanics. Thus, in a later section of the analysis, I compare trends and differentials in college entry between men and women and among the racial and ethnic groups in the full model with trends and differentials under simpler specifications of the association of dependency status, sex, and race-ethnicity with college entry. The effects of central city and nonmetropolitan residence are expressed as deviations from college entry among suburban residents in large metropolitan areas. Central city residence increases the chances of college entry among the small minority of white youth from central cities (compare figure 3.7) but not among black or Hispanic graduates. The positive effects among blacks and Hispanics are not statistically significant, but they are also not significantly different from the effects among whites. Among all groups, college-going chances are slightly lower outside the large metropolitan areas than in their suburban rings, and there are no significant differences among the groups in these effects. The effects of regional location are expressed as deviations from the chances of college entry in the East. Here, there is substantial heterogeneity among racial-ethnic groups that deserves more detailed study. Whites' college entry chances are similar in the South and East, but they are better in the West 20. For nondependent graduates, the dummy variables for missing data were assigned the arithmetic means of those variables among dependent youth. 21. Because the model interacts sex with dependency status, the main effect of sex pertains to dependent women. For example, among white dependents, the effect of being female is an increase of 0.161 in the log-odds of college entry. Being nondependent reduces the log-odds of entering college by 1.060,and being a nondependent female reduces the log-odds by an additional 0.629. Thus, among white dependents, women have an advantage of 0.161 over men in the logodds of college entry; among white nondependents, women have a disadvantage of (0.161 - 0.629 = ) -0.468.

85 Table 3.1

Trends in College Entry Effects of Sex, Nondependency, and Social Background on College Entry: White, Black, and Hispanic High School Graduates White

Black

Hispanic

Variable

Effect

SE*

Effect

SE

Effect

SE

Sex (female = 1, male = 0)

0.161

0.022

0.243

0.059

0.032

0.088

Nondependency Nondependent Female nondependent

- 1.060 -0.629

0.055 0.054

-0.195 -0.736

0.190 0.190

-0.278 -0.479

0.220 0.220

Metropolitan location (relative to suburban ring) Central city Not large metropolitan

0.186 -0.133

0.047 0.027

0.045 -0.137

0.108

0.106

0.194 -0.053

0.150 0.123

Region (relative to East) North South West

-0.108 -0.041 0.119

0.027 0.028 0.032

-0.271 -0.134 0.264

0.092 0.086 0.113

-0.514 -0.265 -0.308

0.181 0.142 0.133

Age Female-headed household Head without occupation Children in household Head’s education Spouse’s education Head’s occupational status Head’s farm occupation Family income not reported Family income (log) Housing tenure (own = 1, rent = 0)

-0.183 -0.170 0.066 -0.076 0.158 0.145 0.133 0.414 0.038 0.172 0.343

0.010 0.031 0.044 0.009 0.005 0.007 0.007 0.059 0.041 0.020 0.035

-0.127 -0.113 0.042 -0.133 0.059 0.087 0.133 0.094 -0.102 0.066 0.360

0.022 0.068 0.080 0.017

-0.116 0.061 0.045 -0.045 0.063 0.005 0.158 0.385 -0.368 0.203 0.083

0.027 0. I14 0.143 0.029 0.015 0.017 0.035 0.233 0.200 0.074 0.106

Sample size

50,348

6,102

0.011

0.017 0.024 0.229 0.130 0.047 0.067

2,801

Note: Excepting race-ethnicity, sex, age, regional and metropolitan location, and dependency status, all

variables pertain only to dependent graduates. The effect of occupational status is reported for a unit of ten points on the Duncan SEl. Dummy variables for year of high school graduation and for retrospective versus contemporaneous reports are also included in each equation. *Standard error.

and worse in the North than in the East. Blacks’ college entry chances are better in the West than in the East, but they are worse in the North and South than on either coast. Hispanics’ college entry chances are similar in the South, and West, where they are somewhat worse than in the East, and they are worse in the North than in any other region. Thus, all groups fare less well in the North and South relative to the East, while whites and blacks fare better in the West than in the East, and Hispanics fare worse in the West than in the East. In general, the regional differences appear to be larger among minority than

86

Robert M. Hauser

among majority populations; among Hispanics this could be a consequence of regional differences in the origin of the Hispanic population. However, these effects are large, and their interpretation is not obvious. This is one of the reasons I have examined trends in entry separately for each racial-ethnic group before attempting to compare overall trends and differentials. There are highly significant negative effects of age on college entry within each racial-ethnic group, and the effects are significantly less among blacks and Hispanics than among whites. That is, although late graduation is an obstacle to college entry, it is less so among the less privileged groups, where late graduation is more common. This suggests that late graduation may have more heterogeneous sources among minorities-for example, late school entry rather than grade retardation. Another possibility is that the greater prevalence of late graduation among minority groups may make it less of a handicap than it is among majority whites. Differences in age at high school graduation may account for part of the difference between whites and minorities in college entry, and the increasing age at high school graduation among whites may have slowed the growth of college entry. Residence in a female-headed household has a statistically significant negative effect on college entry among whites, but its effect among blacks, though negative, is smaller than that among whites and is not statistically significant. Among Hispanics, the effect of living in a female-headed household is actually positive, but it is not statistically significant. The negative effect among whites is less than the effect of a single year of age and barely larger than the effect of sex. Thus, net of other family and household characteristics, the effect of residence in a female-headed household is moderate in comparison with that of other factors influencing college entry. We do not expect blackwhite differences in intact family to contribute substantially to the black-white differential in college entry, nor do we expect that the increase in femaleheaded households substantially reduced growth in college entry. In none of the racial-ethnic groups does college entry depend on whether the household head has an occupation. The number of children in the household has a significant negative effect on college entry among blacks and whites but not among Hispanics. The effect is also significantly stronger among blacks than among whites or Hispanics. Thus, we would expect the black-white difference in number of children to help explain the difference in college entry, and we would expect the decline in number of children (figure 3.10) to contribute to growth in college entry among blacks and whites. Among whites, both education of head and education of spouse of head have the expected large positive effects on chances of college entry. The effects are only about half as large among blacks, but they are still highly significant statistically. Among Hispanics, the effect of head’s education is similar to that among blacks, but there is no significant effect of spouse’s education. Given the large educational differences among heads of white,

87

Trends in College Entry

black, and Hispanic households and the steady growth of schooling among heads and spouses in white and black households (figure 3.1 l), we expect that differentials and trends in parental schooling will contribute substantially to the explanation of racial-ethnic differentials in college entry and that they will contribute to growth in college entry among whites and blacks. Head’s occupational status has similarly large and highly significant effects on college entry in all three racial-ethnic groups. For example, among whites the effect of a 10-point increase in occupational status on the Duncan scale, 0.133, is similar to that of a one-year increase in mother’s educational attainment, 0.145. Among blacks and Hispanics, the effect of a 10-point increase in occupational status is larger than that of a year of schooling of either parent. Thus, we expect majority-minority differences in household head’s occupational status (figure 3.12) to help explain racial-ethnic differentials in college entry, and we expect growth in head’s occupational status among blacks and whites to contribute to growth in college entry. Having a household head with a farm occupation contributes positively to the chances of college entry among whites. Since the effect of a farm occupation, 0.414, is just about three times that of a 10-point difference in occupational status on the Duncan scale, 0.133, we can say that the effect on college entry of being a farm son or daughter is equivalent to a 30-point increase in the occupational status of farmers, given their placement on the scale. The estimated effect of head’s farm occupation is less than one-quarter as large among blacks as among whites, but it is about the same among Hispanics as among whites; however, none of these intergroup differences is statistically significant. Since the share of households with farm heads is small in each racial-ethnic group and the differences among the groups are not consistent (figure 3.12), we do not expect the effect of farm occupations to explain racial-ethnic differences in college entry. However, the steady decline in heads with farm occupations among whites must contribute a modest negative component to their trend in college entry. Family income has a large and significant effect on college entry among whites and Hispanics but not among blacks, and the effect among blacks is significantly lower than among whites.22 Thus, one might expect whiteHispanic differences in family income to account for a positive difference between their chances of college entry. It is not clear how one ought to interpret the absence of a family income effect among blacks. One way to put the matter is that public policy, perhaps among other factors, has eliminated income as a barrier to college entry among blacks, though not among whites or Hispanics, even though it has not eliminated the influence of other background factors that impede college entry-for example, parental schooling, occupational 22. There is no significant effect of missing data on family income among any of the three racial-ethnic groups. That is, children from households that did not answer the family income question had chances of college entry that were similar to those of households with average levels of family income.

88

Robert M. Hauser

standing, and the presence of other children in the household. On the other hand, one might say that the parents of black children, unlike those of white (or Hispanic) children, are unable to improve the college entry chances of their offspring by earning more money. This is similar to the perverse form of equality of opportunity for intergenerational occupational mobility that was experienced by black men in the 1960s (Duncan 1968).23The family incomes of white graduates declined from the late 1970s through the early 1980s (figure 3.12), so we expect family income to have contributed modestly to the observed trend in college entry among whites. Among blacks, family income declined from the early 1970s through the early 1980s and then recovered, but the effect of family income is so small that we do not expect it to contribute significantly to the observed trend. Home ownership has similarly large and significant effects on college entry among whites and blacks but not among Hispanics. Since black households are less likely than white households to live in owner-occupied dwellings (figure 3.12), we expect the difference in home ownership to contribute to the black-white difference in college entry. However, since there is no reliable trend in home ownership among the racial-ethnic groups, we do not expect it to contribute to the trend in college entry. In summary, among background characteristics, only sex, dependency status, age, head’s educational attainment, and head’s occupational status have consistently significant effects on college entry among whites, blacks, and Hispanics. If we disregard the erratic estimates among Hispanic graduates and consider blacks and whites alone, we can add number of children in the household, spouse’s educational attainment, and home ownership to the set of variables consistently affecting college entry. Differentials in social background on each of these variables may contribute to racial-ethnic differentials in college entry. Positive trends in head’s and spouse’s educational attainment and head’s occupational status, along with the decline in number of children, probably contributed to growth in college entry from the 1970s through the 1980s. At the same time, increases in age at high school graduation and in female headship (among whites) may have depressed chances of college entry. In the next section, we examine the overall effect of changes in social background on the college entry of whites, blacks, and Hispanics.

3.5 lkends in College Entry with Background Controlled For each racial-ethnic group, I estimated two logistic regression equations for college entry. The first equation included only the effects of year of report, sex, year of high school graduation, and the interaction of sex with year of graduation. Estimates from this equation give the logs of the odds-ratio of 23. There are large income differentials in college entry among all three racial-ethnic groups, but they disappear among blacks when other background variables are controlled.

89

Trends in College Entry

college entry for men and women in each year. The second equation includes the same variables as the first but adds effects of the social background variables and nondependency and the interaction effects of nondependency with sex. Thus, the effects of social background, but not those of nondependency, were assumed to be equal for men and women within each ethnic group. With any fixed configuration of social background, predictions from this equation estimate the trend in college entry net of changes in social background and dependency. Using these two equations, I estimated two components of the trend in college entry for each sex and racial-ethnic group. The first component is the trend in college entry net of social background and dependency, given by the effects of year of high school graduation and of the year by sex interactions in the second equation. The second component is the trend in college entry predicted from changes in social background and dependency, which is given by the differences between the corresponding effects of year of high school graduation for each sex in the two equations. Figure 3.13 shows the components of trend in college entry among white men and women. The upper panel shows the trends net of social background and dependency status, and the lower panel shows the trends predicted from social b a ~ k g r o u n d The . ~ ~ graph is constructed so the sex difference in college entry pertains to dependent high school graduates, while the estimated sexspecific trends pertain to both dependents and nondependents. From 1972 through 1975, dependent white women and men had essentially the same chances of college entry. From 1975 to 1980, white men’s chances of college entry declined while women’s chances were stable, but men’s and women’s chances of college entry rose almost in parallel thereafter. As shown later in this paper, the difference in men’s and women’s chances of college entry is largely a result of the differing effects of dependency on their college entry chances: dependent women are more likely to enter college than dependent men, while nondependent women are much less likely to enter college than nondependent men. As shown in the lower panel of figure 3.13, the overall effect of changes in social background and dependency status was an almost linear growth in college entry among white men and women. Thus, the observed decline in college entry of whites in the late 1970s (figure 3.4) was less sharp than it would have been in the absence of changes in social background and dependency, while the observed growth in their college entry after 1980 was faster than it would otherwise have been. As shown in the upper panel of figure 3.14, the chances of college entry among black men and women diverged in the middle 1970s, just as they did among whites, but in the case of blacks the divergence may have been created by rapid growth in women’s chances of college entry. After 1977 there were 24. These estimates are conditional on the assumptions that effects of social background are the same for men and women except in the case of dependency status, that the effects of social background and dependency status are constant from 1972 through 1988, and that trends in college entry among men and women do not differ by dependency status.

Robert M. Hauser

90

Trend Net of Social Background and Dependency Status

t

-2.E

2 -2.8

4.

5 - -3.0

0

/

Q)

s3

-

-3.2

v)

c 4

%/’

/

....................

..a.

....

U U

0 -3.4 6, 0

-3.6

1972

1974

1976

1978

1980

1982

1984

1986

1988

Trend Predicted from Social Background and Dependency Status

I

............

........ ...........

a, U J

= a, 3.4 0

0

:

. I -

/

0

0

0

- \ 4 4

3.2

m

Bg

-1

3 2.8

a

1972

1974

1976

1978

1980

1982

1984

1986

1988

Fig. 3.13 Components of trend in college entry among whites: Recent high school graduates, 1972-1988 Note: Data are three-year moving averages from October CPS.

parallel declines in the college entry chances of black men and women through 1983, after which the chances of college entry may have improved.25 25. The separate estimates for black women and men in 1972 and in 1988 are probably too unstable to permit any firm statements about trends; recall that there are no contemporaneous data for 1973 nor any retrospective data for 1988. When I pool the estimates of trends for black men and women, the series shows increasing college entry chances from 1972 to 1975, a sharp decrease from 1977 through 1983, and apartial recovery from 1973 to 1988.

91

Trends in College Entry

0.0 /-----

2 -0.2 c

C

w

&-0.4

Q)

- \

/

..... / . . . . . . . +..........................

-

.- /

\

\ \

/

\)--

\

I

0

-0.6

c

-

....

0

**.

v)

-1.0

J 0

0.6

/

-. /.....

*. 1.

.*

.a.

I

I

-1.2

L.

.

I

I

1

I

I

I

0.4

-

0.2

-

0

c

0

+Io

B

g-0.2 J -0.4

....................

......................

,.a'

/ #

..............

0

) - - - - # -

/

. -

.........

......................

Q)

-a, 6

I

1

c

C

;

...

**..

g -0.8 -

8rn

J /

\

/----

P -

#

-

-

@

-' -I

I

I

I

I

I

I

I

I

Fig. 3.14 Components of trend in college entry among blacks: Recent high school graduates, 1972-1988 Note; Data are three-year moving averages from October CPS.

The lower panel of figure 3.14 shows that among blacks, as among whites, the overall effect of changes in social background and dependency status was an almost linear growth in the chances of college entry for men and women, and the growth was faster among blacks than among whites. Thus, the observed decline in black chances of college entry from the middle 1970s through the early to mid-1980s was muted by rapid improvement in the back-

92

Robert M. Hauser

ground characteristics of black high school graduates.26 Had there been no such trend in background characteristics, the downturn in chances of college entry among blacks would have been much larger, and it might have been detected earlier. Trends in the components of Hispanics’ chances of college entry are shown in figure 3.15. Unlike whites and blacks, there are no significant sex differences in the effects of year of high school graduation on college entry when social background is controlled (Lz = 23 with 16 df), so I have displayed pooled estimates of the trend in the upper panel of figure 3.15. Excepting a possible small rise and fall in the middle 1970s and some instability after 1985, the series shows essentially no change. There does not appear to be a consistent long-term trend among Hispanics toward improvement in social background composition, at least as it relates to their college-going chances. Although there appear to be improvements early and late in the series of social background composites, I think they are unreliable given the lack of consistent change. That is, because cohorts of parents bear children over a period of years, one would not expect to observe rapid change in the social background of successive cohorts of high school graduates, unless there were also drastic changes in selection into high school graduation. 27

3.5.1 Dependency and College Entry The first three rows of table 3.1 show the interaction effects of sex and dependency on college entry. Among all three racial-ethnic groups, dependent women are more likely to enter college than are dependent men. This effect is small and not statistically significant among Hispanics, but the estimate is also not significantly different from those for whites or blacks. Nondependents are less likely to enter college than dependents, regardless of sex and raceethnicity. The main effect of nondependency, which pertains to males, is much stronger among whites than among blacks or Hispanics; it is not statistically significant in the minority groups. There is an additional effect of nondependency among women in all three groups, and there are no significant differences among the groups in this sex interaction. The effects of dependency would not be of special interest except that my single-equation model may incorrectly specify that nondependency is a cause, rather than an effect, of college entry. Even if the equation were wrong in this way, the error could not seriously affect estimated trends and differentials if 26. As shown by Featherman and Hauser (1976), there is a long-term trend toward improvement in the social background characteristics of blacks-except in the important case of nonintact families-and there is no reason to link that trend specifically to the post-1954 dismantling of legally mandated school segregation (compare Kane 1991b). 27. In my current research on trends in high school dropout, I find no evidence of increases in high school dropout in the early 1970s and late 1980s that could explain sudden growth in the social background characteristics of Hispanic high school graduates.

Trends in College Entry

93

.

Trend Net of Social Background and Dependency Status

1972

-

.

C

-

W

1976

1978

1980

1982

1984

1986

1988

1 -

2

w

1974

0.8

........

0 , . Q,

6 0.6 -

:

.............-*..&.w.&*s*u.. --*.........*.* 4;""" ...a.

0.

.c

0

: 0.4

B

$0.2

O

- ;/

/

---,-,'

....* / /

..0

:/ -, -

I

I

I

I

I

I

I

I

I

Fig. 3.15 Components of trend in college entry among Hispanics: Recent high school graduates, 1972-1988 Note: Data are three-year moving averages from October CPS.

there were very few nondependents among recent high school graduates, but this is not the case. Figure 3.16 shows trends in estimated percentages of nondependents by sex among whites, blacks, and Hispanics. The estimates are based on a logistic regression of dependency status on calendar year, year of report, race-ethnicity, and sex, in which calendar year is permitted to interact with race-ethnicity but not with sex. That is, trends in dependency status

94

Robert M. Hauser 20

c

C

a, 0 C 15 a, Q

-

a,

n

z" 10 0,

cn m

c

c

g 5 a,

a

011

1972

-

I

I

I

I

1

1

1974

1976

1978

1980

1982

1984

. _ _ - .......

--.

.-..-

I

L

1986

1988

.__-

White men White women Black men Black women Hispanic men Hispanic women

Fig. 3.16 Percentage of nondependents among recent high school graduates by sex and race-ethnicity, 1972-1988 Note; Data are three-year moving averages of model estimates from October CPS.

vary across the racial-ethnic groups, but the trends are the same for men and women within each group. Rates of nondependency range from about 5 percent to 20 percent across years and racial-ethnic groups. In general, dependency increased from the early 1970s through the late 1980s. For example, nondependency decreased from 19 percent to 14 percent among white women, and it decreased from 8 percent to 3 percent among black men. The only exception to this trend is an anomalous jump in the series for Hispanics after 1983. Within each racialethnic group, men were consistently less likely to be nondependent than women were; the sex differential was largest among whites and smallest among Hispanics. Within each sex, whites were consistently more likely to be nondependent than blacks were. Presumably, this is a consequence of the greater economic resources and opportunities of young whites; they are more able to afford to set up independent households. Similarly, in most years, Hispanic women were less likely to be nondependent than were white women, but Hispanic men were more likely to be nondependent than were white men. What are the likely implications of these trends and differentials in dependency for rates of college entry? First, given the negative association between nondependency and college entry, decreasing rates of nondependency will tend to increase college entry. This is one component of growth in the background effects for whites and blacks shown in figures 3.13 and 3.14. Second, women are more likely to be nondependent than men, but the sex differential in college entry changes with dependency status. Dependent women have better chances of college entry than dependent men, but nondependent women

95

Trends in College Entry

have worse chances of college entry than nondependent men. Thus, the effect of a failure to control dependency status will vary both with the share of women who are nondependent and the effects of sex within dependency status. Because of the ambiguous causal role of dependency status and the opposite effects of sex on college entry among dependents and nondependents, it may be best to be agnostic about the implications of the present analysis for overall gender differentials in college entry. Third, since whites are more likely to be nondependent than blacks, the black-white difference in college entry chances will be understated if we fail to control dependency status. There may be a similar effect on the Hispanic-white difference in college entry, but that is less clear from the findings in figure 3.16 and table 3.1. 3.5.2 Racial-Ethnic Differentials in College Entry In order to compare differentials in college entry among whites, blacks, and Hispanics, I estimated logistic regressions in which the effects of social background were constrained to be equal in each year, in all three racial-ethnic groups, and among men and women. Because more than 80 percent of high school graduates were white, the data for whites dominate the estimates, and in effect the analysis yields comparisons of white, black, and Hispanic rates of college entry, standardized on the white regressions and conditioned on the social background composition of the three groups.28 Sex and race-ethnicity were permitted to interact in the equation for college entry, and each of those variables was permitted to interact with dependency status, but there were no three-way interactions. The effect of sex differed by calendar year (L2 = 40 with 16 df) as did the effect of race-ethnicity (L2 = 120 with 32 df),but there was no three-way interaction among sex, race-ethnicity, and calendar year (L2 = 50 with 32 df). That is, just as in the analyses without controls for social background, when social background and dependency status were controlled, sex differences in the trend in college entry were similar among whites, blacks, and Hispanics. The estimated trends and differentials in college entry among white, black, and Hispanic men and women are shown in figure 3.17. The estimates are normed on dependent youth in their year of high school graduation, and I have evaluated the trends and differentials near the average proportion of college 28, Cameron and Heckman’s complaint about these intergroup comparisons seems overblown. Obviously, to compare groups only at the grand mean of the regressors would be somewhat arbitrary in the presence of strong interactions; the significance of the interactions between raceethnicity and the effects of social background, globally or severally, does not establish their importance in the assessment of trends and differentials. For example, note that the trend lines in figure 3.17 for each racial-ethnic group are similar to those previously estimated independently within each racial-ethnic group (compare the top panels of figures 3.13, 3.14, and 3.15). The apparent insensitivity of these trends to the source of estimated background effects is ignored in Cameron and Heckman’s discussion. It would be useful to supplement the trend analyses reported here with a parallel analysis, comparing blacks, whites, and Hispanics but using the black regressions as the standard. Having carried out many similar exercises in the past, I doubt that this one would alter the present conclusions in any significant way.

96

Robert M. Hauser 0.65

&

0.6

al -

s

0.55

Ol C

.-kl

0.5

c)

c

l U 0.45

c

0 .-

r0

0.4

n 0.35 0.3

'

1972

I

I

I

I

I

I

I

I

1974

1976

1978

1980

1982

1984

1986

1988

_---

W h e n White women Biack.men Bla? L o p n Hisp&n.iF-men

Hisp?ns-Komen

Fig. 3.17 College entry net of social background by sex and race-ethnicity: Dependent high school graduates, 1972-1988 Nore: Data are three-year moving averages of model estimates from October CPS.

entry among male black dependents. Thus, the trend lines for other groupswhite or Hispanic, male or female-show estimated proportions of college entry among persons in each group with the characteristics of black male dependents. Figure 3.17 shows four striking findings. First, the estimates in the full model yield much larger intergroup differentials in college entry than are observable in the original data. Second, among black and white dependents, women's chances for college entry steadily improve relative to those of men. Third, controls for dependency status and social background eliminate or reverse the original racial and ethnic differentials (compare figure 3.4). For example, nearly half of black male dependents entered college in the middle 1970s, but fewer than half of white male dependents with the same social characteristics entered college. Among dependent youth, through most of the 1970s and 1980s, when social background is controlled, Hispanics enjoy better chances of college entry than blacks of the same sex, and blacks in turn have better chances of college entry than whites of the same sex. Differentials in social background, including dependency status, are more than sufficient to account for white advantage in access to college in the first year after high school graduation. Fourth, while chances for college entry grew among whites, they declined among blacks. After the late 1970s, the chances of black college entry declined from a situation of net black advantage to the point where there essentially was parity between dependent blacks and whites of each sex and the same social and economic background. Figure 3.18 shows the trend in college entry among dependent white men, controlling social background, along with the trends in the black-white,

Trends in College Entry

97

1.2

-2.2

1

t.

$

-2.6

5

c

5 5

'C C

0.8

3 0.6 ? 3 0.4 s

j

-5?

-2.4

t

z

-2.8 m

0.2

-3

'5 5

8

W Y=

0

-3.2 g, -I

-0.2 1972

1974

1976

1978

B l a c k w h i t e Hispa!!

1980

ys; white

1982

1984

1986

1988

W 0 ~ n . ymen ~ . WEi men

&

Fig. 3.18 Effects of sex and race-ethnicity on college entry net of social background and dependency status: Recent high school graduates, 1972-1988 Nore: Data are three-year moving averages of model estimates from October CPS.

Hispanic-white, and female-male contrasts.29 To facilitate comparison, these effects are shown on the same scale as the comparable contrasts in figure 3.5, where dependency status and social background were not controlled. There is substantial similarity in the shape of each of the corresponding contrasts in figure 3.5 and in figure 3.18, but there are also important differences in the shape and location of the contrasts. For example, there is little difference in the shape of the baseline trend among dependent white men. However, peak attendance in the late 1980s is higher than peak attendance in the early 1970s in the observed data of figure 3.5, but not in the adjusted data of figure 3.18, because of the contribution of improved social background to the growth of college entry among whites. The shape of the black-white contrast is also similar in figures 3.5 and 3.18, but the troughs of the early 1970s and the late 1980s are at almost the same level-about 0.4 less than whites-in the observed data but not in the adjusted data. In figure 3.18, even in the early 1970s, once social background was controlled, dependent black men had much better college entry chances than dependent white men; in the late 1980s, dependent black men had very nearly the same college entry chances as dependent white men.3o The improvement in the black-white contrast in the 1970s was less steep in the ad29. As displayed, the black-white and Hispanic-white contrasts pertain to dependent men, while the female-male contrast pertains to dependent whites. Under the model, the trends in the black-white and Hispanic-white contrasts are the same for men and women regardless of dependency status, while the trends in the female-male contrasts are the same among whites, blacks, and Hispanics regardless of dependency status. 30. On the average, the black-white contrast is 0.082 larger among dependent women than among dependent men, so black women may have better college entry chances relative to white women than those shown in figure 3.18.

98

Robert M. Hauser

justed than in the observed series, while the decline in the black-white contrast in the 1980s was sharper in the adjusted than in the observed series. The poor social backgrounds of blacks relative to whites account for the overall shift in the contrast from negative to positive between the observed and adjusted series, while the more rapidly improving social backgrounds of blacks account for the differing shapes of the observed and adjusted black-white contrasts. In the case of the Hispanic-white contrast, there is no consistent trend either in the observed series of figure 3.5 or in the adjusted series of figure 3.18; but rather than hovering about zero, the Hispanic-white contrast in the adjusted series is about 0.8. That is, among dependent men of equal social background, the college entry chances of Hispanic men far exceed those of white men.31 The shape of the female-male contrast in the observed series of figure 3.5 is virtually the same as in the adjusted series of figure 3.18. We would expect this from the virtually identical social background distributions of dependent men and women. However, there is a difference in the vertical location of the two contrasts. In the adjusted series, unlike in the observed series, women have better college entry chances than men in every year after 1975. As explained earlier, dependent women are more likely to enter college than are dependent men, but the observed series of figure 3.5 also reflects the experience of nondependent women. Women are more likely to be nondependent than men, while nondependency is associated with poor college entry chances, especially among women. To illustrate this, figure 3.19 shows the female-male contrasts under three alternative The gross effect is taken from the model of figure 3.5, and the effect in the full model is taken from figure 3.18. The third series is estimated from a model in which dependency status, but no other background variables, enters the model of college entry. There is a sharp upward shift in the female-male trend line when dependency status is controlled, but there is almost no shift in the trend line when the other background variables are added to the model. The ambiguous causal standing of dependency status may leave us wondering about the overall college entry chances of women relative to men, but the evidence is clear that from the 1970s to the 1980s the college entry chances of women improved sharply relative to those of men. The effects of dependency status on the black-white and Hispanic-white contrasts are much different than its effects on the female-male contrast. Figures 3.20 and 3.21 show estimates of the racial-ethnic contrasts that correspond to the female-male contrasts of figure 3.19. Both the black-white and Hispanic-white contrasts become larger when dependency status is controlled; thus, it is most unlikely that a failure to control dependency status would lead 3 1. On the average, the Hispanic-white contrast is 0.118 smaller among dependent women than among dependent men, but this difference accounts for only a small share of the Hispanic advantage. 32. The effects shown in figure 3.19 are pooled across whites, blacks, and Hispanics.

Trends in College Entry

99 1

-

P w o 0.5

-5?

-

s p -0.5 -

8

I

1972

I

1974

I

1976

-

I

1978

I

1980

I

1982

I

I

1984

1986

I

1988

Gross effect Effecl net dependency Effect (n.!u!l.model

_-__

Fig. 3.20 Black-white differences in college entry: Recent high school graduates, 1972-1988 Note: Data are three-year moving averages of model estimates from October CPS.

us to overestimate the black-white or Hispanic-white differences in college entry.33 33. Curiously, Cameron and Heckman’s discussion of dependency status and college enrollment ignores the analyses of figures 3.16 through 3.19.

100

Robert M. Hauser

1972

1974

1976

-

1978

1980

1982

1984

1986

1988

Gross effect Effect net dependency Effect ....... in full model

_---

Fig. 3.21 Hispanic-white differences in college entry: Recent high school graduates, 1972-1988 Note: Data are three-year moving averages of model estimates from October CPS.

3.6 How Much Equality of Opportunity? This analysis confirms some previous findings about differentials and trends in college entry while adding some new findings about them as well. More important, the analysis also opens a few new areas of ignorance. By combining contemporaneous and retrospective reports of post-high school activity, we confirm earlier findings of a fall and rise in the college-going chances of white men from the early 1970s to the late 1980s. Contrary to some previous analyses, we find clear evidence of divergent trends in college entry between blacks and whites during this period. Relative to those of whites, black college entry chances rose during the 1970s and declined from the late 1970s through the 1980s. While data for Hispanics are less reliable, they appear to show little difference between whites and Hispanics in trends in college entry. Thus, whatever explanation one offers for the divergent trends among blacks and whites ought not to apply equally to the contrast between Hispanics and whites. Several explanations have been offered for the decline in college entry chances of blacks relative to those of whites. These include changes in the incomes of black families, changing gender differentials in college entry, differential recruitment of blacks and whites into the armed forces, changes in academic performance or in the selectivity of high school graduation, changes in plans and desires to attend college, and changes in the cost of a college education. The available evidence rules out all of these except changes in the net cost of college attendance. While black-white differences in social back-

101

Trends in College Entry

ground are more than sufficient to account for black-white differences in college entry, we found no support for the hypothesis that trends in the blackwhite difference in college entry can be explained by trends in family income-or by trends in any of the other social and economic characteristics available in the CPS. Likewise, we have seen that the decline in college entry chances has affected both black men and black women. Evidence from other sources rules out some of the other explanations (Hauser, in press). Although almost all entrants into military service since the late 1970s have been high school graduates, black rates of entry into military service declined during the same period that black chances of college entry declined. High school completion has increased among blacks, but the change has not been dramatic, and it has been accompanied by improvements in the academic performance of blacks relative to whites. Plans and aspirations to attend college have grown among blacks, as among whites (Hauser and Anderson 1991). Thus, changes in cost appear to be the most likely explanation of changing black and white chances of college entry. Hauser (in press), among others, points to the shift in support from grants toward loans as a source of change that would be especially disadvantageous to blacks, while Kane (199 1b) emphasizes changes in net cost associated with rigidity in the size of Pel1 While the overall chances of men and women are obscured by the complex relationships among gender, dependency status, and college entry, the evidence is clear that there has been steady improvement in the college entry chances of women relative to those of men. In addition, contrary to a great deal of received opinion, there is no substantial evidence that, at any point in time, gender differences in college entry differ among blacks, whites, and Hispanics or that trends in gender differences in college entry differ among blacks, whites, and Hispanics. That is, differentials and trends in the effect of gender cut across racial and ethnic lines, and differentials and trends in the effect of race and ethnicity cut across gender lines. Among whites and blacks but not among Hispanics, global improvements in social background composition have been a steady source of growth in college entry. Growth in parents’ levels of schooling and declines in family size were major components of these improvements, which were not dampened by increasing rates of family disruption. Improvements in social background slowed both the observed decline in white chances of college entry in the 1970s and the observed decline in black chances of college entry in the 1980s; obversely, they accelerated the growth of white chances of college entry in the 1980s and of black chances of college entry in the 1970s. Differences in the effects of social background characteristics among racial and ethnic groups raise questions that should be pursued in additional analyses of these or other bodies of data. For example, why do the effects of family 34. One difficulty with Kane’s explanation is that it would appear to apply equally to blacks and whites.

102

Robert M. Hauser

income differ among whites, blacks, and Hispanics? Is this a consequence of nonlinearity in the effects of income across the different segments of the income distribution occupied by the three groups, or are there real differences in income-specific preferences, in the impact of financial aid policy, or in the location of populations relative to schools with differing costs? Similarly, why are the effects of parental schooling larger among whites than among blacks or Hispanics? Why is living in a female-headed household a greater obstacle to college entry among whites than among blacks or Hispanics? Why do regional differences in college entry appear to differ substantially among whites, blacks, and Hispanics? Why is independence from the parental household associated with much lower chances of college entry among whites than among blacks or Hispanics? Even without controlling academic performance, traditional differentials in the chances of college entry between whites and blacks or Hispanics are reversed when a full set of social background characteristics is controlled. Once social background is controlled, even though blacks’ chances of college entry declined sharply relative to those of whites after the late 1970s, they never fell far below those of white^.^' When social background is controlled, Hispanic chances of college entry are consistently much higher than those of blacks or of whites. One reason for the net advantage of Hispanic high school graduates may be the selectivity of high school graduation: only about 60 percent of Hispanics graduate from high school. However, the selectivity of high school graduation is much less among blacks and whites, and the difference between black and white graduation rates is much less than that between white or black and Hispanic graduation rates. These findings raise, in a rather pointed way, the question of how much compensation for preexisting population differences ought to be the goal of public policy. If, among persons with the same social background, minority chances of college entry exceed those of the white majority, is there a rationale for expanded efforts to improve the relative chances of minorities? On the negative side, one might argue that the goal of public policy ought to be limited to establishing parity among groups. Majority groups are often as quick to condemn reverse discrimination as minority groups are to object to traditional patterns of discrimination. On the positive side, given the evidence of persistent disadvantage among minorities in other social and economic processes, as well as the large and persistent differences in social background between majority and minority groups, one might argue that we should con35.Given the well-documented differences in academic performance favoring majority whites (Jaynes and Williams 19891, it is reasonable to assume that the present findings underestimate the college entry chances of blacks relative to those of whites with the same academic performance and social background. On the other hand, since the academic performance of black high school students has improved rapidly since 1980, the net decline in the college entry chances of blacks may be even greater than the decline that 1 have estimated without controlling academic performance.

103

Trends in College Entry

tinue or expand public efforts to improve the status of minorities where the promise of success is greatest. If, during the late 1970s, economic and social conditions gave a real advantage to black youth in one of the most significant transitions from youth to adulthood, should we not attempt to understand and reestablish those conditions? These questions go well beyond the present data, but the data show that they are not merely of hypothetical interest.

References Bishop, John H., and Shani Carter. 1990. The worsening shortage of college graduate workers. Center for Advanced Human Resource Studies, School of Industrial and Labor Relations. Working Paper no. 90-15. Ithaca, N.Y.: Cornell University. Carter, Deborah J., and Reginald Wilson. 1990. Ninth annual status report: Minorities in higher education. Washington, D.C.: American Council on Education. Duncan, Otis Dudley. 1961. A socioeconomic index for all occupations. In Occupations and social status, ed. Albert J. Reiss, Jr., 109-138. New York: Free Press. . 1968. Patterns of occupational mobility among negro men. Demography 5(1):11-22. Featherman, David L., and Robert M. Hauser. 1976. Changes in the socioeconomic stratification of the races, 1962-1973. American Journal of Sociology 82(November):621-51. . 1978. Opportunity and change. New York: Academic Press. Frase, Mary J. 1989. Dropout rates in the United States: 1988. Analysis Report no. NCES 89-609. Washington, D.C.: National Center for Education Statistics. Hauser, Robert M. 1991a. Educational transitions and the future workforce: Federal statistics to inform public policy. Institute for Research on Poverty Notes and Comments. Madison: University of Wisconsin-Madison. . 1991b. Measuring adolescent educational transitions among black Americans, Hispanics, and whites. Institute for Research on Poverty Discussion Paper 95 1-91. Madison: University of Wisconsin-Madison. . 199Ic. Uniform October Current Population Survey person-household file, 1968-1988. Codebook for machine-readable data file. Madison, Wis.: Center for Demography and Ecology. . In press. The decline in college entry among African Americans: Findings in search of explanations. In Prejudice, politics, and race in America today, ed. Paul Sniderman, Philip Tetlock, and Edward Carmines. Stanford, Calif.: Stanford University Press. Hauser, Robert M., and Douglas K. Anderson. 1991. Post-high school plans and aspirations of black and white high school seniors, 1976-1986. Sociology of Education 64(0ctober):263-77. Hauser, Robert M., and David L. Featherman. 1976. Equality of schooling: Trends and prospects. Sociology of Education 49(April):99-120. Hauser, Robert M., Shu-Ling Tsai, and William H. Sewell. 1983. A model of stratification with response error in social and psychological variables. Sociology of Education 56(January):20-46. Hu, Arthur. 1991. Hu’s on first. Asian Week, February 22:21. Jaynes, Gerald David, and Robin M. Williams, Jr., eds. 1989. A common destiny: Blacks and American society. Committee on the Status of Black Americans, Com-

104

Robert M. Hauser

mission on Behavioral and Social Sciences, National Research Council. Washington, D.C.: National Academy Press. Kane, Thomas J. 1991a. College cost and the timing of college entry. Cambridge, Mass.: Kennedy School of Government, Harvard University. Manuscript. . 1991b. College entry by blacks since 1970: The role of tuition, financial aid, local economic conditions, and family background. Cambridge, Mass. : Kennedy School of Government, Harvard University. Manuscript. Koretz, Daniel. 1990. Trends in the postsecondary enrollment of minorities. Santa Monica, Calif.: Rand Corporation. Mare, Robert D. 1979. Social background composition and educational growth. Demography 16(February):55-71. Mortenson, Thomas G. 1990. College entrance rates for recent high school graduates. ACT Financial Aid Research Briefs, vol. 5 . Iowa City, Iowa: American College Testing Program. . 1991. Equity of higher educational opportunity for women, black, Hispanic, and low income students. ACT Student Financial Aid Research Report Series, no. 91-1. Iowa City, Iowa: American College Testing Program. Murphy, Kevin, and Finis Welch. 1989. Wage premiums for college graduates: Recent growth and possible explanations. Educational Researcher 18(May):17-26. Pelavin, Sol H., and Michael Kane. 1990. Changing the odds: Factors increasing access to college. New York: College Entrance Examination Board. Raftery, Adrian E. 1986. Choosing models for cross-classifications. American Sociological Review 5 l(February): 145-46. Sewell, William H., and Robert M. Hauser. 1975. Education, occupation, and earnings: Achievement in the early career. New York: Academic Press. Siege], Paul M. 1971. Prestige in the American occupational structure. Doctoral diss. University of Chicago. Stevens, Gillian, and Joo Hyun Cho. 1985. Socioeconomic indexes and the new 1980 census occupational classification scheme. Social Science Research 14(June):14268. Stevens, Gillian, and David L. Featherman. 1981. A revised socioeconomic index of occupational status. Social Science Research 1O(December):364-95. U.S. Bureau of the Census. 1974. Social and economic characteristics of students: October 1972. Current Population Reports. Population Characteristics, Series P-20, no. 260. Washington, D.C.: U.S. Government Printing Office. . 1988. The Hispanic population in the United States: March 1985. Current Population Reports. Population Characteristics, Series P-20, no. 422. Washington, D.C.: U S . Government Printing Office. . 1990. School enrollment-social and economic characteristics of students: October 1988 and 1987. Current Population Reports. Population Characteristics, Series P-20, no. 443. Washington, D.C.: U.S. Government Printing Office. U.S. Department of Education. 1990. National goals for education. Washington, D.C.: U.S. Government Printing Office.

105

Trends in College Entry

Comment

Stephen V. Cameron and James J. Heckman

Robert Hauser presents a fascinating and comprehensive analysis of trends in college entry by high school graduates from different demographic groups. The CPS data he uses are the largest available data sets that enable analysts to track year-by-year variation in college entry. Sample size may not compensate for intrinsic limitations in the data, however. In this paper, we question the strength of the evidence supporting the main conclusions set forth by Hauser and others who use CPS samples. Hauser is well aware of many of the limitations of the October CPS and addresses some of them in this and other work. We examine several of these shortcomings by looking at college enrollment data based on the National Longitudinal Survey of Youth (NLSY), which we briefly summarize below. Although these data contain substantially fewer persons than the CPS and cannot be used to estimate aggregate annual enrollment rates reliably, they are longitudinal in nature, contain much more information on family background characteristics, and enable analysts to construct measures of family resources for all persons-not just those who are “dependents” as classified by the CPS. Using these richer data, it is possible to estimate more interpretable models of schooling attendance that do not support many of Hauser’s main conclusions. We report our analyses elsewhere (Cameron and Heckman 1991a, 1991b, 1992a) and summarize them here. We also raise an important interpretive problem that plagues this paper and many others in this literature. Following a long tradition, Hauser estimates the parameter of one transition equation in a multistate educational process: the transition from high school to college. High school graduation is a selective process. Many more whites graduate from high school than do blacks or Hispanics. In the presence of unmeasured ability and motivational variables, his estimates of behavioral equations for the transition from high school to college confound the effects of variables in placing a person in the category of being a high school graduate and being eligible for college with the effects of those variables in “causing” persons to go to college. Cameron and Heckman (1991a, 1992a) present evidence indicating the importance of measuring or controlling for such motivational models in explaining the determinants of education transitions and call into question the validity of behavioral models Stephen V. Cameron is a research associate at the Center for Social Policy Evaluation, Harris School of Public Policy, University of Chicago. James J. Heckman is Henry Schultz Professor of Economics and Public Policy at the University of Chicago, director of the Center for Social Policy Evaluation at the Hams School, and a member of the NBER. Portions of this paper are based on joint work reported in Cameron and Heckman (1990, 199 1a, 1991b, 1992a). This research was sponsored by NSF Grant SES-91-11455 and by a grant to the Harry and Lynde Bradley Foundation to the Harris School of Public Policy at the University of Chicago. The authors thank Michael Rothschild and Robert Hauser for helpful comments on this paper.

106

Robert M. Hauser

of educational transitions estimated on repeated-cross-section data sets such as CPS. This evidence and Hauser’s own evidence render suspect his analysis of racial-ethnic differences in college entry. It is not appropriate to constrain family background effects to equality in accounting for racial/ethnic trends. Such a procedure amounts to implicitly picking one value of background characteristics to compare outcome differences between nonparallel college attendance equations. The benchmark value chosen is only implicitly defined, has no compelling justification, and in fact has a peculiar and unintuitive property. A better method of comparison would be to examine the difference in estimated nonparallel equations at a range of explicitly stated and plausible values. The widely held intuition that slope-constrained equations pass through the overall sample mean is false. Our comments are presented in the following order. First, we discuss the limitations of the CPS data. Second, we summarize the relevant dissonant results from our own research. Finally, we discuss the peculiar properties of Hauser’s method of comparing determinants of demographic differences in schooling attainment.

Limitations of the CPS Data The CPS data are less than well suited for establishing the link between family income and educational decisions. The problem is that young persons who do not live at home and who do not live in group quarters if they attend college are assigned their own income (or the income of the young person and associated spouse) rather than that of their parents. This problem has given rise to a convention in the CPS-based determinants of schooling literature to restrict samples to dependents who are high school graduates in order to estimate equations determining who goes to college. It also gives rise to a focus on college enrollment rather than on college graduation, despite the greater importance of the latter in determining career outcomes. The CPS dependency link between youth and their parents becomes much weaker for youth making postsecondary schooling decisions beyond initial entry decisions. Two distinct problems are created by this convention: (1) excluding nondependents, one cannot ascertain random-sample or population family income effects on schooling participation, and (2) conditioning on a choice variable (dependency status) generates a standard simultaneous equation problem since dependency status is likely to be affected by the same unobservables governing college attendance decisions. By using dependency as a causal, or “right-hand side,” variable, Hauser produces biased estimates of the impact of socioeconomic variables on college attendance. Putting the simultaneity problem to one side, by conditioning on a choice variable, Hauser underestimates (in absolute value) the effect of any variable that moves college attendance and dependency status together and overstates (in absolute value) the population effect of variables that have opposite effects on attendance and dependency.

107

Trends in College Entry

Some evidence on the empirical importance of the first problem for blacks, whites, and Hispanics is presented in tables 3C. 1 and 3C.2. The data consist of recent high school completers, both males and females, between the ages of 14 and 24.’ Using the NLSY data, it is possible to construct a CPS-like measure of family income that measures the parent’s family income if the individual is a dependent and measures his or her family income otherwise. This measurement is taken in the year following high school completion, when college attendance is measured. In addition, we present a measure of parents’ family income for both dependents and nondependents by using parents’ family income measured at the interview in the year earlier.2 Table 3C. 1 presents means of both “correct” NLSY and dependency-based CPS measures for 14- to 24-year-old recent high school completers-the same group studied by Hauser. For whites and Hispanics enrolled in college, mean parental family income of dependents is higher than it is for nondependents, but the difference is not large. For blacks the difference is essentially zero. For Hispanics and whites not enrolled in college, this gap is maintained. For blacks, the dependentnondependent gap in parental income is much less for those not enrolled in school than it is for those enrolled in school. As expected, the CPS dependency-based family income measure badly understates true parental income for nondependents. Table 3C.2 provides the ingredients for answers to three questions: 1 . What is the effect of conditioning on dependency status? 2. Conditioning on dependency, what is the effect of using family income concurrent with enrollment rather than in the previous year, when college plans are being crystallized? 3. Given that nondependents have missing data for family background and family income, what is the effect on the estimates of adding an indicator variable for nondependency status and imputing the missing data? Table 3C.2 reports estimates of logit models for attendance or nonattendance in college in the spring following high school graduation that are similar to those employed in our other work (Cameron and Heckman, 1992a, 1992b). Estimates for college attendance in the October following high school graduation are qualitatively similar and for the sake of brevity are deleted. To answer the first question, column 1 in each panel gives the “true” family income effect; column 2 gives the effect of excluding nondependents from the analysis. In all cases, the effect of family income is significantly underestimated when nondependents are excluded, especially so for blacks. The log-odds ratio for income declines by 25 percent to 35 percent for different groups and 1. More precisely, individuals who either graduated high school or obtained an equivalency degree in the previous year and are now eligible for college entry for the first time. 2. October CPS and NLSY measures of family income are equally poor on one account: both ask householders a single question about total family rather than a series of questions, as is done in March CPS or the Survey of Income and Program Participation.

Table 3C.1

Means of Parent’s Family Income and Current Family Income by Race and Dependency Status for NLSY 14- to 24-Year-Old Males and Females Completing High School in the Previous Year, Excluding Individuals Joining the Military (thousands of 1988 dollars; means with standard errors of means in parentheses) Whites

Blacks

Hispanics

Dependent *

Nondependent

Dependent *

Nondependent

Dependent*

Nondependent

Enrolled in College: Parent’s family income* Current family incomet N (% of total race group)

$38.4 (0.7) $38.5 (0.7) 1,167 (31%)

$32.7 (2.7) $8.0 (0.8) 107 (4%)

$22.7 (0.8) $21.8 (0.8) 537 (35%)

$22.4 (3.0) $ 4.1 (0.5) 55 (4%)

$27.5 (1 .O) $27.5 ( I .O) 303 (35%)

$24.2 (2.6) $ 8.1 (0.9) 46 (5%)

Not enrolled in college: Parent’s family income* Current,family income** N (% of total race group)

$30.0 (0.5) $3 1. I (0.6) 1,019 (37%)

$24.4 (0.9) $10.2 (0.4) 460 ( 17%)

$18.2 (0.5) $18.2 (0.5) 688 (44%)

$15.0 (0.5) $ 7.2(0.8) 270 (17%)

$22.1 (0.9) $24.0 (0.9) 371 (42%)

$17.8 (1.3) $10.6 (0.9) 157 (18%)

~~~~

~

~

~~

~

~

Note: To exclude individuals who had graduated high school before the survey began, only individuals between the ages of 13 and 17 in 1978, the beginning of the first NLSY wave, were included in this sample. *Not married living at home, in a dormitory, or in jail. **Measured from the previous interview or when the individual was last a dependent. ?Counts parent’s family if the individual is a dependent, and counts the individual’s family if not a dependent. Measures the same characteristics of income as the CPS measure, except the measure reported here includes food stamps.

Table 3C.2

March College Enrollment Probabilitiesfor NLSY Males and Females Ages 14 to 24 Completing High School in the Past Year, Excluding Individuals Joining the Military: Logistic Probabilities(T-valuesin parentheses) Using Both Dependents and Nondependents and Family Income from Previous Year

Variable*

(1)

Using Dependents Only and Family Income from Previous Year (2)

Using Dependents Only and Family Income from Current Year (CPS Method) (3)

Using Both Dependents and Imputed Data for Nondependents (4)

A. Whites

Intercept Female Number of siblings HGC of father HGC of mother Family income** Broken home Farm age 14 South age 14 County average waget I979 1980 1981 1982 1983 Nondependent Nondependent female - 2** log-likelihood N (continued)

-6.824 (14.6) 0.199 (2.2) -0.119 (4.7) 0.189 (10.3) 0.254 (9.3) 0.012 (4.4) 0.157 (1.1) 0.290 (1.5) 0.343 (3.3) -0.013 (0.6) 1.343 (5.5) 0.836 (3.5) 1.051 (4.5) 0.897 (3.8) 1.194 (5.0) 3000.0 2650

-6.753 (12.3) 0.290 (2.9) -0.124 (4.5) 0.190 (9.2) 0.268 (8.7) 0.009 (3.8) 0.188 (1.1) 0.211 (1.0) 0.300 (2.6) -0.021 (0.8) 1.347 (4.0) 0.936 (2.8) 1.155 (3.5) 1.081 (3.2) 1.486 (4.4) -

-6.776 (12.5) 0.290 (2.9) -0.123 (4.5) 0.194 (9.5) 0.273 (8.9) 0.006 (2.4) 0.109 (0.7) 0.188 (0.9) 0.305 (2.7) -0.017 (0.7) 1.342 (4.1) 0.966 (3.0) 1.148 (3.6) 1.075 (3.3) 1.475 (4.5)

-

2478. I 2171

2469.9 2171

-6.508 (13.5) 0.293 (3.0) -0.123 (4.5) 0.192 (9.3) 0.273 (8.9) 0.005 (2.3) 0.097 (0.6) 0.146 (0.8) 0.320 (3.1) -0.012 (0.6) 0.912 (3.7) 0.535 (2.2) 0.684 (2.9) 0.603 (2.7) 1.042 (4.4) -0.963 (5.4) -0.840 (3.3) 2938.9 2650

Table 3C.2

(continued)

Variable*

Using Both Dependents and Nondependents and Family Income from Previous Year (1)

Using Dependents Only and Family Income from Previous Year (2)

Using Dependents Only and Family Income from Current Year (CPS Method) (3)

Using Both Dependents and Imputed Data for Nondependents (4)

B. Blacks Intercept Female Number of siblings HGC of father HGC of mother Family income** Broken home Farm age 14 South age 14 County average waget 1979 1980 1981 1982 1983 Nondependent Nondependent female - 2** log-likelihood N

-4.052 (7.9) 0.555 (4.8) -0.081 (3.7) 0.016 (0.8) 0.148 (5.1) 0.014 (3.0) 0.009 (0.1) 0.012 (0.0) 0.138 (1.1) -0.011 (0.7) 1.167 (4.2) 0.818 (3.2) 0.882 (3.6) 0.976 (3.9) 0.661 (2.5) 1809.5 1499

-3.534 (6.2) 0.500 (4.0) -0.078 (3.3) 0.022 (1.0) 0.128 (4.1) 0.009 (2.2) 0.089 (0.7) -0.073 (0.2) 0.041 (0.3) -0.013 (0.8) 1.214 (3.8) 0.934 (3.1) 0.975 (3.3) 1.152 (3.8) 0.785 (2.5) 1535.4 1220

-3.516 (6.2) 0.503 (4.0) -0.077 (3.3) 0.024 (1.1) 0.125 (4.0) 0.011 (2.9) 0.077 (0.6) -0.036 (0.1) 0.035 (0.3) -0.013 (1.1) 1.252 (3.9) 0.947 (3.1) 1.015 (3.4) 1.175 (3.8) 0.813 (2.6) 1529.2 1220

-3.610 0.508 -0.075 0.025 0.125 0.011 0.074 0.034 0.095 -0.011 0.958 0.688 0.744 0.850

(6.9) (4.1) (3.2) (1.1) (4.0) (3.0) (0.6) (0.1) (0.8) (1.0) (3.4) (2.7) (2.9) (3.3) 0.640 (2.5) -1.244 (4.4) 0.189 (0.5) 1773.2 1499

C. Hispanics

Intercept Female Number of siblings HGC of father HGC of mother Family income** Broken home Farm age 14 South age 14 County average waget 1979 1980 1981 1982 1983 Nondependent Nondependent female - 2** log-likelihood N

-2.329 (3.9) 0.089 (0.6) -0.029 (0.9) 0.050 (2.0) 0.004 (0.1) 0.011 (2.1) -0.495 (2.5) -0.338 (0.8) -0.028 (0.2) -0.007 (0.3) 1.195 (3.3) 0.834 (2.6) 1.207 (3.9) 1.053 (3.4) 0.941 (2.9) 1058.9 848

- 1.303 0.152 -0.040 0.053 O.Oo0

(2.0) (0.9) (1.1) (1.9) (0.0) (1.5) (2.3) (0.6) (0.4) (0.3) (2.2) (1.6) (2.9) (1.9) (1.6)

0.008 -0.512 -0.307 -0.081 -0.009 0.862 0.569 1.041 0.686 0.576 871.1 673

-1.236 (1.9) 0.145 (0.9) -0.040 (1.1) 0.056 (2.1) 0.003 (0.1) 0.005 (0.9) -0.558 (2.5) -0.307 (0.6) -0.093 (0.5) -0.014 (0.4) 0.900 (2.3) 0.598 (1.6) 1.059 (3.0) 0.689 (2.0) 0.586 (1.6) 872.5 673

-1.827 (3.0) 0.151 (0.9) -0.035 (1.0) 0.059 (2.1) 0.002 (0.1) 0.004 (0.8) -0.556 (2.5) -0.188 (0.4) -0.055 (0.3) -0.012 (0.4) 1.010 (2.8) 0.701 (2.1) 1.103 (3.4) 0.906 (2.9) 0.881 (2.7) 0.952 (3.0) -0.214 (0.5) 1043.1 848

Note: Enrollment is measured at the yearly interview date-February or March for most people. If the interview was after March, then March enrollment was determined from the monthly school enrollment data. Only individuals 13 to 17 at the beginning of 1978-the initial year of coverage of the NLSY-were included, to exclude individuals who had completed high school before 1978. Approximately 2 percent of each racial-ethnic group was dropped because parent’s family income was missing. In addition, another 1 to 2 percent was dropped due to missing values in the highest grade completed of either the father or the mother. Finally, approximately 2 percent of the total sample .individuals were dropped, as they had joined the regular military since completing high school. *Definitions of variables: Female is an indicator for females. Number of siblings is the total number of siblings. HGC of FatherlMother is the highest grade completed by the individual’s father or mother when the individual was age 14. Family income is the total family income of the parents of the individual. Broken Home is an indicator variable coded 1 if either parent was not in the household when the individual was age 14. Farm age 14 and South age 14 are binary variables indicating the individual at age 14 lived in a farm residence or the southern states respectively. Counry average wage is defined in the footnote below. The year variables are indicators for the year in which the individual is at risk of college entry; years 84-88 are the left-out variable. Nodependent indicates that the individual has hidher own household and was not in the parents’ household in March. Nodependent female is the nondependency indicator interacted with the female indicator. **Denominated in thousands of 1988 dollars. tThis variable measures the average county wage rate for individuals with high school-level skills. It is denominated in thousands of 1988 dollars. See Cameron and Heckman (1992a, 1992b) for a detailed description of the construction.

112

Robert M. Hauser

becomes statistically insignificant for Hispanics. Column 3 of the table addresses the second question. Instead of using the family income for dependents only at the time decisions are made (column 2), it is based on a sample that uses contemporaneous family income for dependents and produces results analogous to what one could estimate with the October CPS. For whites and Hispanics, estimates decline to half the “true” value. Estimates for blacks constitute the only exception to this rule. Note, however, that among the list of socioeconomic variables, only the family income variable demonstrates extreme sensitivity to the treatment of dependency status. (For blacks, the family income variable weakens when October rather than spring attendance is analyzed.) The third question we ask is whether one can improve on these estimates by employing the solution of Hauser to missing family background variables for nondependents. He imputes the values of the missing data for nondependents at the mean value of those variables for dependents. He also includes a dummy variable for nondependency status. In a linear regression model, this procedure can be shown to be equivalent to estimating all slope coefficients from the sample of dependents. In the last column of table 3C.2, we include an indicator variable for nondependency status and for nondependency interacted with “female,” and for each racial-ethnic group we impute family income and other family income variables using the means of the dependent group. The results from this procedure can be seen by comparing columns 3 and 4. There is virtually no change in the point estimates. This is not surprising: no new information is added to the model by this technique. It is worth noting at this point that we have estimated several other specifications. We do not follow Hauser and condition on housing tenure, because it is unavailable in the NLSY. We also do not employ as a regressor the occupational status of the household head. In a variety of models that include a full set of region, central city, and age variables, the empirical results were virtually identical to those presented in table 3C.2. One final adjustment to the specification-transforming family income from natural units to log form, as in Hauser’s table 3.1-again results in little quantitative difference in the estimates for whites and Hispanics. However, the coefficient on family income for blacks becomes small and insignificant for all specifications analogous to those in our table 3C.2, columns 1-4. This raises yet another question about the robustness of Hauser’s main conclusion that black family income changes cannot account for the changes in black college enrollment. Using a Box-Tidwell transformation (see, e.g., Heckman and Polachek 1974), we find that Hauser’s log specification is incorrect for blacks and whites. In fact, linear specifications are statistically significant in the black equations, but our log specifications are statistically insignificant at conventional levels. 3. This is the specification analogous to that reported in Hauser’s table 3.1.

113

Trends in College Entry

A Summary of the Main Findings of Our Related Work In our own research we use the NLSY data to estimate school enrollment and graduation equations for whites, blacks and Hispanics. The NLSY data are much richer than the CPS data. By following the same persons over time, we can study the determinants and consequences of educational selectivity, which is an especially acute problem among minority group members. In a series of papers, (Cameron and Heckman 1991a, 1991b, 1992a, 1992b), we set forth, and estimate, econometric models that control for selectivity in educational attainment. We develop a tractable econometric model of schooling in which family resources, parental environmental variables, local labor market alternatives, and federal and state tuition costhubsidy variables are introduced as explanatory variables. For blacks, Hispanics, and whites, it is possible to produce an econometric model that fits the data and is consistent with a simple neoclassical economic model. Better market opportunities for unskilled workers inhibit educational attainment. Better family resources, better home environments, and lower college tuitions promote schooling attendance. Failure to control for educational selectivity greatly reduces the role of socioeconomic variables in explaining minority college attendance. Our analysis is relevant to the interpretation of Hauser’s evidence. First, our evidence on the importance of controlling for educational selectivity casts doubt on Hauser’s estimates of behavioral parameters that do not control for selectivity in estimating schooling transitions. Second, our weak estimates of black responses to tuition costs and family resources cast doubt on Hauser’s conclusions about the importance of college tuition costs in accounting for black-white differences in college attendance. Third, differences in estimated behavioral coefficients across different demographic groups call into question Hauser’s methodology in accounting for racial/ethnic differences in college entry-a topic to which we now turn.

Comparing Nonparallel Regression Lines In the section 3.5.2., “Racial-Ethnic Differentials in College Entry,” Hauser performs a standard statistical exercise. He seeks to compare the vertical difference between two nonparallel log-odds regression lines. He does this by constraining the slope coefficients for socioeconomic variables to a common value for all demographic groups, allowing only demographic-specific intercepts to be free. These adjusted intercepts then form the basis for his interpretive analysis. He reports the result that when family background variables have the same effect on all demographic groups, minorities are more likely to enter college than are whites. In light of the evidence summarized above and presented in Hauser’s table 3.1, it is incorrect to constrain the slope coefficients to equality across demographic group^.^ In our table 3C.3, we test these 4. There is the additional problem that sample sizes are very different for different demographic groups. Using a common significance level across different demographic groups produces test procedures with power that depends on sample size. A more reasonable procedure recognizes that

114

Robert M. Hauser Testsof Aggregation(using the specificationin column 1 of table 3C.2)

Table 3C.3

A. Sex (chi-square P-values below)

Slope coefficients equal

White

Black

Hispanic

.I3

.58

.92

B. Race (chi-square P-values below) White = Hispanic

.00

.oo

.oo

.oo

.00

.oo

.oo

.00

Slope coefficients equal with race dummies free Slope coefficients equal with dummies for race interacted with sex

White = Black

Black = Hispanic

White = Black = Hispanic

aggregations formally for the full sample of dependents and nondependents, using the “correct” family income and the specification found in our table 3C.2. Though little behavioral difference is found between males and females across racial-ethnic groups (table 3C.3A), aggregation of any combination of racial-ethnic groups is strongly rejected (table 3C. 3B). The implications of these differences are explored in Cameron and Heckman (1992b). It is important to take care in interpreting Hauser’s finding, which also appears in Cameron and Heckman (1991b, 1992b). In a linear regression setting, the problem “solved” by Hauser is to pick a point of evaluation of the vertical difference between two nonparallel regression lines. Clearly, at some point the lines cross. For simplicity, suppose that there are two groups: “1” = the minority group; “0” = the majority group. Each group’s outcome measure is characterized by a separate regression line with a different slope and possibly a different intercept. Assume only one regressor. Letting d = 1 if an observation comes from the minority population ( = 0 otherwise), a conventional regression specification writes

+ YO(1

+ y d + p X + A(dX) + U . The equation for the majority outcome is Yo = cx + PX + U, while the equation for the minority outcome is Y 1 = a t + y + (p + A)X + U. Con(1)

Y

=

Y’d

-

d)

=

straining A = 0, pooling the samples and denoting the least squares estimators by “””, we establish in the Appendix that in a linear regression setting, the estimated vertical difference between the majority and minority regression lines is

significance levels for pretest estimators of the type used by Hauser should be adjusted for sample size. See Donohue and Heckman (1991), where this issue is discussed and examples of the effects of such adjustments are studied.

115

Trends in College Entry

where k l = E(X I d = 1) is the mean of X in the minority population po[= E(X 1 d = O)] is the mean of X in the majority population, and oo =

a“,U &,(l

- P) - P) c r y

+

where a0, is the variance of X in the majority population “0,” IT; is the variance of X in the minority population “1 ,” and P = E ( d ) is the minority population proportion. Thus, in a linear regression setting, Hauser’s adjusted intercept amounts to choosing the following point of evaluation for the vertical difference between two nonparallel regressions:

(2)

Wok1

+ (1

- %)Po

and adjusting y by the difference in slopes between “1” and “0” sample members: “OF1

+ (1

-

WO)POl.

There is no reason for preferring this point of evaluation over another. See figure 3C. 1 for a graphical exposition of the case A > 0, y < 0 and pl < po. Just as the value of the vertical difference at X = 0 in the unconstrained model (i.e., y) has no logical priority as “the” distance between two nonparallel lines, neither does the point expressed in equation (2). In fact, the implicit point used by Hauser and many others has very peculiar properties. The larger the minority proportion ( P ) and the larger the variance in X among minority members (i.e., the greater the proportion of explained variation in the minor-

Line Mean of Majority Outcome

Line:

“+8.PO

ro

WO’Pl+

I

Mean of Minority w1

Regressor

M;;gengPe:

Po I

;s

X

*

j or it y

Point o f Comparison Selected by OLS

Fig. 3C.1 The implicit point of comparison selected by ordinary least squares (OLS) of the vertical distance between two nonparallel regression lines

116

Robert M. Hauser

ity regression), the greater the weight placed on the majority mean as the point of evaluation of the vertical difference! Put another way, as the proportion of minority members declines (P -+ O), Hauser’s implicit point of evaluation in (2) increasingly weights the minority mean. If, for example, X is parental income, “1” refers to Hispanics, “0” refers to whites, and p, < p,,, then A > 0 (see Hauser’s table 3.1) and 4 overstates y, with the degree of overstatement increasing with the difference between the variance in majority and minority parental income and decreasing with increases in the proportion of minorities in the sample. A better way to summarize differences in nonparallel lines would be to pick values at the center (median, mean, etc.) of the X distributions for majority and minority groups and to evaluate sets of these differences. Using a regression-defined point of evaluation introduces the risk of constructing contrasts between outcomes of majority and minority outcome equations at points of evaluation of little interpretive interest. Of course, this analysis is only suggestive. Hauser fits a nonlinear logit model and not linear regressions, and he uses multivariate X rather than scalar X . These departures from the simple model just presented further obscure the interpretation of the implicit point of evaluation. In any case, it would be clearer to present the full array of differences in vertical contrasts rather than to pick a particular (unknown) point and base strong interpretations on it.

Summary Hauser’s comprehensive survey of trends in college attendance by demographic groups makes fascinating and informative reading. The limited nature of the CPS data hampers his analysis. In a previous draft of this paper, Hauser recognized many of the limitations of his analysis. He has been a vocal proponent of improving the October CPS and encouraging the National Center for Education Statistics to implement new periodic, longitudinal surveys to filI the current gap in education data. It is unfortunate that the editors censored his comments on these vital issues. Our analysis of the NLSY data suggests that many of Hauser’s substantive conclusions will not stand the test of better data and better analytic models.

117

Trends in College Entry

Appendix The Bias in Estimating Differences between Majority and Minority Group Outcomes by Falsely Constraining Slope Coeficients to Equality While Letting Group Intercepts Be Free For simplicity, consider a model with only one slope coefficient. Let Y be an outcome of interest. Let d = 1 if a person is black (or a member of a minority). E ( d ) = I! Let X be an explanatory variable with E(X) = F. U is a disturbance with E(U) = 0. Write E(X I d = 1) = pl. Then, for the regression model,

Y = a E ( Y ) = (Y

+ yd + p X + AdX + U + yP + pF + AP F I .

A is the difference in slope parameters between majority and minority group members. Thus, the equation for minorities is Y = ((Y y) (p A)X U and for the majority group is Y = a p X + U. Suppose, as is conventional in the literature, that we measure the difference between majority and minority outcomes by falsely imposing equality of slope coefficients (A = 0) but allow for group-specific intercepts. What is the effect of imposing false constraints on the estimated value of y? At what point (value of X ) are we evaluating the contrast (vertical distance) between the two nonparallel lines characterizing the majority and minority populations? A standard specification error analysis reveals that in the population,

+

+

+

Y =a

+

+

+ yd + p X + (AdX + U)

where the term in parentheses is the new composite error term. Then, assuming finite second moments, and random sampling and denoting regression estimators of a,y and p by ‘‘A’’,

where

Observe that

118

Robert M. Hauser

where

Thus plim $ = y

(‘41)

(A3)

plim & =

(Y

+

+ yP + pp. - (plim $)P - (plim 6)p

= ( Y - A [P(1 -P)uLPI +(P2Pl +PcI.o)uLl u&(l- P ) + u Y

These relationships have a straightforward interpretation. Equation (A2) is the most familiar. The estimated value of p lies between p and p A. It gets closer to p + A the higher P is and the higher the variance in population “1” (uL). As the minority population proportion goes to zero (P + 0) or as the dispersion of X in the minority population (a;) goes to zero, converges to the majority slope coefficient. From (Al), the estimated contrast between minority and majority outcomes, plim (+), is the contrast at zero (y) plus the difference in slope coefficients (A) times a value of X intermediate between p, and pa.This simply measures the difference between the two regression lines at a value between pl and podetermined by the relative size of the minority in the population and the variability of X in the minority population relative to that in the majority population. Observe that if

+

6

119

Trends in College Entry

and WI

then wo

=

- P)

&,(l

+ w1 = 1 and 0 5 wo,

WI

plim f = y

5

+ cry '

1, so we may rewrite (Al) as

+ A(w0kl + wlko).

The bias in the contrast moves in the opposite direction to the bias in the slope. Thus as P 4 1 or crYa0, 4 m, wo + 0 and the implicit point of evaluation comes close to k0.That is, the greater the fraction of minority members in the population or the greater the variance of X among the minority members, the closer we come to using the minority mean IJ.,as the point of evaluation! There is no compelling reason for using this point of evaluation to measure the difference between two nonparallel lines. For example, if X is parental income and A > 0, 0 < k, < po< a ~ then , overstatement in .$I increases with the variability of X in the minority population and the relative size of the minority population. In fact, if y < 0, plim f could be positive. The interpretation of plim 8 and the extension to a multivariate equation are straightforward and hence are deleted. Relaxation of the random sampling assumption is also straightforward and hence deleted for the sake of brevity.

References Cameron, Stephen, and James Heckman. 1991a. The determinants of high school graduation and college attendance. Manuscript. . 1991b. The nonequivalence of high school equivalents. NBER Working Paper no. 3804. Institute for Research on Poverty, University of Wisconsin, February. Revised August and forthcoming in Journal of Labor Economics, January 1993. . 1992a. Life cycle schooling decisions: Theory and evidence. University of Chicago. Manuscript. . 1992b. Sequential models of schooling. Paper presented at the Institute for Research on Poverty, University of Wisconsin, June 1990. Revised February. Donohue, John, and James Heckman. 1991. The impact of the functional form of earnings equations in accounting for black economic progress. American Bar Foundation, October. Memorandum. Heckman, James, and Solomon Polachek. 1974. Empirical evidence on the functional form of the earnings-schoolingrelationship. Journal of the American Statistical Association 69 (June 1974):350-54.

This Page Intentionally Left Blank

4

The Growing Concentration of Top Students at Elite Schools Philip J. Cook and Robert H. Frank

Colleges and universities compete to matriculate the most able students. But while there are literally hundreds of schools pursuing each year’s crop of top high school seniors, the competition is by no means even. Top students are attracted to the schools with the best reputations, the most prestige, and the greatest past success in matriculating good students. Schools further down in the academic hierarchy continue to attract a limited number of top students because of compensating advantages such as location, low tuition, or family tradition. But a remarkably high and growing proportion of top students end up in a small number of elite schools Qualitatively, the current interuniversity allocation of students resembles the intraschool tracking systems employed within many of the nation’s elementary schools. The social desirability of sorting by ability has been hotly debated in the elementary school case but has received much less serious attention in the university context. Indeed, closer examination of the tracking issue at the university level seems especially timely in light of the evidence we present here. This evidence suggests that the concentration of top students in elite schools may have increased substantially from the 1970s to the 1980s. Philip J. Cook is professor of economics and public policy at Duke University, where he holds a joint appointment in the department of economics and the Sanford Institute of Public Policy. Robert H. Frank is the Goldwin Smith Professor of Economics, Ethics, and Public Policy at Cornell University, where he holds a joint appointment in the department of economics and the Johnson Graduate School of Management. The authors gratefully acknowledge the assistance of those who made a special effort to provide unpublished data for this project, including Gary Barnes, Richard Herrnstein, Judy Kowarsky, Oscar Larson, Carol Luszcz, Charles Murray, Susan Murphy, and Judy Siders. Alan Durell, Gina Triplett, and Jose Chavez supplied able research assistance. The authors also thank Tom Devlin and Kim Alexander of the Cornell Career Services Office for help with their survey of Cornell recruiters. Charles Clotfelter provided many useful suggestions along the way. Henry Aaron, James Hamilton, Michael Rothschild, Larry Litten, Paula Berger, and the conference participants provided helpful comments on the original draft.

121

122

Philip J. Cook and Robert H. Frank

And there is other evidence suggesting that this recent increase is a continuation of a process that began in the immediate post-World War I1 period.’ In addition to whatever policy concerns are raised by these results, they may also be of interest as an illustration of a competitive process known in the economics literature as a “tournament.” The interesting features of this process are that it yields only a few winners from among many contestants and that the relative standing of the contestants, much more than their absolute quality, determines their prize. We suggest that the competition for top students can be viewed as a tournament, one characterized by a dynamic process with positive feedback. In recruiting good students, success breeds success, and the result has been a high degree of concentration. Among the possible explanations for why concentration has grown during recent years, we will discuss more intense marketing by elite schools, increasing wealth and reduced family size, lower long-distance telephone rates and real air travel costs, and a shift in the recruiting practices by elite employers. The next section provides background information on the value of a degree from an elite institution and the consequences of having a large proportion of top students at one of these institutions. Section 4.2 presents some statistics on the current distribution of top students. Section 4.3 then discusses changes in the environment of higher education that appear to have influenced the equilibrium distribution of students among schools; this section also describes a positive feedback process whereby small shifts in the environment may produce relatively large shifts in the distribution of students. Sections 4.4 and 4.5 provide evidence on the trend in concentration and the apparent consequences of this trend for recruiting by employers.

4.1 Background There is no mystery about which colleges and universities constitute the elite in American higher education. As noted by Kingston and Lewis (1990, xx), “prestige is a somewhat amorphous asset. Yet, for all the shadings of eliteness, there is remarkable continuity and consistency-among raters and over time-in the rankings of undergraduate schools.” There is a group of perhaps three dozen schools that are at the top of the rankings (with only minor variations) every year in college guides and news magazines and that are overwhelmingly successful in attracting top students. The students who do matriculate at these schools and graduate are on the “high status track” (Kingston and Lewis 1990);they tend to earn more than others and to have a much greater chance of achieving high rank in government or business. A recent survey by Fortune documents the extent to which graduates of elite schools hold the top positions in the business world (Caminiti 1990). Fortune 1. For example, author and commentator Charles Murray has pointed out to us that the annual reports of the Harvard Dean of Admissions document a vast improvement in SAT scores between 1952 and 1960.

123

The Growing Concentration of Top Students at Elite Schools

obtained responses from nearly 1,500 current and former chief executive officers of Fortune 500 and Service 500 companies. Almost all (93 percent) had graduated from college, and the seven schools that led the list were Yale, Princeton, Harvard, Northwestern, Cornell, Columbia, and Stanford-all elite private universities. These seven schools claim 166 CEO respondents, or over 10 percent of the total, as undergraduate alumni. The author notes that “the dominance of the Ivy League is, if anything, increasing: Whereas 14% of the former CEOs surveyed hold Ivy League undergraduate degrees, nearly 19%of the current CEOs do” (p. 121). Of course, relatively few alumni from any school, elite or otherwise, become CEOs of Fortune 500 companies. But taken as a whole, graduates of elite schools are more successful in the labor market than are graduates of other colleges and universities. This is no surprise, given that elite schools select students because of many personal qualities that happen also to predict success on the job. It is a matter of dispute whether elite schools have greater value added than other schools in terms of subsequent earnings and career accomplishment. One summary of the social science research on this subject states that “although graduates of higher-quality institutions do have demonstrably more successful careers, their greater success largely reflects greater intellectual and personal endowments and advantaged family backgrounds” (Kingston and Smart 1990, 148).* But even if a degree from an elite school served only to alert employers to the presence of these attributes, it would be a valuable asset indeed. The best evidence on the value of an elite degree comes from an unusually rich data set, the National Longitudinal Survey of the High School Class of 1972, which followed this cohort through 1986. James et al. (1988) report their analysis of a subsample consisting of 1,241 males who had graduated from college and who worked for an employer in 1985. Earnings in that year were regressed on vectors of individual and family characteristics, institutional characteristics, and higher educational experience. The authors’ primary conclusion is that “while institutional characteristics do not explain a large proportion of the variance in earnings, other aspects of the higher educational experience such as choice of major, number of math credits taken, GPA and postgraduate degree matter a great deal” (p. 21). Nonetheless, the overall selectivity of the school (as measured by average SAT scores of the freshman class) did have a considerable effect-each additional 100 points of combined SAT scores increased earnings by about 4 percent. And alumni of private eastern schools earned a few percent more than others even after controlling for this measure of ~electivity.~ According to James et al., if these 2. For similar conclusions, see Astin (1968) and Griffin and Alexander (1978). 3. James et al. control for whether or not a subject has an advanced degree as if the decision were exogenous. A more complete analysis would model that decision as a function of the characteristics of the undergraduate institution and family characteristics. Kingston and Smart (1990) studied a large sample of college graduates who were interviewed in 1971 as freshmen and again

124

Philip J. Cook and Robert H. Frank

proportional differences in salary persist throughout the career, they are more than sufficient to justify the higher tuition at elite schools. Why is a degree from an elite private school of greater value than a degree from a less selective institution? One explanation, mentioned briefly above, is that it serves as a signal of quality that can be observed at low cost by employers, customers, and other potential transactors. On this view, the reputation of the school generates tangible benefits for graduates, independent of their own abilities and knowledge. While the social value of such signaling mechanisms is positive, it tends to be less than its total private value to the individuals who obtain elite degrees. After all, those who fail to obtain such degrees bear the stigma of a negative labor market signal, a cost that is external from the perspective of elite-degree seekers. The result is that, on signaling grounds, top students face too large an incentive to expend resources in pursuit of elite degree^.^ (For a discussion, see Arrow 1973.) A second explanation for the value of an elite degree is suggested by the economic literature on tournaments. This literature has focused on cases where employers deliberately link compensation to rank-order performance measures as a means of eliciting greater effort from their worker^.^ There may be a similar, albeit unintended, consequence of the intense competition among top students for admission to elite universities. For example, such competition undoubtedly induces many top students to devote more time to their schoolm r k . Elite schools also emphasize that they are looking for well-rounded students, which may cause some students to join organizations or go out for athletic teams and spend less time experimenting with drugs or playing video games. But the consequences of tournament incentive schemes are by no means uniformly positive. Often such schemes result in contestants engaging in a variety of “arms races” that contribute little or nothing to output. For example, much private expense is incurred, with little resulting increment in social value, when high school seniors take Stanley Kaplan courses in order to boost their scores on the Scholastic Aptitude Test.

in 1980. They found that, other things equal, the alumni of “prestige” schools were more likely to go on to an advanced degree. Questions of possible endogeneity of certain characteristics, and of generally unobserved heterogeneity, create substantial specification uncertainty for this type of study. For example, it is possible that personal characteristics that are generally unobservable by labor economists may be taken into account in the admissions decisions of elite schools. If such characteristics play an important role in salary determination, the observed earnings premiums of elite-school graduates may tend to overstate the additional value added by elite schools. 4.By all accounts, the Japanese system is much more extreme than the American system in this respect. The contest for admission to a prestigious university is fierce and expensive for students and their parents alike. Yet the quality of education offered even at top-ranked Tokyo University is quite poor. “Actually the entrance examinations themselves perform one of the university’s most significant functions, for they, more than a student’s work while at the university, help sort Japanese out for their lifetime careers” (Reischauer 1988, 195). Many businesses invite only candidates from the more prestigious universities to take their employment examinations. 5. See, for example, Lazear and Rosen (1981).

125

The Growing Concentration of Top Students at Elite Schools

There are still further possible effects of the interuniversity tracking of students. For example, top students may receive a better education at elite schools for a variety of reasons, including the following: 1. An outstanding student body helps the university recruit outstanding faculty, since faculty members tend to prefer teaching bright students. 2. The curriculum and the standards for student performance in the classroom will be influenced by the quality of the student body, so that a highly selective school will be more likely to challenge bright students. 3. Students learn from each other outside the classroom, and the quality of such interaction at a highly selective school will tend to be more educational than at other schools. 4. The alumni of a school form a network of mutual assistance, and the value of this network tends to increase with the success of its members. If we take as given that this network results in a disproportionate number of graduates of elite universities being allocated to the most important industry and government jobs, then having additional top students in the network is of social as well as private value. Consideration of these mechanisms suggests that educational tracking is productive, at least for those students who are tracked into the top schools. But there is another side of the tracking coin-namely, that it deprives students at nonelite schools of whatever personal or organizational benefits derive from additional contact with top students. For example, it diminishes the value of the honors curriculum that many large state universities offer to their best students. And when outstanding faculty members are drawn to an elite school by the effects of tracking, students in the nonelite schools no longer receive the benefit of their services. A related cost of tracking is that it diminishes the opportunities for “late bloomers”-those whose true high academic potential becomes apparent only after beginning college-to interact with other students and faculty of high ability. The increasing concentration of top students in elite schools thus involves a trade-off: greater value added in the elite schools comes at the expense of diminished value added in some nonelite schools. Given the nature of the externalities involved, there is no assurance that the private decisions of individual students will resolve this trade-off in a socially optimal way. Suppose, for example, that the social product of having more elite students in any given university exhibits sharply diminishing returns. If our goal is to maximize net output from the educational system as a whole, such a technology might favor a relatively even distribution of top students across universities. Yet the private incentives that encourage top students to concentrate in a small number of elite universities would operate with the same force under that technology as under any other technology.6 6. The extensive literature on two-sided matching (see, for example, Roth and Sotomayor 1990) is primarily concerned with the ability of particular mechanisms to produce stable allocations when individual preferences are not affected by the assignment of other participants. (See

126

Philip J. Cook and Robert H. Frank

The allocation of students across universities has implications not only for efficiency but also for equity. Some commentators worry, for example, that family income plays too large a role in the process by which students are distributed among schools. The average family income of students attending elite colleges and universities is far higher than the average family income of students at nonelite schools. Several studies (including Hearn 1990; Spies 1990) report that family income is an important predictor of who applies to and attends an elite school, even after controlling for high school grades, standardized test scores, parents’ education, and other personal characteristics. This difference persists even though postwar admissions policies at elite private schools have become largely meritocratic. The equity concern is that students in the top quintile of the income distribution are able to take advantage of the high returns to an investment in an elite education, while middle-class students of equal ability are relegated to an education with significantly lower value. Kingston and Lewis (1990) object that this pattern has the effect of perpetuating class differences, although they note that only a small percentage of the students from any socioeconomic status category, including the top group, attend elite schools. The net impact of these opposing normative concerns is far from clear. A more complete account of the issues would require an analysis of the technology of higher education, and of the extent to which students with the most to offer their college peers can be identified on the basis of high school records. Such an analysis is well beyond the scope of our effort here. But the distribution of top students surely affects both the productivity of the educated work force and the extent to which ability determines economic success. For these reasons alone, it is an issue well worth studying.

4.2 Current Concentration

To what extent are students with the greatest scholastic aptitude and prornise currently concentrated at the elite schools? In this section, we offer several statistics suggesting that a large percentage of students who are qualified for admission to one of the elite schools actually matriculate at those schools. We then offer some suggestions about the process that produces this result. One way to identify college-bound seniors with the greatest promise is to utilize the lists of winners of national merit-based prizes. We obtained data on the Westinghouse Science Talent Search and the Presidential Scholars Program.’ The Westinghouse Science Talent Search, initiated in 1942, is a program that identifies high school seniors talented in science, mathematics, and Gale and Shapley [1962] for an early application to college admissions.) Our concern is with the more difficult case where externalities are important. 7. We also contacted the National Merit Scholarship corporation but were unable to obtain data from them on the trend in college choice by National Merit Scholars.

127

The Growing Concentration of Top Students at Elite Schools

engineering. Each year 40 finalists are selected on the basis of applicants’ reports on an independent research project, their high school records, and standardized test scores. Among other benefits, the finalists receive a five-day, all-expense-paid trip to Washington, D.C.; a chance to receive a Westinghouse scholarship; and a letter of recommendation in support of their college applications. Using data provided by the Science Service, we were able to calculate the number who matriculated at each college during the three decades since 1960. The top seven schools on this list were all elite private universities; fully half (50.4 percent) of the finalists matriculated at one of these schools (table 4.1). The most popular choice for finalists during this period was Harvard, where one-fifth of all finalists matriculated. The Presidential Scholars Program was established in 1964 “to recognize and honor our nation’s most distinguished graduating high school seniors.”8 Under current procedures, 2 winners are selected from each state and up to 15 winners chosen at large on the basis of standardized test scores, high school transcripts, essays, and other materials. The White House Commission on Presidential Scholars provided us with data on college choices by winners for the period 1987-89. As in the case of the Westinghouse finalists, the top seven choices accounted for half (49.7 percent) of the total (table 4.2). Harvard alone matriculated 18 percent of the Presidential Scholars, and the top five universities are the same as for the Talent Search winners. We also sought information on college choices by the much larger group of high school seniors who have not necessarily won one of these prizes but have credentials sufficient to gain admission to one of the most selective schools. One method for identifying members of this group (albeit with a large number of errors of both types) is the Scholastic Aptitude Test. SATs are taken by all but a few students who intend to apply to a selective college. Unfortunately, the College Board cannot provide data on college choice of high school seniors taking the test. However, we were able to approximate this distribution by using the tabulations of freshmen scores provided in Peterson’s Guide to Four-Year Colleges. This guide reports the fraction of each freshman class that scored above 500, 600, and 700 on each of the SAT tests (verbal and math). The most selective of these six categories, which best identifies the top students, is defined by a score above 700 on the SATV. In 1989 only 9,510 (less than 1 percent) of the 1.1 million seniors who took the SAT scored this high. Of this group, we estimate that 4,075 (42.8 percent) matriculated at 1 of the 33 colleges and universities designated as “most competitive” by Burr~n’s.~ Since these schools matriculated only 2.4 percent of the seniors taking the SAT in that year, this result demonstrates an extraordinary degree of concen8. Quote from the White House Commission on Presidential Scholars, The 1990 United States Presidential Scholars Program fact sheet. 9. We used the list from the 1980 edition of Burron’s.Colleges were rated by several factors to determine the competition for admission, including entrance exam scores and high school grades of the freshman class, as well the proportion of applicantsto whom the college offered acceptance.

128

Philip J. Cook and Robert H. Frank

Table 4.1

Entering Freshmen: Westinghouse Science Talent Search Finalists, 1960-89 College Harvard MIT F'rinceton Stanford Yale Cal Tech Cornell All others Total

Number of Finalists

Percentage of Total

229 103 72 58 50 38 32 573 1,155

19.8% 8.9 6.2 5.0 4.3 3.3 2.8 49.6 100.0

Cumulative 19.8% 28.7 35.0 40.0 44.3 47.6 50.4 100.0

Source: Calculated from unpublished data provided by Science Service.

Table 4.2

Entering Freshmen: Presidential Scholars, 1987-89 Number of Scholars

Percentage of Total

Harvard Princeton Stanford Yale

54 28 26 12

MIT Duke

II

18.2% 9.5 8.8 4.1 3.1 3.0 2.4 50.3 100.0

College

Michigan All others Total

9 7 149 296

Cumulative 18.2% 27.7 36.5 40.6 44.3 47.3 49.7 100.0 ~~

Source: Calculated from unpublished data provided by the White House Commission on Presidential Scholars.

tration.l0 If anything, this measure tends to be an understatement because some of the seniors with a high SATV were not qualified for admission to an elite school. If it were possible to exclude them from our tabulation, the resulting measure of concentration would be still higher. The top four universities (Harvard, Princeton, Stanford, and Yale) had a combined freshman class equal to only 0.5 percent of all those who took the SAT but included 17.5 percent of all those scoring above 700 on the SATV. The statistics on top students suggest that college-bound seniors who are qualified for admission to the most selective colleges and universities are likely to attend one of these schools. Since elite universities are not cheap, it appears that a large portion of the relevant market agrees that the product offered by these schools is of relatively high quality. 10. The same list of schools matriculated 61 percent of the Presidential Scholars for the period 1987-89 and 60 percent of the Westinghouse Talent Search winners for the period 1960-89.

129

The Growing Concentration of Top Students at Elite Schools

4.3 Quality and College Choice In this section, we argue that there is an interaction between perceived quality and the distribution of top students and that this interaction produces a positive feedback effect that tends to magnify or reinforce initial differences in perceived quality. A recent review (Conrad and Blackburn 1985) found that while quality is “analogous to pornography in its elusiveness” (p. 284), the evidence suggests that perceived quality in higher education is closely related to student achievement. According to one survey of top high school students and their mothers, students tended to judge college quality primarily on the basis of the achievements of the student body, while mothers tended to place greater emphasis on the admissions rates of alumni to graduate and professional schools (Litten and Hall 1989).” In practice, of course, the two measures are highly correlated. In a related finding based on college choices of high school seniors, Fuller, Manski, and Wise (1982) report that applicants tend to prefer colleges that matriculate students whose SAT scores exceed their own. I 2 These findings imply the existence of a positive feedback process. If applicants judge a school’s quality partly by the accomplishments of its students (both in high school and after graduation from college), then an upgrade in student quality improves the reputation of the college and demand for its services, thereby making it easier for the college to improve the quality of its other resources as well (McPherson and Winston 1988). Thus, an initial improvement in reputation for whatever reason will generate improvements in the quality of the student body, which in turn leads to a further improvement in reputation.13 As Brian Arthur (1990) observes, an industry in which such positive feedback processes are important may evolve in certain distinctive ways, including “lock-in’’ through historical events and no guarantee of shared markets. These possibilities seem to be at least partly realized in the market for prestige in undergraduate education. l4 To illustrate the workings of the feedback process, consider a simple model with two types of students, “ordinary” and “top.” Top students are always

11. See Krukowski (1985) for evidence that students’ definition of quality shifted during the early 1980s to focus more on the postgraduate success of the student body. 12. The authors found that for a student with given ability, the utility of an alternative first increases fairly linearly with its “performance standard but eventually turns down. Based on which college they choose, students appeared indifferent between schools with the same SAT and those that are 300 points higher, given that they had applied and been admitted to both. The optimum appears to be 100 points higher than the score of the student. The authors employed data from the National Longitudinal Survey of High School Seniors, Class of ’72. 13. Commenting on the University of Pennsylvania’s campaign to broaden its market and improve its image during the early 1980s, Provost Thomas Ehrlich noted: “The wonderful thing is that the more successful you are, the more successful you are. The more you hear Penn is the institution of choice, the more you want to come” (Walton 1986). 14. One piece of evidence concerning the importance of history is the geographical mismatch between elite colleges and students. The fact that most of the elite colleges are located in the Northeast reflects the geographic distribution of college students in the 18th century.

130

Philip J. Cook and Robert H. Frank

admitted to and attend elite schools if they apply. In figure 4.1, D, represents top students’ initial demand curve for slots in elite schools, shown as upward sloping to represent the fact that elite schools become more attractive to top students when there are more such students enrolled in the elite schools. The initial equilibrium occurs at El, where the proportion of top students seeking admission to elite schools is exactly in balance with the proportion of top students enrolled in these schools. Now imagine an upward shift in the demand for elite schools by top students (the result, say, of increased recruiting by elite schools). At S , (the original proportion of top students in elite schools), there will now be a proportion N, > S, of top students who desire positions in top schools. Note that at the new equilibrium, E,, the increase in the proportion of top students enrolled in elite schools (S, - S,) is larger than the original upward shift in the demand schedule (N, - N,). This difference represents the feedback, or multiplier, effect discussed earlier. I s

4.4 ’Iirends in Concentration While the reputational ranking of colleges and universities is very similar now to what it was several decades ago, there is evidence that the importance of reputation in the competition for top students has increased in recent years. We discuss this evidence in this section. For two of the measures of concentration presented above, we have sufficient data to determine whether there has been any shift over the last several decades. For the Westinghouse Science Talent Search finalists, we present a decade-by-decade comparison in table 4.3. While there was only a small rise in concentration between the 1960s and 1970s, the 1980s showed a substantial increase: 59 percent of 1980s finalists chose one of the top seven schools, compared with 48 percent in the 197Os.I6 Between 1979 and 1989 there was an increase from 32 to 43 percent of students scoring over 700 on their SATV who chose one of the “most competitive” colleges on the Burron’s list (table 4.4), even though the number of matriculants at these schools increased only slightly during this period.” The counts in this table tell the story in more detail. First, the number of students taking the SAT was approximately the same in the two years, but the number who scored above 700 on the SATV dropped from 12,879 to 9,510. Second, the number of these high scorers who matriculated at one of the elite schools remained roughly constant (4,166 in 1979 compared with 4,075 in 1989). Thus the elite schools captured a larger percentage of a smaller pool in 1989, which accounts for the increase in the concentration statistics. Another way 15. The equilibria shown in figure 4.1 are stable because the demand curves are less steep than 45 degrees. If they were steeper, no interior equilibria would be observed. 16. This difference is statistically significant at the 1 percent level.

17. The list of schools was the same for 1979 and 1989 and was taken from the 1980 Barron’s.

131

The Growing Concentration of Top Students at Elite Schools

Number of top students seeking admission to elite schools (as a % of elite school capacity)

/

H

,

I

I

I

I

I I I

I I 1

I

% of elite school class who are top students

s2

S1

Fig. 4.1 The multiplier model Table 4.3

Entering Freshmen: Westinghouse Science Talent Search Finalists, by Decade College

1960s

1970s

1980s

Harvard Top three* Top seven**

17.7% 32.8 46.5

18.9% 33.5 48.1

21.6% 38.6 58.9

Number of finalists

396

370

389

Source: Calculated from unpublished data provided by the Science Service. *Harvard, MIT, Princeton **Hanard, MIT, Princeton, Stanford, and Cal Tech were among the top seven during each of the three decades. The other two were Columbia and Chicago (1960s), or Yale and Cornell (1970s and 1980s).

of seeing this is that in 1979, the elite schools drew one-sixth of their combined class from the top 1.3 percent of the SATV distribution, whereas in 1989 they drew almost the same fraction of their combined class from the top 0.9 percent of the SATV distribution. The most plausible explanation for this increase in concentration was that the top students were more likely to seek admission to one to the elite schools in 1989 than they were a decade earlier.18 This explanation is supported by other data, presented below. 18. It is interesting to note the corresponding trends for those scoring above 700 on the SATM (mathematics). This category, which included fully 30,539 students in 1979, was not nearly as exclusive as having an SATV score that high in 1979, and it became still more common in 1989,

Philip J. Cook and Robert H. Frank

132 Table 4.4

Entering Freshmen Nationwide, 1979 and 1989 Number of Freshmen At Most Competitive Schools*

Percentage

Category

Total

1979 SATV of 700-800 All SATV scores

12,879 99 1,405

4,166 25,004

32.4% 2.5

1989 SATV of 700-800 All SATV scores

9,510 1,088,796

4,075 25,796

42.8 2.4

Sources: Number of freshmen in 1989 (College Board 1990); number of freshmen in 1979 (College Board 1980); distribution of freshmen by SAT for 1989 (Peterson’s Guide 1990); distribution of freshmen by SAT for 1979 (Peterson’s Annual Guide 1981); nationwide distribution of SAT scores (College Board 1979, 1989). *“Most competitive schools” are those listed in that category by Barronk in 1980. Bowdoin and Dartmouth were omitted due to missing data. The distribution of SAT scores for Harvard freshmen was assumed to be the same as in 1989. For Mount Holyoke, the distribution of SAT scores for 1989 freshmen was taken from 1986.

There is also evidence that the trend toward increased concentration of top students in at least some leading universities began well before the 1980s. For example, the median combined SAT score for entering freshmen at Harvard, which stood at 1191 in 1952, had already risen to 1388 by 1965. In absolute terms, the Harvard total has actually shown no further significant increases during the last two decades. For freshmen males who entered in fall 1990, it stood at an even 1400. But because average SAT scores fell throughout the same period, these figures imply a continuing improvement in the relative quality of Harvard’s freshmen. The increase in concentration of top students at Harvard and other elite schools does not appear to be the result of a change in relative prices of private and public education. On the contrary, because the price of attending an elite private school has been increasing in relative terms over the last two decades (Schenet 1988; Clotfelter 1990), the observed increase in concentration must have resulted from an increase in demand for elite universities. Clotfelter (1990) argues that such a demand shift has resulted in part from the substantial increase in the income and wealth of households in the top fifth

when the number was 52 percent higher at 46,435. The number of these students who matriculated at one of the elite schools increased by 34 percent during the decade, from 8,548 to 11,446. These results do not tell us much about the preferences of top students, since many of these high SATM scorers were not qualified for admission to the elite schools. The number of students scoring above 700 on the SATM exceeded the number of slots in the elite schools’ combined freshman class in 1979 and greatly exceeded it in 1989.

133

The Growing Concentration of Top Students at Elite Schools

of the income distribution, which supply a disproportionate share of the students for the elite schools (Schapiro, O’Malley, and Litten 1990). He notes that between 1977 and 1987 the average income of households in the top quintile increased in real terms by 12.5 percent. Stock and real estate values increased sharply during this period, and there were two cuts in the top rate of the federal income tax. We add that this period was also notable for the reduction in the average number of siblings of college students, making a high-cost education more affordable. For example, according to results from the American Freshman Survey, the number of freshmen at private universities whose parents had five or more dependents fell from 45 percent in 1979 to 28 percent in 1987. But affordability is not the whole explanation. A recent study by Richard Spies (1990) finds a large increase in recent years in the probability that a student with given characteristics, including family income, would apply to an elite private school.19 He conducted two surveys of high school seniors with high PSAT scores, one in 1976 and the second in 1987. Based on their responses, he reports estimates of the probability of application to at least one of a group of 33 elite colleges and universities (all of them selective, private, and expensive). His ordinary least squares (OLS) regression results control for a variety of factors, including family income. Using these results to compare the two years, we estimate the probability of application for a student with the following characteristics: white Protestant male only child, financially dependent on his parents, resident of the Middle Atlantic states, public high school graduate in the top 10 percent of his class, father with college degree, applicant for financial aid, family income is $40,000 in 1987 dollars.20Holding all of these factors fixed, the probability of application depends as follows on the SAT score and the year2’: SAT Score 1976 1987

1200 .25 .41

1300 .36 .56

1400 .50 .72

19. In 1987, about 56 percent of those who applied to an elite private school matriculated at such a school. See table 4 of Schapiro, O’Malley, and Litten (1990). 20. The consumer price index doubled between 1976 and 1987, so the nominal value for the earlier year was $20,000. 21. It should be emphasized that the data are for application rather than matriculation. It is possible that there was a downward shift in the likelihood that a student who applied and was accepted to at least one of these schools actually matriculated at one of them. There is evidence, however, that it is relatively uncommon for a student who is accepted at an elite school to decline in favor of a lesser-ranked institution. For example, a 1987 survey (Schapiro, O’Malley, and Litton 1990, 22) indicated that 71 percent of the students admitted to at least one COFHE school (a group of elite private institutions very similar to Burron’s “most competitive” list) matriculated at one of these schools.

134

Philip J. Cook and Robert H. Frank

Although most institutions on Barron’s list of most competitive universities are private, a significant number of public schools also have strong academic reputations. We have made some preliminary attempts to check whether demand has also shifted toward relatively more prestigious public institutions. Tables 4.5 and 4.6 provide data on the SAT scores of entering freshmen at the eight campuses of the University of California. All the UC campuses charge the same tuition but differ with respect to prestige and reputation. The campus with the strongest reputation of the eight is Berkeley, and the results demonstrate that there has been a marked increase during the 1980s in the concentration of students with the greatest scholastic aptitude on that campus. For example, the odds ratio that a UC freshman with an SATV of at least 700 would matriculate at Berkeley increased from 2.3 in 1980 to 5.8 in 1988.22 In sum, there is considerable evidence that students qualified for admission to an elite school were more likely to choose such a school in the late 1980s than they were a decade earlier. There is also evidence that the trend toward increased concentration began well before the 1980s. Further, the observed changes cannot be accounted for in terms of trends in tuition and other costs; nor did they result solely from changes in the income distribution. There are many other possible explanations for the observed increases in concentration, several of which we list below: Numerous social commentators have described the 1980s as a time of increased materialism, conspicuous consumption, and brand-name consciousness. The colleges with the most prestigious brand names may have been the beneficiaries of this general cultural shift. The proliferation of publications offering national rankings of colleges and universities may be one quantifiable aspect of this shift. During this period, there was a considerable increase in colleges’ and universities’ expenditures on recruiting students, brought on in part by concerns engendered by the declining population of 18-year-olds. This effort may have encouraged college-bound seniors to consider schools that they otherwise would have ignored. We know that college applicants as a group invested more in “shopping” for the right option: in 1988, 37 percent of college freshmen said they had applied to three or more colleges, a higher percentage than ever before (Astin et al. 1988, 8). Only 15 percent applied to that many in 1968. The shift may be related to trends in the job market for entry-level managers and professionals, including greater emphasis on educational credentials and a relative decline in preference for graduates of local colleges and uni22. A similar analysis of freshmen in the University of North Carolina system (where Chapel Hill has the strongest reputation of the 16 campuses) revealed a different pattern. Unlike what we saw in the California system, there has been little change in the degree of concentration of top students in the UNC system.

The Growing Concentration of Top Students at Elite Schools

135 Table 4.5

Entering Freshmen, University of California: Distributionby SAT Scores, 1980, 1984, 1988 Verbal

SAT Score

InUC System

At UCB

86 362 856 1,773

31 110 271 456

17,732

2,943

68 365 899 1,901

Math Percentage

InUC System

At UCB

36.05% 30.39 31.66 25.72

410 1,114 2,033 3,178

165 387 498 616

16.60

17,729

2,943

16.60

47 181 374 635

69.12 49.59 41.60 33.40

438 1,538 2,775 3,622

244 621 848 810

55.71 40.38 30.56 22.36

20,714

4,040

19.50

20,716

4,041

19.51

53 514 1,479 2,643

38 245 518 622

71.70 47.67 35.02 23.53

713 2,369 3,806 4,423

349 723 686 498

48.95 30.52 18.02 11.26

23,393

3,441

14.71

23,393

3,441

14.71

Percentage

I980 750-800 700-740 650-690 600-640 All freshmen with SAT scores

40.24% 34.74 24.50 19.38

I984 750-800 700-740 650-690 600-640 All freshmen with SAT scores 1988 750-800 700-740 650-690 600-640 All freshmen with SAT scores

Source: Calculated from unpublished data provided by the Office of the President, University of California.

Table 4.6

Entering Freshmen, University of California: Odds-Ratioof Attending Berkeley, 1980,1984, 1988 Verbal

SAT Score 750 700 650

+

+ +

Math

1980

1984

1988

1980

1984

1988

2.8 2.3 2.3

9.2 4.6 3.4

14.7 5.8 3.7

3.4 2.9 2.1

5.2 3.2 2.3

5.5 3.1 2.0

~~

Source: Calculated from data in table 4.5. Note: The Odds-ratio is defined as the odds that a UC freshman with a given SAT score attends Berkeley, divided by the odds that any UC freshman attends Berkeley.

136

Philip J. Cook and Robert H. Frank

versities. (We present some relevant data on recruiting practices in the next section.) With the data available to us we can do little to quantify the relative strength of these explanations. But regardless of the sources of exogenous shifts in college preference, we suspect that these sources by themselves account directly for only part of the observed increase in concentration. The indirect effects of the positive feedback process described earlier, whereby college choices by top students influence and are influenced by colleges’ reputations, may also figure prominently in this process.

4.5 On-Campus Recruiting One factor that may influence top students’ college decisions is their perception of the extent to which attendance at different schools helps them land favored jobs. Thus, for example, top students should find a university more attractive if favored employers actively recruit at that university. For their part, elite employers have an obvious incentive to focus on universities that attract top students. The causal relationship between the elite recruiter’s choice of universities and the top student’s choice of universities thus runs in both directions-another positive feedback loop of the sort mentioned earlier. As top students become more concentrated in elite universities, elite firms will concentrate more of their recruiting in those universities. And this makes elite universities still more attractive to top students. As part of our inquiry into the causes of increased concentration of top students, we conducted a survey of past, current, and expected future recruiting practices of a sample of firms who recruit at Cornell University. Cornell appears in Burron’s list of most competitive universities, and the recruiters who visit the campus gave evidence in their survey responses that much of their other college recruiting takes place on similar campuses. One of our questions, for example, was “Roughly what percentage of the colleges you will visit this year are among those that consistently rank among the top 25 national universities?” Responses averaged 49.7 percent ( N = 60). We also asked respondents to report whether the ratio of top-ranked campus visits to total campus visits has increased, decreased, or remained the same over the past ten years. 35 percent of our respondents reported an increase, only 13 percent a decrease. The remaining 51 percent reported no change (N = 82).23 When asked how they expected their proportion of visits to elite universities to change in the future, 22 percent of our respondents expected an in23. There is a potential selectivity bias in our sample of Cornell recruiters. As we explain below, this bias does not appear significant.

137

The Growing Concentration of Top Students at Elite Schools

crease, while only 10 percent expected a decrease. 68 percent expected no change from the current ratio (N = 82). We observed an essentially similar pattern of responses when we asked how the proportion of interviews (as opposed to campus visits) conducted at topranked universities has changed over the past 10 years. Thirty-six percent of respondents reported an increase in this proportion while only 14 percent reported a decrease ( N = 80). When asked about their expectations concerning future changes in the proportion of total interviews conducted at top-ranked universities, 24 percent expected an increase, while only 8 percent expected a decrease ( N = 80). Although it appears safe to assume that companies recruiting at Cornell are actively in the market for top students, there was significant variation in the extent to which our respondents confined their attention to such students. Some respondents reported that as little as 5 percent of their recruiting took place on top-ranked campuses, while others reported 100 percent. Similarly, while most firms who recruit at Cornell would probably not bother to do so if they felt they had little chance of appealing to top students, not all firms in our sample are equally attractive to such students. With both of these sources of variation in mind, we constructed a subsample of firms that were either more selective than others in their recruiting efforts or more likely than others to appeal to top students. Our goal in constructing this subsample of elite firms was to test the hypothesis that the increase in concentration of recruiting efforts at top-ranked universities is more pronounced for elite firms than for other firms in our sample. On the selectivity dimension, a respondent was included in the elite-firm subsample if it conducted at least 70 percent of its campus visits and total interviews at top-25 universities. In terms of attractiveness to top students, a respondent was included in the subsample if it met at least one of the following criteria: (1) it appeared on the Levering list of “the 100 best companies to work for”; ( 2 ) it was one of the top three firms in its four-digit industry in terms of annual sales revenue. The result was a subsample consisting of 39 respondents, which we call “elite firms.” 43 respondents were excluded on the basis of these criteria. As shown in figure 4.2, the observed pattern of changes is the one we expected. During the last decade, 41 percent of elite firms had increased their proportion of visits to top-ranked universities, while only 8 percent of elite firms decreased that proportion ( N = 39). The corresponding figures for other firms are 30 percent and 19 percent, respectively ( N = 43). We now consider the possibility that the reported changes in behavior for our sample of current Cornell recruiters may not be representative of the changes in behavior for recruiters as a whole. Suppose, for example, that in the total population of firms that recruit on college campuses, some have become more likely to recruit at elite campuses in the past ten years, others less

138

Philip J. Cook and Robert H. Frank

Change in the Ratio of Top-25 Visits to Total Campus Visits (%)

40 35

30 25 20 15

10

5 0

5

4

Elite Firms

Other Firms

Fig. 4.2 Changes in the proportion of recruiting visits to top-ranked universities: Elite firms versus other firms.

likely. Our 1990 cross section of Cornell recruiters may then contain disproportionately many representatives from the former category, causing our estimates to overstate the increase in concentration. To explore this possibility, we searched Cornell’s placement records and found that 36 firms from our sample have been recruiting at Cornell for at least the past ten years. Focusing on this subsample alone, we found that 17 of these firms had increased the proportion of their total interviews conducted on elite campuses, while only 6 had reduced that proportion; 13 firms reported no change. We conclude that selectivity bias does not seem to be a serious problem in this instance. Cornell’s undergraduate placement director, Thomas Devlin, told us that he has observed a steady trend toward more targeted recruiting over the past two decades. He reports that firms have become steadily less likely to choose campuses on the basis of geographic proximity and increasingly likely to choose on the basis of student characteristics. His impressions are thus consistent with the responses of the firms we surveyed. Both lend support to the more general claim that top students have more to gain now than in the past by attending an elite university. The increased focus of elite corporate recruiters on elite campuses suggests a specific mechanism whereby the signaling function of elite schools can generate large costs that would otherwise be avoidable. For example, a top student might once have found it attractive to attend a nearby state university because the presence of other top students there meant that he would be sought

139

The Growing Concentration of Top Students at Elite Schools

out by employers upon graduation. But once sufficiently many top students migrate from state universities to elite schools, this is no longer a safe presumption. By going to the nearby state university, the top student may be much more likely to be overlooked by elite employers and graduate schools. The elite university’s higher tuition and greater distance from family represent painful sacrifices for many top students; but they are sacrifices that many feel they can no longer avoid.

4.6

Concluding Remarks

In this paper, we have presented evidence that a large and growing proportion of the nation’s top students are concentrated in a relatively small number of top-ranked universities. We have also noted that the process whereby students choose universities is characterized by an assortment of externalities and positive feedback effects at several levels. Finally, we have suggested that there are many possible social welfare consequences of increased concentration-some positive, others negative. Given the externalities and positive feedback effects inherent in the individual’s college choice, there can be no presumption that the current aggregate distribution of students is socially optimal. In view of the apparent trend toward increased concentration of top students, additional research on the welfare effects of such concentration deserves high priority.

References Arrow, Kenneth. 1973. Higher education as a filter. Journal of Public Economics 2:193-216. Arthur, W. Brian. 1990. Positive feedbacks in the economy. Scientijic American, Febr~ary:92-99. Astin, Alexander W. 1968. Undergraduate achievement and institutional “excellence.” Science 161:661-68. Astin, Alexander W., K. C. Green, W. S . Kom, M. Schalit, and E. R. Berz. 1988. The American freshman: National norms for fall 1988. Higher Education Research Institute, University of California, Los Angeles. Microfiche (ERIC ED303 133). Caminiti, Susan. 1990. Where the CEOs went to college. Fortune, June 18: 120-22. Clotfelter, Charles T. 1990. Undergraduate enrollments in the 1980s. Duke University Center for the Study of Philanthropy and Volunteerism, Durham, N.C. Photocopy. College Board. 1979. National college-bound seniors, 1979. New York: College Entrance Examination Board. . 1980. The College handbook 1981. New York: College Entrance Examination Board. . 1989. National report: College-bound seniors: 1989 profile of SAT and achievement test takers. New York: College Entrance Examination Board. . 1990. The college handbook 1991. 28th ed. New York: College Entrance Examination Board.

140

Philip J. Cook and Robert H. Frank

Conrad, Clifton F., and Robert T. Blackburn. 1985. Program quality in higher education: A review and critique of literature and research. In Higher Education: Handbook of Theory and Research, ed. John Smart. Vol. 1. New York: Agathon Press. Fuller, Winship C., Charles F. Manski, and David A. Wise. 1982. New evidence on the economic determinants of postsecondary schooling choices. Journal of Human Resources 27(4):477. Gale, David, and Lloyd S. Shapley. 1962. College admissions and the stability of marriage. American Mathematical Monthly 69(January): 9-1 5. Griffin, Larry, and Karl Alexander. 1978. Schooling and socioeconomic attainments: High school and college influences. American Journal of Sociology 84:3 19-47. H e m , James C. 1990. Pathways to attendance at the elite colleges. In The high-status track: Studies of elite schools and stratijication, ed. Paul W. Kingston and Lionel S. Lewis. Albany: State University of New York. James, Estelle, Nabeel Alsalam, Joseph C. Conaty, and Duc-Le To. 1988. College quality and future earnings: Where should you send your children to college? State University of New York at Stony Brook. Photocopy. Kingston, Paul William, and Lionel S. Lewis, eds. 1990. The high-status track: Studies of elite schools and stratijication. Albany: State University of New York Press. Kingston, Paul William, and John C. Smart. 1990. The Economic pay-off of prestigious colleges. In The high-status track: Studies of elite schools and stratijication, ed. Paul W. Kingston and Lionel S. Lewis. Albany, State University of New York. Krukowski, Jan. 1985. What do students want? Status. Change, May-June:21-28. Lazear, Edward, and Sherwin Rosen. 1981. Rank-order tournaments as optimal labor contracts. Journal of Political Economy 89(5):841-64. Litten, Larry H., and Alfred E. Hall. 1989. In the eyes of our beholders: Some evidence on how high-school students and their parents view quality in colleges. Journal of Higher Education 60(3):302-24. McPherson, Michael S . , and Gordon C. Winston. 1988. Reflections on price and quality in U.S. higher education. Williamstown, Mass.: Williams College. Draft. Peterson’s annual guide to undergraduate study. 1981. Princeton, N.J. Peterson’s guide to four year colleges, 1991. 1990. 21st annual edition. Princeton, N.J. Radner, Roy, and Leonard S. Miller. 1975. Demand and supply in US.higher education. New York: McGraw-Hill. Reischauer, Edwin 0. 1988. The Japanese today: Change and continuity. Cambridge, Mass.: Harvard University Press. Roth, Alvin E., and Marilda Sotomayor. 1990. Two-sided matching: A study in gametheoretic modeling and analysis. Econometric Society Monograph Series. Cambridge: Cambridge University Press. Schapiro, Morton O., Michael P. O’Malley, and Larry H. Litten. 1990. Tracing the economic backgrounds of COFHE students: Has there been a middle-income melt? Williamstown, Mass.: Williams College. Draft. Schenet, Margot A. 1988. College costs: Analysis of trends in costs and sources of support. CRS Report for Congress, Washington, D.C. Spies, Richard R. 1990. The effect of rising costs on college choice. Princeton, N.J.: Princeton University. Walton, Mary. 1986. How Penn became a hot school. Inquirer: The Philadelphia Inquirer Magazine, April 13.

141

The Growing Concentration of Top Students at Elite Schools

Comment

Malcolm Getz

The proper education of a society’s elite has been a subject of discussion in the social sciences at least since Plato. The discussion is naturally linked to issues of defining elite, how one becomes a member of “the” elite, and what good is having an elite, anyway? “The Growing Concentration” is in a long and worthy tradition. These issues have particular relevance for a society in which the results of popular elections have great influence in shaping institutions and in which we have ongoing lovelhate relationships with our elite. Reflection upon the names Lincoln, Roosevelt, Kennedy, and Reagan should make the point. My comments can be summarized by three questions: 1. Is the collegiate education of the “elite” being concentrated at elite institutions? 2. Why might such a development be occurring? 3. What difference might such a development make?

Is the Collegiate Education of the “Elite” Being Concentrated at Elite Institutions? Before answering this question, one might first want to attempt to define our society’s elite. Cook and Frank look at the chief executive officers of large corporations and at the earnings of graduates of elite private schools. Members of Congress, governors, and members of the federal judiciary might be a political elite. The persons with the highest income or greatest wealth (which would include certain entertainers, sports figures, and a variety of persons other than CEOs) might be included in a monied elite. The medical field has its own elite, as do other fields such as the arts, science, engineering, and the military. The overlap among elite groups may be relatively small, an important observation in itself. The degree of concentration of collegiate education of each of the several elite groups might then be examined separately with a view to whether the finding for CEOs in Cook and Frank holds true for the elite of other groups as well. The point is that our society’s elite is pluralistic, and no single metric will measure all of it. Given this pluralistic character, one might find that the concentration of the elite at “elite” schools is less than Cook and Frank indicate. Having defined the elite, we turn to the question of whether their collegiate education is concentrated among elite institutions. What is an elite institution? The essay identifies seven schools for some purposes and uses Barron’s “most competitive” (forty-four schools in the 1990 edition) for other purposes. The lists include at least one public institution and a mix of research universities, doctoral institutions, and liberal arts colleges. This group does not by itself Malcolm Getz is associate provost for information services and technology, director of the Jean and Alexander Heard Library, and associate professor of economics at Vanderbilt University.

142

Philip J. Cook and Robert H. Frank

seem monolithic. The contrast pointed out by Cook and Frank between the Japanese system and the American system is striking. By comparison, the elite collegiate institutions in America seem to have a decidedly diluted concentration of elite students. Moreover, the rates of concentration of top students at the American elite institutions is hardly alarming. That 18 percent of 300 Presidential Scholars chose Harvard is interesting but is probably not enough to give Harvard a monopoly on elite students. The essay portrays a trend toward increased concentration of top students in elite institutions. It would be helpful to know when the phenomenon may have started. Although higher education has used entrance exams for centuries, the goal prior to 1945 was principally to assure adequate preparation for collegiate work. After 1945, entrance examinations functioned to promote selectivity in an effort to increase the average ability of the entering class. If the phenomenon dates back no further that 50 years, its full consequences may not yet be manifest. The notion of changing patterns of concentration may have more meaning in a longer historic perspective. Do the elite schools know their admission and yield rates over the last century? How does their experience compare with that at somewhat less selective institutions? Does the phenomenon the authors demonstrate for top institutions hold true to lesser degrees down the line?

Why Might Such a Development Be Occurring? The essay cites evidence that graduates of elite institutions enjoy higher earnings, perhaps even sufficiently higher to justify their cost. As the essay notes, this phenomenon could reflect the ability of the schools to identify able people, and it could reflect a higher quality of education. In support of the latter, the essay gives several plausible explanations to which I might add two. First, the institutions may be better managed than others and therefore do a better job of turning dollars into educational experiences. They may be less bound by civil-service rules, more aggressive in selecting and rewarding faculty, and more successful in identifying and responding to student and faculty interests. They may also do a better job of marketing. Second, private institutions tend not to have their fortunes rise and fall with the yield of state taxes. Therefore, their operations may be less susceptible to economic downturns than are public institutions whose own health may rise and fall with the state’s fiscal health. The essay also identifies a “snowball” mechanism, wherein the successes of graduates make the institution more attractive to subsequent applicants. As the school is more successful, it can be more selective and so enhance its future appeal and success. I might suggest a second such mechanism, albeit one with a longer period for realization. As successful graduates advance in their careers, they can contribute more generously to their alma maters. As an institution’s fund-raising appeals are more successful, the institution becomes better off financially and can use its improved resources to make itself more

143

The Growing Concentration of Top Students at Elite Schools

attractive to prospective students. It is hardly coincidence that the most selective institutions are also the best endowed.

What Difference Might Such a Development Make? Even a social welfare function as egalitarian as that proposed by John Rawls would value better education for the elite when the role of the elite is important in advancing the welfare of all. If advances in science, medicine, management, politics, warfare, and public policies arise from the good ideas of, and their successful implementation by, the elite, then education that enhances the performance of the elite will be likely to advance not only their own fortunes but also the welfare of the least well off. If the elite schools provide better education for the elite, then increasing concentration of members of the elite in such institutions would seem to be a good thing. The important point here may be that success-bound individuals define the better institutions by selecting them in open competition. If an institution fails to provide suitable experiences, the success bound will choose other, more effective institutions. Concentration would be alarming if it reached the point where institutions had a significant monopoly position, such as might occur under some regulatory scheme guaranteeing that position (e.g., a single national military academy). In such a circumstance, the institution need only satisfy its regulators. Students would not be in a position to exercise selection, and so the function of the marketplace would be lost. There may be such institutions in relatively narrow niches in the United States, but the national elite institutions are not so confined. Concentration of collegiate education for the elite within several dozen institutions may yield market discipline. However, if the institutions are essentially similar in culture and outlook, excess concentration might homogenize members of the elite so as to make them more resistant to “foreign” ideas. One might think of the problem as cultural inbreeding. However, this does not seem to be a particular problem for the American elite schools. The institutions described have some diversity. The level of concentration of collegiate education of the elite is not high enough to suggest that members of the elite are all educated in the same way. Moreover, collegiate education is not a requirement for “elitehood.” The current governor of Tennessee does not have a college degree. Finally, one might be concerned about the effect of the increasing concentration of outstanding students in outstanding schools on the distribution of income and opportunity. What is the bottom-line equity consequence of concentration? That bright people are somewhat more likely to have higher income, and that they are likely to earn even more when they are well educated, does not necessarily violate any ethical principle. Equity concerns might arise in thinking about criteria other than ability that might influence admission to selective colleges. Race, sex, parental wealth, and ethnicity all have influ-

144

Philip J. Cook and Robert H. Frank

enced admissions decisions from time to time. How such considerations should influence-both positively and negatively-admission to college, and by extension admission to the elite, will continue to be controversial. Such controversy will likely surround any and all pathways to elitehood. Given the financial aid, aggressive marketing, and quest for diversity of the dozens of institutions involved, concentration of collegiate education at the levels described in this essay may well improve prospects for access to elitehood for persons of disadvantaged backgrounds or who are otherwise underrepresented. Moreover, public policies and philanthropic institutions are available to modify the balance, should it become desirable to do so.

5

Future Graduate Study and Academic Careers Jerry R. Green

5.1

Introduction

Demographic and economic forces have conspired to create the potential for a significant gap between the need for faculty in the arts and sciences and the supply of new Ph.D.'s to fill these positions.' Many avenues have been discussed through which this shortfall can be accommodated. Among these are increases in student-faculty ratios, decreases in retirements, increases in foreign-trained faculty, decreases in the percentage of high school students continuing to college, and increases in the number of domestically trained Ph.D.'s. To the extent that the last of these is relied upon, the quality of these future teachers and scholars is equally of concern to the future of higher education, especially over the longer run. Bowen and Schuster (1986), for time periods from the end of World War I1 to 1985, document the decrease in the frequency of plans to pursue college teaching among several elite groups: Harvard seniors, Rhodes scholars, and members of Phi Beta Kappa. They do not disaggregate across fields of study. Interestingly, their survey evidence does not indicate a reported decrease in the quality of new graduate students and junior faculty, as perceived by existing tenured members of departments and department chairmen. Bowen and Sosa (1989), whose data are broken down by field of study, show that Graduate Record Examination (GRE) test scores for new Jerry Green is David A. Wells Professor of Political Economy at Harvard University and a research associate of the National Bureau of Economic Research. The author would like to thank Henry Rosovsky, Cecilia Rouse, Barbara Carroll, Larry Litten, and Martha Leape for their assistance in obtaining the data for this study, and Robert Scheinerrnan for tireless and insightful research assistance. 1 . Earlier work along these lines includes Radner and Miller (1975). Fernandez (1978), and Bowen and Sosa (1989). Bowen and Sosa disaggregate the supply of new academics by field of study, which is also the principal focus of this paper.

145

146

Jerry R. Green

potential faculty members declined precipitously up to 1982 but that verbal scores have recovered since then. The authors go on to examine whether this pattern could have been produced by changes in the frequency of test taking in the population of college seniors rather than by an underlying change in the quality of graduates desiring academic careers. The present paper is an attempt to detect trends in the quality of people choosing, or potentially choosing, academic careers. The data for this study come from the Harvard Senior Survey, a virtually complete sample of graduating seniors from Harvard and Radcliffe for the years 1985-90.* The nature of these data is summarized in section 5.2. I pay special attention to certain aspects of college graduates’ knowledge of their own future plans which I think are very important, both for the population of all graduates and for the Harvard students who are the source of the data in the present study. First, one must recognize that many if not most students are quite uncertain about their own plans. Therefore, their choices about what to do upon graduation will reflect a preference for flexibility and a need for information gathering as well as an expression of unalterable career intent. Second, demand for various postgraduation activities must be in balance with supply. Students who choose to work after graduation, at least in certain jobs and under certain conditions, might not always be able to get what they want. Part of the function of working is often to learn about career possibilities. If the “right” kind of job is not available, students might go directly to graduate school instead. Third, even in a time span as short as the six years of data covered in this paper, there are exogenous forces that impinge differentially on students with various career plans. A stock market crash (1987), a recession (1990), a decline in medical school applications (reasons chronicled elsewhere), and a boom in biotechnology and related fields all are reflected in the immediate postgraduation plans and in longer-term, more indefinitely held plans. The principal finding in this paper is that, over the period covered by these data, there has been no overall trend in the prevalence of plans to pursue an academic career in the arts and sciences. However, the most recent year, 1990, may be the beginning of a return to academe, although one cannot be sure. Despite the overall absence of a trend, when disaggregated by field of study and academic record, several systematic tendencies do emerge. Humanities concentrators have greatly increased their interest in academic careers, especially those with middle-ranking grade records. In certain scientific specialties there is also an increase, in this case due primarily to the more outstanding students. This paper also attempts to measure the uncertainty students have about their own careers and the way this is expressed in immediate and subsequent postgraduation plans.

2. Older senior surveys are available, but significant variations in the wording of certain questions make comparisons of responses across years more problematic.

147

Future Graduate Study and Academic Careers

Section 5.3 discusses the overall time patterns of career choice and choice of undergraduate concentration. Particular attention is paid to the academic careers and to students who are unsure of their career plans but list “academic” among several possibilities. Section 5.4 looks at these data in more detail, stratifying by various measures of performance while in college. Section 5.5 is a yet more detailed look, examining trends by fields and relating these changes to the principal alternatives to academic careers for individuals with a given undergraduate concentration. Section 5.6 looks at the paths that individuals plan to take to the goal of an academic career. Some will attend graduate school immediately. Others will work first, or travel, or volunteer. Of these, some plan to take a graduate degree later, continuing to an academic career. Section 5.7 is a check on the match between the source of these data and the inferences we ask it to provide. We explore whether the Harvard undergraduates are perhaps unrepresentative of the supply of Ph.D. candidates nationally, as captured in the more broadly based Consortium on Financing Higher Education (COFHE) survey.

5.2 Description of the Harvard Senior Survey The data used in this paper come primarily from the Harvard Senior Survey for the six years 1985-90. (The senior survey questionnaire for 1990 is reproduced as the Appendix to this chapter.) This is an unusually rich source of data for several reasons. The senior survey is completed by approximately 98 percent of all graduating seniors from Harvard and Radcliffe colleges. The completion rate is so high because one cannot receive commencement tickets without it. Seniors in a given year include all those graduating in that year. Students in earlier entering classes who have taken a leave of absence and are graduating later than most of their classmates are part of the senior survey in the year of their graduation. Correspondingly, members of a given class who have not completed all graduation requirements are not included in the senior survey with their class. Approximately 15 percent of a given cohort do not graduate with their class and are thus not in the population in that year. Our use of the senior survey is to examine plans and expectations for activities immediately after graduation and beyond, including plans to return to graduate school at a later date. For this purpose, it seems best to focus on cohorts who leave college together, as the senior survey does, rather than on those who entered together. To the extent that postgraduation plans have been affected by events during the college years, graduates from earlier classes will have had a different experience than those who are completing a normal fouryear program. Nevertheless, since they will have had time to adjust their plans to the most recent events (e.g., economic conditions), we feel it is best to treat them as the senior survey does and keep contemporaneous graduates, rather than entering classes, together.

148

Jerry R. Green

Apart from the obvious advantage of having a virtually complete sample, the Harvard Senior Survey has a number of other good qualities as a source of data for the problem we are examining. It is annual, and the questions are quite consistent although not perfectly identical across years. This allows us, first, to have more data and, second, to be able to test for the effects of surprise events that will affect a given class but will not be durable in nature. Most important, we can match individuals’ responses in the senior survey to the academic record of the student, to data on their application from high school (including standardized test results), and to their actual career choices as later reflected in reports to the Harvard alumni o f f i ~ e . ~ It must be recognized, however, that data from the Harvard Senior Survey are idiosyncratic to the extent that Harvard students may be atypical. To ascertain the extent of any bias, we compare our results to data from the COFHE 1984 ~ u r v e y We . ~ compare the responses to the COFHE questions asked of Harvard students to responses to similar questions in the Harvard Senior Survey. We also compare responses by Harvard students in the COFHE data to those by students from other colleges and universities. In this way, we hope to learn whether there is any significant response bias in the COFHE sample, which varies in response rate quite widely across institutions, and whether and to what extent Harvard graduates in 1985 are different from those of other institutions in 1984. This is a way of checking our results and provides a better means of extrapolating anything learned from our data to broader populations. Further discussion on this point is deferred to section 5.7. Our focus will be on the choice of an academic career. In the senior survey, a student can list up to three possible career choices, without indicating a priority or relative likelihood among them. Therefore, it is not always possible to make an unambiguous inference about the choice of an academic career. The strategy we adopt is to create three variables related to the revealed choice of career path. Students are often uncertain about their career plans, and to some extent our approach is intended to reflect their level of uncertainty about their own choice. Anyone who lists “academe” as one of their first three possible career choices is classified as “yes” for the variable Academic. Among these, if they plan to attend a graduate school in an arts and sciences field within the next 12 months, we classify them as Trueac. Academics who are not Trueacs are of two types. Either they have no plan for graduate school-for example someone who is going to medical school and answers academe as a career 3 . I feel this will be a significant advantage as this work is pursued further. Many potential academics, even among those who begin an academic career by going to graduate school, do not have lifetime careers as college teachers and researchers. In the future, we hope to use alumni records to trace the career paths of individuals, determine where the best people are lost to academic careers, and uncover perhaps what can be done to retain them. 4. This is a year of a complete COFHE survey for which a follow-up (1989) is available, which we hope to use in later work.

149

Future Graduate Study and Academic Careers

choice, by which he or she may mean a clinical professorship in a medical school-or else they plan to go to graduate school in an arts and sciences field at a later date. We also create a third category, one that is a subset of Academic but not of Trueac, which we call Onlyac. These are people who give academe as their only answer to the career choice question. This is an attempt to single out those who have definite academic career plans. The reason that Onlyacs include some non-Trueacs is that there are a number of Academics who are not going to an arts and sciences graduate school immediately. These include some who plan to work immediately, but more important from the point of view of the Onlyac group are those whose postgraduation plans include travel, volunteer work, or government service. This group includes many highly capable individuals who have won fellowships that afford them this “time off’ before graduate school; indeed most such fellowships require plans of this sort. Many of these people are motivated toward academic careers. In section 5.6, we will look in considerable detail at the exact nature of the postgraduation plans of individuals in the Academic, Trueac, and Onlyac groups.

5.3 Time Pattern of Career Choices and Postgraduation Plans The most striking pattern in all of our results is the surprising constancy of the percentage of respondents who select academic careers. Table 5.1 shows the numbers and percentages of graduates for all career choices, including multiple responses to the career-choice q u e s t i ~ nThe . ~ fraction of seniors who are Academics is essentially constant from 1985 to 1989, varying only from a low of 20.1 percent to a high of 23.5 percent in this interval. In 1990, however, there is a jump to 27.3 percent. Of the most common career choices, medicine and business are quite clearly declining in popularity, while law is increasing. Since at Harvard the training for prelaw and prebusiness is almost indistinguishable (there is no undergraduate business major), it is quite likely that the decline in one is offset by the increase in the other. The decrease in the number of premed students is more complex and may be more related to changes in the number of academics over the longer run. We will not speculate here on the reasons for the decline in medical careers among Harvard undergraduates. However, our data reveal a number of interesting interactions connected with the training expected of medical school applicants. First, as can be seen in table 5.2, there is a massive decrease in the number of premeds among life science concentrators. In 1985 and 1986 there 5 . In survey question V.1 only the first three eventual vocations listed were coded into the database. However, for question V.2, all multiple responses were retained because the priority or importance of different responses could not be determined.

150 Table 5.1

Jerry R. Green Number of Graduates Choosing Career Option

Option Academe Arts Business Communications Desigdarchitecture GovernmenVpolitics Health Helping professions Law Library/museum Medicine Religion Science/technology Skilled tradedfanning Teaching/administration Other Undecided Number of graduates Percentage academic Average number of choices

1985

1986

1987

1988

1989

1990

321 189 369 194 38 213 36 62 238 15 22 1 37 106 10 43 27 15 1,479 22.1%

343 I62 361 161 30 215 56 77 236 20 227 21 87 11 68 30 93 1,457 23.5%

336 160 405 186 48 205 69 82 252 12 22 1 24 92 7 51 39 93 1,488 22.6%

320 180 399 209 43 239 59 89 320 13 203 28 98 10 61 38 103 1,592 20.1%

365 170 364 197 37 244 44 87 320 16 202 24 85 16 68 40 99 1,608 22.7%

424 184 332 214 50 229 47 84 303 17 215 18 99 13 76 41 106 1,555 27.3%

1.49

1.51

1.53

I .52

1.48

I .58

Note: Students may select all that apply.

Table 5.2

Number Choosing Career Option: Life Sciences

Option Academe Arts Business Communications Desigdarchitecture GovernmenVpolitics Health Helping professions Law Librarylmuseum Medicine Religion Science/technology Skilled tradedfarming Teaching/administration Other Undecided

1985

1986

1987

1988

1989

1990

54 I1 12 2 2 6 9 5 4

58 4 9

50 5 25 3 5 7 15 I 5

53 7 17 7

55 5 15 2 0 9 5 3 9 I 83 2 12 2 5 2 3

48 3 8 2

1

0 4 17 4 3

1

1

1

135

I40 0 9

119 2 16 0 2

1

19 1 1

2 2

Note: Students may select all that apply.

1

3 3 8

1

7

1

5 8 5 8 2 122 0 17 1

2 0 8

1

3 8 0 2 0 84 1

15 1

5 I 4

151

Future Graduate Study and Academic Careers

were 135 and 140 premeds in this group, whereas in the two most recent years, 1989 and 1990, these numbers are 83 and 84. In each of the other concentration areas, the number of premeds has actually increased (see tables 5.3, 5.4, and 5.5). Table 5.3

Number Choosing Career Option: Physical Sciences

Option Academe Arts Business Communications Desigdarchitecture Government/politics Health Helping professions Law Library/museum Medicine Religion Science/technolog y Skilled trades/farming Teaching/administration Other Undecided

1985

1986

1987

1988

1989

1990

79 14 69 7 8 11 5 5 12 1 23 4 77 1 5 3 12

59 11 49 7 2 9 1 9

73 5 48 7 6 5 3 3 8 1 17

59

81 11 45 5 1 6 2

25 2 55 1 3 2 6

85 23 23 16 9 11 3 5 15 2 36 1 62 2 6 3 12

10 1 18 1 66

0 6 2 15

I 52 1 2 3 14

10 38 12 3 12 1 2 13 1 18 3 59 0

2 3 12

0 8

0

Nore: Students may select all that apply.

Table 5.4

Number Choosing Career Option: Humanities

Option

1985

1986

1987

1988

1989

1990

Academe Arts Business Communications Desigdarchitecture Govemment/politics Health Helping professions Law Library/museum Medicine Religion Science/technology Skilled tradeslfarming Teaching/administration Other Undecided

119 131 88 133 22

146 115 103 105 21 84 24 31 118 12 44

10

115 117 113 115 24 67 20 26 110 6 35 13 6 6 28 17

36

40

130 117 106 117 24 66 20 40 137 8 32 15 10 10 31 11 46

144 120 125 145 28 96 18 35 153 13 50 15 9 9 40 12 54

176 118 109 124 21 80 16 38 133 9 44 12 16 16 43 19 52

64 7 24 99 10 37 22

2 2 22 11 38

Nore: Students may select all that apply.

18 5 5 48

152

Jerry R. Green

Table 5.5

Number Choosing Career Option: Social Sciences

Option

1985

1986

1987

1988

1989

1990

Academe Arts Business Communications Designlarchitecture Governmentipoh tics Health Helping professions Law Libraty/museum Medicine Religion Scienceitechnology Skilled tradedfarming Teachindadministration Other Undecided

75 32 200 52 6 132 15 28 123 3 25 10 8 3 15 11 22

80 29 199 46 7 118 14 33 104 6 25 2

94 32 215 58 13 I22 31 43 127 4 48 8 7 2 17 17 32

78 46 236 71 15 153 28 41 161 2 31 10 12 2 25 23 37

82 33 177 43 8 131 18 48 149 2 50

115 40 190

7 6 11 15 34

5 8 7 20 23 36

70 19 I34 20 41 152 6 44 4 6 4 22 18 37

Note: Students may select all that apply.

Admission to medical school is still competitive but is not nearly as difficult as it once was. Medical school applicants at Harvard now may feel that they can select a specialty other than life sciences and still gain admission with a very high probability. Of course, as they do so they may become exposed to ideas and fields of study that they find attractive and may abandon their original medical school aspirations.6 Table 5.1 reveals that the mean respondent listed approximately 1.4 career choices in every year. Academics, however, are somewhat more uncertain of their choice, listing on average 1.8 responses, as shown in table 5.6. Breaking this down by field of concentration, we see that the uncertainty is greatest among those in the humanities. In the life sciences, although the average number of careers listed by Academic respondents is above average, this is due primarily to a large number of people who list exactly two responses-academe and medicine. Since for these individuals the actual career to be followed is the same, the issue being only whether one obtains an appointment in a medical school, this probably reflects a lower degree of uncertainty than their average of 1.75 responses per individual would reflect. My conclusions at this quite aggregated level are the following: 1 . The percentage of Harvard graduates contemplating academic careers is 6. It might be possible to test this hypothesis using data from their high school admission records which contain a question on prospective careers and fields of academic concentration, and we hope to do so in future research.

153

Future Graduate Study and Academic Careers

Isble 5.6

Number of Vocations Listed by Academics, 1985-1990 1985

1986

1987

1988

1989

1990

Total One Two Three Total Average

116 109 63 288 1.82

116 126 51 293 1.78

121 68 299 1.82

100 108 62 270 1.86

160 109 59 328 1.69

148 144 73 365 1.79

Life sciences One Two Three Total Average

12 32 6 50 1.88

18 35 5 58 1.78

20 22 8 50 1.76

17 30 1 48 1.67

23 23 6 52 1.67

14 29 4 47 1.79

Two Three Total Average

40 28 7 75 1.56

26 27 3 56 I .59

34 22 12 68 1.68

29 15 9 53 1.62

50 21 7 78 1.45

41 30 4 75 1.51

Humanities One Two Three Total Average

39 34 28 101 1.89

49 35 30 114 1.83

36 35 23 94 1.86

34 42 32 108 1.98

53 42 27 122 1.79

63 44 39 146 1.84

Social sciences One Two Three Total Average

25 15 22 62 1.95

23 29 13 65 1.85

30 31 24 85 1.93

20 21 20 61 2.00

33 22 18 73 1.79

30 41 26 97 1.96

110

Physical sciences

One

Note: There are only three eventual vocations listed for each student.

constant, except perhaps for 1990. If this most recent year is not an outlier, an increase in the preference for academic careers would reflect a very early sign of a break in a trend of constancy. 2. The most significant change in career plans is a decrease in the number of graduates with plans to pursue medicine. 3. In addition, the number of premeds concentrating in life sciences has decreased. More-varied backgrounds are sought by this group. The number of premeds in non-life-science areas has actually increased. It is possible that some individuals who were premed in their intent before graduation have pursued non-life-science concentrations and are not premed by the end of their

154

Jerry R. Green

senior year, when they fill out the senior survey. Thus it is possible, but by no means established, that a significant source of academics in arts and sciences, other than life sciences, consists of people who tried a non-life science concentration and liked it so much that it changed their career plans away from medicine and toward an academic career in the discipline to which they were exposed.

5.4 Academic Career Plans and Undergraduate Performance In this section, we attempt to discover if there is any change over time in the announced decision to follow an academic career as a function of two measures of performance while in college. These measures are grade point average (GPA) and high honors. Harvard uses a somewhat unusual grade point system in which there are one-point gaps between adjacent letter grade categories:A = 1 5 , A - = 1 4 , B + = 12, B = 11,B- = 1O,C+ = 8 , C = 7, and so on. Over the six years of this study, there has been a bit of grade inflation, steadily over time. This somewhat beclouds our results, and we will discuss it further below. The salient result is that there is no discernible trend in the percentages of high-performance or low-performance individuals to desire an academic career, to pursue an arts and science graduate degree to this end immediately upon graduation, or to list academe as the only career possibility (see tables 5.7 and 5.8). The latest year, 1990, has the highest overall percentage choosing academic careers (23 percent), but this seems equally distributed among all GPA groups. Indeed, although among the very highest group (those with GPAs above 14) the percentage selecting academic careers is 44 percent, this was exceeded in 1987, an otherwise average year for academic careers. Actually it is among those with GPAs between 11 and 13, basically the “straight B” students, that the percentage of Academics has risen the most and is significantly higher than in any other year-22 percent in 1990 as opposed to 18 percent for 1985-89 combined. Interpretation of these numbers is difficult in the presence of grade inflation, Whether or not students think highly enough of their own academic potential to desire an academic career depends to some extent on their interpretation of their own grades. If students do not know that there is ongoing grade inflation, then as more students get higher grades there is, in effect, more of this positive feedback. Under this hypothesis, conditional on GPA, the progression to an academic career should be constant over time. However, to the extent that students’ self-assessments or tastes for academic work are independent of grades received, grade inflation reduces the quality of the pool in any GPA category and should reduce the progression to graduate school in that category. Thus, our observation of roughly constant progression rates could be consistent with an exogenous increase over time in the likelihood that any

155

Future Graduate Study and Academic Careers

Table 5.7

’Qpes of “Academic Career” Choices by Undergraduate Performance, 1985-1990 Academic

GPA

Trueac

N

Onlyac

N

%

51 54 74 77 32 201 43

11% 17 20 31 40 14 51

10 14 21 28 14 87 21

36

9 17 19 35 43 20 45

11 14 24 33 18 100 19

3

8

5

6

6 12 19 7 23

17 17 9 57

II

32 21 38 57 27 175 37

7 7 10 20 30 12 35

20 11 20 26 16 93 19

4 9

10

%

N

%

Total

1%

46I 324 367 247 80 1,479 84

1985

1 1 .O and below 11.01-12 12.01-13 13.01-14 14.01and up Total High honors

2% 4 6 11 18 6 25

6 7 16 29 9 67 16

2 4 12 11 5

19

I986

11 .O and below 11 .OIL12 12.01-13 13.01-14 14.01and up Total High honors

51

74 92 40 293 37

11

2 2

4 6 10 4 13

407 301 390 265 94 1,457 83

I987 11 .O and below

11.01-12 12.01-13 13.01-14 14.01and up Total High honors

I988 1 1 .O and below

11.01-12 12.01-13 13.01-14 14.01and up Total High honors

52 43 73 89 42 299 55

14 20 32 46 20 51

35 42 72 84 37 270 47

13 17 29 36 17 45

18 30 41 55 31 175 38

19 30 11 36

17 33 19 86 25

43 56 79 99 51 328 62

10 17 19 31 45 20 54

11 11 29 44 32 127 36

3 3 7 14 28 8 31

6 6 19 28 18 77 20

40 75 93

11 21 23

4 17 30

1

5 8

2 9 20

8

10

7

4 4 5

9 18 6 18

473 310 373 280 91 1,488 107

2 3 4 11 18 5 24

442 336 423 287 104 1,592

1

414 339 421 320 114 1,608 115

I05

1989

1 1 .O and below 11.01-12 12.01-13 13.01-14 14.01and up Total High honors

1990 11 .O and below

11.01-12 12.01-13 (continued)

2 5

9 16 5 17 1 3 5

352 352 398

156

Jerry R. Green

Table 5.7

(continued) Academic

GPA 13.01-14 14.01 and up Total High honors

N

%

100 57 365 53

31 44 23 50

Trueac

N 41 29 121 32

Onlyac %

13 22 8 30

N 29 17 77 17

%

Total

9 13 5 16

324 129 1,555 107

Note: See section 5.4 for a description of Harvard’s grading system and section 5.2 for a description of Academic, Trueac, and Onlyac.

Table 5.8

Qpes of “Academic Career” Choices by Undergraduate Performance: Pooled Data Academic

Trueac

Onlyac

N

%

N

%

N

%

Total

1I .O and below 11.01-12 12.01-13 13.01-14 14.01 and up Total High honors

139 148 22 1 258 114 793 135

10% 16 20 33 43 18 49

53 49 83 118 59 362 77

4% 5 7 15 22 8 28

34 24 53 72 34 217 46

3% 3 5 9 13 5 17

1,341 935 1,130 792 265 4,424 274

1988-90 11.O and below 11.01-12 12.01-13 13.01-14 14.01 and up Total High honors

118 173 244 283 145 963 162

10 17 20 30 42 20 50

33 58 100 140 92 423 106

3 6 8 15 27 9 32

15

1

25

2 5 10 16 5 19

1,208 1,027 1,242 93 1 347 4,755 327

GPA

56 90 54 240 62

Note: See section 5.4 for a description of Harvard’s grading system and section 5.2 for a description of Academic, Trueac, and Onlyac.

given individuals would select an academic career, holding constant their assessment of their own potential.’ The increasing interest in academic careers among the B students does not 7. To assess this hypothesis and look for a silver lining in terms of extra supply of academics in what would otherwise be a flat picture (with the possible exception of 1990), we looked at GPA rank rather than GPA itself. This procedure is designed to capture the idea that individuals know that there is grade inflation and take account of it in drawing inferences about their own scholarly potential and hence in the interest they take in academic careers. This analysis also shows that there is very little difference between GPA and GPA rank. The relative rank effect, if it is present, is too weak to be captured in our data. Whatever little time trend is revealed by this method is surely most prevalent in the lower portion of the grade distribution, roughly the B or B - range, where academic careers are relatively rare in any event.

157

Future Graduate Study and Academic Careers

persist to their choice to go to graduate school immediately. In 1990 only 7 percent of students with GPAs between 11 and 13 are Trueacs, which is exactly the same as the percentage in 1985-89. Quite possibly this means that these individuals will be planning to work after graduation, to take additional coursework, and to bolster the quality of their application to a graduate school at a later date. High honors is another measure of performance. I had thought that this would be more indicative of an interest in research, since to graduate with high honors one must write a thesis. The experience of such a major piece of research while an undergraduate may serve to heighten interest in an academic career. However, although the prevalence of Academics among recipients of high honors is much higher than among any other group, even than those with GPAs above 14 (who are about equal in number to the high-honors recipients), there is no discernible trend. Indeed there are some marked curiosities in the time pattern. For example, in 1988, which was in many respects the low point of interest in academic careers (e.g., lowest percentage among those with high GPAs), there was the highest percentage of Onlyac responses. Over half of the high-honors recipients who are Academics are also Onlyacs, whereas in an average year this ratio would be closer to one in three. And for 1990, where there is a heightened interest in academic careers in general and among the B students in particular, there is actually somewhat of a decrease in interest among those earning high honors and a very sharp decrease in the Onlyac category.

5.5 Academic Career Plans and Undergraduate Performance by Field The results of the previous section might make us a bit pessimistic about finding trends toward academic careers, especially among the higher performance categories of undergraduates. We have seen that 1987, 1988, and 1990 are all somewhat interesting and atypical years, but no consistent pattern is evident. The goal of this section is to examine these trends more carefully, breaking things down by field. We will see here that there are some differences between years that are detectable at the field level and that these differences manifest themselves in different ways from one field to another. Because of the danger of creating cell sizes that are too small, I pool 198587 and 1988-90 to create two subpopulations of approximately equal size. Because the stock market crash of October 1987 would have affected the responses of the classes of 1988 and after, this breakpoint might allow the testing of additional hypotheses in this regard. However, one must be cautious to not overinterpret the data, and 1 will try hard in this direction although I may not entirely succeed. The pattern of progression to academic careers is quite different in the four fields of study. Life sciences, as mentioned in section 5.3, have experienced a major decrease in total enrollments-588 degrees conferred in 1985-87 as

158

Jerry R. Green

compared to 471 in 1988-90. Most of this decrease is attributable to a drop in the number of premeds and a shift in the background of premeds to non-lifescience fields. Despite this, the number of Academics coming from the life sciences is roughly constant, and the percentage of graduates who are Academics is necessarily up, as can be seen in table 5.9. More surprising, and more significant for arts and science graduate schools and departments, the number of Trueacs is up, from 44 to 59, and the percentage of Trueacs has almost doubled, from 7 to 13 percent. The situation for Onlyacs is similar22 from 1985-87 and 28 from 1988-90. My interpretation of this is as follows: Life science majors are really two groups, premeds and nonpremeds. Interest in medicine may be down, but molecular biology, genetic engineering, and related areas in chemistry and biochemistry are hot topics. These are fields in which jobs are plentiful and relatively well paid. There is a sense of intellectual excitement. But these are jobs for Ph.D.’s with good postdoctoral training. The better undergraduates with these interests are still there. And there may be some additional scientifically talented and well-motivated people who would have been premeds in other times, some who might have chosen computer science or other areas in applied mathematics, and possibly some who might have been in nonscientific concentrations altogether. Although some of the cell sizes are small, there is fairly sound statistical evidence that the constancy of the number of academics and the doubling of the number of Trueacs and Onlyacs is coming at least as Table 5.9

Pooled Data on “Academic Career” Choices by GPA: Life Sciences Academic

Trueac

Onlyac

N

%

N

%

1985-87 11.O and below 11.01-12 12.01-13 13.01-14 14.01 and up Total High honors

33 21 44 42 18 158 18

20% 19 29 36 41 27 46

14 4 11 12 3 44 4

9% 4 1 10 1 7 10

1988-90 1 1.O and below 11.01-12 12.01-13 13.01-14 14.01 and up Total High honors

16 27 49 39 16 147 24

16 21 38 37 42 31 53

9 12 18 13 7 59 10

9 12 14 12 18 13 22

GPA

N

%

Total

9 2 7 3

163 111 I54 I16

22 2

6 2 5 3 2 4 5

588 39

6 8 1 4 3 28 5

6 8 5 4 8 6 11

99 101 128 105 38 47 1 45

1

44

Note: See section 5.4 for a description of Harvard’s grading system and section 5.2 for a description of Academic, Trueac, and Onlyac.

159

Future Graduate Study and Academic Careers

much from the elite part of the distribution as it is from the B students. (However, one must also be aware that the average GPA in life sciences is higher than in the college as a whole.) Physical sciences present a picture similar to that in the life sciences, though trends are somewhat muted. The results are shown in table 5.10. There has been no mass exodus paralleling that of the premed decrease in the life sciences. However, the number of computer sciences concentrators has dropped slightly, and this accounts for most of the 8 percent decrease in the number of physical science concentrators. The number of those with intentions to follow academic careers is about the same. Moreover, as in the case of the life sciences, both the Trueac and the Onlyac categories are up slightly in numbers and therefore up more than slightly as a percentage of concentrators. All three measures of the propensity to follow academic careers show approximately the same distribution across performance categories in the two time periods. The only item of statistical significance is the rise of Onlyacs in the high-GPA categories (13 and up). However, the numbers are relatively small-42 in 1985-87 versus 57 in 1988-90-and I said I would try not to overinterpret them, so I will go no farther here. Humanities concentrators make up 39.8 percent of respondents in 198890, as opposed to 37.2 percent in 1985-87. There is little time pattern in the percentages of Academics in various performance categories. However, the Table 5.10

Pooled Data on “Academic Career” Choices by GPA: Physical Sciences Academic

Trueac

Onlyac

GPA

N

%

N

%

N

%

Total

1985-87 11.O and below 1 I .01-12 12.01-13 13.01-14 14.01 and up Total High honors

32 28 46 54 39 199 51

15% 26 36 47 62 32 66

15 11 29 39 28 122 38

7% 10 23 34 20 49

12 4 20 26 16 78 24

6% 4 16 23 25 13 31

209 108 127 114 63 62 1 77

19 27 47 72 41 206 48

15 26 35 53

9 10 29 49 32 129 40

10 22 36 45 23 50

4 6 19 34 23 86 27

3 6 14 25 32 15 34

127 105 134 136 71 573 80

1988-90 11.O and below

11.01-12 12.01-13 13.01-14 14.01 and up Total High honors

58

36

60

44

I

Note: See section 5.4 for a description of Harvard’s grading system and section 5.2 for a description of Academic, Trueac, and Onlyac.

160

Jerry R. Green

number going to graduate school immediately rather than later is up almost 50 percent from 105 to 162, as shown in Table 5 . 1 1 . This might be due to a decrease in the opportunities to work after graduation. Although the senior survey has a number of questions about the effect of loan balance and other financial measures on choice of work or graduate school (the intent of which is to inquire about the need to earn and save money or reduce indebtedness before going farther in one’s education), it does not address the question of the availability of work or of the quality of available work. Whatever the reason for the increased propensity to go to graduate school, the increase seems equally distributed across all GPA categories, except perhaps the lowest, which is in any case only a small contributor to the short-run progression to graduate study. Social science concentrators have remained at almost exactly the same proportion of seniors in the two subperiods. See table 5.12. It is in the social sciences that the grade inflation is most severe, so some of the results within GPA categories may be a by-product of this phenomenon and do not reflect the progression probabilities for academic careers or graduate school as discussed in section 5.2. The percentage of social science concentrators who are Academics is essentially the same. However, unlike the case of the humanities concentrators, who now seem more likely to go to graduate school directly, social science concentrators have had no significant change in this regard. Of the four fields of specialization, social scientists are the least likely to go to Table 5.11

Pooled Data on “Academic Career” Choices by GPA: Humanities Academic

GPA 1985-87 11 .O and below 11.01-12 12.01-13 13.01-14 14.01 and up

Total High honors 1988-90 11 .O and below 11.01-12 12.01-13 13.01-14 14.01 and up Total High honors

N

%

32 60 85 97 35 309 45

8% 17 18 31

42 68 96 112 58 376

60

41 19 48

10 17 18 29 45 20 49

Trueac

N 11 16 28 32

18 105 24

8 23 39 55 37 162 38

Onlyac %

N

3% 4 6 10 21 6 26

4 17 24 12 65 16

2 4 8 14 4 17

2 6 7 14 29 9 31

I

0

6 24 38 19 88 18

1 4 10 15

8

%

Total

I%

412 363 469 318 86 1,648 93

5 15

440 402 537 388 128 1,895 122

Nore: See section 5.4 for a description of Harvard’s grading system and section 5.2 for a descnp-

tion of Academic, Trueac, and Onlyac.

161

Future Graduate Study and Academic Careers

Table 5.12

Pooled Data on “Academic Career” Choices by GPA: Social Sciences Academic

Trueac

Onlyac

N

%

N

%

N

%

Total

1985-87 1 1.O and below 11.01-12 12.01-13 13.01-14 14.01 and up Total High honors

42 38 45 65 22 212 21

8% 11 12 27 31 13 34

13 17 14 25 10 79 11

2% 5 4 11 14 5 18

9 9 9 9 5 41 4

2% 3 2

555 347 275 237 71 1,585 61

1988-90 11.O and below 11.01-12 12.01-13 13.01-14 14.01 and up Total High honors

41 50 50 60 30 23 1 30

8 12 12 20 28 13 38

7 13 14 23 16 73 18

1 3 3 8 15 4 23

4 5 6 14 9 38 12

GPA

4 7 3 7 1 1

1 5 8 2 15

537 413 432 300 109 1,791 78

Note: See section 5.4 for a description of Harvard’s grading system and section 5.2 for a description of Academic, Trueac, and Onlyac.

graduate school directly-less than half as likely as in any other field, even adjusted for GPA. I find it quite surprising that the humanities and social sciences are so different in these regards. Both humanities and social sciences are common prelaw concentrations. It would seem therefore that the increase in law as a prospective career should have affected humanities and social sciences similarly, but it did not. This is a trend that should be followed closely in the future.

5.6 Paths to Academic Careers The prevalence of delaying graduate school plans and the multiplicity of careers contemplated by graduates make an analysis of the paths to academic careers interesting. Figure 5.1 shows alternative paths which could have led to the answer Academic. The choice of what to do immediately upon graduation is shown across the top row. Work is by far the modal option. Only 10 percent of all graduates are going to an arts and sciences graduate school within 12 months. The majority of Academics arrive by different routes. Since the passage to an academic career by any route is uncertain, it is important to know what routes are being chosen.8 8. In future work, I hope to examine the attrition rates along various paths. Even for those who go to graduate school directly, the attrition rate i s very high. There are many possible slips along

162

Jerry R. Green

I

COLLEGE

L 10.0

GRADUATE SCHOOL

62.41

18.1 I

L

GRADUATE SCHOOL

ACADEMIC CAREER path probability

6.3

4.4

1.4

0.5

3.0

Fig. 5.1 Paths to academic careers, 1985-1990 Nofe: Percentages shown on each branch of the tree are the conditional probabilities of choosing as indicated. The products of these probabilities down each path of the tree, leading to an academic career. are shown at the bottom.

I assume that in order to be an academic one must have gone to graduate or professional school at some point. Of those not going to graduate school immediately, 16 percent have plans to attend an arts and sciences graduate school eventually. The conditional probabilities of this attendance are shown ernanating from the second level of the tree, “Graduate School” (eventually). It is also possible to become an academic following a professional degree, usually each of the other paths as well. Policies designed to increase the number of qualified faculty members in the future should address each of these leakages. The initial intention to pursue an academic career as expressed by college seniors is not definitive.

163

Future Graduate Study and Academic Careers

law, medicine, or public health. Finally, of those who attain a graduate degree in the arts and sciences or a professional degree, some will have plans to become academics and some will not. The conditional probability of listing an academic career among these possibilities is shown on the next to last line of percentages in the figure. As these conditional probabilities show, those with plans to attend graduate school immediately are the most likely to go on to an academic career. The differences across groups are, however, smaller than I would have initially thought. Over 40 percent of those deferring graduate school in favor of work are still interested in academic careers. The senior survey does not ask about the highest degree one anticipates receiving. Therefore, it could be that many of those who are working first and returning to graduate school are not in Ph.D. programs. Thus, of those iq Ph.D. programs, the commitment for academic careers could be as high among those working initially as it is among those going to graduate school directly. Individuals who work and then return to graduate school are a significant possible source of future academics. Their numerical significance results from the large number of those who plan to work after graduation. Since only 17 percent of the graduating class states that their decision to work is affected by financial considerations such as a loan balance, I assume that the majority of those who plan graduate school after working are doing so because the work will be beneficial to their careers. It may be a good credential. Or it may allow them to explore the career option they have in mind without having to spend the time in graduate school first. For many, time spent working may simply be time away from school in which they can decide on a career, or on a field of graduate study, as may best suit their interests and personal circumstances. Those whose plans are listed as “other” are a very interesting group. Some have won fellowships that allow, or require, travel or public service. These tend to be very highly selective awards, and the students in this category have much better than average performance. Their average GPA is 12.16, and 13.0 percent of them have degrees with high honors, in contrast to samplewide averages of 11.86 and 6.6 percent. Not surprisingly, many of them want to return to graduate school and have the intent to follow academic careers. Indeed they are more than twice as likely to go to graduate school than the sample as a whole. Finally, there is the relatively small group who plan to go to an arts and sciences graduate school after a professional school. These can be people in certain joint degree programs (e.g., M.D./Ph.D.). Most of them have academic plans, though it is difficult to say for certain that they are a potential source of arts and sciences faculty. By multiplying the conditional probabilities down each path in the tree, we can see the sources of academics as a percentage of all graduates. This is shown in the last line of figure 5.1. Figures 5.2 and 5.3 show the same conditional probabilities displayed as

164

Jerry R. Green

I

ISCHOOL

60.t

1 16.0

r i

I

GRADUATE SCHOOL

ACADEMIC CAREER I

I

path probability

6.0

4.2

1.4

0.2

3.5

Fig. 5.2 Paths to academic careers, 1985-1987 Nore: Percentages shown on each branch of the tree are the conditional probabilities of choosing as indicated. The products of these probabilities down each path of the tree, leading to an academic career, are shown at the bottom.

trees for the two subperiods. The differences are not significant. Indeed the initial choice upon graduation is almost exactly the same. A slight increase in the probability of returning to graduate school after working is offset by a decrease in the probability of becoming an academic after this sequence. As a percentage of all graduates, the probability of this route rises from 4.2 percent to 4.5 percent, which is insignificant by any measure. In order to see if the behavior or anticipated behavior of different groups might be changing, we break the population down by performance. Again in order not to create too many small cells, we take a simple division into GPAs

165

Future Graduate Study and Academic Careers

-J

GRADUATE SCHOOL

I

1

ACADEMIC CAREER

I

J

path probability 6.5

4.5

1.4

0.9

2.6

Fig. 5.3 Paths to academic careers, 1988-1990 of the tree are the conditional probabilities of choosing as indicated. The products of these probabilities down each path of the tree, leading to an academic career, are shown at the bottom.

Note; Percentages shown on each branch

above 13 (high GPA) and those below 13 (low GPA).9 The results are shown in tables 5.13 and 5.14. It is not a surprise that those with high GPAs are more likely to go to graduate school immediately and more likely to plan to return to graduate school following work. It is a bit more of a surprise, but not markedly so, that they are more likely to plan an academic career after completing 9. Recall that in the Harvard system, 13 does not correspond to any letter grade. It is in the gap between A - and B , Roughly, a student with a GPA of 13 has as many A’s as B’s and nothing worse than B. About a quarter of all students are above this cutoff.

+

Table 5.13

Paths to Academic Choice: GPA 13.00 and above Graduate School

1988-90; Total = 1,312 Plans Plans for grad. school Conditional probability of Academic Academic Onlyac

Other

Professional School

Professional Academic*

N

%

N

%

N

%

N

%

43% 25 57

129 49 30

12% 38 61

302 4 4

27%

83

28%

72

470 116 66

1 100

82

13 57

12

2

0 50

260 260 186

20% 100 72

66

22%

122

14 66

N 1985-87: Total = 1,102 Plans Plans for grad. school Conditional probability of Academic Academic Onlyac

Work

20 1 201 144

%

18% 100

545 151 79

12

6 18 42% 28 52 6 15

3 175 57 26

5

3 10 13% 33 46 2 19

332 37 21

6

25% 11 57 2 29

Dejinitions of parameters: Plans = Number and percentage making indicated choice Plans for grad. school = Number and percentage planning graduate school, given an initial choice of work, other, or professional school Conditional probability of academic = Number and percentage indicating “academic” as a percentage of those planning graduate school in the given path Academic % = Number of Academics + Total Onlyac: N = Number of Onlyacs % = Number of Onlyacs + Number of Academics *Graduates planning to follow an academic career in a professional school (e.g., teach in a medical school).

Table 5.14

Paths to Academic Choice: GPA under 13.00 Graduate School

1985-87: Total = 3,361 Plans Plans for grad. school Conditional probability of Academic Academic Onlyac

1988-90: Total = 3,443 Plans Plans for grad. school Conditional probability of Academic Academic Onlyac

Work

Other

Professional School

Professional Academic*

N

%

N

%

N

%

N

%

N

%

243 243 125

7% 100 51

2,152 315 122

64%

432 89 34

13% 21 38

534 6

16% 1 83

73

14%

15 39

78

4 62

19

4 16

8

1 24

3

65 18 33

445 98 40

13 22 41

526 35 21

58

12

55

2,252 411 137

60

4 58

14

4 10

3

1 8

7

1 33

220 220 122

71

6 100

5

6

0 0 15 1

Definitions of parameters: Plans = Number and percentage making indicated choice Plans for grad. school = Number and percentage planning graduate school, given an initial choice of work, other, or professional school Conditional probability of academic = Number and percentage indicating “academic” as a percentage of those planning graduate school in the given path Academic % = Number of Academics + Total Onlyac: N = Number of Onlyacs % = Number of Onlyacs + Number of Academics *Graduates planning to follow an academic career in a professional school (e.g., teach in a medical school)

168

Jerry R. Green

graduate school in an arts and sciences field. This may, however, be a statistical artifact of the fact that we cannot distinguish between plans to get master’s degrees in arts and sciences fields as opposed to doctorates. Finally, of those going down each branch of the tree, we show the number and percentage of Onlyacs-that is, those who list no occupation other than “academe.” The percentage of Onlyacs among the high-GPA group is marginally greater than among the low-GPA group. This probably reflects a level of self-assuredness that a strong academic record engenders. The significance is, however, slight. Comparing these transition probabilities across time periods reveals little change at all. Therefore, we examine the data by field, paralleling in tables 5.15-5.18 the analysis in section 5.5. Life science concentrators have, as might be expected, increased their probability of going to graduate school immediately. There is a slight decrease in the probability of choosing an academic career given this initial decision. This is probably due to the excellent research opportunities in biotechnology firms for Ph.D.-level scientists in the life sciences. Otherwise, the analysis of postgraduation plans over time is unremarkable in the life sciences. lo For the other three groups of concentrations, no significant pattern is detected over time in any of the conditional probabilities. Thus, I conclude three things: 1. In certain very specific specialities, such as molecular biology, there may be very marked changes in the probability of progression to graduate school and an academic career and in the quality of students making this choice. These changes are masked when one looks at broader categories of undergraduates, such as all life science concentrators. 2. There is some evidence that quality in the social sciences and humanities may be decreasing, while quality in the sciences may be increasing. 3. It is possible that 1990 is the start of a different trend, perhaps due to the current recession. The data do not allow us to say this definitely, but the possibility should be followed closely this year and thereafter.

5.7 Comparison of the Harvard and COFHE Data In this section, we compare the data from the Harvard Senior Survey of 1985, analyzed above, to the data from the COFHE survey of 1984. We also compare the subset of the COFHE survey that came from Harvard students with two other subsets of COFHE respondents: other universities and coeducational colleges. We restricted attention to institutions where the response 10. Fifteen students in the later time period report a plan to go to graduate school following professional school, as compared to zero in the earlier period. However, there is a corresponding decrease in the number planning an academic career after professional school alone. This is more likely to reflect a change in the degree qualifications for certain research and clinical positions in medical schools than to denote any real change in academic career plans.

Table 5.15

Paths to Academic Choice, by Field: Life Sciences Graduate School

1985-87: Total = 588 Plans Plans for grad. school Conditional probability of Academic Academic Onlyac 1988-90: Total = 471 Plans Plans for grad. school Conditional probability of Academic Academic Onlyac

N

%

52 52 35

100

16 58 58 45

23

Work

Other

Professional School

Professional Academic*

N

%

N

%

N

%

N

%

35% 15 61

42 8 5

7% 19 63

291 0 0

49% 0

70

24%

67

203 31 19

6 46

3

3 16

3

34 22 53

42

21

78

158 34 18

10 51

0

4 0

9%

12 100

0

I 60

0

0

I

8 23 78

216 15 12

46 7 80

2

1 29

3

3 25

39 9

0

Definitions of parameters: Plans = Number and percentage making indicated choice Plans for grad. school = Number and percentage planning graduate school, given an initial choice of work, other, or professional school Conditional probability of academic = Number and percentage indicating “academic” as a percentage of those planning graduate school in the given path Academic % = Number of Academics + Total Onlyac: N = Number of Onlyacs % = Number of Onlyacs + Number of Academics *Graduates planning to follow an academic career in a professional school (e.g., teach in a medical school).

Table 5.16

Paths to Academic Choice, by Field: Physical Sciences Graduate School

1985-87: Total = 621 Plans Plans for grad. school Conditional probability of Academic Academic Onlyac 1988-90: Total = 573 Plans Plans for grad. school Conditional probability of Academic Academic Onlyac

Work

Other

Professional School

Professional Academic*

N

%

N

%

N

%

N

%

N

%

155 155 108

25% 100 70

355 117 43

57% 33 37

44 12 3

7% 27 25

67 1 1

11% 1 100

25

38%

71

17 66

5

7 12

I

0 33

1

0 100

149 149 111

26 100 74

284 97 39

50 34 40

49 23 12

9 41 52

91 12 7

16 13 58

66

84

17

19 69

6

1 1.5

0

2 0

3

1 43

Definitions of parameters: Plans = Number and percentage making indicated choice Plans for grad. school = Number and percentage planning graduate school, given an initial choice of work, other, or professional school Conditional probability of academic = Number and percentage indicating “academic” as a percentage of those planning graduate school in the given path Academic % = Number of Academics f Total Onlyac: N = Number of Onlyacs % = Number of Onlyacs + Number of Academics *Graduates planning to follow an academic career in a professional school (e.g., teach in a medical school).

Table 5.17

Paths to Academic Choice, by Field: Humanities Graduate School

1985-87: Total = 1,648 Plans Plans for grad. school Conditional probability of Academic Academic Onlyac 1988-90: Total = 1,895 Plans Plans for grad. school Conditional probability of Academic Academic Onlyac ~

N

%

151 151

80

9% 100 53

50

63

9

171 171 103

9

1,141

100

275

60 5 62

N

970 183 74

Professional School

Professional Academic*

%

N

%

N

%

N

%

59% 19 40

284 88 39

17% 31 44

243 7 6

15% 3 86

30

13%

39

16

4 12

3

2 8

3

50

100

60 24 36

303 81 28

16 27 35

280 34 17

15 12 50

16

5 16

2

7

6

35

5

64

Other

Work

0

1

1 ~

Definitions of parameters: Plans = Number and percentage making indicated choice Plans for grad. school = Number and percentage planning graduate school, given an initial choice of work, other, or professional school Conditional probability of academic = Number and percentage indicating “academic” as a percentage of those planning graduate school in the given path Academic % = Number of Academics + Total Onlyac: N = Number of Onlyacs % = Number of Onlyacs iNumber of Academics *Graduates planning to follow an academic career in a professional school (e.g., teach in a medical school).

Table 5.18

Paths to Academic Choice, by Field: Social Sciences Graduate School

Work

Other

Professional School

Professional Academic*

N

%

N

%

N

%

N

%

N

%

82 82 45

5% 100 55

1,081 95 51

68% 9 54

188 30 17

12% 16 57

234 2 2

15%

31

13%

100

23

3 51

13

3 25

4

1 24

I

0 50

1,198 152 57

67 13 38

226 41 18

13 18 44

265 I1 6

15 4 55

30

12

49

6 100 48

29

3 59

4

7

4

1 22

1

0 17

1985-87: Total = 1,585

Plans Plans for grad. school Conditional probability of Academic Academic Onlyac 1988-90: Total = 1,791 Plans Plans for grad. school Conditional probability of Academic Academic Onlyac

102

102

3

I

Definitions of parameters: Plans = Number and percentage making indicated choice Plans for grad. school = Number and percentage planning graduate school, given an initial choice of work, other, or professional school Conditional probability of academic = Number and percentage indicating “academic” as a percentage of those planning graduate school in the given path Academic % = Number of Academics + Total Onlyac: N = Number of Onlyacs % = Number of Onlyacs + Number of Academics *Graduates planning to follow an academic career in a professional school (e.g., teach in a medical school).

173

Future Graduate Study and Academic Careers

Table 5.19

UndergraduateField of Study: COFHE Data 1984, Harvard Senior Survey 1985 (percentages)

Life sciences Physical sciences Humanities Social sciences Other Total number included

Table 5.20

Life sciences Physical sciences Humanities Social sciences Other

COFHE Harvard

COFHE Colleges

COFHE Universities

Senior survey

18.1% 11.5 19.6 35.0 27.1 260

9.6% 15.5 29.1 27.4 20.7 1,665

8.8% 17.0 28.8 28.6 16.8 2,119

13.2% 16.4 36.2 34.1 0.1 1,479

Comparison of Questions Related to Progression to an Arts and Sciences Graduate School (percentage) COFHE Harvard

COFHE Colleges

COFHE Universities

Senior Survey

27.0% 33.3 9.8 17.6 11.1

13.1% 27.1 8.7 9.0 8.4

15.0% 24.4 11.7 7.3 7.6

12.6% 24.0 9.1 7.1

-

rate was over 50 percent, as it was for Harvard. We also eliminated MIT from the sample, due to the special nature of the undergraduate program there. Table 5.19 shows the comparison of the fields of study of the students in the three subsamples of COFHE respondents and in the Harvard Senior Survey. We can see that a large number of Harvard students are classified as in “other” fields by COFHE (27.7 percent), whereas our fields have been constructed so that there are essentially no Harvard concentrations outside the life science/physical science/humanities/social science groupings. This should be kept in mind in interpreting the progression plans reported by COFHE respondents in each field. Across the three COFHE subsamples, we can see that Harvard is not markedly atypical. The principal differences are that Harvard has a somewhat larger percentage of life science and social science concentrators. This is probably due to a higher percentage of prelaw and premed students than at other COFHE institutions. The COFHE survey has a general question about intentions to attend graduate school in the arts and sciences. The most similar question in the senior survey asks whether one expects to go to an arts and sciences graduate school immediately. The comparison is shown in table 5.20. Table 5.20 reveals that the Harvard students are more likely to go to graduate school in each field, though the difference is slight in the humanities.

174

Jerry R. Green

Comparing Harvard students who replied to the COFHE survey and the complete sample in the senior survey is interesting. We expect a lower rate of positive answers to the question about progression to graduate school immediately in the senior survey, and this is the case except for humanities concentrators, where the percentages are virtually the same. This similarity could be due to the fact that there is no sample selection bias for humanities students (i.e., humanities students who returned the COFHE survey are not any different than the population as a whole), whereas in the other fields those who returned the COFHE survey may have been the more “academic” types. It is not due to a difference in the questions about plans to go to graduate school immediately rather than on a delayed basis, as can be seen by comparing table 5.17 (humanities) with tables 5.15, 5.16, and 5.18. Future academics in the humanities are at least as likely to delay graduate school for work or another activity as are the rest of Harvard undergraduates. For those students who respond positively to the question about attending an arts and sciences graduate program, the COFHE survey asks about the highest degree expected. This is a good proxy for the intention to continue to an academic career. The responses are shown in table 5.21. We can see that given attendance at graduate school, Harvard students are more likely to expect to attain a Ph.D. However, except in the physical sciences this is not a very pronounced difference. The disparity in the physical sciences may be due to the fact that Harvard has very few engineering concentrators (which is why we deleted MIT from the COFHE sample of universities). Engineers who can get very good jobs at the top of their professions without a Ph.D. are probably less likely than other physical scientists to pursue that course. Therefore, the difference between Harvard students and the other COFHE respondents, although positive, is probably slight and not related to field of study. Based on this comparison of data sources I conclude that the Harvard Senior Survey is probably a good indicator of trends for a wider group of institutions and that the sample selectivity in the COFHE data, due to the nonmandatory responses at most institutions, is not likely to be a very important problem for the study of the progression to graduate school and academic careers. Table 5.21

Final Degree Expected by COFHE respondents: Ph.D. (percentage)

Life sciences Physical sciences Humanities Social sciences Other

COFHE Harvard

COFHE Colleges

COFHE Universities

29.8% 43.3 29.4 20.9 22.2

28.1% 34.5 21.7 20.2 18.8

26.2% 31.4 23.5 13.9 17.7

175

Future Graduate Study and Academic Careers

Appendix

176

Jerry R. Green

177

Future Graduate Study and Academic Careers

178

Jerry R. Green

179

Future Graduate Study and Academic Careers

References Bowen, H. R., and J. H. Schuster. 1986. American professors: A national resource imperiled. New York: Oxford University Press. Bowen, W. G., and J. A. Sosa. 1989. Prospects for faculty in the arts and sciences. Princeton, N.J.: Princeton University Press. Fernandez, L. 1978. U S .faculty a f e r the boom: Demographic projections to 2000. Carnegie Council on Policy Studies in Higher Education, Technical Report no. 4. Berkeley, Calif. Radner, R., and L. Miller. 1975. Supply and demand in US.higher education. New York: McGraw-Hill.

Comment

Charlotte v. Kuh

Jerry Green has dug up what can be described as a treasure trove of data about the intentions of Harvard undergraduates to attend graduate school. Harvard undergraduates may be a special group, but he has a virtual census of them and information concerning their postgraduation plans. As we look at these data, we need to remember that they are data about intentions gathered at one point in time. To ascribe broader significance to this data set, we need to look at other data to ask whether Harvard intentions mirror undergraduate intentions more broadly, how those intentions translate into outcomes, and what sorts of models could give rise to the data that have been presented. It is these three questions that I would like to address in my discussion. Before doing that, however, I would like to mention briefly why these data are so interesting. The paper mentions the mounting concern and debate over the adequacy of Ph.D. supply to meet the increased demand for new faculty that is expected later in the 1990s. A broader source of concern is that production of highly trained American citizens is not keeping up with the growth of our population, let alone the growth of those parts of the economy that are relatively intensive in highly trained people. Between 1978 and 1988, the U.S. population of age 25 to 34 grew by 26 percent (NCES 1990). In the same period, for U.S. citizens: Master’s degrees declined by 4 percent Doctoral degrees grew by 8 percent First professional degrees grew by 6 percent On the other hand, the rate of participation in graduate education of highability students has not changed over the past decade. Specifically, of those Charlotte V. Kuh is executive director of the Graduate Record Examinations Program at Educational Testing Service.

180

Jerry R. Green

who scored in the top quartile in an ability test given as a part of two national surveys of high school seniors (High School and Beyond survey of 1980 and National Longitudinal Survey of High School Seniors in the Class of 1972), 12 percent of both groups were enrolled in graduate study seven years after high school graduation (Hilton and Pollack 1991). Although graduate school attendance rates have been relatively constant, choice of undergraduate major has undergone large shifts in the past decade. Nationally, bachelor’s degrees declined by 5 percent in the humanities and social sciences, by 29 percent in the biological sciences, by 23 percent in the physical sciences, and by 33 percent in education. The growth fields were computer science and engineering (96 percent) and business (52 percent). These changes in baccalaureate degrees are also reflected in higher degrees. (NCES, 1990.) Undergraduates do seem to gravitate, in impressive numbers, to those fields that reputedly will bring higher economic rewards. Although the only data set on senior intentions to undertake graduate study is the COFHE data set that is discussed in the paper, we can look at other data to see how similar the field choice of Harvard graduates who obtain Ph.D.’s is to field choice nationally. These data would relate to the question, How indicative of national trends are the field choices of Harvard undergraduates? If they track fairly closely, then the Harvard data can be generalized and be used as a leading indicator of changes in Ph.D. supply to particular fields. These distributions are shown in table 5C. 1. With the exception of the biological sciences, to which there was a much greater shift by Harvard baccalaureates, there are similarities in both the direction and the size of the change in percentage distribution by field of the 196082 cohorts and the most recent (1986-88) cohorts. For Harvard, the source of the boom in biological science Ph.D.3 comes primarily at the expense of Ph.D.’s in nonscience and engineering fields, principally the humanities. Table 5C.2 recasts the Green data to examine the question of what the recent direction of field switching has been. We note that the fields in Green’s paper are not as disaggregated as the National Science Foundation data for Ph.D.5 shown in table 5C.1. I looked at two groups that overlap in some respects but differ in “quality”: those who are directly graduate school bound (Trueac) and those who are academically gifted (high honors) as shown by both grades and an undergraduate research project. Approximately 25 percent of Trueacs receive high honors, while approximately 33 percent of highhonors graduates classify themselves as Trueac. What may be trends in field choice are evident. We see continued movement into the biological sciences and humanities by Harvard seniors and movement away from the physical sciences. Most worrisome, the movement out of the physical sciences is greatest in the high-honors group, although a higher percentage of this group still majors in these fields than of the less select Trueac group. The NSF data indicate, weakly, that Harvard graduates change fields in ways similar to Ph.D.’s nationally. Thus, we can conclude, at least tentatively, that we may

181

Future Graduate Study and Academic Careers

Table 5C. 1

Percentage Distribution of Ph.D’s by Broad Field Other Physical Biological Social Science and Sciences Engineering Sciences Sciences Psychology Engineering

All Other Fields

Ph.D. year 1960-82

All Harvard B.A.’s

11.5% 13.8

9.9% 2.6

11.5% 10.2

9.3% 17.0

8.1% 7.5

8.5% 11.4

41.6% 37.5

8.1 14.1

9.6 8.4

8.7 11.6

40.2 30.0

1.5 0.9

0.2 0.2

198688

All HarvardB.A.’s Change in distribution All Harvard B.A.’s

9.8 11.4

11.5 3.2

12.0 21.0

- 1.7

1.6 0.6

10.8

-2.4

0.5

-1.2 -2.9

-1.4 -7.5

Source: National Science Foundation.

Table 5C.2

Percentage Distribution of Harvard Seniors Physical Sciences

Biological Sciences

Social Sciences

All Other Fields

34.8% 30.4

12.5% 13.9

22.6% 17.2

30.0% 38.2

Trueac 1985-87 1988-90

High honors 49.3 37.7

5.1 9.4

14.3 17.0

31.1 35.8

-4.4 11.6

1.4 4.3

-5.4 2.7

8.2 4.7

1985-87 1988-90

Change in distribution Trueac High honors

-

see an upswing in Ph.D.3 in the biological sciences and humanities in the mid-1990s. A worry about the supply of Ph.D.’s in the physical sciences may also be justified. The evidence for the social sciences is mixed. I find section 5.6 on intended paths to academic careers to be among the most interesting parts of the paper. About 25 percent of Harvard seniors intend to go to graduate school at some point, and of these about half see an academic career in their future. The next step, which I hope Green will pursue, is to find out the slippage between intentions and actual outcomes. For Harvard seniors, some of this can be learned by matching reunion reports with senior surveys. One thing that I have found striking for my own class of Radcliffe women is how few of us are now doing what we planned to do on graduation. It would also be interesting to match these data with additional data about the same individuals. Do choices on the senior survey vary systematically with earlier ability measures, such as SAT scores? Is socioeconomic status

182

Jerry R. Green

(SES) important? In the national longitudinal surveys of high school seniors mentioned above, students from high SES backgrounds are much more likely than other students to attend graduate school (Ekstrom et al. 1991). Finally, I would like to see Green apply his talents as an economic theorist to the question of what models of career choice and uncertainty are consistent with these data. I can outline three possible models: 1 . A matching model. This model, familiar to economists, is one of rational career choice. A student gathers information about his or her abilities in college, becomes knowledgeable about alternative career options, and makes a rational choice. This model would use transcript data combined with salary data as the information that conditions career choice. 2. A social structure model. Here socioeconomic status would be a primary determinant of postgraduation choice. Regardless of their demonstrated ability, we might expect students from higher-SES families to choose higherSES careers. 3. An uncertainty and postponement model. Students who seem to follow this model may be found concentrated in the humanities. Undergraduate performance does not give them enough information to predict with any certainty whether they will succeed in the academic labor market. They may be more likely to choose postgraduation alternatives that are ways of postponing making a career commitment. What Green has observed in his very interesting data are student choices at an important moment in their lives. To know what gives rise to these choices, and how these choices evolve at later points in time, may be helpful as we try to understand fluctuations in the supply of Ph.D.’s-a rare and important social resource. References Ekstrom, Ruth, Margaret Goertz, Judith Pollack, and Donald Rock. 1991. Undergraduate debt and participation in graduate education: The relationship between educational debt and graduate school aspirations, applications, and attendance among students with a pattern of full-time, continuous postsecondary education. GRE Board Research Report no. 86-05R. Princeton, N.J. Hilton, Thomas, and Judith Pollack. 1991. Talent flow in higher education: A longitudinal study of 1980 high school graduates and the sub-group taking the graduate record examination. GRE Board Research Report no. 86-16. Princeton, N.J. National Center for Education Statistics (NCES). 1990. The condition of education, 1990. Vol. 2, Postsecondary education. Washington, D.C.

6

How Would Universities Respond to Increased Federal Support for Graduate Students? Ronald G. Ehrenberg, Daniel I. Rees, and Dominic J. Brewer

6.1 Introduction

Projections of forthcoming shortages of Ph.D.’s, and thus new faculty for the academic sector, abound (e.g., see Bowen and Sosa 1989; National Science Foundation 1989; National Research Council 1990; and Atkinson 1990). The demand for new faculty is projected to grow due to increased retirements from an aging professoriate and projected rises in college enrollments. On the supply side, while the number of Ph.D.’s granted by U.S. universities has been roughly constant in recent years, nonacademic job opportunities are increasingly available to Ph.D.’s. Ph.D. recipients are also increasingly nonU.S. citizens whose observed probabilities of obtaining employment in the United States are low (see Ehrenberg 1991, chap. 7). Integration of supply and demand forces leads to the projections of forthcoming shortage; one major book projected at least a 43 percent underproduction of new doctorates in the arts and sciences as a whole during the 1997-2002 period (Bowen and Sosa 1989, table 8.5). American college graduates are much less likely to receive doctorates today than they were 20 years ago. The ratio of doctorates granted by American Ronald G. Ehrenberg is the Irving M. Ives Professor of Industrial and Labor Relations at Cornell University and a research associate of the National Bureau of Economic Research. Daniel 1. Rees is a visiting assistant professor of economics at Queen’s University (Canada). Dominic J. Brewer is a Ph.D. candidate in labor economics at Cornell University. The authors are grateful to Alan Fechter at the National Research Council for encouraging them to undertake this study and to Cornell University, the National Bureau of Economic Research, and the National Science Foundation for research support. An earlier version of the paper was presented at the May 1991 NBER Conference on the Economics of Higher Education, and the authors are also grateful to the discussant, Michael McPherson, other participants at the conference, and numerous colleagues for their comments on that version. The views expressed are solely those of the authors and do not necessarily reflect the views of any of the above-mentioned institutions or individuals.

183

184

R. G. Ehrenberg, D. I. Rees, and D. J. Brewer

universities to bachelor’s degrees granted by American colleges and universities six years earlier, .064 in 1970-71, fell to .035 in 1978-79 and has remained roughly constant at the lower level since then (Ehrenberg 1991, table 6.4). Numerous factors probably contribute to this decline in the propensity of American college graduates to receive doctorates; however, one important factor may well be the increase in the length of time necessary for doctorate students to complete their programs. The median registered time to degree for new Ph.D.3 granted in the United States in 1968 was 5.5 years. By 1988 this figure had risen to 6.9 years. The increase has been even more dramatic in some fields; for example, median registered time to degree in the social sciences rose from 5.1 to 7.4 years and in the humanities from 5.5 to 8.5 years during the same period (National Research Council 1989, table l).’ Among the policies urged to prevent future Ph.D. shortages is increased federal support for graduate students. Such a policy would reduce the private costs of doctoral study and thus hopefully should increase the number of college graduates willing to undertake graduate study. To the extent that financial support reduces the time students need to complete degrees and increases their probability of completing doctoral programs, the future supply of Ph.D.’s should further increase. While conceptually these roles of financial support on the supply of doctorates are clear, empirical evidence on the effects of financial support on doctoral production actually is quite scanty (see Ehrenberg 1991, chap. 8). Lost in the policy debate, however, has been any concern for the possibility that changes in federal, or other external to the institution, support for graduate education may simply induce an academic institution to redirect its own financial resources in a way that at least partially frustrates the intent of such a policy. For example, increased federal support for graduate students in the sciences may lead an institution to cut back somewhat on (or not increase as rapidly as it had planned) its own internal support for graduate students in the sciences and to use the funds saved either to support graduate students in other disciplines or for other purposes (e.g., non-graduate student expenditures or moderating planned tuition increase). Conversely, faced with cutbacks in federal or other external support, institutions may react by attempting to partially offset the cutbacks by increasing their own internal support for graduate education. To the extent that changes in external financial support for graduate education lead institutions to alter their own support levels or allocations across fields, the resulting changes in the field composition and total number of doctorate students supported may be different than policymakers intended. The issue being raised here is very similar to one confronted by policymakers in 1. Bowen, Lord, and Sosa (1991) have shown that part of the reported increase in times to degree in the humanities is a statistical artifact caused by the grouping of individuals by year of degree rather than by year of program entrance, during a period in which the size of entering cohorts was decreasing.

185

How Would Universities Respond to Increased Federal Support?

the 1970s and early 1980s when concern was expressed that the net job creation effects of public-sector employment (PSE) programs, programs in which the federal government gave state and local governments funds to increase their employment levels, were considerably less than the number of positions funded. Empirical studies of what became known as the displacement efect, orjscal substitution efect, of PSE programs did indeed find that on average an increase in PSE program positions typically led to a smaller increase in public-sector employment levels (see, e.g., Johnson and Tomola 1977; Borus and Hamermesh 1978; Adams et al. 1983).* To fully evaluate the likely effects of an increase in federal support for graduate students, an analysis of the extent to which the federal funds would displace institutional funds is required. Such an analysis is undertaken in this paper, using institutionally based data for science (including social science) and engineering fields. Unfortunately, data do not exist that would permit similar analyses for the humanities and for professional fields other than engineering. We begin in the next section with a discussion of the aggregate time-series evidence on how support for graduate students in science and engineering has changed. While this evidence suggests that federal policies may influence institutional support levels, causation cannot be inferred from these aggregate data. In section 6.3, we present institutionally based econometric analyses of the determinants of the number of full-time graduate students in science and engineering fields that receive institutional support. The analyses are extended in section 6.4 to field-specific data, and attempts are made to ascertain if increased external support to one field may influence internal support allocations to other fields. Section 6.5 further extends the analyses and addresses how different types of external support (e.g., fellowships and traineeships, research assistantships, teaching assistantships) influence the distribution of types of internal support. The brief concluding section summarizes our findings and proposes an agenda for future research.

6.2 Aggregate TCme-Series Evidence Table 6.1 presents evidence for the 1966-88 period on the number of fulltime science and engineering graduate (FTSEG) students in doctorategranting institutions whose major source of support came from the federal government each year. Psychology and the social sciences are included among the sciences for the purposes of this table and those that follow. The data in columns labeled A come from a National Science Foundation 2. More generally, economists have a long tradition of analyzing how various types of federal grants influence state and local government expenditure and taxation decisions (see Gramlich and Galper 1973); recently, economists have also analyzed the extent to which changes in state aid to local school districts influence teacher salaries, student-teacher ratios, and local property tax rates (see Ehrenberg and Chaykowski 1988).

186

R. G . Ehrenberg, D. I. Rees, and D. J. Brewer

Table 6.1

Full-Time Science and Engineering Graduate Students with Federal Support in Doctorate-GrantingInstitutions Number with Federal Support

Year (fall) I966 1969 1970 1971 1972 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988

A 44,612 5 1,620 50,256 45,101 45,029 43,089 48,365 48,508 50.308

B

47,989 48,249 48,594 50,378 51,273 52,874 52,939 50,897 47,206 47,333 47,476 48,716 5 1,060 53,093 54,852

Share with Federal Support

Total Number A 118,273 141,199 145,970 142,169 149,937 169,145 210,641 215,355 218,226

B

195,455 210,321 214,094 2 17,454 216,613 223,414 230,535 234,194 236,939 243,661 245,530 248,782 258,055 263,003 268,385

A ,377 ,366 ,344 .317 ,300 ,255 ,230 ,225 .231

B

,246 ,229 ,227 .232 ,237 ,237 ,230 ,217 ,199 .194 ,193 ,196 ,198 ,202 .204

Sources for data used in authors' computations: Column A-National Science Foundation, Graduate Student Support and Manpower Resources in Graduate Science Education, Fall 1965 and Fall 1966, f i g . 9; Fall 1969, table ClOa; Fall 1970, table C81; Fall 1971, table C9; National Science Foundation, Graduate Science Education: Student Support and Postdoctorals, Fall 1972, table C14; National Science Foundation, Graduate Science Education: Student Support and Postdoctorals, Detailed Statistical Tables, Fall 1974, table B13; Fall 1975, p . 11; Fall 1976, table B10; Fall 1977, table B10. Column B-National Science Foundation, Academic SciencelEngineering: Graduate Enrollment and Support, Fall 1988, table C14; Fall 1991, table C14.

(NSF) survey, the scope of which changed over time. For example, in 1972 the survey was expanded to include graduate students in doctorate-granting institutions in departments that granted only master's degrees, while in 1973 it was expanded to include graduate students in medical and clinical sciences. Response rates to this survey varied over time. The data in columns labeled B come from a separate but similar National Science Foundation survey. Response rates to this survey also varied over time. The two surveys overlapped during the 1974-77 period, yielding virtually identical aggregate numbers for those years. During the 1966-88 period, the number of FTSEG students at doctorategranting institutions whose major source of support came from the federal government fluctuated in the 43,000-to-almost-55,000 range. In recent years,

187

How Would Universities Respond to Increased Federal Support?

however, there has been a clear upward trend. The number of students on federal support rose steadily between 1982 and 1988, and the 1988 level of 54,852 was over 16 percent higher than the 1982 level of 47,206. As the second panel indicates, however, the total number of FTSEG students enrolled in doctorate-granting institutions increased throughout the period, rising (using the consistent series B) from about 195,500 in 1974 to almost 268,400 in 1988. As a result, the share of FTSEG students in doctorate-granting institutions whose major source of financial support came from the federal government fell from almost 38 percent in 1966 to slightly over 19 percent in 1984. Between 1984 and 1988, as the number of FTSEG students with federal support increased, the share with federal support increased slightly to 20.4 percent. However, this is still well below the shares experienced in the late 1960s and early 1970s. Table 6.2 repeats the percentage of FTSEG students with federal support data and adds information on the percentages whose major source of support was institutional funds, other outside funds, and self-support. In these NSF data, institutional funds include funds coming from state governments and administered by the institutions, other outside support includes funds derived from foundation and corporate as well as from foreign sources, while selfsupport includes loans, family support, and earnings from outside the university. Quite strikingly, the fall from 1974 to 1988 in the percentage of FTSEG students whose major source of support was the federal government from 24.6 to 20.4 was substantially offset by the increase in the percentage of FTSEG students whose major source of support was in~titutional.~ As noted above, while this suggests that changes in federal support for graduate students may induce institutions to alter their own support levels, causation should not be inferred from these aggregate time-series data.4 As the distribution of FTSEG students by major source of support has changed, so has the distribution of support recipients changed by type of support. Table 6.3 presents 1968, 1974, and 1988 information, in total and for 3. If the percentage of FTSEG students whose major source of support came from the federal government remained at its 1974 level of 24.6, about 11,000 more FTSEG students would have been supported by federal funds in 1988. About 113,000 FTSEG students’ major source of support was institutional funds that year. If the percentage of FTSEG students whose major source of support came from institutional funds had remained at its 1974 level of 38.5,about 10,000 fewer students would have been supported by institutional funds in 1988. 4. We must also caution that these data refer to students’ major sources of support. Suppose, for example, a student who was initially receiving a $15,OOO tuition waiver from an institution subsequently received a supplementary $16,000 fellowship stipend from the federal government. The student’s reported major source of support would shift from the institution to the federal government. However, no reduction in institutional support would have occurred. Thus, the use of these “major source of support” data may overstate the extent of substitution of external for institutional funds. The reader should keep this in mind when drawing conclusions from the econometric models presented below. Unfortunately, data are not collected on the variety of sources from which a student receives any support.

188

R. G. Ehrenberg, D. I. Rees, and D. J. Brewer Percentage of Full-Time Science and Engineering Graduate Students by Major Source of Support in Doctorate-Granting Institutions

Table 6.2

Federal Funds Year

A

1966 1969 1970 1971 1972 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988

40.9% 36.6 34.4 31.7 30.0 25.5 22.5 23.1

B

24.6% 22.9 22.7 23.2 23.7 23.7 23.0 21.7 19.9 19.4 19.3 19.6 19.8 20.2 20.4

Institutional Funds A

35.0% 35.7 36.9 37.0 38.6 39.9 37.0 36.9

B

38.5% 36.7 37.0 37.0 36.8 37.1 37.6 38.5 39.4 39.5 40.6 41.0 41.6 41.9 42.2

Other Outside support A

6.1% 9.0 9.2 8.8 8.3 8.9 8.2 8.5

B

8.4% 8.0 8.3 8.4 8.9 9.0 9.1 9.6 10.0 10.0 10.0 10.6 10.2 9.5 9.5

Self-support A

18.0% 18.6 19.5 22.4 23.1 25.8 32.3 31.6

B

28.6% 32.4 32.0 31.5 30.6 30.3 30.3 30.2 30.8 31.0 30.1 28.9 28.4 28.4 27.8

Sources: See table 6.1.

selected major fields, on the percentages of FTSEG students in doctorategranting institutions by major type of support. The fellowship category includes fellowships and research traineeships, the RA category represents research assistantships, the TA category represents teaching assistantships, and the other category includes tuition waivers and self- upp port.^ In the aggregate, a steep decline between 1968 and 1988 in the percentage of students supported by fellowships has been offset by a small increase in the percentage supported by research assistantships and by a large increase in the percentage who are on other types of support. Focusing on the 1974-88 period, the almost 6-point decline in the percentage of students supported by fellowships was offset by a slightly larger increase in the percentage of students supported by research assistantships. However, patterns of change vary widely across fields. For example, during the 1974-88 period, the decline in the percentage of students in the social sciences supported by fellowships was offset primarily by an increase in the percentage supported by teaching assistantships. 5. In the NSF data, federal fellowships are offered to students who then decide which institution to attend, while traineeships are granted to institutions who then decide to which students to offer the awards.

189

How Would Universities Respond to Increased Federal Support?

Table 6.3

Percentages of Full-Time Science and Engineering Graduate Students in Doctorate-GrantingInstitutions, by Field and Qpes of Major Support: 1968, 1974, 1988

A Field

B

1968

1974

1974

1988

Total Fellowship* RA* TA* Other

32.0% 22.1 23.3 22.6

20.1% 21.9 24.7 33.3

19.7% 20.3 23.6 36.4

14.0% 27.4 22.9 35.7

Engineering Fellowship RA TA Other

29.4 29.5 13.2 27.9

15.2 34.2 15.2 35.4

14.3 33.0 15.4 37.3

8.7 37.8 17.7 35.8

11.6 30.1 47.3 10.9

8.5 42.6 40.4 8.5

Physical science Fellowship RA TA Other Agriculture** Fellowship RA TA Other

21.6 47.6 8.3 32.5

10.1 45.9 9.0 35.0

10.1 45.8 7.8 36.3

5.8 51.1 9.6 33.4

Biology Fellowship RA TA Other

38.0 9.5 30.2 22.3

24.7 9.6 35.8 29.8

25.7 20.3 26.5 27.5

23.4 36.4 21.6 18.6

Health Fellowship RA TA Other

39.6 5.5 11.0 43.9

27.3 12.1 9.2 51.4

Environmental Science Fellowship RA TA Other

10.7 32.0 24.2 33.1

9.1 38.6 24.6 27.7

9.5 10.3 46.5 33.7

7.5 15.6 40.2 36.9

Math and CIS Fellowship RA TA Other (conrinued)

27.2 8.6 41.3 22.8

10.6 11.3 50.4 27.7

190

R. G . Ehrenberg, D. I. Rees, and D. J. Brewer ~~

Table 6.3

~

(continued)

B

A

Field

1968

1974

1974

1988

Psychology Fellowship RA TA Other

41.1 15.2 21.2 24.5

24.7 12.4 21.6 41.2

24.2 12.1 20.8 42.9

11.0 14.9 22.0 52. I

Social sciences Fellowship RA TA Other

36.2 10.5 18.5 34.8

22.4

21 .o

11.3

11.0

19.4 46.9

17.5 50.5

17.4 11.8 20.2 50.6

Sources: See table 6.1. *Fellowship includes fellowships and research traineeships; RA = research assistantships; TA = teaching assistantships; Other includes tuition waivers and self-support. **I969 figures are reported in the 1968 column.

Changes in external support of a particular type may well affect more than one type of institutional support. For example, an increase in the number of federally funded research assistantships received by an institution might prompt the institution to reduce the number of research assistantships it awards out of institutional funds but, in an effort to attract top students, to increase the number of fellowships it awards out of institutional funds. We provide estimates of such substitution across types of support in section 6.5. Finally, it is worth noting that tables 6.1, 6.2, and 6.3 all refer to full-time students. In the aggregate, as the data presented in table 6.4 show, the percentage of science and engineering graduate students at doctorate-granting institutions who are enrolled part-time has risen over the 1974-88 period. No increase in the proportion of part-time students occurred in the field of engineering, where the sum of the proportions of students on fellowships, research assistantships, and teaching assistantships was higher in 1974 than it was in 1988 (table 6.3). In contrast, an increase in the proportion of part-time graduate students in the fields of psychology and the social sciences has occurred. Although we do not pursue the topic further here, analysis of how changes in federal and other external support levels influence the proportion of graduate students who are enrolled part-time is also of obvious interest.

6.3 Institutionally Based Analyses Consider the following simple equation that seeks to explain the number of FTSEG students in institution j in academic year t supported by institutional funds ( f j f ) .

191

How Would Universities Respond to Increased Federal Support? Percentage of Science and Engineering Graduate Students Enrolled PartTime at Doctorate-GrantingInstitutions

Table 6.4

Total Year

A

1965 1968 1971 1974 1977 1980 1983 1988

27.4% 23.1 21.9 21.4

Engineering

B

26.3% 29.1 30.9 32.0 31.5

A 43.9% 41.1 36.2 31.5

B

40.5% 43.6 40.2 38.8 36.4

Psychology A 16.3% 12.5 12.5 20.4

B

24.0% 24.2 26.6 28.4 28.5

Social Sciences A 23.2% 22. I 24.8 25.7

B

28.0% 31.1 35.8 33.5 34.3

Sources: See table 6.1.

Here XJris the number of undergraduate students that the institution expects to enroll in science and engineering courses during the academic year; F,, is the number of science and engineering faculty employed by the institution in the academic year; A,, is the number of FTSEG students in the institution supported by federal government and other external funds in the academic year; vJris an institution-specific error term; and E ~ is , a random error term. Presumably an increase in undergraduate student enrollments will increase the institution’s demand for teaching assistants, so a, is expected to be positive. While an increase in science and engineering faculty size will similarly increase the institution’s demand for graduate research assistants, holding undergraduate enrollments constant, it might decrease the institution’s demand for teaching assistants. Thus, the sign of u2is a priori indeterminate. The key variable in the model is the number of FTSEG students supported on external funds. At one extreme, if the number of students the institution supports is independent of the number that the federal government and other external sources support, no displacement takes place and a, will be zero. In contrast, if the institution reduces the number of students it supports by exactly the number that the federal government and other external sources support, displacement will be complete and u3will equal minus one. Values of u3 between zero and minus one indicate partial substitution of external for institutional funds. In theory, equation (1) can be estimated using a single year’s data for a cross section of doctorate-producing universities. However, the institution-specific error term presents a problem. Surely there are many other variables besides an institution’s undergraduate enrollments and its faculty size that should affect its willingness to finance graduate students out of its own internal funds. Omission of these variables, which are captured by the institution-specific error term, may lead to biased coefficient estimates. For example, suppose institutions that place a high value on graduate education and research simultaneously support above-average (given their size)

192

R. G. Ehrenberg, D. I. Rees, and D. J. Brewer

numbers of graduate students and hire first-rate faculty, who succeed in attracting above-average levels of support for graduate students from federal government and other external research grants. In the context of equation ( l ) , this can be interpreted as high values for the institution-specific error term (v,,) simultaneously causing the numbers of FTSEG students supported by external (A,,) and institutional (I,,) funds to be high. Thus, a spurious positive correlation will arise between the numbers of FTSEG students supported by institutional and external funds, and if we ignore the institution-specific error term, our estimate of a, will likely be biased. One way around the problem is to try to make the institution-specific error term “observable” by including other variables with which it is likely to be correlated in the analyses (e.g., prestige measures of science and engineering fields in the institution and, in the case of private-sector institutions, measures of the institution’s wealth). While we intend to pursue such strategies in later research, here we adopt a more parsimonious approach. If one is willing to treat the institution-specific error term as fixed over time (v,, = v,), one can obtain data for two time periods ( t and s ) , write equation (1) down for both periods, and then take first differences to obtain

Estimation of ( 2 ) , in which all variables are expressed as changes, will yield unbiased estimates of the parameter of interest, a,, because the unobserved fixed effect has been eliminated from the model. Alternatively, one can obtain unbiased estimates by using the two years of data and estimating an augmented version of the original model that includes institution-specific intercept terms. Table 6.5 presents estimates that use the latter approach and data from 200 doctorate-producing universities on the number of FTSEG students supported on institutional funds during fall 1984 and fall 1983. In each of columns 1, 2, and 3, the number of FTSEG students supported on external funds in the fall of each year is divided into the number supported on federal government funds (GTOT), the number supported on foreign funds ( R O T ) , and the number supported on other U.S., primarily corporate and nonprofit organization, funds (OTOT). In columns 4, 5, and 6, these three sources are aggregated to get a total number of FTSEG students supported on external funds (ATOT). Support is defined here to include fellowships, traineeships, research assistantships, teaching assistantships, and other types (primarily tuition waivers). These data come from the annual National Science Foundation Survey of Graduate Science and Engineering Students and Postdoctorates. Data on enrollments in undergraduate science and engineering courses by institution are not available. What is available from the annual National Center for Education Statistics’ Higher Educational General Information Survey (HEGIS) is the total number of bachelor’s degrees awarded in science and

193

How Would Universities Respond to Increased Federal Support? Determinants of Institutional Support for Full-TIme Science and Engineering Graduate Students in Research and Doctorate Universities, Fall 1983 and Fall 1984: Fixed Effects Model (absolute value &statistics)

Table 6.5

ITOT 1

GTOT OTOT FTOT ATOT TD FTE TD2 FTE2 TDA FTEA

- ,214 (1.9) - ,209 (2.0) -.286(1.9)

2

- .248 (2.3) - .210 (2.0) -.238(1.6)

3

4

5

6

- .240 (2.0) - .I99 (1.9) -.228(1.5) - ,224 (3.3) - ,001 (0.0) .116(3.5)

- ,001 (0.0) ,113 (3.4)

- ,231 (3.5)

- ,221 (3.4)

.082 (2.0) ,217 (4.2)

,080 (1 .O)

,218 (4.0)

.075 ( I .4) ,207 (4.5)

.074 (1.3) ,208 (4.5)

..................................................................................... R* FICE/DOF*

,997 200/194

,991 1971190

,997 1971189

,991 2001191

.997 197/192

,997 197119I

Sources for data used in authors’ computations: ITOT, FTOT, OTOT, GTOT: National Science Foundation, Survey of Graduate Science and Engineering Students and Postdoctorates: Fall 19XX. R E , FTE2, FTEA: National Science Foundation, Survey of Scientific and Engineering Personnel Employed at Universities and Colleges: January 19XX. TD, TD2, TDA: National Center for Education Statistics, Higher Educational General Information Survey (HEGIS):Academic Year 19XX. All of these are available as part of the National Science Foundation’s Computer Aided Science Policy Analysis and Research Database System (CASPAR). However, ITOT is not reported in CASPAR and the underlying data tapes must be used to obtain this variable. Note: All specifications in this table are estimated using the ABSORB command in Proc GLM in SAS. Definitions: ITOT = Number of full-time science and engineering graduate (FTSEG) students supported by instituitonal and state funds on fellowships, traineeships, research assistantships, teaching assistantships, or other types (primarily tuition waivers) of support in the fall of year t GTOT = Number of FTSEG students supported by federal government funds in the fall of year t FTOT = Number of FTSEG students supported by foreign funds in the fall of year t OTOT = Number of FTSEG students supported by other U.S. (primarily corporate and nonprofit) funds in the fall of year t ATOT = Sum of GTOT, R O T , and OTOT TD = Total bachelor’s degrees in science and engineering awarded by the institution in the academic year TD2 = Same as TD but for academic year t 1 = Average of TD and TD2 TDA FTE = Total full-time scientific and engineering personnel employed by the institution in January of year t FTE2 = Same as FTE but for January of year t 1 FTEA = Average of FTE and FTE2

+

+

194

R. G. Ehrenberg, D. I. Rees, and D. J. Brewer

engineering fields by an institution in each academic year.6 While there is not necessarily a one-to-one relationship between changes in course enrollments and changes in graduating majors, the latter is the best proxy available for the former. Changes in degrees granted may well also lag changes in undergraduate enrollments. Hence, it is not clear, for example, whether bachelor’s degrees granted in science and engineering in 1983-84 (TD) or those granted in 1984-85 (TD2) should be the best predictor of the demand for graduate teaching assistants in fall 1984. Results are presented in table 6.5 for specifications that use both measures, as well as their average (TDA). Finally, no data exist by institution on the number of faculty employed in science and engineering fields. However, from 1973 to 1985, the National Science Foundation’s Survey of ScientiJic and Engineering Personnel Employed at Universities and Colleges collected information from doctorategranting institutions in January of each year on the total number of full-time scientists and engineers employed.’ These headcounts are not restricted to faculty nor even to doctorates, but they probably provide a reasonable approximation to the scale of research and teaching activity in science and engineering fields in the institution. Restricting the headcount to full-time employees assures that graduate assistants are not included in the total. Again, it is not a priori obvious whether the best predictor of the demand for research and teaching assistants in the fall of a year would be the number of full-time scientists and engineers employed in the institution in January of that year (FTE), which represents the previous academic year, or in January of the next year (FTE2), which represents the current academic year. Specifications are thus again estimated using both measures, as well as their average (FTEA). The results displayed in table 6.5 suggest that changes in external support for FTSEG students do influence institutional support levels. The institutional responses to changes in the various sources of external support (GTOT, OTOT, and R O T ) reported in columns 1, 2, and 3 appear to be quite similar; indeed, formal F tests indicate one cannot reject the hypothesis that they are all equal. When the various sources are aggregated (ATOT), the specifications in columns 4, 5 , and 6 suggest that for every 100 additional FTSEG students supported by external funds, institutions reduce the number of FTSEG students supported by institutional funds by 22 to 23. Whether the money saved was used to support graduate students in other fields or for other purposes cannot be determined from these data. The above results assume instantaneous adjustment of the number of FTSEG students supported on institutional funds, the number of degrees granted, and faculty size. However, commitments to support graduate students are often made, at least implicitly, for more than one year at a time. As 6. In recent years, the scope of the HEGIS has been expanded, and it is now called the Inregrated Postsecondary Education Data System (IPEDS). 7. The cessation of this survey in January 1985 precludes us from using more-recent data on institutional and external support for graduate students in our analyses.

195

How Would Universities Respond to Increased Federal Support?

such, considerable inertia may be built into the process, and the substitution of external for internal funds may be greater in the long run than in the short run. One way to test for this is to build a lagged adjustment process directly into the model. Suppose that equation (1) is replaced by

(3)

I;

=

b,

+ bJ,, + b,FIf+ bd,, + v, + E,~,

where 1; is the number of FTSEG students that institution j desires to support out of its own funds in year t . Because of the inertia caused by multiyear commitments to graduate students and the institution’s goal of maintaining relatively stable graduate enrollments and financial commitments to graduate students, I; is assumed to adjust to its desired number of institutionally supported FTSEG students only gradually, Specifically, suppose that

I,,

(4)

-

I,,-l = q; - I,,-l).

where A (0 CA 5 1) is the fraction of the adjustment between this year’s desired and last year’s actual number of FTSEG students supported on institutional funds that the institution makes in the year. Substitution of (3) into (4) yields that ( 5 ) I,, = Ab,

+ Ab,X,, + Ab,F,, + Ab,A,, + Av, + ( I - A)Zlf- + A&,,. I

First, differencing to eliminate the unobservable fixed effects, one finds that (6)

I,, -

= Ab,(X,,-X,,-l)

Ab,(A,,-A,,-

I)

+

(1 -A)U,,-

+

Ab2(F,,-F,,-,) + I -I,,-,) + A(&,, - &-,

1)

Equation (6) differs from equation (2) in that the lagged change (from t - 1 to t - 2) in the number of FTSEG students supported by institutional funds appears on the right-hand side of (6). With three adjacent years’ data on the number of students supported on institutional funds (here, data for fall 1984, 1983, and 1982), one can obtain consistent estimates both of the magnitude of the lagged adjustment term (A) and of the extent to which external support substitutes for internal support (from b,). To achieve this, an instrumental variable estimator must be used for I,,- I to remove the spurious negative correlation between that variable and the error term A(&,, - E,,- that the first differencing causes.8 Estimates of equation (6) appear in table 6.6 for the specifications that correspond to those found in columns 1,2, and 3 of table 6.5. Column A for each specification uses the actual value of the lagged one-year change in the number of FTSEG students supported on institutional funds as an explanatory variable, while column B in each specification uses an instrumental variable 8. The variables used as instruments include I,,-, and the values from periods t - I and t - 2 of all the other explanatory variables in the model.

196

R. G. Ehrenberg, D. I. Rees, and D. J. Brewer

Table 6.6

Determinants of Institutional Support for Full-Time Science and Engineering Students in Research and Doctorate Universities: Lagged Adjustment Model with Fixed Effects (absolute value t-statistics) CITOT la

CGTOT CFTOT COTOT CFTE CTD CFTE2 CTD2 CFTEA CTDA CITOTL*

- ,223 (2.1) - ,147

(1.O) - .298 (2.8) -096 (3.1) - ,031 (0.9)

.096(1.6)

Ib

2a

2b

- .242

- ,249 (2.4)

- .I73

- ,150 (1

(2.3) (1.2) - ,254 (2.5) .095 (3.0) - .015 (0.4) .149 (2.7) ,041 ( I .O)

-.005(0.0)

3b

- ,247 (2.4) - ,124 (0.8)

- ,260 (2.5)

- ,241 (2.3)

- ,260 (2.5)

- ,233 (2.3)

,163 (3.5) ,002 (0.0) -.017(0.1)

,164 (3.5) ,024 (0.4) ,071 (1.1)

- ,257 (2.4)

.O) - ,174 (1.1)

- .254 (2.4)

3a

- ,146 (1 .O)

,152 (2.7) ,047 ( I . 1)

.055(0.9)

-.OO2(O.O)

. . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . R2

DOF

,090 187

,078 187

,083 188

,079 187

,010 187

,093 187

Sources: See table 6.5. Notes: CITOT = ITOT(84) - ITOT(83) CFTE FTE(84) - FTE(83) CFTEA = [FTE(85) - FTE(83)]2 = TD(84) - TD(83) CTDA = [TD(85) - TD(83)]2 CGTOT = GTOT(84) - GTOT(83) CTD CFTOT = flOT(84) - FTOT(83) CFTE2 = FTE(85) - FTE(84) CITOTL = ITOT(83) - lTOT(82) COTOT = OTOT(84) - OTOT(83) CTD2 = TD(85) - FTE(84) See table 6.5 for variable definitions.

estimator. The “C” in front of each variable name indicates that each is in firstdifference form. Quite strikingly, in no case can one conclude that A is statistically significantly different from zero. Put another way, institutions appear to fully adjust, to their desired levels the number of FTSEG students they support out of internal funds each year. The first three columns of table 6.7 report similar estimates for the specifications that aggregate the various external support sources into a single variable (ATOT). Given the statistical insignificance of the lagged change in the number of students supported on internal funds in the previous table, only the specification that uses the actual lagged change is reported here. Again, adjustment appears to be complete (X = o), and displacement appears to be in the range of - .23. The last three columns of table 6.7 report estimates of specifications in which the extent to which the number of FTSEG students supported by institutional funds varies with the number supported on external funds is allowed to vary across public and private institutions and across research I universities and other doctorate-granting institutions. Research I universities are those that award at least 50 Ph.D. degrees annually and receive at least $33.5 million

197

How Would Universities Respond to Increased Federal Support? Determinants of Institutional Support for Full-Time Science and Engineering Students in Research and Doctorate Universities: Lagged Adjustment Model with Fixed Effects and All External Support Sources Aggregated Together (absolute value t-statistics)

Table 6.7

CITOT 1

2

3

4

5

6

.143 (0.8) - ,382 (2.9) - ,143 (0.9)

,128 (0.7)

~~

- ,239 (3.7) CATOT CATOT*Rl CATOT*P CFTE ,093 (3.0) - ,026 (0.7) CTD CFTE2 ,146 (2.7) CTD2 ,041 (1.0) CFTEA CTDA CITOTL ,078 (1.4) R2

DOF

,095 189

- ,234 (3.7)

- .231 (3.7) - ,406 (3.0) - ,127 (0.8)

,121 (0.7) - ,405 (3.1) - ,145 (0.9) ,083 (2.7) -.039 (1.1) ,140 (2.6) ,045 (1.1)

,158 (3.4) .006 (0.1) ,047 (0.8)

,060 (1.0)

.046 (0.8)

.090 I90

,106 189

,131 187

,145 (3.1) - ,007 (0.1) ,009 (0.1) ,128 188

,028 (0.4) ,138 187

Dejinirions; All variables are defined in tables 6.5, and 6.6 save for:

+

CATOT = [GTOT(84) FTOT(84) + OTOT(84)l R I = I for research I institutions, 0 for other P = I for public institutions, 0 for other

-

[GTOT(83) + FTOT(83) + ITOT(83)J

annually in federal research support. Most award considerably more than 50 science and engineering Ph.D.’s each year.9 These specifications suggest that substitution of external for institutional funds supporting graduate students occurs only at the relatively large (in terms of doctorates produced and external research support generated) research I institutions. No such substitution tends to occur in those institutions with smaller-scale doctorate and research programs. Furthermore, the extent of substitution of external for institutional funds at research I institutions does not appear to differ between public and private institutions.

6.4 Disaggregation by Field To conclude that, in the aggregate, when the number of FI’SEG students supported by external funds increases by 100, institutions reduce the number of FTSEG students they support out of institutional funds by about 22 to 23 is not to say that the response will be the same across all fields. To address the latter issue requires that separate analyses be undertaken by field. A first approach is to estimate variants of equation ( l ) , using field-specific data. Data on institutional and external FTSEG student support levels, the 9. For example, in 1988, 70 institutions awarded at least 100 science and engineering Ph.D.’s, with Berkeley alone awarding 576 (see National Science Foundation 1989, table 10).

~~

198

R. G. Ehrenberg, D. I. Rees, and D. J. Brewer

number of full-time scientific and engineering personnel employed, and the number of bachelor’s degrees granted were collected by institution for seven broad science and engineering subfields. Field-specific equations were estimated, and the coefficients of the external support variables that were obtained are displayed in panels A and B of table 6.8. The coefficients of the external support variables for each field in panel A come from field-specific specifications similar to the specification found in column 1 of table 6.5. The effects on internal support levels of changes in federal government, other U.S., and foreign support levels often appear to differ from each other at this level of disaggregation. Formal F tests indicate that this is indeed the case. l o Only for the engineering and mathematical sciences fields can one not reject the hypothesis that the marginal effects of changes in the number of FTSEG students supported by the various external funding sources are equal. Nonetheless, it is interesting to aggregate the external support variables and estimate what the “average” substitutability of internal for external support is for each field. The results obtained when one does this are found in panel B of table 6 . 8 ; the coefficient estimates presented there come from field-specific variants of the model estimated in column 4 of table 6.5. External support appears to partially substitute for internal support in six of the seven fields. This substitution is statistically significant in five of these six fields. The magnitude of the substitution ranges from almost 50 percent in the physical sciences, where an additional 100 FTSEG students supported on external funds are estimated to reduce the number of internally supported students by about 48, down to about 10 percent in the mathematical sciences. Only for the relatively small environmental sciences fields do increases in external support appear to be associated with increases in internal support. I ’ There is weak evidence that fields which, on average, have a greater share of their students supported on institutional funds tend to reduce their own internal support for FTSEG students the most when the number of externally supported students is increased.’* The model that underlies the estimates presented above treats each field separately and does not allow for the possible interdependency of internal 10. The computed F-statistics are: Engineering Physical sciences Life sciences Social sciences

F(2,137) = 1.22 F(2,180) = 3.48 F(2,183) = 4.61 F(2.179) = 3.59

Environmental sciences Psychology Mathematics

F(2.147) = 5.47 F(2,176) = 3.73 F(2.181) = 1.04

In each case, the critical value to reject the null hypothesis at the .05 level is 3.09. 11. In October 1984, only 4.6 percent of all FTSEG students in doctorate-granting institutions were enrolled in environmental science fields (National Science Foundation 1990, table CI). 12. Across the seven fields, the correlation of the average proportion of supported students in a field supported by institutional funds and the estimate of the substitution of external for internal funds in the field (the coefficients in panel B) is - .32. However, if one drops environmental sciences from the sample, the correlation across the six remaining fields falls to under - .2.

How Would Universities Respond to Increased Federal Support?

199

Determinants of Institutional Support for Full-Time Science and Engineering Graduate Students in Research and Doetorate Universities, Fall 1983 and Fall 1984: Fixed Effects Model, by Field

Table 6.8

Physical Sciences

Engineering

Life Sciences

Social Sciences

Environmental Sciences

Psychology

Mathematical Sciences

.316(2.6) .434(2.1) - .I08 (0.8)

-.659(4.6) 1.479(1.6) - .I63 ( 1 .O)

-.044(0.2) .193(0.8) - .22 1 (1.6)

A

GTOT -.148(1.1) R O T -.059(0.3) OTOT - ,270 (2.9)

-.522(6.5) -.425(1.2) - .480 (3.4)

-.486(7.3) -.464(2.1) .036 (0.2)

.094(0.7) -.346(3.0) - .450 (3.3)

.............................................................................................. 142

FICE

187

190

186

185

151

I88

~

B ATOT-.204(2.7)

-.479(7.1)

-.380(7.4)

-.219(3.2)

ATOT TOT

- .653 (6.7) - .021 (1.0)

-.328 (3.7) -.019 (0.4)

-.318 (3.3) .042 (1.0)

- ,199 (1.9) - ,028 (0.6)

.251 (2.8)

-.412(3.7)

-.104(1.2)

,232 (2.0)

- ,499 (4.5)

- ,001 (0.1)

,018 (1.4)

-.I62 (1.6) ,038 (2.0)

.............................................................................................. D ATOT -.149(1.5) TOT - . W ( I . O )

-.633(6.8) -.024(1.2)

-.310(3.7) -.026(0.5)

-.318(3.4) .043(1.0)

.233(2.1) -.001 (0.0)

-.517(4.8) .019(1.5)

-.I71 (1.8) .038(2.0)

....................................................................................... 113

FICE

113

1 I3

I I3

1 I3

113

1 I3

Notes: Panel A, same specification as table 6.5, column I , but all data field-specific; Panel B, same specification as table 6.5, column 4, but all data field-specific; Panel C, same as B, but TOT added; Panel D, same as C, but seemingly unrelated regression method used, where TOT = sum of ATOT across all seven fields and FICE = number of institutions included in the analyses.

support levels across fields. So, for example, an increase in the number of students supported on external funds in one field might induce an institution to reduce the number of students it supports out of institutional funds in that field and then use all or part of the savings to fund more graduate students out of internal funds in other fields. One way to test whether such interdependencies exist is to estimate a system of equations of the form (7)

‘jkf

=

‘Ok

+

‘t&jkr

+

‘ Z j k f

+

‘3djkf

+

‘&jr

+

‘jk

+

‘jk,

k = 1 , 2 . . . . . 7.

In the above equations, the subscript k indexes the field of study. The number of students in the field supported out of institutional funds (Iju) is assumed to depend on both the number of students in the field supported by external funds (Ajk,)and the number of students supported by external funds in the institutions as a whole (Ajf).Other factors staying the same, an increase of 100 in the number of students in field k supported by external funds would lead to a change of 100(a3, + a4& in the number of students in field k supported by internal funds. Similarly, an increase of 100 in the number of FTSEG students

200

R. G . Ehrenberg, D. I. Rees, and D. J. Brewer

supported in the institution as a whole by external funds, with no increase in the number of students in field k supported by external funds, would lead to a change of 100a,, in the number of students in field k supported by internal funds. A positive estimate qt thus indicates that part of any increase in external support for graduate students elsewhere in a university is implicitly used to support graduate students in field k. Given two years of data, one can first difference the data to eliminate the assumed institutiodfield fixed effects (v,J and obtain consistent estimates of the parameters from the system of equations in (7). The coefficients that result for the number of FTSEG students with external support in the field (ATOT) and in the institution as a whole (TOT) are displayed in panels C and D of table 6 . 8 . The data used here come from a sample of 113 institutions that reported data in both years for all seven fields. The estimates reported in panel D use the seemingly unrelated regression method to improve efficiency by taking account of the correlation of the error terms across fields within an institution. In most cases, these estimates vary only marginally from the estimates reported in panel C. Of key interest are the estimated coefficients for TOT. These estimates suggest that increases in the overall number of students supported by external funds in the science and engineering fields are used partially to subsidize graduate education in the social sciences, psychology, and mathematical sciences. However, only the latter effect is statistically significantly different from zero. Other factors being equal, an increase of 100 in the number of FTSEG students supported by external funds outside of these fields leads to an increase in the number of students supported on institutional funds of roughly 4 in the social sciences, 2 in psychology, and 4 in the mathematical sciences. As noted in earlier sections, whether a similar subsidization of graduate education in the humanities occurs cannot be ascertained from these NSF data because they lack information on graduate student support in humanities fields.

6.5 Disaggregation by ‘Qpe of Support

FTSEG students who are supported from external funds often have different types of support than those who are supported from institutional funds. For example, the former are more likely to receive research assistantships, while the latter are more likely to receive teaching assistantships. l 3 13. More generally, in fall 1984 the proportions of FTSEG students supported from institutional and external funds, by type of support, in our sample were:

Institutional External

Fellowship/ Traineeship

Research Assistantship

Teaching Assistantship

,140 .279

.I16 ,515

.574 ,018

Other .110 ,188

201

How Would Universities Respond to Increased Federal Support?

It is possible that an institution that receives an increase in one type of external support for FTSEG students may reduce the number of students that it supports out of institutional funds on that type of support and use some or all of the savings to increase the number of FTSEG students it supports internally on other types of support. So, for example, an increase in external support for research assistants may lead an institution to reduce the number of research assistantships it offers out of institutional funds but to increase its allocation of internal funds to teaching assistantships and fellowships. To allow for this possibility, equation (1) can be generalized to the fourequation system (8)

tI‘

= ~ o T+

a,&

+

+ ‘A&Z,r

+ vT,r +

E,n

T, Z = SUM, TA, RA, OTH. The numbers of FTSEG students supported from institutional and external funds are decomposed in each case into the numbers supported on fellowships and traineeships (SUM), on teaching assistantships (TA), on research assistantships (RA), and on other types-primarily tuition waivers-of support (OTH). Assuming that the institutionspecific error terms are fixed over time (vT,, = vT,), with two years of data one can again estimate the equations in first-difference form to obtain unbiased estimates. Estimates of this system appear in table 6.9. While an increase in the number of FTSEG students supported by externally funded research assistantships is associated with a decrease in the number of FTSEG students supported by institutional research assistantships, a large share of these “saved” institutional funds are redirected toward increasing the number of students supported by institutional teaching assistantships and fellowships. An increase in external funding for teaching assistantships leads to a substantial reduction in institutional teaching assistantships. In contrast to the research assistantship results, however, none of these “saved’ institutional funds appears to be diverted to other types of support for graduate students. Finally, changes in external fellowships and traineeships and in other types of support each seem to affect primarily other, rather than the same, internal types of support. Similar estimates of the coefficients of the various external types of support variables appear in table 6.10 for analyses done separately by field. Increases in external support for fellowships and traineeships lead to statistically significant reductions in institutional support for fellowships in five of the seven fields. Similar statistically significant “own substitution” effects occur in four of the seven fields for research assistantships and five of the seven fields for teaching assistantships. Many statistically significant “cross-substitution” effects are present, although the pattern is not always consistent across fields. For example, an increase in external fellowship support is associated with an increase in institutional teaching assistant support in the life sciences but with a decrease in such support in the social sciences. Findings of this type confirm the need to undertake separate analyses by field.

R. G. Ehrenberg, D. I. Rees, and D. J. Brewer

202

Determinants of Institutional Support for Full-Time Science and Engineering Graduate Students in Research and Doctorate Universities, Fall 1983 and Fall 1984: Fixed Effects Model, by Qpe of Support (absolute value t statistics)

Table 6.9

ISUM ASUM ARA ATA AOTH FTE

-.066(1.2) .091 (2.1) - ,172 (0.8) - ,148 (3.3) .I09 (1.3) ,001 (0.0)

TD RZ RCWDOF

,990 200/188

IRA

- .039 (0.4) - .205 (2.5) - ,602 ( 1.5) - ,189 (2.3) .058(2.3) - .008 (0.3) ,990 2001188

ITA

IOTH

- .086(1.2) .106(1.8) - ,796 (2.7) -.067 (1.1) .015 (0.8) - .013 (0.7)

.122(2.1) -.082 (1.8) -.267(1.1) .013 (0.3) ,027 (0.1) .001 (0.1)

,998 2001188

,961 2001188

Dejnition'S.' ISUM = Number of FTSEG students supported by institutional and state funds on fellowships

and traineeships

ITA

= Number of FTSEG students supported by institutional and state funds on teaching as-

IRA

= Number of mSEG students supported by institutional and state funds on research as-

IOTH

= Number of FTSEG students supported by institutional and state funds on other (primar-

sistantships sistantships ily tuition waivers) types of support ASUM = Same as ISUM but supported by federal government, foreign, or other US (FFO) funds ARA = Same as IRA but supported by FFO funds = Same as ITA but supported by FFO funds ATA AOTH = Same as IOTH but supported by FFO funds Other variables are defined in table 6.5.

6.6 Concluding Remarks This paper has demonstrated that doctorate-producing universities respond to changes in the number of FTSEG students supported on external funds by altering the number of FTSEG students that they support on institutional funds. While institutional adjustment to changes in external support levels appears to be quite rapid, in the aggregate the magnitude of these responses is quite small. A increase of 100 in the number of FTSEG students supported by external funds is estimated to reduce the number supported on institutional funds by 22 to 23. Since some of the institutional funds that are "saved" may be redirected to support graduate students in the humanities and other fields not represented in the data, the total effect of such a policy change on institutional support for graduate students is probably somewhat smaller. Two qualifications are in order here. First, institutions are likely to react quite differently to changes in external support levels that they perceive as being transitory as opposed to changes that they perceive as being permanent.14 Transitory increases, which are not expected to recur in future years, 14. We owe this point to Michael McPherson.

203 Table 6.10

Field Engineering ASUM ARA ATA AOTH

How Would Universities Respond to Increased Federal Support? Determinants of Institutional Support for Full-TLmeScience and Engineering Graduate Students in Research and Doctorate Universities, Fall 1983 and Fall 1984: Fixed Effects Model, by Field and 'Qpe of Support (absolute value t statistics) ISUM

IRA

ITA

IOTH

-.167 (1.4) .042 (0.9) - ,870 (2.7) ,072 (1.2)

- .011 (0.2) - .144 (5.2) - .437 (2.3) -.042 (1.2)

.089 (0.8)

- ,116 (2.4) -.022 (1.1) - ,063 (0.5) - .048( 1.9)

.080 (0.6) - .089 (1.6) - .657 (1.7) - .045 (0.6)

Physical Sciences ASUM - ,189 (3.5) ARA - ,090(2.8) ATA - .028 (0.1) AOTH - ,014 (0.2)

-.139 (1.7) - ,243 (4.9) .181 (0.4) ,041 (0.4)

- 1.609 (2.6) - .447 (3.0)

- ,003 (0.1) ,030 (1.8) - .120 (0.8) ,016 (0.5)

Life Sciences ASUM ARA ATA AOTH

,112 (3.5) - ,190 ( 1 . 1 ) ,075 (1.3)

-.311 (3.1) - ,452 (7.0) - ,480 (1.4) - ,336 (2.9)

.222 (3.2) - .087 (2.0) - .200 (0.8) .050(0.6)

.055 (1.3) - ,053 (2.0) - ,067 (0.5) .043 (0.9)

Social Sciences ASUM - .174 (3.2) ARA ,125 (0.9) ATA .402 (0.7) AOTH - ,302 (3.9)

,064(1.3) - .lo4 (0.8) - ,299 (0.6) - ,052 (0.8)

- .094 (2.0) - ,016 (0.1) -.517 (1.1) .003 (0.1)

- .069 (1.2) ,143 (1 .O) 1.220 (2.1) ,092 (1.2)

Environmental Sciences ASUM ,057 (0.8) ,029 (0.5) ARA .080 (0.3) ATA .073 (0.8) AOTH

- .018 (0.2) - ,079 (1.4) - .222 (0.8) - .030 (0.3)

,264 (2.7) ,074 (1.0) - ,231 (0.6) ,188 (1.5)

,051 (0.7) - ,017 (0.3) - 1.379 (5.2) ,079 (0.9)

Psychology ASUM ARA ATA AOTH

- ,126 (1.1) - ,241 (2.1) .099 (0.3) - .I20 (1.9)

.069 (0.6) - .126 (1.1) - .088 (0.3) .037 (0.6)

- ,215 (1.7) - ,096 (0.8) - 1.072 (3.3) - .090 (1.3)

- ,609 (3.3) ,025 (0.1) - .214 (0.5) - ,050(0.5)

Mathematical Sciences ASUM - S48 (4.5) ARA .I27 (1.5) ATA - ,178 (0.6) AOTH . I 14 ( 1.4)

.106(1.1) - .444(6.7) -.326(1.5) .061 (1.0)

.310 (1.7) .182(1.5) - 1.317 (3.2) ,024 (0.2)

,284 (4.1) -.I81 (3.7) ,162 (1.0) - .015 (0.3)

- .126 (2.5)

- ,310 (4.5)

Note: The underlying model is the same as that estimated in table 6.9, save that all variables are fieldspecific. See table 6.9 for variable definitions.

204

R. G. Ehrenberg, D. I. Rees, and D. J. Brewer

are unlikely to lead to large reallocations of institutional funds. Institutions may treat such increases as windfalls and compensatingly reduce their own expenditures for graduate support temporarily. In contrast, permanent increases, which institutions may view as fundamentally altering their wealth levels, are likely to lead to larger institutional commitments to graduate education and thus to less substitution of external for institutional funds. To the extent that the variation in changes in external support levels across institutions during a two-year period reflect primarily transitory fluctuations, our estimates may thus well overstate the extent to which institutions would reduce their own internal support for FTSEG students in response to an increase in external support that was perceived to be more permanent. Second, changes in external support levels in one year may affect the intertemporal allocation of institutional funds to support FTSEG students. l 5 For example, the provision of external fellowships to support first-year entering graduate students in a field in year t might induce an institution to reduce its internal support for entering students in the field in year t . However, to the extent that substitution was not one for one, the size of its entering class will have increased and thus the number of advanced FTSEG students who “need” support will increase in subsequent years. To the extent that an institution uses some, or all, of the “saved” internal funds in year t to support an increased number of FTSEG students in subsequent years, focusing on contemporaneous responses (as we have done) will overstate the extent of substitution of external for institutional funds. A similar result would occur if institutions that previously provided support to students for four years used some of the saved internal funds in year t to provide fifth-year support in year t + 4 for some of the new students who entered in year t. Policymakers also need be concerned that the magnitudes of the responses appear to differ significantly across fields. There is also evidence that even within science and engineering there is some fungibility of external support across fields. In particular, institutional support for the social sciences, psychology, and the mathematical sciences appears to increase somewhat in response to increases in external support to other science and engineering fields which permit institutions to reduce their own support to these other fields. Finally, policymakers need be concerned that changes in external support levels influence the distribution of institutional support by type of support. For example, in the aggregate an increase in the number of FTSEG students supported by externally funded research assistantships is associated with a decrease in the number of FTSEG students supported by institutional research assistantships. However, a share of these “saved” funds is redirected to increasing the number of students receiving teaching assistantships out of institutional funds. It is often conjectured, although it has not been proven, that 15. We owe this point to Robert Hauser,

205

How Would Universities Respond to Increased Federal Support?

teaching assistantships slow down degree progress relative to research assistantships (see Ehrenberg 1991, chap. 8). As such, the latter shift may partially frustrate the goal of policymakers when they increase external support for research assistantships for FTSEG students. The analyses reported in this paper are only a start at addressing the issues we pose. To a large extent, they focus on changes in external and institutional support levels between fall 1983 and fall 1984. While this was a period in which approximately half of the institutions in the sample faced increases in external support and half faced decreases, one wonders whether institutional responses would differ in periods when external support changes all tended to move in one direction and, more generally, whether institutional responses are stable over time. As discussed above, our focus on this one-year period also precluded us from distinguishing between institutional responses to transitory and permanent changes in external support for graduate students and from analyzing how such changes influence institutions’ intertemporal decisions on allocating internal funds. Subsequent research by us will attempt to use a panel of 11 years’ data (1974-84 period) from these institutions to address these issues. Throughout the paper, differences in institutional characteristics that might influence universities’ desire and willingness to support graduate students are, for the most part, “buried” in the unobservable fixed effects. Generalizations of the empirical models could productively be explicitly tied to models of university utility maximization subject to budget constraints (see, e.g., Garvin 1980; James 1990). One implication that likely flows from such an approach is that institutional support for graduate students should depend on the “wealth” levels of institutions. This suggests that measures of state budgetary tightness (in the public sector) or endowment strength (in the private sector) are candidates to be added to the empirical models. Similarly, an institution’s willingness to support graduate students in a field may well depend upon the “quality,” or the recent change in the “quality,” of the field and of other fields in the institution. As such, estimation of whether the extent that external funds substitute for internal funds varies with field quality measures is also clearly warranted.

References Adams, Charles, et al. 1983. A pooled time series analysis of the job creation impact of public service employment grants to large cities. Journal of Human Resources 18(Spnng):283-94. Atkinson, Richard C. 1990. Supply and demand for scientists and engineers: A national crisis in the making. Presidential address delivered to the American Association for the Advancement of Science. New Orleans, La.

206

R. G . Ehrenberg, D. I. Rees, and D. J. Brewer

Borus, Michael, and Daniel Hamermesh. 1978. Estimating fiscal substitution by public service employment programs. Journal of Human Resources 12(Fall):561-65. Bowen, William G., Graham Lord, and Julie Ann Sosa. 1991. Measuring time to the doctorate. Proceedings of the National Academy of Sciences 88(February):7 13-17. Bowen, William G., and Julie Ann Sosa. 1989. Prospects for faculty in the arts and sciences. Princeton, N. J.: Princeton University Press. Ehrenberg, Ronald G. 1991. Academic labor supply. Part 2 of Economic challenges in higher education, ed. Charles Clotfelter, Ronald Ehrenberg, Malcolm Getz, and John Siegfried. Chicago: University of Chicago Press. Ehrenberg, Ronald G., and Richard P. Chaykowski. 1988. On estimating the effects of increased aid to education. In When public sector workers unionize, ed. Richard B. Freeman and Casey Ichniowski. Chicago: University of Chicago Press. Garvin, Donald. 1980. The economics of university behavior. New York: Academic Press. Gramlich, Edward M., and Harvey Galper. 1973. State and local fiscal behavior and federal grant policy. Brookings Paper on Economic Activity 4( 1):15-58. James, Estelle. 1990. Decision processes and priorities in higher education. In The Economics of American universities, ed. Stephen Hoenack and Eileen Collins. Albany: State University of New York Press. Johnson, George, and James Tomola. 1977. The fiscal substitution effects of altemative approaches to public service employment. Journal of Human Resources 12(Winter):3-26. National Research Council. 1989. Summary report 1988: Doctorate recipients from United States universities. Washington, D.C. : National Academy Press. . 1990. Biomedical and behavioral research scientists: Their training and supply. Vol. 1, Findings. Washington, D.C.: National Academy Press. . 1989. Future scarcities of scientists and engineers: Problems and solutions. Washington, D.C.: National Science Foundation, Division of Policy Research and Analysis, Directorate for Scientific, Technological andd International Affairs. Mimeograph. , 1990. Academic science engineering: Graduate enrollment and support, fall 1988. Washington, D.C.: National Science Foundation.

Comment

Michael S . McPherson

It is useful, I think, to locate this valuable paper in relation in two literatures. One is the stream of research in public finance that concerns itself with the “flypaper effect”-the proposition that grant money provided by an external agency to an institution or a lower level of government, even when it is fungible in principle, tends to “stick where it hits.” ’ The other stream of literature is the small but recently growing set of empirical studies of the behavior of colleges and universities. A classic in this literature is David Breneman’s Michael S. McPherson is a professor of economics at Williams College. 1 . In addition to the studies mentioned by the authors concerning the employment effects of federal support for public-sector employment, there are numerous studies on topics ranging from welfare and health care to education. For a brief survey, see McPherson and Schapiro (1991), chapter 4.

207

How Would Universities Respond to Increased Federal Support?

dissertation on the impact of institutional incentives on time to degree for graduate students at the University of California at Berkeley-a study later reported in an NBER volume (Breneman 1976). Recent work includes a fine study by Steve Hoenack and Dan Pierro (1990) of the interaction between the University of Minnesota and the state legislature, which includes empirical estimates of the legislature’s reaction functions, and some work Morton Schapiro and I (McPherson and Schapiro 1991, chap. 4) have done studying the impact of changes in external funding on universities’ allocation of resources across a set of activities. An examination of either or both of these literatures points to the real difficulties in doing this kind of thing well. One is trying to extract a single set of behavioral relationships from among a set of mutually dependent decisions made by an institutional actor. Many relevant variables are hard to measure, and theory gives us only limited guidance about suitable empirical specifications. When the actor whose behavior is being modeled is a governmental or not-for-profit institution, we are in even worse shape for theory, since we don’t have behavioral models of this kind of “firm” that we have much confidence in. However, the only way to make progress is to try to push ahead on both empirical and theoretical fronts, and Ehrenberg, Rees, and Brewer’s study provides an interesting first attempt to examine universities’ behavioral responses to fluctuations in external support for graduate students. The authors find evidence of moderate “substitution” of federal for institutional graduate student support and evidence that the degree of substitutability differs across fields.

Modeling and Empirical Implementation There are a couple of issues worth noting concerning the connection between the authors’ theoretical discussion and their empirical implementation. Dollars versus Bodies Both the general public finance problem of fungibility from which this paper takes off and indeed its own theoretical discussion run in terms of dollars: How does external funding for one activity influence the institution’s own funding for that activity and its funding for other activities? The data, however, apparently compel the authors to do their empirical work in terms of “bodies”-numbers of students-rather than dollars. The fact that some sorts of support are probably more costly than others makes for some difficulties in translating the authors’ findings into conclusions about dollar substitutability. Suppose, to illustrate, that a student who would have gotten a $15,000 university fellowship gets a federal fellowship of equal value instead. The university might take $lO,OOO of the $15,000 it would have spent on aiding this student and use it for some other purpose and then use the other $5,000 in

208

R. G. Ehrenberg, D. I. Rees, andD. J. Brewer

support to another student who would otherwise have gotten nothing. In terms of body counts, the federal money shows no substitutability-the same number of students get institutional support whether or not the federal support is provided. But in dollar terms, two-thirds of the federal money is substituted out of student aid into support of other activities. It is interesting in this context that when the authors examine the impact of different types of external support on different types of internal support (section 6.5), some interesting patterns emerge. In particular, more federal research assistantships lead to fewer institutionally funded research assistantships and more institutionally financed teaching assistantships. Although we lack data on the cost to the institution of these different types of support, the results seem likely to be consistent with the story I have told here. Given the available data, there is not much the authors can do to translate their bodies results into dollars results, but there is need for caution in interpreting the results. Dynamics of University Behavior A second, very interesting, problem concerns the authors’ use of a partial adjustment model to examine the dynamics of university behavior. Reflection on this modeling strategy-which seems a natural one to try-is worthwhile, since it may shed light on some of the underlying mechanisms at work within the university. The partial adjustment model seems to make sense with regard to adjustments of institutional support for students to factors such as number of undergraduate degrees granted. If your undergraduate population rises, you will increase the number of graduate students on university-supported teaching assistantships, with a lag. But does it make as much sense for variation in external funding? The university gets a new stream of federal funding for students. The partial adjustment model says that the university continues to support students out of institutional funds for awhile, presumably using the new stream of federal funding to support added students, and then gradually reduces the number of added students, instead switching students over to federal funding. This gives more substitution in the long run than in the short run, as the partial adjustment model supposes. This could happen, but a different kind of story seems more likely. The university gets a new stream of federal funding and instantly reshuffles its budget to finance more existing students from federal resources, either saving the institutional money or devoting it to other purposes. Then, if the federal support continues, the university uses that new funding base to expand its total graduate student operation, expanding the number of students and increasing the number on institutional funds. Here there is more substitution in the short run than in the long run. I’ve never been a dean (and never hope to be one), but it seems to me this is what I would try to do if more federal student support funds became available to my budget. As the authors note in their conclusion, following up on discussion at the

209

How Would Universities Respond to Increased Federal Support?

conference, the nature of the university’s response here will depend on whether changes in federal support are viewed by the institution as permanent or transitory. The story as I have told it fits the case of a permanent increase. A transitory increase would presumably lead to a lot of short-run substitution as the university tries to “save” its windfall gain by using federal money to substitute for institutional money.* The larger point, though, is that if the dynamics of university funding decisions follow the pattern described here, the partial adjustment model used in this paper may not be a very good way to capture what is going on. What seems to be true is that the university is making (roughly) two kinds of decisions: one about how “big” to be, in terms of the total number of graduate students the institution wants to have around, and another about how to finance them. The first decision, one suspects, could be handled well by some kind of partial adjustment model into which changes in federal support would enter. The second decision, about how to finance existing students, seems more likely to depend on contemporaneous variables. Both aspects of the model would need to be sensitive to the issue of distinguishing responses to permanent and transitory changes. Elaborating models that can capture more of these complexities is an important agenda for future work. Notice that it is inducing schools to expand the total size of their programs-or, more precisely, to expand their “output” in terms of completed degrees-that is the policy goal of federal support. Thus, looking at the relation between (permanent) variations in federal student support and total enrollments or (lagged) degree production might be an interesting approach in future work.

Directions for Future Research As these remarks perhaps suggest, I would strongly endorse the authors’ proposal that future work attempt to connect these kinds of empirical investigations with more explicit and complete models of university behavior. These don’t necessarily have to be optimizing models, in which the university maximizes some objective function subject to constraints. There is, I think, a great deal of room for models that view university decisions as the outcome of contending political forces or as the result of behavioral rules that need not necessarily be derived from a single objective function. There are at least two good reasons for striving to embed these kinds of empirical investigations in a more comprehensive picture of university functioning. One is the familiar problem of getting the equations specified correctly. For example, it is likely that increases in federal research assistant support are correlated with changes in the institution’s success in winning grants. 2. I’m therefore puzzled by the authors’ statement in section 6.6 that “transitory increases , . . are unlikely to lead to large reallocations of institutional funds.” This seems to be contradicted by their next sentence.

210

R. G. Ehrenberg, D. I. Rees, and D. J. Brewer

That sort of success will influence a number of other financial flows that may wind up affecting the institution’s allocation of resources to graduate student aid. It is easy to think of more potential interdependencies than one can plausibly incorporate in an empirical analysis; but better modeling of the university’s behavior may help us keep track of the more important possibilities. The second point is that it really is important to try to keep track of what universities do with resources that are freed up by increases in external support. In work that I have done with Morton Schapiro, we have attempted to trace impacts of changes in research funding on the growth in tuitions, institutional support for student aid, and spending on instruction. If external money is fungible, where it gets “funged” to is an important question for empirical investigation and for public policy consideration. To do the empirical work, one needs a more comprehensive picture of the choices the institution faces about where to devote its resources. References Breneman, David. 1976. The Ph.D. production process. In Higher education as an industry, ed. Joseph Froomkin, Dean Jamison, and Roy Radner. Cambridge, Mass.: Ballinger. Hoenack, Stephen A., and Daniel J. Pierro. 1990. An econometric model of a public university’s income and enrollments. Journal of Economic Behavior and Organization 14:403-23. McPherson, Michael, and Morton Owen Schapiro. 1991. Keeping college affordable: Government and educational opportunity (Washington, D.C.: Brookings Institution).

7

Optimal Investment Strategies for University Endowment Funds Robert C. Merton

7.1 Introduction

To examine the question of optimal investment strategies for university endowment funds, one must of course address the issue of the objective function by which optimality is to be measured. My impression is that practicing money managers essentially sidestep the issue by focusing on generically efficient risk-return objective functions for investment which are just as applicable to individuals or nonacademic institutions as they are to universities. Perhaps the most common objective of this type is mean-variance efficiency for the portfolio’s allocations. Black (1976) provides a deeper approach along those lines that takes account of tax and other institutional factors, including certain types of nonendowment assets held by institutions. The Ford Foundation study of 1969 gave some early practical (if ex post, somewhat untimely) guidance for investment allocations. Much of the academic literature (which is not copious) seems to focus on appropriate spending policy for endowment, taking as given that the objective for endowment is to provide a perpetual level flow of expected real income (cf. Eisner 1974; Litvack, Malkiel, and Quandt 1974; Nichols 1974; Tobin 1974). Ennis and Williamson (1976) present a history of spending patterns by universities and a discussion of various spending rules adopted. They also discuss the interaction between spending and investment policies. Fama and Jensen (1985) discuss the role of nonprofit institutions as part of a general analysis of organizational forms and investment objective functions, but they do not address the functions of endowment in such institutions. In contrast, Hansmann (1990) provides a focused and comprehensive reRobert C. Merton is George Fisher Baker Professor of Business Administration at Harvard University and a research associate of the National Bureau of Economic Research.

211

212

Robert C. Merton

view of the various possible roles for a university’s endowment. Despite the broad coverage of possibilities ranging from tax incentives to promoting intergenerational equity, he is unable to find compelling empirical evidence to support any particular combination of objectives. Indeed, he concludes that “prevailing endowment spending rules seem inconsistent with most of these objectives” (p. 39). Hansmann goes on to assert (pp. 39-40): It appears, however, that surprisingly little thought has been devoted to the purposes for which endowments are maintained and that, as a consequence, their rate of accumulation and the pattern of spending from their income have been managed without much attention to the ultimate objectives of the institutions that hold them. The course taken here to address this question is in the middle range: it does not attempt to specify in detail the objective function for the university, but it does derive optimal investment and expenditure policy for endowment in a context which takes account of overall university objectives and the availability of other sources of revenue besides endowment. In that respect, it follows along lines similar to the discussion in Black (1976, 26-28). In addition, our model takes explicit account of the uncertainties surrounding the costs of university activities. As a result, the analysis reveals another (perhaps somewhat latent) purpose for endowment: namely, hedging against unanticipated changes in those costs. Formal trading rules for implementing this hedging function are derived in sections 7.3 and 7.4. However, the paper neither assesses which costs, as an empirical matter, are more important to hedge nor examines the feasibility of hedging those costs using available traded securities. The interested reader should see Brinkman (1981, 1990), Brovender (1974), Nordhaus (1989), and Snyder (1988), where the various costs of universities are described and modeled, both historically and prospectively. Grinold, Hopkins, and Massy (1978) develop a budget-planning model which also integrates endowment returns with other revenue and expense flows of the university. However, their model differs significantly from the one presented here, perhaps because their focus is on developing policy guidelines for expenditures instead of optimal intertemporal management of endowment. In the section to follow, we describe the basic insights provided by our analysis and discuss in a qualitative fashion the prescriptions for endowment policy. The formal mathematical model for optimal expenditures and investment that supports those prescriptions is developed in sections 7.3 and 7.4. It is based on a standard intertemporal consumption and portfolio-selection model. Hence, the formal structure of the optimal demand functions is already widely studied in the literature. It is the application of this model to the management of university endowment which is new. For analytical simplicity and clarity, the model is formulated in continuous time. However, it is evident from the work of Constantinides (1989), Long (1974), and Merton (1977) that a discrete-time version of the model would produce similar results.

213

Optimal Investment Strategies for University Endowment Funds

7.2 Overview of Basic Insights and Prescriptions for Policy As indicated at the outset, a standard approach to the management of endowment is to treat it as if it were the only asset of the university. A consequence of this approach is that optimal portfolio strategies are focused exclusively on providing an efficient trade-off between risk and expected return. The most commonly used measure of endowment portfolio risk is the variance (or equivalently, standard deviation) of the portfolio’s return. As is well known, the returns on all mean-variance efficient portfolios are perfectly correlated. Thus, a further consequence of treating endowment as the only asset is that the optimal endowment portfolios of different universities should have quite similar risky investment allocations, at least as measured by the correlations of the portfolio returns. Universities, as we all know, do have other assets, both tangible and intangible, many of which are important sources of cash flow. Examples of such sources are gifts, bequests, university business income, and public- and private-sector grants. Taking explicit account of those assets in the determination of the endowment portfolio can cause the optimal composition of that portfolio to deviate significantly from mean-variance efficiency. That is, two universities with similar objectives and endowments of the same size can nevertheless have very different optimal endowment portfolios if their nonendowment sources of cash flow are different. A procedure for selecting the investments for the endowment portfolio that takes account of nonendowment assets includes the following steps: 1. Estimate the market value that each of the cash flow sources would have if it were a traded asset. Also determine the investment risk characteristics that each of those assets would have as a traded asset. 2 . Compute the total wealrh or net worth of the university by adding the capitalized values of all the cash flow sources to the value of the endowment. 3. Determine the optimal portfolio allocation among traded assets, using the university’s total wealth as a base. That is, treat both endowment and cash flow-source assets as if they could be traded. 4. Using the risk characteristics determined in step 1, estimate the “implicit” investment in each traded-asset category that the university has as the result of owning the nonendowment (cash flow-source) assets. Subtract those implicit investment amounts from the optimal portfolio allocations computed in step 3, to determine the optimal “explicit” investment in each traded asset, which is the actual optimal investment allocation for the endowment portfolio. As a simple illustration, consider a university with $400 million in endowment assets and a single nonendowment cash flow source. Suppose that the only traded assets are stocks and cash. Suppose further that the university estimates in step 1 that the capitalized value of the cash flow source is $200 million, with risk characteristics equivalent to holding $100 million in stock

214

Robert C. Merton

and $100 million in cash. Thus, the total wealth of the university in step 2 is (400 + 200 = ) $600 million. Suppose that from standard portfolio-selection techniques, the optimal fractional allocation in step 3 is .6 in stocks and .4 in cash, or $360 million and $240 million, respectively. From the hypothesized risk characteristics in step 1, the university already has an (implicit) investment of $100 million in stocks from its nonendowment cash flow source. Therefore, we have in step 4 that the optimal amount for the endowment portfolio to invest in stocks is $260 million, the difference between the $360 million optimal total investment in stocks and the $100 million implicit part. Similarly, the optimal amount of endowment invested in cash equals (240 100 =) $140 million. The effect on the composition of the optimal endowment portfolio induced by differences in the size of nonendowment assets can be decomposed into two parts: the wealth effect and the substitution effect. To illustrate the wealth effect, consider two universities with identical preference functions and the same size endowments, but one has nonendowment assets and the other does not. If, as is perhaps reasonable to suppose, the preference function common to each exhibits decreasing absolute risk aversion, then the university with the nonendowment assets (and hence larger net worth) will prefer to have a larger total investment in risky assets. So a university with a $400 million endowment as its only asset would be expected to choose a dollar exposure to stocks that is smaller than the $360 million chosen in our simple example by a university with the same size endowment and a nonendowment asset valued at $200 million. Such behavior is consistent with the belief that wealthier universities can “afford’ to take larger risks with their investments. Thus, if the average risk of the nonendowment assets is the same as the risk of the endowment-only university’s portfolio, then the university with those assets will optimally invest more of its endowment in risky assets. The substitution effect on the endowment portfolio is caused by the substitution of nonendowment asset holdings for endowment asset holdings. To illustrate, consider again our simple example of a university with a $400 million endowment and a $200 million nonendowment asset. However, suppose that the risk characteristics of the asset are changed so that it is equivalent to holding $200 million in stocks and no cash. Now, in step 4, the optimal amount for the endowment portfolio to invest in stocks is $160 million, the difference between the $360 million optimal total investment in stocks and the $200 million implicit part represented by the nonendowment asset. The optimal amount of endowment invested in cash rises to (240 - 0 =) $240 million. If instead the risk characteristics of the asset had changed in the other direction to an equivalent holding of $0 in stocks and $200 million in cash, the optimal composition of the endowment portfolio would be (360 - 0 =) $360 million in stocks and (240 - 200 = ) $40 million in cash. Note that the changes in risk characteristics do not change the optimal deployment of total net worth ($360 million in stocks and $240 million in cash).

215

Optimal Investment Strategies for University Endowment Funds

However, the nonendowment assets are not carried in the endowment portfolio. Hence, different risk characteristics for those assets do change the amount of substitution they provide for stocks and cash in the endowment portfolio. Thus, the composition of the endowment portfolio will be affected in both the scale and fractional allocations among assets. With the basic concept of the substitution effect established, we now apply it in some examples to illustrate its implications for endowment investment policy. Consider a university that on a regular basis receives donations from alums. Clearly, the cash flows from future contributions are an asset of the university, albeit an intangible one. Suppose that the actual amount of gift giving is known to be quite sensitive to the performance of the general stock market. That is, when the market does well, gifts are high; when it does poorly, gifts are low. Through this gift-giving process, the university thus has a “shadow” investment in the stock market. Hence, all else the same, it should hold a smaller portion of its endowment in stocks than would another university with smaller amounts of such market-sensitive gift giving. The same principle applies to more specific asset classes. If an important part of gifts to a school that specializes in science and engineering comes from entrepreneur alums, then the school de fact0 has a large investment in venture capital and high-tech companies, and it should therefore invest less of its endowment funds in those areas. Indeed, if a donor is expected to give a large block of a particular stock, then the optimal explicit holding of that stock in the endowment can be negative. Of course, an actual short position may not be truly optimal if such short sales offend the donor. That the school should optimally invest less of its endowment in the science and technology areas where its faculty and students have special expertise may seem a bit paradoxical. But the paradox is resolved by the principle of diversification once the endowment is recognized as representing only a part of the assets of the university. The same analysis and conclusion apply if alum wealth concentrations are in a different class of assets, such as real estate instead of shares of stock. Moreover, much the same story also applies if we were to change the example by substituting government and corporate grants for donations and gift giving as the sources of cash flows. That is, the magnitudes of such grant support for engineering and applied science may well be positively correlated with the financial performance of companies in high-tech industries. If so, then the prospect of future cash flows to the university from the grants creates a shadow investment in those companies. The focus of our analysis is on optimal asset allocation for the endowment portfolio. However, the nature and size of a university’s nonendowment assets significantly influence optimal policy for spending endowment. As shown in section 7.4, for a given overall expenditure rate as a fraction of the university’s total net worth, the optimal spending rate out of endowment will vary, depending on the fraction of net worth represented by nonendowment assets, the

216

Robert C. Merton

expected growth rate of cash flows generated by those assets, and capitalization rates. Hence, neglecting those other assets will generally bias the optimal expenditure policy for endowment. In addition to taking account of nonendowment assets, our analysis differs from the norm because it takes account of the uncertainty surrounding the costs of the various activities such as education, research, and knowledge storage that define the purpose of the university. The breakdown of activities can of course be considerably more refined. For instance, one activity could be the education of a full-tuition-paying undergraduate, and a second could be the education of an undergraduate who receives financial aid. The unit (net) cost of the former is the unit cost of providing the education less the tuition received, and the unit cost of the latter is this cost plus the financial aid given. As formally demonstrated in section 7.3, an important function of endowment investments is to hedge against unanticipated changes in the costs of university activities. Consider, for example, the decision as to how much (if any) of the university’s endowment to invest in local residential real estate. From a standard mean-variance efficiency analysis, it is unlikely that any material portion of the endowment should be invested in this asset class. However, consider the cost structure faced by the university for providing teaching and research. Perhaps the single largest component is faculty salaries. Universities of the same type and quality compete for faculty from the same pools. To be competitive, they must offer a similar standard of living. Probably the largest part of the differences among universities in the cost of providing this same standard of living is local housing costs. The university that invests in local residential housing hedges itself against this future cost uncertainty by acquiring an asset whose value is higher than expected when the differential cost of faculty salaries is higher than expected. This same asset may also provide a hedge against unanticipated higher costs of off-campus housing for students that would in turn require more financial aid if the university is to compete for the best students. Note that this prescription of targeted investment in very specific real estate assets to hedge against an unanticipated rise in a particular university’s costs of faculty salaries and student aid should not be confused with the often-stated (but empirically questionable) assertion that investments in real estate generally are a good hedge against inflation. See Bodie (1976, 1982) for empirical analysis of the optimal assets for hedging against general inflation. Similar arguments could be used to justify targeted investment of endowment in various commodities such as oil as hedges against unanticipated changes in energy costs. Uncertainty about those costs is especially significant for universities located in extreme climates and for universities with major laboratories and medical facilities that consume large quantities of energy. The hedging role for endowment can cause optimal investment positions that are in the opposite direction from the position dictated by the substitution

217

Optimal Investment Strategies for University Endowment Funds

effects of nonendowment assets. For example, consider a specialized institute of biology that receives grants from biotech companies and gifts from financially successful alums. As already explained, such an institute has a large shadow investment in biotech stocks, and it should therefore underweight (perhaps to zero) its endowment investments in such stocks. Suppose, however, that the institute believes the cost of keeping top faculty will rise by considerably more than tuition or grants in the event that there is a strong demand for such scientists outside academe. Then it may be optimal to invest a portion of its endowment in biotech stocks to hedge this cost, even though those stocks’ returns are highly correlated with alum gifts and industry grants. As demonstrated in section 7.3, the hedging role for endowment derived here is formally valid as long as there are traded securities with returns that have nonzero correlations with unanticipated changes in the activity costs. However, the practical significance for this role turns on the magnitude of the correlations. As illustrated in Bodie’s (1976, 1982) work on hedging against inflation, it is often difficult to construct portfolios (using only standard types of traded securities) that are highly correlated with changes in the prices of specific goods and services. Nevertheless, the enormous strides in financial engineering over the last decade have greatly expanded the opportunities for custom financial contracting at reasonable costs. As we move into the twentyfirst century, it will become increasingly more common for the financial services industry to offer its customers private contracts or securities that allow efficient hedging when the return properties of publicly traded securities are inadequate. That is, implementation of the quantitative strategies prescribed in sections 7.3 and 7.4 will become increasingly more practical for universities and other endowment institutions. See Merton (1990b, chap. 14; 1990c, 264-69) for a prospective view on financial innovation and the development of custom financial contracting. There are of course a variety of issues involving endowment management that have not been addressed but could be within the context of our model. One such issue is the decision whether to invest endowment in specificpurpose real assets such as dormitories and laboratories instead of financial (or general-purpose physical) assets. The returns on those real assets are likely to be strongly correlated with the costs of particular university activities, and thereby the assets form a good hedge against unexpected rises in those costs. However, because the real-asset investments are specialized and largely irreversible, shifting the asset mix toward such investments reduces flexibility for the university. That is, with financial assets, the university has more options as to what it can do in the future. In future research, I plan to analyze this choice problem more formally by using contingent-claims analysis to value the trade-off between greater flexibility in selecting future activities and lower costs in producing a given set of activities. Another issue not explicitly examined is the impact long-term, fixed liabilities such as faculty tenure contracts have on the management of endowment.

218

Robert C. Merton

Our formal model of sections 7.3 and 7.4 that uses contingent-claims analysis (CCA) can handle this extension. See McDonald (1974) and Merton (1985) for CCA-type models for valuing tenure and other wage guarantee contracts. In summary, the paper explores two classes of reasons why optimal endowment investment policy and expenditure policy can vary significantly among universities. The analysis suggests that trustees and others who judge the prudence and performance of policies by comparisons across institutions should take account of differences in both the mix of activities of the institutions and the capitalized values of their nonendowment sources of cash flows. The overview completed, we now turn to the development of the mathematical model for the process and the derivation of the quantitative rules for implementation.

7.3 The Model The functions or purposes of the university are assumed to be a collection of activities or outputs such as education, training, research, and storage of knowledge. We further assume that the intensities of those activities can be quantified and that a preference ordering exists for ranking alternative intertemporal programs. In particular, the criterion function for this ranking can be written as m

where Q,(t) denotes the quantity of activity j per unit time undertaken at time t , j = 1, . . . ,m;the preference function U is assumed to be strictly concave in ( Q , , . . . ,QJ and E, denotes the expectation operator, conditional on knowing all relevant information as of time t. This preference ordering satisfies the classic von Neumann-Morgenstern axioms of choice, exhibits positive risk aversion, and includes survival (of the institution) as a possible objective. The infinite time horizon structure in (1) implies only that there need not be a definite date when the university will liquidate. As shown in Merton (1990b, 149-51, 609-ll), U can reflect the mortality characteristics of an uncertain liquidation date. The intertemporally additive and independent preference structure in ( 1 ) can be generalized to include nonadditivity, habit formation, and other pathdependent effects on preferences, along the lines of Bergman (1985), Constantinides (1990), Detemple and Zapatero (1989), Duffie and Epstein (1992), Hindy and Huang (1992), Sundaresan (1989), and Svensson (1989). However, as shown in Merton (1990b, 207-9), those more realistic preference functions do not materially affect the optimal portfolio demand functions. Moreover, just as Grossman and Laroque (1990) show for transactions costs in consumption, so it can be shown here that imposing adjustment costs for changing the levels of university activities does not alter the structure of the

219

Optimal Investment Strategies for University Endowment Funds

portfolio demand functions. Hence, because the focus of the paper is on optimal investment (rather than on optimal expenditure) strategies, we assume no adjustment costs for activities and retain the additive independent preference specification to provide analytical simplicity, Let S,(t) denote the (net) cost to the university of providing one unit of activity j at time t , j = 1, . . . ,m.For example, i f j = 1 denotes the activity of having full-tuition-paying undergraduates, then S,would be the unit cost of providing the education minus the tuition received. If j = 2 denotes the activity of having undergraduates who receive financial aid, the unit cost S, would equal S,plus the financial aid given. In general, all costs and receipts such as tuition that are directly linked to the quantities of specific activities undertaken are put into the activity costs or prices, (S,). As will be described, fixed costs and sources of positive cash flows to the university that do not depend directly on the activity quantities are handled separately. As in Merton (1990b, 202, 499), we assume that the dynamics for these costs are described by the stochastic differential equations: for S = (S,,. . . ,S,,,),

+ g,(S,t)S,dq,

. . . ,m , where8 is the instantaneous expected rate of growth in S,, g, is the instanta(2)

dS,

= &(S,t)S,dr

, j = 1,

neous standard deviation of the growth rate, and dq, is a Wiener process with the instantaneous correlation coefficient between dq, and dq, given by u,,, i, j = 1, . . . ,m.f, and g, are such that dS, 2 0 for S, = 0, which ensures that S,(t) 2 0. Especially since (S) has components that depend on tuition, financial aid, and other variables over which the university has some control, one would expect that the dynamic path for those costs would be at least partially endogenous and controllable by the university, even though competition among universities would limit the degree of controllability. However, as specified, (2) is an exogenous process, not controlled by the university. Alternatively, it can be viewed as the “reduced-form” process for S after optimization over nonportfolio choice variables. The university is assumed to have N nonendowment sources of cash flows, which we denote by Y,(t)dt for the kth source at time r. As noted in section 7.2, examples of such sources are gifts, bequests, university business income, and public- and private-sector grants. It can also be used to capture transfer pricing for the use of buildings and other university-specific assets where Y, is the rental rate and this rental fee appears as an offsetting charge in the (S,) for the appropriate university activities. The dynamics for these cash flows are modeled by, for Y = ( Y , , . . . , Y J , (3) where pk and 6, depend at most on the current levels of the cash flows and the unit costs of university activities and de, is a Wiener process, k = 1, . . . ,N. Equation (3) can also be used to take account of fixed costs or liabilities of the university such as faculty tenure commitments, by letting Yk < 0 to reflect a

220

Robert C. Merton

cash outflow. However, the focus here is on assets only, and therefore we assume that p, and 6, are such that dY, ? 0 for Y, = 0, which implies that Y k ( f ) 2 0 for all t . By inspection of (2) and (3), the dynamics for (Y,S) are jointly Markov. A more realistic model would have p k and 6, depend on both current and historical values of Q ,, . . . ,Qm.For example, if a university has undertaken large amounts of research activities in the past, it may attract more grants and gifts in the future. The university may also affect the future expected cash flows from nonendowment sources by investing now in building up those sources. Thus, the dynamic process for Y should be in part controllable by the university. However, again for analytical simplicity, the Y process is taken as exogenous, because that abstraction does not significantly alter the optimal portfolio demand functions. If for k = 1, . . . , N , Vk(t)denotes the capitalized value at time t of the stream of future cash flows, Yk(7)for T 2 t , and if K(t) denotes the value of the endowment at time t , then the net worth or wealth of the university, W ( t ) is given by n

(4)

W ( t ) = K(t)

+ C V k ( t ). 1

A model for determining the Vk(t)from the posited cash flow dynamics in (3) is developed in section 7.4. The endowment of the university is assumed to be invested in traded assets. There are n risky assets and a riskless asset. If P,(t) denotes the price of thejth risky asset at time t , then the return dynamics for the risky assets are given by, f o r j = 1 , . . . ,n, (5)

dP, = aJP,dt

+ oJPJdZJ,

where a,is the instantaneous expected return on assetj; uJis the instantaneous standard deviation of the return; and dZ, is a Wiener process. The instantaneous correlation coefficients (p,,,qk,,{,,) are defined by, f o r j = 1, . . . ,n, (5a)

dZ,dZ, = ~ , ~ ,d it = 1, . . . ,n dqkdZJ = qkJdt, k = 1, . . . ,m de& = &dt , 1 = 1, . . . , N .

For computational simplicity and to better isolate the special characteristics of endowment management from general portfolio management, we simplify the return dynamics specification and assume that (aJ,uJ,p,) are constants over time, i, j = 1, . . . ,n. As shown in Merton (1990b, chaps. 4, 5, 6), this assumption of a constant investment opportunity set implies that [P,(t + ~ ) / P , ( t )jl , = 1, . . . ,n,for T > 0 are jointly lognormally distributed. The riskless asset earns the interest rate r , which is also constant over time. Optimal portfolio selection for general return dynamics would follow along the lines of Merton (1990a, sec. 7; 1990b, chaps. 5, 15, 16).

221

Optimal Investment Strategies for University Endowment Funds

To analyze the optimal intertemporal expenditure and portfolio-selection problem for the university, we begin with a further simplified version of the model in which the university's entire net worth is endowment (i.e., Yk[t] = V,(t) = 0, k = 1, . . . ,N and W ( t ) = K [ t ] ) .The budget equation dynamics for W ( t )are then given by

where wj(t) = the fraction of the university's wealth allocated to risky asset j at time t , j = 1, . . . ,n;the fraction allocated to the riskless asset is thus 1 2 w,. Trustees, donors, and the government are assumed not to impose explicit limitations on investment policy for the endowment, other than general considerations of prudence. In particular, borrowing and short selling are permitted, so the choice for (w,) is unrestricted. We further posit that spending out of endowment is not restricted, either with respect to overall expenditure or with respect to the specific activities on which it is spent. However, we do impose the feasibility restrictions that total expenditure at time t , E;, QJ,, must be nonnegative and zero wealth is an absorbing state (i.e., W[t] = 0 implies W [ t + T ] = 0 for 7 > 0). At each time t , the university chooses a quantity of activities (Q, , . . . ,Q,) and a portfolio allocation of its wealth so as to maximize lifetime utility of the university as specified in (1). Just as for the case of multiple consumption goods analyzed in Breeden ( 1979), Fischer ( 1975), and Merton ( 1990b, 205), so the solution for the optimal program here can be decomposed into two parts. First, at each t , solve for the utility-maximizing quantities of individual activities, (Q,, . . . ,ern),subject to an overall expenditure constraint, C ( t ) = C;, Q,(t)S,(t). Second, solve for the optimal level of overall expenditures at time t and the optimal portfolio allocation of endowment. The first part is essentially the static activity-choice problem with no uncertainty

subject to C(t) = Z;, Q$k(t). The first-order conditions for the optimal activity bundle (Q;, . . . ,Q;) are given by, forS,(t) = s&,

with C(r) = C;lQ;S,, where subscripts on U denote partial derivatives (i.e., 0, = dU//aQ,). It follows from (8) that the optimal quantities can be written as Q; = Q;[C(t),S(t),t], k = 1, . . . ,m . Define the indirect utility function U by U[C(t),S(t),t]= U(Qr, . . . ,em,t). By substituting U for U , we can rewrite (1) as

222

Robert C. Merton m

max E o [ l U[C(r),S(t),rIdt} ,

(9)

where the “max” in (9) is over the intertemporal expenditure path [C(t)]and portfolio allocations [wj(t)].Thus, the original optimization problem is transformed into a single-expenditure choice problem with “state-dependent’’ utility (where the “states” are the relative costs or prices of the various activities). Once the optimal total expenditure rules, [C*(t)],are determined, the optimal expenditures on individual activities are determined by (8) with C*(t) = cyQ;Sk.

The solution of (9) follows by applying stochastic dynamic programming as in Merton (1990b, chaps. 4, 5 , 6 ) . Define the Bellman, or derived-utility, function J by m

conditional on W(t) = W and S(t) 181, 202), J will satisfy

=

S. From Merton (1990a, 555; 1990b,

(10)

WiWJUijW m

n

subject to J(O,S,t) = J; u(0,. . . ,O,T)&, where subscripts on J denote partial derivatives with respect to W , t , and Si, i = 1, . . . ,mand uo = pijuiuj, the instantaneous covariance between the return on security i andj. A is a Kuhn-Tucker multiplier reflecting the nonnegativity constraint on C, and at the optimum it will satisfy A *C* = 0. The first-order conditions derived from (10) are (1la)

0 = U,(C*,S,t)

+ A*

- J,(W,S,t)

and

0

=

J , ( y - r)

+ J,C

” W~WU,~ I

+ C Jkwg$kuiqki,i =

1,

.

. . ,n ,

1

where C* = C*(W,S,t) and w: = w*(W,S,t) are the optimal expenditure, and portfolio rules expressed as functions of the state variables and subscripts on U denote partial derivatives.

223

Optimal Investment Strategies for University Endowment Funds

From ( 11a), the optimal expenditure rule is given by (12)

UC(C*,S,f)= J,(W,S,t) for C* > 0 x* = max [0,J,(W,S,t) - U,(O,S,t)]

From ( 1 lb), the optimal portfolio allocation can be written as m

(13)

w,*W = Ab,

+ 2 H , h , , i = 1, . . . ,n , I

where 6, = C; vI,(a, - I ) ; h, = C; u$J,q,,v,,; v,, is the i j element of the inverse of the instantaneous variance-covariance matrix of returns (u,,);A = - J J J , (the reciprocal of absolute risk aversion of the derived-utility function); and H , = - J,JJ,,, k = 1, . . . ,m. A and H , depend on the individual university’s intertemporal preferences for expenditures and its current net worth. However, 6, and h,, are determined entirely by the dynamic structures for the asset price returns and the unit costs of the various activities undertaken by universities. Hence, those parameters are the same for all universities, independent of their preferences or endowment size. To provide some economic intuition about the optimal allocation of endowment in (13), consider as a frame of reference the “standard” intertemporal portfolio-selection problem with state-independent utility, U = U(C(t),t).As shown in Merton (1990b, 131-36), given the posited return dynamics in ( 5 ) , all such investors will hold instantaneously mean-variance efficient portfolios as their optimal portfolios. For aUlaSk = Uk = 0, H, = 0 , k = 1, . . . ,m. Hence, in this case, (13) becomes w,”W = Ab,, and w,“Wlw,”W = b,/b,, the same for all investors. This is the well-known result that the relative holdings of risky assets are the same for all mean-variance efficient portfolios. However, the state-dependent preferences for universities induced by the uncertainty surrounding the relative costs of undertaking different desired activities causes the more complex demand structure in ( 13). To better understand this differential demand, w,*W - Ab, = Z;l HJl,,, it is useful to examine the special case where for each cost S, there exists an asset whose instantaneous return is perfectly correlated with changes in S,. By renumbering securities if necessary, choose the convention that q, = 1 in (5a), k = 1, . , . ,m(m< n ) . As shown in Merton (1990b, 203-4), it follows that in this case, h, = gJ@, for k = 1, . . . ,m and h , = 0 for k # j . Hence, we can rewrite (13) as wTW = Abi

gi + Hi -i = si

1,

...

mi

= Ab,

i = m + 1, . . . ,n .

By the strict concavity of U with respect to C , J is strictly concave in W. Hence, J,, < 0 and Hi = - J,JJ,, is positively proportional to JiW. Thus,

224

Robert C. Merton

relative to a “normal” investor with state-independent preferences (i.e., H , = 0, i = 1, . . . ,m) but the same current level of absolute risk aversion (i.e., - J d J , ) , the university will optimally hold more of asset i if J , , > 0 and less if J,, < 0, i = 1, . . . ,m. If J,, > 0, then at least locally the university’s marginal utility (or “need”) for wealth or endowment becomes larger if the cost of undertaking activity i increases, and it becomes smaller if this cost decreases. Because the return on asset i is perfectly positively correlated with the cost of activity i, a greater than expected increase in S, will coincide with a greater than expected return on asset i. By holding more of asset i than a “normal” investor would, the university thus assures itself of a relatively larger endowment in the event that S, increases and the need for wealth becomes more important. The university, of course, pays for this by accepting a relatively smaller endowment in the event that S, decreases and wealth is less important. The behavioral description for J,, < 0 is just the reverse, because the need for endowment decreases if the cost of activity i increases. To perhaps help in developing further insights, we use (12) to interpret the differential demand component in (14) in terms of the indirect utility and optimal expenditure functions. By differentiating (12), we have that, for C*(W,S,t) > 0 ,

J,,

= U,,(C*,S,t)

Jkw =

ucc(C*,S,t)

ac* ~

aw

ac* ~

+ U,-,(C*,S,t)

ask

- ac*

for k = 1, . . . ,m.Because U,, < 0 and aC*/aW > 0 for C* > 0, we see that the sign of Hkis determined by the impact of a change in the cost of activity k on two items: the optimal level of total current expenditure and the marginal utility of expenditure. So, for example, if an increase in S, would cause both a decrease in optimal expenditure (aC*/aS, < 0) and an increase in the marginal utility of expenditure (U,, > 0), then, from (15), Hk> 0 and the university will optimally hold more of asset k than the corresponding investor with a mean-variance efficient portfolio. Following (14) causes the university’s optimal portfolio to be meanvariance inefficient, and therefore the return on the endowment will have

225

Optimal Investment Strategies for University Endowment Funds

greater volatility than other feasible portfolios with the same expected return. However, the value of the endowment or net worth of the university is not the “end” objective. Instead, it is the “means” by which the ends of a preferred expenditure policy can be implemented. Viewed in terms of the volatility of the time path of expenditure (or more precisely, the marginal utility of expenditure), the optimal strategy given in (14) is mean-variance efficient (cf. Breeden 1979; Merton 1990b, 487-88). That is, because aC*laW > 0, the additional increment in wealth that, by portfolio construction, occurs precisely when S, increases will tend to offset the negative impact on C* caused by that increase. There is thus a dampening of the unanticipated fluctuations in expenditure over time. In sum, we see that in addition to investing in assets to achieve an efficient risk-return trade-off in wealth, universities should optimally use their endowment to hedge against unanticipated and unfavorable changes in the costs of the various activities that enter into their direct utility functions. In closing this section, we note that the interpretation of the demand functions in the general case of (13) follows along the same lines as for the special case of perfect correlation leading to (14). As shown for the general case in Merton (1990a, 558-59; 1990b, 501-2), the differential demands for assets reflect attempts to create portfolios with the maximal feasible correlations between their returns and unanticipated changes in the S,, k = 1, . . . ,111. These maximally correlated portfolios perform the same hedging function as assets 1, . . . ,m in the limiting case of perfect correlation analyzed in (14). Furthermore, if other state variables besides the various activities’ costs (e.g., changes in the investment opportunity set) enter a university’s derived utility function, then a similar structure of differential asset demands to hedge against the unanticipated changes in these variables will also obtain. 7.4 Optimal Endowment Management with Other Sources of Income

In the previous section, we identified hedging of the costs of university activities as a reason for optimally deviating from “efficient” portfolio allocations when endowment is the only means for financing those activities. In this section, we extend the analysis to allow other sources of cash flow to support the activities. To simplify the analysis, we make two additional assumptions. First, we posit that pk and 6, in (3) are constants, which implies that Yk(t)lYk(0) is lognormally distributed, k = 1, . . . , N . Second, we assume that for each k there exists a traded security whose return is instantaneously perfectly correlated with the unanticipated change in Y,, k = 1, . . . ,N. By renumbering if necessary, we use the convention that traded security k is instantaneously perfectly correlated with Y,. Hence, it follows that ,{ = 1 in (5a) and (16)

de,

=

dZ,,

k = 1,. . . , N .

226

Robert C. Merton

These two assumptions permit us to derive a closed-form solution for the capitalized values of the cash flows, [V,(t)],using contingent-claims analysis. As will be shown, those valuation functions are independent of the university’s preferences or wealth level. From (3), (3,and (16) with p, and 6, constant, we have that the cash flows can be written as a function of the traded asset prices as follows, for k = 1, . . . ,N, (17)

‘k(‘)

= Yk(o)exp(- $ k f )

[pk(r)lpk(o)lPk

7

+,

where = p, (a,- a:/2) - (p, - 6:/2) and p, = 6Ja,. That (17) obtains can be checked by applying Id’s Lemma. We now derive the capitalized value for Y,, following Merton (1990a, 562-63; 1990b, 415-19). Let Fk (P,, t ) be the solution to the partial differential equation, for 0 5 t IT,, 0 = 112 a F i F : ,

(18)

+ rP,Fi

- rFk

+ F: + Y,

subject to the boundary conditions: (1 9 4 ( 19b)

Fk(0,t) = 0 Fk/(P,)Pk bounded as P ,

(19c)

Fk(pk,Tk) = 0

4m

9

where subscripts on Fk in (18) denote partial derivatives with respect to its arguments P , and r; Y, is given by (17); and T, is the last date at which the university receives the cash flows from source k , k = 1, . . . , N . It is a mathematical result that a solution exists to (18)-( 19) and that it is unique. Moreover, for Y, 2 0, Fk 2 0 for all P, and t . Consider a dynamic portfolio strategy in which F:[P,(t),t]P,(t)is allocated to traded asset k at time rand V(r) - F,k[P,(t),t]P,(r)is allocated to the riskless asset, where V(t) is the value of the portfolio at time t . Furthermore, let the portfolio distribute cash (by selling securities if necessary) according to the flow-rate rule (20)

D,(Pk,t)

= yk(t)

as given by (17). Then the dynamics of the portfolio can be written as, for P,(t) = P, and V(t) = V , (21)

dV

= Ff(P,,t)dP,

+ { [ V - Ff(P,,t)P,]r - D,(P,,t)}dt .

Since Fk satisfies (18), it is a twice continuously differentiable function and therefore, by ItB’s Lemma, we can write the dynamics for Fk as (22)

dFk

=

112 a: Pi F:,

+ F: dt + FfdP, .

But Fk satisfies (18) and hence, 1/2 a: Pi Ff, Substituting into (22), we can rewrite it as

+ Ft

= rFk - rP,F:

- Y,.

227

(23)

Optimal Investment Strategies for University Endowment Funds dFk = Ft dP,

+ (rFk - rP$t

- Y,)dt

From (21) and (23), we have that (24)

dV - dFk =(rV - rP& - D, - rFk + rP,F:

+ Y,)dt

= r(V - Fk)dt

because D, = Y,. By inspection, (24) is an ordinary differential equation with solution (25)

V(t) - Fk[Pk(t),tl=

{v(o) - ~

k [ ~ ~ ( ~ ) ~ .~ l ) e ~ p ( ~ ~ )

Thus, if the initial investment in the portfolio is chosen so that V(0) = F,[P,(O),O],then for all t and P,(t), we have that (26)

V(t) = Fk[P,(t),t].

To ensure that the proposed portfolio strategy is feasible, we must show that its value is always nonnegative for every possible sample path for the price P , and all f , 0 5 t I T,. Because Fk is the solution to (18) and Y, 2 0, Fk 2 0 for all P, and t . It follows from (26) that V(t) 2 0 for all P , and t. We have therefore constructed a feasible dynamic portfolio strategy in traded asset k and the riskless asset that produces the stream of cash flows Y,(t)dt for 0 5 t 5 T, and has zero residual value [V(T,) = 01 at Tk. Because the derived strategy exactly replicates the stream of cash flows generated by source k , it is economically equivalent to owning the cash flows Y,(t) for t 5 T,. It follows that the capitalized value of these cash flows satisfies (27)

vk(t)

=

Fk[P,(t),tl

for k = 1, . . . ,N. Note that by inspection of (18)-(19), Fk, and hence V,(t), does not depend on either the university’s preferences or its net worth. The valuation for source k is thus the same for all universities. Armed with (27), we now turn to the optimal policy for managing endowment when the university has N nonendowment sources of cash flows. The procedure is the one outlined in section 7.2. To derive the optimal policy, note first that even if those nonendowment sources cannot actually be sold by the university for legal, ethical, moral hazard, or asymmetric information reasons, the university can achieve the economic equivalent of a sale by following the “mirror image,” or reverse, of the replicating strategy. That is, by (short selling or) taking a -Flk[Pk(t),r]Pk position in asset k and borrowing (Fk - F,,P,) of the riskless asset at each t , the portfolio will generate a positive amount of cash, Fk(P,,t), available for investment in other assets at time t. The entire liability generated by shorting this portfolio is exactly the negative cash flows, ( - Y,dt), for t 5 T,, because V,(T,) = Fk(P,, T,) = 0. But, since the university receives Y,dt for t 5 T, from source k , this short-portfolio liability is entirely offset. Hence, to undertake this strategy beginning at time

228

Robert C. Merton

t is the economic equivalent of selling cash flow source k for a price of Vk(t) = F';[P,(t),t].

As discussed more generally in Merton (1990b, sec. 14.5, esp. 465-67), the optimal portfolio strategy will be as if all N nonendowment assets were sold and the proceeds, together with endowment, invested in the n risky traded assets and the riskless asset. This result obtains because it is feasible to sell (in the economic sense) the nonendowment assets and because all the economic benefits from those assets can be replicated by dynamic trading strategies in the traded assets. Hence, there is neither an economic advantage nor a disadvantage to retaining the nonendowment assets. It follows that the optimal demand for the traded risky assets is given by (13) and the demand for the riskless asset is given by (1 - Z;w,*)W(t),where from (4) and (27)

Because, however, the university has not actually sold the nonendowment assets, the optimal demands given by (13) and (28) include both implicit and explicit holdings of the traded assets. That is, the university's ownership of nonendowment cash flow source k at time t is equivalent to having an additional net worth of Fk[P,(t),t], as reflected in (28), and to having F~[P,(t),t]P,(t) invested in traded asset k and {Fk[P,(t),t] - F;[P,(t),t]P,(t)} invested in the riskless asset. Thus, ownership of source k causes implicit investments in traded asset k and the riskless asset. Optimal explicit investment in each traded asset is the position actually observed in the endowment portfolio, and it is equal to the optimal demand given by (13) and (28) minus the implicit investment in that asset resulting from ownership of nonendowment assets. Let D*(t) denote the optimal explicit investment in traded asset i by the university at time t . It follows from (13) that m

D*(t) = Ab,

(29)

+ C H,hki - F;[P,(t),tIP,(t) , i = 1, . . . ,N I

"

= Ab,

+ 2 Hkhki

, i = N + 1, . . . ,n ,

I

where W ( t )used in the evaluation of A and H,is given by (28). If we number the riskless asset by "tz 1 ," then explicit investment in the riskless asset can be written as

+

n

(30)

D*,+,(t) = [1 - C W ~ ( t ) l W ( f-) 1

N

C{m,(t),tI- F;[P,(r),tIP,
By inspection of (29), it is apparent that, in addition to the hedging of activity costs, the existence of nonendowment sources of cash flow will cause further

229

Optimal Investment Strategies for University Endowment Funds

differences between the observed holdings of assets in the optimal endowment portfolio and the mean-variance efficient portfolio of a "standard" investor. Similarly, from (30), the observed mix between risky assets and the riskless asset will differ from the true economic mix. To explore further the effects of those nonendowment sources of cash flows, we solve the optimal expenditure and portfolio-selection problem for a specific utility function, U . However, in preparation for that analysis, we first derive explicit formulas for the capitalized values of those sources when Y,(t) is given by (17). As already noted, there exists a unique solution to (18) and (19). Hence, it is sufficient to simply find a solution. As can be verified by direct substitution into (18), the value of cash flow source k is given by, for k = 1 , . . . ,N,

where P,,

4, are as defined in (1 7) and 0, = r

(3 1 4

It follows from (31) that, fork (32)

+ &(a, - r) - p., =

.

1, . . . ,N,

F:[P,(t),tlPk(t) = ppk[p,(t)3t1

7

which implies that the capitalized value of source k has a constant elasticity with respect to the price of traded asset k . Equation (32) also implies that the replicating portfolio strategy is a constant-proportion or rebalancing strategy which allocates fraction p, of the portfolio to traded asset k and fraction (1 p,) to the riskless asset. In the case when positive fractions are allocated to both assets (i.e., [ l - p,] > 0 and p, > 0), then P is a strictly concave function of P,. If p, > 1 , then Fk is a strictly convex function of P,, and the replicating portfolio holds traded asset k leveraged by borrowing. In the watershed case of p, = 1 , Fk is a linear function of P,, and the replicating portfolio holds traded asset k only. Using (17) and (27), we can rewrite (31) to express the capitalized value of source k in terms of the current cash flow it generates:

(33)

V,(t) = Y,(t)

1 - exp[ - 0,(T, - t)]

, k = 1 , . . . ,N .

'k

From (17), (31a), and (32), it is a straightforward application of It6's Lemma to show that the total expected rate of return for holding source k from t to t dt is given by

+

(34) = [I

+ pk(a,- r)]dt .

230

Robert C. Merton

Thus, if the rights to the cash flows Y, between t and Tk were sold in the marketplace, the expected rate of return that would be required by investors Pk(ak- r ) . Therefore, 8, equals the to bear the risk of these flows is r required expected rate of return (the capitalization rate) minus the expected rate of growth of the cash flows, p,. By inspection of (33), Vk(t)can be expressed by the classic present-value formula for assets with exponentially growing cash flows. For 0, > 0, the perpetual (T, = w) value is Y,(t)lO,, and the limiting “earnings-to-price” ratio, Y,(t)lVk(t),is 8,, a constant. Applying the closed-form solution for P,we can by substitution from (27) and (32) into (29) and (30) rewrite the optimal demand functions as

+

m

D f ( t ) = Ab, (354

+ 2 HJt,,

- P,V,(t) , i = 1, .

. . ,N

1

m

=

Ab,

+ 2 H,h,,

,i

=

N

+

1 , . . . ,n

l

and

Having derived explicit formulas for the values of nonendowment assets, we turn now to the solution of the optimal portfolio and expenditure problem in the special case where the university’s objective function is given by

with p > 0 and r, 2 0, j = 1 , . . . ,m.Without loss of generality, we assume that Zyri = 1. From (8), the optimal Q, satisfy (37) From (36) and (37), the indirect utility function can be written as m

(38)

v(c,s,t)= exp(-pt){iog c - Crj[iog s, - log (r,)]} . I

It follows from (1 la) that the optimal expenditure rule is

(39)

C*(t) = exp( -pt)

1

J,(W,S,d .

It is straightforward to verify by substitution into (lo), (1 la), and (1 lb) that (40)

J(W,S,t) =

1 -

P

exp( - pt)log W

+ Z(S,t)

231

Optimal Investment Strategies for University Endowment Funds

for some function Z(S,t). By the verification theorem of dynamic programming, satisfaction of (lo), (1 la), and (1 lb) is sufficient to ensure that J in (40) is the optimum. It follows from (40) that Jkw = 0 and hence that H, = 0 in (13) and ( 3 3 , k = 1, . . . ,m.Therefore, for the log utility specified in ( 3 6 ) , there are no differential hedging demands for assets to protect against unanticipated changes in the costs of university activities. The optimal allocation of the university’s total net worth is thus instantaneously mean-variance efficient. Noting that A = -J,/J,, = W , we have that (35) can be written in this special case as (414

Df(t) = b,W - P,V,(t) , i = b,W ,i

1, . . . ,N N + 1 , . . . ,n

= =

and N

(41b)

D,*+,(t)

(1 - CbJW I

-

2

1 - P,Y,(t)

‘

I

By inspection of (41), in the absence of nonendowment assets, the fraction of endowment allocated to risky asset i in the university’s optimal portfolio is b,, i = 1, . . . ,n, and the fraction allocated to the riskless asset is (1 - q b , ) , independent of the level of endowment. If x,* = D f ( t ) / K ( t )is the optimal fraction of endowment invested in asset i, then from (41) the difference in fractional allocations caused by the nonendowment assets is (424

xf(t) - b, = R(b, =

Rbl

-

P,XJ , i = 1, ,i =N

. . . ,N

+

1 , . . . ,n

and

where A, = V,(t)/X:V,(t) is the fraction of the capitalized value of the university’s total nonendowment assets contributed by cash flow source k at time t , k = 1, . . . , N , and R = XyVl(t)/K(t)is the ratio of the values of the university’s nonendowment assets to its endowment assets at time t . As discussed in section 7.2, the differences in (42) are the result of two effects: (1) the “wealth” effect caused by the difference between the net worth and the endowment of the university and (2) the “substitution” effect caused by the substitution of nonendowment asset holdings for traded asset holdings. Suppose, for concreteness, that the expected returns, variances, and covariances are such that a positive amount of each traded risky asset is held in mean-variance efficient portfolios. Then, 6, > 0, i = 1 , . . . ,n. It follows that the impact of the wealth effect in (42a) and (42b), (RbJ, is unambiguous: it causes a larger fraction of the optimal endowment portfolio to be allocated

232

Robert C. Merton

to each risky asset and therefore a smaller percentage allocation to the riskless asset. Because p, 2 0 and A, > 0, i = 1, . . . ,N, we have that the impact of the substitution effect in (42a) and (42b), (RPzA,), is also unambiguous: for those traded assets 1 , . . . ,N for which the nonendowment assets are substi1, . . . ,n, tutes, the fractional allocation is smaller; for the traded assets N the fractional allocation is unchanged; and the allocation to the riskless asset thus increases. Because the wealth and substitution effects are in opposite directions for b, > 0, whether the optimal endowment portfolio allocates an incrementally larger or smaller fraction to traded asset k depends on whether 6, > P,A, or b, < PJ,. P,h, is the fraction of the total increment to net worth (from nonendowment assets) that is implicitly invested in asset k as the result of owning cash flow source k . If that fraction exceeds the optimal one for total wealth, b,, then the optimal endowment portfolio will hold less than the meanvariance efficient allocation. Indeed, if A, > (1 + R)b,/(RP,), then x,*(t) < 0 and the university would optimally short sell traded asset k in its portfolio. This is more likely to occur when R is large (i.e., nonendowment assets are a large part of university net worth) and A, is large (i.e., cash flow source k is a large part of the value of nonendowment assets). The implications of (42a) and (42b) for optimal endowment fit the intuitions discussed at length in section 7.2. For instance, if a significant amount of gift giving to a particular university depends on the performance of the general stock market, then in effect that university has a “shadow” investment in that market. Hence, all else the same, it should hold a smaller portion of its endowment in stocks than another university with smaller amounts of such market-sensitive gift giving. As noted in section 7.2, much the same substitution-effect story applies to concentrations in other assets, including real estate. The same analysis also follows where grants from firms or the government are likely to be strongly correlated with the financial performance of stocks in the related industries. However, the underweightings in those assets for substitution-effect reasons can be offset by sufficiently strong demands to hedge against costs, as is illustrated by the biotech example in section 7.2. The analysis leading to (29) and (30) requires that there exist traded securities which are instantaneously perfectly correlated with the changes in Y , , . . . , Y,. If this “complete market” assumption is relaxed, then the capitalized values of those nonendowment cash flow sources will no longer be independent of the university’s preferences and endowment. However, the impact on endowment investments will be qualitatively similar. This more general case of nonreplicable assets can be analyzed along the lines of Svensson (1988). We can use our model to examine the impact of nonendowment cash flow sources on optimal expenditure policy. From (39) and (40), we have that the optimal expenditure rule is the constant-proportion-of-net-worth policy

+

233

Optimal Investment Strategies for University Endowment Funds

(43)

C*(t) = pW(t)

However, current expenditure from endowment will not follow a constant proportion strategy. Optimal expenditure from endowment at time t is [C*(t) C y Y,(t)]dt,which can be either positive or negative (implying net saving from nonendowment cash flow sources). If s*(t) denotes the optimal expenditure rate as a fraction of endowment (= [C*(t) - Ey Y , ( t ) ] / K ( t ) ) then , from (4) and (43),

where R(t) is as defined in (42a) and (42b) and y ( t ) = [Cy Y,(t)]/[C; V,(t)] is the current yield on the capitalized value of the nonendowment sources of cash flow. In the special case of (33), where the cash flows are all perpetuities (i.e., T, = ~0 and 8, > 0, k = 1, . . . ,M, V,(t) = Y,(t)/e,and the current yield on source k is constant and equal to Or. In that case, y(r) = Cy X,8, the value-weighted current yield. From (31a), Ok will tend to be smaller for assets with higher expected growth rates of cash flow, (KJ. If on average the current yield on nonendowment assets is less than p, then the current spending rate out of endowment will exceed p. If the current yield is high so that y ( t ) > p, then s * ( t ) < p. Indeed, if y ( t ) > p( 1 R)/R, then s*(t) < 0 and optimal total expenditure is less than current cash flow generated by nonendowment sources. Because both R(t) and A,(t) change over time, we have from (44) that the optimal current expenditure rate from endowment is not a constant, even when expected returns on assets, the interest rate, and the expected rate of growth of nonendowment cash flows are constants. We can also analyze the dynamics of the mix of the university’s net worth between endowment and nonendowment assets. If a = r 2;b,(az - r ) denotes the instantaneous expected rate of return on the growth-optimum, mean-variance efficient portfolio, then, as shown in Merton (1990b, 169-71), the resulting distribution for that portfolio is lognormal with instantaneous expected return a(> r ) and instantaneous variance rate equal to (a - r ) . It follows from (6), (41), and (43) that the dynamics for the university’s net worth are such that W(t)/W(O)is lognormally distributed with

+

+

(45)

If X,(t) = V,(t)/W(t)denotes the fraction of net worth represented by nonendowment cash flow source k , then, because V , and W are each lognormally distributed, X,(t) is lognormally distributed, and from (33) and (45)

234

Robert C. Merton

4 l 0 g x ~ ]= [Pi + P

(a -

+ r + Si) 2

Ir

fork = I , . . . , N . From (46), the fraction of total net worth represented by all sources of nonendowment cash flow, X ( t ) = Xy X,(t) = R(t)/ll R(t)], is expected to grow or decline depending on whether p > 8,,, or p < On,,, where Om,, =- min(O,), k = 1, . . . , N . In effect, a university with either a high rate of time preference or at least one (perpetual) high-growth nonendowment asset (i.e., with p > Om,,) is expected to “eat” its endowment. Indeed, it may even go to a “negative” endowment by borrowing against the future cash flows of its nonendowment assets. Whether this expected growth in X ( t ) is the result of declining expected net worth or rising asset values can be determined from (45). Because a > r, if p Ir , then both the arithmetic and geometric expected rates of growth for net worth are positive. For p < O,,,,, it follows that E,[X(r)] -+ 0 as rwm. Hence, in the long run of this case, endowment is expected to become the dominant component of the university’s net worth. Of course, these “razor’s edge” results on growth or decline reflect the perpetual, constant-growth assumptions embedded in nonendowment cash flow behavior. However, this special case does capture the essential elements affecting optimal portfolio allocation and expenditure policies (cf. Tobin 1974). The formal analysis here assumes that endowment is fungible for other assets and that neither spending nor investment policy are restricted. Such restrictions on endowment could be incorporated, using the same Kuhn-Tucker type analysis used in section 7 . 3 to take account of the constraint that total expenditure at each point in time is nonnegative. The magnitudes of the KuhnTucker multipliers at the optimum would provide a quantitative assessment of the cost of each such restriction. However, including those restrictions is not likely to materially change the basic insights about hedging and diversification derived in the unrestricted case. The model can also be integrated into a broader one for overall university financial planning. Such integration would permit the evaluation of other nonendowment financial policies such as whether the university should sell forward contracts for tuition.

+

References Bergman, Yakov. 1985. Time preference and capital asset pricing models. Journal of Financial Economics 14(March):145-59.

235

Optimal Investment Strategies for University Endowment Funds

Black, Fischer. 1976. The investment policy spectrum: Individuals, endowment funds and pension funds. Financial Analysts Journal 32(January-February):23-3 1. Bodie, Zvi. 1976. Common stocks as a hedge against inflation. Journal of Finance 3 l(May):459-70. . 1982. Investment strategy in an inflationary environment. In The changing roles of debt and equity in financing U.S.capital formation, ed. Benjamin Friedman, 47-64. Chicago: University of Chicago Press. Breeden, Douglas T. 1979. An intertemporal asset pricing model with stochastic consumption and investment opportunities. Journal of Financial Economics 7( September):265-96. Brinkman, Paul T. 1981. Factors affecting instructional costs at major research universities. Journal of Higher Education 52(May-June):265-79. . 1990. Higher education cost functions. In The economics ofAmerican universities, ed. Stephen Hoenack and Eileen Collins, 107-28. Albany: State University of New York. Brovender, Shlomo. 1974. On the economics of a university: Toward the determination of marginal cost of teaching services. Journal of Political Economy 82(MayJune):657-64. Constantinides, George. 1989. Theory of valuation: Overview and recent developments. In Theory of valuation: Frontiers of modern finance theory, ed. Sudipto Bhattacharya and George Constantinides. Vol. 1, 1-23. Totowa, N.J.: Rowman and Littlefield. . 1990. Habit formation: A resolution of the equity premium puzzle. Journal of Political Economy 98(June):519-43. Detemple, Jerome B., and Fernando Zapatero. 1989. Optimal consumption-portfolio policies with habit formation. Working Paper. New York: Graduate School of Business, Columbia University. Duffie, Darrell, and Larry Epstein. 1992. Asset pricing with stochastic differential utility. Review of Financial Studies 5:411-36. Eisner, Robert. 1974. Endowment income, capital gains and inflation accounting: Discussion. American Economic Review 64(May):438-41. Ennis, Richard, and J. Peter Williamson. 1976. Spending policy for educational endowments. Research and Publication Project. New York: Common Fund, January. Fama, Eugene, and Michael C. Jensen. 1985. Organizational forms and investment decisions. Journal of Financial Economics 14(March):101-19. Fischer, Stanley. 1975. The demand for index bonds. Journal of Political Economy 83(June):509-34. Ford Foundation Advisory Committee on Endowment Management. 1969. Managing educational endowments: Report to the Ford Foundation. New York: Ford Foundation. Grinold, Richard, David Hopkins, and William Massy. 1978. A model for long-range university budget planning under uncertainty. Bell Journal of Economics 9(Autumn):396-420. Grossman, Sanford, and Guy Laroque. 1990. Asset pricing and optimal portfolio choice in the presence of illiquid consumption goods. Econometrica 58(January):25-52. Hansmann, Henry. 1990. Why do universities have endowments? Journal of Legal Studies 19(January):3-42. Hindy, Ayman, and Chi-fu Huang. 1992. Intertemporal preferences for uncertain consumption: A continuous time approach. Econometrica 6O(July):781-801. Hoenack, Stephen, and Eileen Collins, eds. 1990. The economics of American universities. Albany: State University of New York Press.

236

Robert C. Merton

Litvack, James, Burton Malkiel, and Richard Quandt. 1974. A plan for the definition of endowment income. American Economic Review 64(May):433-42. Long, John B. 1974. Stock prices, inflation, and the term structure of interest rates. Journal ofFinancial Economics l(Ju1y): 131-70. McDonald, John. 1974. Faculty tenure as a put option: An economic interpretation. Social Science Quarterly 55(September):362-7 1 . Merton, Robert C. 1977. A reexamination of the capital asset pricing model. In Risk and return injnance, ed. Irwin Friend and James Bicksler, Vol. 1, 141-60. Cambridge, Mass.: Ballinger. . 1985. Comment: Insurance aspects of pensions. In Pensions,labor and individual choice, ed. David A. Wise, 343-56. Chicago: University of Chicago Press. . 1990a. Capital market theory and the pricing of financial securities. In Handbook of monetary economics, ed. Benjamin Friedman and Frank Hahn, 498-581. Amsterdam: North-Holland. . 1990b. Continuous-time~nance.Oxford: Basil Blackwell. . 1990c. The financial system and economic performance. Journal ofFinancial Services Research 4(December):263-300. Nichols, Donald. 1974. The investment income formula of the American Economic Association. American Economic Review 64(May):420-26. Nordhaus, William. 1989. Risk analysis in economics: An application to university finances. New Haven, Conn. : Cowles Foundation, Yale University, May. Unpublished paper. Snyder, Thomas. 1988. Recent trends in higher education finance: 1976-77 to 198586. In Higher education administrative costs: Continuing the study, ed. Thomas Snyder and Eva Galambos, 3-23. Washington, D.C.: Office of Educational Research and Improvement, Department of Education. Sundaresan, Suresh. 1989. Intertemporally dependent preferences and the volatility of consumption and wealth. Review of Financial Studies 2:73-89. Svensson, Lars. 1988. Portfolio choice and asset pricing with nontraded assets. NBER Working Paper no. 2774. Cambridge, Mass.: National Bureau of Economic Research, November. . 1989. Portfolio choice with non-expected utility in continuous time. Economic Letters 39(0ctober):3 13-17. Tobin, James. 1974. What is permanent endowment income? American Economic Review 64(May):427-32.

Comment

George M. Constantinides

In addressing the complex problem of optimal investment strategies for university endowment funds, Robert Merton adopts the view that the university is an economic agent maximizing the expected utility of a set of activities subject to a budget constraint. In so doing, he is able to frame the problem in George M. Constantinides is the Leon Carroll Marshall Professor of Finance at the Graduate School of Business, University of Chicago, and a research associate of the National Bureau of Economic Research. The author benefited from the discussion at the NBER Conference on the Economics of Higher Education, May 17-19, 1991, at the Kingsmill Resort and Conference Center, Williamsburg, Virginia. In particular, he benefited from the comments of Michael Rothschild.

237

Optimal Investment Strategies for University Endowment Funds

the standard microeconomiclfinance paradigm and draw heavily on the finance literature to which he is a seminal contributor. Standard finance theory dictates that the university allocate its endowment plus capitalized nonendowment future income in a mean-variance efficient portfolio of financial assets, modified by an overlay of hedging portfolios designed to hedge against unanticipated shifts in the state variables. The extent of hedging depends on the prices and price expectations of financial assets and on the sensitivity of the university’s marginal utility of wealth to the shifts of the state variables. I begin by discussing the university’s objective in managing the endowment. Next I review Merton’s model and principal results. I then present some generalizations and offer some concluding remarks.

The Objective in Managing the Endowment The task of managing a university’s endowment ought to be placed in the general context of the goals of a university. In addressing the goals of the university, we discuss three questions: (1) What, if any, objective function do market forces impose on the university? (2) What objective function do the university trustees and administration apply in practice? (3) What is the socially desirable objective function of a university? In particular, we consider two paradigms of the university, first as a utility-maximizing agent and second as a profit-maximizing firm. The Objective Function Imposed by Market Forces Is it plausible to model the university as an economic agent with a university-wide increasing and concave utility function? Even if we assign a utility function to each one of the agents who make up the university, what devices exist which equalize their marginal rate of substitution and lead to the existence of a university-wide utility function? To some extent, universities compete for students, funding, and the services of faculty, officers, and staff. There are also transfer payments between the undergraduate and graduate divisions of the college, professional schools, and research groups. But I very much doubt that these market mechanisms are adequate to equalize the marginal rate of substitution and give rise to a university-wide utility function. Is it plausible to model the university as a profit-maximizing firm? A university does not have a clearly defined group of residual claimants. There are diverse groups of claimants to the services of a university which include the past, current, and future generations of students, the faculty, staff, and industry. Furthermore the threat of a takeover or reorganization, which leads corporations toward the goal of profit maximization, does not apply with equal force to universities, I view the university as a nexus of contracts among economic agents which include the state government and legislature in the case

238

Robert C. Merton

of state universities; the alumni, trustees, and officers; a highly individualistic and fragmented faculty; and last but not least a heterogeneous student body with “overlapping generations” features. Rothschild and White, in chapter 1 of this volume, present an insightful discussion of these issues. The Objective Function Perceived by Universities A search through the academic literature on the objectives of the university provides valuable insights but falls short of providing answers on what objectives universities do or should adopt. In addressing the investment income formula of the American Economic Association, Nichols (1974) assigns a utility function to the association and outlines the implications of the standard Fisherian analysis. Tobin (1974, 427) formulates the objective of the trustees of an endowed institution as “the guardians of the future against the present. Their task is to preserve equity among generations. . . . In formal terms, the trustees are supposed to have a zero subjective rate of time preference.” Tobin neither endorses nor justifies these perceived goals. Litvack, Malkiel, and Quandt (1974) suggest that endowment management should (1) seek to make investment management independent of the spending decisions of the university, (2) protect the real value of the endowment fund, and (3) stabilize spendable income. Finally, Hansmann (1990) surveys a number of possible theories to explain endowment accumulation and explores their strengths and weaknesses. Hansmann is unable to find a plausible and rational explanation for observed endowment accumulation policies. The lack of consensus on the perceived objectives of universities is hardly surprising. Universities are a diverse group of institutions with heterogeneous objectives. In fact, as we argue next, there is no compelling reason why universities should have uniform objectives. The Socially Desirable Objectives of a University

I argue that universities provide diverse services to the society, and therefore different universities serve the society best by adopting different objectives. As my first example, consider a state university system. Its financial collapse has major adverse effects on the current and future generations of students, the academic community, and the economic and cultural life of the state and beyond. It seems socially desirable that this university should follow a prudent policy of diversifying its instructional and research activities and also diversify its endowment and capitalized nonendowment income and hedge it against future contingencies. Merton’s paradigm of the university as a utilitymaximizing agent has implications which, in this example, are socially desirable. As my second example, consider a small, research-oriented university located in a rural area where the economic life is dominated by the local spe-

239

Optimal Investment Strategies for University Endowment Funds

cialized industry. Suppose that this university relies heavily on the local industry for research contracts, the supply of students, and endowment income. If we adopt Merton’s paradigm, we conclude that the university should diversify its research and instructional activities away from the local industry and should hedge its current and capitalized nonendowment income by selling short, if possible, the stock of the local industry firms. Is this policy socially desirable? A case can be made that society is best served by the exact opposite policy: specialize the research and instructional services to serve best the demands of the local industry and invest the endowment in the stock of the local industry firms. If the local industry declines into oblivion, so does the university whose social function was to serve this industry. If on the contrary the local industry flourishes, the university is best suited in terms of its specialization and financial strength to serve the local industry. Which of the two diametrically opposite policies is socially preferred? There is no simple answer. But I hope that the examples illustrate that diversification is not always an obvious social attribute of the university’s objective. Furthermore, society is served best by different universities following different, even diametrically opposite, policies.

A Review of the Model and Principal Results Merton defines the university’s objective as the maximization of a von Neumann-Morgenstern, time-separable, and concave utility function of the level of a set of activities Q(t) = [Q , ( t ) ., . . ,Q,,,(t)l as ~

The price (or net cost) of activityj is S,(t). No distinction is made between the margin21 and average cost of an activity; we therefore interpret the supply of activities as perfectly elastic. The vector of activity prices S(t) = [&(t), . . . ,S,(t)] is an exogenous autoregressive process which Merton models as a continuous-time diffusion process. The endowment capital at the beginning of period t is K(t). The nonendowment income at t is Y ( t ) , an exogenous stochastic process. Merton models [S(t), Y(t)] as a diffusion process. Merton assumes that the nonendowment income is spanned by the returns of the financial assets. Then the nonendowment income stream [Y(t),Y(r l ) , . . . ] may be capitalized with value f(t). The university’s wealth is defined as w(t)= K ( t ) f(t),the sum of endowment capital and capitalized present and future nonendowment income. The expenditure on activities at time t is Q’(t)S(t),where the prime denotes the transpose. The wealth net of the currentexpenditure on activities is y(t)

+

+

Robert C. Merton

240

- Q’(t)S(t)and is allocated among the financial assets with portfolio weights r(t)= [El(& .

. . , ~ , ( t ) ]which , sum up to one. The supply of financial assets is perfectly elastic, and the asset returns over one period are denoted by R(t 1) = [&,(t l), . . . Jn(r l)]. Merton models the joint process of financial asset prices, activity prices, and nonendowment income as a diffusion process. The wealth dynamics is

+

(2)

+

W(t

-

+

+

1)

=

[W(r) - Q’ (t)S(t)]w’(t)R(t

+ 1) .

Zero wealth is an absorbing state; that is, W ( t ) = 0 implies zero investment in the activities and in the financial assets at all future times. The control variables are the activity levels Q ( t ) and the portfolio weights vv(t). Stated formally, the university maximizesthe expected utility in (1) by the sequential choice of activity levels and portfolio weights subject to the sequence of budget constraints, ( 2 ) , and the constraint of nonnegative wealth. Essentially, Merton models the university’s problem as the standard intertemporal consumption and investment problem, which has been studied extensively in the finance literature (see Fama and Miller 1972; Ingersoll 1987; Merton 1990). Merton proceeds along familiar lines to define the indirect utility of consumption as

u[c(t),

(3)

tl = maxl@(t), Q(t)

r1

subject to Q’(t)S(t)= C ( t )and then define the derived utility of wealth as

(4)

J[W(r),S ( t ) , tl

=

max E,{ [w(t),C(t)l

i:‘ u [ c ( ~ ) ,

~ ( 7 1TI} ,

=

subject to the budget constraint. Merton’s primary focus is on the portfolio allocation, that is, the control variables E(t). In general, the optimal portfolio consists of a mean-variance efficient portfolio of the endowment plus the capitalized nonendowment income, modified by an overlay of hedging portfolios designed to hedge against unanticipated shifts in the state variables. In the special case where the indirect utility of consumption is the sum of the logarithm of consumption and a function of the state variables, a myopic policy is optimal: the university invests the endowment plus the capitalized nonendowment income in a meanvariance efficient portfolio, without an overlay of hedging portfolios.

Generalizations The University’s Production Function Whether we choose to view the university as a consumer of activities, a profit-maximizing firm, or a nexus of contracts, we should explore the pro-

241

Optimal Investment Strategies for University Endowment Funds

duction function of the university. The inputs are the expenditures on teaching, faculty, research, physical plant, and public relations-to mention just a few. The outputs are the activities in Merton’s terminology. We should recognize that it takes many years to build a reputation, to attract the best student applicants, to build a superior faculty, or to create a niche in a certain academic field. Therefore, the production function should incorporate adjustment costs and “time to build.” For a minority of activities, such as student aid or visiting faculty, the price per activity unit is exogenous and well defined. But for the majority of university activities, the prices are endogenous. Merton is aware of this and points out that his exogenous price processes of activities can be viewed as “reducedform” equilibrium price processes. Still it remains unclear whether Merton views these prices as marginal or average. In his budget equation (6), the activity expenditures are the sum of the product of activity levels and prices; therefore, prices are interpreted as average prices. In his first-order equations (8), the same prices play the role of marginal prices. Therefore, the distinction between the marginal and average price of an activity is not drawn. The distinction can be drawn by introducing a production function. The Intertemporal Complementarity and Substitutability of the Activities in the University’s Preferences Some university activities exhibit strong intertemporal substitutability: a university basks in the glory of a Nobel laureate among its ranks long after the laureate has retired. Other activities exhibit strong complementarities: a university is more disturbed by the lowering of its academic ranking than by the maintenance of a steady but low ranking. These effects can be modeled in one of two ways. The first is to draw a distinction between university outputs and activities. The university outputs are durable goods which produce a stream of activities over time. In this case, the stream of activities is not directly controllable, and Merton’s analysis needs to be modified accordingly. The second way to model these effects is to define the university’s preferences over the outputs rather than over the activity flows from these outputs. But then the preferences are no longer time separable, and we can no longer define a time-separable indirect utility function of consumption as in (3), except in simple cases.

Concluding Remarks In the context of a simplified, or “reduced-form,’’ model of a university as a utility-maximizing agent, Merton has demonstrated that the basic principles of finance apply and, in particular, endowment funds should be managed according to the principles of diversification and hedging. I have argued that universities are a diverse group of institutions with heterogeneous functions in the society. Whereas diversification of instructional

242

Robert C. Merton

activities, research activities, and endowment and capitalized nonendowment income may be reasonable and socially desirable objectives of some universities, it is an open question whether these objectives are reasonable and socially desirable for the whole spectrum of universities. References Fama, Eugene F., and Merton H . Miller. 1972. The theory offinance. New York: Holt, Rinehart and Winston. Hansmann, Henry. 1990. Why do universities have endowments? Journal of Legul Studies 19(Januaryj:3-42. Ingersoll, Jonathan E. 1987. Themy offinancial decision making, Totowa, N.J.: Rowman and Littlefield. Litvack, James M., Burton G. Malkiel, and Richard E. Quandt. 1974. A plan for the definition of endowment income. American Economic Review 64(May):433-37. Merton, Robert C. 1990. Continuous-timefinance. Oxford: Basil Blackwell. Nichols, Donald A . 1974. Endowment Income, Capital Gains, and Inflation Accounting. American Economic Review 64(May):420-26. Tobin, James. 1974. What is permanent endowment income? American Economic Review 64(Mayj:427-432.

8

Public Choices in Public Higher Education John M. Quigley and Daniel L. Rubinfeld

8.1 Introduction Public institutions of higher education have grown in prominence in the United States over the past 200 years. By the mid-l980s, total public enrollments were roughly twice the level of private enrollments. Important as they are, these aggregate trends mask the substantial and systematic state-by-state variation in public and private enrollments that is the primary focus of this paper. The outputs associated with public higher education are notoriously difficult to conceptualize and to quantify. We therefore concentrate attention on input measures that can proxy for educational output: enrollments per capita and expenditures per student. To a large extent, current enrollment levels reflect a historical set of decisions by state legislatures concerning the appropriate “supply” of public higher education. But they also depend on the demand for higher education by both residents and nonresidents. In this paper, we relate the 1985 statewide pattern of publicly provided higher education to the political conditions and choices that have confronted legislatures, along with the labor market conditions and other economic forces that affect students’ (and families’) demands for higher education.

’

John M. Quigley is professor of economics and public policy at the University of California, Berkeley. Daniel L. Rubinfeld is professor of economics and law at the University of California, Berkeley. Support for this research was provided by the National Bureau of Economic Research and the Center for Real Estate and Urban Economics, University of California, Berkeley. The paper benefited from the suggestions of Norton Grubb and Charles Clotfelter. Maya Ibser provided valuable research assistance. 1. Public enrollments averaged 1,166,934 per state, while private enrollments averaged 620,87 I .

243

244

John M. Quigley and Daniel L. Rubinfeld

Section 8.2 provides the conceptual overview. We sketch out some of the alternative political-economic theories that might serve to explain the current pattern of student enrollments. Section 8.3 begins the empirical analysis by describing the statewide public enrollment pattern as it has developed historically and as it relates to other input and output measures (expenditures per pupil and a quality index). In this section, we flesh out the public-choice problem of the legislature which causes states to provide alternative packages of postsecondary education services. Section 8.4 describes the regression analyses that attempt to sort out the effects of legislative supply variables from the important demand-oriented variables. Some brief concluding remarks appear in the final section.

8.2 Why Public Provision? Enrollment patterns are in part historically determined; additionally, they are intricately related to the availability of private institutions in or near each state. As a result, there is no single, simple theory that is likely to explain fully the variation in interstate public enrollment rates. In this section, we sketch out a number of alternative theoretical views that might help to explain the existing spending-enrollment pattern. The theories also serve to explain why states choose to provide subsidized higher education at all, in light of the economic evidence suggesting that most of the benefits of college training are fully captured by the graduates themselves.2 One approach emphasizes human capital and mobility. According to this view, states wishing to import valuable human capital will provide relatively high levels of public spending and enrollment opportunities, at least to the extent that they believe they can convince graduates of the public institutions to continue to reside in their states. A second public-choice explanation concentrates on the political benefits associated with the provision of higher education. According to this view, politicians use logrolling to trade for forms of public higher education that benefit their constituencies. California’s increased emphasis on the provision of two-year institutions is consistent with this view; logrolling may have led to the creation of a large number of new institutions in a sufficiently large number of legislative districts so as to make the entire educational package politically viable. This theory is also consistent with the Leviathan theory of big government, according to which political support by self-interested bureaucrats may be sufficient to cause the subsidized public sector to grow. A third perspective (taken by Bowles and Gintis 1976, for example) suggests that subsidies to higher education are simply a means that the capitalist 2. See, for example, the discussion in Douglas (1977). He argues that the average private economic return on higher education has been falling and is just above the break-even point. This does suggest, however, that for the lower-paying professions, the return is negative, which might provide an additional motive for subsidizing higher education-especially two-year institutions.

245

Public Choices in Public Higher Education

class has of distributing state resources to the middle class. Karabel (1974) follows this line of reasoning to explain the emergence of two-year institutions as a means of keeping the working class “in their place.” The growing emphasis on two-year institutions, especially in the West, could be consistent with this view, but the theory does not explain why such institutions have been so successful in other areas of the country. Yet another explanation for higher-education subsidies relies on the inherent optimism of the lower and middle socioeconomic classes concerning their own prospects of moving up the economic ladder. This optimism, we believe, helps to explain the widespread historical support for assessing relatively low marginal tax rates on upper-middle-income individuals. In general, to the extent that lower- and middle-income groups are unusually optimistic, they may vote (directly or indirectly) to support subsidies which tum out ex post to be detrimental to their narrow “class interests.” Finally, it is important to recognize the rather unusual place that education plays in our society. To some, education is a form of private secular religion; as such it may receive substantial political support, regardless of the calculus of immediate benefits.

8.3 8.3.1

Interstate Variation in Public Higher Education A Historical Perspective

Higher education in the United States was provided entirely through the private sector in the early years of the republic. Beginning with Harvard University in 1636, a total of nine private institutions had come into existence by the time of the American Revolution. The 75 years following the revolution was a period of great expansion, and by 1861 over 800 colleges had been founded. However, most of these colleges lacked either the faculty, students, or funding to survive; in 1900 only about 180 were in existence (Westmeyer 1985). The first public institution of higher education was founded in 1816 (it is now the University of Virginia). At the time, there was little demand-side pressure, since relatively few students were completing college preparatory programs. However, the role of public institutions became far more important with the passage of the first Morrill Act in 1862, creating the land-grant colleges. The Morrill Acts (the second was in 1890) mandated support for at least one college devoted to agriculture and the mechanical arts in every state, with land and funds provided by the federal government. Some states used this support to expand existing colleges; others used it to establish new institutions. Public institutions would most likely have been prominent earlier, were it not for the precedent of the Dartmouth College case of 1819. In that case, the U.S. Supreme Court ruled that states had no authority over private institutions

246

John M. Quigley and Daniel L. Rubinfeld

and that assumption of such authority amounted to a breach of contract with those institutions. If the court had ruled otherwise, it seems likely that instead of the subsequent proliferation of private institutions, states would have taken control of (and made “public”) many existing institutions. The next great expansion in the number of institutions of higher learning came after World War 11, when the GI Bill facilitated great increases in enrollments. Since that war, enrollments have fluctuated, in part in response to economic and demographic change^.^ 8.3.2

Variations in Public and Private Enrollments

Much of the empirical research on higher education has focused on enrollments, in both four-year and two-year institutions (see, e.g., Christensen, Melder, and Weisbrod 1975; Corazzini, Dugan, and Grabowski 1972; Corman 1983; Hopkins 1974; Hoenack and Pierro 1990). Enrollment studies, in turn, have tended to concentrate on the micro decisions of students (enrollment is related to income, tuition, and the opportunity cost of attending school) (see Feldman and Hoenack 1969; Galper and Dunn 1969; Kohn, Manski, and Mundel 1976; Mattila 1982; Sulock 1982; Weiss 1972). The “supply side” is usually given exogenously in terms of tuition, quality of schools, and state and federal support for public education. The only endogenous variable involves the decision rule for admitting students. We take a broader view in this paper, emphasizing the legislative decision involved in providing a statewide system of public higher education. We use the level of student enrollments per 1,000 population in 1984-85 as a means of comparing the opportunities for public education across states. (Throughout the paper, enrollments per 1,000 population will be referred to as enrollment per ~ a p i t a . ) ~ The range and variability of public and private enrollments is striking. Per capita enrollments range from a low of 25.1 in Georgia to a high of 61.7 in Arizona (and 55.4 in California), with a mean of 39.3. The coefficient of variation is 20.8 percent. The variability of private enrollments is even greater, ranging from per capita enrollments as low as 0 in Wyoming and 0.32 in Nevada to enrollments as high as 40.5 in Massachusetts (and 35.8 in Rhode Island). In contrast to public enrollments, the coefficient of variation for per capita private enrollments is 78.2 percent. Figure 8.1 shows how public enrollments per capita vary by region of the 3. For a more detailed description and analysis of the sources of enrollment changes, see Clotfelter (1991), chapters 1 and 5. 4. Enrollment per capita allows one to conceptualize education as benefiting all state residents. The alternative enrollment rate, as measured by the ratio of enrollments to population aged 1830, focuses more directly on the choices that potential students make. This choice is unsatisfactory, since it fails to account for the fact that older students are a growing portion of college students. According to Corman and Davidson (1984), over 33 percent of college students were 25 and older in 1979, and the number is expected to increase to 43 percent by 1990. In any case, the results were essentially unchanged, whichever variable was utilized.

247

Public Choices in Public Higher Education

Slate

Fig. 8.1

Public enrollment per capita by region

country. (Each of the 50 states appears on the horizontal axis.) Clearly, the mean enrollment rate increases as we move from the East to the South, Midwest, and West. Within each region, however, there remains substantial variation. In the East, for example, enrollment rates range from a low of 25.4 in Pennsylvania to a high of 46.0 in Maryland. By contrast, the range in the West is from a low of 34.7 in Idaho to a high of 61.7 in Arizona. There is also a pronounced regional pattern in private enrollments, as figure 8.2 shows. Per capita private enrollments are substantially higher in the East than elsewhere. Among the other three regions, enrollment rates are highest in the Midwest, followed by the South, and then the West. There remains substantial variation within the East, with the lowest private enrollment rate of 7.1 in Connecticut and the highest, 40.5, in Massachusetts. By contrast, with the exception of Utah in the West, there is relatively little within-region variation in private enrollment rates. These important regional differences in higher education were largely historically determined, as figures 8 . 3 and 8.4 suggest. These figures describe enrollment rates by order of statehood, from 1st to 50th. The positive relationship between public enrollment and order of statehood, and the corresponding negative relationship for private enrollment rates, is immediately clear.5 From a cross-sectional viewpoint, one might view each state as making (or having made) a “public choice” about the mix of public and private enrollments to provide. (Alternatively, the legislature makes a public enrollment 5. When we attempted to sort out geography (region) from history (order of statehood) by examining the pattern of enrollment rates within region by order of statehood, we found no discernible relationship.

248

John M. Quigley and Daniel L. Rubinfeld

Slale

Fig. 8.2 Private enrollment per capita by region

State

Fig. 8.3 Public enrollment per capita by order of statehood

choice, conditioned on the availability of private alternatives.) The enrollment trade-off is shown in figure 8.5. Overall, there is a negative relationship between public and private enrollment rates, ranging from the low-public, highprivate extreme of Massachusetts to the high-public, low-private alternative of Arizona. The relationship appears to be nonlinear, as shown by the best-

State

Fig. 8.4 Private enrollment per capita by order of statehood 70

60

..........

...

L..

......................

......

I

.

2 40

-8..

.......

.........

n

a

30

20 0

10

20

30

Private Enrollment per capita

Fig. 8.5 The public-private education bundle

40

50

250

John M. Quigley and Daniel L. Rubinfeld

fitting hyperbola that appears in the graph.h The relationship would not be substantially changed if the South, where all enrollment rates are relatively low, were removed from the data set. 8.3.3 The Mix of Educational Opportunities Within the public sector, states vary substantially in the mix of educational opportunities that they provide. Figure 8.6 provides one interesting perspective by illustrating the generally negative relationship between enrollment rates for two-year and four-year institutions. This pattern has come about as the result of a rapid increase in the enrollments in two-year institutions that occurred during the 1970s.’ By 1980,91 percent of students in two-year institutions were in public as opposed to private schools; the comparable figure for four-year schools was 67 percent (Grubb 1988, 301-2). Four-year institutions should themselves be broken down into those that are primarily liberal arts teaching institutions (hereafter “colleges”) and those that emphasize research and offer extensive programs of doctoral studies. Of the 1,993 four-year institutions in the United States, 150 are classified as research institutions; 38 percent of those are private and 62 percent public. Of the group of 1,843 colleges, 75 percent are private.* Once again, the current mix of institutions reflects historical development. Institutions with graduate programs are a phenomenon of the last 100 years. Although the first Ph.D. was awarded at Yale in 1861, by 1930 only 2,024 Ph.D.’s had been awarded. Graduate enrollments increased more substantially, from 198 in 1878 to 2,382 in 1890, 9,370 in 1910, and 47,255 in 1930. Two-year, or community, colleges are even more recent in their origin but have grown much faster. The first junior colleges were founded at the turn of the century, with 52 in existence in 1920, 610 in 1941, and 1,100 in 1970. 8.3.4 The Public versus Private Choice There is, of course, substantial variation in the quality of both public and private institutions. To examine the quality issue, we rely upon the Gourman (1987) index measuring the proportion of rated institutions in each state classified as “strong” or “good.”’ The Gourman index is based on the opinions of a substantial number of individuals active in the field of higher education about the quality of faculty, students, individual departments within a school, school administration, and library facilities. While such a subjective measure should be viewed with some skepticism, it is reassuring to note that the Gour6 . The regression estimated curve in figure 8.5 plots the relationship: log(pub1ic enrollment per capita) = 3.8959 - . I181 log(private enrollment per capita). 7. According to Grubb (1988), enrollments in two-year institutions grew at a rate of 1 1 percent per year during the 1970s, as compared to a 2 percent annual rate of growth for four-year colleges. The source is the Statistical abstract o/the United States, table 260. 8. Of the 1,305 two-year institutions, only 28 percent are private. 9. An index of the number of institutions rated “good’ and “strong” yielded similar results.

Public Choices in Public Higher Education

251

.'

' .

10

20

. *

30

40

4-Year Public Enrollment per capita

Fig. 8.6 The two-year-four-year public education bundle

man index is reasonably highly correlated with other objective measures of quality. I n According to our constructed measure of quality, Arizona, California, and Iowa have the highest-quality public institutions, and Colorado, Rhode Island, and Massachusetts the highest-quality private institutions." Figures 8.7 and 8.8 show that there is a generally positive relationship between quality and enrollment rates; those states with institutions of the highest quality tend to have the highest enrollments. The public choice between public and private, high- and low-quality insti10. Solmon (1973, table 1) found the overall Gourman index to have a correlation of .80 with average faculty salary; .62 with the SAT math score of enrolled students; .75 with the departmental research, instruction, and library expenditures; and .71 with basic expenditures. 1 I . The Gourman rating system considers all schools in a state. The quality index used here is unweighted. The results were quite similar when an index of the total number of highly rated schools was used in its place.

252

John M. Quigley and Daniel L. Rubinfeld 0.75

UI

-

.-0 5 ._ L

0.50

....

0.25

. . ..... ~

C

0 ._

E a

c

.2. m

6 c

-

/

C

'. ,.

,.

0.00

20

Fig. 8.7

30

40 50 Public Enrollment per capita

60

70

Public enrollment per capita and institutional quality

tutions is more complex than any single graph can show. The listings in table 8.1 make this clear.'* A group of 11 states, located primarily in the South, offer educational bundles consisting of low public and low private enrollments in schools that are not highly rated. A second group, 8 states primarily in the East, offer low public enrollment and high private-high-quality bundles. At the opposite end of the spectrum, a third group, 18 states located primarily in the Midwest and West, offer high public, often high-quality, but low private enrollment rates. Only 1 state, Illinois, would be classified as providing both high public and high private enrollment rates. The remaining 12 states lie someplace in between these sharply contrasting combinations. An alternative measure of the quality of public and private institutions is given by an input to the educational production function: the dollar level of expenditures per student. Figure 8.9 shows that four-year private liberal arts colleges exhibit a clear, positive relationship between per student expendi12. This classification was accomplished by dividing each enrollment series into three approximately equal parts-high, moderate, and low.

Public Choices in Public Higher Education

253 0.6

0.5

c u) ._ 5 .-

0.4

I

ul

-

-J (u

2

B

0.3

.2. -

4

0 3

$-

0.2

0.1

0.0

0

10

20

30

40

50

Private Enrollment per capita

Fig. 8.8 Private enrollment per capita and institutional quality

tures and enrollment rates. Other things equal (e.g., tuition), this suggests that states with “better” schools have higher enrollments. I 3 However, as figure 8.10 illustrates, the pattern is just the opposite for public education. Here we see a negative relationship between expenditures per student and enrollment rates.I4 The suggestion is that some states opt to offer a high-quality, lowquantity public education alternative, while others offer a low-quality, highquantity combination. Is Part of the public-choice decision that a legislature must make is the (joint) choice of the level of tuition and the level of public subsidy. To measure the subsidy for all schools, we use the state appropriation per full-time equivalent 13. The same pattern holds for private four-year universities. 14. Alaska, with a per capita expenditure of $17,042, has been excluded as an outlier. The negative relationship is somewhat less pronounced when Alaska is included. Note, in addition, that the relationship is essentially the same when expenditures per full-time equivalent (FTE) student at four-year universities is related to four-year public enrollments per capita. 15. For two-year private institutions, there is no relationship between expenditures per student and enrollment rates, while for two-year public institutions, there is a slight positive relationship.

254

John M. Quigley and Daniel L. Rubinfeld

Table 8.1

Classification of Enrollment by States Private Enrollment

Public Enrollment

Low

Low

Alabama Arkansas Florida Georgia Hawaii Idaho Kentucky Louisiana Mississippi South Carolina West Virginia

Moderate

Montana

Modcrate

High

Missouri New Jersey South Dakota Tennessee

Connecticut Maine Massachusetts New Hampshire New York Pennsylvania Rhode Itland Vermont

lndEE

Utah

~~~

hWd

Minnesota North Carolina Ohio

High

Alaska Arizona California Colorado Delaware Kansas MaryIand Michigan Nevada New Mexico North Dakota Ok!ahoma OEiP Texas Virginia Washington Wisconsin Wyoming

Nebraska

Illinois

Nores: States rated as having high-quality public institutions are underlined; those with highquality private institutions appear in bold print. A state may rate high in both categories. A state has high-quality public or private education if two or more institutions are rated good or excellent.

student (SUPPB). I h (See the Appendix to this chapter for the definitions of and data sources for variables presented hereafter.) For two-year schools and four-year colleges (excluding research universities), we calculated the subsidy 16. Unfortunately, we were unable to find direct meamres orthe subsidy per two-year and fouryear student.

Public Choices in Public Higher Education

255 40

. m .-c Q

30

8 t

n c

c

.

E -

*

. *

g 20 c

W

9 5 (I) c

Q

._ ?

a 10

2500

5000

7500

10000

12500

Private 4-Year Liberal Arts Expenditure per student

Fig. 8.9

Private four-year per capita enrollment and per student expenditure

per full-time student (SUPP2 and SUPP4) as the difference between per student expenditures and average per student tuition. Subsidies vary substantially among states. For two-year schools, they range from a low of $2,056 per student to a high of $6,590 (with a mean of $3,723); for four-year colleges, they range from $1,864 to $15,101 (with a mean of $4,412). Tuition is a major component of the price of a public education to a prospective student.” Figure 8.1 I shows a negative relationship between the tuition at two-year public institutions and enrollment rates; states with the lowest tuition (and, other things equal, the lowest subsidies) have the highest enrollment.IXThe pattern is the same with respect to tuition and enrollment at fouryear colleges (see figure 8.12).l9 17. But see Nerlove (1972) for a more complete discussion of the relationship between tuition and the price that efficiently allocates scarce resources. 18. Two-year public tuition was highest in Pennsylvania ($3,595) and Vermont ($2,525) and lowest in California ($250), North Carolina ($382), and Montana ($420). 19. Four-year tuitions ranged from a high of $3,547 in Vermont and $3,202 in New Hampshire to a low of $739 in Oklahoma and $740 in West Virginia. There is little relationship, however, between doctoral tuition and four-year enrollment rates.

256

John M. Quigley and Daniel L. Rubinfeld

.

.

. . 2500

5000

7500

10000

Public 4-Year Liberal Arts Expenditure

Fig. 8.10 Public four-year per capita enrollment and per student expenditure

The level of tuition and the subsidy represent two sides of a coin from a legislature’s point of view; one can offer a low tuition and easy access to the system of higher education at a high budgetary cost, or one can cut budgetary costs by offering a high tuition. To see how these policy choices vary among states, we calculated the ratio of tuition to per student expenditures individually for both two-year and four-year nonuniversity public institutions. The variation in both was substantial. For two-year schools, the tuition-toexpenditure ratio varied from a low of 6.3 percent in California to a high of 63.6 percent in Pennsylvania (with a mean of 22.5 percent). For four-year colleges, the range ran from a low of 11.4 percent in Alaska to a high of 63.2 percent in New Hampshire (with a mean of 28.3 percent). Interestingly, these tuition rates are positively correlated with private per capita enrollments (the correlation coefficient is 0 . 3 1 for two-year institutions and 0.29 for four-year colleges). Thus, legislatures in states with substantial private alternatives tend to charge higher tuition than do states with relatively little to offer in the private sector. These results are apparently supply-side rather than demand-side driven. Presumably, providing easy access to the

257

Public Choices in Public Higher Education

Average Tuition at 2-Year Institutions

Fig. 8.11 lbo-year public tuition and enrollment

public sector is most important in states whose residents have little in the way of in-state private-sector options. Note the clear regional pattern to the levels of tuition charged by public systems of higher education. As the bar cart in figure 8.13 shows, public tuitions are lowest (and subsidies highest) in the West, while tuitions are at their peak in the East.*O Public tuitions in all regions are small in relationship to private tuitions. Figure 8.14 illustrates this, along with the fact that private tuitions are at their peak in the East and are lowest in the South. To pursue the public-private analysis along a further dimension, we distinguished between public tuition for state residents and nonresidential tuition. The general pattern can be seen in figures 8.15 and 8.16. The former shows the statewide variation in the ratio of residential tuition to expenditures per FTE student; the latter shows the comparable ratio for nonresidential tuition. Because of comparability issues, it is more instructive to make relative com20. The pattern is unchanged when we deflate tuition by a cost-of-living index.

John M. Quigley and Daniel L. Rubinfeld

258 40

-

.. .

m c .n

m

30-

a,

4

'

Q

..

n 2

a

0 '

-*

.

. 10.

Fig. 8.12 Four-year public tuition and enrollment

State

Fig. 8.13 Tuition at public institutions by region

259

Public Choices in Public Higher Education

state

Fig. 8.14

Public and private tuition by region

Stale

Fig. 8.15 Ratio of nonresidential tuition to expenditure per FTE public student

parisons than to reflect on the difference between these ratios and the fullfunding fraction, 1.O.*' 21. The former was given by TUITRIEXPLAPUB, the latter by TUITNRIEXPLAPUB. In both cases, we attempted to make the 1985 tuitions comparable to the 1988 expenditures, using the overall consumer price index to adjust.

260

John M. Quigley and Daniel L. Rubinfeld

state

Fig. 8.16 Ratio of residential tuition to expenditure per FTE public student

There is substantial variation in the tuition ratios. Of the states reporting, resident tuition ratios are the highest in South Dakota, New Hampshire, and Indiana. The residential tuition ratios are several multiples higher than the states with the lowest tuition ratios-California, New Mexico, and Tennessee. Similarly, the highest nonresident tuition ratios occur in New Hampshire, Colorado, Massachusetts, and Oregon, while the lowest occur in Alabama, Missouri, South Carolina, and Tennessee. Overall, these two ratios are positively correlated as one would expect: states can generally be labeled as hightuition or low-tuition states. One might expect to find that states use tuition rates ( a ) as a means of attracting talented out-of-state students who might choose to reside permanently in the state and ( 6 ) as a means of extracting payment from talented in-state students who might choose to leave the state after receiving a subsidized education. However, we found very little correlation between mobility and other demographic variables and the tuition ratio variables.22 Finally, we conclude this section with an unsettling, yet important, issue. We have carried out our statistical analysis on the presumption that there is a clearly defined distinction between public and private higher education. In fact, the distinction has always been a fuzzy one. The definition of what is public and what is private actually differs from state to state; the key distinction is control, not funding.23 A public institution is publicly controlled 22. The exception is the proportion of the population that is black and the proportion that is Hispanic. Both were negatively correlated with both tuition ratios. 23. The distinction was a central issue in the Darrmoufh College case, in which colleges, once chartered, were protected from state control. See Hofstadter and Smith (1961) for details.

261

Public Choices in Public Higher Education

State

Fig. 8.17 Per student state support for private institutions

whether publicly or privately funded. Correspondingly, a private institution is privately controlled even if partly subsidized by public funds. As evidence of the confusion that would arise if funding were the source of the public-private distinction, examine figure 8.17. The bar graph illustrates the extremely wide divergence in per student public support for private institutions. Public support is very substantial in the eastern and midwestern states of New Jersey, Michigan, Maryland, and New York, yet essentially nonexistent in a large number of states, many of which are located in New England and the West. Overall, state public support for private higher education is highest in the Midwest and the East and lowest in the West. It is strongly negatively correlated with the mobility of the student population, a result consistent with the view that such subsidies may be most advantageous when given to a group of students that reside within the state. Having discussed the relationships between enrollments and some of the important enrollment determinants, we now move to the statistical analysis of the statewide pattern of public enrollments.

8.4 An Econometric Model of Public Choices in Public Higher Education The per capita public enrollment rate in a given state will be determined by the interaction of demand (students’ choices) and supply (legislative choices). Further, both choices are highly dependent on the historical development of

262

John M. Quigley and Daniel L. Rubinfeld

private colleges and ~ n i v e r s i t i e s .Unfortunately, ~~ the theories that we have sketched out do not provide us with a clear set of restrictions that allow us to identify demand and supply variables. We have attempted to make these distinctions, nevertheless, in order to relate our work to earlier literature and to provide some preliminary tests of the underlying theory. Caution should be used in reaching conclusions about structural parameters from the analysis that follows. From the demand perspective, we expect the dependent variable, the per capita public enrollment rates (ENB), to be negatively related to the opportunity cost of getting an education, as represented by the unemployment rate (UNEMP), so that the coefficient of UNEMP would be positive. Similarly, as Hoenack and Weiler (1979) suggest, a higher unemployment rate may provide evidence of an increased value of a college degree (especially at a two-year school). On the other hand, higher unemployment rates may be associated with lower family incomes, which would reduce the demand for higher education. We also expect the ENB to be positively related to the opportunity to find jobs. The number of service jobs per capita (SERV) provides one jobopportunity index. Given the employment trend away from manufacturing and toward the service sector, states with a higher proportion of service jobs are more likely, other things the same, to have job openings when economic conditions are good.25 With respect to the financing of public higher education, we expect enrollment rates to be negatively related to the direct out-of-pocket cost of education (average public tuition, TUITB, or the vector of tuition variables TCLAPUB [median tuition at four-year colleges] and T2YRPUB [median tuition at twoyear public institution^]).^^ Enrollments should also be positively related to the quality of public institutions in the state, as represented by the quality rating (QUALB) and by the level of expenditures per student (EXB). For all public schools-and twoyear schools, where location is an important issue-we would also expect the cost of alternative private education (TUITV, or TCLAPRI-the median tuition at four-year colleges-and T2YRPRI-the median tuition at two-year institutions) and the level of competing private expenditures per student (EXV) to be relevant. Finally, we have also included as a demand-determining variable the pro24. Thus, order of entry into statehood is highly correlated with enrollment rates. Order of entry is also highly correlated with a number of “demand” and “supply” variables. Consequently, we have chosen to present our results in terms of the latter. If order of statehood were added as an additional explanatory variable, it would be highly insignificant in both the demand and supply equations. 25. We found a host of other job-mix variables to be statistically insignificant and to have insubstantial coefficients. 26. Corman and Davidson (1984) relate enrollment rates in two-year and four-year colleges to tuition, unemployment, and income. Sulock (1982) has a similar analysis of community colleges.

263

Public Choices in Public Higher Education

portion of the population aged 18-24 (YOUNG). This variable not only measures the demand of the most relevant subgroup of the population but also has the effect of translating our dependent variable, enrollments per capita, into an enrollment rate per population aged 18 to 24.27 From the legislative supply perspective, we expect higher enrollment rates to provide direct social benefits of higher education to the state as well as benefits to legislators who are responding to the interest groups made up of potential students (and their families) and potential employers.28Thus, we would expect a positive relationship between the enrollment rate and income (INC) and the percentage of the population that is in the 18-24 age group (YOUNG). We would also expect a positive relationship between the enrollment rate and the elasticity of the state’s tax base (ELAST80);29a state is likely to find it politically easier to raise revenues to finance public education if state revenues are likely to increase substantially with growth (thus obviating the need to raise tax rates).30In addition, we expect a negative relationship between the relative price of goods in the state (PRICE) and the public subsidy to education (SUPPB). Other variables that would arguably affect legislative supply are the proportions of the population that are black (BLACK) and Hispanic (HISP), as minority groups might provide political pressure, especially with respect to the provision of two-year educational opportunities; the proportion of the population that is located in metropolitan areas (METRPOP), since the more urbanized the population, the greater access that students are likely to have to public higher education opportunities, especially for two-year schools; and a measure of the mobility of the population (MOBIL), because the more mobile the population, the lower the proportion of public benefits that reach citizens of the state.31This effect of mobility is likely to be masked, however, by the fact that MOBIL also provides a measure of the recent growth in a state’s population, which we would expect to be positively related to enrollment supplyWe have also included the (endogenous) level of public tuition in the legislative supply equation, but its effect is certainly ambiguous. On one hand, a higher tuition provides more revenue and therefore lowers the financial pressures on the state. (Part of this effect will be reflected statistically, by the inclusion of the state appropriations variable.) On the other hand, a lower tuition 27. The translation would be exact if the enrollment variables were in logarithmic form. Thus, log(ENB/POP) - lOg(POP18-24/POP) = log(ENB/POP18-24). 28. See Hoenack and Pierro (1990) for a recent application of interest group theory to the legislative supply of public higher education in Minnesota. 29. It is, of course, possible that ELASTLO could itself be affected by the enrollment rate. We found that making ELAST80 an endogenous variable did not change any results significantly. 30. Clotfelter (1976) tests a different version of the “fiscal illusion” hypothesis. He finds a very small, insignificant relationship between his measure of tax complexity (a Herfindahl index calculated using nine categories of taxation) and per capita expenditures on higher education.

264

John M. Quigley and Daniel L. Rubinfeld

provides greater opportunities for students of all socioeconomic strata to obtain a public education, an important goal for the state systems of public higher education. The former argument would be consistent with a positive coefficient on TUITB, while the latter would suggest a negative coefficient. Finally, we have chosen not to include a direct measure of private-sector alternatives in our supply equation; we see the private-public distinction as more of a demand-side phenomenon. Nevertheless, we have tested alternative models in which per capita private enrollment (ENV) does appear as an additional variable. We believe that tuition and public expenditures per student are most reasonably specified as endogenous variables. We also wanted to allow for a nonzero error covariance between demand and supply equations. We therefore estimated the student demand and legislative demand equations using three-stage least squares (3SLS) as well as ordinary least squares (OLS), the latter for comparison purposes. All of the variables that appear in the regression analyses are defined in table 8.2. The OLS and 3SLS results for overall enrollment rates are given in tables 8.3 and 8.4.32Results that are specific to two- and four-year institutions appear in tables 8.5 and 8.6. These were estimated using seemingly unrelated estimation, to account for the expected negative crossequation negative error c ~ r r e l a t i o n . ~ ~ 8.4. I

Student Demand

As expected, public tuition has a negative effect on the enrollment rates. The price elasticity of demand is about - 0.20, obtained from the three-stage least squares regression, and suggests that overall student enrollment demand is price insensitive. Note, however (from table 8.5) that student demand for two-year enrollment has a substantially higher price elasticity, one that is consistent with several earlier studies. (Obviously, four-year enrollment is very insensitive to price; the elasticity is - 0.04.) Finally, note that overall public enrollment is a substitute for private enrollment, since the sign on the private tuition variable is positive. (The cross-price elasticity, however, is quite low) Many of the other variables in the demand equation had the expected effect on enrollment demand. Higher per student expenditures and quality both increase student demand substantially. Consistent with the view that public and private education are substitutable to some extent, we find that higher private expenditures, tuition held constant, are associated with lower public enrollments. 34 31. Clotfelter (1976) finds a negative relationship between mobility and per capita public expenditures. He measures mobility by the probability that a recent graduate living in a given state will move out of the state during at least one five-year age period during his or her working lifetime. 32. The residual correlation between demand and supply equations was 0.65. 33. The correlation was -0.38 in the demand system and -0.53 in the supply system. 34. A measure of the quality of private institutions had essentially no effect on enrollment demand and was dropped from the regression.

265

Public Choices in Public Higher Education Definition of Variables in Regression Analyses

Table 8.2

Endogenous variables:

ENB ENB4 ENB2 EXB EXPLAPUB EXP2YR EXP2YRB TCLAPRI TCLAPUB T2YRPRI TZYRPUB TUITB

Public enrollment per capita in 1984 Four-year enrollment per capita in 1984 Two-year enrollment per capita in 1984 Public expense per capita, 1984 Comprehensive and liberal arts: total expense per FTE student, 1988 (median), public Two-year: total expense per FTE student, 1988 (median), private Two-year: total expense per FTE student, 1988 (median), public Comprehensive and liberal arts tuition, 1988 (median), private Comprehensive and liberal arts tuition, 1988 (median), public Two-year tuition, 1988, private Two-year tuition, 1988, public Average undergraduate tuition and fees, 1985, public

Student demand:

EXPLAPRI EXPZYR EXV QUALB SERV TUITV UNEMP YOUNG

Comprehensive and liberal arts: total expense per FTE student, 1988 (median), private Two-year: total expense per FI’E student, 1988 (median), private Expense per capita, 1984, private Percentage of institutions rated good or excellent in Gourman report Percentage of population employed in service industries Average undergraduate tuition and fees, 1985, private Unemployment rate Percentage of population 18 to 24

Legislative supply:

BLACK ELAST 80 HISP INC MOBIL METRPOP PRICE SUPPB SUPP4 SUPP2

Percentage of population that is black Elasticity of combined income and sales tax liability Percentage of population that is Hispanic Median family income, 1979. Percentage of residents of state in 1980 that were not residents in 1975. Percentage of population in metropolitan areas Geographical price difference State appropriation per FTE public student, 1988-89. State appropriation per FTE four-year public student, 1988 (EXPLAPUB TCLAPUB) State appropriation per FTE two-year public student, 1988 (EXP2YRB T2YRPUB)

Other variables:

ASSC C BACH DOCT EAST MAST MlDW ORDER SOUTH WEST

Percentage of population receiving associate of arts degree Constant Percentage of population receiving bachelor’s degree Percentage of population receiving doctoral degree CT, DE, ME, MD, MA, NH, NJ, NY, PA, RI, VT Percentage of population receiving master’s degree IL, IN, IA, KS, MI, MN, MO, NE, ND, OH, SD, WI Rank order of entrance into statehood AL, AR, FL, GA, KY, LA, MS, NC, SC, TN, VA, WV AK, AZ, CA, CO, HI, ID, MT, NV, NM, OK, OR, TX, UT, WA, WY

Table 8.3

Student Demand for Public Enrollment

Variable

Ordinary Least Squares

C

Three-Stage Least Squares

14.69 (20.71) - 0.0085* (0.0025) 0.078* (0.019) -0.014 (0.01 1) 6.58 (6.27) 0.0012

TUITB EXB EXV QUALB TUITV

24.1 1 (22.54) -0.0076* (0.0038) 0.087* (0.027) -0.0051 (0.01 1) 2.45 (6.51)

O.Oo068 (0.00075) -65.18 ( 194.78) -9.85 (42.83) 98.12* (28.71) .68 5.16

(0.oO080) YOUNG

1.14 ( 162.23)

UNEMP SERV Adjusted R2 Standard error of regression

-3.71 (49.87) 116.08* (32.45) .63 5.03

*Coefficient is more than twice its standard error.

Table 8.4

Legislative Supply of Public Enrollment ~

Variable C TUITB INC

Ordinary Least Squares

ELASTI0 PRICE MOBIL BLACK HISP METRPOP SUPPB Adjusted R2 Standard error of regression

Three-Stage Least Squares

- 9.35

- 6.67

(17.47) - 0.0097* (0.0025) 0.0020*

( 16.74) -0.011* (0.0038) 0.0018*

(O.Ooo48) YOUNG

~~~

261.19* (1 16.73) 8.17* (2.63) -0.20 (0.14) 0.13 (0.14) - 3.67 (10.16) 21.59 (13.45) 2.87 (4.12) -0.0023* (0.00091) 0.58 5.28

*Coefficient is more than twice its standard error.

(O.OOO44) 211.56* (109.43) 7.22* (2.51) -0.20 (0.14) 0.13 (0.13) -4.74 (9.17) 9.89 (13.16) 1.55 (3.47) -0.0014 (0.00094) 0.53 5.67

267

Public Choices in Public Higher Education

Table 8.5

Student Demand for W o - and Four-Year Enrollment

Variable

Ro-Year

C TCLAPUB TZYRPUB EXPLAPUB EXPZYRB TCLAPRI TZYRPRI EXPLAPRI EXP2YR YOUNG QUALB UNEMP SERV Adjusted R* Standard error of regression

- 16.23 (23.50) 0.00082 (0.0023) -0.0050* (0.0024) 0.0026* (0.00088) - 0.0024 (0.0014) o.Ooo40 (0.0013) -0.0012 (0.00094) 0.00078 (0.0015) 0.00062 (0.00069) 115.44 (158.33) 17.64* (6.57) 20.31 (59.04) 87.97 (1 07.44) 0.34 6.01

Four-Year

- 15.10 (21.66) -0.0012 (0.0021) -0.00097 (0.0022) -0.0018* (0.00082) 0.0026 (0.0013) -0.0013 (0.0012) -0.00081 (0.00086) -0.00047 (0.0014) 0.00070

(O.OOO64) 260.89 (145.91) - 4.24 (6.06) 32.39 (54.41) 106.24 (99.02) 0.021 5.54

*Coefficient is more than twice its standard error.

We find, as expected, that the number of service jobs per capita is positively related to enrollments. Surprisingly, the sign on the unemployment rate variable is negative. This could reflect the fact that individuals cannot afford the cost of education. More likely, however, this is the result of averaging, since the unemployment rate is positively related to enrollment rates in both the two-year and the four-year enrollment regressions of table 8.5. Finally, the percentage of the population that is age 18 to 24 is positively correlated with demand in the overall ordinary least squares regressions and in the two-year and four-year regressions. The sign change in the two-stage least squares regression is puzzling, although it presumably results from the correlation of the unemployment variable with several of the instruments used (e.g., METRPOP). 8.4.2

Legislative Supply

In the two-stage least squares regression, legislative supply is negatively related to the level of public tuition per student, as well as to the level of

268

John M. Quigley and Daniel L. Rubinfeld

Table 8.6

Legislative Suply: lko- and Four-Year Enrollment

Variable

Two-Year

Four-Year

~~~

C TCLAPUB T2YRPUB TCLAPRI T2YRPRI INC YOUNG ELASTIO PRICE MOBIL BLACK HISP METRPOP SUPP2

18.00 (22.26) 0.0031 (0.0019) 0.0063* (0.0019) 0.00024 (0.00076)

- 0.001 1 (0.00055) 0.00146* (0.00049) -93.04 (1 30.02) 7.09 (3.01) -0.39* (0.18) -0.23 (0.16) 29.22* (1 1.42) 86.29* (1 9.48) 6.76 (4.36) -0,00044 (0.00095)

0.0021 (0.0015)

- O.O0044* (0.0014) -0.0010 (0.00056) -0.00025 (0.00042) 0.00096* (0.00037) 312.41 (96.43) 1.02 (2.07) -0.19 (0.13) 0.14 (0.12) - 26.53 (8.55) -23.31* ( 15.09) -5.58 (3.38)

- 0.00035

SUPP4 Adjusted Rz Standard error of regression

-5.12 ( 16.90)

0.50 5.10

(0.00044) 0.50 3.82

*Coefficient is more than twice its standard error.

appropriation per student. The former result is difficult to interpret; given the difficulties of identification, it could merely reflect the negative relationship between student demand and tuition. In any case, it is consistent with a pattern in which states that choose to support public education by providing for high enrollments also support education by offering relatively low tuitions. The latter result suggests that legislatures recognize the direct trade-off between offering a low subsidy to a large number of students and offering a higher subsidy to a smaller group. The elasticity of -0.15 is similar in magnitude to the elasticity of student demand, suggesting that legislatures and students are only mildly cost-sensitive. We also find that higher-income states offer substantially higher enrollments, as do states whose populations tend to be centered in metropolitan

269

Public Choices in Public Higher Education

areas and that have substantial Hispanic populations (the Southwest). However, states with substantial black populations (including the South) tend to offer lower levels of public e n r ~ l l m e n t . ~ ~ Finally, the elasticity of the tax base is, as expected, a positive and highly significant determinant of enrollment rates, a result that is consistent with the view that states’ budgeting decisions are sensitive to the political ease with which taxes can be raised. (PRICE and YOUNG also have the expected effects.) When we look at the two-yeadfour-year breakdown of enrollments from the legislative point of view, we find that two-year enrollment rates are very sensitive to tuition rates (at two-year schools), while four-year enrollments are much less so (in relation to four-year college tuition).36Other distinctions worth noting include the effect of mobility (the more mobile the population, the lower the two-year enrollments, other things equal) and the state appropriation (a positive but very small correlation between appropriations and four-year enrollments). The most important difference between the two- and four-year enrollment equations lies with the race variables. Both BLACK and HISP (and also a related variable, METRPOP) have substantial positive effects on two-year enrollment rates. This may reflect the fact that black and Hispanic populations have formed effective interest groups in terms of achieving access to public higher education in a number of states; the result is consistent with the analysis of Grubb (1988). The result is surprising; however, it does not seem spurious-there is a negative but small simple correlation between the percentage of blacks in a state and the per capita enrollment in two-year institutions. Also surprising is the substantial negative relationship between BLACK and HISP and four-year enrollments. The result for blacks may reflect to a substantial degree the fact that black populations are highest in the South, where four-year enrollments are low. However, the strong relationship between the Hispanic population and four-year enrollments is surprising to us, since the overall simple correlation between the two variables is essentially zero.

8.4.3 Expenditures As a final exercise, we attempted to explain the statewide variation in expenditures on public higher education. The first column in table 8.7 describes overall per capita expenditures, while the second relates to per student expenditures on two-year and four-year colleges. In addition to some of the con35. When we allowed for a direct interaction between legislative supply and private-sector alternatives by including ENV as an explanatory variable, we found the coefficient to be negative and marginally significant in the ordinary least squares regression, and negative and insignificant when two-stage least squares was used. 36. The two- and four-year equation system was estimated using seemingly unrelated regression. The cross-equation residual correlation was -0.3 1.

270

John M. Quigley and Daniel L. Rubinfeld

Table 8.7

Public Per Capita Expenditures

~

~

Variable

All Higher Education

Two-Year and Four-Year College

2.09 ( I .90)

2.92 (1.69)

C (two-year)

3.24

C (four-year) 0.088 (0.15)

LSUPPB BACH

-0.020 (0.040)

ASSC MAST DOCT

0.016 (0.051) - 0.058 (0.12) 0.29 (0.65) 0.31 (0.38)

LINC YOUNG ELAST8O LPRICE

MOBIL

(1.68) 0.26* (0.11) 0.10 (0.028) - 0.10* (0.044)

-0.19 (0.11) 0.18 (0.58) 0.21 (0.24)

17.98* (4.50) 0.25* (0.095) -0.97* (0.4)

0.21* (0.069) 0.141 (0.37)

-0.0030 (0.0040)

-0.0033 (0.0035)

LENB

7.71* (3.00)

0.41 (0.32)

-0.17* (0.042)

LEN2/LEN4 Adjusted R2 Standard error of regression

0.71 0.15

0.20

*Coefficient is more than twice its standard error.

trol variables that we used in the supply-demand equations, we have included five additional variables. The first is an endogenous variable, the enrollment rate, which reflects the effect of size on per capita expenditures: a negative elasticity would suggest that there are scale economies associated with the provision of higher education, whereas a positive elasticity could reflect the additional scope of programs associated with larger enrollments. The remaining four variables are included to reflect the degree mix of the students attending public institutions. The equations in table 8.7 were estimated using two-stage least squares and three-stage least squares, respectively. A number of variables, shown with a prefix L, were introduced in logarithmic form to allow for the direct estimation of elasticities. For the two-year and four-year schools, the most important

271

Public Choices in Public Higher Education

program-mix variable was the percentage of students receiving associate of arts degrees, the usual two-year degree offered. In addition, the percentage receiving master’s and doctor’s degrees had the expected negative effect on spending; surprisingly, however, there was no relationship between per student expenditure and the percentage of bachelor’s degrees offered. It is not unexpected to find a strong positive relationship between state appropriations and expenditures, nor perhaps to find a similar relationship between the elasticity of the tax base and spending (the coefficient of ELAST80 was high in both expenditure equations). It is more surprising, however, to find a relatively low income elasticity of demand for higher education (expenditures), although our elasticity of .31 overall is higher than that found by Clotfelter (1976). Note also that the price variable had the expected negative effect in the overall equation, but not when the model was restricted to two-year schools and four-year colleges. The migration variable had a negative coefficient in both equations, consistent with the view that a more mobile population leads legislatures to spend less money per pupil on public higher education, other things equal. 8.5

Concluding Remarks

Our analysis of the statewide patterns of public enrollments and expenditures has emphasized the close link between the public and private sectors. Because private higher education was dominant in the first 100 years of our history, public higher education developed in its shadow. As a result, public higher education enrollments and spending have been highest primarily largely in the West and Midwest, where private educational opportunities have historically been limited. When seen from a cross-sectional point of view, this historical pattern shows up as a negative relationship between public and private enrollment rates. A group of primarily eastern states offer a high private enrollment (often high-quality), low public enrollment bundle of higher education opportunities, while a substantial group of primarily western and midwestern states offer high public (often high-quality), low private enrollment rates. The important exception to this general rule is the South, where most states offer low public, low private (and generally lower-quality) enrollment bundles. What light do our empirical results shed on the alternative theories of public choice that we sketched out at the beginning of the paper? The human capital and mobility approach is supported, but only weakly, by the negative coefficient on the mobility variable that we obtained in the two-year institutions equation (table 8.6) and by the negative coefficients in the per capita expenditure equations (table 8.7). There is also some support for the second public-choice explanation, in which politicians use logrolling to trade for forms of public higher education

272

John M. Quigley and Daniel L. Rubinfeld

that benefit their constituencies. The growth of two-year institutions is consistent with this view; logrolling may have led to the creation of a large number of new institutions in a sufficiently large number of legislative districts so as to make the entire educational package politically viable. Note, however, that there is only very mild empirical support for the Leviathan perspective on the growth of government; in the legislative supply equation, we find higher enrollments and higher expenditures to be positively related to the elasticity of the tax base. The growing emphasis on two-year institutions, especially in the West, is also consistent with the view that public education is a means of distributing state resources to the middle class. But this theory does not explain why such institutions have been so successful in other areas of the country. A clean, convincing test of these and other theories that explain public choices in public higher education awaits further research. We hope that this paper has helped to mark the way.

References Becker, G. S. 1983. A theory of competition among pressure groups for political influence. Quarterly Journal of Economics 98:37 1-400. Bishop, J. 1977. The effect of public policies on the demand for higher education. Journal of Human Resources 12:283-307. Bowles, S., and H. Gintis. 1976. Schooling in capitalist America: Educational reform and the contradictions of economic life. New York: Basic Books. Campbell, R., and B. Siege]. 1976. The demand for higher education in the United States, 1919-1964. American Economic Review 57:482-94. Christensen, S . , J. Melder, and B. Weisbrod. 1975. Factors affecting college attendance. Journal of Human Resources 10: 174-88. Clotfelter, C. T. 1976. Public spending for higher education: An empirical test of two hypotheses. Public Finance 31:177-95. Clotfelter, C. T. 1991. Demand for undergraduate education. Manuscript. Corazzini, A., D. Dugan, and H. Grabowski. 1972. Determinants and distributional aspects of enrollment in U.S. education. Journal of Human Resources 7:39-59. Corman, H. 1983. Postsecondary education enrollment responses by recent high school graduates and older adults. Journal of Human Resources 17:247-67. Corman, H., and P. Davidson. 1984. Economic aspects of post-secondary schooling decisions. Economics Education Review 3: 131-39. Douglas, G. K. 1977. Economic returns on investments in higher education. In Investment in learning: The individual and social value of American higher education, ed. H. R. Bowen. San Francisco: Jossey-Bass. Feldman, P., and S. Hoenack. 1969. Private demand for higher education in the United States. In The economicJinancing of higher education in the U.S., 375-95. Washington, D.C.: U.S. Congress Joint Economic Committee. Galper, H., and R. Dunn. 1969. A short-run demand function for higher education in the United States. Journal of Political Economics 77:765-77. Garvin, D. A. 1980. The economics of university behavior. New York: Academic Press. Gourman, J. 1987. The Gourman report: A rating of undergraduate programs in Amer-

273

Public Choices in Public Higher Education

ican and international universities. 7th ed. Los Angeles: National Education Standards. Grubb, W. Norton. 1988. Vocationalizing higher education: The causes of enrollment and completion in public two-year colleges, 1970-1980. Economics of Education Review 7:301-19. Hoenack, S. A., and D. J. Pierro. 1990. An econometric model of a public university's income and enrollments. Journal of Economic Behavior and Organization 14(3): 1-2 1 . Hoenack, S . A., and W. C. Weiler. 1979. The demaiid for higher education and institutional enrollment forecasting. Economic Inquiry 17:89-113. Hofstadter, R., and W. Smith, eds. 1961. American higher education: A documentary history. Vols. 1 and 2. Chicago: University of Chicago Press. Hopkins, T. 1974. Higher education enrollment demand. Economic Inquiry 1253-65. Jackson, G. A,, and G. B. Weathersby. 1975. Individual demand for higher education: A review and analysis of recent empirical studies. Journal of Higher Education 46~623-52. Johnson, J. M. 1980. Resident and nonresident undergrad and graduate tuition andlor required fees, public universities, colleges and state universities, and community colleges. Olympia: Washington State Council for Postsecondary Education. Karabel, J. 1974. Protecting the portals: Class and the community college. Social Policy 5(1):12-19. Kelly, R. L. 1940. The American colleges and the social order. New York: Macmillan. Koepplin, L. W., and D. A. Wilson. 1985. The future of state universities. New Brunswick, N.J.: Rutgers University Press. Kohn, M., C. Manski, and D. Mundel. 1976. An empirical investigation of factors which influence college-going behavior. Annals of Economic and Social Measurement 5(4):391-419. Leslie, L., and G. Ramey. 1986. State appropriations and enrollments. Journal of Higher Education 57: 1-19. Manski, C., and D. Wise. 1983. College choice in America. Cambridge, Mass.: Harvard University Press. Mattila, J. P. 1982. Determinants of male school enrollments: A time-series analysis. Review of Economic Statistics 6:242-5 1. McPherson, M. S. 1978. The demand for higher education. In Publicpolicy andprivate higher education. ed. D. W. Breneman and C. E. Finn, Jr, Washington, D.C.: Brookings Institution. Nerlove, M. 1972. On tuition and the costs of higher education: Prolegomena to a conceptual framework. Journal of Political Economy 80:5178-5218. Peltzman, S . 1973. The effect of government subsidies-in-kind on private expenditures: The case of higher education. Journal of Political Economy 81. Radner, R., and L. S. Miller. 1975. Demand and supply in US. higher education. New York: McGraw-Hill. Solmon, L. C. 1973. The definition and impact of college quality. In Does College Matter? ed. L. C . Solman and P. J. Taubman. New York: Academic Press. Sulock, J. 1982. The demand for community college education. Economics of Education Review 2:351-61. Taussig, M. 1987. Educational quality, access, and tuition policy at state universities. Journal of Higher Education 58:2 15-37. Thwing, C. F. 1906. A history of higher education in America. New York: D. Appleton. Weiss, Y. 1972. The risk element in occupational and educational choices. Journal of Political Economics 80:1203-1 3. Westmeyer, P. 1985. A history of American higher education. Springfield, 111.: Charles C Thomas.

274

John M. Quigley and Daniel L. Rubinfeld

Appendix Sources and DeJinitions of Variables Variable

DefinitiodExplanation

ASSC

Percentage of population receiving associate of arts degreehumber of earned associate degrees As ASSC, but bachelor’s degree As ASSC, but doctoral degrees Northeastern regionConnecticut , Delaware, Maine, Maryland, Massachusetts, New Hampshire, New Jersey, New York, Pennsylvania, Rhode Island, Vermont Elasticity of combined income and sales tax liability/income elasticity of the sum of personal income and sales tax liabilities Enrollment in public institutions per capita, 1984lstudents whose programs of study are creditable toward a bachelor’s or higher degree and also undergrads in one-, two-, three-, or four-year occupational programs which are not chiefly creditable toward a bachelor’s degree, per capita Enrollment in four-year public institutions, per capita, fall 1984 Enrollment in two-year public institutions, per capita, fall 1984 Order of statehood Enrollment in private institutions per capita, 1984/as ENB Enrollment in four-year private institutions, per capita, fall 1984 Enrollment in two-year private institutions, per capita, fall 1984 Total expenditures per FTE student, 1988, in public institutions/median of reported institutions As EXPLAPUB, but private

BACH DOCT EAST

ELASTSO

ENB

ENB4 ENB2 ENTER0 ENV ENV4 ENV2 EXB

EXPLAPRI

*See listing below for numbered list of data sources.

Source* 18

18 18 8

14

1

1 1

1 1

1 1 3

17

275

Public Choices in Public Higher Education

EXPLAPUB EXP2YR

EXP2YRB EXV INC MAST METROPOP

MIDW

MOBIL

PRICE

POV QUALB

QUALV SERV

SOUTH

Total expenditures per FTE student, 1988, in comprehensive and liberal arts public institutiondmedian of reported institutions Total expenditures per FTE student, 1988, in two-year private institutiondmedian of reported institutions As EXP2YR, but public As EXB, but private Median family income, 1979/median money income of families As ASSC, but master’s degrees Percentage of population in metropolitan areadpercentage in 1 of 261 metropolitan statistical areas and 20 consolidated metropolitan statistical areas Central region/Illinois, Indiana, Iowa, Kansas, Michigan, Minnesota, Missouri, Nebraska, North Dakota, Ohio, South Dakota, Wisconsin Percentage of residents of state in 1980 that were not residents in 1975/only includes residents over five years of age in 1980 Geographical price differences/cost of government index, 1988: prices and wages that would be paid for a fixed market basket of public services

16

Poverty normalized/percent of children 1-1 8 years old below poverty line, 1980 Proportion of high-quality public institutions/ percentage of Gourman-rated public institutions rated strong or good, rating based on size, quality of faculty, depth and breadth of curriculum, athletics, etc. Proportion of high-quality private institutions/ as QUALB Percentage of population employed in service industries, 1980/ratio of service employees to population Southeastern regiodAlabama, Arkansas, Florida, Georgia, Kentucky, Louisiana, Missis-

15

17

16 3 9 18 10

8

9

15

5

5 12

8

276

John M. Quigley and Daniel L. Rubinfeld ~~

SUPP SUPPB SUPP4

SUPP2 TCLAPRI TCLAPUB

T2YRPRI T2YRPUB TUITB

TUITNR

TUITR

TUITV

UNEMP WEST

YOUNG

sippi, North Carolina, South Carolina, Tennessee, Virginia, West Virginia State appropriation per FTE private student, 1988-89 State Appropriation per R E public student, 1988-89 State appropriation per FTE comprehensive or liberal arts student, 1988 (EXPLAPUB TCLAPUB) State appropriation per FTE two-year public student, 1988 (EXP2YRB - T2YRPUB) As TCLAPUB, but private Tuition in comprehensive and liberal arts public institutions, 198Wmedian of reported institutions As T2YRPUB, but private Tuition in two-year public institutions, 19881 median of reported institutions Average undergraduate tuition and fees, public, 1985/mean tuition and fees by students enrolled Tuition of nonresidents, 1979-80/average tuition of nonresidents as undergraduates in public institutions Tuition of residents, 1979-80/average tuition of state residents as undergraduates in public institutions Average undergraduate tuition and fees, private, 1985/mean tuition and fees by students enrolled Proportion unemployed, 1980/percentageof labor force not employed Western regiodAlaska, Arizona, California, Colorado, Hawaii, Idaho, Montana, Nevada, New Mexico, Oklahoma, Oregon, Texas, Utah, Washington, Wyoming Proportion of college-age population/percentage of population in 1980 and 18- to 24-year age range

15 15 16

16 17 16

17 16

3

20

20

3

11 8

9

277

Public Choices in Public Higher Education

Data Sources 1. U.S. Department of Education, National Center for Education Statistics. Fall enrollment in higher education. 2. U.S. Department of Education, National Center for Education Statistics. Survey of residence and migration of college students. 3. U.S. Department of Education, National Center for Education Statistics. Financial statistics of institutions of higher education. 4. U.S. Department of Education, National Center for Education Statistics. Education directory: Colleges and universities. Special tabulation. 5. The Gourman Report. 3rd ed. 1987. Monograph. 6. Education Commission of the States. State postsecondary education structures handbook 1986: State coordinating and governing boards, no. PS-85-1. 7. U.S. Department of Education, National Center for Education Statistics. Revenues and expenditures for public elementary and secondary education (annual). Statistics of public elementary and secondary day schools (annual). 8. National Education Association. Estimates of school statistics (annual) (copyright) and unpublished data. Washington, D.C. 9. U.S. Bureau of the Census. Current Population Reports, series P-25. 10. U.S. Bureau of the Census. Press release (CB 86-1 18) and unpublished data. 11. U.S. Department of Labor, Bureau of Labor Statistics. Geographic profile of employment and unemployment. 12. U.S. Department of Labor, Bureau of Labor Statistics. Employment and earnings. 13. Advisory Commission on Intergovernmental Relations. Tax capacity of the states. 14. National Bureau of Economic Research. State personal income and sales taxes, 1977-1983. Daniel R. Feenberg and Harvey S. Rosen. 15. Research Associates of Washington. State profiles: Financing public higher education 1978 to 1990. 16. Research Associates of Washington. Higher education revenues and expenditures, vol. 1: Public Institutions. 17. Research Associates of Washington. Higher education revenues and expenditures, vol. 2: Private Institutions. 18. U. S. Department of Education, National Center for Education Statistics. Degrees and other formal awards conferred surveys. 19. Paul Westmeyer. 1985. A history of American higher education. Springfield, Ill.: Charles C Thomas. 20. Jackie M. Johnson. 1980. Resident and nonresident undergrade and graduate tuition andlor required fees, public universities, colleges and state

278

John M. Quigley and Daniel L. Rubinfeld

universities, and community colleges. Olympia: Washington State Council for Postsecondary Education.

Comment

Helen F. Ladd

Spending on public higher education represents a substantial component of many state budgets. In 1989, spending on higher education accounted for 20 percent of total direct general spending by states and exceeded 25 percent in 11 states. Given its significance for state budgets, state legislators must continually grapple with issues such as how much public higher education to provide, of what kinds, at what quality, and at what price to students. John Quigley and Daniel Rubinfeld vividly describe the tremendous variation in state policies toward higher education. They show, for example, that states in the West typically provide extensive access to public colleges and universities (as measured by enrollments per 1,000population) while states in the East typically rely less on public provision in favor of private provision. Consistent with this finding is a negative relationship between public and private enrollments. Tuitions for public universities as a share of public spending vary across states from a low of 6.3 percent in California to a high of 63.6 percent in Pennsylvania, and the quality of public education varies from very high (according to one source) in states such as Arizona, California, and Iowa to low in many southern states. The mix between two- and four-year colleges also varies across states, with a generally negative relationship between twoyear and four-year enrollment rates. This variation raises a variety of interesting questions about the “legislative supply decision,” which is the central focus of the Quigley and Rubinfeld (henceforth QR) paper. In examining public choices in public higher education, the authors have taken an exploratory approach. They have intentionally avoided the traditional format of presenting and then testing a particular theory. The result is a paper with a tremendous wealth of information, a model that tries to sort out the supply-side determinants from the demand determinants of enrollments, and a few reflections about alternative theoretical views that might be consistent with the enrollment patterns they observe. I have no difficulty with the authors’ decision not to test specific theories (the theories they discuss in section 8.2 are all quite general and hard to test with any precision). However, as I discuss below, the paper’s lack of focus leaves room for additional speculation about the underlying objective function of state legislatures and research focused on specific policy-related questions. Helen F. Ladd is a professor of public policy studies and economics at Duke University.

279

Public Choices in Public Higher Education

Descriptive Results Three major conclusions emerge from the descriptive section of the paper. The first has already been mentioned, namely, the tremendous variation in policies across states. The second is that public higher education differs quite notably from that controlled by the private sector. Major differences include the significantly lower tuitions at public universities and the large presence of community, or two-year, colleges. (Another major difference not discussed by QR is the higher proportion of in-state students. In my home state of North Carolina, for example, the legislature limits the percentage of out-of-state students to 8 percent, while the proportion of such students at Duke University, a private institution, is about 85 percent). These differences suggest that public and private higher education are not perfect substitutes and that, as we try to model the legislative supply decision, we need to think quite carefully about the nature of the service being provided. The third conclusion to emerge from the descriptive analysis is that decisions about higher education are heavily path dependent and reflect cultural and historical factors that are hard to model. The observation that private colleges and universities emerged before public institutions of higher education makes it unsurprising that older states rely more heavily on private than on public education and that the reverse is true for younger states. Somewhat surprisingly, QR seem to ignore this key role of history in the model they present in the following section of the paper. The notion that 1985 levels of variables such as income, employment, tuition, and unemployment are the key determinants of 1985 levels of public enrollments seems inconsistent with the story they tell in the first section. Consequently, I would be much more comfortable with a panel data set that would allow them to focus on how changes in the explanatory variables affect changes in enrollments or other variables during, say, a 20- or 25-year period. Assembling a panel data set of this type, however, would be a formidable task.

The QR Model One of the authors’ contributions is to shift the primary focus away from expenditures onto a more appropriate measure of the supply of higher education, namely enrollments. Although enrollments are a measure of inputs, QR argue that they are a reasonable proxy for output. Even this measure, however, is not without its problems. Presumably, the legislature does not control enrollments directly. Instead it controls variables such as admissions policy (e.g., admission to all applicants meeting some minimal standard), the amount of resources in the educational system, and tuitions which, together with student demand, determine enrollments. Nonetheless, the use of enrollments can be viewed as a reasonable first step in understanding public choices about higher education.

280

John M. Quigley and Daniel L. Rubinfeld

The starting point for QR's demand equation appears, quite sensibly,' to be a model of investment in human capital. That is, students weigh the costs of attending college against the benefits. Costs are measured by tuition and foregone opportunities as proxied by the unemployment rate. However, the unemployment rate could be a proxy for other variables as well and actually enters some of the estimated equations with an unexpected positive sign. Benefits are measured by an index of quality, expenditures per pupil, and jobs in the service sector. The low skill requirements of many service jobs make me question how well this variable proxies job opportunities for college graduates, but at least it enters the equation with the expected positive sign. In addition, the equation appropriately includes the cost of alternatives and the proportion of the population in the relevant age range. Alternatively, we could view higher education as a consumer good. According to this approach, family income should be in the demand equation. Similarly, the racial mix of the population may also belong in the equation. I suspect that these variables were excluded from the demand equation purely for purposes of identification. The specification of the supply side also raises questions, some related to included variables and others to important variables that are missing. The equation correctly includes the income of state residents and a measure of their mobility, but both deserve further refinement. Because students from the highest-income families probably choose private colleges and those from the lowest-income households often do not attend college, the demand for public institutions most likely emanates from the middle class. This observation suggests that the equation should include measures of the distribution, as well as the level, of income. With respect to the mobility variable (measured as the proportion of residents living in the state less than five years), the authors point out that the sign should be negative unless it picks up the effects of population growth, in which case it should be positive. The obvious question then becomes, why not control for growth directly in order to sort out the two effects? Missing from the supply equation is the education level of the state population. Presumably, the proportion of the population with college degrees could be an important indicator of tastes for public services. Another variable missing from the equation is the supply of private-sector alternatives. The omission of this variable is surprising, given the centrality of the public versus private trade-off that emerges from the descriptive analysis. The presence of private-tuition variables in the demand and some of the supply equations helps, but I am not convinced that the tuition variable alone captures the full effect of private universities on public enrollments. Three main results emerge from the model. The first is the low price elasticity of demand for higher public education. According to the model, this elasticity is - .20 overall and close to 0 for four-year colleges. These estimates seem low relative to the consensus from other studies of about - .70. In addition, the extreme inelasticity for four-year colleges poses a puzzle of why

281

Public Choices in Public Higher Education

states do not raise tuitions. A second result is the role of the elasticity of the tax base. I must admit that I am suspicious of the causal link between the elasticity of the tax base and enrollments, since it runs directly counter to recent work by Feenberg and Rosen (1987), whose careful study based on panel data provides no support for the hypothesis that the elasticity of the tax base affects state spending. I suspect that the QR finding reflects a simultaneity problem: states that provide above-average supplies of public higher education may happen to be those with elastic tax structures.' Finally, the race variables affect two-year and four-year enrollments quite differently. The greater the proportion of blacks and Hispanics, the greater the two-year enrollment but the lower the four-year enrollment. Why minority groups would push for two-year but oppose four-year institutions is surprising and deserves further analysis.

What Might We Learn from Additional Work on this Topic? Additional work on the supply of public higher education could take two different approaches. The first would be to start with a more explicit objective function. Three possible objective functions come to mind. First, in order to promote state economic development, legislators could be trying to maximize the enrollment of state college-age residents. If this were the goal, they would be indifferent between private and public enrollments. Alternatively, in order to garner voter support, legislators may be interested in maximizing enrollment in public institutions. Through the vehicle of public education, they transfer resources to many middle-class residents in the form of in-kind subsidies. Finally, state legislators may be motivated by the desire to maximize political support from employees of state institutions. Working through the implications of each of these objective functions might well lead to equations that can be more precisely specified than those of QR and that yield specific testable hypotheses. Second, additional work could focus on some of the policy-related questions raised by the wide variation across states in the provision of public higher education. The first set concerns the relationship between public and private colleges and universities. The second set concerns tuition policies, and the third concerns the mix of public enrollments between two-year and fouryear institutions. QR provide descriptive information on these questions, but their model provides little explanation of the differences across states. Substitutability between Public and Private Options Why do some states provide extensive systems of public higher education, while others do not? The simple answer suggested by QR's descriptive analy1. QR assert that the finding remains when they make the elasticity variable endogenous, but further exploration of this finding is warranted.

282

John M. Quigley and Daniel L. Rubinfeld

sis is that states will provide more public education the fewer are the privatesector opportunities in the state. For example, the New England states, which have many high-quality private universities and colleges, have chosen to provide limited amounts of higher education through public institutions, while the reverse is true in many western states. But this answer is insufficient since it does not explain how states with private institutions are able to resist the pressure for public education. As illustrated by QR, private education is not the same as public education, especially in terms of tuition. To think about the effect of private institutions on public institutions, consider how legislators in South Carolina might respond if a wealthy donor established a new, highquality private university in that state. Given the differences between public and private education, it is entirely plausible that state legislators might not change their support of public higher education at all. Thus, the first interesting question is the extent to which the presence of private institutions affects the provision of public institutions. Despite the puzzles about public-private substitutability that emerge from the descriptive part of the paper, QR hardly address this issue in the modeling portion of the paper. As noted earlier, neither the quantity nor the quality of private-sector options is included in the public-sector supply equation. Consequently, the QR model can provide no insights about how legislators adjust their support for public colleges in response to the differences in private institutions. Additional work in this area would be useful. A related question concerns the responsiveness of private enrollments to changes in the quality and quantity of public opportunities. The current state budget crunch has forced many states to reduce appropriations for public higher education and to raise tuitions. To the extent that public and private education are substitutes, one would predict that the deterioration in the public-sector options induced by the budget crunch would lead to increased demand for private-sector options. To determine the magnitude of this response, the level of enrollments in the private sector would need to be modeled explicitly as a function of, among other variables, the quality, quantity, and price of public-sector options. Tuition policy Why do tuitions at public universities vary across states? Why, for example, does North Carolina charge low tuitions and Michigan high tuitions? The difference in the average income of residents in the two states probably provides a substantial part of the answer. However, even if income differences account for a large portion of the cross-state variation in tuition policy, one would still want to explore the reasons why income matters. What goals are legislators trying to achieve in providing low-price public colleges and universities? Are they keeping the price low to transfer resources to middle-income taxpayers or because they believe low tuitions will foster economic growth? Taken at its face value, the QR conclusion that the price elasticity of demand for four-year

283

Public Choices in Public Higher Education

colleges is zero suggests that maintaining low tuitions at four-year colleges has no effect on enrollments and consequently may have little impact on the supply of human capital in the state. Additional insights about goals might also emerge from a close examination of the ratio of in-state to out-of-state tuitions and about admissions policies. In any case, there is room here for more research. In addition to understanding more about the variation in tuition levels across states, it would be desirable to learn more about the political or budgetary pressures that induce states to increase tuition at public universities. It would also be interesting to find out how, if at all, the structure of tuitions is changed as tuitions are raised. A comparative analysis of changes in tuition policies during both the 1990-91 recession and the 1981-82 recession might produce some interesting insights into the political economy of tuition changes. Composition of Public Systems Finally, it would be worthwhile to try to understand more about the composition of public systems. QR describe the recent dramatic growth in twoyear colleges. Why did they grow faster in some states than in others? Again, a better understanding of the objectives of state policymakers would be helpful. Related to the two-yeadfour-year mix is the more general question of how states allocate funds among the different parts of the university system. How are flagship campuses treated relative to other campuses? Are there swings in support over time for flagship campuses and, if so, why? In this regard, it might be interesting to look at the role graduates of state schools play in the state legislature. More generally, one might investigate the lobbying power of employees in state higher education. Presumably, once a state system is set up, the education establishment provides a natural lobbying group in the form of state employees.

References Feenberg, Daniel, and Harvey Rosen. 1987. Tax structure and public sector growth. Journal of Public Economics 32: 185-201.

This Page Intentionally Left Blank

Contributors

Dominic J. Brewer New York State School of Industrial and Labor Relations Cornell University Ithaca, NY 14851 Stephen V. Cameron Department of Economics University of Chicago 1126 East 59th Street Chicago, IL 60637

Charles T. Clotfelter Sanford Institute of Public Policy Duke University Box 90245, Duke Station Durham. NC 27706 George M. Constantinides Graduate School of Business University of Chicago 1101 East 58th Street Chicago, IL 60637 Philip J. Cook Sanford Institute of Public Policy Duke University Box 90245, Duke Station Durham, NC 27706

285

Ronald G. Ehrenberg New York State School of Industrial and Labor Relations Cornell University Ithaca, NY 14851 Martin Feldstein President and Chief Executive Officer National Bureau of Economic Research 1050 Massachusetts Avenue Cambridge, MA 02138 Robert H. Frank Department of Economics Cornell University Ithaca, NY 14853 Malcolm Getz Department of Economics Vanderbilt University Box 1819, Station B Nashville, TN 37235 Jerry R. Green Department of Economics Littauer Center 309 Harvard University Cambridge, MA 02138 Eric A. Hanushek Department of Economics University of Rochester Rochester, NY 14627

286

Contributors

Robert M. Hauser Institute for Research on Poverty University of Wisconsin 1 180 Observatory Drive Madison, WI 53706

Robert C. Merton Graduate School of Business Harvard University Soldiers FielaMorgan 397 Boston. MA 02163

James J. Heckman Department of Economics University of Chicago 1126 East 59th Street Chicago, IL 60637

John M. Quigley Graduate School of Public Policy 2607 Hearst University of California Berkeley, CA 94720

Charlotte V. Kuh Educational Testing Service Room U-350 Rosedale Road Princeton, NJ 08541

Daniel I. Rees New York State School of Industrial and Labor Relations Cornell University Ithaca. NY 14851

Helen F. Ladd Institute of Policy Sciences and Public Affairs Duke University Box 4875, Duke Station Durham, NC 27706

Michael Rothschild Office of the Divisional Dean, Social Sciences University of California at San Diego 9500 Gilman Drive La Jolla, CA 92093

Charles F. Manski Department of Economics University of Wisconsin Social Sciences Building 1 180 Observatory Drive Madison, WI 53706

Daniel L. Rubinfeld School of Law 225 Boalt Hall University of California Berkeley, CA 94720

Michael S . McPherson Department of Economics Williams College Denison Gatehouse Williamstown. MA 02167

Lawrence J. White Stern School of Business New York University 44 West 4th Street New York, NY 10012

Author Index

Adams, Charles, 185 Alchian, Armen A , , 21 Alexander, Karl, 123n2 Anderson, Douglas K., 64,791116, 101 Arrow, Kenneth J., 124 Arthur, Brian, 129 Astin, Alexander W., 123n2, 134 Atkinson, Richard C., 183

Caminiti, Susan, 122-23 Carter, Deborah J., 64 Carter, Shani, 62 Charlier, Marj, 231114 Chayowski, Richard P., 185n2 Cho, Joo Hyun, 791117 Christensen, S., 246 Clotfelter, Charles, 1, 2nn3,4, 4, 132, 246n3, 263n30, 264113 1,27 1 Cohn, Elchanan, 16x15 Coleman, James S., 2n2 College Board, 132t Collins, Eileen L., 2n3, l l n l Conrad, Clifton F., 129 Constantinides, George, 212, 218 Corazzini, A., 246 Corman, Hope, 24, 246, 262n26 Cotter, William R., 23 Crawford, Vincent P., 18

Bandura, A., 47 Bank, B., 47 Becker, William B., 12n2 Beneman, David, 206-7 Bennett, William, 2 Bergman, Yakov, 218 Bemdt, T., 46 Biddle, B., 47 Bishop, John H., 62 Black, Fischer, 211,212 Blackbum, Robert T., 129 Bodie, Zvi, 216, 217 Bok, Derek, 12 Borus, Michael, 185 Bowen, Howard R., 12, 145 Bowen, William G., 2, 145, 183, I84nl Bowles, S., 244 Breeden, Douglas T., 221, 225 Brinkman, Paul T., 12n2, 212 Brovender, Shlomo, 212

Davidson, P., 246n4, 262n26 Detemple , Jerome B ., 2 18 Donohue, John, 113-14n4 Douglas, G. K., 244112 Duffie, Darrell, 218 Dugan, D., 246 Duncan, Otis Dudley, 79, 88 Dunn, R., 246

Camerer, C . , 47n2 Cameron, Steven, 5,65n3,67n7, 83n19, 951128.991133, 105, 107, 110-llt,113, 114

Ehrenberg, Ronald, 24,26, 28n26, 183, 184, 185n2, 205 Eisner, Robert, 21 1 Ekstrom, Ruth, 182

287

288

Author Index

Ennis, Richard, 21 1 Epstein, Larry, 218

Hutchins. Robert Maynard, 3 Hyman, H., 47

Fallows, James, 6 Fama, Eugene, 21 1, 240 Faulhaber, Gerald R., 12, 14 Featherman, David L., 72, 79, 81, 92n26 Feenberg, Daniel, 277, 281 Feldman, P., 246 Fernandez, L., 145111 Finn, Chester, 2 Fischer, Stanley, 221 Frase, Mary J., 73 Fraumeni, Barbara, 221112 Freeman, R., 43.47, 52n5 Froomkin, Joseph T., 2n3 Fuller, Winship C., 24, 26, 129

Ingersoll, Jonathan E., 240 James, Estelle, 12, 13-14, 123-24, 205 Jaschik, Scott, 23 Jaynes, Gerald David, 64,68, 102n35 Jensen, Michael C., 211 Johnson, George, 185 Johnson, J. M., 277 Jorgenson, Dale W., 221112 Juster, T., 56 Kahneman, D., 47n2 Kane, Michael, 64 Kane, Thomas J . , 62,64, 71,72, 92n26, 101

Gale, David, 17, 125-26n6 Galper, Harvey, 185n2, 246 Garvin, David A,, 12, 241118 Garvin, Donald, 205 Gintis, H., 244 Goldberger, A., 52 Gourman, 250 Grabowski, H., 246 Gramlich, Edward M., 185n2 Griffin, Larry, 123n2 Griliches, Zvi, 45 Grinold, Richard, 212 Grossman, Sanford, 2 18 Grubb, W. Norton, 250, 269 Hall, Alfred E., 129 Hamermesh, Daniel, 185 Hansen, W. Lee, 14 Hansmann, Henry, 21 1-12, 238 Hartigan, John A,, 20n8 Hauser, Robert M., 62,63, 64,65,67, 68, 71,72, 79, 81, 82, 92n26, 101 Heam, James C., 26, 126 Heckman, James, 5 , 4 5 , 65113, 67n7, 83n19, 95n28,99n33, 105, 107, 110-llt, 112, 113, 114 Hight, Joseph E., 24 Hilton, Thomas, 180 Hindy, Ayman, 218 Hoenack, Stephen A., 2n3, I l n l , 241117, 207, 246, 262, 263n28 Hofstadter, R., 260n23 Hopkins, David, 212 Hopkins, T., 246 Hu, Arthur, 64 Huang, Chi-fu, 218

Karabel, J., 245 Kelso, Alexander S., Jr., 18 Kerr, Clark, 15 Kingston, Paul William, 122, 123, 126 Kohn, M., 246 Koretz, Daniel, 64, 701111 Kramer, John R., 29n27 Krukowski, Jan, 1291111 Kunreuther, H., 47n2 Laroque, Guy, 2 18 Lazear, Edward, 124n5 Leibenstein, Harvey, 34 Leslie, Larry L., 12n2 Lewis, Lionel S., 122, 126 Litten, Larry H., 129, 132, 133nn19,21 Litvack, James, 21 1, 238 Long, John B., 212 Longanecker, David, 26 Lord, Graham, 184111 McClain, David, 241117 McDonald, John, 218 Machlup, Fritz, 56 McPherson, Michael S., 241116, 33n34, 129, 206n1, 207 Malkiel, Burton, 21 1, 238 Manski, Charles F., 24, 26.44.45.47, 49n3, 56, 129,246 Mare, Robert D., 72 Massy, William F., 12, 33, 212 Mattila, J. P., 246 Melder, J., 246 Merton, Robert C., 212, 217, 218, 219, 220, 221, 222, 223, 225, 226, 228, 233, 240

289

Author Index

Mickelson, R., 46 Miller, K., 46 Miller, Leonard S., 2n3, 24 Miller, Merton H.,240 Mortenson, Thomas G., 64,68 Mueller, Willard F., 34n35 Mundel, D., 246 Murphy, Kevin, l n l , 46,47, 59.62 National Center for Education Statistics (NCES), 179, 192, 193t, 194n7 National Research Council, 183, 184 National Science Foundation, 181t. 183, 192, 193t, 194, 198nl1 Nelson, Richard A,, 21 Nerlove, Marc, 2551117 Neuberger, Egon, 12, 13 Nichols, Donald, 21 I , 238 Nordhaus, William, 212 Oi, Walter Y., 28 O’Malley, Michael P., 132, 133nn19, 21

Shapley, Lloyd S.,17, 125-26n6 Sherman, Daniel R., 24, 26 Siegel, Paul M., 791117 Slavings, R., 47 Smart, John C., 123 Smith, H., 46 Smith, R., 2601123 Snyder, Thomas, 212 Solmon, L. C., 2511110 Sosa, Julie Ann, 2, 145, 183, 184nl Sotomayor, Marilda A. Oliveira, 18, 125n6 Spence, A. Michael, 19, 29 Spencer, Leslie, 211111, 23n14 Spies, Richard R., 241116, 126, 133 Stevens, Gillian, 791117 Stiglitz, Joseph, 19 Stocking, George W., 34n35 Streufert, P., 50n4 Sulock, J., 246, 2621126 Sundaresan, Suresh, 218 Svensson, Lars, 218, 232 Tobin, James, 21 1, 234, 238 Tomola, James, 185 Tsai, Shu-Ling, 82 Tversky, A,, 47n2

Pelavin, Sol H., 64 Peterson’s Annual Guide, 132t Pierro, Dan, 207, 246, 263n28 Polachek, Solomon, 112 Pollack, Judith, 180 Posner, Richard A,, 34n35 Powell, H., 46 Putka, Gary, 23

U. S. Bureau of the Census, 1 , 6 6 U. S. Department of Education, 1, 21nn9,10, 24n17, 25, 31t, 32t, 33t, 61

Quandt, Richard, 21 1, 238

Vance, Bradley, 24n17

Radner, Roy, 2n3, 24, 145x11 Raftery, Adrian E., 69n10 Reischauer, Edwin O., 124114 Rhine, Shenie L. W., 16115 Robb, R., 45 Rohlen, Thomas P., 6 Rosen, Harvey, 277, 281 Rosen, Sherwin, 44,46,47, 124115 Rosovsky, Henry, 3, 16n6 Roth, Alvin E., 17, 18, 125116 Rothschild, Michael, 19

Walton, Mary, 129n13 Washington Post Weekly, 2 Weiler, William C., 241117, 262 Weisbrod, Burton A,, 14, 246 Weiss, Y., 246 Welch, Finis, I n l , 46, 47, 49, 62 Westmeyer, P., 245, 277 White, Lawrence J., 23 Wigdor, Alexandra K.,20118 Williams, Robin M., Jr., 64,68, 102n35 Williamson, J. Peter, 21 1 Willis, R., 44, 46, 47 Wilson, Charles, 19 Wilson, Reginald, 64 Winston, Gordon C., 241116, 33n34, 129 Winter, Sidney G., 21 Wise, David A., 24, 26, 44,47, 129 Wood, Elizabeth, 24n17

Salop, Steven C., 23 Santos, Maria A,, 16115 Schapiro, Morton O., 132, 133nn19,21, 206111, 207 Schenet, Margot A., 132 Schuette, H. L., 21 Schuster, J. H., 145 Sewell, William H., 79, 82

Zapatero, Fernando, 218

This Page Intentionally Left Blank

Subject Index

Academic careers: alternative paths to, 16168; progression by field of study to, 157-61, 169-72tt; relation of choice to academic performance, 154-57 Administrators, university: in competitive pricing, 23; lack power and incentive, 38-41. See also Assets, university (endowment and nonendowment); Decision making; Endowment management; Endowment portfolio Admissions policies, 16-20, 129-30 Alumni representatives, 38 American Freshman Survey, 133 Assets, university (endowment and nonendowment), 213-18,219-20,239-40 Black students: as high school graduates, 6566; trend in college entry for, 64-65, 69-70, 89-92 Budget constraints, 15,21 Career choices: estimates of undergraduate, 148-54; uncertainty about, 146, 182. See also Academic careers Choice: factors influencing student, 23-27, 59-60; for quality of higher education, 250-61; relation of quality to, 129-30. See also Career choices; Feedback process; Preference function; Public choice model; Schooling choices COFHE. See Consortium on Financing Higher Education (COFHE) survey COFHE schools, 133n21

291

Colleges/universities: brand-name, 24-25, 28, 34, 134; effect of concentration of elite in, 143; for-profit motivation, 3940; importance of reputation for, 1303 1, 134; intertemporal complementarity and substitutibility of activity, 220-23, 240, 241; marketplace decisions of, 3033; as monopolies, 33-34; prices of activities, 241; as profit maximizers, 2022; resistance to technological change, 22-23. See also Competition; Endowment management; Endowment portfolio Competition: among different kinds of undergraduate education, 15-16; among universities, 20-34, 121-22; circumstances for loss in educational marketplace of, 143; effect of price discrimination on, 26-27; in higher education marketplace, 3-4, 38-39; in market for educational services, 23. See also Price system Consortium on Financing Higher Education (COFHE) survey, 147-48 Costs: defining relevant marginal, 40-41; differentials in education, 27-28; incentive to reduce, 38; uncertainty of university activity, 216. See also 'hition levels CPS data. See Current Population Survey (CPS) data Cross-subsidization: among university activities, 12; arguments for and against, 1315

292

Subject Index

Current Population Survey (CPS) data, 64, 66-68, 105-7, 112, 116 Dartmouth College case, 245-46, 260 Data sources: for analysis of college entry trends, 64, 66-68, 105-7, 112, 116; for analysis of educational selectivity, 1 13, 116; for choices in public higher education, 274-78; for estimates of F E E G student financial support, 192-94; for estimates of school enrollment and graduation, 113; for estimates of top students in elite schools, 123, 126-28; Harvard Senior Survey as, 146-49 Decision making: in colleges and universities, 38-39; factors influencing student, 12939; under uncertainty, 58; for university asset allocation, 213-18 Degrees: decline in Ph.D., 183-84; trend in doctorates granted, 183-84; value of elite school, 122-24 Demand: for educated work force, 62; for educational credentials, 134; effect on public enrollment rate, 261-64, 267; for elite universities, 132-33; for faculty, 22, 145, 183; for nonendowment assets, 225-34; in public choice model, 264, 267; for university output, 22 Duncan socioeconomic index for occupations, 79

Economies of scope (in higher education), 15-16 Education: differences in costs of undergraduate, 27; effect on economy of higher, 1; graduate and undergraduate as joint product, 14-16; reasons for better, 125 Educational attainment, 64-65, 72-83 Elite in society, 141 Elite schools: quality of education at, 125; questioned definition of, 141-42; trend in concentration of top students in, 13036; value of degree from, 122-24. See also Feedback process; Nonelite schools Employer recruiting, 136-39 Endowment management: issues in, 217-18; objectives of, 237-39, 241-42; with other sources of income, 225-234; standard approach to, 2 13 Endowment portfolio: investment selection for, 213-14; optimal asset allocation for, 215-18; wealth and substitution effects on, 214

Endowment portfolio model: description of, 218-25; with other income sources, 225-34; review of, 239-40 Enrollment: determinants of rate of, 261-71; of part-time science and engineering graduate students, 190, 191t Enrollment levels: comparison of public and private school, 243, 246-50; determinants of, 244-61; impact of state legislation on, 243 Entry: of college/university into marketplace, 30-33; dependency and nondependency as variables in college, 92-95; equality of opportunity for college, 100-103; factors influencing college, 62-64, 72-88; influence of social background on college, 83-88; student trend in college, 89-92 Ethnicity. See Black students; Hispanic students; White students Exit (of college/university from marketplace), 32-33 Expectations: effect of misspecified, 54-55; measurement of, 46-47 Expectations formation: assumptions about, 43-44; econometrics of, 45-46; theories of, 47. See also Human capital model Faculty: power of, 38; relative autonomy of, 2; shortages of, 2-3 Feedback process, 129-30, 136 Female students: trend in college entry for, 64, 69-71 Financial support: effect on university of external, 184-85; shifting levels of external and institutional, 187-205, 207-9 For-profit institutions, 39-40 Full-time science and engineering graduate (FTSEG) students: determinants of support for, 197-200, 202-3tt; disaggregation by type of support, 200-202; sources of financial support for, 185-90; supported by institutional funds, 190-91 Gourman index, 250-51 Graduate students: effect of increased federal support for, 185; federal government support for FTSEG, 185-90 Harvard Senior Survey: comparison of data with COFHE survey, 168, 173-74; description of and use for, 147-49

293

Subject Index

Hedging (of university assets), 216-17, 237. See also Assets, university; Endowment management Higher education: contribution of, 42; historical development of private, 245-46; public and private choice for quality in, 250-61. See also Colleges/universities Higher education, public: factors influencing enrollment in, 243; historical development of, 245-46; model of public choices in, 261-71; public choice for, 244-45; variation in opportunities offered, 250-51; variations in state expenditures for, 269-7 1, 282-83 High School and Beyond survey (1980). 180 Hispanic students: as high school graduates, 65; trend in college entry for, 65, 6970, 92 Human capital: absence of good markets for, 21-22; education as investment in, 43 Human capital model, 2, 48-54, 59 Incentives: lack of, 38-42; to reduce costs, 38 Income, realized, 50, 53-54 Income returns to schooling, 54, 59-60 Inputs: in higher education marketplace, 2223; incoming students as, 13; to university production function, 241 Investment: education as, 43; higher education as, 21-22. See also Endowment management; Endowment portfolio Legislation: for provision of public higher education, 244-46; reflecting public school enrollment, 243. See also Higher education, public; Spending, state; Subsidies; Tuition levels

Nonprofit institutions: in higher education, 21; lack of competitive incentive for, 21. 38-42; in private higher education, 1, 3 Output: demand for university, 22-33; of universities, 23-25, 218; of university production function, 241 Part-time science and engineering graduate students, 190, 191t Preference function: with endowment and nonendowment assets, 241; for university activity, 218-1 19, 241 Presidential Scholars Program, 126, 217, 128t Prices (of university activity), 241 Price system: to allocate students in higher education, 16-20; competitive among universities using, 23-25; effect of competition in, 26; effect within university of uniform, 27-28; using discrimination in, 25-27 Private sector: competition with public education sector, 38; control of institutions in, 261; nonprofit higher education in, 1, 3 Privatization, 41-42 Production function, university, 240-41 Public choice model, 261-71 Public policy: impact on college entry, 87-88; proposed for federal graduate study support, 184; for supply and demand at state level, 261-71. See also Legislation Public sector: control of institutions in, 2606 1; in higher education, 1, 8; subsidies to higher education in, 14

Minority groups. See Black students; Hispanic students Monill Acts (1862, 1890). 245

Quality: of elite school education, 125, 128; of public higher education, 262; relation to college choice, 129-130; signal of, 124; variation in public and private higher education, 250-55

National Intern Matching Program (NIMP), 17-18 National Longitudinal Study of High School Class of 1972, 123, 180, 182 National Longitudinal Survey of Youth (NLSY), 105, 113 NIMP. See National Intern Matching Program (NIMP) Nonelite schools, 125-26

Race. See Black students; Hispanic students; White students Recruiting: college expenditures for, 134; employer on-campus, 136-39 Returns to schooling: defined, 43; expectations formation for, 45-46; results of human capital model to analyze, 48; youth perceptions of, 44-45. See also Income returns to schooling

294

Subject Index

Schooling choices: with expectations based on ability, 50-53, 59; with misspecified expectations, 54-55 Spending, state, 269-71 Students: benefits of homogeneous or diverse groups of, 18-20; characteristics of elite school applicants, 133-34; competition among, 124; competition for top, 12122; concentration of elite, 141-42; factors influencing school choices of top, 129-39; as input to university production, 13; trend in college-level entry, 89-92; trend in concentration of top, 130-36. See also Black students; Fulltime science and engineering graduate (FTSEG) students; Graduate students; Hispanic students; White students Subsidies: in higher education budget constraint, 15; legislature choices for levels of, 253-61; to public colleges and universities, 38, 39. See also Crosssubsidization Substitution effect, 214

Supply: legislative choices, 261-69; of new Ph.D.S, 145, 183-84 Tuition levels: choice of legislature for, 25361; relative increases in, 2; relative uniformity among universities, 27-3 I Undergraduate education: institutions producing only, 14, 15-16; relation of performance to academic career choice, 154-61; shift in choice of major in, 180; as subsidy to graduate education, 13-16 Utility maximization, 39 Wages/salaries: differentials dependent on educational level, 62; for elite school graduates, 122-24; premium for college education, 59 Wealth effect, 214 Westinghouse Science Talent Search, 126-27, 128t

White students: as high school graduates, 66; trend in college entry for, 69-70, 89; trend in college entry for male, 64 ~

This Page Intentionally Left Blank