Immigration and the Work Force: Economic Consequences for the United States and Source Areas 9780226066707, 9780226066332

Citation preview

This Page Intentionally Left Blank

Immigration and the Work Force

A National Bureau of Economic Research Project Report

Immigration and the Work Force Economic Consequences for the United States and Source Areas

Edited by

George J. Borjas and Richard B. Freeman

The University of Chicago Press

Chicago and London

GEORGE J. BORJASis professor of economics at the University of California at San Diego and a research associate of the National Bureau of Economic Research. RICHARD B . FREEMAN is professor of economics at Harvard University and director of the Labor Studies Program at the National Bureau of Economic Research.

The University of Chicago Press, Chicago 60637 The University of Chicago Press, Ltd., London 0 1992 by the National Bureau of Economic Research All rights reserved. Published 1992 Printed in the United States of America 01009998979695949392

123456

ISBN (cloth): 0-226-06633-9

Library of Congress Cataloging-in-Publication Data Immigration and the work force : economic consequences for the United States and source areas / edited by George J. Borjas and Richard B. Freeman. p. cm.-(A National Bureau of Economic Research project report) Includes bibliographical references and indexes. 1. Alien labor-United States. 2. Labor market-United States. 3. Emigrant remittances. I. Borjas, George J. 11. Freeman, Richard B. (Richard Barry). 111. Series. HD8081.A5152 1992 92-15620 33 I .6'2'0973-d~20 CIP

@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 o$cer

Geoffrey Carliner, executive director Charles A. Walworth, treasurer Sam Parker, director ofjlnance and administration

Directors at Large John H. Biggs Andrew Brimmer 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 G. Cain, Wisconsin Franklin Fisher, Massachusetts Institute of Technology Jonathan Hughes, Northwestern Saul H. Hymans, Michigan Marjorie B. McElroy, Duke

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

Directors by Appoinment 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 Labor and Congress of Industrial Organizations Dean P. Phypers, Comrniteefor Economic Development Charles A. Walworth, American Institute of CertiJied Public Accountants

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

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

George B . Roberts Willard L. Thorp William S. Vickrey

Relation of the Directors to the Work and Publications of the National Bureau of Economic Research 1. The object of the National Bureau of Economic Research is to ascertain and to present to the public important economic facts and their interpretation in a scientific and impartial manner. The Board of Directors is charged with the responsibility of ensuring that the work of the National Bureau is carried on in strict conformity with this object. 2. The President of the National Bureau shall submit to the Board of Directors, or to its Executive Committee, for their formal adoption all specific proposals for research to be instituted. 3 . No research report shall be published by the National Bureau until the President has sent each member of the Board a notice that a manuscript is recommended for publication and that in the President’s opinion it is suitable for publication in accordance with the principles of the National Bureau. Such notification will include an abstract or summary of the manuscript’s content and a response form for use by those Directors who desire a copy of the manuscript for review. Each manuscript shall contain a summary drawing attention to the nature and treatment of the problem studied, the character of the data and their utilization in the report, and the main conclusions reached. 4. For each manuscript so submitted, a special committee of the Directors (including Directors Emeriti) shall be appointed by majority agreement of the President and Vice Presidents (or by the Executive Committee in case of inability to decide on the part of the President and Vice Presidents), consisting of three Directors selected as nearly as may be one from each general division of the Board. The names of the special manuscript committee shall be stated to each Director when notice of the proposed publication is submitted to him. It shall be the duty of each member of the special 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 manuscript may be published, though approved by each member of the special manuscript committee, until forty-five days have elapsed from the transmittal of the report in manuscript form. The interval is allowed for the receipt of any memorandum of dissent or reservation, together with a brief statement of his reasons, that any member may wish to express; and such memorandum of dissent or reservation shall be published with the manuscript if he so desires. Publication does not, however, imply that each member of the Board has read the manuscript, or that either members of the Board in 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 Board 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 adopred October 25, 1926, as revised through September 30, 1974)

Contents

Preface

1.

2.

3.

4.

5.

6.

7.

vii

ix

1 Introduction and Summary George J. Borjas and Richard B. Freeman National Origin and the Skills of Immigrants 17 in the Postwar Period George J. Borjas Out-Migration and Return Migration of 49 Puerto Ricans Fernando A. Ramos The Assimilation of Immigrants in the 67 U.S. Labor Market Robert J. LaLonde and Robert H. Topel The Fertility of Immigrant Women: Evidence 93 from High-Fertility Source Countries Francine D. Blau Mass Emigration, Remittances, and Economic Adjustment: The Case of El Salvador in the 1980s 135 Edward Funkhouser When the Minimum Wage Really Bites: The Effect of the U.S.-Level Minimum 177 on Puerto Rico Alida J. Castillo-Freeman and Richard B. Freeman

On the Labor Market Effects of Immigration and Wade George J. Borjas, Richard B . Freeman, and Lawrence F. Katz

213

viii 8.

Contents

The Effect of Immigrant Arrivals on Migratory Patterns of Native Workers Randall K. Filer

245

Contributors

27 1

Author Index

273

Subject Index

277

Preface

The papers collected in this volume are the final product of a research project organized by the Labor Studies Program of the National Bureau of Economic Research. Support for the project came from the Ford Foundation. We are grateful to Jennifer Amadeo-Holl for organizing the conference and to Ann Brown and Julie McCarthy for assistance in shepherding the volume through the editorial process. Any opinions expressed in this volume are those of the respective authors and do not necessarily reflect the view of the National Bureau of Economic Research or the sponsoring organization. George J. Borjas and Richard B . Freeman

ix

This Page Intentionally Left Blank

Introduction and Summary George J. Borjas and Richard B. Freeman

After decades during which the influx of immigrants to the United States declined relative to the growth of the native work force, immigration has once more become a major factor in the U.S. labor market. Since the 1930s, the immigrant flow has risen by about one million immigrants per decade. By the 1980s, about 600,000 persons were legally admitted to the United States per year, which added approximately 400,000 workers to the labor force annually. In addition, a steady flow of illegal entrants produced over three million undocumented aliens who qualified for amnesty under the provisions of the 1986 Immigration Reform and Control Act. Even though only 5 percent of the U. S . population was foreign born in 1970, by 1988 over 9 percent of the labor force was foreign born. As a result of these trends and the concurrent slowdown in the growth of the native work force, immigrants accounted for more than onequarter of new labor market entrants between 1980 and 1988. Key provisions of the 1990 Immigration Act and the continued influx of illegal immigrants, together with slower growth of the female work force and other demographic changes, ensure that the representation of immigrants in the U.S. labor force will increase through the beginning of the twenty-first century. Who are the immigrants? How do they perform in the U.S. labor market? How do they affect the employment opportunities of natives? How do the labor market effects of immigration compare to those of international trade? What does immigration to the United States do to the economies of the sending countries or regions? These questions, which are central to any assessment of the economic effect of immigration, guided this and the preceding NBER research project on George J. Borjas is professor of economics at the University of California, San Diego, and a research associate of the National Bureau of Economic Research. Richard B. Freeman is professor of economics at Harvard University and director of the Labor Studies Program at the National Bureau of Economic Research.

1

2

George J. Borjas and Richard B. Freeman

immigration. The first research report (Abowd and Freeman 1991) analyzed the effect of the post-1965 wave of immigration on the U.S. labor market and, for comparative purposes, studied immigration in two other major immigrant host countries, Canada and Australia. The current study provides additional analyses of the economic effects of immigration on the United States through the late 1980s and explores the links between immigration to the United States and selected source area economies. Both research projects concentrated on the “new immigration” that followed the 1965 immigration act, which scrapped the national origins quota system and thus greatly altered the number and characteristics of immigrants. 1. The New Immigration Prior to 1965, immigration to the United States was guided by the national origins quota system. Under this system, which dated from the 1920s, almost all available visas were given to applicants from northern and western European countries. The 1965 Amendments to the immigration and Nationality Act removed the national origin quotas and made family reunification the main objective of immigration policy. As a consequence of the 1965 Amendments and of major changes in economic and political conditions in source countries relative to conditions in the United States, the national origin mix of the immigrant flow entering the United States has changed substantially over the past few decades. Table 1 lists the “top ten” source countries during the 1950s and during the 1980s. There is a substantial amount of turnover in this list: only three counTable 1

“Top Ten” Source Countries in the 1950s and 1980s 1950s

Rank 1 1

L

3 4 5 6 7 8 9 10 Total flow from top ten countries (%)

Source Country Germany Canada Mexico U.K. Italy Cuba Austria Netherlands France Ireland

1980s

Size of Flow ( I ,000s)

477.8 378.0 299.8 202.8 185.5 78.9 67.1 52.3 51.1 48.4

Source Country Mexico Philippines China Korea Vietnam India Dominican Republic Jamaica U.K. Cuba

73.2

Source: U.S.Immigration and Naturalization Service (1990, table 2).

Size of Flow ( I ,000s)

975.7 477.5 306.1 302.8 266.0 222.0 209.9 184.5 140. I 135.1 55.5

3

Introduction and Summary

tries that were important sources of immigration during the 1950s remained so during the 1980s (Mexico, Cuba, and the United Kingdom). During the 1950s, 53 percent of immigrants originated in Europe, 25 percent in Latin America (i.e., the Western Hemisphere except for Canada), and only 6 percent in Asia. In contrast, during the 1980s, only 11 percent of immigrants originated in Europe, whereas 42 percent came from Latin America and 42 percent from Asia. While lifting restrictions on immigration from countries in Asia allowed the migration of Asians to occur, and the cutback in the number of visas for western European countries reduced the potential size of the immigrant flow from those countries, U.S. immigration policy is not the sole cause of changes in the national origin mix of immigrants. Availability of visas aside, potential migrants come to the United States on the basis of the benefits and costs of such a major decision in their lives. Economic and political conditions in the source countries and in the United States (as well as opportunities available in other immigrant-receiving countries) are potentially important determinants of these decisions and thus of the national origin mix of the immigrant flow. In addition to the legal immigrants, a sizable flow of illegal aliens entered the United States in the 1970s and 1980s, primarily from Mexico. As noted above, in 1986 Congress enacted the Immigration Reform and Control Act (IRCA), which gave amnesty to over three million illegal aliens, indicating that at least that many had entered illegally, mostly since the mid-1970s. Although IRCA included employer sanctions designed to deter the future entry of illegal immigrants, the flow of illegals did not appear to have slowed substantially by the end of the 1980s. In 1989, for example, the Border Patrol apprehended 954,000 persons attempting to enter the United States illegally, about the same number as it apprehended in 1982. The 1990 Immigration Act introduced several provisions designed to alter the size and composition of the immigrant flow. By 1995, the number of legal immigrants admitted annually (excluding refugees) will increase from about 500,000 to 675,000. While the bulk of these visas (480,000) will be awarded to relatives of persons already residing in the United States, the number of visas awarded to persons on the basis of skills will increase greatly: from 54,000 to 140,000. In effect, about half the additional visas will be awarded to skilled workers. Finally, to generate a more ethnically diverse foreign-born population, the remaining 55,000 visas will be allocated to persons originating in countries that have sent few immigrants to the United States since 1965. This provision is designed primarily to benefit visa applicants from European and African countries. Immigration reduces the size and alters the skill endowments of the labor force available to the source countries, with positive or negative consequences depending on the state of their labor markets. In some cases, immigration to the United States has greatly depleted the population of small source areas in Central America and in the Caribbean (where about 10 percent of Jamaicans

4

George J. Borjas and Richard B. Freeman

and one-third of Puerto Ricans, who have citizenship in the United States, have chosen to emigrate). Even among larger countries, immigration to the United States can have a nonnegligible effect on the size and composition of the labor force, particularly among selected skill groups. In the 1980s, 1.4 percent of Mexicans (most of them unskilled) migrated legally, and perhaps as many illegally, to the United States.’ While relatively few Indians have come to the United States, virtually all Indian migrants have been college graduates, many of them doctors. For a complete analysis of the economics of immigration, it is necessary to consider the effects of immigration not only on the economy of the United States but also on the economies of the source areas. This volume seeks to do this for selected areas where the effect of immigration is likely to be large. 2. The NBER Project

Motivated by the changing national origin composition of the immigrant flow admitted to the United States and by the fact that immigration affects the economic well-being of workers in source countries as well as in the United States, the NBER undertook the research project whose results are presented in this volume. Concern with both sides of the immigration “trade” led project researchers to develop data on the labor force and economies of the source countries or areas and also to analyze data from the U.S. Census of Population, which has long been the mainstay of information on immigrants to this country, and from various Current Population Surveys that include questions on immigration. In some cases, this involved adding variables about aggregate economic or demographic conditions in source countries to the Census files. In one case, it involved linking Census and labor force survey files from the source area, Puerto Rico, which has a similar statistical base as the United States proper, to the relevant U.S. data sets. In another case, the analyst exploited available surveys on the labor force in El Salvador and conducted field and survey research in the country. The use of computerized data sets on immigrants in the United States and on persons in the source countries is rare in the analysis of immigration, as any survey of the voluminous U.S. research makes clear. Such data sets allowed this NBER project to derive firmer conclusions about the characteristics and economic effects of immigrant flows than could possibly be reached from data sets covering only the United States or any single source country. Finally, to deal with issues of assimilation of immigrants from different countries to the United States, researchers combined data from decennial Censuses to create “synthetic cohorts”-persons in a given age group in one Cen1. In the 1980 Census, counted and uncounted illegal immigrants from Mexico exceeded the number of legal immigrants (see Borjas, Freeman, and Lang 1991).

5

Introduction and Summary

sus and those in that age group plus ten in the succeeding Census. In investigations of how immigrants progress in the economy relative to natives, synthetic cohorts offer a research tool that is superior to the more commonly used cross-sectional comparisons between immigrants who came in one period and those who came years earlier or later. The NBER research project paid considerable attention to migration between Puerto Rico and the mainland even though (or perhaps more properly because) Puerto Rico is not a foreign country but rather an integral part of the United States. Migration from Puerto Rico, which has a different language and culture than the mainland, is free from political impediments and thus provides a natural experiment to assess individual decisions to migrate, absent the need for entry or exit visas. If economic analysis cannot account for Puerto Rican migration decisions, its relevance to immigration from other areas, where the immigration decision is distorted by both U.S. immigration policy and source-country emigration policy, is subject to question. The analysis of the economic effects of immigration on source economies focused on potentially instructive “natural experiments” where the outflows to the United States were large: Puerto Rico, which has a larger proportion of persons born outside the United States residing in the country than any other locale; and El Salvador, a small country whose migrant flows are sufficiently sizable to have potentially large economic effects. The initial research design envisioned that the political problems in El Salvador-repression and civil war-might give an “exogenous” shock to immigration that could help identify its effects on the economy and that the resultant immigration would differ greatly from immigration from Puerto Rico. In fact, one of the major results of the analysis is that, despite some differences in the pattern of immigration from these areas, economic incentives rather than exogenous political factors appear to dominate migrant flows from both areas. In capsule form, the research in this volume adds to our stock of knowledge about immigration in several areas. It shows that labor market opportunities in source areas relative to opportunities in the United States help determine the magnitude and composition of immigrant flows (Borjas; Ramos; Funkhouser; Castillo-Freeman and Freeman). It also shows that the changing national origin mix of immigrants over time-that is, the greater influx of persons from less-developed countries with lower education and income levels than the traditional European source countries-is the prime determinant of the decline in the skills of immigrants relative to native workers in the United States (Borjas). The research on assimilation shows that immigrants assimilate fairly rapidly in the U.S. job market, with the result that, a few years after arrival in the United States, immigrants earn roughly as much as comparably skilled nativeborn workers of the same national origin, but not necessarily as much as the typical native worker (LaLonde and Topel). This finding reinforces the importance of the initial skill composition of immigrant flows in calculating their

6

George J. Borjas and Richard B. Freeman

contribution to the nation’s skill endowment. The research on fertility behavior yields an even more striking example of “assimilation”; it finds that immigrant women from high-fertility countries have essentially the same fertility behavior as native American women (Blau). The low fertility of these immigrants prior to immigration suggests a strong element of selectivity or preadaption to the expected U.S. economic conditions. With respect to the economic effects of immigration on source economies, immigration to the United States is a major element in the economic development of these areas (Funkhouser; Castillo-Freeman and Freeman) and thus a potentially important policy tool to spur development. In addition to the direct effects of immigration on the size and composition of the labor force, remittances from immigrants in the United States to the source countries induce changes in the labor supply behavior of remaining family members. On the U.S. side, the increasing number of immigrants with less than a high school education appears to have substantially affected the job market for the declining number of natives in that schooling category (Borjas, Freeman, and Katz). This finding suggests that the effect of immigration on the earnings and employment opportunities of native-born workers is much greater than was reported in the first NBER volume (Abowd and Freeman 1991), which based its assessment on the relatively small differences in the economic position of natives across localities with differing immigrant flows. One possible reason for the difference in results is that natives adjust their migration behavior to the influx of immigrants, with the result that total labor supplies in immigrantintensive areas are little affected by immigration (Filer). All told, this volume shows that immigration links the labor markets of the United States and of major source areas in important ways, ways that affect the well-being of workers in both the destination and the source economies. Thus, the economic development and human capital formation policies of the major source countries should be matters of concern to Americans as well as to citizens of those countries sending large numbers of immigrants to the United States. We summarize next the specific findings that underlie these broad conclusions and the evidence and logic on which they are based. Some of the findings are new because previous research either did not address the issue or addressed it with less-adequate data than were available to this research project. Some of the findings disagree sharply with earlier work. When that is the case, we note the disagreement, try to pinpoint the causes of the disagreement, and consider which results appear to be more valid. Source Countries and Immigrants to the United States 1. The changing national origin mix of immigrantjows is the major reason for the decline in the skills of immigrants relative to natives. At the time of entry into the United States, the typical worker who migrated in the late 1950s had about 0.4 years more schooling than natives. By the late

7

Introduction and Summary

1970s, the typical newly arrived immigrant had 0.7 years fewer schooling than natives (Borjas, table 1.4). In terms of earnings, the late-1950s immigrant earned 13 percent less than natives at the time of arrival, while the late 1970s newly arrived immigrant earned nearly 30 percent less than natives. A single factor, the changing national origin mix of the immigrant flow, is responsible for most of this decline in relative immigrant skills (Borjas). This conclusion is based on two facts. First, the skills and labor market performance of immigrants vary significantly by national origin. For instance, the 1980 Census reveals that newly arrived immigrants originating in the United Kingdom have 2.5 years more schooling and earn about 22 percent more than natives, while newly arrived immigrants from Mexico have 6.1 fewer years of schooling and earn about 61 percent less than natives (Borjas, table 1.8). Second, the national origin composition of immigrant flows has shifted from the more-developed to the less-developed countries (see table 1 above). Given large and reasonably constant differences in the education and U.S. labor market performance of immigrants from different countries, the contribution of the changing national origin mix to the decline in relative immigrant skills can be obtained simply by applying different distributions of immigrants by country to any particular year’s difference between the education and earnings of natives and immigrants by country. The result of such calculations clearly shows the massive effect of the actual change in the source country mix on the average labor skills of immigrants. 2. Migration from source areas to the United States is consistent with an economic analysis of immigration based on labor supply considerations. As noted above, migration from Puerto Rico to the United States presents a natural experiment for testing economic theories of immigration because Puerto Ricans face no quotas or legal impediments to moving to the mainland. Ramos applies the self-selection model of immigration introduced by Borjas (1987b) to the Puerto Rican case and finds striking confirmation of the model’s stress on the effect of differing rates of return to skills between source and destination areas on the skill composition of migration flows. According to the model, highly skilled persons migrate to countries that offer a relatively high rate of return to their skills, while the less skilled prefer countries with more egalitarian income distributions. Since Puerto Rico’s income distribution offers a much higher payoff to skills than does that of the mainland United States, the analysis implies that highly skilled Puerto Ricans will, on average, prefer to remain in Puerto Rico and that the less skilled will migrate to the United States. Ramos makes use of information on the education and earnings of migrants to the United States and of residents in Puerto Rico, distinguishes between those who never migrated and return migrants, and finds that migrants to the United States are less skilled than nonmigrants and that return migrants tend to be the most-skilled workers from the initially unskilled flow. In a related analysis of the Puerto Rican case, Castillo-Freeman and Free-

8

George J. Borjas and Richard B. Freeman

man explore the potential effects of one policy innovation-enactment of high minimum wages-on immigration to the United States. They show that the introduction of a U. S.-level minimum wage reduced employment substantially on the island and that immigrants to the United States consisted disproportionately of persons lacking work on the island and of those with characteristics that made them especially subject to the minimum. They found that the tendency for migrants to be less educated than nonmigrants developed in the 1970s as the Puerto Rican minimum rose toward U. S. levels and as return to schooling rose on the island relative to the mainland. Funkhouser’s analysis of migration from El Salvador lends additional support to the notion that economic factors are critical in determining immigration flows. While noting the difficulty of differentiating the effects of structural economic problems from civil war-related political repression on immigration, he attributes much of the immigration flows to economic conditions per se. Here, however, it is the more educated who come to the United States-a fact that can be attributed to the massive difference in income levels between El Salvador and the United States and the information and transportation costs of migration for the less educated. Castillo-Freeman and Freeman’s finding that those at the very bottom of the educational attainment ladder in Puerto Rico are also unlikely to migrate in large part because they speak no English helps reconcile the Puerto Rican and Salvadoran cases. 3. Immigration and trade alter U.S. factor proportions in the same direction: increasing the supply of less-skilled labor. The classic Heckscher-Ohlin model of trade predicts that flows of factors and goods will operate in the same direction to equate relative factor proportions across’economies. The United States has an exceptionally large proportion of highly educated workers, suggesting that trade and immigration should act to increase the relative number of less-skilled workers. Borjas, Freeman, and Katz show that this is in fact the case in their comparisons of the “implicit skill composition” of trade and the educational composition of immigrants. On the immigrant side, the greatest number of immigrants are persons with less than a high school degree. On the trade side, import-intensive industries tend to employ relatively low-skill workers, including many immigrants. Assimilation into the United States 4. Immigrant earnings may reach parity with those of their native-born ethnic counterparts, but not with the earnings of the typical native-born worker. As new entrants to the U.S. labor market, immigrants invariably start their work lives in the United States at a disadvantage compared to otherwise similar natives (i.e., natives with the same educational attainment, age, etc.). The speed with which immigrants assimilate to the labor market, measured by the rate at which their earnings catch up to the earnings of natives as they accumulate labor market experience in the United States, has long been a key

9

Introduction and Summary

indicator of the ability of the economy to absorb immigrants and of immigrants’ ability to adapt to U.S. economic conditions. Because a single cross section cannot disentangle the economic effects of assimilation from those of changes in the unmeasured characteristics of cohorts who come to the country at different times (Borjas 1985), the importance of assimilation can be determined only by analyzing longitudinal data or “synthetic” cohort data created from a series of Census cross sections. LaLonde and Topel show that the extent of economic “assimilation” depends critically on the groups of native workers to which immigrants are compared. From 1970 to 1980, new immigrants did not catch up with typical U.S. workers in terms of earnings. By contrast, they did catch up with native-born workers from similar ethnic groups. In addition, the state of the labor market when immigrants arrive can also affect how well they do. Less-skilled immigrants are likely to do worse if they come to the United States when the job market for the less skilled is deteriorating, as in the 1980s, than if they come when that job market is improving; similarly, how well skilled immigrants are likely to do will depend in part on the market for skills when they enter the country. LaLonde and Topel’s evidence differs from Borjas’s (1985) decomposition of immigrant wage growth into cohort and assimilation effects. Borjas generally finds a smaller rate of earnings adaptation within national origin groups. The related questions of how best to define economic assimilation and how best to measure it have attracted substantial attention. Recent work by Smith (1991) and Friedberg (1991) argues that the intercensal comparisons that are the basis of the findings both in Borjas and in LaLonde and Topel are subject to error because they fail to control for the age at which persons migrated. For any given time-of-arrival cohort, the composition of the working-aged sample will change across Censuses because later Censuses include a larger number of immigrants who migrated as children and who thus did not go through the same process of labor market adaptation as persons who migrated as adults. The evidence shows that rates of growth of earnings of immigrants differ depending on whether those immigrants came as children or as adults. Persons who came to the United States as children look similar to native-born workers. Persons who migrate as adults do not experience that much economic assimilation. Mixing the two groups overestimates the rate of growth of immigrant earnings relative to native earnings. 5 . The fertility rate of immigrant women is similar to the fertility rate of native-born women. Earnings assimilation, while important, is not the only way in which immigrants adjust to the U.S. economic conditions. Blau compares the fertility of immigrant women in the United States to average rates of fertility in those women’s source countries and to the fertility of similarly aged native-born women. The initial expectation of the study was that immigrant fertility rates would lie between those in the source country and those in the United States,

10

George J. Borjas and Richard B. Freeman

indicating partial adjustment of fertility over time, consistent with BenPorath’s (1973) findings for Israel. In fact, the results are much more striking. The average married immigrant woman had 2.4 children in 1980, compared to 2.2 children for similarly aged women born in the United States and to an average of 5.5 children for a woman residing in the typical source country (Blau, tables 4.1 and 4.3). Even prior to migrating to the United States, the fertility behavior of immigrant women tends to resemble that of native-born women, not that of their compatriots who remain in the source countries. After arriving in the United States, immigrant women make further adjustments to their fertility, so that after a few years in the United States the number of children borne by immigrant women is quite similar to the number borne by native-born women. This remarkable similarity in fertility behavior raises important questions about the way in which the flow of immigrant women is self-selected or adapts to expected U.S. economic conditions even in their home country and about the extent to which immigrant women learn and adapt to the U.S. environment after arrival. Repercussions of Immigration for Source Countries

6 . Immigration to the United States has major direct and indirect effects on the labor markets of small source areas. The most immediate and direct effect of immigration to the United States on the labor market of a small source economy is through the reduction in the labor supply of migrants. However, there are also important indirect modes of adjustment that can augment or offset the direct reduction in labor supply (Funkhous6r). Remittances from immigrants raise the income of family members remaining in the source country, with the standard income effect of reducing the family’s labor supply. At the same time, the family may substitute for the migrant in the local labor market, which is particularly likely if wages or other earnings opportunities rise owing to immigration or if the reduction in family size “frees up” time previously devoted to household production. In El Salvador, the income effect on family labor supply caused by remittances was significant, dominating other effects, with the result that the reduction in labor supply exceeded the direct decline due to immigration. Because of economic dislocation resulting from civil war, these effects have eased the unemployment problem rather than creating a labor shortage problem. In Puerto Rico, the flow of immigrants to the United States was so large as to alter significantly the island’s capital/labor ratio and thus to affect the overall level of real earnings. Castillo-Freeman and Freeman estimate that a minimum of one-fourth and probably much more of the long-term upward trend in real earnings in Puerto Rico is due to migration to the United States. Moreover, since migration was concentrated among the less educated and those lacking jobs, the migrant flow also reduced the rate of unemployment on the island.

11

Introduction and Summary

7. By providing an important mode of labor market adjustment, immigration alters the set of policy tools available to governments in source and destination countries. When a large number of persons migrate to another locale but leave family members behind, their remittances can contribute substantially to the nation’s supply of foreign currency. Funkhouser reports that remittances to El Salvador were two-thirds as great as exports and equal to the entire trade deficit, making immigration more important than any single industry in generating foreign currency. This produced a major parallel currency market in the country, which limited the ability of the Salvadoran government to control the money supply. In Puerto Rico, migration to the United States made the imposition of high minimum wages possible because it provided an outlet beyond unemployment for those displaced by the minimum. For the United States, immigration offers a mode of adjusting the skills of the work force to changes in the domestic economy, as in the 1990 effort to increase the influx of skilled migrants through additional visas issued on the basis of skills. However, such fine-tuning of the flows has traditionally been weak owing to the focus on family reunification in awarding visas and the lack of success in preventing sizable illegal flows. Economic Repercussions for the United States

8 . In the 1970s, native migration Jrows ofset immigration to local labor markets, suggesting that the internal migration decisions of native workers may respond to the pattern of immigration among areas. Immigrants typically concentrate in relatively few source cities, such as Los Angeles, New kork, and Miami. In the previous NBER research volume, Bartel and Koch (1991) found that immigrants tended to stay in these initial areas rather than disperse across the country. This concentration of immigrants in limited areas has led many researchers to assess the effects of immigration on the earnings and employment opportunities of native workers by examining differences in native wages between cities with many immigrants and those with few immigrants. These studies, for the most part, find an insignificant correlation between the presence of immigrants in a locality and the earnings of natives in that locality (Grossman 1982; Borjas 1987a; Card 1990; Altonji and Card 1991; Butcher and Card 1991; LaLonde and Tope1 1991). One possible explanation of this result is that natives attenuate the negative effect of immigration by choosing to reside in other localities. Using 1980 Census data, Filer finds a negative correlation between the in-migration rates of natives to particular cities and the presence of immigrants in those cities. This result suggests that natives respond to the increase in immigrant labor supply by moving to other cities. In effect, the internal migration of natives dissipates the effect of immigration on particular labor markets and makes it difficult to determine the effects on immigration by spatial analyses. These results are consistent with Card’s (1990) finding that the Marie1 Boat-

12

George J. Borjas and Richard B. Freeman

lift had no noticeable effect on the population growth in Miami, but they are not consistent with studies of the relation between internal migration and immigration in other time periods. Bronars (1989) found an inverse correlation between immigration and native mobility among states in 1980 but not in earlier Census years. Using the Current Population Surveys, Butcher and Card (1991) report a positive correlation between the in-migration rates of natives to particular cities and immigration flows in the 1980s. The evidence thus suggests that Filer’s results may reflect the particular historical circumstances of the late 1970s rather than a general structural pattern of response. Even if this is the case, however, it casts doubt on inferences of the broad economic effects of immigration from cross-city comparisons. If native labor mobility offsets the effects of immigration on native workers in local labor markets in one period, perhaps flows of capital or immigrant-induced expansion of demand for goods offsets it in other periods. 9. Immigration of workers with less than a high school education reduced the rate of decline in the less-educated work force of the United States, with important consequences for the US.earnings distribution. From the 1970s through the 1980s, the real earnings and employmentpopulation rates of less-educated American men deteriorated at a historically unprecedented rate. Earnings differentials between more- and less-educated workers skyrocketed, and inequality rose among workers in given skill categories, producing a more unequal income distribution. Borjas, Freeman, and Katz find that the flow of less-educated immigrants and the indirect increment in labor supply due to the importation of goods produced by less-educated workers contributed substantially to the rise in earnings differentials across education groups. Because of the concentration of immigrants among workers with less than a high school education, more than one-quarter of American workers with fewer than twelve years of schooling in 1988 were immigrants. The massive trade deficit experienced by the United States during the 1980s contributed an additional increment to the implicit labor supply of less-educated workers, with the result that in 1988 the combined effect of immigration and trade was to increase the “supply” of high school dropouts in the U.S. economy by about 30 percent. By contrast, the effect of immigration and trade on the supply of more-educated workers was relatively modest. Given reasonable responses of wages and employment opportunities to an increase in the ratio of less-educated to more-educated workers, this massive change in relative supplies must have had a sizable adverse effect on the economic well-being of less-skilled workers. Borjas, Freeman, and Katz find that perhaps as much as half the 10 percentage point decline in the relative weekly wage of high school dropouts between 1980 and 1988 can be attributed to trade and immigration flows. This conclusion contrasts sharply with the inference drawn from spatially oriented studies, including those in the previous NBER volume, that compare

13

Introduction and Summary

the earnings and employment of natives in localities where large numbers of immigrants reside and of natives in localities with limited numbers of immigrants. Those studies found only modest and statistically insignificant differences in the position of natives relative to the level of immigration, leading to the inference that immigration does not affect native workers. There are several potential explanations for the differing implications of the results from the aggregate analysis in this volume and the cross-area analyses that dominate the literature. First, note that in the spatial literature many of the point estimates of the effect of immigration on native earnings are statistically insignificant and thus not inconsistent with the evidence in this volume. Consider, for instance, the findings in Altonji and Card (1991), which is representative of the literature based on spatial comparisons. Using the 1980 Census, Altonji and Card find that a 10 percentage point increase in the fraction of immigrants in a standard metropolitan statistical area (SMSA) reduces the log wage of white male high school dropouts in that locality by about 1.8 percent, with a standard error of 2.1 percent (row 6 of their table 7.9). Looking at changes between the 1970 and the 1980 Censuses, they obtain a larger but still statistically insignificant effect of a ten-point increase in the immigrant share on earnings of 3.6 percent. One interpretation of the statistical insignificance is that the elasticity has a very wide confidence interval: a 5 percent confidence interval would range from - 6 to 2 percent for their 1980 Census result and from - 12 to + 5 percent for their cross-sectional change estimate. The estimated declines in real wages for dropouts in Borjas, Freeman, and Katz lie at the lower end of these ranges.2 Another possible reason for the perceived difference in the results is the continued growth of immigrant flows (relative to the native labor supply) in the latter part of the 1980s, which necessarily makes immigration more significant in the 1980s than one would infer from comparisons of local labor markets in the 1980 Census. If this is correct, comparisons of natives in cities with more or less immigration in the 1990 Census or with increasing immigration from 1980 to 1990 should yield higher estimates of immigrant effects than did estimates based largely on the 1980 Census. The third explanation of the difference between the spatial studies and the more aggregate analyses in this volume is that local labor markets adjust rapidly to the increased supply of immigrants. If markets clear rapidly, the effects of immigration will be dispersed throughout the economy and thus cannot be readily discovered by spatial analyses. Suppose, for example, that immigrants

+

2. In the Borjas, Freeman, and Katz study, the immigrant share of high school dropouts rose by about 10 percentage points between 1980 and 1988, with a resultant increase in the pay of dropouts (relative to high school graduates) of 2.4-6.1 percent (table 7.8). A ten-point increase in the immigrant share of all workers comparable to Altonji and Card’s (1991) analysis would raise the immigrant share of high school dropouts by roughly twice as much, given the larger proportion of dropouts among immigrants than among natives, implying greater responses, of 412 percent.

14

George J. Borjas and Richard B. Freeman

and natives are substitutes in production. Then, as immigrants enter the labor market, native wages fall, and natives will move to cities where they face less competition from immigrants (or not move to cities with large immigrant populations, in accord with Filer’s findings). Alternatively, the fall in wages will induce capital flows to the immigrant-receiving areas. As a consequence, native wages are equalized across cities, yielding no measurable correlation between native wage rates and the presence of immigrants across labor markets. This hypothesis implies that both the spatial correlations and the macro findings are “correct.” The findings differ because they address very different questions. The spatially based studies correctly tell us that immigrants have no measurable effect on particular labor markets, but they are not informative about the economy-wide effects of immigration. The aggregate analysis indicates that immigration affects economy-wide labor supplies, with a sizable effect on the aggregate economic oppoytunities of natives.

3. Conclusion The 1980s witnessed a resurgence in the importance of immigration as a determinant of demographic change in the United States, a rebirth of the political debate over how many and which types of immigrants this country should admit, and a renewal of interest in the “economics of immigration.” It is likely that all these trends will intensify as we enter the twenty-first century. The 1990 Immigration Act ensures that the size of the immigrant flow will increase substantially during the 1990s and that legal immigration will reach, if not surpass, the historically high levels recorded in the early 1900s. This increased immigration will be taking place at a time when the growth of the native-born work .force has greatly slowed down, both because of the aging of the baby boomers and because female labor force participation rates are reaching a plateau. In addition, despite the enactment of IRCA, the flow of illegal aliens continues unabated. It is likely, therefore, that the same concerns that sparked the debate over immigration policy in the 1980s will resurface in the next few years, with renewed pressure for further changes in policy concerning both legal and illegal immigration. The increasing importance of foreign-born workers to the U.S. labor market, and the growing awareness that these workers have a substantial economic effect not only in the United States but also in source countries, will surely motivate and guide an extensive research agenda designed to measure these effects. The papers in this volume have highlighted a number of research areas that deserve further analysis: the reasons for the small cross-sectional correlations between native employment opportunities and the presence of immigrants in labor markets compared to the larger effect inferred from the aggregate change in labor supplies due to immigration; the definition and measurement of immigrant assimilation. The reintroduction of immigration

15

Introduction and Summary

into mainstream economic analysis on a par with trade and capital flows has only just begun.

References Abowd, John M., and Richard B. Freeman, eds. 1991. Immigration, Trade, and the Labor Market. Chicago: University of Chicago Press. Altonji, Joseph, and David Card. 1991. The Effects of Immigration on the Labor Market Outcomes of Less-skilled Natives. In Abowd and Freeman (1991). Bartel, Ann P., and Marianne J. Koch. 1991. Internal Migration of U.S. Immigrants. In Abowd and Freeman ( 1991) . Ben-Porath, Yoram. 1973. Economic Analysis of Fertility in Israel: Point and Counterpoint. Journal of Political Economy 8 1 (MarchlApril): S2024233. Borjas, George J. 1985. Assimilation, Changes in Cohort Quality, and the Eamings of Immigrants. Journal ofLabor Economics 3 (October): 463-89. . 1987a. Immigrants, Minorities, and Labor Market Competition. Industrial and Labor Relations Review 40 (April): 382-92. . 1987(b). Self-selection and the Eamings of Immigrants. American Economic Review 77 (September): 531-53. Borjas, George J., Richard B. Freeman, and Kevin Lang. 1991. Undocumented Mexican-born Workers in the United States: How Many, How Permanent? In Abowd and Freeman (1991). Bronars, Stephen G. 1989. Immigration, Internal Migration, and Economic Growth: 1940-1980. University of California, Santa Barbara. Mimeo. Butcher, Kristen F., and David Card. 1991. Immigration and Wages: Evidence from the 1980s. American Economic Review 81 (May): 292-96. Card, David. 1990. The Impact of the Marie1 Boatlift on the Miami Labor Market. Industrial and Labor Relations Review 43 (January): 245-58. Friedberg, Rachel. 1991. The Labor Market Assimilation of Immigrants in the United States: The Role of Age at Arrival. Massachusetts Institute of Technology. Mimeo. Grossman, Jean. 1982. The Substitutability of Natives and Immigrants to Production. Review of Economics and Statistics 54 (November): 596-603. LaLonde, Robert J., and Robert H . Topel. 1991. Labor Market Adjustments to Increased Immigration. In Abowd and Freeman (1991). Smith, James. 1991. Hispanics and the American Dream: An Analysis of Hispanic Male Labor Market Wages, 1940-1980. Santa Monica, Calif.: Rand Corp. Mimeo. U. S. Immigration and Naturalization Service. 1990. Statistical Yearbook of the Immigration and Naturalization Service, 1989. Washington, D.C.: U.S. Government Printing Office.

This Page Intentionally Left Blank

1

National Origin and the Skills of Immigrants in the Postwar Period George J. Borjas

Immigration is an increasingly important component of demographic change in the United States. Since the Great Depression, the size of the legal immigrant flow has increased at the rate of approximately one million persons per decade. Although only 500,000 immigrants entered the United States during the 1930s, approximately eight million immigrants were admitted legally during the 1980s (U.S. Immigration and Naturalization Service 1988, pp. 1-2). A more revealing way of describing the growing importance of immigration is to contrast the size of the immigrant flow with the number of live births that occur in the United States. The immigranvbirth ratio was only .02 in the 1930s; it increasdd to .06 during the 1950s and to .16 during the 1980s.' As a fraction of births, therefore, immigrant flows today are near the record levels achieved in the early 1900s, when the immigranvbirth ratio was almost .20. Furthermore, these statistics understate the current importance of immigration as a determinant of demographic change because they ignore the large numbers of illegal aliens who entered the United States in the past two decades. The significant role played by immigration in recent years sparked the development of a large and growing literature analyzing a fundamental aspect of the immigrant experience: how immigrants perform in and adapt to the American labor market. Using the 1970 and 1980 Public Use Samples of the U.S. Census, for the most part these studies find that earlier waves of immigrants have relatively high earnings in the labor market but that more recent waves are less successful. George J. Borjas is professor of economics at the University of California, San Diego, and a research associate of the National Bureau of Economic Research. The author is grateful to John Abowd, Geoffrey Carliner, Richard Freeman, James Smith, and Stephen Trejo for insightful comments. He is also grateful to the National Science Foundation (grant SES-8809281) for financial support. 1. These statistics are drawn from Borjas (1990, 6).

17

18

George J. Borjas

Although this empirical finding is robust with respect to methodological approach and time periods analyzed, its interpretation is less clear.*The early studies (Chiswick 1978; Carliner 1980; DeFreitas 1980) used single crosssectional data sets and stressed the concept of immigrant assimilation or adaptation in explaining the empirical evidence. As immigrants accumulate experience in the U.S. labor market, their age/earnings profiles converge to those of comparable natives, and earlier immigrant waves can be expected to experience more favorable economic outcomes. More recent studies (Borjas 1985, 1987; Abbott and Beach 1987; Jasso and Rosenzweig 1988) suggest that different immigrant waves have substantially different skills, and the empirical results may be revealing a shift in the earnings capacities or underlying abilities of successive cohorts of immigrants entering the United States. The skill differentials among successive immigrant waveskan arise for a number of reasons. First, immigrants have very high out-migration rates. At least 20-30 percent of the foreign born return to their birthplace (or migrate elsewhere) within a decade or two after their arrival in the United States.’ If these immigrants are, on average, persons who did not perform well in the labor market, the earlier waves overrepresent “successes” and have higher earnings than more recent waves. The stylized fact discussed above is consistent with this alternative hypothesis. Skill differentials among immigrant waves may also be generated by the major changes in immigration policy that occurred in the postwar period. Prior to the 1965 Amendments to the Immigration and Nationality Act, immigration to the United States was guided by the national origins quota system. Under this system, the number of visas allocated to countries was based on the repiesentation of that national origin group in the U.S. population as of 1920. The 1965 Amendments abolished the “discriminatory” quotas (where the discrimination was based on national origin) and established a system under which visas are allocated mainly to applicants who have relatives already residing in the United States. Finally, it is likely that changing economic and political conditions in the source countries, relative to those in the United States, altered the national origin mix and skill characteristics of immigrant flows. After all, even if visas are freely available, many persons will not find it profitable to migrate to the United States. The increasing income levels and political stability attained by 2. In fact, most of the studies in the literature generate the stylized fact using data from the 1970s and 1980s. Blau (1980) and Eichengreen and Gemery (1986) use data from the late 1800s and early 1900s. These studies, unlike those based on the recent data, reach conflicting conclusions. Blau reports age/earnings trajectories for immigrants that greatly resemble those obtained from the 1970-80 data. By contrast, Eichengreen and Gemery report that more recent immigrant waves perform as well as, if not better than, earlier immigrant waves. 3. For estimates of the out-migration rate, see Warren and Peck (1980), Jasso and Rosenzweig (1982), and Borjas and Bratsberg (1990). The type of selection that characterizes out-migrants is addressed by Jasso and Rosenzweig (1988), Borjas (1989), and Borjas and Bratsberg (1990). Probably because of data problems, these studies do not reach a consensus on the type of selection that characterizes out-migrants.

19

National Origin and the Skills of Immigrants

western European countries in the postwar period probably reduced the incentives of these national origin groups to migrate to the United States. Similarly, political upheavals in many parts of the world such as Cuba or Southeast Asia also affected the nature of the immigrant flow. The postwar years, therefore, witnessed fundamental shifts in the size, national origin mix, and skill composition of immigrant flows. Remarkably, there has been little systematic study of these trends. Using the five decennial Public Use Samples available between 1940 and 1980, this paper documents the effect of changes in the “immigration market” on the skills and labor market performance of the foreign born in the United States. The empirical analysis of the five decennial Censuses yields two substantive results. First, almost all the measures of skills or labor market success available in the data document a steady deterioration in the skills and labor market performance of successive immigrant waves over the postwar period, with this trend accelerating since 1960. More important, the study suggests that a single factor, the changing national origin mix of the immigrant flow, is almost entirely responsible for this trend. In fact, the empirical analysis presented below reveals that, if the national origin mix of the immigrant flow had not changed over the postwar period, the decline in skills and the deterioration in the labor market performance of successive immigrant waves would not have occurred.

1.1 U.S. Immigration Policy Before proceeding to the empirical analysis, a brief summary of the changes that occurred in immigration policy during the postwar period will be instructive. This description helps establish the institutional background that regulates the size and composition of immigrant Immigration to the United States was largely unregulated during the first century after independence. The first restrictive legislation was passed in the 1870s, in response to the entry of large numbers of Chinese immigrants into the western states. Responding to the resultant political pressure, Congress moved to restrict the admission of certain groups into the United States. By 1917, these statutes banned the entry of large numbers of persons, including all Asians, political radicals, persons with tuberculosis, and polygamists. As the immigrant flow from Asia was completely cut off, a major shift occurred in the national origin composition of European immigrants. Traditionally, the immigrant flow had originated in northwestern European countries, such as the United Kingdom and Germany. Economic and political factors shifted the origin of the immigrant flow toward southern and eastern European countries, such as Italy, Poland, and Russia. To redirect the origin of the immigrant flow, Congress enacted the national origin quota system in 4. Hutchinson (1981 ) presents a comprehensive history of American immigration policy up to 1965

20

George J. Borjas

the 1920s. The number of entry visas allocated to countries in the Eastern Hemisphere depended proportionately on their representation in the national origin composition of the U.S. population in 1920. Because the ancestors of the great majority of U.S. residents originated in northwestern Europe, the United Kingdom was allocated 65,721 visas (almost half the 150,000 available visas) and Germany 25,957, while Italy was allocated 5,802 visas and Russia 2,784. The national origins quota system applied only to visa applicants originating in countries in the Eastern Hemisphere. Applicants from Western Hemisphere countries were exempt from the quotas and faced no numerical restrictions on the number of visas, presumably because of the close economic and political ties between the United States and its neighbors. These visas were awarded on a first-come, first-served basis as long as the applicants satisfied the growing list of health, moral, and political requirements. A review of immigration policy in the immediate postwar period led to the reaffirmation of the national origins quota system in the Immigration and Nationality Act of 1952. In addition, the 1952 statutes included a preference system as a means of allocating quota visas among the Eastern Hemisphere applic~nts.~ Preference was given to applicants whose skills were “needed urgently” in the country, and half of all visas were allocated to such persons. The remaining visas were allocated to relatives of U.S. residents. A melange of laws, regulations, and private bills diminished the importance of the national origins quota system over time. In their review of immigration policy, Abrams and Abrams (1975, p. 7) conclude that, “although the national origins system was theoretically the heart of American immigration policy until 1965, by the 1950s two thirds of all immigrants were being admitted under exceptions to it.” The 1965 Amendments to the Immigration and Nationality Act (and subsequent revisions in the immigration laws through the 1980s) regulated the process of legal immigration until the enactment of the 1990 Immigration Act. Table 1.1 summarizes the main components of current law and reports the number of legal immigrants admitted in 1987 under the various provisions. The United States currently permits the entry of 270,000 persons per year, with no more than 20,000 immigrants originating in any particular country of origin. Instead of focusing on national origin as the key determinant of admission, the 1965 Amendments made family reunification the central objective of immigration policy. This was accomplished through several provisions. First, 80 percent of the 270,000 numerically limited visas go to “close” relatives of U.S. citizens or residents. These close relatives include unmarried adult children of U.S. citizens, siblings of adult U.S. citizens, and spouses of resident aliens. The remaining 20 percent of the visas are allocated to persons on the 5 . A preference system was already in place as a result of the statutes enacted in the 1920s (see Hutchinson 1981, p. 580).

21

National Origin and the Skills of Immigrants

Table 1.1

Provisions of U.S. Immigration Law and Number of Immigrants Admitted in 1987

Preference

No. Admitted (in 1,000s)

Immigrants subject to numerical restrictions (270,000visas) First: Unmarried adult children of U.S. citizens and their children (20% of visas are allocated to this category) Second: Spouses and unmarried children of permanent resident aliens and their children (-26% and any visas not used above) Third: Professional or highly skilled persons and their spouses and children (10%) Fourth: Married children of U.S. citizens and their spouses and children (10%and any visas not used above) Fifth: Siblings of adult U.S. citizens and their spouses and children (24% and any visas not used above) Sixth: Needed skilled and unskilled workers and their spouses and children (10%) Nonpreference and other (visas not used above and other special admissions) Subtotal

~

Immigrants not subject to numerical restrictions Spouses, parents, and minor children of adult U.S. citizens Refugees and asylum seekers Other Subtotal Total

218.6 96.5 15.3 330.4 601.5

11.4

110.8

26.9 20.7 69.0 27.0 5.4 271.1

~

Source; U.S. Immigration and Naturalization Service (1988, pp. 8-1 I )

basis of their skills. A large number of these 54,000 visas, however, are allocated to the families of the skilled workers who qualify for a visa. Furthermore, parents, spouses, and minor children of adult U.S. citizens can bypass the numerical restrictions specified in the legislation. These “immediate” relatives automatically qualify for entry into the United States and need not apply for one of the 270,000 numerically limited visas. As table 1.1 shows, more immigrants (219,000) entered under this single provision of the law than under all the family reunification preferences combined (2 17,000). Owing to the combination of the kinship bias in the preference system and the unregulated entry available to immediate relatives, only 4 percent of the legal immigrants admitted in 1987 actually entered the United States because of their skills. The postwar period also witnessed the entry of large numbers of refugees and asylum seekers. Prior to 1980, the United States defined a refugee as a person fleeing a Communist country, a Communist-dominated area, or the Middle East. Over two million permanent residents entered the United States as refugees (or asylum seekers) since 1946 (U.S. Immigration and Naturali-

22

George J. Borjas

zation Service 1988, p. 62).6 The largest refugee flow originated in Cuba (473,000) and the second largest in Vietnam (41 1,000). Refugee admissions have become increasingly important since the 1960s. The fraction of total immigration that can be attributed to refugee admissions increased from 6 to 19 percent between the 1960s and the 1980s and is rapidly approaching the level reached immediately after World War 11 (25 percent), when a large flow of displaced persons entered the United States. The most noticeable consequence of the disintegration of the national origins quota system, of the enactment of the 1965 Amendments, and of changing political and economic conditions both in the United States and abroad is the shift that occurred in the national origin mix of the immigrant flow in the postwar period. Table 1.2 summarizes the national origin distribution of the immigrant flows admitted in each decade between 1931 and 1980. During the Great Depression, when the size of the immigration flow was at a record low, nearly two-thirds of the immigrants originated in Europe, and the remainder originated in the Western Hemisphere. By the 1950s, the fraction of persons originating in Europe had declined to about half, the percentage originating in the Americas had increased to about 40 percent, and the size of the Asian immigrant flow became nontrivial (6 percent of immigrants). During the 1970s, the share of Europeans declined further to roughly 18 percent, the share of Western Hemisphere immigrants was 44 percent, and Asian countries were responsible for just over one-third of the immigrant flow. The change in the national origin of immigrants is strikingly revealed by a more disaggregated look at the national origin mix of immigrants. Table 1.3 presents a “top ten” list of the source countries responsible for immigration in the period 1931-80. Even though German immigrants were the largest national origin group in each decade between 1931 and 1960, German immigration was not sufficiently large to place it among the top ten flows in the 1970s. On the other hand, six of the countries in the top ten in the 1970s (the Philippines, Korea, Vietnam, India, the Dominican Republic, and Jamaica) were not important source countries as recently as the 1950s. It is erroneous to attribute this shift in the national origin mix of the immigrant flow solely to changes in U.S. immigration policy. Obviously, the lifting of the restrictions on immigration from Asia is responsible for the increased Asian migration, and the cutback in the number of visas allocated to western European countries reduces the potential size of the immigrant flow from those countries. However, even if visas are freely available, potential migrants will not come to the United States unless they gain from the move. Even prior to the 1965 Amendments, quotas allocated to many European countries went unfilled. For instance, during the first half of the 1960s, the United Kingdom was allocated over 65,000 visas per year, but the annual flow 6. These data are not accurate counts of the number of refugees because many of the refugees never adjust their status to permanent residence.

23

National Origin and the Skills of Immigrants

Table 1.2

Period

Legal Immigration, 1931-80, by Origin No. of Immigrants (in 1,000s)

1931-40 1941-50

528.4 1.035.0

% of Immigrant Flow Originating in:

Africa .3 .7

Asia

Americas

Europe

3.0 3. I

30.3 34.3 39.6 51.6 44.1

65.8 60.0

1951-60

2,515.5

.6

6.1

1961-70 1971-80

3,321.7 4,493.3

.9 1.8

12.9 35.3

52.1 33.8 17.8

Source: U.S. Immigration and Naturalization Service (1987, pp. 2-5).

Table 1.3

Source Countries with Ten Largest Immigrant Flows, 1931-80 (size of flow in thousands) 193 1-40

Rank I

Country

1951-60

No. of Immigrants

10 Source:

U.S.Immigration and Naturalization Service (1987, pp. 2-5).

103.5 68.0 31.6 22.3 17.0 14.4 12.6 11.0 9.1

Germany Canada Mexico U.K. Italy Cuba Austria Netherlands France Ireland

No. of Immigrants Country

Germany Canada Italy U.K. Mexico Poland Czechoslovakia France Ireland Greece

2 3 4 5 6 7 8 9

I 14.1

Country

1971-80

477.8 378.0 299.8 202.8 185.5 78.9 67.1 52.3 51.1 48.4

Mexico Philippines Korea Cuba Vietnam Canada India Dominican Republic Jamaica U.K.

No. of Immigrants 640.3 355.0 267.6 264.9 172.8 169.9 164.1 148.1 137.6 137.4

averaged fewer than 28,000 persons (U.S. Immigration and Naturalization Service 1965, p. 34). In other words, American immigration policy and economic conditions in the United States are not the only variables that influence the mobility decisions of potential migrants. The alternative opportunities provided by economic and political conditions in the source countries also affect the size and composition of immigrant flows. As will be seen below, the postwar shift in the national origin mix of the immigrant flow is the most important single factor that explains the changing economic effect of immigrants on the United States.

1.2 Data and Descriptive Statistics The empirical analysis uses the Public Use Samples of the decennial Censuses available since 1940. In each of the Censuses, the study is restricted to

24

George J. Borjas

men aged 25-64 who do not reside in group quarters. The entire 1/100 samples of immigrants and natives contained in the 1940 and 1960 Public Use Samples and the “sample line” extract of the 1/100 1950 Public Use Sample are used in the analysis.’ The 1970 immigrant extract contains a 21100 sample (obtained by pooling the 1/100 state and standard metropolitan statistical area [SMSA] files from the 5 percent questionnaire), while the 1980 immigrant extract contains the entire 5/100 A File. Random samples of the native base are drawn for studying the periods 1970 and 1980.8 Table 1.4 presents descriptive statistics for both the native and the immigrant populations in each of the five Censuses. The variables summarized in the table are the number of years of completed schooling, the labor force participation rate in the Census week, the unemployment rate (defined as the fraction of labor force participants looking for work in the Census week), the logarithm of weeks worked in the calendar year prior to the Census (calculated in the subsample of persons who worked in that year), the logarithm of annual earnings in the year prior to the Census (calculated in the subsample of persons who worked in that year), the logarithm of the wage rate (calculated in the subsample of workers and defined as the ratio of annual earnings to annual hours worked), and the logarithm of the wage rate adjusted for differences in observable socioeconomic characteristics (including education, age, marital status, and metropolitan residence) between immigrants and native^.^ The top panel of the table gives the average values of the variables under analysis for native men. The middle panel of the table presents the difference in the various variables between the average immigrant enumerated in the Census and the average native. The trends in these differences reflect two factors. Ovet time a particular cohort or wave of immigrants adapts or “assimilates” in the labor market, and the differences between immigrants and natives would be expected to narrow. At the same time, however, newer immigrant cohorts are replacing older cohorts. Because the new immigrant cohorts may differ from the old, the composition of the immigrant pool is changing across Censuses. Four of the five decennial Censuses provide information on the year of immigration for the foreign born. In particular, the 1970 and 1980 Censuses report the calendar year of immigration (in intervals), while the 1940 and 1960 Censuses report the place of residence five years prior to the Census.10 7. The Public Use Sample of the 1950 Census is a lil00 sample. Many of the variables required for the analysis reported below, however, are available for only a subsample of the respondents. 8. The 1970 sample of natives is a lil000 extract, while the 1980 sample of natives is a 1/2500 extract. 9 . To compute the adjusted wage differentials, I estimated separate wage regressions for natives and immigrants in each Census. The adjusted wage differential is evaluated at the sample mean of immigrants in each Census. 10. The 1950 Census does not provide any information on the place of residence five years prior to the Census; hence, a comparable sample of recent immigrants cannot be constructed from

25

National Origin and the Skills of Immigrants

Table 1.4

Mean Characteristics of Native and Immigrant Men, 1940-80 Natives

Variable Years of schooling Labor force participation Iate Unemployment rate Log (weeks worked) Log (annual earnings) Log (wage rate) Sample size

1940

1950

1960

I970

1980

8.8 ,949 ,012 3.777 6.978 - .549

9.5 .937 ,035 3.825 7.865 ,297

10.3 ,940 ,041 3.845 8.432 ,850

11.3 .921 ,026 3.886 8.989 1.365

12.7 ,892 ,049 3.843 9.608 2.041

149,477

60,541

260,537

28,978

15,071

Difference between Immigrants and Natives Years of schooling Labor force participation rate Unemployment rate Log (weeks worked) Log (annual earnings) Log (wage rate) Adjusted log (wage rate) Sample size

- 1.861

- ,020 ,027 - .032 .060 ,138 ,124 26,989

- 1.755 - ,016 ,011

- ,014 ,030 ,067 ,062 6,316

- 1.289 - ,018 ,003 - ,018

- ,727

,001

,005 - ,026 - ,037

,045 ,049

- ,001

- ,827 ,007 ,003 - ,036 - ,149 - .083 - ,062

17,566

32,491

134,252

- ,001

,010

Difference between Recent Immigrants and Natives Years of schooling Labor force participation rate Unemployment rate Log (weeks workeq) Log (annual earnings) Log (wage rate) Adjusted log (wage rate) Sample size

,153 - ,033 ,028 - ,073 - ,091 - ,031 - ,026

544

... ... ...

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

,412 .006 ,003 -.I16 - .263 - ,128 -.I13

- ,222 - ,005 ,012 - ,094 - ,289 -.I60 - ,149

- ,664 - .056 ,018 -.137 - ,480 - ,299 - ,224

1,886

6,205

26,781

The data thus allow the creation of a “recent immigrant” sample in each of these Censuses-that is, a sample of immigrants who arrived in the five-year period prior to the Census. Thus, it is possible to net out the assimilation effect in the intercensal comparisons and focus the analysis on the contrast among successive immigrant waves. The bottom panel of table 1.4 presents the averthese data. To the extent possible, I attempted to create comparable samples across the various Censuses. The most problematic of the Censuses (in terms of comparability) is the 1940 Census (where, e.g., weeks worked last year is defined in terms of full-time equivalent weeks). In every Census, the sample is restricted to persons who are not self-employed, and wage rates are defined by dividing annual earnings by annual hours worked, where annual hours worked is the product of weeks worked and hours worked last week in the 1940-70 Censuses and is defined as the product of weeks worked and usual hours worked per week in the 1980 Census. In every Census, the sample of workers used for the wage regressions is defined in terms of whether they reported hours worked in the past year.

26

George J. Borjas

age difference in skills and labor market characteristics between recent immigrants and natives. The descriptive statistics presented in table 1.4 tell an interesting story. Consider initially the educational attainment of the native and immigrant populations. Completed years of schooling increased steadily for natives throughout the period, from 8.8 years in 1940 to 12.7 years in 1980. The schooling gap between immigrants and natives declined until 1970 and then began to rise again. For instance, the typical immigrant in 1940 had 1.9 fewer years of schooling than natives; this difference declined to 0.7 years in 1970, but it increased to 0.8 years in 1980. The changes in the educational attainment of the immigrant population are, of course, much more pronounced in the comparison among successive immigrant waves. In 1940, the typical immigrant who had just arrived in the United States had about 0.8 years more schooling than the typical native. This educational advantage narrowed over time, and by 1970 the typical immigrant who had just entered the United States had slightly less schooling than natives. The decline in the relative schooling of immigrants accelerated during the 1970s, with the result that the most recent immigrant wave enumerated in the 1980 Census had 0.7 years fewer schooling than natives. Table 1.4 documents qualitatively similar trends in the performance of recent immigrants (relative to natives) in many other measures of labor market success. For instance, the labor force participation rate of natives declined from about 95 percent in 1940 to about 89 percent in 1980. The decline observed in the participation rate of recent immigrants is even steeper. In the 1940 Census, at the end of the Great Depression, recent immigrants had participation rates that were 3 percentage points below those of natives. This difference vanishes in the 1960 Census, but by 1980 the participation rate of recent immigrants was 5.6 percentage points below that of natives. The trend in the unemployment rate statistics tells the same story. In 1940, the unemployment rate of native men aged 25-64 was 7.2 percent, while that of recent immigrants was about 2.8 percentage points above that of natives. The difference between the two groups’ unemployment rates vanishes by 1960, reappears in 1970, and becomes large (almost two-thirds of the Great Depression difference) in 1980. Exactly the same trend is revealed by the weeks-worked data: recent immigrants in 1980 worked about 14 percent fewer weeks than natives, while recent immigrants in 1940 worked only 7.3 percent fewer weeks than natives. The annual earnings and wage data provide striking confirmation of recent findings in the literature suggesting that recent immigrant waves have lower earnings capacities than earlier waves (Borjas 1985, 1987). Table 1.4 indicates that this trend can be observed over the entire postwar period, not simply for the post-1960 cohorts. In 1940, the wage rate of recent immigrants was about 3.1 percent lower than that of natives. The wage differential increased

27

National Origin and the Skills of Immigrants

to 12.8 percent by 1960, to 16.0 percent by 1970, and to 29.9 percent by 1980. Finally, table 1.4 shows that the drop in the relative immigrant wage cannot be explained by the relative decline in immigrant educational attainment (or by changes in other observable demographic characteristics). In 1940, the typical recent immigrant earned about 2.6 percent less than a demographically comparable native. The wage disadvantage of recent immigrants relative to comparable natives increased to 11.3 percent in 1960, to 14.9 percent in 1970, and to 22.4 percent in 1980. Table 1.5 continues the descriptive analysis by documenting the changes in the occupational distributions of immigrant and native workers during the postwar period. These data use the one-digit occupational categories defined in the 1980 Census, where the occupational categories reported for the 194070 decennial Censuses have been redefined to match those in the 1980 Census as closely as possible. The secular trend in the native occupational distribution, of course, reflects structural changes in the U.S. economy. In particular, the fraction of the native labor force working in managerial and professional specialty occupations increased from 16.8 to 25.4 percent over the period, while the fraction employed in precision production, craft, and repair occupations increased from 15.6 to 22.5 percent. Conversely, the fraction employed in farming, forestry, and fishing occupations decreased from 22 to 4 percent. Table 1.5 shows that the changes observed in the occupational distribution of successive immigrant waves do not necessarily mirror those experienced by natives. For instance, the most recent immigrants in 1940 were about 8.6 percentage points more likely to be managers or professionals than were natives. By 1970, this statistic had declined to about 4.3 percentage points, and by 1980 the most recent immigrant wave was 0.5 percentage points less likely to be in managerial or professional jobs than were natives. Similarly, the most recent immigrants in 1940 or 1960 were about as likely as natives to be employed in the craft and repair occupations, but by 1980 the most recent immigrant wave was 6.4 percentage points less likely to be employed in these types of jobs. In contrast, recent immigrants in 1940 were half as likely to be in agricultural jobs as natives (10.9 percent of immigrants vs. 22 percent of natives). But by 1980 the most recent immigrants were slightly more likely to be in agricultural jobs than were natives (4.6 vs. 4.0 percent, respectively). Moreover, recent immigrants in 1940 were slightly less likely than natives to be employed as operators, fabricators, and laborers, but by 1980 recent immigrants were slightly more likely than natives to be in these occupations. Table 1.6 portrays the industrial distribution of immigrants and natives during the postwar period. These data reinforce the conclusion that the agricultural sector provides ample job opportunities for more recent immigrant

28

George J. Borjas

Table 1.5

Occupational Distributions of Native and Immigrant Men, 1940-80 ~~

Natives Occupation

1940

1950

1960

1970

1980

Managerial and professional Technical, sales, and administrative support Service Farming, forestry, and fishing Precision production, craft, and repair Operators, fabricators, and laborers

,168

,201

.230

.274

,254

,135 ,055

,132 ,052

,132 ,053

,139 ,068

,187 ,071

,220

,133

,074

,043

.040

,156

,206

.212

.231

,225

,262

,273

,258

,245

,222

Difference between Immigrants and Natives Managerial and professional Technical, sales, and administrative support Service Farming, forestry, and fishing Precision production, craft, and repair Operators, fabricators, and laborers

,009

.010

,006

,018

,006

-.049 ,036

-.042 ,038

-.024 ,045

-.027 ,037

-.030 ,040

-.I29

-.073

-.031

-.014

-.001

-.019

,063

,036

,022

-.003

,070

,031

,004

-.010

,003

Difference between Recent Immigrants and Natives Managerial and professional Technical, sales, and administrative support Service Farming, forestry, and fishing Precision production, craft, and repair Operators, fabricators, and laborers

,086

...

.026

.043

,014 ,033

... ...

-.029 ,043

-.043 ,049

-.I11

...

-.035

-.020

- ,015

...

,009

-.050

- ,007

...

.003

.021

-.005

-.024 ,061 ,006 -.064 ,026

waves. Apart from this fact, however, there are remarkably few other discernible trends in the immigrant industrial distribution (relative to the secular trends in the native distribution). Therefore, the historical trends in the occupational and industrial distributions indicate that, except for agriculture, the growing divergence between immigrants and natives does not lie in which sector of the economy they are employed. Rather, the divergence is occumng in the kinds of tasks that immigrants and natives perform on the job. More recent immigrant waves are less likely to be employed in the types of jobs that

Industrial Distributions of Natives and Immigrant Men, 1940-80

Table 1.6 ~~~~

~

~

~

1940

Industry Agriculture Mining Construction Manufacturing Transportation Trade Finance Business service Personal services Entertainment Professional servic:es Government Other

Natives

.227 ,029 .062 ,241 .096 .151

,032 .024 ,030 .008 ,045 .045 .010

Immigrants

.097 ,030 ,068 ,334 .08 I ,209 .03I .020 .059 ,009

,032 ,020 .010

1950 Recent Immigrants .I18 ,019 ,052 ,255 .063 ,221 ,023 .026 ,073 .018 .086 ,024 .022

Natives

.139’ ,026 ,084 ,277 .I02 ,169 ,030 ,034 ,025 ,008 .048 .054

,004

1960

Imrnigrants

Natives

,067

,081

.018

,016 ,088 ,314 ,093

,087 .348 ,084 ,222 .029 ,024 .047 ,009 ,038 .024 .003

,155

,035 ,030 ,021 .006 .068 .061 .032

Imrnigrants

,048 .006 .083 .359 ,072 ,196 .035 ,027 .049

~~

1970 Recent Irnmigrants

,045 .003 .095 ,386 ,048 .I53 .037 ,029

,009

,046 ,005

.076 .029 .012

,121 ,019 ,013

Natives .051

,013 ,102 ,300 ,094 ,163 ,042 .035 .020 ,007 .098 .070 .Ooo

Immigrants ,038 ,005 ,092 ,328 ,065 ,189 ,042 ,038 ,036 ,009 ,122 ,032 .Ooo

~~~~

1980 Recent Immigrants

.030 .005 .019 .347 ,049 .167 .039 .040

.033 ,011

.177 .020 .Ooo

Natives

Immigrants

,040

.040

.017 .I02 .283 ,106 ,159 ,045 ,045 ,012 .008 . I19 ,062

,006 ,089 ,305 ,070 ,189 ,046 ,052 ,029 ,010 ,132 .031

,001

,001

Recent Immigrants

,045 ,008 .069 .316 ,048 .200 .044

.057 .032 ,011

.142 ,027 .001

30

George J. Borjas

require relatively high levels of skills (such as managerial or craft jobs) and more likely to be employed in jobs that require fewer skills (such as the operators and laborers occupation and agriculture). Of course, the intercensal comparisons discussed above make an implicit assumption about period effects. It is well known that neither longitudinal data nor a series of cross sections provides sufficient degrees of freedom to estimate aging, cohort, and period effects without an identifying restriction. I have assumed that, by differencing immigrants’ wage and employment outcomes from those experienced by natives, I have netted out the effect of the business cycle, of shifting skill prices, and of other macroeconomic fluctuations on the skills and labor market performance of immigrants. In effect, I have assumed that period effects are the same for immigrants and natives. It is unlikely, however, that immigrants and natives respond equally to cyclical changes in the economy or that secular changes in the rental price of skills are the same for both groups. For instance, it may be the case that immigrant wages and labor market opportunities are much more sensitive to economic downturns than are those of natives. This hypothesis would provide an alternative explanation of why immigrant labor market performance lagged in 1980 (although the hypothesis would be hard pressed to explain the 1940 data). To determine the sensitivity of intercensal comparisons to changes in the native base, I calculated the immigranthative differences using alternative reference groups. The top panel of table 1.7 presents the estimated differences between recent immigrants and young native men (aged 18-24). These two groups have one factor in common: both have just entered the U.S. labor market. If new labor market entrants are more sensitive to changing economic conditions, intercensal comparisons of the skills and labor market performance of recent immigrants that adjust for the changes experienced by young native men should provide better estimates of the secular trends. Alternatively, one can argue that recent immigrants should be compared, not to young native men, but to native men who are roughly in the same stage of the working life. In fact, a disproportionately large number of the recent immigrants in my sample are between the ages of 25 and 44 (in 1980, e.g., 85.5 percent of the recent immigrants are in this age group, as compared to 48.5 percent of the natives). Hence, an alternative base is the group of native men aged 25-44. The bottom panel of table 1.7 reestimates the various differences using these natives as the reference group. Despite the major changes in the way that period effects are accounted for, the results in table 1.7 qualitatively resemble those discussed above. For instance, the typical recent immigrant in 1960 earned about 33.1 percent more per hour than a young native man. The immigrant advantage over natives aged 18-24 declines to 24.9 percent in 1970 and to 21.5 percent in 1980. Over the period 1960-80, therefore, the relative immigrant wage declined by about 12 percent. Similarly, recent immigrants in 1960 earned 12.6 percent less than

31

National Origin and the Skills of Immigrants

Table 1.7

Differences between Recent Immigrants and Alternative Reference Groups Census Year

1940

1960

1970

1980

Base group: Young native men (18--24) - ,231 Education ,082 Labor force participation rate - ,071 Unemployment rate ,127 Log (weeks worked) ,692 Log (annual earnings) ,521 Log (wage rate) ,353 Adjusted log (wage rate)

- ,285 ,119 - ,042 ,112 ,541 ,331 ,472

- ,835 ,159 - ,037 ,084 ,428 ,249 ,210

- ,313 ,013 - ,053

Base group: Native men aged 25-44 Education Labor force participation rate Unemployment rate Log (weeks worked) Log (annual earnings) Log (wage rate) Adjusted log (wage rate)

- ,286 - ,022 ,003 -.116 - ,272 - ,126 - ,143

- ,862 - ,040

- 1.354 -.I13 .017 -.I37 - ,443 - ,257 - ,221

Variable

.326 - ,055 .028 - .072 - ,048 ,011 - ,107

.OI 1

- ,095 - ,288 - ,148 -.I36

,101 ,390 ,215 .I28

'The adjusted log wage controls for differences in education, marital status, and metropolitan residence in panel 1; it also includes age in panel 2.

natives aged 25-44. By 1970, the wage disadvantage had increased to 14.8 percent and by 1980 to 25.7 percent. Between 1960 and 1980, the immigrant relative wage had fallen by 13 percentage points. In table 1.4 above, which used the population of native men aged 25-64 as the base group, the decline in the relative immigrant wage over the period 1960-80 was 17 percent. It seems, therefore, that accounting for differential period effects between the immigrant and the native populations only attenuates the downward trend in immigrant skills and labor market performance.

1.3 National Origin and Declining Immigrant Skills The historical evidence presented in the previous section provides strong evidence of a significant deterioration in the (relative) skill level and labor market performance of successive immigrant waves in the postwar period. I will argue that the main reason for the observed decline in immigrant skills is the changing national origin mix of the immigrant population. Because of shifts in the parameters guiding exchanges in the immigration market, the bulk of the immigrant flow to the United States today is composed of national origin groups that, for a number of reasons, do not perform well in the U.S. labor market. The empirical analysis presented below shows that this hypoth-

32

George J. Borjas

esis does remarkably well in explaining the facts summarized in the last section. Throughout the remainder of the paper, the immigrant population is categorized into forty-one national origin groups as well as a residual “other” category. The forty-one national origin groups account for over 90 percent of the 1951-80 immigrant flow. Moreover, a subset of thirty of these countries accounts for over 99 percent of the foreign-born population enumerated in the 1940 Census. For each of these national origin groups (and for the “other” national origin category), as well as for the most recent immigrants in each of the groups, I calculated the average characteristics of the various skill and labor market variables introduced in the last section. Table 1.8 illustrates the extent of these differences among recent immigrants for forty-one national origin groups (relative to the mean of native men aged 25-64) reported in the 1980 Census. The intercountry variation in skills and labor market performance is huge. Mean years of schooling among recent immigrants (relative to natives) range from - 6.1 for immigrants originating in Mexico to over 3 for immigrants originating in such diverse countries as France, the Netherlands, Egypt, and India. Similarly, the labor force participation rate of immigrants ranges from 40 percentage points below to 5 percentage points above the native rate, the unemployment rate from - .05 to + .11, the logarithm of weeks worked from - .39 to 0.0, and the log wage rate from - .70 to + .33. Similarly, the fraction employed in managerial or professional occupations can be as high as 32 percentage points above the native propensity for Swedish immigrants and as low as 21 percentage points below for immigrants born in Mexico. By jointly analyzing data on the skills and labor market characteristics of national origin groups and data on the shifting source country composition of the immigrant flow, I can document the extent to which the changes in national origin are responsible for the decline in immigrant skills. Let Y,be the average value for a particular skill or labor market characteristic observed in the immigrant population in year r (relative to that observed in the native population). By definition, Y,can be written as

where yj, is the average value for the labor market characteristic observed among immigrants from national origin group j in year r , and pJIis the fraction of the immigrant flow in year r originating in country j . It is useful to define the average labor market performance that would have been observed if a different national origin mix had migrated to the United States, such as the national origin mix observed at time 7,p j T .This is given by

Table 1.8

Skills and Labor Market Characteristicsof Recent Immigrants in 1980 (relative to natives)

Country of Birth

Education

Europe Austria Czechoslovakia Denmark France Germany Greece Hungary Ireland Italy Netherlands Norway Poland Portugal Romania Spain Sweden Switzerland U.K. USSR Yugoslavia

2.06 2.60 2.16 3.21 2.61 - 1.32 .93 1.26 - 1.80 3.23 2.90 .05 -5.86 1.25 .83 2.89 2.79 2.49 I .51 - 1.54

Asia and Africa China Egypt India (continued )

- .96 3.24 3.43

LFP Rate .05 .02 - .07 .00 .02 - .04 - .03 - .01 - .01

.04 - .06 .02 .06

.oo

- .02

.00 .01

.05 - .07 .03

Unemp. Rate

-

.oo .00

- .05 - .03 - .02 .01 .00

.02 .02 - .03 - .03 - .01

.03 .07

- .01 - .03 - .04 - .02 .06

.03

- .08

- .01

- .05 .02

.01

.01

Log Weeks Worked

Log Annual Earnings

Log Wage Rate

-.I8 - .03 -.18 - .07 - .02 - .I6 - .07 -.I1 - .06 .00 - .05 - .I9 - .09 -.19 -.13 - .02 - .07 - .02 - .30 -.I3

- .32

- .10

.07 .29 .20 .27 - .55 -.19 - .27 - .22 .33 .22 - .54 - .40 - .44 - .32 .23 .I7 .24 - .62 - .26

.06 .45 .25 .29 - .31 -.I5 -.I1 -.I3 .31 .27 - .34 - .30 - .26 -.I9 .I9 .24 .22 - .26 -.I4

-.17 -.17 -.I1

- .70 - .51 - .31

- .5l - .26 -.I8

Adjusted Log Wage

Managerial

- .08 -.I0 .43 .I8 .I9 - .20 -.I7

.26 .20 .29 .27 .31 - .05

-.I0 - .06 .22 .27 - .38 - .04 - .35 - .24 .I9 .24 .I4 - .38 - .05

.13 - .05 .33 .29 - .09 - .20

- .46 - .34 - .31

- .03

.01

Crafts - .I4

.09 .03 - .I2 - .06 .03 .I2 - .03 .01

-.I5 - .08

.oo

Operatives

-.i6 -.17 -.19 -.17 -.16 - .04 - .08 - .05

.oo

-.18

-.16 .18

.00

.33

.04

.04

- .02

.05 .32 .23 .34

- .07 - .07

- .06

.04 - .10

.19 .23

- .08

-.lo .06 .08 -.I2 - .I5 -.I6

-.16 -.16 -.16

- .02 .10 -.11 -.13 - .08

Table 1.8

Country of Birth Iran Israel Japan Korea Philippines Americas Argentina Brazil Canada Colombia Cuba Dominican Rep Ecuador Guatemala Haiti Jamaica Mexico Panama Trinidad & Tobago

(continued)

Education 2.46 1.37 2.90 1.49 1.41 1.21 2.67 2.02 - .62 - 1.92 - 3.93 - 1.40 - 3.64 - 2.69 - 1.15 - 6.06 .28 - .62

LF'P Rate

Unemp. Rate

- .40

.05 .02 - .03 - .oo

- .07 - .02 -.04 - .01

-

.oo .oo

- .01 - .I4 - .04 - .03 -.I4 - .01 - .04 .04 - .02

- .01 - .03 - .02 .I1 .04 .06 - .01

.06 .03 .04 .04 .05

- .01

.03 - .04 - .04

Note: LFP = labor force participation; Unemp.

Log Weeks Worked

=

unemployment.

Annual Earnings

Log Wage Rate

Adjusted Log Wage

- .39 - .21 - .03 -.I6 -.I3

- .77 - .48 .21 - .48 - .48

- .20 - .22 .23 - .27 - .29

- .21 - .24

-.I3 -.I1 - .04 -.I6 - .21 - .I6 -.I0 -.I1 -.I1 - .I7 -.I3 - .04 -.I3

- .28

-.I1 .08 .20 - .42

-.I6 .02 .I4 - .35 - .51 - .46 - .40 - .36 - .56 - .28 - .26 - .33 - .32

- .08

.18 - .65 - .73 - .87 - .67 - .14 - .78 - .59 - .77 - .45 - .58

- .51 - .65

- .49 - .57 - .63 - .33 - .61 - .34 - .39

.I1

- .38 - .31

Managerial

Crafts

Operatives

.I3 .I4 .33 - .03 - .07

-.I1 -.I0 - .I6 - .06 - .09

-.i2 - .08 -.i8 .04 - .04

.05 .28 .30 - .07

- .01 - .I2 -.I2 - .04 - .01 - .I0 - .03 .03

- .03

- .08

- .Ol - .I8

- .I5 - .21 - .I9

- .I0 - .I4

.10

.05 .27 .16 .15

.03

- .01

.22 - .04 .20 - .01

- .06

- .03

.05

-.I1 - .21

- .oo - .05

35

National Origin and the Skills of Immigrants

The effect of a changing national origin mix is then given by the difference between equations (1) and (2): (3)

y, - Y(t7

7) =

c

YJ, (P,,- PJJ

I

The decomposition implicit in equation (3) is similar to that commonly used in the discrimination literature (Oaxaca 1973) and has its roots in the statistical literature (Kitigawa 1955). It is well known that this is not the only possible measure of the change in Y due to the shift in the source country composition of the immigrant flow. In particular, there are alternative measures of the vector detailing the economic performance of the various national origin groups. In other words, the vector y,, could have been observed at any other time period, such as time e, and equation (3) could be defined, in general, as

(3‘)

y(e, t ) -

~ ( e7 ,) = C YJe (pJ,- P,J. I

Using this methodological framework, table 1.9 reports the predictions using equation (2) for the various measures of skills and labor market characteristics obtained in the sample of all immigrants, and table 1.10 presents the same statistics for the sample of recent immigrants. To understand the construction of these tables better, it is instructive to discuss in detail the results reported in a particular panel of table 1.10. Consider, for instance, the panel referring to the educational attainment of the immigrant population. As I documented in the last section, the average educational attainment of recent immigrants (relative to natives) was 0.7 years in 1940, 0.4 years in 1960, -0.2 years in 1970, and -0.7 years in 1980. These numbers are given by the diagonal terms in the educational attainment matrix of table 1.10. In effect, the diagonal of the matrix simply reports the result of the calculation defined by equation (1). The off-diagonal terms in the matrix report the results of the calculation using equation (2) for various combinations Of PJI and YJT’ The entries in any single column of the matrix reveal the extent to which changes in the national origin mix of the immigrant population alter the average characteristics of immigrants holding constant the vector of economic outcomes y. The first column of the educational attainment matrix indicates that, if the educational attainment of particular national origin groups (relative to natives) had remained constant over time (i.e., at the level reported in the 1940 Census), the change in the national origin mix of the immigrant flow alone would have led to a decline in the relative educational attainment of successive immigrant waves: 0 by 1960, -0.4 by 1970, and -0.5 by 1980. Thus, the changing national origin mix caused a drop in (relative) educational attainment among successive immigrant waves of about 1.2 years in the postwar period. Alternatively, if the educational attainment of the various national origin groups were held constant in terms of their 1980 values, the last column of

36

George J. Borjas

Table 1.9

Predicted Immigrant Outcomes (relative to natives) under Alternative National Origin Distributions: Sample of All Immigrants Value of y,, Obtained from: 1940 Census

1950 Census

1960 Census

1970 Census

Average education using national origin mix of foreign-born population in: 1940 Census - 1.864 - 1.700 - 1.227 - ,774 - 1.911 - 1.759 - 1.312 - ,847 1950 Census - 1.809 - 1.695 - 1.289 - .820 1960 Census - 1.757 - 1.677 -1.185 - ,755 1970 Census - 2.097 -2.051 - 1.293 - ,831 1980 Census

1980 Census

- ,335 - ,431 - .457 - .562 - .842

Average labor force participation rate using national origin mix of foreign-born population in: 1940 Census - ,020 - ,017 - ,021 - ,005 ,005 1950 Census - ,019 - .016 - ,021 - ,005 .005 - ,018 - ,014 - ,018 - ,003 .008 1960 Census 1970 Census - .017 - ,012 - ,017 - ,002 .010 - .018 - .014 - .023 - ,007 ,007 1980 Census Average unemployment rate using national origin mix of foreign-born population in: 1940 Census .027 ,010 ,003 ,003 - ,003 .004 - ,003 1950 Census ,028 ,011 ,003 ,004 - ,003 1960 Census ,027 .013 ,003 - .002 1970 Census .030 ,015 ,004 ,006 .008 ,003 1980 Census ,039 ,020 ,005 Average weeks worked using national 1940 Census - ,032 - ,034 1950 Census - ,031 1960 Census 1970 Census - ,033 - ,041 1980 Census

origin mix of foreign-born population in: - ,012 - ,014 - ,015 - ,014 - .016 - ,016 - ,013 - ,018 - .019 - ,016 - .033 - ,026 - ,029 - .053 - ,037

- ,013 - ,014 - ,016 - .022 - .036

Average log annual earnings using national origin mix of foreign-born population in: 1940 Census ,060 ,049 .043 ,053 ,057 ,040 1950 Census .036 ,031 ,024 ,037 1960 Census ,015 ,013 ,001 .018 ,019 - ,030 - ,043 - ,073 - ,037 - ,043 1970 Census -.I44 - ,180 -.I19 -.149 1980 Census -.I33 Average log wage rate using national origin mix of foreign-born population in: 1940 Census .138 ,085 ,084 ,092 ,082 1950 Census ,116 ,067 ,068 .077 .067 1960 Census ,083 ,043 ,045 .058 .049 - ,018 - ,012 .010 - .003 1970 Census ,029 - .075 -.I12 - .092 - ,057 - ,083 1980 Census Average adjusted log wage rate using national origin mix of foreign-born population in: 1940 Census ,124 ,071 ,069 ,057 .037 ,031 1950 Census ,113 ,062 ,063 .051 .084 ,040 ,049 ,036 .020 1960 Census - .007 - .004 - .001 - ,018 1970 Census ,043 1980 Census -.011 - ,062 - .052 - ,052 - ,062

37

National Origin and the Skills of Immigrants

Table 1.10

Predicted Immigrant Outcomes (relative to natives) under Alternative National Origin Distributions: Sample of Recent Immigrants Value of y,, Obtained from:

I940

I960

1970

1980

Census

Census

Census

Census

Average education using national origin mix of: 1935-40 flow ,741 ,677 1955-60 flow - ,007 .412 1965-70 flow - ,358 ,641 1975-80 flow - ,492 ,870

,823 ,102 - ,217 ,089

1.363 ,269 - ,448 - .663

Average labor force participation rate using national origin mix of: ,022 1935-40 flow - ,034 ,018 1955-60 flow - ,073 ,006 ,013 1965-70 flow - ,165 - ,017 - ,005 1975-80 flow - ,063 - ,036 - .033

- ,006 - ,021 - ,040 - ,056

Average unemployment rate using national origin mix of: I93540flow ,029 - .003 1955-60 flow ,018 ,003 1965-70 flow ,006 ,001 1975-80 flow - ,003 - ,004

,006 ,012

- .004 ,007 ,020

,014

.018

- ,052 -.lo1

- ,082 - ,093 -.122 - ,137

Average annual earnings using national origin mix of: 193540flow - ,092 - .I56 -.I80 1955-60 flow - .263 1965-70 ROW - ,165 - ,425 1975-80 flow - .358 - ,520

- ,014 -.I42 - ,289 - .349

- ,053 - .192 - ,383 - ,480

Average log wage rate using national origin mix of: 193540flow - ,032 - ,050 1955-60 flow - ,077 - ,128 1965-70 flow - ,097 - .218 1975-80 flow - ,210 - .272

,054 - .05 1 -.160 - ,201

,038 - ,086 - ,233 - ,299

-.I38

- ,001 - ,077 - ,200 - ,224

Average log weeks worked using national origin mix ofi 1935-60 flow - .074 - ,093 1955-50 flow -.I32 -.I16 1965-70 flow -.lo2 -.I64 1975-80 flow -.211 -.191

Average adjusted log wage rate using national origin mix of: 193540flow - .026 - ,093 1955-60 flow - ,030 -.113 1965-70 flow - ,028 - ,225 1975-80 flow - .067 - ,256

,011

- ,065 - ,094

-.I60 - .I49

- ,204

38

George J. Borjas

the matrix indicates that immigrants would have had 1.4 years more schooling than natives in 1940 but that this statistic would have declined to 0.3 in 1960, to -0.4 in 1970, and to -0.7 in 1980. Therefore, the changing national origin mix would be responsible for a drop of over two years in the average educational attainment of the immigrant population. Generally, the educational attainment matrix in table 1.10 indicates that, for most sets of weights used, the changing national origin mix of the immigrant flow is responsible for a sizable decline in the relative educational attainment of successive immigrant waves. The comparison of the entries in any given row of the matrix provides information on how the average educational attainment of a particular mix of national origin groups is changing over time (because it holds constant the national origin mix of the flow). For instance, the top row of the matrix indicates that, if the national origin mix of recent immigrants had remained constant at the 1940 level, the average educational attainment of recent immigrants would have increased from 0.7 in 1940 to about 1.4 in 1980. This indicates that the (relative) education level of immigrants originating in the source countries that formed the bulk of immigration in 1935-40 increased during the postwar period. By contrast, the last row of the matrix indicates that, given the national origin mix of recent immigrants in 1980, the average educational attainment of immigrants declined since 1960. In other words, the average education of an immigrant originating in the countries that make up the bulk of immigration today declined over time. The remaining matrices presented in table 1.9 and particularly in table 1.10 generally reinforce the link between the deteriorating labor market performance of immigrants and the changing national origin mix of the immigrant flow. Consider, for instance, the labor force participation rate matrix. Using the labor force participation data of recent immigrants reported in the 1940 Census, which are heavily influenced by the Great Depression, the participation rate of immigrants (relative to natives) declined from - 3.4 percentage points in 1940 to -6.3 percentage points in 1980. Using the participation data reported in the 1980 Census, the decline is from - 0.6 to - 5.6 percentage points. The changing national origin mix, therefore, generated a 3-5 percentage point drop in the labor force participation rate. The trends revealed by the unemployment rate are less clear, probably because of the pervasive role played by the Great Depression in the unemployment data reported in the 1940 Census. The unemployment data available in either the 1970 or the 1980 Census, however, lead to results more consonant with the thrust of the evidence. These data indicate that the changing national origin mix of immigrants caused a 1-2 percentage point increase in the unemployment rate among successive immigrant waves. The weeks-worked matrix more clearly shows the role of national origin in the employment of immigrants. Using the average (log) weeks worked for the various national origin groups reported by the 1940 Census (relative to na-

39

National Origin and the Skills of Immigrants

tives), the 1940 national origin mix leads to immigrants working an average of about 7.4 percent fewer weeks than natives, but the 1980 national origin mix leads to immigrants working about 21.1 percent fewer weeks than natives. Similarly, using the weeks-worked data reported by the 1980 Census, the predicted decline in immigrant labor supply is from -8.2 percent to - 13.7 percent. Therefore, the changing national origin mix is responsible for at least a 5 percent decline in the number of weeks worked across successive immigrant waves. Perhaps the most revealing results are given by the matrices showing the effect of national origin on (log) annual earnings and (log) wage rates. Using the 1980 data on the relative annual earnings of the various national origin groups, the 1940 national origin mix implies that immigrants earn about 5.3 percent less than natives. The 1980 national origin mix, however, is responsible for an immigrant flow that earns approximately 48 percent less than natives. Similarly, the wage rate data indicate that the 1940 national origin mix would lead to immigrants earning 3.8 percent more than natives, while the 1980 national origin mix leads to immigrants earning 29.9 percent less than natives. The various matrices reported in table 1.10, therefore, unambiguously indicate that shifts in the source country composition of the immigrant flow are responsible for a substantial decline in immigrant skills and for a deterioration in the labor market performance of successive immigrant waves over the postwar period. Moreover, this same factor is responsible for the deterioration in the occupational distribution of immigrants (relative to natives). This finding is documented in table 1.11, which uses the occupational distribution data reported in the 1980 Census to illustrate the nature of the results. In view of the large number of statistics that would be generated if the occupational dis-

Table 1.11

Predicted Occupational Distribution (relative to natives) under Alternative National Origin Distributions

Managerial

Technical

Service

Farming

Predicted propensity using national origin mix of all immigrants in: 1940 Census ,040 - ,033 ,018 -.019

1950 Census 1960 Census 1970 Census 1980 Census

,032 .032 ,024 ,005

- ,034

- ,032 - ,029 - ,030

,021 .024 ,033 ,041

-.017 -.014 -.Oll

-.ooo4

Crafts

Operators

.030 ,028 ,020 .004 -.019

- ,035

- .042 - ,045 - ,052 - ,064

- .081 - ,041 .002 ,026

- .030 - ,029 - ,020 .003

Predicted propensity using national origin mix of recent immigrants in:

1935-40 flow 1955-60 flow 1965-70 flow 1975-80 BOW

,168 .099 ,025 - ,004

- ,030 - ,032

- ,027 - ,024

,009 ,030 ,060 ,061

-.024 -.010 -.008 ,006

40

George J. Borjas

tribution data available in other Censuses were used, the use of a single Census helps focus the results of the analysis. As noted earlier, the 1980 Census reveals that the most recent immigrants are - 0.4 percentage points less likely to be managers than natives. Table 1.11 shows that, if the national origin mix had been the same as that which characterized the 1935-40 flow, the percentage of immigrants who are managers would have been 16.8 percentage points higher than that of natives. Conversely, the 1935-40 national origin mix predicts the immigrants are 8.1 percentage points less likely than natives to be operators or laborers, but the 1975-80 national origin mix implies that recent immigrants will be 2.6 percentage points more likely to be operatives. The changing national origin mix is therefore responsible for a shift in the occupational distribution of immigrants, away from managerial, professional, and craft occupations, and toward service, farming, and laborer occupations. l l It is of interest, of course, to determine the extent to which the changing national origin mix “explains” the decline in immigrant skills. In other words, how important is the change predicted by equation (3) in terms of the total change? Because the data on yj, (for recent immigrants) can be chosen arbitrarily from any of four decennial Censuses, there are a number of answers to this question. To summarize the nature of the evidence easily, I use the vector y,, estimated from the 1980 Census. The choice of alternative vectors does not alter the qualitative nature of the results. Table 1.12 reports the results using the sample of recent immigrants. As before, it is instructive to work through the results on educational attainment in order to understand the implications of the data. Consider, for example, the change’in educational attainment between the 1955-60 and the 1975-80 immigrant waves. During this period, the relative educational attainment of recent immigrants declined by about 1.1 years. Table 1.12 indicates that, on average, national origin alone is responsible for a - .9-year decline over that period, or about 85 percent of the observed decline. The remaining rows of the table indicate that the changing national origin mix “explains” (and, in some cases, overexplains) the 6.2 percentage point drop in the labor force participation rate, the 2.1 percent drop in weeks worked, the 21.7 percent drop in annual earnings, the 17.1 percent drop in the wage rate, and the 11.1 percent drop in the adjusted wage. In addition, the analysis suggests that national origin is responsible for much of the change in the occupational distribution of successive immigrant waves. For instance, the changing source country distribution of immigrants caused a 10.3 percentage point drop in the fraction of immigrants who are in managerial occupations and a 6.7 percentage point rise in the fraction of im11. The data reported in sec. 1.1 indicate that there has been little change in the industrial distribution of immigrants over time (relative to that of natives), with the exception of agriculture. The decomposition of the observed changes in the industrial distribution are uninteresting and are omitted from the paper.

41 Table 1.12 ~~

National Origin and the Skills of Immigrants Decomposition of Changes in Immigrant Outcomes (relative to natives) in the Postwar Period

~

Change between: 1935-40 & 1975-80 Waves

Variable Education Labor force participation rate Unemployment rate Log (weeks worked) Log (annual earnings) Log (wage rate) Adjusted log (wage rate) Fraction managerial Fraction technical Fraction service Fraction farming Fraction crafts Fraction operators

Average Change

- 1.404 - ,022 - ,011 - ,063 - ,368 - ,261 -.I98 - ,091 - ,038 ,028 ,117 - ,049 .033

Change Due to National Origin -2.026

- ,050 .022 - .055 - ,423 - .337 - .223 -.I72 ,006 ,052 .030 - ,022 ,107

1955-60 & 1975-80 Waves

Average Change

1965-70 & 1975-80 Waves

Change Due Change Due to National Average to National Origin Change Origin

1.075

- .932

- ,446

- ,215

- ,062 ,015 - .021 - ,217 - ,171 -.I11 - .031 ,005 ,020 ,041 - .073 ,023

- ,036 ,011 - .044 - ,288 - ,213 - ,147 - ,103

- ,051

- ,016 - .002 - ,015 - ,097 - ,066 - .024 - ,029 ,003

-

,008

,031 ,016 - .019 ,067

,006

- ,043 - ,191 - .I39 - ,075 - ,048 ,019 ,012 ,026 - ,014 ,005

,001

.014 - .012 ,024

migrants who are operators or laborers. Both these changes greatly exceeded the actual changes that occurred over the period.

1.4 Why Does National Origin Matter? The study of the post-1940 decennial Censuses reveals that a single variable, the changing national origin mix of the immigrant flow, provides a coherent (and simple) understanding of many of the trends in the skills and labor market experiences of successive immigrant waves during the postwar period. This result, however, does not provide an explanation of why national origin should matter so much. The importance of national origin as a determinant of the labor market performance of immigrants is the focus of recent research (Borjas 1985, 1987; Jasso and Rosenzweig 1986). This literature is based on the hypothesis that, as long as immigration is motivated by the search for better employment and earnings opportunities, the immigrant flow will be self-selected from the population at risk and will be self-selected differently in different source countries. Moreover, the skills and abilities that the various national origin groups bring with them to the United States are not equally transferrable across countries. Therefore, there is likely to be considerable dispersion in economic opportunities among national origin groups in the United States, even if the

42

George J. Borjas

groups have the same observable socioeconomic and demographic characteristics. Consider the link between the changing national origin mix of the immigrant flow and the decline in the relative schooling level of successive immigrant waves. Partly, this arises because the populations of the source countries responsible for the bulk of immigration today have relatively little schooling. In Mexico, the average schooling level is only six years, while in the Philippines it is eight years. By contrast, persons who migrated in the 1950s or early 1960s tended to originate in countries with a relatively well-educated work force. The typical person living in Germany or in the United Kingdom has about eleven years of schooling. The empirical importance of this insight is documented in the first column of table 1.13, which reports the average schooling level in the country represented by the typical immigrant. The mean educational attainment data for the various source countries are obtained from Borjas (1991, table 2) and gives the average years of schooling in the source country during the 1970s. The statistics presented in the first column of table 1.13 are a weighted average of these educational attainment data, with the weights being the fraction of the immigrant flow originating in a specific country. The data in table 1.13 indicate that the educational attainment of the source country responsible for the “average” immigrant between 1935 and 1940 was 10.2 years. This statistic declined to 9.5 years for the 1955-60 flow, to 8.5 years for the 1965-70 flow, and to 7.7 years for the 1975-80 flow. Therefore, the average educational attainment of the typical source country represented in the immigrant flow declined by about 2.5 years since 1940 and by 1.8 years since i960. This fact alone, therefore, implies that the typical immigrant today-even if he or she were randomly selected from the population of the source countries-would be less educated than earlier immigrants. Table 1.13

Average Characteristics of Source Countries, Weighted by National Origin Mix of Immigrant Flow Variable

Average Using National Origin Mix of Foreign-born Population in:

Education

Income Inequality

Per Capita GNP

1940 Census 1950 Census 1960 Census 1970 Census 1980 Census

10.00 9.89 9.74 9.22 8.33

4.15 4.35 4.84 5.69 7.44

7,598 7,394 7,192 6,260 4,862

193540 flow 1955-60 flow 1965-70 flow 1975-80 flow

10.18 9.45 8.45 7.68

4.32 5.53 6.79 8.77

8,588 6,823 4,566 3,828

43

National Origin and the Skills of Immigrants

However, persons are not randomly allocated into the immigrant flow. The economic theory of self-selection implies that highly educated persons in the country of origin are more likely to migrate if the American labor market rewards their education more than the source country does. Alternatively, the United States will attract less-educated workers if schooling is better rewarded in the source country. Unfortunately, extensive (and reliable) data on intemational differences in the rate of return to education do not exist. For instance, Psacharopoulos’s often-cited (1973) study reports schooling rates of return for only fourteen countries that are important sources of immigration to the United States. In previous research (Borjas 1991, p. 36), I have estimated that a one-year increase in the mean educational attainment of the source country increases the average education level of the self-selected immigrant flow by 0.2 years. Because the populations of the source countries responsible for the new immigration have relatively little schooling, the new immigrants are likely to have less education than the old. In fact, this factor is responsible for a decline of 0.4 years in the average educational attainment of immigrants (relative to natives) between the 1955-60 and the 1975-80 waves. Therefore, if a year of schooling increases earnings by about 10 percent, the increasing gap between immigrant and native educational attainment is responsible for a 4 percentage point drop in the relative earnings of immigrants. Of course, education is only one of a large number of different types of skills and abilities that determine a person’s earnings, and a summary measure of the prices of skills is needed to assess whether a favorable or an unfavorable skill sorting takes place overall. The application of Roy’s (195 1) self-selection model to the study of immigration (Borjas 1987) suggests that such a summary measure is given by the amount of dispersion in a country’s income distribution. An economy with an egalitarian income distribution offers relatively low returns to skills. Because persons migrate to countries that provide the best economic opportunities, the immigrant flow originating in source countries with less income inequality than the United States will have above-average skills or productivities. Alternatively, the returns to skills are higher in source countries that have more income inequality than the United States. Highly skilled persons then face relatively better economic opportunities in the country of origin and have little incentive to migrate to the United States. The immigrant flow, therefore, will contain a relatively large number of unskilled workers. The link between the shape of the income distribution in the source country and the skill composition of the immigrant flow provides an additional explanation of why the old immigrants are relatively more skilled than the new. In the 1940s and 1950s, a large fraction of immigrants originated in western European countries. Today, the immigrant pool is much more likely to originate in Asia or Latin America. The second column of table 1.13 documents

44

George J. Borjas

the change that occurred in the income dispersion of the source countries represented by the typical immigrant during the postwar period. The typical person who immigrated between 1935 and 1940 originated in a country where the ratio of the income accruing to the top 10 percent of households to that accruing to the bottom 20 percent of households was 4.3. This statistic increased to 5.5 for the 1955-60 flow, to 6.8 for the 1965-70 flow, and to 8.8 for the 1975-80 flow. By this measure of income inequality, therefore, the amount of dispersion in the average immigrant’s source country doubled in the postwar period, with most of that increase occurring after 1960. In earlier work (Borjas 1991, table 5 ) , I estimated that a one-unit increase in this measure of income inequality is associated with a -0.004 unit decline in the (log) earnings of immigrants in the United States (after holding constant the demographic characteristics of immigrants). Thus, the increase in income inequality in the source countries responsible for immigration between the 1955-60 and the 1975-80 waves is responsible for a 1.3 percentage point decline in the earnings of immigrant waves over the period. Finally, national origin influences the labor market performance of immigrants in the United States because source countries differ dramatically in their level of industrialization and economic development. Clearly, the kinds of skills that workers acquire in highly developed economies are not the same as those acquired in the less-developed countries. It is likely, therefore, that skills acquired in advanced economies can easily be transferred to the U.S. labor market and that skills acquired in less-developed countries are much less useful to American employers. In fact, even after controlling for differences in demographic characteristics among immigrants, there is a strong positive correlation between immigrant earnings and the level of economic development in the country of origin, as measured by the country’s per capita GNP. Immigrants who originate in highincome countries have higher earnings than otherwise similar immigrants who originate in less-developed countries. In fact, doubling the source country’s per capita GNP increases the lifetime earnings of immigrants in the United States by 5 percent (Borjas 1991, table 5). The last column of table 1.13 reports the 1980 per capita GNP of the source country representing the typical immigrant. The average person who immigrated between 1935 and 1940 originated in a country with a 1980 per capita GNP of $8,588 (in 1980 dollars). By contrast, the respective statistic for the typical immigrant is $6,823 in the 1955-60 flow, $4,566 in 1965-70, and $3,828 in 1975-80. The changing national origin mix of successive immigrant waves cut by more than half the per capita GNP of the country represented by the typical immigrant, with most of this decline occurring after 1960. Because the elasticity of immigrant earnings in the United States with respect to per capita GNP in the source country is .05, immigrants who arrived in the late 1950s will earn about 4 percent more than demographically comparable immigrants who arrived in the late 1970s.

45

National Origin and the Skills of Immigrants

Table 1.4 above implies that there was a 15.7 percent decline in the (relative) immigrant wage rate between the 1955-60 and the 1975-80 immigrant waves.'* The decrease in the level of economic development in the countries responsible for immigration to the United States and the increase in the extent of income inequality characterizing these source countries together account for a 5 percent decline. The deteriorating educational attainment is responsible for an additional 4 percent drop. Therefore, these factors alone explain about 60 percent of the decline in earnings between these two immigrant waves. 1.5

Summary

This paper presented a study of the historical experience of immigrants in the U.S. labor market between 1940 and 1980. The analysis used the five available Public Use Samples of the U.S. Census to study the trends in the skills and labor market performance of successive immigrant waves over the postwar period. The analysis leads to a number of substantive empirical findings. 1. The comparison of successive immigrant waves entering the United States in the last five decades reveals a major decline in their skills and a deterioration in their labor market performance. The most recent waves have significantly lower earnings and labor force participation rates, work fewer weeks, and have higher unemployment propensities than earlier waves. In addition, the data indicate a substantial worsening in the occupational distribution of immigrants, with more recent immigrant waves less likely to be employed in the managerial and professional occupations and more likely to be employed as laborers or operators. 2. One single factor, the changing national origin mix of the immigrant flow, is mostly responsible for these historical trends. Because of changes in immigration policy and in economic and political conditions both in the United States and abroad, the new immigrants are more likely to originate in Latin America and in Asia than earlier waves. The Census data document substantial dispersion in the skills and labor market performance of various national origin groups. The data also indicate that, if the national origin mix of immigrant waves had remained unchanged over the postwar period, the decline in the skills and labor market performance of successive immigrant waves either would not have occurred or would have been greatly tempered. 3. National origin matters because source countries differ in various economic characteristics that are important determinants of the national origin group's labor market performance in the United States. In particular, the new immigrant waves are originating in countries with less-educated populations, 12. This statistic is obtained by taking the antilog of the change in the relative log wage between these two waves.

46

George J. Borjas

lower per capita GNP, and less-egalitarian income distributions. Each of these factors is responsible for a decline in immigrant skills and productivities among successive immigrant waves. Together, these factors account for about 60 percent of the wage differential between the immigrants who arrived in the late 1950s and those who arrived in the late 1970s.

References Abbott, Michael G., and Charles M. Beach. 1987. Immigrant Earnings Differentials and Cohort Effects in Canada. Queen’s University. Mimeo. Abowd, John M., and Richard B. Freeman, eds. 1991. Immigration, Trade, and the Labor Market. Chicago: University of Chicago Press. Abrams, Elliott, and Franklin S. Abrams. 1975. Immigration Policy: Who Gets In and Why? Public Interest 38:3-29. Blau, Francine. 1980. Immigration and Labor Earnings in Early Twentieth Century America. Research in Population Economics 2:21-41. Borjas, George J. 1985. Assimilation, Changes in Cohort Quality, and the Earnings of Immigrants. Journal of Labor Economics 3:463-89. . 1987. Self-selection and the Earnings of Immigrants. American Economic Review 7 7 5 3 1-53. . 1989. Immigrant and Emigrant Earnings: A Longitudinal Study. Economic Inquiry 27:21-37. . 1990. Friends or Strangers: The Impact of Immigrants on the US.Economy. New York: Basic. . 1991. Immigration and Self-selection. In Abowd and Freeman (1991). Borjas, George J., and Bernt Bratsberg. 1990. Who Leaves? The Outmigration of the Foreiin-Born. University of California, San Diego, June. Mimeo. Carliner, Geoffrey. 1980. Wages, Earnings, and Hours of Work of First, Second and Third Generation American Males. Economic Inquiry 18:87-102. Chiswick, Barry R. 1978. The Effect of Americanization on the Earnings of ForeignBorn Men. Journal of Political Economy 86:897-921. DeFreitas, Gregory. 1980. The Earnings of Immigrants in the American Labor Market. Ph.D. diss., Columbia University. Eichengreen, Barry, and Henry A. Gemery. 1986. The Earnings of Skilled and Unskilled Immigrants at the End of the Nineteenth Century. Journal of Economic History 46:441-54. Hutchinson, E. P. 1981. Legislative History of American Immigration Policy, 17981965. Philadelphia: University of Pennsylvania Press. Jasso, Guillermina, and Mark R. Rosenzweig. 1982. Estimating the Emigration Rates of Legal Immigrants Using Administrative and Survey Data: The 1971 Cohort of Immigrants to the United States. Demography 19:279-90. . 1986. What’s in a Name? Country-of-Origin Influences on the Earnings of Immigrants in the United States. Research in Human Capital and Development 4175-106. . 1988. How Well Do U.S. Immigrants Do? Vintage Effects, Emigration Selectivity, and Occupational Mobility. Research in Human Capital and Development 61229-53. Kitigawa, Evelyn M. 1955. Components of a Difference between Two Rates. Journal of the American Statistical Association 50: 1168-94.

47

National Origin and the Skills of Immigrants

Oaxaca, Ronald. 1973. Male-Female Wage Differentials in Urban Labor Markets. International Economic Review 14:693-709. Psacharopoulos, George. 1973. Returns to Education: An International Comparison. San Francisco: Jossey-Bass. Roy, Andrew D. 1951. Some Thoughts on the Distribution of Earnings. Ugord EconomicPapers 3:135-46. U.S. Immigration and Naturalization Service. Various issues. Statistical Yearbook. Washington, D.C.: U.S. Government Printing Office. Warren, Robert, and Jennifer Marks Peck. 1980. Foreign-born Emigration from the United States: 1960-1970. Demography 17:71-84.

This Page Intentionally Left Blank

2

Out-Migration and Return Migration of Puerto Ricans Fernando A. Ramos

The study of the movement of persons between Puerto Rico and the United States differs from most of the studies in the international migration literature in two important ways. First, Puerto Ricans do not face any statutes restricting either their exit from Puerto Rico or their entry into the host country. Because Puerto Rico is a territory of the United States, Puerto Ricans are U.S. citizens and can therefore move freely between the two “countries.”The size and composition of migration flows can, in effect, be attributed entirely to differences in social and economic factors between the sending and the receiving regions. Second, Puerto Rican migration to the United States is characterized by a large probability of return migration. Unlike most return migration flows, the size and skill composition of that to Puerto Rico are well documented in publicly available data sets. As is well known, nonrandom return migration propensities can generate an erroneous portrayal of Puerto Rican assimilation into the United States in cross-sectional data sets. For instance, if only the most successful migrants remain in the United States, the cross-sectionalcorrelation between the U S . earnings of Puerto Ricans and years since migration will be positive, even in the absence of any true assimilation or aging effects.’ It is also likely that many Puerto Ricans did not perceive their migration to the United States as permanent. Many migrated with the expectation that they

Fernando A. Ramos is a manager at the Policy Economics Group of KPMG Peat Marwick. The author wishes to thank George Borjas, Richard Freeman, Edward Funkhouser, Lawrence Katz, Lawrence Summers, and the participants in Harvard University’s Labor Seminar for their valuable comments and suggestions. Financial support from the Ford Foundation is greatly appreciated. The views presented in this paper are those of the author and do not necessarily reflect those of KPMG Peat Marwick. 1. Borjas (1985), Carliner (1980). and Chiswick (1978) study the assimilation of immigrants in the U.S. labor market.

49

50

Fernando A. Ramos

would return to their homeland.* A study of return migration flows can, therefore, significantly increase our understanding of the social and economic consequences of immigrati~n.~ I will use the U.S. Census of Population to investigate migration and return migration decisions of Puerto Rico-born and U.S.-born Puerto Ricans between the United States and Puerto Rico. These data will be used to test a version of the self-selection migration model developed by Borjas (1987). This model is based on the assumption that migration flows are generated as persons choose to reside in countries that maximize their economic wellbeing. The income maximization hypothesis has an important prediction: skilled workers will choose to reside in the country that offers a higher rate of return for their skills. My results indicate that the skill composition of Puerto Rican migration flows is consistent with the predictions of the Borjas model. First, migrants in the United States have less advantageous observable socioeconomic characteristics (such as education). This finding is consistent with the fact that Puerto Rico has a much more unequal income distribution and that it offers higher returns to skills than the United States. Second, return migrants to Puerto Rico tend to be more skilled than the Puerto Ricans who remain in the United States. Furthermore, U.S.-born Puerto Ricans moving to Puerto Rico also have more human capital than U.S.-bornPuerto Ricans who choose to remain in the United States. 2.1

Puerto Rican Migration to the United States

Migration has been an important aspect of Puerto Rico’s economic development for the past four decades. Unfortunately, the only source of historical data on the migratory flows is the net flow of passengers at the airport in San Juan, Puerto Rico. In the 1950s, there was an annual average outflow of 45,800 passengers. This outflow decreased to an average of 27,300 between 1960 and 1969 and 24,300 between 1970 and 1979.4 Table 2.1 shows population figures for Puerto Ricans residing either in Puerto Rico or in the United States proper in 1980. I define Puerto Ricans as individuals who either were born in Puerto Rico or are of Puerto Rican heri2. Intended temporary migration is especially important in the cases of migrant workers (who are permitted by the country of destination to remain for only a limited period of time) and of groups who are allowed unrestricted access between the place of origin and the place of destination (such as internal migration and the Puerto Rican case). Those permitted unrestricted access can move back and forth without the need to make a more permanent decision about their residential choice. For a detailed explanation of the different types of return migration, see King (1986) and Bohning (1984). 3. King (1986) surveys the return migration literature and presents eleven return migration case studies. 4. These data are reported by the Puerto Rico Planning Board (1983).

51

Out-Migration and Return Migration of Puerto Ricans

Table 2.1

Puerto Rican Population Living in Puerto Rico and in the United States in 1980 Living in Puerto Rico

Total population Total Born in: Puerto Rico United States Other % of Puerto Rico born % of U.S. born Age 20-64 Total Born in: Puerto Rico United States Other 8 of Puerto Rico born %ofU.S. born

Living in the United States

N

%

N

%

3,097,000

100.0

2,O 14,000

100.0

2,889,000 177,000 3 1,000

93.3 5.7 1.o

930,600 1,014,500 68,900

46.2 50.4 3.4

75.6 14.9

24.4 85.1

1,520,000

100.0

1,022,000

100.0

1,461,600 45,000 14,400

96.1 3.0 .9

712,100 265,300 44,600

69.8 25.9 4.3

67.2 11.3

32.8 88.7

Source: Author estimates from U.S. (1/100) and Puerto Rico (5/100) Census tapes.

tage (i.e., have at least one parent born in Puerto Rico). Throughout the analysis, I will refer to the United States as a political (and geographic) entity that does not include the island of Puerto Rico. The total Puerto Rican population in Puerto Rico in 1980 was 3.1 million persons, of whom 2.9 million were born in Puerto Rico. There were 2.0 million Puerto Ricans in the United States, 46.2 percent of whom were born in Puerto Rico. If we restrict the calculations only to those persons born in Puerto Rico, we find that 24.4 percent of the Puerto Rico-born population was living in the United States in 1980. Borjas (1987) calculated similar shares for most of the source countries with sizable migration flows to the United States. The results for selected countries are reported in table 2.2. Borjas found that the country with the largest share of its population living in the United States was Jamaica, with 10.3 percent. The share of the Puerto Rican population living in the United States is almost 2.5 times greater than the largest share for other countries. This difference clearly reflects the fact that Puerto Rican migration is not hindered by political restrictions on the ability to enter (or leave) the United States: Puerto Ricans are U.S. citizens, so they have unrestricted access to the United States proper. As an alternative measure, therefore, I calculate the average share of per-

52

Fernando A. Ramos

Table 2.2

International and Internal Migration Flows in the United States, 1980 Migrants as 9% of Population of Origin Country Greece Ireland Canada Cuba Dominican Republic Jamaica Panama Trinidad & Tobago Puerto Rico

2.4 3.5 2.8 6.3 4.3 10.3 2.6 8.0 24.4

Stare

Alaska California Florida Georgia Hawaii Louisiana Massachusetts Michigan Mississippi Missouri Nevada New York Pennsylvania Texas Virginia

49.9 19.6 27.2 29.3 33.7 24.2 29.4 21.8 44.7 36.2 54.3 30.6 29.8 21.8 33.6

All states

31.0

Sources: Country data from Borjas (1987); state data from the 1980 census (1/1OOO).

sons in the United States who live in a state other than the one in which they were born. The results are also presented in table 2.2. The average share for the United States is 31 .O percent, that is, about one-third of the persons in the United States reside in a state different from their state of birth. Puerto Rico’s share exceeds that of only California (19.6), Texas (21.8), Michigan (21 .S), and Louisiana (24.2). The relative size of Puerto Rican migration flows, therefore, is large relative to international flows but small relative to intranational flows. Throughout this paper, I will continue to compare Puerto Rican migration to international flows. The cultural and language differences between h e r t o Rico and the United States are more likely to resemble those encountered by international migrants than those encountered by internal migrants in the United States.

53

Out-Migration and Return Migration of Puerto Ricans

2.2 Selection Model The migration model used in this paper is an application of the Roy model and was introduced into the literature by Borjas (1987).* Following Borjas, migration takes place when expected earnings, net of migration costs, in the new country (country 1) are greater than in the source country (country 0). The earnings distribution in the country of origin is described by log w, = X p ,

(1)

+ e,,

where X is a vector of socioeconomic characteristics, and e, is a normally distributed random variable with mean zero and variance a;. Similarly, the wage structure in the United States is given by (2)

log w , = (1 - M ) X p ,

+ MXp, + el,

where M is a dummy variable equal to one if an individual is foreign born. The vectors p,, and (3, represent the returns to the socioeconomic characteristics of natives and migrants, respectively. These returns can differ because of discrimination or differences in the quality of the characteristics. The random variable el is normally distributed with mean zero and variance a:. A person residing in country 0 will migrate if he or she can earn more in country 1 (net of migration costs). The decision to migrate is summarized by the sign of the index function:

(3)

I = 1% [w,I(w, +

01

=

MP, -

- + (el - e0)L C/w,is a time-equivalent mea-

Po)

where C represents migration costs, and n = sure of these costs. For simplicity, I will assume that n is constant across individuals. Individuals will migrate if I > 0. The probability that individuals with characteristics X will migrate is given by (4)

Po) - TI} = 1 - cp (z), - [ X ( p , - Po) - T]/u,, and is the standard nor-

P ( X ) = pr{v > - [ X ( P , -

where v = el - e,, z = mal distribution function. The conditional expectations E[log w, I X , I > 01 and E[log w , I X , I > 01 give the expected wages of migrants prior to their migration as well as after their migration. Because of the normality assumptions, these conditional expectations are given by (5)

E(l0g W , I X, I > 0) = X p ,

+ [u~u,/u,](~ - u,/u,)X,

(6)

E(l0g W , I X, I > 0) = X p ,

+ [u,u,/u,](u,/u,- p)X,

where X = I$ (z)/P(X),I$ is the density of the standard normal distribution, and p is the correlation between the random variables e, and e l . The condi5 . I will describe the model only briefly. Borjas (1987, 1991) presents a more extensive derivation and discussion of the model.

54

Fernando A. Ramos

tional means in (5) and (6) can be used to identify the types of selection in unobserved characteristics that characterize the migrant flow from Puerto Rico to the United States, depending on the sign of the coefficient of A . As shown by Borjas, three types of selection are possible: positive, negative, and refugee selection. It follows from (5) and (6) that the necessary and sufficient conditions for each type of selection are as follows: a ) Positive selection. A high value of p and a more unequal distribution of income in the United States relative to the country of origin. This selection implies that migrants have above-average earnings both in Puerto Rico and in the United States. b) Negative selection. A high value of p and a more unequal distribution of income in the country of origin than in the United States. This selection implies that migrants have below-average earnings both in Puerto Rico and in the United States. c ) Refugee sorting. A small or negative value of p. This selection implies that migrants have below-average earnings in Puerto Rico and aboveaverage earnings in the United States.

Because of its political and economic association with the United States, Puerto Rico’s economy has adopted many U.S. economic institutions. It is likely, therefore, that p takes on a relatively high value. The Borjas model thus implies that the migration flow from Puerto Rico to the United States should be characterized by either positive or negative selection, depending on which of the two “countries” has a more unequal income distribution (i.e., offers a higher rate of return for skills). One.of the most frequently used measures of the distribution of earnings in a country is the Gini coefficient. The Gini coefficient for the Puerto Rican wage distribution in 1977 was 3.97, while it was only 3.57 for the U.S. wage distribution, thus reflecting a more unequal distribution of wage income in Puerto Rico.6 Furthermore, the Gini coefficient for Puerto Rico underestimates the true amount of wage inequality in the population because it incorporates wage information only on workers. The unemployment rate in Puerto Rico has been historically higher than the unemployment rate in the United States. In 1977, for instance, 19.9 percent of the Puerto Rican labor force was unemployed, while the unemployment rate in the United States was only 7.1 percent. Thus, the exclusion of the unemployed from the calculation of the Gini coefficients underestimates the Puerto Rican Gini coefficient more than the U.S. coefficient. More of the lower tail of the income distribution is truncated in the case of Puerto Rico. One of the reasons for the higher unemployment rate in Puerto Rico is the minimum wage (see Castillo-Freeman and Freeman, in this volume). In 1977, the U.S.-level minimum wage also began to apply to Puerto Rico. Since the 6. For a detailed analysis of the estimated Gini coefficients for h e r t o Rico and the United States, see Mann (1985) and Moroney (1978), respectively.

55

Out-Migration and Return Migration of Puerto Ricans

minimum wage generates unemployment among the least skilled, it truncates the lower tail of the earnings distribution, and the real measure of earnings dispersion will be higher than that estimated by the Gini coefficient. In order to obtain an estimate of the extent of income inequality that is not biased by the truncation due to the minimum wage, I analyzed the sample of individuals with more than a high school education. The truncation problem for this group is minimal because these workers are not likely to be affected by the imposition of a relatively high minimum wage. I calculated the variance of the logarithm of wages for this group. The results are consistent with the implications of the comparisons of Gini coefficients. The log variance of wages of these highly educated workers (high school graduates) is higher in Puerto Rico (0.325) than in the United States (0.298).’ I note, of course, that these calculated variances, which show a more unequal distribution in Puerto Rico than in the United States, do not necessarily measure true population variances. After all, as documented above, a large number of Puerto Ricans moved out of Puerto Rico. In order to measure the true difference in income distribution, one must make the comparison before the migration process began. The earliest comparable measure of inequality for the two economies is for 1947, around the period when Puerto Rican outmigration accelerated. Measured Gini coefficients for family income for 1947 are also higher for Puerto Rico (0.52) than for the United States (0.40).* Throughout the rest of the paper, I will rely on the estimated Gini coefficients and the earnings distribution for the college educated and assume that earnings are more unequally distributed in Puerto Rico than in the United States, both at the time of the initial migration wave and for more recent immigrants. The economic model of migration then predicts that we should observe negative selection on unobserved characteristics for migrants from Puerto Rico to the United States. We should also observe positive selection on unobserved characteristics for return migrants to Puerto Rico. As shown in Borjas (1991), migrants are also selected on observable characteristics. If the education (s) distribution for the population of the country of origin can be written as s = ps e,, where the random variable es is normally distributed with mean zero and variance us, Borjas has shown that the expected value of schooling for migrants can be expressed as

+

(7)

+

where t = (el - e,) (PI - Po)es, and PI and P, are the rates of return to education in the destination country and the country of origin, respectively. 7. An F-test reveals that the difference in the variances is statiistically significant at the 99 percent level of significance. The sample sizes are 3,817 for Puerto Rico and 15,076 for the United States (the U.S. sample was extracted from the 1/100 1980 U.S. Census microfile). The critical value is 1 .OO, which is smaller than the ratio of the variances (1.09). 8. For the Gini coefficient estimates for Puerto Rico and the United States, see Andic (1964) and Budd (1967), respectively.

56

Fernando A. Ramos

This expression predicts that the education level of the migrant pool will depend on the relative return to education in the two countries. If the return to education is higher in the country of destination ([p, - Po] > 0), there will be positive selection in schooling. Highly educated workers born in Puerto Rico as well as U.S.-born Puerto Ricans should migrate to the location with the higher returns to education. There are, therefore, two important testable implications of the selection model. First, we should observe negative selection on unobserved characteristics in the migration between Puerto Rico and the United States. Second, highly educated individuals should migrate to the location with the higher returns to education. For an intuitive explanation of these selection predictions, examine figure 2.1 and the migration index equation (3). Figure 2.1 illustrates the wage distribution of the Puerto Rican population. Since the variance of the U.S. income distribution is less than the variance of the Puerto Rican income distribution, the selection model predicts out-migration from the lower tail of the earnings distribution. The population to the left of d will migrate to the United States. The migration decision expressed in the index function (3) depends on relative wages and migration costs. The migration costs variable (T) includes monetary relocation costs as well as psychological costs of adjustment. The psychological costs include, among other things, adjustment to a different culture, a different climate, and life away from family and friends. While monetary relocation costs are easily observed ex ante, psychological costs are difficult to measure. Individuals may be able to measure adjustment costs correctly only after migrating. We can therefore divide the migration cost variable T into a component observed before migrating (monetary relocation costs) and one observed only after the migration decision (psychological adjustment costs). If observed psychological costs are higher than expected, total migration costs would increase and lead to a change in the sign of the migration index function for

Fig. 2.1 Distribution of unobserved characteristics of the Puerto Rican population

57

Out-Migration and Return Migration of Puerto Ricans

some migrants. Not only are these persons worse off in the United States than they expected, but they may actually gain by migrating back to Puerto Rico. The number of individuals permanently migrating is reduced to the area to the left of a'. Return migrants are measured by the area between a' and ci. Notice that return migration to Puerto Rico is characterized, as in the previous discussion, by positive selection (from the initially unskilled migrant flow). Return migrants are the most skilled among the original migrants.

2.3 Characteristics of Puerto Rican Migrants The data used in this paper are drawn from the 1980 U.S. Census Public Use Samples for Puerto Rico and the United States. To obtain large numbers of observations, my sample of Puerto Ricans residing in the United States is obtained by combining the 1/100 sample with the 5/100 ample.^ The sample of Puerto Ricans residing in Puerto Rico is extracted from 9100 Census for Puerto Rico. I am interested in distinguishing among four main groups in the Puerto Rico-born population: persons who, between 1975 and 1980, (1) migrated from Puerto Rico to the United States, (2) migrated from the United States to Puerto Rico, (3) resided in Puerto Rico, and (4) resided in the United States. I will also analyze two groups of U.S.-born Puerto Ricans: those who reside in Puerto Rico and those who reside in the United States. The questions asked in order to identify the migration status of the Puerto Rico-born population differ between the U.S. Census and the Puerto Rican Census. In the U.S. Census, individuals are asked where they were living in 1975. People'responding that they were living in Puerto Rico are defined as migrants (between 1975 and 1980). People residing in the United States in 1975 are included in the migrant group that moved prior to 1975. In the Puerto Rican Census, individuals were asked if they lived in the United States between 1970 and 1980 and when (what year) they returned to Puerto Rico. People who returned between 1970 and 1980 are included in the migrant group. I assume that those persons who did not reside in the United States in the previous ten years never migrated out of Puerto Rico. It is apparent that the migration variables that can be constructed from the Census have some significant shortcomings. For example, it is possible for recent migrants to the United States to have moved between Puerto Rico and the United States more than once in the past five years, but we observe only 9. For budgetary considerations, I used only 90 percent of the 51100 U.S. Census. Data from the 5/100 Census tape were extracted for twenty-two states-Colorado, Connecticut, Delaware, Florida, Georgia, Idaho, Indiana, Iowa, Maryland, Massachusetts, Nebraska, Nevada, New Hampshire, New Jersey, New Mexico, New York, Ohio, Pennsylvania, Texas, Virginia, Washington, and West Virginia-and the District of Columbia. The Puerto Rican population from these states accounts for 90 percent of all Puerto Ricans in the United States. Furthermore, the labor migration questions were asked only of a 50 percent subsample, so the Puerto Rican sample represents 2.75 percent of the total Puerto Rican population in the United States.

58

Fernando A. Ramos

one move. Similarly, we do not know when those who resided in the United States between 1975 and 1980 migrated to the mainland; we know only that the migration took place prior to 1975. Therefore, we cannot estimate rates of assimilation into the U.S. economy. Furthermore, it is possible that some of these migrants moved to Puerto Rico after 1975 but returned to the United States prior to the 1980 Census. Finally, the data allow me to identify return migrants to Puerto Rico only if they returned after 1970. Individuals who returned prior to 1970 are included in the group that never migrated. My sample is composed of men aged 20-64 who are not members of the armed forces, are not self-employed, are not enrolled in school, report an occupation, and have hourly wages below $lOO.'O The data used below contain 480 recent migrants to the United States, 4,846 migrants to the United States before 1975, 1,650 U.S.-born Puerto Ricans residing in the United States, 12,193 nonmigrants in Puerto Rico, 2,344 recent return migrants to Puerto Rico (1,209 returning between 1975 and 1980 and 1,135 returning between 1970 and 1974), and 381 U.S.-born Puerto Ricans living in Puerto Rico." Tables 2.3 and 2.4 present the average characteristics of the Puerto Rican population in Puerto Rico and the United States, respectively. The average age of the migrants reflects the fact that migration is predominantly undertaken by young persons. Migrants are younger than nonmigrants. Recent migrants to the United States (whose average age is 32.3) are much younger than nonmigrating Puerto Rican residents (38.1); recent return migrants to Puerto Rico (34.2) are younger than nonreturning migrants (38.8); and U.S.-born Puerto Rican migrants (28.5) are younger than those living in the United States (29.6) : Differences in years of schooling, the most important human capital measure available in the data, show that persons who migrated from Puerto Rico to the United States are less educated than Puerto Ricans who migrated in the opposite direction. Recent migrants to the United States (who have an average education of 10.4 years) and men who migrated to the United States before 1975 (9.4 years) are less educated than Puerto Ricans who never left Puerto Rico (10.8 years). It is also worth noting that return migrants to Puerto Rico, who have 9.8 years of schooling, are more educated than the pre-1975 mi-

10. Observations with wages above $100 are considered coding errors. Hourly wages were calculated as (annual earnings)/(annual weeks worked X average weekly hours), 11. The sample sizes reported in tables 2.3 and 2.4 are not fully comparable to table 2.1 above because of the exclusion of females, members of the armed forces, the self-employed, students, and observations without reported occupation or with reported wages greater than $100. Including members of the armed forces, the self-employed, students, and observations without reported occupation or with reported wages greater than $100, the subgroups in tables 2.3 and 2.4 translate to the following population estimates for males (20-64-year-olds): (i) Puerto Rico: return migrants (1975-80), 64,780; return migrants (1970-74), 50,000; never migrated, 573,760; born in the United States, 19,440; (ii) United States: recent migrants, 33,200; migrated before, 288,600; born in the United States, 129,400.

59

Out-Migration and Return Migration of Puerto Ricans

Table 2.3

Characteristicsof the Puerto Rican Population in Puerto Rico (males 20-64), 1980 Return Migrants

Education Hourly wage % Elementary (0-6) % Junior high (7-8) % Some high (9-1 I ) % High school (12) % Some COIL (13-15) % Coll. grad (16-22) % Married % Manager & profess. % Admin. & sales % Service % Other

1975-80

1970-74

1,209 34.2 (9.52) 9.8 (3.80) 3.59 (2.18) 22.2 11.5 23.2 27.8 8.0 7.4 73.7 10.3 12.5 19.4 57.8

1,135 36.8 (9.36) 10.5 (3.83) 3.97 (2.94) .I7 .I0 .20 .31 .II .I1 84.7 16.4 15.5 19.0 49. I

Never Migrated

Born in U.S.

12,193 38.1 (11.2) 10.8 (4.1 I ) 4.14 (2.97) .I8 .09 .I5 .32 .I2

38 1 28.3 (8.06) 12.8 (2.82) 4.56 (3.36) .03 .02 .I3 .37 .23 .22 68.8 28.3 23.5 15.5 32.7

.I5 81.4 19.7 17.1 15.2 48.0

Nore: Standard deviations are given in parentheses

Table 2.4

Characteristicsof the Puerto Rican Population in the United States (males 20-64), 1980

Education Hourly wage % Elementary (9-6) % Junior high (7-8) % Some high (9-1 I ) % High school (12) % Some coll. ( I 3-1 5 ) % COIL grad ( 16-22) % Married % Manager & profess. % Admin. & sales % Service % Other

Recent Migrants

Migrated Before

Born in U.S.

480 32.3 (10.2) 10.4 (3.92) 5.11 (2.95) 15.8 13.1 21.3 28.1 12.5 9.2 65.0 13.3 10.8 11.9 64.0

4,846 38.8 (10.5) 9.4 (3.54) 6.25 (3.09) .20 .I7 .26 .25 .08 .04 74.8 10.2 12.2 19.0 58.6

1,650 29.6 (8.95) 11.8 (2.75) 6.60 (4.20) 3.2 5.1 25.2 38.7 18.1 9.7 55.8 17.8 21 .o 15.5 45.7

Nore: Standard deviations are given in parentheses.

60

Fernando A. Ramos

grants still in the United States (who have 9.4 years). The same pattern is observed for U.S.-born Puerto Ricans. Migrants to Puerto Rico (with 12.8 years of schooling) are more educated than U.S.-born Puerto Ricans who choose to remain in the United States (1 1.8 years). I will discuss below how these conditional means are consistent with the different returns to education in Puerto Rico and the United States, as predicted by the selection model. There are significant differences in the hourly wages of recent migrants to Puerto Rico relative to other Puerto Rican residents and of recent migrants to the United States relative to Puerto Ricans residing on the mainland. The average wage of recent return migrants to Puerto Rico is 13.3 percent lower than the average wage of Puerto Ricans who never migrated, while the average wage of recent migrants to the United States is 18.0 percent lower than the average wage of older migrants. In addition, the wages of U.S.-born Puerto Ricans, both in Puerto Rico and in the United States, are higher than those of the Puerto Rico born. The most interesting pattern in tables 2.3 and 2.4 is the higher level of education for residents of Puerto Rico (both migrants and nonmigrants) relative to residents of the United States (again both migrants and nonmigrants). In table 2.5, I pool data from the U.S. and Puerto Rican Censuses to estimate regressions with years of education as the dependent variable. This analysis attempts to examine whether the observed differences in schooling among the different groups are still significant after controlling for age. In the first column, I only include dummy variables indicating birthplace and migration status as independent variables (the group of persons who were born in Puerto Rico and never migrated to the United States is the omitted dummy variable). Table 2.5

Levels of Schooling by Group: Ordinary Least Squares

Intercept Return migrant

Born in U . S . , live in P.R.

Born in U . S . , live in U . S . Born in P.R., live in U . S . Recent migrants to U.S.

R’ Note: Standard errors are given in parentheses.

10.79 (.035) - ,650 (.087) 2.057 (.2W 1.029 (.101) - 1.346 (.065) - ,376 (.179)

13.58 (.242) - ,828 (.086) 1.383 (.193) ,439 (.101) - 1.291 (.OM) - ,774 (.176) - ,069 (.002)

.04

.07

61

Out-Migration and Return Migration of Puerto Ricans

These coefficients, of course, reproduce the differences in the averages of tables 2.3 and 2.4. Return migrants to Puerto Rico, for example, have 0.65 fewer years of education than Puerto Ricans who never migrated. In column 2, I control for age differences across the various groups. The evidence indicates that the systematic patterns discussed above remain even after controlling for age. In particular, the most-educated individuals still choose to reside in Puerto Rico. Recent migrants to the United States have 0.774 fewer years of education than Puerto Ricans who never migrated; recent return migrants to Puerto Rico have 0.05 more years of education than nonreturning migrants; and U.S.-born Puerto Rican residents of Puerto Rico have 0.94 more years of education than U.S.-born Puerto Ricans who chose to remain in the United States. The selection model described in the previous section predicts that individuals will migrate to the location that best rewards their human capital. The evidence in table 2.5, therefore, indicates that we would expect the returns to education for Puerto Ricans to be higher in Puerto Rico than in the United States. In table 2.6 and 2.7, I present ordinary least squares earnings regressions for Puerto Ricans both in Puerto Rico and on the mainland. The dependent variable is the logarithm of hourly wages. As shown in column 3, the coefficient on years of education is 0.060 in Puerto Rico but only 0.040 in the United States. The higher return to education in Puerto Rico explains not only why the most-educated Puerto Ricans choose to remain in Puerto Rico but also why return migrants are more educated than migrants who choose to remain in the United States. The evidence presented in this section is consistent with the implication of the Borjas (1987) model. The Puerto Rican migration flow to the United States is relatively unskilled simply because skilled Puerto Ricans find better opportunities in the Puerto Rican economy. This same factor also explains why the return migration to Puerto Rico is composed of persons who are more skilled than those Puerto Ricans who choose to remain in the United States.

2.4 Pattern of Migration The selection model not only predicts the type of observable skill characteristics most likely to characterize the immigrant flow but also has equivalent predictions about unobserved skill characteristics. The type of selection on unobserved characteristics depends on the shape of the distribution of earnings in the country of origin and in the country of destination. The variance of these distributions proxies for the return to unobserved skills in the countries. I argued above that, in the case of migration between Puerto Rico and the United States, we would expect negative selection on unobserved characteristics because earnings are more unequally distributed in Puerto Rico. Table 2.6 presents ordinary least squares earnings regressions for Puerto Ricans residing in Puerto Rico in 1980. In these regressions, I compare the

62

Fernando A. Ramos Ordinary Least Squares Earnings Regressions by Location: Puerto Rico

Table 2.6

~~

Constant

(1)

(2)

1.319 (.028)

,144 (.030) ,024 (.001) - .OW3 (.00002) ,077 (.001) - .075 (.025) - ,079 (.027)

- ,0002 (.00002) .060 (.001) - ,061 (.024) - ,061 (.026)

no .25 14,918

.31 14,918

Experience Experience squared Education Never migrated Return migrant Controls R2 N

- ,071 (.029) - .I41 (.031) no .01 14,918

(3) ,373 (.036) ,020

(.ow

Yes

Note: Standard errors are given in parentheses.

earnings of return migrants, nonmigrants, and U.S.-born migrants (this latter group represents the omitted dummy variable). The first column reports the earnings regressions without any controls. Nonmigrant wages are 7.1 percent lower than the wages of U.S.-born migrants and 7.0 higher than the wages of return migrants. After controlling for experience and education, the nonmigrant/Lf.S.-born migrant wage differential increases to 7.5 percent, and the wage differential between nonmigrants and return migrants falls to 0.4 percent. After controlling (in col. 3) for marital status, industry of employment, and occupation, the differential between nonmigrants and U.S.-born migrants falls to 6.1 percent, and the nonmigrant/return migrant differential disappears. To the extent that the migrant dummy variables capture unobserved skill characteristics of each group, the migration of U.S.-born Puerto Ricans to Puerto Rico is characterized by positive selection, while the return migrants have similar unobserved characteristics relative to nonmigrants. Table 2.7 presents similar regressions for the United States with U.S.-born Puerto Ricans as the omitted group. In column 1, I show that the wages of Puerto Rico-born persons who migrated prior to 1975 are 1.7 percent lower than the wages of U.S.-born Puerto Ricans and 23.9 percent higher than those of recent migrants. After controlling for experience, education, marital status, industry, and occupation, the wage differentials change to 8.1 percent and 22.8 percent, respectively. These results imply that Puerto Rican migrants to the United States have less valuable unobserved skill characteristics than U.S.-born Puerto Ricans. The results in tables 2.6 and 2.7 are consistent with the selection model presented above. We observe that the migrant flow mov-

63

Out-Migration and Return Migration of Puerto Ricans Ordinary Least Squares Earnings Regressions by Location: United States

Table 2.7 ~~

~

(1)

Constant

1.743 (.012)

Experience squared Education

Recent migrant

- ,017 (.014) - ,239 (.025)

Controls R2

N

,882 ( ,029)

,033 (.002) - .0004 (.00003) ,049

Experience

Pre- 1975 migrant

(2)

no .01 6,976

- ,075 (.015) - .239 (.024) no .I4 6,976

(3) 1.127 (.045) ,024 (.ow - ,0003 (.00003) .040 - ,081

(.014) - ,228 (.023) Yes .21 6,976

Note: Standard errors are given in parentheses.

ing to the country with the most income inequality is positively selected while the migrant flow moving to the country with the least income inequality is negatively selected. Even if the dummy variables in tables 2.6 and 2.7 accurately measure unobserved skill characteristics of the different migrant groups, we cannot unambiguously conclude that the observed migration patterns reflect different economic rewards in the two economies. While it is instructive to know how Puerto Rican migrants in the United States fare relative to the U.S.-born nonmigrant population, the relevant comparison should be how migrants would have fared in Puerto Rico had they not migrated relative to the nonmigrant population. To make that comparison, we need to predict the wages that migrants would have earned in Puerto Rico. To calculate this prediction, I will follow a procedure described in detail in Lee (1978) and Robinson and Tomes (1984). I first estimate earnings equations for the United States and Puerto Rico controlling for the sampleselection bias introduced by the endogenous decision to migrate. The inverse Mills ratio for the earnings equations is calculated from equation (3). The coefficient estimates for the first-stage probit are reported in the first column of table 2.8. The coefficient estimates of the selectivity-corrected least squares earnings equations (reported in cols. 2 and 3 of table 2.8) are then used to calculate the wage that each individual would earn in each location given his or her observable characteristics. We can make two important observations from the estimates of the selectivity-corrected earnings equations in table 2.8. First, the estimated coefficient

64

Fernando A. Ramos Migration Regressions-Puerto

Table 2.8

Rico Born

OLS Wage Regressions Probit, U.S. = 1 Constant Experience

- ,496 (.054) ,028 (.003)

Experience squared Education Marriage Professional

- ,0005 (.00006) - ,024 (.003) - ,289 (.024) - .363 (.027)

Mills ratio Children

Puerto Rico - ,226

(.072) ,014 ( ,002) - .0001 (.00003) ,073 (.002) .I77

United States 1.101 (.140) .020 (.004) - ,0002 ( .00006) ,040

(.003) ,219

(.018)

(.031)

,251 ( ,020) ,571 (.120)

.I56 (.040) - ,220 (.128)

,153

(.023) -2

X

lOg(X)

761.2

R’ N

19,863

.27 14,537

.I3 5,326

Nore: Standard errors are given in parentheses.

of the inverse Mills ratio in the U.S. equation is negative, indicating negative selection for migrants to the United States, while it is positive in the Puerto Rican equation, indicating positive selection. I 2 This result is consistent with the migration model described above. The migrant flow to the economy with the more egalitarian income distribution (i.e., the United States) is negatively selected. Second, the returns to education in Puerto Rico remain higher than in the United States even after controlling for sample selection. Table 2.9 reports the predicted hourly wages for three Puerto Rico-born groups: return migrants to Puerto Rico; individuals who never migrated; and persons who migrated to the United States. The first and second columns show the predicted wage for each group if those individuals lived in the United States and Puerto Rico, respectively. For example, persons born in Puerto Rico but living in the United States have a predicted log hourly wage of 1.97. If they lived in Puerto Rico, they would earn a predicted log hourly wage of 0.876. The average migrant in the United States has relatively low predicted wages. In particular, the predicted wage for these migrants had they remained 12. The value of the inverse Mills ratio is positive in both regressions, so the sign of the coefficient also represents the sign of the selection.

65

Out-Migration and Return Migration of Puerto Ricans

Table 2.9

Average Predicted Log Hourly Wages Living in United States

Migrant to the U.S. Return migrant to Puerto Rico Live in h e r t o Rico, never migrated

1.967 (.@33) 1.984 (.@34) 2.039 (.002)

Living in Puerto Rico ,876 (.W) .914 1.002 (.003)

Note; Standard errors are given in parentheses.

in Puerto Rico is lower than the predicted wage of Puerto Ricans who chose not to migrate. The results also show the positive selection characterizing the return migrants to Puerto Rico. The predicted wages of return migrants are higher than the predicted wages of Puerto Ricans who chose to remain in the United States (but are still lower than the predicted wages of Puerto Ricans who never migrated to the mainland). 2.5

Conclusion

The special political relationship between Puerto Rico and the United States allows for continuous unrestricted movement across borders and permits a unique test of economic theories of migration, such as that given by the Roy model. The empirical evidence reported in this paper supports many of the predictions OF the model. In particular, the data reveal that relatively unskilled Puerto Ricans migrate to the United States; hence, the out-migration flow is negatively selected. At the same time, however, the return migrant pool tends to be composed of the most skilled of these (relatively unskilled) migrants. These empirical results are consistent with the hypothesis that workers choose to reside in those locations that offer the highest payoffs for their characteristics. The skill composition of Puerto Rican migration flows, therefore, can be understood in terms of the economic incentives created by differences in the rewards to skills between the sending and the destination regions.

References Abowd, John M . , and Richard B. Freeman. 1991. Immigration, Trade, and the Labor Market. Chicago: University of Chicago Press. Andic, Fuat M. 1964. Distribution of Family Incomes in Puerto Rico. Rio Piedras, P.R.: Institute of Caribbean Studies. Bohning, W. R. 1984. Studies in International Labor Migration. New York: St. Martin’s.

66

Fernando A. Ramos

Borjas, George J. 1985. Assimilation, Changes in Cohort Quality, and the Earnings of Immigrants. Journal of Labor Economics 3 (October): 463-89. . 1987. Self-Selection and the Earnings of Immigrants. American Economic Review 77 (September): 531-53. . 1989. Immigration and Self-Selection. In Abowd and Freeman (1991). Budd, Edward, ed. 1967. Inequality andpoverty. New York: Norton. Carliner, Geoffrey. 1980. Wages, Earnings, and Hours of Work of First, Second and Third Generation American Males. Economic Inquiry 18 (January): 87-102. Chiswick, Barry R. 1978. The Effect of Americanization on the Earnings of Foreignborn Men. Journal of Political Economy 86 (October): 897-921. King, Russell, ed. 1986. Return Migration and Regional Economic Problems. London: Croom Helm. Lee, L. F. 1978. Unionism and Wage Rate: A Simultaneous Equation Model with Qualitative and Limited Dependent Variables. International Economic Review 19:415-33. Mann, Arthur J. 1985. Economic Development, Income Distribution, and Real Income Levels: Puerto Rico, 1953-1977. Economic Development and Cultural Change 34:485-502. Moroney, John R. 1978. Income Inequality. Lexington, Mass.: Lexington. Puerto Rico Planning Board. 1983. Balance of Payments. San Juan, P.R. Robinson, C., and N. Tomes. 1984. Self-Selection and Interprovincial Migration in Canada. Canadian Journal of Economics 15:474-502.

3

The Assimilation of Immigrants in the U. S. Labor Market Robert J. LaLonde and Robert H. Topel

The popular image of new immigrants to the United States, impoverished but with great expectations of the future, is now part of our national culture. Since nearly all Americans are descended from immigrants, the “assimilation” of immigrant stock into the U.S. labor market is largely an accepted fact.’ As a generalization, the children of immigrants, and later generations, do The path to this prosperity is not well understood, however. One possibility, implied by the work of Chiswick (1978) and others, is that new immigrants rapidly accumulate skills-language, culture, and other dimensions of human capital-that are specific to the American labor market. Thus, the earnings of the typical immigrant rise quickly after arrival and eventually equal (or overtake) the earnings of similar nonimmigrants. Another possibility is that the

Robert J. LaLonde is associate professor of industrial relations at the University of Chicago Graduate School of Business and a fellow of the National Bureau of Economic Research. Robert H. Topel is professor of business economics and industrial relations at the University of Chicago and a research associate of the National Bureau of Economic Research. Reseach support from the National Science Foundation, the Sloan Foundation, and the William Ladany Research Fund at the University of Chicago Graduate School of Business is gratefully acknowledged. The authors thank William Anderson and Patrick Greenlee for their assistance with the calculations. 1. Sowell’s Ethnic America (1983) is an important narrative of the experiences and assimilation of immigrant groups in the United States. A theme of Sowell’s book is that the earnings of ethnic groups converge to the U.S. norm, at least across generations. Borjas (1990) argues, however, that differences in the earnings of U.S. ethnic groups reflect previous differences in the earnings of first-generation immigrants. 2. Japanese immigrants are a prime example of intergenerational mobility. Most Japanese immigrants had limited formal education and arrived as contract laborers in Hawaii. Many later migrated to the mainland. By 1940, the children of these immigrants (Nisei) had completed more years of schooling, on average, than white natives of the same age (U.S. Census of Population, 1940). Despite the dislocations of the 1940s, Japanese Americans are now among the most prosperous ethnic groups in the United States.

67

68

Robert J. LaLonde and Robert H. Topel

assimilation of immigrant families is mainly intergenerational. On this view, immigrants themselves realize only modest earnings growth after arrival in the United States, but their native offspring prosper. This paper studies the intragenerational assimilation of immigrants to the United States, relying on wage and earnings data from the 1970 and 1980 Censuses of Population. It is well known that in individual Censuses the average earnings of immigrants rise rapidly with time in the United States. New arrivals have substantially lower average earnings than observationally similar immigrants who arrived earlier. One interpretation of this finding is that the earnings of the typical immigrant rise with time in the United States, so that intragenerational assimilation is important. An alternative interpretation is that the average productivity (“quality”) of immigrant cohorts has declined over time. Earlier arrival cohorts earn more because of higher average skills, not because of assimilation. At least for recent data, this interpretation of the evidence is consistent with changes in immigration law such as the 1965 Amendments to the Immigration and Nationality Act, which shifted the emphasis from national origins quotas to family preferences in admission decisions. These alternative hypotheses about the assimilation process cannot be distinguished in a single cross section of earnings data. To break that deadlock, Borjas (1985) charted the earnings growth of immigrant arrival cohorts between 1969 and 1979. He concluded that assimilation is a much less important contributor to earnings growth than would be implied by cross-sectional earnings comparisons. He attributed the difference between the time-series and the cross-sectional estimates of assimilation to “a precipitous decline in the ‘quality’ of immigrants admitted to this country since 1950” (p. 463). The implication of his findings is that the assimilation of immigrant families to the American labor market is mainly due to intergenerational mobility; the assimilation of immigrants themselves is both slow and numerically small. This conclusion is important since it virtually reverses popular and strongly held conceptions about immigrants: they do not assimilate as much as we thought, and they have been getting worse over time. This paper reassesses the evidence on immigrant assimilation and changes in immigrant quality over time. Our estimates of assimilation are based on the relative earnings of different immigrant cohorts in the 1970 and 1980 U.S. Censuses as well as on changes in the average earnings of these cohorts during the 1970s. We have two main findings. First, for most ethnic groups we find very strong evidence of assimilation. The first ten years of experience in the U.S. labor market raise earning capacity of a typical new immigrant by over 20 percent, holding experience and education constant. This estimate is not much different than what cross-sectional earnings comparisons would predict, so that we find little evidence of declining immigrant quality within the ethnic groups that we study. In this sense our conclusions are substantially different than those of Borjas (1985). We also provide evidence in the conclusion that

69

Assimilation of Immigrants in the U.S. Labor Market

overall immigrant quality did decline, but largely as a result of changes in the ethnic composition of new immigrants to the United States. Recent immigrants are from source countries with lower average amounts of human capital, but immigrants from those countries do assimilate into the American labor market. Our second finding is that relative earnings of immigrants are sensitive to aggregate factors that have increased the inequality of wages in the United States. After peaking in the early 1970s, relative wages of less-skilled workers have steadily declined. Since immigrants are typically less skilled than the representative native, this change in relative wages had a disproportionate effect on immigrant earnings. We estimate that changes in the relative returns to skills during the 1970s reduced the relative wages of some less-skilled immigrant groups by between 5 and 10 percent. That decline in immigrant earning power partly offset the wage gains that immigrants received from assimilation. Thus, estimates of immigrant assimilation understate the true amount of human capital accumulation experienced by the typical immigrant. This evidence also reflects on the issue of declining “quality” of immigrants. Among less-skilled immigrant groups such as Mexicans, our evidence is that immigrant wages would have declined even if immigrant quality had remained unchanged. This implies that some of the concern about declining immigrant quality is unwarranted. The paper is organized as follows. The next section provides some empirical foundation for the problem we study, showing trends in immigration, the relative earnings and educational attainment of immigrants, and trends in wage inequality in the U.S. labor market. Section 3.2 describes our empirical methods for isolating the effect of assimilation on earning capacity. Section 3.3 provides initial estimates of assimilation based on both cross-sectional and synthetic panel estimates of immigrants’ earnings growth. Section 3.4 evaluates the effect of aggregate labor market conditions on immigrants’ wages, and section 3.5 concludes.

3.1 Background: Patterns of Immigration and Earnings One of the most striking features of immigration into the United States during the 1970s was the change in the countries from which immigrants migrated. As shown in the first row of table 3.1, in the 1970s, 18 percent of immigrants arrived from either Europe, Canada, or Australia, 23 percent from South and East Asia, 27 percent from Mexico, and 18 percent from Latin America or the Caribbean.3 Those percentages represent a significant departure from the corresponding percentages of immigrants arriving in the United 3. For the purposes of this paper, we consider immigrants from the Middle East as coming from Afghanistan, Pakistan, Iran, and North Africa as well as those countries normally considered the Middle East. Other immigrants come primarily from sub-Saharan Africa and the South Pacific.

70

Robert J. LaLonde and Robert H. Tope1

Table 3.1

Where Do Immigrants Come From? (percentage from region during decade) Place of Origin

Decade Arrived: Census File

1970s: 1980 1960s: 1970 1980 1950s: 1970 1980 Before 1950: I970 1980

Europe

Asia

Middle East

Mexico

Latin America

Other

18

23

6

27

18

7

40 34

13 12

4 4

12 17

27 26

4 6

69 63

6 6

2 3

11

14

9 8

3 6

79 68

6 6

1

7 10

4 6

3 9

I

Note: The place of origin categories are defined as follows: Europe encompasses all European countries and also includes the Soviet Union, Canada, Australia, and New Zealand; Asia encompasses South and East Asia; Middle East encompasses North Africa and Southwest Asia, including Pakistan (see no. 3); Latin America encompasses all of Central and South America (except Mexico) and the Caribbean; Orher encompasses sub-Saharan Africa and all other areas. Census File refers to Public Use Census File used to tabulate the percentages in the table.

States during the 1950s, when approximately two-thirds of all immigrants arrived from Europe, Canada, or Australia. By contrast, only 6 percent arrived from South or East Asia, only 14 percent from Mexico, and only 8 percent from Latin America and the Caribbean. Those changes in the source countries of immigrants also entailed changes in the skills that immigrants brought to the U.S. labor market. As shown in table 3.2, European immigrants typically have slightly less education than comparably aged natives; Asian immigrants typically have more education than natives; and Mexican immigrants typically have substantially less education than natives or even Hispanic natives. Such differences in observable skills suggest that the skill distribution of the immigrant work force has changed with the changing ethnic composition of immigrant flows. Thus, if the average education of new immigrant cohorts were fixed at 1980 levels, the change in relative immigrant shares from the 1950s to the 1970s, shown in table 3.1, would reduce average immigrant years of schooling by about two years, from 12.5 to 10.4. Changes in the immigrant skill distribution potentially confound efforts to estimate the rate of assimilation of immigrants into the U.S. labor market, as differences in the relative earnings of recent and earlier immigrants may reflect differences in skills and not time spent in the United States. That consideration would be particularly important if, for each ethnic group, the skills of successive immigrant cohorts had declined. However, as shown by table 3.2, statis-

71

Assimilation of Immigrants in the U.S. Labor Market

Table 3.2

Years of Completed Schooling (means for selected immigrant groups, 1970 and 1980 Censuses) Years in the United States

Place of Origin and Age Cohort European 1970: 25-34 3544 45-54 1980: 25-34 35-44 45-54 Asian 1970: 25-34 35-44 45-54 1980: 25-34 35-44 45-54 Mexican 1970: 25-34 35-44 45-54 1980: 25-34 3544 45-54

0-5

6-10

I 1-15

16-20

Natives

12.0 11.0 9.4

11.4 11.1 9.9

11.3 11.0 10.7

12.4 11.0 11.0

12.3 11.7 11.2

13.9 13.7 12.1

11.7 11.6 9.9

11.6 12.3 10.4

13.2 12.4 11.8

13.5 13.0 12.3

15.8 14.2 10.8

15.2 14.0 13.0

15.5 14.1 14.0

12.5 12.2 9.5

12.3 11.7 11.2

14.4 13.9 13.0

15.3 16.2 13.7

15.2 16.7 13.9

15.2 16.0 15.4

13.5 13.0 12.3

6.5 5.5 3.4

7.1 5.7 5.3

7.6 6.3 6.0

8.2 6.5 6.1

10.2' 9.0' 8.2*

7.0 6.1 5.5

7.2 6.2 5.3

7.6 6.5 5.9

10.2 7.4 5.7

11.9 10.9"

9.6'

Source: Public Use Files, 1970 and 1980 Census. For selection criteria, see the appendix Nore: For place of origin, see the note to table 3. I . 'The figure is the mean years of completed schooling for Hispanic natives.

tics on educational attainment suggest little change over time in the skills of different immigrant cohorts. In fact, recent European and Mexican immigrants in 1980 have completed more years of schooling than their counterparts in 1970. That finding suggests that changes in the skill distribution of immigrants largely reflect changes in the ethnic composition of immigrant flows and not changes in skills within each ethnic group. The earnings of different immigrant groups reflect the differences in their observed skills. As shown in table 3.3, relative earnings vary significantly with the source country of the immigrant. Among recent arrivals, immigrants of European ancestry have the highest earnings and Mexicans the lowest. That

Table 3.3

Relative Wages of Male Immigrants (differences in mean log weekly wages) Years in the United States

Place of Origin and Age Cohort

1-5

6-10

11-15

16-20

-.19 - .22

- .01 - .08

.01 .04

.08

- .33 - .28 - .37

- .20 - .21 - .40

- .09

.02 .01 - .05

- .05

.13 .09

.09 .12

15

- .04 .I0 - .05

.01

- .08 -.14

.08 .08

.05 .lI

- .07

.08

- .19 - .16

.I2 - .03

.14 .19

- .22 0

- .20 .31 - .37

.03

.06 - .22

.14 .19 - .06

.21 .20 .27

- .34 - .55

- .33 - .31

- .25

- .80 - .58 - .90 - .81

- .44 - .72 - .89

- .26 - .55 - .60

-.16 - .36 - .53

- .32

- .09

- .02

- .02

- .37

-.18

-.18

- .52 - .46 - .69

- .25

-.16

- .33 - .61

- .32 - .38

A 11 immigrants

1970: 25-34 3544 1980: 25-34 3544 45-54

- .08 - .21

.08

Europe

1970: 25-34 3544 1980: 25-34 35-44 45-54

0

16

Asia

1970: 25-34 3544 1980: 25-34 35-44 45-54 Mexico

1970: 25-34' 35-44 1980: 25-34 3544 45-54

- .63

- .33

Latin America

1970: 25-34 3544 1980: 25-34 3544 45-54

.10

.05 - .03 - .21

Source: U.S. Census 1970 and 1980 Public Use Files. Note: Estimates are differences between mean log weekly earnings of immigrants and natives in

the indicated age category. The mean log weekly earnings of natives are 5.02 for 25-34-yearolds in 1970; 5.16 for 35-44-year-olds in 1970; and 5.65 for 25-34-year-olds, and 5.88 for both 35-44-year-olds and 45-54-year-olds in 1980. The appendix discusses the sample.

73

Assimilation of Immigrants in the U.S. Labor Market

finding indicates that the increased shares of Mexican and other similarly skilled immigrants reduced the average earnings of recent immigrants. Because a large share of earlier immigrants came from high-wage groups, whereas recent immigrants have come from low-wage groups, it would uppear in cross-sectional data as though relative earnings of immigrants rose with time in the United States. Thus, among immigrants aged 35-44 in 1970, those who arrived after 1964 earn 22 percent less than similarly aged natives, while those who have been in the country for eleven to fifteen years have reached earnings parity with natives. But if the skills of the immigrant work force have also changed, evidence of assimilation should be less apparent and less systematic when we compare the relative earnings of the same cohort across Census years. Thus, by 1980, the same 1970 cohort of 35-44-yearolds is 45-54 years old and has been in the United States for eleven to fifteen years. That group still earns 21 percent less than natives, which is virtually the same as the 22 percent difference experienced in 1970. This supports the contention that the increase in earnings with time spent in the United States largely reflects changes in immigrant quality rather than assimilation. In addition to the decline in immigrant skills, changes in the U.S. labor market may have reduced the relative earnings of new immigrants. Beginning in the late 1960s, the U.S. labor market has shown a pronounced trend toward increased earnings inequality. As documented by Juhn, Murphy, and Pierce (1989), this trend has meant significantly lower relative earnings for lessskilled workers. The potential effect of increased inequality on the earnings of immigrants is illustrated in figure 3.1. The figure shows that, during the 1970s, the earnings of workers below the median grew more slowly than the earnings of workers at or above the median. The potential effect on certain immigrant groups is implied by their relative positions in the earnings distribution. For example, the median earnings of Mexican immigrants who arrived between 1965 and 1969 was at the eleventh percentile of the 1970 native earnings distribution. Over the decade, persons at the eleventh percentile experienced a 13 percent decline in their relative earnings, so we would predict a substantial decline in the relative earnings of Mexican immigrants between 1970 and 1980. By contrast, the 1970 median earnings of European immigrants who arrived between 1950 and 1959 was at the fifty-fourth percentile of native distribution. For Europeans, figure 3.1 implies only a negligible effect of increasing wage inequality on the relative earnings of a representative immigrant.

3.2 Methodology To estimate the rate of assimilation of new immigrants, we begin with a standard econometric model of wage determination based on cross-sectional data for each Census year, 1970 and 1980:

74

Robert J. LaLonde and Robert H. Topel 0.2

P

0.15

I

C

.-5

-

9

3J c

.-

i$ I

0.1

0.05 0 -0.05

* 0

-.-a, 5

-0.1

4-

-0.15

n -0.2 -0.25

I

1

1

1

1

1

1

1

1

1

1

1

1

1

1

3 1 9 1 1 5 1 2 1 1 2 7 1 3 3 1 3 9 1 4 5 1 5 1 1 5 7 1 6 3 1 8 9 75 81 87 Q36 6 12 18 24 30 36 42 48 54 60 66 d2 d8 $4 d0

Peramtile of the Earnings Distribution

Fig. 3.1 Growth in earnings, 1969-79 Source: U.S. Census Microdata Files for 1970 and 1980. Nore: The figure shows the growth in weekly earnings of each percentile of the native earnings distribution between 1969 and 1979. Earnings changes are for males 25-44 in 1970 and expressed as the difference in log earnings relative to the median.

In ( I ) , y , refers to the log weekly wage of an immigrant from arrival cohort i and Census year t . In the data, date of arrival in the United States is usually recorded in five-year intervals; for example, immigrants in the 1980 Census are recorded as having arrived in 1975-79 ( i = 7 9 , 1970-74 (i = 70), and so on. The vector X refers to a standard list of human capital controls. In writing (l), we have ignored differences among immigrants in place of origin. However, in the empirical work reported below, we allow the prices of these characteristics to vary by country of origin (ethnicity) and over time-but not across arrival cohorts of an ethnic group.4 Thus, p, may be different for Mexican immigrants than for Europeans but is restricted to have the same value for recent and earlier Mexican immigrants. 4. We impose this restriction as a matter of computational convenience. When p varies by both arrival cohort and ethnicity, sample sizes would be small.

75

Assimilation of Immigrants in the U.S. Labor Market

Unobservable factors that affect earnings are decomposed in ( 2 ) . The parameters a,, represent the average level of accumulated, U. S.-specific human capital embodied in members of arrival cohort i . We view these (unobserved) parameters as lying along a time-invariant assimilation profile. Assimilation occurs if the regression-adjusted earnings of a more recent immigrant cohort are smaller than the earnings of an earlier immigrant cohort, a,, < a,-,, ,, or if the regression-adjusted earnings of a cohort rise with time spent in the United States, a,, < a, ,+,,,.Thus, a,, represents the main parameter of interest in this paper. The b,, represent time effects, attributable to overall labor market conditions, that may have differential effects on particular arrival cohorts. One interpretation of the b,, is that they are transitory fluctuations in the value of human capital for various cohorts and so have zero expected value over time. Alternatively, if there are permanent changes over time in the price of skills, the b,, may affect the assimilation profile experienced by the typical immigrant. Finally, u, refers to the cohort-average value of other unobserved factors (talent or immigrant “quality”) that affect productivity but are fixed within an arrival cohort. It is important to highlight the meaning of assimilation implied by (1) and ( 2 ) . In this framework, assimilation occurs if, between two observationally equivalent persons, the one with greater time in the United States typically earns more. This is a different conceptual experiment than the one that was carried out in table 3.3 above, where we asked whether immigrant earnings converged over time to those of comparably aged natives. The age of immigrants and natives was not held fixed for that calculation. Below, we highlight the empirical differences between these alternative definitions of assimilation. It is obvious that, in a single cross section, say 1970, the parameters a,,, b,,, and u, are not separately identified. The problem is the familiar one of identifying time ( b J , vintage (a,,), and cohort (u,) effects from survey data (Griliches 1971). Thus, estimates of the degree of assimilation based on crosssectional data must impose identifying assumptions. For example, compare the estimates of E,, from equation (2) for immigrants who arrived in the United States between 1965 and 1969 (i = 65) to the corresponding estimate for those who arrived between 1955 and 1959 ( i = 55). The estimated effect on earnings of ten years’ residence in the United States is then

This is an unbiased estimate of assimilation so long as (i) there are no time effects on relative earnings for the two cohorts (E[b,,,,- b,,,] = 0 ) and (ii) there are no differences between the cohorts in average levels of “talent” (E[u,, - u6J = 0). Otherwise, estimates of (3) may either overstate or understate the amount of assimilation. For example, if the quality of new immigrants declined over the period 1955-69, then E(u,, - uS5)> 0, and (3) will overstate the rate of immigrant assimilation. This point is implicit in the ar-

Robert J. LaLonde and Robert H. Topel

76

guments of Borjas (1985). In contrast, if transitory changes in market conditions reduce the wages of less-skilled new immigrants proportionally more than their predecessors’ wages, (3) will understate the degree of assimilation. An alternative to the cross-sectional estimator (3) is to form a quasi panel by following the wage growth of an arrival cohort between the 1970 and the 1980 Censuses. In order to use this strategy, secular wage growth of the cohort must be indexed against that of some base group, n (natives, e.g.). Thus, assume that the base group earnings are determined by (4)

Y,,

=

x,,o, + bn, + U”,

where b,, and un are interpreted as above. A panel estimate of the magnitude of ten years’ assimilation on the earning capacity of cohort i is (5)

i‘

=

(‘i.80

-

‘1.70)

-

(&,.SO

-

-

‘n.70)

(bn,80

-

=

(‘t.80

-

‘i.70

+

b~,80

-

’8.70)

’n.70)’

Notice that cohort effects, u,, are eliminated from (5) owing to the differencing procedure. Thus, variation in immigrant quality over time will not affect the estimates. Yet assimilation in the sense of accumulating human capital is not identified without additional assumptions. The identifying assumption necessary to make (5) useful is that relative wage changes caused by changes in market conditions over the decade are factor neutral:

which is to say that there are no time effects on the relative wages of immigrants. Evidence against this assumption was provided in figure 3.1 above, which documented that relative wage changes during the 1970s favored more-skilled workers. Since new immigrants are typically less skilled, this trend toward increased inequality means that inferences drawn from (5) may be sensitive to the choice of a base group, n. For example, if the base group is prime-aged native men, and if the relative wages of new immigrants fall relative to the typical native, then (6) will not be satisfied. In this case, equation (5) will understate the true amount of immigrant assimilation. We adopt two methods of accounting for relative price changes in implementing (5) across Census years. First, we will present estimates of ( 5 ) for various immigrant groups, using different base groups, n, to normalize wage growth. An “optimal” base group is one that, on a priori grounds, would be similarly affected by changes in inequality or relative skill prices. Lacking strong theory or evidence on which group that would be, our strategy is to present alternatives. On the whole, our evidence is that inferences about assimilation are not highly sensitive to the choice of a base group. Our second method adopts a less parametric approach to isolating the effect of changing relative prices. To focus on the essential idea, assume that b,,,, - bn,70= 0, and rearrange (5):

77

Assimilation of Immigrants in the U.S. Labor Market

(7)

‘i

=

+ ’i,80)

(‘i.80

-

+ bi,70).

(‘i.70

Both terms in parentheses can be estimated, but their separate components are not identified without further assumptions. Thus, an estimate of (7) will understate the assimilation of cohort i if bi,80< bi,70.To estimate assimilation, we require an answer to the question, What would be the value of qE0 b,,80 if no assimilation occurred between 1970 and 1980? If we had an estimator of this value, say di,80= bi.80,then (7)could be decomposed as

+

+

(8)

=

‘i

-

-

(‘t,80

+

bi,80

(‘i.80

-

‘i.70)

-

‘ 30 years in 1980, > 20 years in 1970

1965-69 1960-64 1950-59 Average

,293 ,248 ,128 ,223

,158 ,060 .I24 ,114

Borjas sample, experience quadratic: Omitred group is > 30 years in 1980, > 30 years in 1970

1965-69 1960-64 1950-59 Average

.293 ,248 ,128 .223

,256 ,156 ,217 .209

Borjas sample, experience quarric: Omitted group is > 30 years in 1980, > 30 years in 1970

1965-69 196C-64 1950-59 Average

,296 ,250 ,125 .223

,319 ,224 ,265 ,269

Full sample, experience quartic: Omitred group is > 30 years in 1980, > 30 years in 1970

1965-69 1960-64 1950-59 Average

,217 .215 ,146 ,193

,206 ,168 ,156 ,177

Note: Borjas sample refers to individuals between the ages of 18 and 54 in 1970 and 28 and 64 in 1980. Omitted group refers to the immigrant cohort against which the other immigrants’ wages are gauged in estimating the 1970 and 1980 cross-sectional regressions. Other selection criteria are the same as in our earlier analysis.

panel cdrresponds to our unrestricted sample and specification, and it shows that the cross-sectional and panel estimates of assimilation are very similar. The results in table 3.5 above also stand in contrast to the erratic patterns of within-cohort wage growth documented in table 3.3 above. Those calculations suggested that cross-sectional estimates of assimilation are partly an illusion, perhaps accounted for by changing characteristics of immigrants over time. The difference in interpretation can be reconciled, in part, by taking note of two facts. First, immigrants are less skilled than natives. They enter the U.S. labor market with fewer years of schooling and’thus have, for a given age, more years of experience than the typical native. Given the concavity of earnings profiles, that fact implies slower wage growth for immigrants. Second, life-cycle earnings profiles are also flatter for less-skilled workers. Thus, even for immigrants and natives with the same number of years of experience, the typical immigrant will have slower earnings growth. Since table 3.3 allows both immigrants and natives to “age” from 1970 to 1980, both these effects imply smaller relative wage growth for immigrants than for the typical native. Thus, calculations like those in table 3.3 will understate the actual rate of immigrant assimilation. Those points are illustrated by figure 3.3, which depicts the experience-log

85

Assimilation of Immigrants in the U.S. Labor Market

5.8 5.7 5.6 5.5 CD u)

-

.-c

5.4 -

w

5.2 -

E

> Y

r”

CD

j

5.3 5.1

5 4.9 -

4.0 a, r

-

-

4.7 4.6 -

4.5 4.4 4.3

1

1

1

1

1

1

1

1

1

2

3

4

5

6

7

0 9 10 11 12 13 14 15 16 17 10 19 20

1

1

1

1

1

1

1

1

1

1

1

1

Fig. 3.3 Natives’ and Mexican immigrants’ earnings

earnings profiles for three groups: natives with 12.3 years of schooling, Mexicans with 6.5 years of schooling and zero to five years in the United States, and Mexicans with 6.5 years of schooling and eleven to fifteen years in the United States.8 As depicted in the figure, a recent Mexican immigrant just entering the labor market earns 50 percent less (subject to the log approximation) than a typical native worker. As that immigrant ages, he moves up the experience-earnings profile because of human capital gains associated with labor market experience, and he also jumps up to a 20 percent higher profile because of the gains associated with time spent in the United States. Despite that jump, the Mexican immigrant’s earnings remain approximately 50 percent behind the same group of natives because of the steepness of the native profile. From this evidence, we conclude that immigrant and native wages do not necessarily converge over time. Lack of convergence is partly caused by differences in shapes of earnings profiles-immigrant profiles are flatter because immigrants are less skilled to start with. But this finding does not imply lack of assimilation. As we documented above, time in the United States has a 8. The years of schooling chosen for the natives and the Mexicans correspond to the mean years of schooling for 1970 25-34-year-olds in table 3.2 above.

86

Robert J. LaLonde and Robert H. Topel

strong positive effect on earning capacity, holding constant experience and education. The finding does imply that immigrants do not catch up with white natives, so the U.S. labor market is not a “melting pot” in which there are no ethnic wage differences in the long run. But that was known; for example, native Hispanics typically earn less than native whites for reasons unrelated to assimilation.

3.4 Rising Inequality and Changes in Immigrant Wages All the preceding results are based on the assumption that changes in the price of immigrants’ unobservable skills, relative to a normalizing population, are negligible. In this case, within-cohort growth in relative wages identifies the accumulation of unobserved human capital. This assumption is open to question in light of the trend toward greater wage inequality in the United States, which has reduced the relative earning capacity of less-skilled groups. If market conditions caused the relative value of immigrants’ skills to decline between 1970 and 1980, then panel estimates of wage growth will understate the true amount of immigrant assimilation. Our purpose in this section is to assess the importance of this effect. Our main finding is that, although changes in inequality during the 1970s are unimportant for most immigrant groups, they did affect the relative wages of low-skilled immigrants, in some cases by a substantial amount. Table 3.7 illustrates this point. In the table, we apply the methods described in equations (8) and (9) and report adjusted estimates of relative wage growth for six immigrant cohorts that entered the United States between 1950 and 1969. Those estimates measure the change in relative earnings of immigrants that would have occurred in the 1970s in the absence of assimilation, based on the position of immigrants in the 1970 wage distribution. For purposes of these calculations, we applied (8) and (9) to weekly wages; we did not remove the effects of the observables, X . Also, to enhance the sample size for these calculations, we focused on only two immigrant aggregates: (i) the immigrant population with less than ten years of schooling and (ii) Mexican immigrants. The base group (n)for these comparisons is natives of the same age. To illustrate the calculations, consider the Mexican cohort that arrived in 1965-69. In 1970, these individuals earned 71 percent (using the log approximation) less than a representative native of the same age (col. 1). If no assimilation had occurred, we estimate that persons in this cohort would have earned 79 percent less than a representative native in 1980 (col. 2). They actually earned 57 percent less, so our corrected estimate of growth in earning capacity is 23 percent (col. 5). Therefore, panel estimates of relative wage growth understate assimilation of this cohort by about 8 percent, owing to aggregate changes in relative wages that occurred over the decade. 9. Butcher (1990, tables 11, IV, VII) reports similar results for black immigrants based on the 1980Census.

87

Assimilation of Immigrants in the U.S. Labor Market Immigrant Wage Growth Relative to Natives with Adjustment for Changing Inequality, 1970-80

Table 3.7

Immigrant Group Year of Arrival

(1)

(2)

(3)

(4)

(5)

Relative Wage in 1970

Predicted Relative Wage in 1980

Relative Wage in 1980

Relative Wage Growth (3) - (1)

Corrected Growth (3) - (2)

- .47 - .34 - .23

- .02 - .06 -.I3

.03 - .03 -.I3

- .05 - .02

- .57

.14 .03 .01

.23

- .08 - .05 - .03

Immigrants with < 10 years of schooling: -.so 1965-69 - .45 - .31 1960-64 - .28 -.I0 -.I0 1950-59 Mexican immigrants: - .79 1965-69 - .71 - .49 - .44 1 9 M - .33 - .30 195G59

- .41

- .29

.08

.04

(6) Effect of Changing Inequality (2) - ( I )

.o

Note: Relative wage measures the difference between the log weekly earnings of immigrants aged 2544 in 1970 and comparably aged natives. The predicted relative wages are computed as a weighted average of 1980 native wages, where the weights represent the immigrant cohort’s density at each kth percentile of the native 1970 wage distribution.

Note the obvious point that the size of the inequality effect, shown in column 6 of table 3.7, depends on the size of the original wage differential in 1970. In fact, for the sample of immigrants with less than ten years of schooling, there are no adjustments to wage growth for arrivals between 1950 and 1959 (1965-69). Given the magnitudes of the adjustments for the other cohorts, our findings suggest that biases in assessing the role of assimilation that result from increasing wage inequality apply mainly to recent arrivals and others who earn substantially less than the typical native. The upshot is that inferences about assimilation from within-cohort wage growth may be sensitive to changes in relative wages caused by aggregate labor market conditions, especially among unskilled recent arrivals for whom assimilation is likely to be most rapid. 3.5

Conclusion

In this paper, we reexamined the evidence on immigrant assimilation to the U.S. labor market. For the immigrant groups that we studied, our evidence suggests substantia! assimilation in the sense of sharply rising earning capacity after entering the United States, holding constant other observable factors that affect wages. Following fixed cohorts over time, our estimates of assimilation profiles roughly conform to estimates that can be derived from individual cross sections of Census data. In fact, the growth rates that we derive from synthetic panels across Census years sometimes exceed the rates implied by simple wage comparisons in a single cross section. Because of this, we conclude that there is no important evidence of declining immigrant “quality” within the groups that we have studied.

88

Robert J. LaLonde and Robert H. Topel

This is not to say that the overall quality of immigrants has not declined. As we showed in table 3.1 above, the distribution of immigrants by source countries has shifted over time, so the human capital of the average immigrant may have fallen because, say, Mexican immigrants bring a smaller stock of human capital than their European counterparts. In fact, the estimates of wage differentials between immigrants and natives in table 3.3 above strongly suggest this. To address this issue more directly, table 3.8 reproduces our calculations of between- and within-cohort wage growth on the sample of all immigrants, regardless of ethnic background. We perform the calculations both with and without experience and education controls, which turns out to make a difference. Several points about these estimates are noteworthy. First, cross-sectional estimates of assimilation are relatively large when observable characteristics are excluded from the analysis. In the 1980 data, we estimate that ten years of U.S. experience for a new arrival would raise earnings by 3 1 percent. Because that value is substantially larger than the corresponding estimate of withincohort wage growth (9 percent), cohort quality declined over time. Second, two-thirds of the difference between cross-sectional and within-cohort estimates of assimilation is accounted for by observables. After controlling for experience and education, estimates of within-cohort growth are only moderately smaller than the corresponding cross-sectional estimates. Thus, the unobservable skills of immigrants declined only modestly over time. Third, our findings on assimilation rates for each ethnic group indicate that changes in unobservables are accounted for by immigrants’ ethnicity. Thus, we find no evidence that immigrants’ unobserved skills have declined within ethnic groups. Immigrant skills declined because new immigrants are more likely to arrive from countries whose immigrants have always been relatively unskilled. Finally, given important changes in relative wages of skilled and unskilled workers that occurred in the 1970s, panel estimates of assimilation will understate immigrant assimilation among less-skilled groups such as Mexicans. For relatively unskilled new arrivals to the United States, we estimate that these changes in skill prices may have reduced the wages of new immigrants relative to natives by as much as 8 percent. Thus, panel estimates of assimilation may be sensitive to “time effects” caused by economy-wide conditions.

Appendix This study used the 1970 and 1980 Public Use Microdata Samples from the Censuses of Population and Housing (see U.S. Bureau of the Census 1970, 1980; for the technical documentation, see U.S. Bureau of the Census 1973, 1983). The estimates reported in the paper were derived from samples of 16-

89

Assimilation of Immigrants in the U.S. Labor Market Estimates of Immigrant Assimilation: Cross-sectionaland Synthetic Panel Estimates from Pooled Sample of 1970 and 1980 Immigrants

Table 3.8

A. Effects of Years in the United States on Relative Wages Years in the United States Census Year

0-5

Without controls: 1970

1980

- .36 - .58

6-10

11-15

16-20

21-30

-.I8 (.02) - .42

- .09 (.02) - .27

- .06 ( .02) -.16 (.02)

.07 (.02) - .06 (.02)

(.02) With controls for schooling and experience: 1970 - .33 -.I9 -.I1 (.01) (.02) - .39 - .29 -.I9 1980 (.02) (.02) (.02) With controls for schooling, experience, and place of origin: 1970 - .27 -.I2 -.lo (.02) (.02) - .32 - .21 -.I2 1980 (.02) (.02)

B. Estimated Effects of Ten Years’ Residence in the United States from Cross-sectional and Within-Cohort Growth Between-Cohort Growth year of Arrival

1970

I980

Within-Cohort Growth, 1970-80.

Without controls: 1965-69 .24 .31 1960-64 .13 .26 1950-59 .I5 .16 With controls for schooling and experience: 1965-69 .22 .21 1960-64 .I3 .I9 1950-59 .08 .I2 With controls for schooling, experience, and place of 1965-69 .17 .20 1960-64 .06 .I6 1950-59 .08 .06

.09 .02 .01 .14 .09 .05 origin: .I4 .07 .05

Note; The figures in panel A are the estimated coefficients from a regression of log weekly earnings of immigrants on dummy variables for the time in the United States. The left out group is immigrants who have been in the United States for more than thirty years (thirty-five years in 1970). The controls in the second model are for years of schooling, separate quartics in experience for those with less than twelve and twelve or more years of schooling, and schooling and experience interacted. The figures in panel B are derived from those in panel A. Numbers in parentheses are standard errors.

90

Robert J. LaLonde and Robert H. Tope1

64-year-old males who had worked forty or more weeks in 1979 (or 1969) as wage or salary employees or self-employed workers. Unpaid family members, persons with negative self-employment income, persons living in institutional or military quarters, and persons not in the 1980 (or 1970) civilian labor force were excluded from the sample. Table 3.4 in the text presented the estimated coefficients for time in the United States corresponding to equations (1) and (2). The complete set of estimates corresponding to (1) and (2) is presented in table 3A. 1. Besides controls for time in the United States, weekly earnings (annual earnings divided by weeks worked) for immigrants from a given source country were a function of years of completed schooling, a dummy variable indicating whether the workers had less than twelve or twelve or more years of schooling, and separate quartics in experience for each of those two educational groups. In those regressions, experience is measured as age minus schooling minus six. We chose the quartic specification for two reasons. First, the literature indicates that a standard quadratic earnings equation tends to overstate earnings of less-experienced workers (see Murphy and Welch 1990). Second, our data rejected the quadratic specification in favor of the quartic specification. Table 3A.l

Estimate of Earnings Equation (for table 3.4)

Variable

Europeans

Asians

Mideasterners 1970

65-69 60-64 55-59 50-54 3549

grade

HS x grade exP HS

X

exp

exp* HS x exp2

Mexicans

Other Hispanics

Table 3A.1

(continued)

Variable

Europeans

Asians

Mideastemers

.oO018 .00010 ,00019 (.00007) (.0002) (.00002) .00003 HS x exp3 - ,00002 ,00004 (.0002) (.owl) ( ,00004) exp‘ - .OooOo13 - .OOOOO13 -5.6 x 10-7 (.000001) (2.0x 10-7) (6.0x 10-7) HS x exp‘ -7.4x 10-8 - 5 . 8 X lo-’ -4.9 x 10-7 (4.0x lo-’) (.000002) (.00@)01) - ,0011 - ,0005 - ,0010 grade X exp (.002) (.0003) (.OOOS) ,0013 - .0024 HS x grade x exp ,0001 (.002) (.001) (.0004) - 1.45 - .29 HS graduate .01 (.92) ~17) (.46) 3.82 2.64 Intercept 2.95 (.79) (.39) (. 14) .45 .36 Mean standard .34

exp3

error Adjusted R2 N

.19 13,923

.32 1,752

.26 540

Other Hispanics

Mexicans

,00013

.oO033 (.OOOO6) ,00044 - ,00015 (.00014) (.oooo9) - 8 . 7 x 10-7 -2.6 X (4.0x lo-’) (5.0 x 10-7) -4.6 X 1.2x 10-6 (2.0x 10-6) (9.9x 10-7) - ,0014 - ,0028 (.0007) (.0005) ,0017 ,0013 (.002) (.0009) .01 .58 (.49) (.34) 2.85 2.49 (.22) (.28) .32 .37 (.OOOO4)

.18 2,060

.19 2,800

1980 - .20 ~03) -.18 ~03)

65-69 60-64

-.12 J.02) - .07 (.02)

55-59 5c-54 3549 grade HS x grade exP HS

X

exp

exp’

- .05 ( .02) .079 (.01) ,009 (.02) .22 (.02) - ,068 (.02) - ,010 (.001)

HS x exp’

,0024 (.002)

exp3

.00020

HS x exp3

(.oooo3) - ,000026

exp4

- 1.6X

(.00005)

(3.0x (continued)

lo-’)

- .40 (.06) - .25 ~05) - .14

- .53

- ,039 (.06) 0 (.06) .034 03) ,082 (.03) .15 ~04) ,039 (.04) - .0062 (.002) - .0051 (.003) .00010 (.00006) .00021

- .15

(W

(.00008)

- .46

(.06) - .30 (.06) - .22 (.06) -.11 (.06) -.12

(. 13)

- .27

(.I)

- .24

(.I)

(.I) -.15 (.I) ,276 (.06) - .182 ~07) .38

,064 (.02) .04 (.02) .19 ~03) - ,072

(.OW

- ,332

(.W

- ,018 ( ,006) - ,021 (.007) ,00042 (.0002) - ,00064

- ,009 (.002) .005 (.002) .0w20

(.oo@w

- ,00013 (.oooo7)

(.OoW

-6.0 x 10-7 -3.6 X lo-“ (6.0x lo-’) (2.0x 10-6)

1.5 X (4.0 x lo-’) -

92

Robert J. LaLonde and Robert H. Topel

Table 3A.1

(continued)

Variable

Europeans

HS x exp4

grade x exp HS

X

grade

X

HS graduate Intercept Mean standard error Adjusted R2 N

-7.9 x 10-8 ( 5 . 0 x 10-7) - ,0012 (.0004) exp ,0007 (.0005) .27 (.21) 3.49 .41 .21 11,102

Asians

-2.5 X (9.0 x lo-’) - .0007

(.0008) - ,0005 (.0009) - .84 (.35) 4.16 (.33) .41 .29 4,342

Mideastemers

6.8 x (2.0 x 10-6) - ,0065 (.0021) ,0061 (.0023) 2.7 ( ,841 1.31 (. 80)

.60 .25 1,145

Mexicans

-2.0 x 10-6 (1.0 x 10-6) 0

(.oO04) - .0015 (.0011) - .37 (.29) 4.36 (. 16) .53 .I2 5,404

Other Hispanics

1.1 x 10-6 (8.0 x 10-7) - ,0011

(.0006) 0

(.0007) - .16

~27) 3.84 (.23) .46 .21 5,069

Note: Standard errors are given in parentheses

References Borjas, George. 1985. Assimilation, Changes in Cohort Quality, and the Earnings of Immigrants. Journal of Labor Economics 4(0ctober):463-89. . 1990. The Intergenerational Mobility of Immigrants. Working paper. University of California, San Diego, February. Butcher, Kristin. 1990. Black Immigrants to the United States: A Comparison with Native Blacks and Other Immigrants. Industrial Relations Section Working Paper no. 268. Princeton University, August. Chiswick, Barry. 1978. The Effect of Americanization on the Earnings of Foreignborn Men. Journal ofPolirica1 Economy 86(0ctober):897-921. Griliches, Zvi. 197 1. Introduction: Hedonic Price Indexes Revisited. In Price Indexes and Quality Change, ed. Zvi Griliches. Cambridge, Mass.: Harvard University Press. Juhn, Chinhui, Kevin M. Murphy, and Brooks Pierce. 1989. Wage Inequality and the Rise in Returns to Skill. University of Chicago, December. Typescript. Murphy, Kevin M., and Finis Welch. 1990. Empirical Age-Earnings Profiles. Journal of Labor Economics 8(April):202-29. Sowell, Thomas. 1983. Ethnic America. New York: Basic. U.S. Bureau of the Census. 1970. Census of Population and Housing: United States Public Use Microdata Sample. County Group Sample, 1% Sample. Washington, D.C. . 1973. Technical Documentation for the 1970 Census of Population and Housing. Public Use Samples. Washington, D.C.: U.S. Government Printing Office. . 1980. Census of Population and Housing: United States Public Use Microdata Sample (B Sample), 1% Sample (1 CPSR 8170). Washington, D.C. . 1983. Technical Documentation for the 1980 Census of Population and Housing. Public Use Samples. Washington, D.C.: U.S. Government Printing Office.

4

The Fertility of Immigrant Women: Evidence from HighFertility Source Countries Francine D. Blau

Although women have constituted the majority of immigrant flows during most of the post-World War I1 period,’ surprisingly little attention has been devoted to them in the research on immigrants by economists. This study addresses this research need by examining immigrant women’s fertility behavior. I focus on immigrants from the Middle East, Asia, Latin America, and the Caribbean. This is a particularly interesting group to study for two reasons. First, immigration from these areas has increased considerably in recent years, from 29 percent of immigrants in the 1950s to 77 percent in the 1970s (Blau 1986).*Second, fertility rates in many, although not all, of these source countries areconsiderably higher than those in the United States, averaging in excess of 5.5 children per woman during the early 1960s and 1970s, in comparison to 3.3 and 2.0 children per woman, respectively, for these periods in Francine D. Blau is professor of economics and labor and industrial relations at the University of Illinois at Urbana-Champaign and a research associate of the National Bureau of Economic Research . An earlier version of this paper was presented at the Ford/NBER conference “The Determinants and Effects of Immigration on the U.S. and Source Economies,” Cancun, January 1990. The author would like to thank John Abowd, George Borjas, Dagobert Brito, Richard Freeman, Joan Kahn, Lawrence Kahn, T. Paul Schultz, James Smith, and the participants in the Ford/NBER immigration preconference and conference for helpful comments and suggestions. She is indebted to Nichola Dyer, Lisa Reavlin, Ho Hwan Park, and Mann Yoon for excellent research assistance. This research was supported by a grant from the Ford Foundation to the NBER and by the University of Illinois Research Board. 1 . This situation appears to have changed in the 1980s, when women made up slightly less than half of immigrants (U.S. Immigration and Naturalization Service 1986; U.S. Bureau of the Census 1980). 2. Immigrants from Africa were not included in this study both because they constitute a very small proportion of the total and because South Africa and Egypt are the only African countries separately identified in the 1970 Census (and Egypt has been included here as part of the Middle East).

93

94

Francine D. Blau

the United state^.^ If these extremely high fertility rates were maintained by immigrant women in the United States, the implications for the average rate of natural increase of the U.S. population would be substantial. The fertility decisions of immigrant women and their families are also of interest owing to their effect on the economic status of the family. Since fertility tends to be inversely related to female labor force participation (see, e.g., Smith 1980), large family size may be expected to have an adverse effect on family income. Further, at any given level of family resources, more children imply smaller levels of investment per child and thus lower child quality and reduced earnings for subsequent generations (Becker 1981; Chiswick 1988). But do these immigrant women differ significantly from the native born in their fertility behavior? The answer to this question is far from obvious but will depend on the selectivity of immigrants relative to the source country population, the effect of source country characteristics on immigrants’ behavior in the United States, and the extent and speed of immigrants’ adaptation to conditions in the United States. In addressing these issues, previous research on immigrant women’s fertility (e.g., Bloom and Killingsworth 1985; Kahn 1988; and Ford 1990) has relied on cross-sectional analyses of the data, measuring the effect of length of time in the United States by differences in fertility among immigrants who have resided in the United States for varying lengths of time.4 However, using this approach, it is not possible to distinguish between the true effect of U.S. residence on fertility (i.e., changes over time in the fertility behavior of a given cohort of immigrants as they reside in the United States for additional years) and cohort effects (i.e., cross-sectional differences in the fertility behavior of immigrants who arrived in the United States at different points in time). Given shifts over time in economic and political conditions in source countries and changes in U.S. immigration policies affecting who is admitted into this country, the possibility of cohort effects is a very real one. A particularly interesting feature of this study is that I analyze immigrantnative differences in fertility within a framework initially developed by Borjas (1987) for analyzing male earnings. Through the use of two cross sections of data, the effect of length of residence can be distinguished from cohort effects. In addition, I examine the influence of a wide range of source country char3. Fertility is measured by the total fertility rate. The immigrant average is weighted by the representation of women from that country in the 1970 and 1980 Censuses. 4. An exception is a study of Gorwaney et al. (1989) that calculates incremental fertility between the 1970 and the 1980 Censuses for a number of immigrant groups. The information provided by this study is limited, however, because, apart from age, no factors are controlled for and because no baseline measure of immigrant fertility relative to native fertility is obtained (i.e., natives are not included in the study). So, e.g., the authors claim that larger incremental fertility over the period for immigrants who were recent arrivals in 1970 than for longer-term residents is indicative of assimilation. However, without an initial comparison of fertility to otherwise similar natives or a knowledge of incremental fertility of natives over the period, it is not possible to ascertain which of a number of alternative views of immigration (discussed below) this finding supports.

95

The Fertility of Immigrant Women

acteristics on immigrant women’s fertility. I also compare the fertility of immigrants to that of otherwise similar natives. Studies by demographers have tended to focus solely on immigrants (e.g., Kahn 1988; Gorwaney et al. 1989; Ford 1990). The comparison to natives is not only of interest from a policy perspective but, as we shall see, is essential to distinguishing among alternative views of immigrant women’s fertility behavior. The plan of the paper is as follows. I begin with an overview of the data and of the immigrant-native fertility comparisons. In the next section, I consider alternative views of immigrant fertility behavior. I then present the framework for the empirical analysis followed by my findings regarding immigrant-native fertility differences, the effect of years of residence on immigrant fertility, and the effect of source country variables on immigrant fertility. I conclude with a summary of findings and a discussion of their implications.

4.1 Overview of Data and Immigrant-Native Comparisons Fertility is analyzed using data from the 1 Percent County Group Public Use Sample of the 1980 Census and the 1 Percent State Public Use Sample of the 1970 Census. The sample is restricted to women aged 18-54, and inmates in group quarters are excluded. Native-born women are also excluded if they were born abroad, at sea, or in outlying areas of the United States. Immigrant women are excluded if information on the period of their immigration is missing.5 The complete sample was used in creating the immigrant extracts; however, random samples were employed for some native race/ethnic groups.6 An overview of immigrant-native fertility differences by length of residence is presented separately for all women and for married spouse-present women in table 4.1. Fertility is measured by number of children ever born. The unadjusted figures are simply the observed means. We also show results adjusted for age on the basis of regression equations controlling only for age and (for immigrants) dummy variables for years since migration (YSM). The regressions are evaluated at the immigrant mean age in each year. While overall immigrant-native differences in age are small, YSM is highly positively correlated with age. Thus, in order to identify fertility patterns by YSM group, it is important to net out age effects. The most striking result of the table is the relatively small difference in fertility between immigrants and natives in each year. In 1970, the fertility of 5 . This exclusion is required to ensure consistency between the two Censuses. In the 1970 Census, “not reported” is explicitly listed as a code for the period of immigration variable, whereas, in the 1980 Census, period of immigration was imputed in such cases. 6. The following sampling percentages were employed: 0.1 percent of Hispanics and nonHispanic Indians and 0.05 percent of non-Hispanicblacks and whites. The full (1 Percent) sample of non-Hispanic other nonwhites was included. In the results presented below, native means are weighted by the inverse of the sample probabilities.

96

Francine D. BIau

Table 4.1

Overview of Immigrant-Native Differences in Fertility by Length of Residence 1970 All

1980 Married

All

Married

Means Unadjusted: Natives lmmigrants Adjusted for age? Natives Immigrants YSM

0-5

6-10 11-15 YSM 16-20 YSM 20 -k YSM

YSM

2.098 2.028

2.508 2.454

1.764 1.943

2.225 2.357

2.196 2.029 1.776 1.939 2.208 2.313 2.636

2.571 2.454 2.172 2.322 2.647 2.759 3.049

1.812 1.943 1.983 2.023 1.771 1.999

2.168 2.351 2.401 2.440 2.207 2.188 2.426

1.808

Differentials Unadjusted Adjusted for age:' All cohorts YSM 0-5 YSM 6-10 YSM 11-15 YSM 16-20 YSM 20 -k

- ,070

- ,054

,179

,132

- .167 - ,419 - ,257 ,012 . I 17 ,440

-.117 - ,399 - .249 ,076 .188 ,478

.I24 .171 .211

,183 ,233 ,272 ,039 ,020 .258

- ,004 - ,041 ,186

'Based on fertility regressions, including controls for age, age squared, and (for immigrants) dummy variables for length of residence, evaluated at the immigrant means in each year.

immigrants was about the same as natives; in 1980, it was only .18 higher. These small unadjusted differentials are surprising in light of the high average fertility rates in the countries of origin and the substantial differences in the individual characteristics of immigrants and natives detailed below. One important focus of my empirical work will be to shed light on the reasons for the relatively small immigrant-native differences in fertility. This finding of small unadjusted differentials is important in and of itself in that, from a policy perspective, it is to some extent the unadjusted differentials that are of particular interest.' That is, fertility differences between immigrants and natives have potential effects on the domestic economy regardless of whether they are 7. This finding must be regarded with some caution, however, given the censored nature of the fertility variable. Many of the women in the sample have not yet completed their fertility, and, as we shall see, the fertility of immigrant women may follow a very different time path than that of native women. Thus, even after adjusting for age, cross-sectional comparisons may give a misleading picture of eventual fertility outcomes. This issue is considered in greater detail below.

97

The Fertility of Immigrant Women

due to differences in the characteristics (means) of the two groups or to differences in the immigrant-native response to those characteristics(coefficients). The immigrant-nativedifferentials for 1970 and 1980 also indicate that relative immigrant fertility trended upward over the 1970s. Table 4.1 indicates that this was the result of larger declines in fertility for natives than for immigrants over the decade. The age-adjusted fertility of natives declined by .384 among all women and by .403 among married spouse-present (MSP)women compared to declines of .086 and .103, respectively, for immigrants. The trends among immigrants were in turn tied to an increase in the relative fertility of new arrivals (YSM 0-lo), who had higher fertility compared to natives in 1980 than in 1970 and whose fertility was absolutely greater in 1980 than that of new arrivals in 1970. The reason for these trends and their implications for future immigrant-nativefertility differences will be considered below. Finally, the data in table 4.1 suggest that cross-sectional differences in ageadjusted fertility by YSM group may not be a very good indication of the actual effect of years of residence on fertility. Cross-sectional comparisons in 1970 suggest that fertility increased with years of residence, while the 1980 figures suggest roughly declining fertility with longer residence. However, when we compare the relative fertility of recent arrivals in 1970 (YSM 0-5 and YSM 610) to the relative fertility of the same group in 1980 (YSM 11-15 and YSM 1620), we find that fertility compared to natives increased over the decade. This pattern is consistent with the notion that immigration initially disrupts fertility. Temporal patterns of immigrant fertility will be analyzed in considerably greater detail below, holding constant a variety of other determinants of fertility.

4.2 Immigration and Fertility: Alternative Views Economic models of fertility (see, e.g., Becker 1981; and Schultz 1981) suggest that the major determinants of the demand for children are the woman’s potential market wage, her husband’s income (if she is married) or other sources of nonlabor income, costs of market inputs into producing children, and the tastes for children of the woman and her family. Increases in the woman’s own wage and her husband’s income have both income and substitution effects on the demand for children. Given the traditional division of labor in most families, with the wife providing the major time inputs into child rearing, however, the wife’s wage is expected to serve primarily as an indicator of the (opportunity) cost of time inputs. Thus, the effect of increases in the wife’s wage on the demand for children is expected to be negative, ceteris paribus (Butz and Ward 1979). Husband’s income, on the other hand, is expected to represent primarily the income available to the family. Its sign is uncertain a priori since an increase in income is expected to raise the demand for child quality as well as child quantity and thus to have an ambiguous effect on the number of children (Becker 1981).

98

Francine D. Blau

In the context of this model, immigrant fertility may differ from that of otherwise similar native-born women owing to differences in tastes or differences in wages (and husbands’ incomes) of immigrant and native women with similar characteristics. The prevailing view in the few previous economic analyses of immigrant fertility emphasizes what I shall term the assimilation model (Ben-Porath 1973; Bloom and Killingsworth 1985). In the case of highfertility source countries, immigrant women’s fertility is expected to exceed that of their native-born counterparts initially (reflecting conditions in the country of origin) but to approach native fertility over time with increasing residence in the United States.s Kahn (1988) finds cross-sectional evidence from the 1980 Census that is consistent with the notion of assimilation. BenPorath (1973) presents evidence from Israel that also supports this modeL9 This is certainly an intuitively appealing view. The climate in which a woman is reared is likely to influence her tastes and preferences for children. In addition, the type of human capital investments made by women in highfertility countries (controlling for their level) may be less market oriented, thus lowering their potential market wages and the opportunity cost of children. Finally, women who emigrate after reaching adulthood may have already begun their families, thus imparting a fairly direct relation between immigrant women’s fertility and conditions in the source country. Over time, we would expect these differences to diminish as immigrant women respond to economic conditions and opportunities in the United States and are increasingly exposed to prevailing attitudes toward fertility. This reasoning also suggests that the effect of source country characteristics should be greater for women who emigrated to the United States as adults. The assimilation model contains three separable although interdependent sets of predictions regarding (1) the initial level and time path of relative immigrant fertility, (2) (somewhat more indirectly) the level of overall fertility of a cross section of immigrants relative to otherwise similar natives, and (3) the effect of source country characteristics on immigrant fertility in the United States. While the assimilation model is intuitively appealing, there are plausible alternatives with respect to each of these components that need to be considered. lo The picture of immigrant-native differences in fertility and of the immigrant fertility adjustment process may in fact be quite different than that suggested by the assimilation model. I consider each of these predictions, and the corresponding alternatives, below. First, with respect to the initial level and time path of immigrant fertility, 8. Average fertility in the country of origin may be influenced by such underlying economic conditions as infant mortality rates and per capita income levels and as well as by tastes for children. The consequences of the reason for the higher fertility rates in the source country for immigrant fertility behavior in the United States are considered below. 9. See also the findings of Gorwaney et al. (1989). 10. Some of these are derived from the demographic literature. For useful summaries, see Kahn (1988) and Gonvaney et al. (1989).

99

The Fertility of Immigrant Women

the pattern predicted by the assimilation model may not be observed if the process of immigration results in a disruption or postponement of fertility. Such disruption may occur for two sets of reasons. First, there may be what can be termed economic disruption: wife’s wage and husband’s income may initially be depressed. While lower wife’s wage could be associated with greater fertility (owing to a lower opportunity cost), lower husband’s income may temporarily decrease fertility, depending on the sign of the income effect. Second, disruption may arise because of such demographic factors as delayed marriages or temporary separations of husbands and wives. If economic or demographic disruption occurs, the observed fertility of recent immigrants will be below their desired levels, and their relative fertility is expected to increase over time as actual fertility is adjusted to desired levels. While the assimilation model carries the strong prediction that the initial fertility of immigrants is above that of otherwise similar natives, the disruption model is focused solely on the time path of fertility and is consistent with either a positive or a negative initial differential. Ford (1990) and Bloom and Killingsworth (1985) find cross-sectional evidence that is consistent with disruption.” Second, with respect to the overall level of ceteris paribus immigrant-native fertility differentials, the assimilation model implies that, unless assimilation is virtually instantaneous (see below), the overall fertility of a cross section of immigrants from high-fertility source countries will be higher than that of otherwise similar natives since the fertility of more recent cohorts is expected to be unambiguously higher than that of natives, while the fertility of earlier cohorts will approach (but presumably not fall below) that of natives. On the other hand, if disruption is pronounced or has permanent consequences (women whose fertility is delayed may never attain desired levels), the overall fertility of immigrant women may be lower than that of natives after controlling for observed characteristics. Additional insight into immigrant-native differentials may be gained by considering the selectivity hypothesis. Immigrant women may be a selfselected group whose fertility is low relative to others in the source country owing either to tastes or to characteristics associated with labor market success. This may simply be the case because women who (for whatever reason) have fewer children are more mobile. In addition, where immigration is selective of relatively highly educated women, overall country average fertility rates are less likely to reflect the labor market opportunities and preferences of the group of women who actually emigrate to the United States.12 Consideration of the selectivity hypothesis suggests an additional factor 1 1 . In Bloom and Killingsworth (1985). fertility is modeled solely as a function of age. As noted above, in Ford’s (1990) study the fertility of recent immigrants is compared only to that of longer-term immigrants, not to that of natives. 12. Kahn (1988) finds that, the more selective immigration is of college-educatedwomen, the lower the fertility of the group in the United States.

100

Francine D. Blau

that would work to lower the fertility of immigrants relative to women in the country of origin. Immigration may be viewed as a form of human capital investment (Chiswick 1978). Thus, it may be selective not only of those who for various reasons have higher benefits relative to costs but also of individuals who are more future oriented (i.e., have lower discount rates). Such individuals may be more prone to engage in other types of human capital investments as well. This may be one reason why immigrants tend to have above-average levels of education relative to those prevailing in their countries of origin.13 An additional manifestation of this future orientedness might be a greater willingness to invest in child quality.I4 This could result in lower fertility levels not only relative to women in the country of origin but also relative to otherwise similar women in the United States. A final reason for expecting small fertility differences for immigrant women even initially is quite straightforward. To the extent that immigration is anticipated, it is possible that immigrant women’s fertility is adjusted to conditions in the United States prior to immigration. We may term this the instantaneous assimilation model. To the extent that this is the case, initial immigrant-native differences would be reduced, as would the responsiveness of immigrant women’s fertility to time spent in the United States. Third, a reasonable implication of the assimilation model is that, within the group of immigrants, fertility will vary systematically with source country characteristics, the most obvious one being the fertility rate in the country of origin. That is, the assimilation model suggests that source country variables will influence the fertility behavior of immigrant women in the United States. As a consequence, immigrants from different nationality groups may exhibit quite different fertility behavior in the United States, depending on conditions in their country of origin. An effect of source country characteristics on immigrant fertility behavior is also consistent with the disruption and selection models. It is not, however, consistent with the instantaneous adjustment model.

4.3 Empirical Framework The empirical work can be divided into two major sections. In the first, I estimate individual level fertility regressions in’each year separately for immigrants and natives. These regression results are first used to shed light on

13. For evidence of this, see Borjas (1991) as well as the results presented in table 4.3 below. 14. Schultz (1984) presents cross-sectional evidence consistent with this view. He finds that, while the children of recent immigrants have fewer years of schooling than children of otherwise similar natives, children of earlier immigrant cohorts receive somewhat more schooling than the children of native parents. He also finds that “virtually every measure of child health limitation, condition and disability is found less frequently among children of immigrant parents than among children of native parents” (p. 281), although, as Schultz points out, these subjective measures of health as reported by parents may be subject to cultural biases.

101

The Fertility of Immigrant Women

the extent of ceteris paribus immigrant-native fertility differentials. This is of interest in light of the second prediction of the assimilation model discussed above. A key question here is whether immigrants from these on average high-fertility source countries do indeed have higher overall fertility than natives, ceteris paribus, or whether, owing to disruption or self-selection, they constitute a low-fertility group relative to otherwise similar natives. I also consider the sources of observed trends over the period in the fertility of immigrants relative to natives (i.e., the rising relative fertility of immigrants over the period 1970-80). The regression results are then used to investigate the effect of years of residence in the United States on immigrant women’s fertility by constructing synthetic cohorts. This portion of the analysis is of interest in light of the first prediction of the assimilation model discussed above that immigrant women’s fertility will initially be high relative to natives but will approach that of native women over time. In the second portion of the analysis, a two-stage procedure is employed to examine the effect of source country variables on ceteris paribus fertility differentials by nationality among immigrants. The synthetic cohort approach (Borjas 1987) involves utilizing two nationally representative cross sections of data, in this case the 1970 and the 1980 Censuses, to track a particular cohort over time. The fertility of otherwise similar natives is used to determine the period effect. A drawback of this approach for which there is no obvious solution is that this comparison may be biased by the selectivity of the group included in each year owing to return migration on the one hand (see, e.g., Jasso and Rosenzweig 1990) and the 1970 Census undercount on the other.I5 As noted above, an assimilation effect would be implied if immigrant women’s initial fertility is high relative to their native-born counterparts but approaches that of native-born women over time. Alternatively, if immigrant women’s fertility tends to increase over time relative to otherwise similar natives (regardless of its initial level), disruption will be suggested. Finally, where considerable adjustment to conditions in the United States occurs prior to immigration, initial immigrant-native differences will be small, and there is expected to be little effect of length of residence on immigrant-native fertility differentials. The synthetic cohort approach is also helpful in addressing a problem that arises owing to the censored nature of the fertility variable. While some of the 15. On net, the undercount appears quantitatively more important than return migration for immigrants from these regions. The number of immigrant women aged 18-44 in 1970 who arrived in the United States before 1970, was 7,747 in the 1980 Census sample, compared to 7,317 in the 1970 Census sample. On the other hand, the number of native women aged 18-44 in 1970 in the sample declined slightly over the period (from 19,371 to 18,969), as would be expected owing to mortality and out-migration. The potential bias due to return migration or the undercount will depend on whether those who leave the United States and those who were omitted from the 1970 Census but included in the 1980 Census are a self-selected group with respect to their fertility behavior.

102

Francine D. Blau

women in the sample will have completed their fertility, the reproductive life of many is ongoing. In addition, for the reasons indicated, the fertility of immigrant women may follow a very different time path than that of native women so that cross-sectional comparisons, even between women with similar observed characteristics, may give a misleading picture of eventual fertility outcomes. Following a given cohort over time sheds light on the direction and magnitude of any differences in fertility between immigrants and natives that are likely to result as the assimilation or disruption process plays itself out. To make the analysis of individual fertility as comprehensive as possible, I estimate a reduced-form fertility model for all women, regardless of current marital status. I also estimate both reduced-form and structural fertility models for married spouse-present (MSP) women, a group for whom the determinants of fertility are better understood and better measured. The following equations are estimated separately for immigrants and natives in each year using ordinary least squares (OLS): (1)

(2)

FERTILITY, =

+ AGE-H,B, +

AGE,B,

+ X,B, + FERTILITY, = AGE, b,

EDUCATION,^?,

+

Y S M , B ~ ~e,,, ~

+ LNYHAT-H, by + LNWGHAT~b,

+ X,b, + YSMtbYSM+ e,2,

where FERTILITY, is a measure of cumulative fertility (number of children ever born) for individual i, l 6 AGE and AGE-H denotes the age and age squared of the woman and her husband (where present), EDUCATION includes the education of the woman and her husband (where present), X is a vector of control variables, YSM is a vector of years-since-migration dummy variables included for immigrants, LNYHAT-H and LNWGHAT are predicted values of the natural log of husband’s income and wife’s wage,17 and e,, and e,, are stochastic error terms. The reduced-form model includes controls for the underlying determinants of the woman’s potential wage and her husband’s income. The structural model explicitly includes the predicted values of LNWGHAT and LNYHAT-H as explanatory variables to understand their role in determining fertility patterns better. Variable definitions and means are shown in table 4.2. 16. The measure of fertility is truncated at zero. However, when selected specifications were estimated using tobit, a more appropriate technique under these circumstances, the results were quite similar. 17. LNYHAT-H and LNWGHAT were estimated on the basis of regression equations including controls for the individual’s education, potential experience and potential experience squared, disability status, racekthnicity, and years of residence in the United States (for immigrants), as well as region and standard metropolitan statistical area (SMSA) residence. Separate regressions were estimated for immigrants and natives (and their spouses) in each year. When a selectivitybias correction was included in the LNWGHAT regression (see Heckman 1980), it was found to be significant; however, the estimated magnitudes of the coefficients were implausibly large for some groups when FERTILITY was omitted from the first-stage regression (as would be appropriate in this case) and quite sensitive to specification. Given general concerns over the lack of robustness of this correction (Manski 1989), the OLS results were used.

103

The Fertility of Immigrant Women

The woman’s own age is obviously an important determinant of her cumulative fertility given the life-cycle pattern of childbearing. In the reduced form, the age and education variables are included as determinants of the wages of women (and the incomes of their husbands). The husband’s age and the education variables of the woman and her husband are thus excluded from the structural model. The control variables include race and ethnicity (HISPANIC, BLACK, and OTHERNW) as well as the nativity of the husband (FOREIGN-H), included for both immigrant and native women, as proxies for group differences in tastes (and incomes in the reduced form). I also include marital history (AGEMAR and TMSMAR) and marital status (MSP) variables to adjust for differences in tastes for children and in the costs and benefits of childbearing across these various states. Finally, location variables (SOUTH, NCENT, WEST, and SMSA) are used to control for differences in the costs of market inputs across locations. Since the marital history variables-age at first marriage (AGEMAR) and whether married more than once (TMSMAR)-may plausibly be outcomes rather than determinants of women’s fertility choices (Schultz 198 l ) , I also estimate reduced-form equations excluding these variables. Following Ben-Porath (1973), I explore the issue of the effect of level of maturity at the time of immigration by distinguishing between immigrant women whose first marriage occurred abroad (MARABR = 1) and those whose first marriage occurred in the United States (MARHERE = 1).l9Not only has the former group been subject to the effect of conditions in the source country for a longer period of time, but in addition these women may have begun childbearing before immigrating to the United States. They are also more likely to be “tied movers,” a factor that would lower their expected market wage in the United States (Mincer 1978). Since the Census variables giving information on when the woman arrived in the United States specify only the interval during which immigration occurred (e.g., between 1975 and 1980), I also define a group for whom it is not possible to determine whether the woman’s first marriage occurred in the United States or abroad (MARSAME = I ) . Since slope coefficients may differ across these groups, I examine their effects by estimating additional regressions separately for each group. The variable means shown in table 4.2 indicate that a high proportion of immigrant women from these source countries were recent arrivals in each year. In 1970, 65 percent had arrived in the preceding ten years; this was true of 58 percent in 1980. In addition, in each year approximately three-quarters of the married women arrived in the United States subsequent to or at about the same time as their first marriage. A review of the other variable means indicates that, compared to natives, immigrant women from these areas had lower average levels of education (1.6 years less in 1970 and 1.7 years less in 18. Ben-Porath (1973), on the other hand, favors the inclusion of marriage age since it may be exogenously delayed for immigrants. 19. See also the findings of Kahn (1988) regarding “adult” vs. “child” immigrants.

Table 4.2

Means of Individual Variables ~

1970 All Variables number of children ever born age AGE^ = age squared (100s) AGE-H = age of husband if married; 0 otherwise A G E ~ - H = age squared of husband if married (100s); 0 otherwise EDUCATION = years of school completed EDUCATION-H = education of husband if married; 0 otherwise LNYHAT-H = predicted natural log of husband’s income in 1979 dollars (1,OOOs) LNWGHAT = predicted natural log of women’s hourly wage in 1979 dollars TMSMAR = 1 if married more than once; 0 otherwise AGEMAR = age at first marriage if ever married; 0 otherwise MSP = 1 if married spouse present; 0 otherwise OTHMAR = 1 if separated, divorced, or widowed; 0 otherwise HISPANIC = 1 if Hispanic; 0 otherwise BLACK = 1 if black non-Hispanic; 0 otherwise OTHERNW = 1 if other nonwhite non-Hispanic; 0 otherwise FERTILITY = AGE =

~~

~

~

_

_

_

1980 Married

Married

All

Immigrants

Natives

Immigrants

Natives

Immigrants

Natives

Immigrants

Natives

2.028 33.91 1 12.445 26.765 1 1.264

2.098 34.536 13.137 27.599 11.792

2.454 35.155 13.145 39.203 16.499

2.508 36.377 14.262 39.519 16.884

1.943 33.515 12.189 25.594 10.676

1.764 33.271 12.192 23.953 10.042

2.347 34.939 13.01I 38.825 16.195

2.225 35.962 13.898 38.853 16.288

10.019 7.345

11.591 8.088

9.872 10.775

11.584 11.582

10.717 7.6650

12.388 7.816

10.606 11.620

12.375 12.677

...

...

2.502

2.703

...

...

2.446

2.695

...

...

1.452

1.492

...

...

1.448

1.457

,074 18.485

,109 16.902

,085 22.790

.123 20.53 1

,063 18.092

,124 15.624

,070 22.800

.I54 20.524

,683 ,130

,698 ,126

... ...

,659 ,136

,617 .149

,592 ,056 .224

,029

.028 ,080 ,009

.532 ,075 ,317

,036 ,127 ,012

,111 .010

... ... .590 .037 .244

...

...

...

...

,524 .050

,347

,033 .076 ,011

_

_

1 if South; 0 otherwise 1 if North Central; 0 otherwise WEST = 1 if West; 0.otherwise SMSA = I if SMSA resident; 0 otherwise FOREIGN-H = 1 if immigrant husband; 0 otherwise YSM 0-5 = 1 if 0-5 years since immigration; 0 otherwise YSM 6-10 = 1 if 6-10 years since immigration; 0 otherwise YSM 11-15 = 1 if 11-15 years since immigration; 0 otherwise YSM 16-20 = 1 if 16-20 years since immigration; 0 otherwise YSM 21-25 = 1 if 21-25 years since immigration; 0 otherwise YSM 25 = 1 if more than 25 years since immigration; 0 otherwise YSM 21-30 = 1 if 21-30 years since immigration; 0 otherwise YSM 30 = I if more than 30 years since immigration; 0 otherwise MARABR = 1 if first marriage occurred before immigration; 0 otherwise MARSAME = I if first marriage occurred at the same time as immigration; 0 otherwise MARHERE = 1 if first marriage occurred after immigration; 0 otherwise SOUTH =

NCENT =

+ +

N

...

,243 ,114 ,384 .857 ,653 .353

,254

...

.266

,150

...

.161

,089

...

,094

,053

...

,059

...

...

...

...

...

,062

...

,067

...

...

...

...

...

,241 .I06 .370 .869 ,446 ,393

.319 .211 ,168 .631

-

,015

.322 .286 .169 .616 .021

...

...

,244 ,101 ,417 ,942 ,489 ,316

.342 ,263 .I84 ,741 .014

.258 ,110

.350 ,273 ,181 ,715 ,023

...

,421 ,934 ,742 ,293

...

,263

...

,271

...

.I90

...

,193

...

,120

...

,121

...

...

...

.090

...

,097

...

...

...

.022

...

,024

...

...

...

.374

...

...

...

.337

...

...

...

.385

...

...

...

,403

...

...

...

.240

...

...

,259

...

25,549

6,034

17,697

8,838

Nore: Native means are weighted in inverse proportion to sampling probabilities.

... 22,786

30,298

15,021

18,402

106

Francine D. Blau

1980) and a greater representation of Hispanics and other nonwhites. As might be expected on the basis of these differences in personal characteristics, they had lower husband's income and own wages, although the magnitude of the latter difference is surprisingly small given the size of the immigrantnative educational differential. They were also, in 1980, more likely to be married spouse present and, in both years, tended to get married later-over two years later on average. I now turn to a description of my examination of the effect of source country variables on immigrant fertility. The analysis proceeds in two stages. In the first stage, reduced-form immigrant fertility functions are estimated including country dummy variables for each of the source countries identified in both the 1970 and 1980 Censuses. The following regressions were estimated separately by Census year (1970 and 1980) for all women and, for the MSP group, both combined and separately by the stage in the life cycle when they immigrated (i.e., MARHERE = 1, MARABR = 1 , and MARSAME = 1): (3)

FERTILITYl =

xiB

+ DIC + e,,

where FERTILITY~is the fertility of individual i, X includes the explanatory variables in (1) above, D is a vector of dummy variables for each source country, and ei is a stochastic error term.2oSince equation (3) does not include a constant term, C is essentially a vector of country-specific constant terms. The included countries as well as the frequencies of the immigrant sample by country are shown in appendix table 4A. 1. The sample is reduced somewhat in these analyses, primarily because the 1970 Census identifies considerably fewer specific source countries than does the 1980 Census.2' However, 93 percent of the original sample in each year is included, and the means of the country sample are quite similar to the full sample (see app. table 4A.2). In the second stage of the analysis, the coefficients on the country dummy variables, C,,, are regressed on source country variables in a pooled 1970 and 1980 regression in order to explain ceteris paribus differences in fertility by nationality group. The following country-level regressions are run separately for each group (i.e., all women and MSP = 1 , MARHERE = 1 , MARABR = 1 , MA MAR SAME = 1):

(4)

C,, = Z,,B,

+ YR70,T + en,,

where C,, is the coefficient on the country dummy variable for nationality n in year t estimated from equation (3), Z is a vector of source country variables, 20. AGEMAR and TMSMAR are excluded from these regressions so that the effect of intercountry differences in these factors on fertility will be captured by the country dummies. In addition, the race/ethnicity variables are excluded from these regressions because they may be considered to some extent an intrinsic characteristic of the country of origin. 21. Immigrants from Paraguay were deleted from this analysis because there were too few of them to permit meaningful analysis.

107

The Fertility of Immigrant Women

YR70 is a dummy variable for 1970, and en, is a stochastic error term.22No substantive interpretation is given to differences in the level of the country dummies between 1970 and 1980. Sources, definitions, and means of the source country variables are shown in table 4.3. As may be seen in the variable definitions, source country variables were measured at different points in time, depending on the period of immigration. Table 4.3 reports means weighted by the number of individuals in each source country-ysM cell in each Census year. The number of observations is given as the number of individuals (rather than countries). The source country variables employed in the country regression analyses are computed by weighting the level of the source country variables in each source country-ysM cell by the distribution of immigrants of that nationality across YSM categories. This was done separately for each group: all immigrants as well as those who, were married spouse present (MSP = l), married in the United States (MARHERE = l), married at about the same time as immigration (MARSAME = l), and married abroad (MARABR = 1). Following previous work (Kahn 1988; Ford 1990), I include a measure of source country fertility, the total fertility rate (TFR), to capture the effect of the source country environment on immigrant women’ taste for children. Obviously, a positive sign is anticipated and has been obtained in previous work. However, high fertility in the source country may be due not only to tastes but also to economic conditions. It seems reasonable to expect that immigrant women’s fertility will respond relatively quickly to changing economic conditions but that it will take considerably longer for their tastes to adapt. By controlling for the major economic determinants of the source country fertility rate, per capita GNP ( G N P ) , ~and ~ infant mortality rate (MORT), we can better measure the effect of the relatively more permanent taste effect of source country TFR on immigrant women’s fertility. The inclusion of these source country variables is expected to increase the estimated coefficient on TFR by making it a better measure of tastes. PROPEDUC, the proportion of women in the source country with the same or a higher level of educational attainment,24 22. Regressions are weighted by the inverse of the standard errors of the dependent variable. An alternative approach would have been to include the source country variables directly as explanatory variables in (3). However, it is likely that the regression errors will be correlated within groups (countries), in which case Moulton (1986) has shown that the (downward) bias in OLS standard errors on group-level variables ( i s . , the source country variables) can be quite large (see also Borjas 1990). By aggregating up to the country level, this problem is eliminated. 23. Unfortunately, reliable price-adjusted GNP data were not available for many of the source countries prior to the 1960s. 24. Note that this measure is based on enrollment data and does not take completion of the indicated level of schooling into account. Unfortunately, prior to 1960, enrollment ratios (as opposed to levels) for higher education are not available. In addition, it was frequently not possible to obtain enrollment data separately by sex for many of these countries in the pre-1960 period. The sex breakdown is important in that there are substantial differences in enrollment rates by gender in many cases.

Weighted Means and Sources of Country Variables

Table 4.3

Variables

1970

1980

TFR =

the total fertility rate: the average number of children that would be born to a hypothetical cohort of women if they experienced throughout their reproductive years the age-specific fertility rates prevailing in the indicated period: 1970-75 (1970s immigrant cohort), 1960-65 (1960s immigrant cohort), 1950-55 (pre-1960 immigrant cohort). (Source: United Nations, Demographic Indicators of Countries: Estimates and Projections as Assessed in 1980 [New York, 19821, and Demographic Yearbook [New York], various issues.)’

5.690

5.528

GNP =

average per capita GNP in 1979 U.S. dollars for 1973-75 (1970s immigrant cohort) or 1963-65 (pre-1970s immigrant cohort). (Source: U.S. Arms Control and Disarmament Agency, World Military Expenditures and Arms Transfers: 1971-80 [Washington, D.C.], and World Military Expenditures and Arms Transfers: 1963-73 [Washington, D.C.].)

1.149

1.258

,910

,795

MORT =

annual number of deaths of infants under 1 year per 1 ,OOO live births for 1970-75 (1970s immigrant cohort), 1960-65 (1960s immigrant cohort), and 1950-55 (pre-1960 immigrant cohort). (Source: United Nations Secretariat, “Infant Mortality: World Estimates and Projections, 195&2025,” Population Bulletin of the United Nations 14 [ 19821; and United Nations, Demographic Yearbook [New York], various issues.)

PROPEDUC =

the . proportion of women in the source country with the . same or higher educational attainment as the respondent. Estimated on the basis of enrollment data by level (i.e., primary, secondary, and higher) for 1970 (1970s immigrant cohort) and 1960 (pre-1970 immigrant cohort). (Source: Unesco, Trends and Projections of Enrollment by Level of Education and by Age, [September 19771, and Statistical Yearbook, various issues.)

,334

,333

DISTANCE =

number of kilometers direct distance (in thousands) between the country’s capital and the nearest U.S. gateway (Los Angeles, Miami, or New York). (Source: Gary L. Fitzpatrick and Marilyn J. Modlin, Direct-Line Distances, international ed., [Metuchen, N.J.: Scarecrow, 19861.)

4.586

5.278

,091

,086

8,274

21,232

RT =

refugees as a proportion of total immigrants during the 1970s (1970s immigrant cohort), 1960s (1960s immigrant cohort), 1950s (1950s immigrant cohort), and 1940s (pre-1950 immigrant cohort). (Source: U.S. Immigration and Naturalization Service, Statistical Yearbook of the Immigration and Naturalization Service, various issues, and Annual Report, various issues.)

No. of individuals

Note: Means are weighted by the number of individuals in each source country-ysM cell in each Census year. Sources listed were supplemented by World Bank, World Development Report (Washington, D.C.), various issues; and Statistical Yearbook of the Republic of China, various issues. “The total fertility rate was not available for some countries or periods. In order to have a consistent series, it was approximated by the gross reproduction rate multiplied 1/.488 in all cases.

109

The Fertility of Immigrant Women

is an inverse measure of the educational selectivity of immigrants and is expected to be positively related to fertility. Source country infant mortality rate (MORT) is expected to be inversely related to the fertility of immigrant women in that, at given levels of TFR, an increase in MORT reflects a smaller expected number of surviving children and presumably a lower demand for children. Since in most cases the United States infant mortality rate is considerably below that in the source country, this would reduce fertility in the United States. Since the fertility behavior of immigrant women who arrive as refugees may differ from that of economic immigrants, I include a measure of the proportion of the group composed of refugees ( R T ) . * ~It is unclear a priori whether a positive or a negative sign on RT is expected. On the one hand, the conditions that give rise to refugee flows may be expected to disrupt fertility leading to temporarily or permanently lower levels. On the other hand, immigration is less likely to be anticipated for this group than for economic immigrants; thus, they are more likely to have adjusted their fertility to the higher levels that are, appropriate to source country conditions. They are also more likely to anticipate permanent immigration to the United States, although it is unclear whether this factor would increase or reduce their fertility.26Finally, the direct distance between the source country and the United States (DISTANCE) is also included as a proxy for permanency of residence in the United States. The means of the source country variables suggest that there were no major shifts in average source country characteristics between the two Census years among immigrants from these areas. The increases in GNP per capita and declines in infant mortality probably reflect the effect of secular trends. There was also an increase in DISTANCE due to the increased proportion of immigrants from Asian countries (see app. table 4A. 1). It may be noted that, while immigrant women have considerably less education on average than the native born, they are a positively selected group relative to other women in the source country: in each Census year immigrant women were on average from about the top third of the source country educational distribution. Refugees comprised about 9 percent of total immigrants in each year. 25. An obvious weakness of this measure is that it relies on official definitions of refugees. 26. On the one hand, those who anticipate return migration may be less likely to “put down roots” in the United States and may postpone some or all of their childbearing. On the other hand, their fertility levels may be determined by source country conditions to a greater extent than that of permanent immigrants. In addition to measuring the substantive effect of MORT and RT on immigrant women’s fertility, the coefficients on these variables may also reflect underreporting of number of children ever born when the child dies in infancy or lives in other households (owing to the separation of family members among refugees). The omission of children whb have died or who live in other households is the most important source of error in fertility measures of this kind (United Nations 1983). My inclusion of MORT and RT controls for this possible bias in measuring the effect of other source country characteristics on fertility.

Table 4.4

Regression Results for the Reduced-Form Model 1970

Immigrants Variables

Coeff.

t

1980

Natives Coeff.

Immigrants t

Coeff.

Natives

t

Coeff.

t

16.69 -7.51 21.13 - 19.53 - 26.55 - 12.14 - 6.02 -51.15 2.85 57.88 6.52 12.72 1.40 4.29 7.31 12.16 -8.35 6.87 .04 -4.89 -5.83

,115 - ,071 ,150 -.164 - ,094 - ,021 - ,020 - ,094 .273 3.001

16.30 -7.29 20.25 - 18.30 -25.91 -5.98 - .81 -38.80 1.81 52.45 15.09 26.18 2.67 -5.95 3.29 .65 -8.90 1.88

All Women AGE

AGE^ AGE-H AGE*-H EDUCATION EDUCATION-H TMSMAR AGEMAR MSP OTHMAR HISPANIC BLACK OTHERNW SOUTH NCENT WEST SMSA FOREIGN-H

6-10 YSM 11-15 YSM 16-20 YSM 21-251 2 1-30

YSM

,196 - ,200 ,142 - .I45 - ,060 - ,052 - .435 -.118 1.562 3.952 ,216 ,408 - ,009 ,186 .374 .413 - ,172 ,024 .004 ,229 ,305

12.16 - 8.96 10.70 -9.55 - 11.00 - 8.69 -5.96 -29.16 5.27 34.91 3.57 4.20 -.I3 3.49 5.53 8.38 -3.11 .48 .08 3.96 4.30

,403

,276 - ,330 ,163 - ,171 - ,077 - ,038 - .292 -.I13 ,770 3.668 .540 ,718 .170 - ,095 ,127 ,025 - ,209 .165

30.84 -26.77 20.20 -17.97 -17.03 -8.82 -8.44 -38.46 4.50 50.44 11.16 20.00 4.28 -3.22 4.26 .76 -9.65 2.27

,147

- .091

...

... ... ...

,170 -.182 -.078 -.039 -.257 -.116 .495 3.587 ,263 ,682 .057 .I30 ,283 ,337 -.362 ,205 .001 - ,150 - ,211

4.51

...

...

- ,106

- 2.55

...

...

,595

6.59

...

...

- ,006

- .08

...

...

-3.023 ,383 8,838

- 10.98

-22.45

- 1.973

- 12.60

- ,920 ,458 30,298

-7.94

27.64 -20.02 6.07 -7.07 -17.11 -11.09 -10.64 -51.56 6.59 9.71 1.54 3.40 6.95 10.32 -7.32

.294 -.266 ,029 -.048 -.OM -.024 -.I61 -.I07 ,653 ,674 ,069 -.170 ,095 ,021 -.I57

22.97 -16.40 2.80 -4.18 -16.01 -5.69 -5.29 -36.46 13.84 15.66 1.68 -5.39 2.90 .60 -6.43

,510

,682 ,077 - ,137 ,079 ,016 - ,168 ,108 .

I

.

...

YSM25+/

30 +

Constant Adjusted R2 N

- 3.411 ,377 25,539

.456 22,786

Married Women AGE

AGE^ AGE-H AGE~-H EDUCATION EDUCATION-H TMSMAR AGEMAR HISPANIC BLACK OTHERNW SOUTH NCENT WEST SMSA

,427 -.460 ,035 -.047 -.061 -.043 -.744 -.146 ,313 .512 -.008 ,203 ,456 ,505

-.157

17.73 -14.46 2.30 -2.79 -8.31 -6.24 -8.77 -29.83 4.15 3.69 -.09 3.01 5.51 8.14 -2.37

,470 -.539 ,038 -.058 -.079 -.036 -.442 -.I27 ,596 ,767 ,115 -.I31 ,134 ,048 -.231

33.01 -30.01 3.52 -4.94 -12.57 -7.28 -11.02 -37.13 9.85 15.14 2.26 -3.54 3.60 1.15 -8.68

,385 -.363 ,058 -.075 -.068 -.042 -.537 -.142 .328 ,716 ,077 .131 .330 ,365 -.373

111 Table 4.4

The Fertility of Immigrant Women (continued)

1970

1980

Immigrants Coeff.

f

Immigrants

Natives Coeff.

f

Coeff.

Natives

f

Coeff.

f

4.82 -2.10 -7.64 -8.77

,108 -.157 ,108

1.78 -6.43 1.78

...

...

-5.48

...

...

Married Women FOREIGN-H

6-10 YSM 11-15 YSM 16-20 YSM 21-251 21-30 YSM 25+1 30 YSM

+

Constant Adjusted R2

N

,023 -.012 ,244 ,282 ,307 ,570 -3.161 ,336 6,034

.44 -.20 3.38 3.18 2.83 5.11 -8.14

.197 -.231 .197

2.60 -8.68 2.60

...

...

.150 -.072 -.299 -.408

..

...

-.285

.

... -3.394 .281 17,697

... -17.32

-.209

-2.33

-2.998 .407 15,021

-13.82

... -1.298 ,347 18,402

... -7.76

4.4 An Analysis of Immigrant-Native Differences The reduced-form regression results for the estimation of equation (1) are shown in table 4.4 for all women and married women, separately. Looking first at the results for the years since migration (YSM) variables in 1980, we see a pattern that appears to support the assimilation model. Controlling for the individual characteristics of the immigrants (including age at first marriage), fertility is highest for recent arrivals and declines with time spent in the United States. However, the findings for 1970 show a very different picture. Fertility appears to be lowest for recent arrivals and to increase with additional time in the United States. Reduced-form results omitting the marital history variables (AGEMAR and TMSMAR) are shown in appendix table 4A.3. The coefficients on the years since migration dummy variables are considerably increased by the exclusion of age at first marriage (AGEMAR). Although age at first marriage was actually fairly constant across immigration cohorts, owing to the high proportion of married women whose first marriage occurred at or prior to the period of migration, the YSM variables tend to become proxies for marital duration when AGEMAR is omitted. In other respects, the results for both reduced-form specifications are quite similar. Finally, the regression results for the structural model estimated for married women are shown in table 4.5. For both immigrants and natives, husband’s income and wife’s wage (LNYHAT-H and LNWGHAT) are found to have significant negative effects on fertility. The former finding is consistent with considerable cross-sectional evidence and suggests a larger income elasticity of demand for child quality than for child quantity (Becker 1981). The coefficients

112

Francine D. Blau

Table 4.5

Regression Results for the Structural Model: Married Women 1970

Immigrants Variables AGE AGE* LNYHAT-H LYWGHAT

TMSMAR AGEMAR HISPANIC BLACK OTHERNW

SOUTH NCENT WEST SMSA FOREIGN-H

6-10 11-15 YSM 16-20 YSM 21-251 2 1-30 YSM25+/30+ YSM

YSM

Constant Adjusted R2

N

Coeff.

t

,493 -.549 -.385 -1.646 -.763 -.148 ,065 ,523 -.209 -.197 ,411 ,275 ,078 -.016 .147 .616 ,621

23.30 -19.04 -4.29 -11.25 -9.00 -30.71 .79 3.70 -2.54 -2.70 4.93 4.12 1.14 -.29 2.38 8.11 6.75

.604 .968

5.45 8.30

-1.339 ,334 6,034

-3.27

1980

Natives Coeff. ,544 -.640 -.200 -1.083 -.450 -.127 ,556 ,586 .203 -.292 ,050 -.029 -.082 ,151

Immigrants t

... ... ...

... ...

,515 -.542 -.334 -2.128 -.559 -.I45 ,168 ,714 ,088 -.164 .407 ,389 -.084 ,152 ,315 ,253 ,204

...

... ...

,424 ,653

-16.78

-1.920 ,405 15,021

... -3.265 ,278 17,697

48.75 -42.86 -3.83 -14.90 -11.20 -37.61 9.01 10.65 3.93 -7.57 1.33 -.70 -2.83 2.01

Coeff.

...

t

43.43 -33.67 -7.38 -23.24 -11.08 -53.56 3.29 9.56 1.76 -4.06 8.49 11.03 -1.60 4.85 8.52 5.64 3.83 7.05 6.72 -8.99

Natives Coeff. ,377 -.377 -.204 -1.176 -.I78 -.lo9 .690 ,694 .I38 -.267 ,077 .029 ,023 ,088

t

38.77 -29.38 -4.84 -17.59 -5.84 -37.41 14.58 14.90 3.32 -8.32 2.32 .81 .86 1.44

.. ... ...

... ...

... ...

...

...

...

-1.619 -10.02 .344 18,402

on the years since migration dummies (YSM) are considerably increased when controls for husband’s income and women’s wage are included. The difference between the coefficients on the YSM variables in the reduced-form and structural models reflects the indirect effect of years since migration on fertility via its effect on wages and incomes. This indirect effect appears to be strongly negative: increases in wages and incomes across immigration cohorts lower fertility. Further clarification of the relation of these findings to the time pattern of immigrant fertility will be gained when I explicitly track the various immigration cohorts across the two Censuses below. Here, I focus on the magnitude of immigrant-native differences in the cross section of each Census as well as on trends over time in the magnitude of this difference. Table 4.6 decomposes the immigrant-native fertility differential into a portion accounted for by immigrant-native differences in means and a portion that is “unexplained,” that is, is due to differences in the response of immigrants and natives to the same characteristics. It may be recalled that immigrantnative differences in fertility were found to be relatively small. In this portion

113

The Fertility of Immigrant Women

Table 4.6

Decomposition of Immigrant-NativeDifferences in Fertility (native functions) Married Women All Women, Reduced Form

(1)"

(ab

Reduced Form (1)"

(ab

Structural

(3)

1970 Due to means Age Education Marital status Raceiethnicity Location Income/wages Unexplained Total

,462 .027 ,198 ,025 ,278 - ,066

,306 - ,168 ,148 ,012 ,372 - ,059

,557 ,071 ,238 t

.

.

,315 - ,067

,327 - ,249 .I64 ,017 ,453 - ,058

...

...

...

...

- ,530 - ,068

- ,373 - ,068

- ,608 - ,052

- ,379 - .052

,281 - ,238

... .017 ,430 - ,012 ,083 - ,333 - ,052

1980 Due to means Age Education Marital status Race/ethnicity Location Income/wages Unexplained Total

,420

,457 ,171 ,201 - ,077 ,198 - ,037

,337 - ,062 .I60 - ,025 .292 - .028

...

...

- ,274

- ,154

,182

,182

- .284 ,135

- ,044

,257

... ,250 - ,043

,264 - ,305 .180 .013 ,404 - .029

- .129 ,135

.230 - ,299

... ,015

,429 ,024 ,061 - ,095 ,135

Note: The immigrant-native difference due to means is M = 2, B,NX,, - Z,B,NX,N,where X, and B, are the mean and estimated regression coefficient of variable i, and subscripts N and I denote natives and immii.e., (I = grants, respectively. The remainder of the total differential (T) is considered unexplained (U), T - M . Age includes AGE, AGE^, AGE-H,A G E ~ - H , and AGEMAR. Education includes EDUCATION and EDUCATION-H. Marital status includes TMSMAR and also MSP and OTHMAR (where applicable). Race/ ethniciry includes HISPANIC, BLACK, OTHERNW, and FOREIGN-H. Location includes SOUTH, NCENT, WEST, and SMSA. Incomelwages includes LNYHAT-H and LNWGHAT. 'Excludes AGEMAR and TMSMAR. bIncludes AGEMAR and TMSMAR.

of the analysis, we may ascertain whether this is due to a similarity (or offsetting differences) in characteristics or whether it appears to be due to behavioral differences. The latter would suggest some form of selectivity, while a comparison of the reduced-form and structural models will shed light on the relative importance of selection associated with labor market outcomes and that associated with tastes. It was also found that immigrant fertility had increased relative to native fertility over the period. This trend may similarly be related to changes in characteristics of immigrants relative to natives or to shifts in the responses of each group to these characteristics.

114

Francine D. Blau

Looking at the reduced-form results for specification (1) excluding the marital history variables-age at first marriage (AGEMAR) and whether married more than once (TMSMAR)-we see that immigrants have characteristics that are associated with higher fertility; that is, substantial positive mean effects are obtained. The lower education of immigrant women and their husbands and the effect of immigrant-native differences in race/ethnicity-chiefly, the considerably higher representation of Hispanic women (a high-fertility group) among immigrants-are the primary factors working to increase immigrant women’s fertility relative to natives. In both years, however, immigrant women had fewer children ever born than otherwise similar natives-that is, negative coefficient effects were obtained. Immigrant fertility was .553 (all women) to .608 (married women) lower in 1970 and .274 (all women) to .284 (married women) lower in 1980. A comparison of the unexplained differentials under the various specifications shown in table 4.6 suggests some of the factors that are responsible for these large ceteris paribus immigrant-native differences. With respect to marital history, there are effects in opposing directions. The higher age at first marriage of immigrants tends to lower their fertility relative to natives, while their lower incidence of marital breakup increases their fertility. On net, however, immigrant-native differences in marital histories lower relative immigrant fertility and thus can account for a substantial portion of the unexplained differential. Among married women, when AGEMAR and TMSMAR are included (specification 2), the unexplained differential is decreased in absolute valu by 37.7 percent (from - .608 to - .379) in 1970 and by 54.5 percent (from - .284 to - .129) in 1980. Finally, comparing the results from the structural model (specification 3) for married women, we find that, in each year, the unexplained difference between immigrants and natives is somewhat smaller in absolute value in the structural model than in the reduced-form model (specification 2): the unexplained differential is now - .333 for married women in 1970 and - .095 in 1980. This suggests that selection of immigrants with respect to unobserved characteristics associated with labor market outcomes does play a role in reducing their fertility relative to otherwise similar natives. The remaining unexplained difference is presumably due to selfselection with respect to tastes (although large disruption effects could also contribute to this; see below). An examination of the structural model results suggests one important source of such unexplained differences. The coefficients on the income/wage variables are considerably larger in absolute value for immigrants. Differences in these two coefficients alone are more than sufficient to account for the unexplained differential. Evaluated at the immigrant means, immigrant-native differences in the coefficients on LNYHAT-H and LNWGHAT would work to reduce the fertility of immigrants relative to natives by over one child in each year. This is indirect evidence in support of the view that immigrants have a higher demand for child quality than otherwise similar natives, with the larger nega-

115

The Fertility of Immigrant Women

tive coefficient on LNWGHAT for immigrants further suggesting that immigrant women are more responsive to labor market opportunity costs than their native-born counterparts.

4.5 Rends in the Immigrant-Native Differential We now consider the sources of the increase in the fertility of immigrants relative to the native born that occurred between 1970 and 1980. (The total predicted differential increased by .250 among all women and .187 among married women.) The results in table 4.6 indicate that this was principally due to a decline in the absolute value of the unexplained differential over the period and, by implication, a reduction in the extent of immigrant self-selection relative to natives. However, it is not clear whether this decrease was due to shifts in immigrant behavior or shifts in native behavior or a combination of both. Some light can be shed on this issue by considering fertility levels of immigrants relative to similar natives by years of residence in each of the two years. In table 4.7, the immigrant and native fertility functions are evaluated at the immigrant means. The results are presented separately by years-sincemigration group and for reduced-form (specification 2 in table 4.6) and structural (specification 3 in table 4.6) models. I focus initially on the reducedform model and on fertility levels (rather than immigrant-native differences). The figures suggest that the trend toward increasing relative fertility of immigrants may reflect period shifts in fertility behavior within the United States to a greater extent than changes in the selectivity of immigrants relative to others in the source country. Over the decade, the predicted fertility of a native woman with mean immigrant characteristics declined by .304 among all women and by .348 among married women. At the same time, among immigrants, fertility decreased as well, but to a lesser extent, by .085 among all women and by .097 among married women. The trend toward lower fertility among the native born of course reflects declining domestic birth rates as the “baby bust” followed the “baby boom” in the United States. There was, however, no comparable trend in average fertility in these source countries-although there were declines in individual cases. Thus, while TFR for the United States fell from 3.32 in the early 1960s to 1.97 in the early 1970s, average source country TFR remained roughly constant at 5.53-5.69. Two additional pieces of data support the notion that conflicting fertility trends in the United States and source countries are responsible for the rising relative fertility of immigrants. First, the relative increase in immigrant fertility that occurred over the decade was primarily due to an increase in the fertility of recent arrivals compared to otherwise similar natives. Whereas, all else equal, recent arrivals (YSM 0-10) had about half a child less than natives in 1970, in 1980 they had about the same number of children among married women and only .09 fewer among all women. There was, however, no nar-

116

Francine D. Blau

Table 4.7

Adjusted Immigrant-NativeDifferences in Fertility by Years since Migration (immigrant means) All Women 1970

1980

RF

RF

Married Women

-~ RF

1970

1980 Struct.

RF

Struct.

Predicted Means Natives: Native means Immigrant means All immigrants

0-5 YSM 6-10 YSM

11-15 16-20 YSM 20-k YSM

YSM

2.096 2.401

1.761 2.097

2.506 2.833

2.506 2.787

2.222 2.485

2.221 2.452

2.018 1.908 1.912 2.136 2.213 2.414

1.943 2.006 2.007 1.856 1.795 1.920

2.454 2.335 2.323 2.579 2.618 2.783

2.454 2.157 2.305 2.773 2.779 2.955

2.357 2.516 2.444 2.217 2.109 2.246

2.357 2.141 2.456 2.394 2.345 2.610

- ,129 .031 - ,041 -.268 -.377 - ,240

- ,095 - .311 - ,005 -.058 -.I07

Adjusted Differentials All immigrants YSM

0-5

6-10 11-15 YSM 16-20 YSM 20 -k

YSM

YSM

- ,373 -.494 -.490 - ,265 - ,188 ,013

- ,154 -.091 -.090 - ,242 .303 - ,178

-

-.379 - ,498 -.510

-.254 -.216 - ,051

-.333 - ,630 -.483 -.014 -.009 ,168

,158

Note: Adjusted immigrant-native differences are equal to z,E,,X,,- z,B,X,,, where X , and B, are the mean and estimated regression coefficient of variable i, and subscripts N and I denote natives and immigrants, respectively. The reduced-form model includes controls for AGE, AGE^, AGE-H, A G E ~ - H ,EDUCATION, EDUCATION-H, TMSMAR, AGEMAR, MSP, OTHMAR, HISPANIC, BLACK, OTHERNW, SOUTH, NCENT, WEST, SMSA, and FOREIGN-H. The structural model includes controls for LNYHAT-H and LNWGHAT as well as AGE, AGE^, TMSMAR, AGEMAR, HISPANIC, BLACK, OTHERNW, SOUTH, NCENT, WEST, SMSA, and FOREIGN-H. Combined YSM categories are weighted averages of the relevant YSM categories where the

weights are based on the immigrant frequencies across categories in each year. RF = reduced form; Struc. = structural.

rowing of the ceteris paribus immigrant-native differential in the case of longer-term residents. The findings for the structural model show a similar picture of rising relative fertility of recent arrivals relative to natives, all else equal. While U.S. trends appear to be relatively more important in generating the increase in the fertility of immigrants relative to comparable natives, differences in the degree of selectivity of immigrants relative to the source country population may also have played a role. The fertility of recent arrivals (YSM 0-10) with mean immigrant characteristics increased over the period, by about .10 among all women and .12 (YSM 6-10) to .18 (YSM 0-5) among married women for the reduced-form results. There was also a small increase for the YSM 6-10 category for the structural model.

117

The Fertility of Immigrant Women

Table 4.8

Adjusted Immigrant-Native Differences in Fertility by YSM Category and Where Married I970 MARHERE

MARABR

1980 MARSAME

MARHERE

MARABR

MARSAME

Predicted Means Natives: Native means Immigrant means: Age only' All variablesb Immigrants

2.506

2.506

2.506

2.222

2.222

2.222

2.326 2.405 2.252

2.894 3.514 3.058

2.397 2.440 1.994

1.892 2.077 1.778

2.565 3.171 3.310

2.014 2.175 1.931

Differentials Unadjusted Adjusted for: Age only' All variablesb

- ,254

.552

- ,512

- ,443

1.088

- ,290

- .074 -.I52

,164 - ,456

- ,403 - ,446

-.I14 - ,298

,745 ,139

- ,083 - ,244

*Estimatedfrom regression equations including controls for AGE and AGE' only bEstimated from regression equations including controls for all variables in the reduced-form model (eq. UI).

A second piece of data supporting the notion of conflicting fertility trends is given in table 4.8, which shows ceteris paribus immigrant-native fertility differentials separately by the stage of the life cycle at which immigration occurred. Recall that we expect women who were married prior to or at approximately the same time as immigration to exhibit the strongest influence of source country characteristics on their fertility. Thus, if differences in fertility trends between the United States and source countries were responsible for the rise in relative immigrant fertility over the period, we would expect this to be manifested by rising relative fertility of the married-abroad group (MARABR = 1) and, to a lesser extent, of the group who married at about the same time as immigration (MARSAME = 1) rather than the group who married in the United States (MARHERE = 1). This is indeed precisely what we find. All three groups of immigrants had lower fertility than otherwise similar natives in 1970. However, the fertility of the MARABR and MARSAME groups increased relative to natives over the period, while the fertility of the MARHERE group decreased further relative to natives, all else equal. By 1980, the fertility of the MARABR group was .14 higher than otherwise similar natives, and the fertility differential between the MARSAME group and comparable natives had been cut by almost 50 percent. Trends in the United States appear to be responsible for the decreased immigrant-native fertility differential for the MARSAME group. The fertility of these women declined somewhat over the period, but not as rapidly as that of natives with similar characteristics. As in

118

Francine D. Blau

the case of recent arrivals, however, there is evidence that both period effects in the United States and shifts in the degree of selectivity of immigrants played a role in increasing the relative fertility of immigrant women who were married abroad. While the fertility of natives with the mean characteristics of the MARABR group declined by .34 over the period, the fertility of the MARABR group was .25 higher in 1980 than in 1970. Finally, as noted above, from a policy perspective, immigrant-native differences due to differences in characteristics are also of interest. Adjusting only for age, we find that, among married immigrants, women whose first marriage occurred in the United States or at about the same time as immigration have lower fertility than natives in both years. However, the fertility of women who were married abroad was .164 higher than natives in 1970 and .745 higher in 1980. The latter constitutes a substantial difference.

4.6 The Effect of Years of Residence: Assimilation versus Disruption As we have seen, the assimilation model predicts that the fertility of immigrants from high-fertility source countries will initially exceed that of otherwise similar natives but will approach that of the native born over time. The instantaneous assimilation model suggests that initial immigrant-native differences will be small and relatively constant over time. Finally, the disruption model leads us to expect that, since immigrant women’s fertility is initially below desired levels, regardless of its initial level, it will increase relative to otherwise similar natives over time. The results in table 4.7 above are not consistent with the prediction that immigrants from these, on average, highfertility source countries will have relatively high ceteris paribus fertility on arrival in the United States. In 1970, recent immigrants exhibit lower fertility than native-born women with similar characteristics. Although the extent of this difference declined over the decade, even in 1980 the fertility of recent immigrant women was no higher than that of their native counterparts. This is powerful evidence against the assimilation model, for even if we were to find declining relative fertility of immigrants over time, the notion of assimilation would not make much sense if immigrant fertility were not approaching native fertility. In order to examine the fertility behavior over time of given immigration cohorts, we must take into account the results from both the 1970 and the 1980 cross sections. Assuming that the underlying parameters are constant over time, the intertemporal pattern can be ascertained by comparing immigrant-native differentials for the particular cohort in 1970 to the differential that prevailed for that group in 1980 (Heckman and Robb 1985; Borjas 1987). These results are presented for the red~ced-form~~ and structural models in table 4.9. Note that immigration cohort is given by year of arrival (ARR) variables. 27. The reduced-form specification includes controls for AGEMAR and TMSMAR.

119

The Fertility of Immigrant Women Synthetic Panel Estimates of Immigrant-Native Differences in Fertility with and without Aging, 1970-80 (1970 immigrant means)

Table 4.9

Without Aging’

With Agingb

Married All, RF

RF

Married

Structural

(1)

All,

RF

RF

(2)

Structural (1)

(2)

1970 -.494 -.490 -.236 .013 -.373

65-70 60-64 ARR 50-59 ARR

ARR

ARR PRE-50

All

-.498

-.510 .-.240 -.051

-.379

-.630 -.483 -.012 .I68 -.333

-.581 -.577 -.324 -.075 -.460

-.537 -.478 -.081

,082

...

-.668 -.680 -.410 -.220 -.549

-.723 -.576 -.lo5 ,074 -.426

-.630 -.571 -.I75 -.012

-.I97 -.305 -.I83 -.I06 -.211

-.I26 -:175

-.098 -.I86 -.012

...

~

1980 65-70 60-64 ARR 50-59

ARR

ARR

ARR PRE-50

All

-.253 -.314 -.208 - ,108 -.236

-.289 -.397 -.275 -. 198 -.298

-.I07 -.I56 ,064 ,293 -.023

-.079 -.167 .007 .179

...

-.205 -.266 -.160 -.060 -.193

,045

,274 -.045

,160

...

Change 197WO

65-70 60-64 ARR 50-59 ARR

ARR

ARR PRE-50

All

,241 ,176 ,028 - ,121 ,137

,209 ,113 - ,035 - .I47 ,081

.523 ,327 ,076 ,125 .310

,458 .311 .088 ,097

.376 .311

...

,267

,164 ,015

,471 ,375 ,227 . I 14 ,338

,597 ,401 ,150

,200 ,381

.532 ,385 ,162 ,172

...

Nore: RF = reduced form. ‘Fertility, husband’s income, and wage functions for each group (natives and immigrants) and each year (1970 and 1980) are evaluated using 1970 immigrant means, including the 1970 distribution across period of arrival categories. In structural specification 1, LNYHAT-H and LNWGHAT are evaluated at their overall mean levels for immigrants arriving prior to or during 1970. In structural specification 2, LNYHAT-H and LNWGHAT are evaluated for immigrant women and their husbands who arrived during the indicated period (1965-70, 1960-64, etc.). bComputed as above but evaluating the immigrant and native functions at AGE = 25 and AGE-H = 28.142 in 1970 and AGE = 35 and AGE-H = 38.142 in 1980. The age difference between husband and wife is based on the 1970 differential for immigrants. In addition, in the structural model, LNYHAT-H is evaluated at 11.367 years of potential experience in 1970 and 21.367 years in 1980. LNWGHAT is evaluated at 9.128 years of potential experience in 1970 and 19.128 years in 1980. Potential experience at the indicated ages is calculated on the basis of immigrant means for education in 1970.

Fertility, husband’s income, and wage functions are evaluated at 1970 immigrant means throughout. Two formulations are employed. The first, “without aging,” simply matches up the appropriate cohorts across the Censuses. Thus, for example, the cohort that arrived in 1965-70 had YSM 0-5 in 1970 and YSM 11-15 in 1980. The results are similar to those given in table 4.7;

120

Francine D. Blau

however, some additional computations are performed for YSM groups not included in that table (e.g., the YSM ~ O + / A R RP R E - ~ Ogroup in 1980), and 1970 immigrant means are employed. In the second formulation, aging is allowed for by evaluating the fertility functions at AGE = 25 in 1970 and AGE = 35 in 1980. Other variables, including the age of the husband and the potential experience at which LNYHAT-H and LNWGHAT are evaluated, are set at appropriate levels to correspond with these ages, given 1970 immigrant means. In addition, the evaluation of the structural model is presented in two ways. In specification 1 , LNYHAT-H and LNWGHAT are evaluated at the mean level for all immigrants arriving during or prior to 1970, given the distribution of immigrants across YSM categories in 1970. In specification 2, LNYHAT-H and LNWGHAT are evaluated for immigrant women and their husbands who amved during each indicated period (i.e., 1965-70, 1960-64, etc.).28The results for specification 2 reflect the tendency of incomes and wages to increase across immigrant cohorts who have resided in the United States longer. The results in table 4.9 are strongly consistent with the disruption model. The relative fertility of immigrants tended to increase over the decade, with the largest and most consistent increases obtained for the recent arrivals in 1970 (ARR 60-70), the group most likely to have had its fertility disrupted by the immigration process.29A comparison of the results for the reduced-form and structural models suggests that demographic rather than economic factors underlie the observed pattern of disruption. We find that estimates of disruption (gains in relative fertility over the period) are actually larger in the structural model, where wages and husband’s income are controlled for. This is inconsistent with the view that disruption is caused by initial economic dislocation. Qn the contrary, the indirect effect of the economic assimilation of immigrants (their rising relative wages and husbands’ incomes) over time appears to have a negative effect on their fertility.3oThis result is not surprising in that negative coefficients on both LNYHAT-H and LNWGHAT were obtained in the structural model. Finally, we may address the issue of selectivity by asking whether immigrant fertility remains below that of otherwise similar natives as this disruption process is worked out or whether it eventually catches up to and even exceeds native fertility. The reduced-form results in table 4.9 indicate that immigrant fertility remains below that of natives with similar personal characteristics, although 1980 immigrant fertility is higher relative to natives 28. These estimates are obtained using the estimated coefficients on the YSM variables and assuming the overall 1970 immigrant means for the remaining explanatory variables. 29. Evidence of disruption is also obtained for the reduced-form model estimated excluding AGEMAR and TMSMAR. For the estimates without aging, this specification produces somewhat larger estimates of disruption than those reported for the reduced-form model in table 4.9. For the estimates with aging, this specification produces somewhat smaller estimates. 30. The direct effects of immigration cohort (ARR)on fertility given by the structural results are larger than the total effects given by the reduced-form results. This implies a negative indirect effect on fertility via wages and incomes.

121

The Fertility of Immigrant Women

when aging is taken into account than when it is not. Nonetheless, these results suggest that some of the lower ceteris paribus fertility of immigrants observed in the 1970 cross section, particularly for the relatively recent arrivals, represents the effect of disruption. Overall, when fertility is evaluated at age 25 in 1970, immigrant women are estimated to have .46 fewer children than native women with similar characteristics. By 1980, at age 35, women in these immigration cohorts are estimated to have only .193 fewer children. Immigrant fertility is also found to rise relative to natives in the estimates without aging effects, although the estimated 1970-80 increase is smaller (.137 vs. .267). While the findings for the reduced-form model imply that immigrant fertility remained below that of natives with similar personal characteristics, the results from the structural model suggest that, by 1980, immigrants who arrived during or prior to 1970 had about the same fertility as natives with similar wages and husband’s income. This in turn implies that, for married women, immigrant-native differences in labor market outcomes are sufficient to explain the remaining lower fertility of immigrants with similar personal characteristics in 1980. The structural results further indicate that immigrant fertility may eventually surpass that of natives with similar wages and husband’s incomes; specifically, a positive immigrant-native difference is obtained for immigrants who arrived before 1950. However, these differences are relatively small (. 16-. 18 children) when structural specification 2 is employed, that is, when incomes and wages are evaluated for immigrant women and their husbands who arrived during the indicated period. Broadly, I take these results as consistent with the notion of selectivity of immigrants aa a low-fertility group due to unobserved characteristics associated with labor market outcomes. The fact that immigrant women appear to be a selected low-fertility group relative both to source country populations (see below) and to natives with similar personal Characteristicsdoes not mean, however, that they have the same fertility as they would have had in their countries of origin. The pattern of disruption makes it difficult to discern the process by which immigrant women adapt their desired family size to economic conditions in the United States, and particularly the speed with which this adaptation occurs, but it does not necessarily mean that such adaptation is absent. Some further light is shed on these issues in my consideration of the effect of source country characteristics on immigrant women’s fertility below. My finding of a substantial disruption effect among the cohort of immigrants who arrived in the 1960s raises the issue of the future fertility of immigrant women from these source countries who arrived in the 1970s. As we have seen, they had higher adjusted fertility relative to both natives and longer-term immigrants in 1980 than had recent arrivals in 1970. If the fertility of recent arrivals in 1980 follows the disruption model, as did that of their predecessors in 1970, their fertility relative to natives could increase considerably over the next decade. On the other hand, the fact that their initial fertil-

122

Francine D. Blau

ity level is comparable to that of similar natives (rather than below the native level, as had been the case in 1970) may exert some downward pressure on the extent of their relative fertility increase, even if it follows the disruption model. 4.7 The Effect of Source Country Variables The potential importance of source country characteristics is indicated by the results presented in table 4.10, which are based on fertility regressions including country dummy variables. Specification 1 controls only for the women’s own age (and the YSM variables) and specification 2 for all variables included in the reduced-form model (see eq. [3] above).31In order to compare immigrant women’s fertility in the United States to source country fertility measured by the total fertility rate (TFR), a proxy for completed fertility, the equations have been evaluated at age 45 for women and, in specification 2, at 48.142 for husbands (if present). (The latter figure reflects the husband-wife age difference among immigrants in 1970.) It may be noted that, in times of shifting fertility patterns, the use of the coefficients on the age variables in this fashion has similar drawbacks to the use of the YSM variables to measure the effect of residence on immigrant women’s fertility. That is, to some extent age measures the effect of belonging to a cohort that had its children during a period of relatively high or low fertility. Since fertility was decreasing during this period, completed fertility is likely to be overstated. Selectivity of immigrants relative to women in the source countries, as well as their adaptation to conditions in the United States, is suggested by the large number. of cases in which the estimated completed fertility of immigrants is considerably lower than the overall average for the source country during the relevant period. Since the estimate of immigrant women’s completed fertility is most likely overstated, these may be considered conservative estimates of the true differences in fertility. The results for specification 2 in table 4.10 also suggest that there exist considerable differences in fertility by source country, even after controlling for individual characteristics. These differences are examined in my estimation of equation (4) shown in table 4.11 for all women and in table 4.12 for all married women and separately for the MARHERE, MARABR, and MARSAME groups. It may be recalled that the dependent variable in this analysis is C,,, the coefficient on the dummy variable for country n from a regression controlling for individual characteristics estimated for the indicated group of immigrants in year t ( t = 1970, 1980); see equation (3) above. Looking first at the findings for all women shown in table 4.11, the first specification presents the results including only a measure of source country 3 1. For the reasons noted above, cluded from these regressions.

AGEMAR, TMSMAR,

and the racekthnicity variables are ex-

Table 4.10

Predicted Fertility of Immigrant Women at Age 45 by Country of Origin: All Women Predicted Fertility 1970

Predicted Fertility 1980

TFR

TFR

1960-65

(1)

(2)

1970-75

(1)

(2)

Argentina Bolivia Brazil Chile China Columbia Costa Rica Cuba Dominican Republic Ecuador Egypt El Salvador Guatemala Haiti Honduras India Iran Israel Jamaica Japan Jordan Korea Lebanon Mexico Nicaragua Pakistan Panama Peru Philippines Syria Trinidad & Tobago Turkey Uruguay Venezuela Vietnam

3.074 6.617 6.148 5.020 5.216 6.721 6.947 4.652 7.498 6.988 5.982 6.844 6.844 6.148 7.357 6.283 7.254 3.832 5.451 2.008 7.172 5.430 6.352 6.742 7.316 7.032 5.738 6.844 6.617 7.459 4.939 6.002 2.928 6.701 5.799

2.376 2.686 2.278 2.474 2.776 2.704 2.875 2.343 3.117 2.706 2.472 2.314 2.267 2.809 2.873 2.548 2.671 2.583 2.521 2.322 4.015 2.618 2.488 4.009 2.894 2.546 2.578 2.719 2.668 3.073 2.546 2.296 1.876 2.352 3.062

2.239 2.735 2.268 2.542 2.556 2.581 2.884 2.080 2.735 2.499 2.374 2.238 2.339 2.791 2.703 2.550 2.420 2.544 2.630 2.273 3.320 2.450 2.461 3.425 2.753 2.607 2.485 2.575 2.722 2.860 2.536 2. I06 1.789 2.297 2.847

2.951 6.494 5.080 3.320 3.859 4.775 4.262 3.463 6.189 6.496 5.123 6.332 6.148 6.064 7.377 5.635 6.537 3.750 5.428 2.070 7.377 4.426 4.916 6.189 6.924 6.755 4.836 5.840 5.490 7.480 3.379 5.307 2.992 5.143 5.717

2.547 2.858 2.498 2.798 2.760 2.694 2.891 2.470 3.323 3.082 2.646 3.109 3.103 3.146 3.017 2.715 2.712 3.110 3.063 2.268 3.512 2.694 3.245 4.113 3.090 3.07 1 3.086 2.679 2.806 2.893 2.978 2.694 2.195 2.751 3.432

2.389 2.897 2.532 2.629 2.656 2.563 2.717 2.182 2.968 2.932 2.588 2.937 2.894 3.156 2.703 2.681 2.709 3.01 1 3.146 2.267 3.179 2.570 2.953 3.429 2.876 2.844 3.087 2.702 2.961 2.592 3.045 2.445 2.169 2.825 3.222

Immigrant average United States

5.690 3.320

2.950 3.017

2.695 2.728

5.528 1.967

3.189 2.915

2.920 2.624

Source Country

Note: Predicted fertility is estimated on the basis of regression equations including country dummy variables. Specification 1 is calculated from regression equations that include controls for AGE and AGE^ and the YSM variables only. Specification 2 is calculated from regression equations that include controls for AGE, AGE^, AGE-H, A G E ~ - H ,EDUCATION, EDUCATION-H, MSP, OTHMAR,SOUTH, NCENT, WEST, SMSA, FOREIGN-H, and the YSM variables. Equations are evaluated for women who are age 45 and whose husbands (if present) are age 48.142. This reflects the average age difference among husbands and wives for immigrants in 1970. The remaining variables are fixed at their mean values for immigrants in each year. (Predicted fertility for the United States is based on native regression coefficients and native means.) The immigrant average for TFR is weighted by immigrant frequencies across source countries in each Census year.

124

Francine D. Blau Regression Results Including Source Country Variables: AIL Women

Table 4.11

(1) Variables

(2) Coeff.

Coeff.

t

TFR

,117

4.71

GNP

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

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

MORT PROPEDUC RT DISTANCE

Adjusted R2

t

.167 ,007 - ,216

4.48 .I5 - ,196

...

... ...

...

...

,893 70

N

(3) Coeff.

t

,159 - ,023 ,007 1.179 - ,236 ,011

4.00 - .51 .05 3.54 - 1.07 1.38

,896 70

,910 70

Note: Regressions include a constant term and a year dummy variable.

Table 4.12

Regression Results Including Source Country Variables: Married Women MSP =

Variables TFR GNP MORT PROPEDUC RT DISTANCE

Adjusted R2 N

Coeff. ,177 - ,028 ,072 1.325 - ,246 .019 ,793 70

1

1

MARHERE =

t

3.36 -.47 .43 3.11 -.82 1.77

Coeff. -.007 -.118 ,096 ,562 -.718 ,011 ,576 69

t

-.lo - 1.33 .49 .81

-1.40 .76

MARABR =

Coeff. ,328 ,0002 -.013 1.274 -.438 ,017 ,354 69

1 t

3.66 .002 -.04 2.18 -.89 .87

1

MARSAME =

Coeff. ,126 -.053 -.110 ,677 -.530 ,009

f

2.47 -.91 -.68 1.79 -1.66 .90

,214 70

Note: Regressions include a constant term and a year dummy variable.

fertility (TFR).Specification 2 presents the results when source country per capita GNP and infant mortality rate (MORT)are added, and specification 3 includes the full set of source country variables.32 In general, source country characteristics are found to be significant determinants of immigrant women’s fertility. As expected, the coefficient on TFR is significantly positive and increases in magnitude when controls for its major source country determinants (GNP and MORT) are added, so it represents a pure taste effect to a greater extent. TFR remains significant and of roughly similar magnitude when the full set of source country variables is included (specification 3). We find that, all else equal, an increase of one child in the source country TFR is associated with an increase of .16-. 17 children for immigrant women in the United States. In specification 2, MORT is found to be signifi32. Interactions of the source country variables with the year dummy were not found to be significant.

125

The Fertility of Immigrant Women

cantly negative, although it becomes insignificant when additional variables are added to the regression owing to problems of multicollinearity and small sample size. At given levels of TFR, an increase in infant mortality in the source country is associated with a smaller number of surviving children. To the extent that this reflects source country differences in desired family size, fertility in the United States, where the infant mortality rate is, in most cases, considerably lower, would be reduced. The findings for the specification incorporating the full set of explanatory variables are quite similar for all women and married women (see tables 4.11 and 4.12). In addition to the positive and highly significant effect of TFR, the most important result is the positive and significant coefficient on PROPEDUC, the inverse measure of educational selectivity: all else equal, the higher the rank of women of a particular nationality in the source country educational distribution, the lower their fertility in the United States. The coefficient on DISTANCE is positive and significant in the regression for married women and larger than its standard error in the regression for all women. This effect may be due to an increased propensity to remain in the United States (i.e., a lower probability of return migration) under the assumption that groups that intend to remain will have more children (i.e., be more likely to “put down roots”). Finally, while the proportion of immigrants who are refugees (RT) is not significant in either regression, it is negative in both and larger than its standard error in the regression for all women. The lower fertility of refugees may reflect the disruptive conditions in the source country at the time of immigration. 33 Finally, I examine the extent to which the influence of source country characteristics depends on the effect of the stage of the life cycle reached at the time of immigration. Table 4.12 presents the results of estimating separate regressions for the MARHERE, MARABR, and MARSAME groups. The results strongly suggest that the effect of source country characteristics is influenced by maturity at the time of immigration. The increasing effect of source country TFR with increasing maturity is particularly notable. The coefficient on TFR is small and insignificant in the MARHERE regression but positive and significant for the MARABR and MARSAME groups and larger for the former (.33) than for the latter (.13). A similar pattern emerges for PROPEDUC. There is one interesting exception to the pattern of larger effects of source country variables for immigrants who were married abroad. Although the coefficient on RT is not significant in any of the regressions, it is considerably larger than its standard error for women who were married here or at about the same time as immigration. This suggests that refugee conditions are most likely to disrupt the fertility of women who are unmarried or recently married at the time of immigration, possibly by inhibiting them from starting their families. 33. The negative coefficients on MORT and RT may also reflect biases in the fertility variable due to the omission of children who have died in infancy or who live in other households.

126

Francine D. Blau

My examination of the effect of source country characteristics on immigrant women’s fertility suggests that fertility differences across nationality groups in the United States are related to conditions in the country of origin. It is particularly notable that higher fertility rates in the source country do appear to increase the fertility of immigrant women in the United States. The results in this section also lend support to the view that both self-selection and assimilation affect immigrant women’s fertility in the United States. Evidence for the former is found in the significant positive effect on fertility of PROPEDUC, my inverse measure of the educational selectivity of immigrants relative to the source country population. At the same time, while the coefficients on source country TFR are positive and significant, they are considerably less than one. (The largest coefficient, obtained for women who were married abroad, is .33.) This suggests an attenuation of the taste effect in response to the U.S. environment. 4.8

Conclusion

Using data from the 1970 and 1980 Censuses, I examined the fertility of immigrant women from the Middle East, Asia, Latin America, and the Caribbean, where fertility rates averaged in excess of 5.5 children per woman during the period of immigration to the United States. Perhaps the most interesting finding of this study is that immigrants from these on average high-fertility source countries were found to have very similar unadjusted fertility to nativeborn women. The number of children ever born was .07 lower for immigrants than natives in 1970 and only .18 higher in 1980. This finding of small unadjusted differentials is important in that, from a policy perspective, it is the unadjusted differentials that are to some extent of particular interest. The small immigrant-native differential appears to reflect self-selection of immigrants as a low-fertility group relative both to source country populations and to native-born women with similar personal characteristics. Evidence in support of the former is the finding that, as a group, immigrant women are positively selected in terms of education, coming on average from the top third of the source country educational distribution. Controlling for years of schooling and other variables, the higher the average ranking of women of a particular nationality in the source country educational distribution, the lower their fertility. Evidence for the latter is the finding that immigrant women have fewer children than native women with similar characteristics (a relatively high-fertility group), .37-.53 fewer in 1970 and .15-.27 fewer in 1980. The finding that immigrants have lower fertility than otherwise similar natives is in turn consistent with a higher demand for child quality among immigrants. Indirect evidence for this possibility is provided by the finding that, for married women, the negative coefficient on husband’s income is larger in absolute value for immigrants than for natives, as is the negative coefficient on wife’s

127

The Fertility of Immigrant Women

wage. The latter further suggests that immigrant women tend to be more responsive to labor market opportunity costs than their native counterparts. Immigrant fertility is also depressed relative to natives in the 1970 cross section by the tendency of immigration to disrupt fertility. In 1970, new arrivals to the United States (those arriving in the past decade) had considerably lower fertility than natives, ceteris paribus. Tracking the relative fertility of synthetic cohorts of immigrants across the 1970 and 1980 Censuses, I found evidence in support of the disruption model: immigrant fertility, especially of the most recent cohort of immigrants in 1970, increased relative to natives over the decade. Despite this increase in relative fertility, reduced-form results indicated that the fertility of these immigrants remained below that of natives with similar personal characteristics in 1980. However, findings from the structural model estimated for married women indicated that, by 1980, the fertility of immigrant women was roughly the same as natives with similar wages and husband’s incomes, all else equal. I take these results as consistent with the notion of selectivity of immigrants as a low-fertility group due to unobserved characteristics associated with labor market outcomes. The pattern of disruption makes it difficult to discern the process by which immigrant women adapt their desired family size to economic conditions in the United States and particularly the speed with which this adaptation occurs, but it does not mean that such adaptation is absent. Some indirect evidence suggesting that immigrant women have fewer children in the United States than they would have had in the source country is suggested by the finding that, while level of source country fertility (TFR) is positively associated with a nationality’s fertility in the United States, all else equal, its coefficient is considerably less than one. This suggests an attenuation of the taste effect in response to the U.S. environment. A consideration of the disruption effect raises the issue of the relative fertility of immigrant women from these source countries in the future. One trend of interest is that recent arrivals had higher adjusted fertility relative to both natives and longer-term immigrants in 1980 than in 1970. This in part represents the effect of declining birth rates in the United States, a trend in which immigrants residing in this country appear to have participated. However, source country fertility rates remained on average fairly constant over this period, and this appears to have resulted in an increase in the fertility of new arrivals relative to the native born. Regardless of the sources of this increase, if the fertility of recent arrivals in 1980 follows the disruption model, as did that of their predecessors in 1970, the relative fertility of immigrant women could have increased considerably over the 1980s. On the other hand, the fact that their initial fertility level was comparable to that of similar natives in 1980 (rather than below the native level, as had been the case in 1970) may exert some downward pressure on the extent of their relative fertility increase, even if it adheres to the disruption model.

128

Francine D. Blau

Appendix Table 4A.1

Frequencies of Immigrants by Country of Origin: 1970 and 1980 1970

Country

All

Argentina Bolivia Brazil Chile China Colombia Costa Rica Cuba Dominican Republic Ecuador Egypt El Salvador Guatemala Haiti Honduras India Iran Israel Jamaica Japan Jordan Korea Lebanon Mexico Nicaragua Pakistan Panama Peru Philippines Syria Trinidad & Tobago Turkey Uruguay Venezuela Vietnam

,017 ,003

N

,008 ,008 ,072 ,029 ,010 .175 .024 .016 .007 .007 .011 .013 ,007 .017 .006 ,012 ,034 .079 ,003 .025 .004 ,274

,008 .002 ,014 .01I ,071 ,003

,008 ,008

1980

Married ,018 ,003 ,009

,008 ,078 ,023 ,008 .177 ,021 .017

,008 ,006 .007 ,009 ,007 ,020 ,007 ,012 ,021 ,089 .003 ,030 ,003 ,289 ,007 ,002 .011 ,010 ,068 ,003 ,006

,008

,003

.004

,005

.005 ,004

,005 8,274

5,688

All ,011 ,003 ,007 .007 ,069 ,027 ,006 ,095 ,030 ,016 ,006 ,018 ,010 ,015 ,007 ,033 ,014 ,009 ,033 ,046 ,002 ,055 ,005 ,297 .009

,004

Married .012 ,002 ,007 .007 ,075 ,023 ,006 ,093 ,022 ,014 .007 ,012

,008 ,011 ,007 ,042 ,012 ,010 .022 ,053 ,002 ,063 ,005 ,313 ,006 ,006 ,010 .009 ,090

,011 .009 ,088 ,003 ,012 ,004 ,002 ,006 ,030

.009

21,232

14,108

,004 ,005

,002 ,004 ,027

129

The Fertility of Immigrant Women

Table 4A.2

Means of Individual Variables for Immigrants: Country Sample 1970

1980

Variables

All

Married

All

Married

FERTILITY

2.055 33.988 12.496 26.964 1 1.350 9.921 7.332 .075 18.570 ,687 .130 ,618 ,045 .220 ,248 ,103 ,376 ,867 ,448 .253 ,152 .090 ,053

2.471 35.168 13.149 39.222 16.51 1 9.782 10.666 ,086 22.764

2.363 34.995 13.054 38.886 16.245 10.594 I 1.558 ,071 22.786

,061

,614 .028 ,242 ,250 ,111 .389 ,856 .651 ,267 ,162 ,095 ,058 ,066

1.952 33.571 12.229 25.839 10.794 10.685 7.680 ,064 18.183 ,664 ,135 ,565 ,057 ,309 ,248 ,099 ,426 ,942 ,493 ,263 ,192 ,122 ,093 ,022

8,274

5,688

2 1,232

14,108

AGE

AGEWOO AGE-H AGE’-H/lOO EDUCATION EDUCATION-H TMSMAR AGEMAR MSP OTHMAR HISPANIC

BLACK OTHERNW SOUTH NCENT WEST SMSA FOREIGN-H

6-10 11-15 YSM 16-20 YSM 21-25/21-30 YSM 25 i 130 + YSM YSM

N

... ,552 ,038 ,339 .263 ,107 ,429 ,933 ,742 ,271 ,196 ,123 .loo ,025

Table 4A.3

Reduced-Form Regression Results Excluding AGEMAR and TMSMAR 1970

1980

Immigrants Variables

COeff

Natives

COeff

t

Immigrants

CXff

t

Natives t

COeff

t

All Women AGE

~ ~ ~ ~ 1 1 0 0 AGE-H

AGE'-HI~~~ EDUCATION

EDUCATION-H MSP OTHMAR HISPANIC BLACK

OTHERNW SOUTH

NCENT WEST SMSA FOREIGN-H

6-10 YSM 11-15 YSM 16-20 YSM 21-25121-30 YSM 25 + 130+ YSM

,165 - ,192 ,152 - ,154 - ,077 - .058

- 1.261 1.473 .I89 ,227 - ,089 .I97 ,372 .394 - .205 .I51 .054 ,296 .458 ,678 ,925

9.80 -8.21 10.95 -9.71 - 13.50 -9.28 -4.30 18.80 2.98 2.23 - 1.29 3.53 5.26 7.63 - 3.55 2.88 1.09 4.90 6.18 7.30 9.87

.283 - ,360 ,156 - .162 - ,103 - ,050 - 1.250 1.429 ,474 ,640 ,029 - ,025 ,175 .087 - ,232 ,093

30.81 - 28.48 18.81 - 16.47 - 22.49 -11.30 - 7.45 31.74 9.53 17.34 .71 - .81 5.72 2.56 - 10.41 1.25

...

...

... .

I

.

... ...

,102 - .061 ,186 - ,198 - .096 - .043 - 2.41 1 1.176 ,199 ,536 - .033 ,177 .321 ,345 - ,390 .367

11.03 -4.79 21.91 - 20.16 -31.18 - 12.70 - 13.91 27.86 4.67 9.48 - .76 5.55 7.85 11.81 - 8.54 11.81

.I22 ,061 - ,202 .01I

.m

.32 -3.62 -2.85 2.66 3.96

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

-.I17 - .lo9 ,116 ,306

,115

- ,081 .I50 - ,162 -.117 - ,035 - 1.462 1.166 .457 ,612 - ,006 - .080

15.96 - 8.16 19.70 - 17.59 -32.13 -9.84 -9.92 36.12 13.19 22.99 - .20 -3.39 4.97 2.35 - 10.44 .I9

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

Constant

Adjusted R2

N

-2.045 ,323 8,838

-7.16

-3.063 ,341 25,539

- 19.64

-.752 ,393 22,786

-4.61

- ,536 .430 30,298

- 4.55

13.91 - 10.02 14.58 - 13.87 -22.78 - 10.28 4.40 5.41 - .34 4.95 7.51 9.90 -7.81 11.68 .04 -4.06 - 2.83 2.41 3.31

,241 - ,222 .066 - .080 - .I25 - ,033 ,576

18.44 - 13.33 6.22 -6.77 -22.58 -7.67 11.80 12.38 - 1.83

17.95

-2.560 .300 18,402

Married Women ,2818

AGE

~ ~ ~ ~ 1 1 0 0 AGE-H

ACE~-H/IOO EDUCATION EDUCATION-H

- ,3418 ,1064 - ,1089 - ,0873 - .0477

6-10 11-15 YSM 16-20 YSM 21-25 YSM 25

,2403 .1258 - ,1270 .2485 .4601 .4809 ,2181 ,2164 ,0967 ,4147 .6058 .7984 1.1155

Constant Adjusted R2 N

-4.5980 ,2376 6,034

HISPANIC BLACK OTHERNW SOUTH NCENT WEST

-

SMSA FOREIGN-H YSMYSM

+

11.15 -10.11 6.51 -6.04 -11.16 -6.42 2.97 .85 - 1.47 3.44 5.19 7.23 - 3.08 3.80 1.51 5.39 6.41 6.96 9.47 -11.15

,4256

28.89 - 27.97 6.16 -6.53 - 18.46 -8.51 8.22 1 I .92 - 1.94 - 1.06 5.22 3.14 -9.71 1.53

- ,5213

,0686 - ,0791 -.1185 - .0432 S164 .6257 - ,1023 - .w .2008 .I351 - .2683 ,1198 .. . . t

.

...

.

... -5.0134 ,2242 17,697

-25.29

.

.203 -.193 .149 -.157 - ,097 - ,042 .237 .43 1 - ,018 ,206 ,387 ,380 - ,432 ,385 ,001 - ,172 -.142 .135 .320 -4.201 ,302 15,021

-

,550

- ,077 - .088 .159 .092 - ,218 .011

-2.71

4.71 2.55 - 8.61 .17

... ... ... -15.11

132

Francine D. Blau

References Abowd, John M., and Richard B. Freeman, eds. 1991. Immigration, Trade, and the Labor Market. Chicago: University of Chicago Press. Becker, Gary S . 1981. A Treatise on the Family. Cambridge, Mass.: Harvard University Press. Ben-Porath, Yoram. 1973. Economic Analysis of Fertility in Israel: Point and Counterpoint. Journal of Political Economy 81 (MarcWApril, pt. 2): S2024233. Blau, Francine D. 1986. Immigration and the U.S. Taxpayer. In Essays on Legal and Illegal Immigration, ed. Susan Pozo. Kalamazoo, Mich.: Upjohn Institute. Bloom, David E., and Mark R. Killingsworth. 1985. Dynamic Analysis of Immigrant Fertility: Preliminary Results. Columbia UniversityiRutgers University, December. Typescript. Borjas, George J. 1987. Self-Selection and the Earnings of Immigrants. American Economic Review 77 (September): 53 1-53. . 1990. Self-Selection and the Earnings of Immigrants: Reply. American Economic Review 80 (March): 305-8. . 1991. Immigration and Self-Selection. In Abowd and Freeman (1991). Butz, William P., and Michael P. Ward. 1979. The Emergence of Countercyclical U.S. Fertility. American Economic Review 69 (June): 3 18-28. Chiswick, Barry R. 1978. The Effect of Americanization on the Earnings of Foreignborn Men. Journal of Political Economy 86 (October): 897-921. . 1988. Differences in Education and Earnings across Racial and Ethnic Groups: Tastes, Discrimination and Investments in Child Quality. Quarterly Journal of Economics 103 (August): 571-97. Ford, Kathleen. 1990. Duration of Residence in the United States and the Fertility of U.S. Immigrants. International Migration Review 24 (Spring): 34-68. Gonvaney, Naintara, Maurice D. Van Arsdol, Jr., David Heer, and Leo A. Schuerman. 1989. Assimilation, Disruption, or Selectivity? A Comparison of Alternative Hypotheses Regarding the Fertility of Immigrants in the United States: 1970-1980. Paper presented at the annual meeting of the Population Association of America, Baltimore, 27 March. Heckman, James J. 1980. Sample Selection Bias as a Specification Error. In Smith (1980). Heckman, James, and Richard Robb. 1985. Using Longitudinal Data to Estimate Age, Period and Cohort Effects in Earnings Equations. In Analyzing Longitudinal Data for Age, Period and Cohort Effects, ed. S. Feinberg and W. Mason. New York: Academic. Jasso, Guillermina, and Mark R. Rosenzweig. 1990. Self-Selection and the Earnings of Immigrants: Comment. American Economic Review 80 (March): 298-304. Kahn, Joan R. 1988. Immigrant Selectivity and Fertility Adaptation in the United States. Social Forces 67 (September): 108-27. Manski, Charles F. 1989. Anatomy of a Selection Problem. Journal of Human Resources 24 (Summer): 343-60. Mincer, Jacob. 1978. Family Migration Decisions. Journal of Political Economy 86 (October): 749-73. Moulton, Brent R. 1986. Random Group Effects and the Precision of Regression Estimates. Journal of Econometrics 32 (August): 385-97. Schultz, T. Paul. 1981. Economics of Population. Reading, Mass.: Addison-Wesley. . 1984. The Schooling and Health of Children of U.S. Immigrants and Natives. Research in Population Economics 5:25 1-88.

133

The Fertility of Immigrant Women

Smith, James P., ed. 1980. Female Labor Supply: Theory and Estimation. Princeton, N.J.: Princeton University Press. United Nations. 1983. Manual X : Indirect Techniques for Demographic Estimation. New York. U.S. Bureau of the Census. 1980. Statistical Abstract of the United States. Washington, D.C. U.S. Immigration and Naturalization Service. 1986. Statistical Yearbook of the Immigration and Naturalization Service. Washington, D.C.

This Page Intentionally Left Blank

5

Mass Emigration, Remittances, and Economic Adjustment: The Case of El Salvador in the 1980s Edward Funkhouser

Among the most important labor market issues to confront the small developing economies of Central America and the Caribbean in the 1990s is the increasing volume of emigration. The continuing decline in real wages relative to those in the United States and increasing political tensions are likely to create further pressures to migrate. This outflow of workers will have large effects on the supply of labor, especially of educated workers, in the sender countries. In addition, for many countries, remittances will be a major source of foreign exchange and household income. In retrospect, one of the striking disparities in these countries was the absence of policies concerning international migration during the 1970s and the critical importance that migration has subsequently assumed. In this paper, I examine the effect of migration and remittances on one of these countries, El Salvador. The magnitude of migration and remittance flows-and the effect on national income, labor markets, and foreign exEdward Funkhouser is assistant professor in the Department of Economics at the University of California, Santa Barbara. The author would like to acknowledge financial support from the National Bureau of Economic Research and the Ford Foundation as well as the helpful comments of Richard Freeman, Larry Katz, Anne Case, George Borjas, Alan Krueger, Fernando Ramos, and Steve Trejo. In addition, this paper would not have been possible without the assistance of the following persons in El Salvador. Mauricio Alens and Joel Santiago at the Household Survey Division of the Ministry of Planning generously provided the 1985 and 1988 household surveys and suggestions. Edgar Leonel Saballos and Miguel Angel Figueroa of the Central Bank assisted in the preparation of remittance data. Raymundo Alvarado of the Population Desk of the Ministry of Planning supplied population projections. Victor Ramirez and Juan Castellano, also of the Ministry of Planning, provided general economic data and information about current planning. Wage data of the social security system was provided by Lic. Soto Mejiva of the Instituto Salvadoreno de Seguridad Social. Eduardo Federico Calderon of the Ministry of Labor assisted in the collection of published material from earlier household surveys. Leda Arguedas and Manuel Rincon of CELADE in San JosC provided useful documents and discussion. Finally, the author has benefited greatly from data provided by, discussions with, and the example of Segundo Montes.

135

136

Edward Funkhouser

change markets-is large; 10-15 percent of the population has emigrated, and the money these people remit amounted to 9 percent of GDP and 67 percent of exports in 1988. I look at the effects of these flows on brain drain, labor force participation, and wages in the Salvadoran labor market using six data sources on emigrants from El Salvador and two data sources for the native Salvadoran population.’ I find that the massive emigration from El Salvador had significant effects on the labor force participation of remaining household members: this occurs primarily because of the income effect from receiving remittances. I also find evidence that wages were higher in the areas with the greatest number of international migrants. The initial out-migration, the labor force participation response, and the wage response to migration suggest that unemployment rates were lower with the large emigration than they would have been otherwise. Although the motives for migration from other small sender countries may differ from those in El Salvador, the effects of emigration are likely to be similar to those in El Salvador as emigration and remittances increase.

5.1 The Migration and Remittance Experience in El Salvador in the 1980s From 1979 until 1981, the levels of out-migration, capital flight from El Salvador,2 and the inflow of remittances to El Salvador from emigrants increased significantly. Although capital flight resumed its previous level in 1981, migration and remittances have remained high throughout the 1980s. Estimates of the volume of migration from El Salvador, summarized in appendix A , range from 194,000 Salvadorans outside the country (MIPLAN 1986) to over one million (Montes 1987). The official data, presented in column 1 of table 5.1, indicate that, between 1978 and 1987, there were 653,200 more exits than entrances by citizens of El Sal~ador.~ The peak in net migration in 1982 was followed by a slight decline until 1986. By 1988, emigration began to increase again. 1. The data sources on emigrants are the 1985 Multiple-Purpose Household Survey (MIPLAN 1986). the 1987 University of Central America (UCA) Survey (Montes 1987) in El Salvador and the United States (see n. 1I below), Salvadorans in the 1980 U.S. Census, Salvadorans in the April 1983 Current Population Survey, and Salvadorans in the June 1988 Current Population Survey. The data sources on natives are the 1985 Multiple-Purpose Household Survey and the 1988 Multiple-Purpose Urban Household Survey (MIPLAN 1989b). These sources are discussed in more detail in app. B. 2. Capital flight, measured by the sum of the lines in the balance of payments accounts for “Short-Term Capital: Other Sectors” and “Net Errors and Omissions,” registered as particularly large outflows of $216.7 million in 1979 and $406.7 million in 1980. 3 . This number may undercount the number of emigrants who left for Honduras through areas affected by the civil war. But since there is no legal restriction on Salvadoran exit from El Salvador or Salvadoran entrance to Guatemala, there is not likely to be a large volume of illegal emigration that is not captured by the official statistics. The period of exception is 1979-80. The official data show that this was a year of extremely low migration, suggesting an undercount of forty to sixty thousand net emigrants.

Table 5.1

Annual Levels of Emigration and Remittance Flows for El Salvador, 1978-87: Official Data and Estimates by Author Remittances

Migration

Author’s Estimates

Estimates

1978 I979 1980 1981 1982 1983 1984 1985 1986 1987

Official Registered Total (1)

Medium Variant (2)

68.4 68.8 83.2 40.5 129.4 73.0 55.6 51.5 52.6 30.2

35.6 43.0 46.6 66.9 37.8 28.8 26.6 27.2 15.6

High Variant (3)

Central Bank Balance of Payments (4)

IMF Balance of Payments (5)

60.5 73.5 79.2 113.9 64.2 48.9 45.3 46.3 26.6

45.3 49.2 10.9 7.0 20.5 11.2 19.6 11.7 9.2 14.5

44.9 44.9 17.4 39.2 51.7 97.4 118.0 129.4 149.6

Central Bank Estimate (6)

(7)

(8)

(9)

49.2 59.6 74.7 87.3 97.0 121.4 101.9 134.5 168.7

132.7 176.4 240.9 289.1 333.9 379.9 429.5 470.2

94.4 143.5 217.2 266.4 309.3 352.4 398.6 433.1

148.3 215.8 318.5 393.8 463 .O 534.4 611.9 674. I

Sources: Column 1, MIPLAN (1988); and published data. Columns 2 and 3, app. A: medium variant = 450,000 emigrants in 1987; high variant = 650,000 emigrants in 1987. Columns 4 and 6, Central Bank of El Salvador, unpublished data. Column 5 , IMF, lnrernarional Financial Srarisrics (various issues), private unrequited transfers. Columns 7-9, app. A: low variant = remittances from emigrants in the United States ($600 per year in 1978; $900 per year in 1987) and from other emigrants ($0 per year); high variant = remittances from emigrants in the United States ($600 per year in 1978; $1,200 per year in 1987) and from other emigrants ($250 per year in 1978; $400 per year in 1987). Column 7, medium variant for migration, high variant for remittances. Column 8, high variant for migration, low variant for remittances. Column 9, high variant for migration, high variant for remittances.

138

Edward Funkhouser

I show two additional estimates of the net emigration flow in columns 2 and 3. These estimates are calculated by interpolation between a base stock of emigrants of 164,000 in 197fii4 and medium (450,000) and high (650,000) estimates of the number of Salvadorans outside the country in 1987. The shape of the migration flow for these estimates is that of the official statistics adjusted for the low 1979-80 outflow. I estimate the total stock of emigrants in 1988 to be between 500,000 and 750,000, or 10-15 percent of the 1988 population of 5.3 million persons. Of these, it is estimated that 85 percent came to the United States. There was also a sizable shock to the economy from the inflow of remittances. From column 4 of table 5.1, it can be seen that, following 1979, private exchangers ceased the use of official channels to change dollars. The level of remittances must, therefore, also be estimated. Existing estimates range to over $1.3 billion per year (Montes 1987). Two official estimates of remittances are shown in columns 5 and 6-the line item for “Private Unrequited Transfers” in the balance of payment account reported to the International Monetary Fund and the Central Bank’s own estimate. The data show an increase in remittances from less than $50 million in 1979 to over $194 million in 1988 and a total of over $1 billion in remittances during the 1980s. These estimates are based on a constant level of remittances per emigrant over time. Remittance per emigrant has not remained constant, however. Calculations from the 1987 University of Central America (UCA) survey (see app. A) suggest that average remittance per emigrant, including nonworkers, has increased to approximately $100 per month. To calculate the volume of remittances shown in columns 7-9 of table 5.1, therefore, I used an interpolation of remittance per emigrant that is then multiplied by an estimate of the stock of emigrants at each point in time. The three estimates represent different variants of high and medium estimates of remittances and emigration as described in the notes to table 5.1. The estimates range from $400 to $600 million per year, or an increase from double the Central Bank estimates in 1980 to over three times the Central Bank estimate in 1986. The lower amount is equal to 67 percent of exports, 99 percent of the trade deficit, and 8.6 percent of GDP in 1987.5 Over this period, remittances became an important source of foreign exchange. While these estimates of migration and remittances are crude, their orders of magnitude suffice to document the key fact that motivates this study: the large outflow of persons from El Salvador and the correspondingly large inflow of remittance funds. 4 . This figure is the official estimate of the Central Bank. 5. In addition, for all years except 1984, estimated remittances exceed U.S. economic aid to El Salvador (see Gallardo and Lopez 1986; and MIPLAN 1989a).

139

Mass Emigration, Remittances, and Economic Adjustment

5.1.1 Characteristics of Emigrants Emigrants are not a random draw from the Salvadoran population. In table 5.2, I compare the characteristics of the emigrant population with the characteristics of the Salvadoran population using all available data sources. Emigrants are much more likely to come from an urban area, are more likely to be of working age, and are more educated than the nonmigrant population. There are two ways to analyze further the characteristics of emigrants relative to the native population: in terms of their personal attributes and in terms of the attributes of their families. In column 1 of table 5.3, I estimate probit equations that relate emigrant status to individual characteristics using data obtained in the 1985 Salvadoran Household Survey, which asked limited questions about the characteristics of emigrants. In column 2, I estimate similar probits for a pooled data set consisting of natives from the 1985 Household Survey and migrants from the 1987 UCA Survey conducted in El Salvador, which contains information about the education and marital status of emigrants not found in the 1985 Household Survey. The results in the two columns confirm the simple tabulations in table 5.2: persons who are urbanized, more highly educated, married, male, and young are more likely to be migrants. All these characteristics suggest that migrants had greater human capital and labor force participation than nonmigrants. In addition, migrants are likely to be a generation younger than the household head and to come from larger households. In column 3 of table 5.3, I show how households with emigrants differ from households without members abroad. I include the characteristics of the household head (age, education, marital status, sex), household income,6 household size, urban status, and department.’ As in the individual data, urban households are more likely to have emigrant members. The coefficients on the characteristics of the household head suggest that adult children are those who migrate-households in which the head is married and over the age of 50 are those most likely to have had members emigrate. Female-headed households are also more likely to have members abroad. Unfortunately, the data do not specify relationship of the emigrant to the household head. Therefore, I cannot test directly whether migration has contributed to family split up. Two robust findings in the household estimates are that higher-income 6 . Remittances are not identified separately in the 1985 data. It is not likely that respondents included remittances in the “Other Income” category in the 1985 survey. The 1988 Urban Household Survey (MIPLAN 1989b) does ask about assistance from people outside the country. Only a small proportion of persons report such assistance, and for those persons the predicted remittance from the UCA Survey (419 colonies per month) exceeds total income (255 colonies per month). 7 . In El Salvador, there are fourteen regional divisions called departments. The eight departments located in the eastern half of the country are those that have been most affected by the civil war.

Table 5.2

Characteristics of Salvadoran Population and Emigrants (%) Total Population

Emigrants UCA Surveys

Salv. Household Surveys Salv. Census,

U.S. Census,

April CPS,

Salv. Household Survey,

Salv.,

US.,

June CPS,

1983

1985

1987

1987

1988

3.2 94.2 4.7

.9 92.7 6.4

9.7 84.7 5.6

38.3

1.1 94.2 2.6 61.1 38.9

58.7 41.3

58.9 41.1

51.8 48.2

51.4 48.6

36.8 21.1 36.9 5.1

35.8 25.8 29.4 9.0

35.7 14.3 37.5 12.5

64.7 60.4

87.7 65.5

8.9 36.3 22.6 32.2

4.6 55.3

1971

1975

1978

1980

1985

1980

46.4 46.3 7.3 39.5 60.5

45 .O 46.4 8.6 40.3 59.7

44.9 46.1 9.0 41.7 58.3

43.7 46.1 10.2 41.9 58.1

39.7 49.2 11.1 48.4 51.6

11.8 81.5 6.7

48.6 51.4 10: 90.9 5.0 2.7 1.3

47.0 53.0

47.3 52.7

47.0 53.0

45.5 54.5

43.6 56.4

44.1 55.9

83.1 8.8 5.8 2.3

80.7 10.4 7.0 2.0

75.1 11.9 10.1 3.0

32.3 20.2 27.5 19.7

41.2 23.5 23.5 11.8

69.4 29.2

70.1 34.4

66.6 34.7

79.2 59.8

86.7 65.8

5.1 85.7

4.9 86.3

6.6 81.4

6.3 63.2

8.8

10.4

30.5

Age: 0-14 15-54 Over 54

Urban Rural Pop. 15-54 Male Female Education level of pop. over 0-6 years 7-9 years 10-12 years 13 + years Labor force participation: Male Female Occupation: Professional Workers/employees Domestic Service

Source: See app. B.

68.4 25.3

[ 9.1

95.6 4.4 61.7

5.7 88.0 6.3)

1.3 57.1 14.1 27.6

[ 40.11

Mass Emigration, Remittances, and Economic Adjustment

141 Table 5.3

Probit Estimates for Migrant Status: Emigrant Data Pooled with Nonmigrant Data from National Household Survey 1985 Salvadoran Household Data Pooled with: 1985 Migrants Probit (1)

-2.324 (.074) ,299 (.027)

Intercept Urban

Change (Ib)

,032

Education: 0-6 years

7-9 years 1@- I2 years

,060

Age

,003

(.004) - ,101

Agez/100 X

male

Married

X

female

Probit (2)

-3.407 (.112) .I49 (.030)

- .379 (.061) ,326 (.064) ,540 (.061) ,105

Change (2b)

.015

,080

,039 ,001

,131

Female Household incomeil ,OOO

.483 -.274 (.024)

-.030

- ,252 (.033)

Persons over 10 Department dummies

Log likelihood N

Yes 6,231.0 30,530

Yes 5,856.2 30,534

1985 ~ousehold Data, Migrant Member, Probit (3) -3.214 (.207) ,257 (.042) ,308 (.Ill) ,505 (.125) ,166 (.121) .041

(.007) - .028 (.007)

-.I39 (.007) .388 (.038)

(.006) Married

1987 Montes Data

,026

,084 (.043) .349 (.OM) . I24 (.026) .03 I (.031) Yes 2,954.2 8,095

Note: In cols. Ib and 2b, “Change” refers to the change in probability resulting from a one-unit increase in the independent variable. For the education variables, the change is movement from the next lowest education group. In col. 3, characteristics are those of household head. Observations weighted by expansion factors. Numbers in parentheses are standard errors.

households and larger households-even after exclusion of the emigrants themselves-are more likely to have household members outside the country. Higher-income households can more easily afford the transportation costs of migration. Larger households can more easily afford the psychic costs or more easily insure themselves against loss when there are credit constraints.

142

Edward Funkhouser

5.1.2 Brain Drain

Of special concern is the possibility that the number of more educated migrants exceeds the replacement capacity of the educational system. I use published data from the 1980 Household Survey and official education statistics to calculate whether natural increase has replaced emigrants in the Salvadoran labor market. Between 1980 and 1986, the natural increase in the potential labor force was slightly over 100,000.8This is double the high estimate of the number of emigrants who would have been in the labor force had they not migrated. I find that, even for the more educated population, the educational system has supplied at least as may graduates at each educational level as the number of people at that level who have left the country. For those with ten or more years of education, there was an increase of thirty-two thousand people per year between 1980 and 1986, or approximately seven thousand more than the estimated number of migrants from this education group.9 This evidence suggests that increases in nonmigrant graduates from the educational system at higher education levels have compensated for the effects of migration on the stocks of persons in each education group. The extent of shortages in skilled labor has been lessened because there has not been growth in aggregate demand for labor-replacement of the educated work force has been sufficient to avoid shortages in most areas. The net effect is that the pool of persons in the work force of a given education level becomes increasingly younger and commands less labor market experience, as emigrants are replaced by recent graduates. 5.1.3 The 1980s Migration An important question is the extent to which the changes following 1979 can be attributed to political changes and the subsequent violence or to economic decline. The more the migration is due to economic forces, whether politically induced or not, the more likely the results of this study can be generalized to other countries. In table 5.4, I compare the characteristics of Salvadorans who migrated before and after 1979, in four data sets. Despite the lack of comparability due to sample selection, the data show a higher proportion of men, a higher proportion of single persons, and lower levels of education among those who migrated after 1979 than among those who migrated before 1979.IoAlthough the lower education level of post-1979 mi8. The increase in the potential labor force is calculated by subtracting the age 60-64 cohort from the age 10-14 cohort and averaging over the five-year bracket to account for the six-year period. 9. Between 1978 and 1986, there were an average of 43,200 Salvadorans per year completing the ninth grade, 21,660 completing the twelfth grade, and 10,OOO in each university year (MIPLAN 1986, table K1, pp. 147-48). 10. It should be noted that those counted as having left before 1979 are, in fact, those who left before 1979 and did not return. Thus, there is not equal selection in both cohorts.

Table 5.4

Characteristicsof Migrants before and after 1979: Evidence from the April 1983 CPS, the June 1988 CPS, and the 1987 UCA Survey CPS

UCA

April 1983

Mean age at migration Sex (%): Male Female Married (%) Education (%):

c-6 7-9 10-12 Over 12 Working in U.S. (%): Male Female (continued )

June 1988

Salv., 1987

U.S., 1987

Before 1979

After 1978

Before 1980

After 1979

Before 1979

After 1978

Before 1979

After 1978

24.8

29.3

20.4

25.6

24.9

24.8

25.7

26.3

44.4 55.6 47.2

41.4 58.6 65.5

40.0 60.0 56.7

47.7 52.3 47.7

45.9 54. I 72.5

62.1 38.9 40.2

38.8 62.2 54.4

56.4 43.6 44.3

38.9 19.4 22.2 19.4

48.3 31.0 20.7

13.3 6.7 43.3 36.7

35.4 21.5 41.5 1.5

34.2 16.7 44.0 5.0

35.9 23.9 35.6 4.6

40.2 22.9 31.8 5.1

33.2 27.1 29.6 10.2

75.0 66.7

80.6 55.9

90.7 77.1

71.1 65.1

84.9 70.8

62.0 58.2

...

Table 5.4

(continued) CPS June 1988

April 1983

Mean U.S. earnings ($1980): Week Year Married emigrant left spouse (%): Male Female Legal status (%): Legal Undocumented In legalization process Migration for (%): Economic reason Political reason or both Plan to return to El Salvador (%) Plan to bring family to U.S. (%)

UCA Salv., 1987

U.S., 1987

Before

After

Before

After

Before

After

Before

After

1979

1978

1980

1979

1979

1978

1979

1978

8,548

7,110

306

209

246

191

17.1 15.8

40.1 21.0

12.7 12.5

31.6 22.9

75.1 11.8 13.2

24.4 56.0 19.6

41.6 36.8 15.6

14.1 62.9 23.0

70.5 3.1 9.4 45.4

15.9 12.2 32.1 52.7

58.3 17.4 40.3 63.4

31.5 47.3 53.4 53.2

145

Mass Emigration, Remittances, and Economic Adjustment

grants is consistent with political emigration from a more conservative government, the data are more suggestive of economic motivations to migration affecting a greater cross section of the population. Further evidence is provided from the UCA surveys,” which asked motive for migration, in the bottom half of table 5.4. Although the responses do show a higher proportion reporting political motivation among those emigrating after 1979, even during the early 1980s, the years of greatest conflict, most emigration was economically motivated. Not surprisingly, however, a higher proportion of those surveyed in the United States, where reprisal is less likely, stated political motivations for migration. One of the principal determinants of “economic” migration-an increase in the wages in the destination country relative to those in the source countrydid provide greater incentive for emigration from Central America in the 1980s. In figure 5.1, I present data on earnings in El Salvador and the United States, calculated from social insurance data for El Salvador (ISSS 1978, 1987) and Employment and Earnings for the United States. The ratio of mean weekly earnings in the United States to those in El Salvador increased from 8.3 in 1980 to over 14 in 1986. However, the decline in relative earnings alone does not explain why migration levels from El Salvador are higher than those from other countries with similar declines in relative real wages. To examine the contribution of earnings incentives to the composition of the emigrant pool, I estimated log earnings equations for Salvadorans in the United States and those in El Salvador using the 1980 Census of the United States and the 1985 Encuesta de Hogares de Propositos Multiples (MIPLAN 1986) of El Salvador: lnw,;

=

+ p,X, + E,,,

a/

in which w,, is the earnings in location 1 of person i , a,and p, determine the reward structure in location 1, X , is a vector of characteristics of individual i, and E,; is the random component of the wages of individual i in location 1. The results of these estimations are shown in table 5 . 5 . In both countries, there is an inverted U-shaped age-wage profile, a positive relation between 11. Segundo Montes, then director of the Instituto de Investigationes of the University of Central America, conducted parallel surveys with similar questions about emigration patterns of households in El Salvador and of emigrants in the United States. The survey conducted in El Salvador asked questions about relatives who had emigrated. The survey conducted in the United States asked questions of the emigrants themselves. (See Montes 1987; and app. A below.) 12. Although real wages have fallen while GDP per capita has increased slightly over the 1980s, data on GDP per capita are more easily available and more comparable across countries. Figures for the increase in real GDP per capita for the period 1983-87 and the officially registered net emigration (as a percentage of the 1983 population) for the period 1983-87, respectively, are as follows: El Salvador, 1.65 and 3.62 percent; Honduras, 0.64 and 1.47 percent; Nicaragua, - 16.55 and 3.18 percent; and Costa Rica, 6.24 and 0.87 percent. The increase in real GDP per capita for the period 1983-87 for Mexico and the United States, respectively, is - 3.35 and 12.93 percent. (Sources: International Monetary Fund, International Financial Statistics; MIPLAN (1988); INEC (1989); and unpublished data.)

146

Edward Funkhouser

Weekly earnings ($1

Ratio U.S./E.S.

72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87

E.S.

YEAR

U.S.-Total ......... .Q .......I

- Ratio-Tot. - - )c- - -

U.S.- Retail Ratio-Ret. -+-+--+ -.-.-. 8.-.-.-

Fig. 5.1 Real wages in El Salvador, the United States, and the retail industry in the United States and the ratio of real wages in El Salvador to real wages in the United States

education and earnings, and a positive marriage effect. The educationearnings profile is flatter in the United States, implying that, in percentage terms, more-educated workers do not do as well as less-educated workers after migration. However, because of the large difference in mean earnings noted above, the difference in earnings in absolute terms provides a greater incentive for more-educated workers to come to the United States. To see this, consider two married males, aged 30-39, considering moving from San Salvador to a western urban area in the United States. One has ten to twelve years of education and the other zero to six years of education. In El Salvador, the more-educated worker will earn 71.2 percent more than the less-educated worker, while in the United States he will earn only 16.3 percent more. However, because mean wages in El Salvador are less than one-tenth those in the United States, the absolute gain favors the educated worker, who will earn $593 more by moving to the United States, over the less-educated worker, who will earn only $531 more.I3 An additional component of the net benefit to migration is the migration cost. The effect of this fixed cost is to make migration unprofitable for those who gain the least in absolute terms. The cost of illegally entering the United 13. Expected wages are likely to show even greater relative benefits to educated workers. In the 1985 data for El Salvador, unemployment rates for educated workers are higher than those for uneducated workers. In the 1980 Census data, unemployment rates for educated Salvadoran immigrants are lower than those for uneducated Salvadorans.

Table 5.5

Estimates of Log Earnings Equations for Salvadorans, El Salvador and the United States Females Aged 16-65

Males Aged 16-65 Wages in El Salvador Urban

UCA

Total,

Intercept Education: 0-6

7-9 10-12

Wages in El Salvador

Wages in the U.S.

1985 (1)

1985 (2)

1988 (3)

6.358 i.058)

6.718

6.884

(.OM) (.056)

Wages in the U.S.

Urban

us. Census

U.S.

Salv.

(4)

(5)

(6)

5.025 (.323)

5.117 ( ,267)

5.215 (.287)

UCA

Total,

- ,535 - ,898 - ,868 - ,736 - .221 -.I58 (.033) (.068) (.136) (.086) (.042) - ,374 - ,080 - ,612 - ,632 - ,481 -.113 (.047) (.050) (.038) (.073) (.143) (.089) - .393 - ,361 - ,334 - ,306 - ,073 - ,063 (.046) (.047) (.039) (.063) (.137) (.085)

(.ow

1985 (1)

1985 (2)

1988 (3)

6.465

6.572

(.080)

(.081)

6.723 (.083)

U.S. Census

U.S. (5)

Salv.

(4) 4.672 ( .306)

4.947 (.407)

4.446 ( ,300)

(6)

1.202 - 1.203 - ,970 - ,227 - ,368 - .649 (.073) ( .202) (.087) (.054) (.043) (.052) - ,530 - ,535 - .893 - .857 - ,688 -.118 (.061) (.065) (.054) (.077) (339) (.OW) -.120 - ,485 - .388 - .339 -.I17 - .394 (.058) (.049) (.072) (.200) (.087) i.056)

-

Age:

lo)

No. children Migration variables: Migrant in household Predicted remittance/100 Proportion migrants in dept. Constant

,077 (.046) - ,045 (.006) ,006 (.002) - ,027 (.010) ,055 (.014)

- ,205 (.056)

,0048 - ,016 (.001)

- ,016 (.001) -.373 (.033)

,232 (.046)

,0495

- ,374 (.033)

,108

,0241

,462 (.034) - ,019 (.004)

,435 (.057) - ,019

,020

,020

(.077) - ,045

(.ow

- .0003

,006

-.1278

,1553

- .0002

- ,027 (.010) ,054 (.014)

- ,0062 ,0118

- .022 (.011)

-.022 (.011)

,012 (.117) - .029 (.014) - ,005

,0027

- ,033 (.043)

,016

,016

.0056

(.OW

- ,731 ( ,256)

- 2.674 (.182)

,149 (.087) -.024 (.010) ,005 (.008) -2.721 (.194)

Department dummies Yes Yes -- 2,806.4 - 2,804.0 Log likelihood N 7,033 7,033

Yes - 5,223.9 8,656

Yes - 5,220.9 8,656

- ,769 (.241)

Increase in Probit from 1-unit change (2b)

- .oO01 - ,0010 (.010)

-.0075

.0523 -.OO01 ,0019

Note: Remittances are translated to colones at average black market rate for 1985, which is 5.66 colones per $US 1.00. The omitted groups are rural, age 50-59, and college education. In col. 2b, change is a one-unit change for education, age, marriage, urban, adults in household, children in household, migrant in household, and proportion of households in region with migrant member. For income and remittances, change is a 1 percent change. Numbers in parentheses are standard errors.

153

Mass Emigration, Remittances, and Economic Adjustment

vey reporting a member outside the country. This leads to use of the following remittance variable: (3) where a and (3 are the coefficients from the estimation of remittances using the 1987 UCA data, X is a vector of household characteristics in the 1985 data, and CPI is the 1987 consumer price index, base 1985. The regression from which this variable is calculated is column 1 of table 5.6. A shortcoming of this method is that households that do not have a member abroad but that receive remittances from another relative have zero predicted remittances since they cannot be identified in the 1985 data.I7 In column 2 of table 5.7, I include the three migration variables-amount of predicted remittances, emigrant from household, and emigrants as a proportion of population in the department-urban cell-as determinants of labor force participation. The strongest effect of emigration is through remittances. For both men and women, there is a significant negative effect of the level of remittances on participation that is approximately the same in magnitude as the domestic nonwage income effect. The remittance effect is the only significant effect of emigration on the labor force participation of men. For women, however, the remittance effect is partially offset by a positive within-household substitution effect. There is also a weak positive coefficient on the proportion of migrants from the local labor market variable that suggests some movement of women into the labor force to fill the jobs of the men who left. The total effect of migration on labor force participation can be calculated from the three separate effects. For men, the negative income effect from remittances dominates all other effects. For females, the positive regional labor market effect, although insignificant, is large enough to outweigh the negative within-household effect (remittance and composition effects) on participation. I can now examine how much of an effect migration had on observed changes in labor force participation in the 1980s. To make this calculation, I use the counterfactual assumption that migration would have stayed at the same proportion of the population as it had been in the 1970s-5 percent instead of 13 percent. Between 1978 and 1985, the labor force participation 17. Unfortunately, all households without an emigrant member in the 1985 survey are candidates to have a nonmember send remittances, and the only overlapping information for these households in both surveys is the department-urban cell of residence. Predicting on the basis of area alone mixes the direct and indirect effects of remittances for these households since all households in a particular area identified not to have an emigrant member will be assigned the same value of remittances. Estimating the direct effects of remittances in this manner is likely to underestimate the true effect since the comparison group includes some households for which there actually were remittances. It is also likely to increase the apparent direct substitution effect since for these households the negative effect of remittances will show up in the variable measuring presence of migrants.

154

Edward Funkhouser

of men aged 20-59 dropped from 92.9 to 89.7 percent and for women aged 20-59 increased from 39.9 to 47.7 percent.l* Under the counterfactual assumption, labor force participation for men of this age group would have fallen 2.1 points instead of 3.2. For women, labor force participation would have risen only 4.3 points instead of the observed 7.8 points.

5.3 Macroeconomic Effects of Migration The macroeconomic implications of the discussion in sections 5.1 and 5.2 are summarized in figure 5.2a, based on the presentation of Bhagwati and Brecher (1981) and Djajic (1986). N T is the production-possibilities frontier between traded and nontraded goods before migration. With balanced commodity trade, production and consumption take place at point C = X. With migration, the production-possibilities frontier moves inward to NT’ . There is a proportionately greater shift inward in the good that is more labor intensive in production. With no remittances, production following migration takes place at point X’, which is above the premigration nontradedkraded ratio if the traded-goods sector is labor intensive and below this line if the nontraded-goods sector is labor intensive. When emigrants remit, income increases. The budget constraint moves outward by the value of the remittances in traded goods to CP.At the existing price ratio, desired consumption is the point Cf. But since remittances cannot purchase nontraded goods from abroad, consumption initially can move only to point C’. Excess demand for nontraded goods results, the price of nontraded goods increases, final production moves to point Xf, and final consumption is pbint Cf. At the optimum, the slope at the points of production Xf and Cf are equal. The locus of points traced out by the maximum consumption point for each level of migration, CC, is depicted in figure 5.2b. The maximum consumption possibility occurs at point C*, where the indifference curve is tangent to the CC locus. This corresponds to inward movement of the postmigration production-possibilities frontier until the postmigration consumptionpossibilities frontier is at its maximum.19 The magnitude and shape of the inward shift of the production-possibilities frontier is the aggregation of the individual responses.2oAbove, I found the characteristics of emigrants to be disproportionately of higher productivity 18. To be comparable with the earlier published data, the probit regression of col. 2 of table 5.3 was reestimated with self-reported participation in the labor force as the dependent variable. 19. The actual shape of the CC curve is not determined. When the marginal gain from remittances is greater than the marginal gain in production, the curve moves to the right. When the opposite holds, the curve bends back toward the origin. 20. This suggestion is smooth if the distribution of household characteristics is sufficiently diverse.

155

Mass Emigration, Remittances, and Economic Adjustment

A. Non-

Traded Goods

Tradeable Goods B. Non-

Traded Goods

Tradeable Goods C. Non-

Traded Goods

N

T

---~

R

Trodeable Goods

Fig. 5.2 Shift in the production-possibilitiesfrontier and the consumptionpossibilities frontier with emigration and remittances

156

Edward Funkhouser

than those of nonmigrants and that labor supply response to emigration is negative for males and positive for females. These considerations lead to the following modification of figure 5.2a, shown in figure 5 . 2 ~ .The initial movement in the production-possibilities frontier in figure 5 . 2 depends ~ on the number of migrants, shown as NT’. For the case of El Salvador, since the labor market characteristics of those migrating are above average, the frontier will be placed closer to the origin that it would have been had the migrants been of average labor market quality, to NP. Since the net effect on participation of remaining members is positive, the production-possibilities frontier after migration is located slightly away from the origin, to N T 3 .

5.3.1 Wages Large-scale migration can also affect wages. To estimate the effect of Salvadoran emigration on wages, I utilize two approaches. First, I add two migration variables-proportion of households in department-urban area with members abroad and member of own household abroad-to the wage equations using the 1985 Salvadoran Household Survey. I present the regressions in table 5.8 for both monthly wages and calculated

Table 5.8

Coefficientson Migration Variables in Wage Regressions for Persons in El Salvador Males Hourly Wage (1)

Proportion migrants X urban Proportion migrants X rural Migrant in household Years education Education’ Years experience Experience2/100

R’ N

Females

Monthly Wage (2)

Hourly Wage (3)

.012

,010

- .001

(.007)

(.006)

.006 (.005) ,083 (.029) ,026 (.OlO) ,004 (.001) ,035 (.003) - ,052 (.004)

,001

(.005) ,081 (.028) ,037 (.OlO) .003 (.001) ,037 (.003) - ,055

(.OW

(.011) ,011 (.008) ,066 (.037) - .016 (.015) ,010 (.OOl) ,035 (.004) - .042 (.006)

.33 6,342

.34 6,425

.32 3,592

Monthly Wage (4)

- ,011 (.010)

,005 (.008) ,088 (.034) ,001 (.014)

,008 (.OOI) .033 (.003) - ,041 .31 3,642

Nore: Controls include department, urban, marital status, and literacy. Numbers in parentheses are standard errors.

157

Mass Emigration, Remittances, and Economic Adjustment

hourly wages. For men, columns 1 and 2 show that wages are higher in urban areas with a high number of migrants but that there is no difference in rural areas. For women, columns 3 and 4 show that the reverse is true. Wages are higher in rural areas with a high number of migrants but not in urban areas. At first glance, the positive relation between wages and having had a member of one's own household migrate is surprising. There is no direct effect on productivity from migration of other household members. One interpretation is behavioral. An increase in nonwage income through remittances increases the reservation wage, which results in lower participation rates and higher wages for those who work. A second interpretation is that this coefficient provides evidence on the type of selection that takes place in migration. Within households, the household will, on average, send members abroad for whom the economic return to the family investment in migration is greatest, leaving behind those with worse labor market characteristics. The significant positive coefficient indicates that positive selection across households dominates the effects of positive selection within households. My second approach is to relate migration to average wages in local labor markets. To do this, I construct a longitudinal data set from department-urbangender cells in the published data from the 1978 Household Survey (MIPLAN 1979) and the tabulated data from the 1985 survey.*' I then estimate the following difference equation:

(4)

wk85

-

Wt78

=a

'

b(zk85

-

zk78)

+ cMk +

ek7

where wkSJ= wk7*is the change in wage in the department-urban cell k between 1978 and 1985, Zk,, - Zk,, is a vector of change in characteristics in the area k, and M k is the proportion of households in area k that have at least one member who left El Salvador between October 1979 and 1985. These regressions, presented in table 5.9, show that areas with greater increase in average education and regions with a high proportion of females in the labor force had larger wage growth. A possible explanation for the lower wage growth in urban areas is the increase in labor supply in cities arising from an increase in rural-to-urban migration. The outflow of migrants is associated with a higher change in the real wage between 1978 and 1985. The effect of a 1 percentage point increase in the proportion of households with migrants who left between the two dates is a 0.6 percent increase in the rate of change of wage. These estimates are consistent with those found in the cross-sectional data.

5 . 3 . 2 Labor Force and Unemployment With 40 percent of its population below the age of 15, El Salvador has a labor force that is projected to continue growing at an average rate of 3.0 21. There are fifty-six such cells, arising from fourteen departments, urban or rural area, and two sexes.

158

Edward Funkhouser

Table 5.9

Regressions of Change in Wage by Department, 1978-85: Change in Logarithm of Wage for Department-Urban-Sex Cell (1)

Intercept

- ,710

(.179) Change in education Change in log employment Change in unemployment rate Female Urban Proportion households with migrants Department dummies R2 N

,097 (.109)

,078 (.032) - ,251 (.057) ,021 (.008) Yes .53 52

(2)

- ,526 (.157) ,143 (.036)

(3)

(4)

- ,635 (.183)

- ,467 (.158) ,146 (.036)

- .209

- .224 (.308) .118

,048

(.092)

.087 (.027) - .147 (.056) ,006 Yes .79 52

(.371) .I10 (.058) - ,248 (.057)

(.@W - ,140 (.054)

,018

,004 (.008)

Yes .68 52

Yes .79 52

Note: There are no data included in the 1985 Household Survey for the rural parts of the depart-

ments of San Vicente and Morazan. Regressions are weighted by economically active population in cell in 1978. Numbers in parentheses are standard errors.

percent per year for the next twenty years.22Especially during periods of poor economic performance, emigration may act to reduce the number of unemployed. ‘In the case of El Salvador, a low estimate of the amount by which emigration between 1978 and 1985 has reduced the labor force of persons aged 20-59 is 185,000. This is nearly 50 percent of the 400,000-person increase in the labor force that would have been expected without migration.23 In other words, working-age out-migrants are equal to 11.2 percent of the potential work force of 1985. Again using the counterfactual assumption that only 5 percent of households had migrants instead of 13 percent, these numbers indicate that the labor force would have increased by approximately 6.9 percent without the increase in emigration. The larger labor force would have put downward pressure on wages, which 22. Calculation based on the data found in MIPLANXELADE (1986) using current labor force participation rates. 23. The expected increase in the labor force from aging of cohorts with labor force participation rates held constant is 315,000. The effect of migration accounts for 40,000 of the increase of 60,000 that can be attributed to changes in labor force participation rates. With the addition of 25,000 immigrants, the increase in the labor force is approximately 400,000. Of this increase, 175,000 is the observed increase in the labor force in El Salvador. The remaining 225,000 is the low estimate of the number of persons 19-59 who would have been working in El Salvador had there not been an increase in migration following 1978. The change in the labor force due to migration is thus 185,000.

159

Mass Emigration, Remittances, and Economic Adjustment

in turn would have created some additional jobs. I estimate this effect by approximating the effect of migration of wages and the effect of wages on employment. I calculate that, for each 1 percentage point increase in migration, the average change in real wages between 1978 and 1985 increased by approximately 4.0-8.0 percent more than they would have otherwise. For a wage elasticity of employment of 0.5, this would imply an increase in employment of between 2.0 and 4.0 percent. These calculations indicate that, had migration not increased, all else equal, the unemployment rate would have increased by an additional 2.9-4.9 (6.9 - 4 to 6.9 - 2) percentage points.

5.3.3 Effect on Foreign Exchange Markets Although remittances to El Salvador do affect household consumption and i n v e ~ t m e n tthe , ~ ~main feature of remittances to Central America in the 1980s is the effect on foreign exchange markets. Although the increase in dollars from remittances should act to relieve current account problems and stabilize the exchange rate because otherwise the economy would have experienced foreign exchange shortages, this was not the case initially in El Salvador. During the period when remittances were increasing, the government was attempting to implement a two-tier exchange ratez5 and to control foreign trade.26Because the success of such a system depends on the proper allocation of dollars between the official and parallel government markets, the increase in the remittances exchanged in the decentralized black market undermined the government attempt to control exchange markets. The deficit in the official exchange market worsened, as only $50 million of the $400 million in remittances in the mid- 1980s was changed in the official exchange market .27 The imbalance eventually led the government to abandon the two-tier system and adopt a 100 percent devaluation of the official rate (to five colonies/US$) during 1985 (U.N. Economic Commission for Latin America, various issues).z8The Salvadoran experience highlights the importance of attracting remittances into the official sector and provides a strong argument against attempts to control foreign exchange markets in countries in which remittances are a significant part of the black market for dollars. 24. In the case of El Salvador, previous research has found that, with the exception of spending on residential construction, nearly all dollars received are used for consumption (Montes 1987). This is in large part due to the decline in the real wage described above. Lopez and Seligson (1990) also find that remittances are an important source of investment for small businesses. 25. With a subsidized rate for the import of essential items, a higher rate for other imports, and the rate for exports in between the two. 26. In late 1979 and early 1980, all import and export activities were nationalized. Subsequently, more stringent restrictions on foreign exchange transactions, including deposits exceeding 100 percent of import value and stiff penalties for violating exchange laws, were implemented. In 1981, unnecessary imports from non-Central American countries were eliminated. 27. A reasonable estimate of imports financed in the black market, based on the excess of the value of imports above import permits issued, is $300 million in 1985. The remaining $50 million in remittances in the parallel market finances capital flight (see Webb et al. 1988). 28. In addition, the government unsuccessfully implemented a preferable exchange rate for remittances through the official sector.

160

Edward Funkhouser

5.4 Summary and Conclusions El Salvador provides a good experiment in the effects of a large outflow of labor combined with an inflow of remittances. Emigrants from El Salvador are disproportionately male, more urban, better educated, and disproportionately aged 20-39 than the Salvadoran population. The main source of this positive selection in labor market characteristics is selection across, rather than within, households. In addition, households from which emigrants leave tend to have more income and to be larger than nonmigrant households. The incentives to migration, measured by the absolute earnings differential for the individual and return on the migration investment for the household, are large. Earnings differentials alone, however, do not provide a complete explanation for the selection in migration because the most important difference in the earnings distributions between El Salvador and the United States is that in mean earnings. Migration has important effects on the sender-country labor market in addition to the direct loss of labor force. The main finding of this paper is that migration from El Salvador has affected labor market activity of nonmigrants. Remittances have strong negative effects on labor market participation of nonmigrants. In addition, the labor force participation rate of women is higher both with the emigration of a household member and with a high proportion of migration in the regional labor market. A second result is that wages are affected by changes in the labor force caused by migration. The effects of shortages of skilled labor are not as severe now as they may become as the demand for labor increases with economic recovery and the skilled emigrants do not rGturn to El Salvador. The findings of reduced labor force participation for males and a wage response to out-migration also suggest that, during the current period of economic decline, emigration has reduced unemployment pressures.

Appendix A Data Sources, Estimates of Volumeof Migration, and Estimates of Remittances: Use of the National Household Survey of 1985 and the UCA Survey of 1987 It remains difficult to estimate the number of Salvadorans who have emigrated. The official migration statistics show over 700,000 more Salvadorans have left the country than have returned. Data from the two available household surveys in El Salvador-the 1985 Encuesta de Hogares de Propositos Multiples (MIPLAN 1986) and the 1987 University of Central America

161

Mass Emigration, Remittances, and Economic Adjustment

(UCA) (Montes 1987)-suggest a much smaller number, at most 500,000 emigrants. Casual estimates based on assistance organizations have resulted in estimates above one million Salvadorans when the United States, Mexico, and other Central American countries are aggregated. What is disturbing about using the household surveys is that it is unlikely that there is a large upward bias to the official statistics. If there is a bias, it is to underestimate-persons fleeing the war unofficially across the border to Honduras and persons who left through any border during the FMLN offensive in 1980-81 are possibly undercounted.

Data Sources: The 1985 Household survey and 1987 UCA Survey In 1985, the Salvadoran government conducted a nationwide survey that included all but two rural areas of the country. The sample design for the survey was that of a similar survey conducted in 1978. Expansion factors were adjusted to correspond to estimates by the Ministry of Planning (MIPLAN) of the population by department in 1985. On the questionnaire, the following question was asked of each household: “Since October of 1979 until the present, is there any member who normally lives in this household and who has left to live in another country?” If the respondent answered yes, the number of household members outside the country was given, and relationship to the respondent, sex, age, and occupation were identified for each emigrant. Using the expansion factors for the household and the information provided in these questions, the calculated tables from this survey show 193,096 Salvadorans who had left after 1979 to have been outside the country in 1985. The second source of household data from El Salvador in which questions about family r’nembers abroad were asked is the 1987 UCA Survey (Montes 1987). Continuing previous surveys on internally displaced persons (Montes 1985, 1986), the 1987 survey asked about relatives living outside the country. For each principal person abroad, forty-three subsequent questions were asked, including family relationship, sex, age, marital status, education, whether the emigrant left a spouse or children in El Salvador, year of arrival in the United States, city of entrance, legal status, reason for emigrating, English ability, labor market status, income, money sent to family members in El Salvador, use of remittances, and desire of emigrant to return to El Salvador. By calculating the proportion of households with members in the United States (35.6 percent), multiplying that proportion by the number of households in El Salvador according to MIPLAN (1,079,245) and by the average number of household members in the United States (2.741), then adjusting for a calculated number of household members that could be cited by more than one household (11.2 percent), Montes arrived at the estimate of between 988,551 and 1,042,340 Salvadorans in the United States. In the first survey, little information about the emigrant is known. In the second, little information about family members in El Salvador is known, and

162

Edward Funkhouser

no information about families without relatives in the United States is given. More important, in both surveys, the responses to the detailed questions suggest that respondents tended to provide information only about adult family members who had emigrated (see table 5A. 1). The calculation based on the 1985 survey is likely to have underestimated the number of Salvadorans abroad because it asks only about former residents of the current household, not about a unit of observation that has not changed over time. The calculation based on the 1987 survey is likely 'to have overestimated the number of Salvadorans abroad since double counting is likely to have occurred. Nonetheless, these two surveys provide the best source-country information available for estimating the number of Salvadorans outside the country.

Estimates of the Volume of Migration Estimates with the 1985 Household Survey In the 1985 Household Survey, 13.3 percent of the 948,000 households represented by the sample, or 126,000 households, stated that at least one household member was living abroad. Three adjustments are necessary with these data. First, these data count only migrants who left after 1979. I adjust by the proportion of households with immediate family members abroad all of whom migrated prior to 1979 in the UCA Survey (21.2 percent) and by the average number of migrants in all households with members abroad who migrated prior to 1979. Second, the number of children identified in the data is implausibly small, suggesting that respondents identified only adult household members abroad. Since this problem is also present in the UCA Survey data, any correction rule will be somewhat arbitrary. The first correction considered is the application of the ratio of Salvadorans older than 10 to the ratio of Salvadorans younger than or equal to 10 found in the emigrant Salvadoran population in the 1980 U. S . Census, 11.8 percent. Third, by asking about usual household members who are outside the country, the survey does not capture whole households that emigrated or, more important, members of a previously restructured household. Estimates with the 1987 UCA Survey The design of the sample for the UCA Survey intended the number of households with family members in the United States to be distributed by department proportionately to the distribution of the total population. For each department, the number of households without family members in the United States is known, but no further information about these households was given. From these sample data and the household information from the 1985 Household Survey, factors of expansion by department and urbdrural area were constructed. A weighted average of the number of family members abroad using the departmental-urban cell factors of expansion differed by only 8.0 percent from the calculation based on a national average.

163

Mass Emigration, Remittances, and Economic Adjustment

Although each family reported the number of family members abroad, the forty-three follow-up questions, including family relationship, were not asked of all such persons. Of the 3,440 persons included in the 1,282 households interviewed with family members abroad, the detailed questionnaire was asked of only 2,121. Because the unit of observation is the household in El Salvador, it is necessary to restrict the emigrants corresponding to a particular household to those persons who would have been living in the household had they not migrated. Unfortunately, this information must be inferred from the familial relationship of the migrant. Of the 663,535 emigrants in the UCA Survey represented by the detailed questionnaires for whom family relationship is known, 38,699 were parents, 180,577 were children, 16,770 were spouses or partners, 179,197 were siblings, 3,735 were grandparents, 18,535 were nieces or nephews, 75,943 were aunts or uncles, and 150,079 were cousins. As a first approximation, emigrants identified as spouse, child, or parent are presumed to be household members. With such a restriction, only 36.1 percent of the original sample of respondents can be included in an expansion using the household as the unit of observation, and only 12.7 percent of the households in El Salvador have members outside the country. A better determination of which emigrants would be household members if they had not emigrated is to examine household structure in El Salvador, using the 1985 Household Survey. Unfortunately, the 1985 survey does not ask family relationship of household members who have emigrated. The generational distribution of emigrants, shown in table 5A. 1, suggests that the “household head” whose relationship to the emigrant is being determined differs in the UCA and Household surveys. Thus, brothers in the UCA Survey most likely correspond to children in the 1985 survey, aunts and uncles in the UCA Survey to brothers in the 1985 Table 5A.1

Age Head Age Emigrant’

< -11

Generational Distribution of Emigrants: Calculations from the 1985 National Household Survey and the 1987 UCA Survey 1985 Survey, Proportion of Emigrants (%)

-10t09

1.8 14.4

10-29

35.8

3-9 Over 50

41.8 6.2

Relationship to Respondent Parents Siblings, cousins, brothers/sisters-in-law Children, nieces/ nephews, daughters/ sons-in-law Grandchildren Great-grandchildren

UCA Survey, Proportion of Emigrants (YO) 17.2 52.2 30.0

.6 ...

aThe difference in age indicates the number of generations that separate the head of the household in El Salvador and the emigrant. For example, an emigrant who is more than eleven years older than the household head in El Salvador is of the generation of the household head’s parents. An emigrant whose age is within ten years of the household head’s is of the same generation.

164

Edward Funkhouser

survey, and so on. Therefore, in order to estimate the number of persons identified as brothers in the UCA sample who might have lived in the respondent’s household had they not migrated, the number of persons one generation younger than an older head (over 50) in houses in which at least one offspring was already living was calculated. Six percent of the families in the 1985 survey in which there was an adult child present also reported an emigrant one generation removed from the household head. This figure corresponds with 12.2 percent of the households in the UCA Survey that indicated a sibling abroad. The average number of siblings abroad per household reporting a sibling abroad is similar between the two surveys (1.28 in the Household survey, 1.35 in the UCA Survey). The resulting estimate of the number of adult brothers or sisters abroad is approximately 49.3 percent of brothers reported, or 88,344 persons. Including these persons as immediate household members abroad increases the estimate of Salvadorans outside the country to 31 1,355. A final adjustment of these data is to account for the migration of children. It is likely that children are included in the persons for whom detailed questionnaires were not recorded. This adjustment is made by applying the proportion of persons under 14 in the age distribution of Salvadoran immigrants in the 1980 Census (1 1.8 percent, table 5.2). The total number of Salvadorans in the United States including this adjustment is 348,05. Randomness of the UCA Sample The expansion factors used in the UCA study were calculated by averaging the number of households encountered without an emigrant before each household in which emigrants were encountered, or ITu

= 1/(1

+ z),

where IT is the proportion of households with emigrants, and Z is the average number of households encountered without an emigrant before encountering a household with an emigrant. In the sample for 1987, Z is approximately 1.82, which implies that 35 percent of Salvadoran households have relatives in the United States. If the samples are random, the distribution of the number of households encountered prior to encountering a household with an emigrant is geometric. The probability of encountering a household with an emigrant after encountering X households without an emigrant is prob(X) = (1 -

IT)”(IT).

The expected number of houses encountered prior to finding a house with an emigrant is W

z = IT 2 X(1 and IT

=

I/( 1

-

IT)X

= (1 -

IT)/IT,

+ Z)gives the true proportion of households with emigrants.

165

Mass Emigration, Remittances, and Economic Adjustment

The extent to which the data represent a random sample can be seen from how well the observed distribution of Z corresponds to expected distribution. In table 5A.2, the actual and expected distributions for households with any family members and immediate household members abroad are presented. This calculation for own household members abroad was made by assuming that, within each department, the order in which the households were surveyed was the same as the order in which they were numbered. Any households that fell at the end of a department ordering but did not have immediate family members abroad were deleted from the sample as they would not have been interviewed under the new emigrant rule. These data suggest that, for both sampling rules, but particularly the sample of persons with any relative abroad, the UCA sample may be slightly biased toward finding households with emigrants. I use the estimates from the household data shown in table 5A.3 as a lower bound and the official migration data as an upper bound to provide a range for emigration from El Salvador in the 1980s. The shape of the official migration data was applied to three variants, and the resulting estimates of the stock of emigrants and the net yearly flow are shown in tables 5A.4 and 5A.5. In each of the variants, it is assumed that the stock of emigrants in 1978 was 164,700 and that the amount of official net migration in 1981 was 90,000 (instead of the 40,000 reported). In the low variant, it is assumed that the stock of emigrants outside the country was 300,000, corresponding to the household surveys. In the high variant, it is assumed that the stock of migrants in 1985 was 650,000, corresponding to the official data. In the medium variant, a stock of 450,000 emigrants in 1985 is used. Table 5A.2

Randomness of the UCA Sample in El Salvador: Comparison of Actual Distribution of Number of Households Encountered without an Emigrant with Random Pattern Any Family Member

Immediate Household Member

Households without Emigrants

Proportion of Sample

Geom. Dist. with 71 = .35

Proportion of Sample

Geom. Dist. w i t h a = ,135

0 I 2 3 4 5 6 7 8 9 10 or more

46.8 13.5 12.5 8.4 4.0 5.3 3.4 2.9 1.1 .4 1.8

35.0 22.8 14.8 9.6 6.2 4.1 2.6 1.7 1.1 .l

19.2 12.0 10.8 1.5 5.9 1.2 5.4 4.5 4.5 2.9 20.1

13.5 11.7 10. I 8.7 1.6 6.5 5.1 4.9 4.2 3.1 23.5

Mean

1.81

.1

1.85

6.24

6.41

166

Edward Funkhouser

Table 5A.3

Estimation of Number of Emigrants Based on 1985 Household Survey and 1987 UCA Survey

No. of households % with emigrants from household No. of households with emigrants Nuclear family members abroad represented Adjustment for emigrants prior to 1979 (21.2%)

1985 Household

1987 UCA

Survey

survey

947,326 13.3 126,105 193,439

1,079,245 13.5 149,711 223,011

234.448

Adjustment for siblings in household (49.2%)

88,344

Adjustment for children not counted (1 1.8%)

27,665 262,113

Total no. of emigrants

Table 5A.4

Table 5A.5

36,740 348,095

Estimate of Stock of Migrants in El Salvador, 1978-87

Low Variant

Medium Variant

High variant

1985 sto(:k

300,000

450,000

650,000

1978 1979 1980 1981 1982 1983 1984 1985 1986 1987

164,700 181,600 202,000 224,100 255,800 273,700 287,400 300,000 312,900 320,300

164,700 200,300 243,300 289,900 356,800 394,600 423,400 450,000 477,200 492,800

164,700 225,200 298,500 377,700 491,500 555,800 604,700 650,000 696,300 722,900

Low Variant

Medium Variant

High Variant

300,000

450,000

650,000

16,900 20,400 22,100 3 1,700 17,900 13,600 12,600 12,900 7,400

35,600 43,000 46,600 66,900 37,800 28,800 26,600 27,200 15,600

60,500 73,500 79,200 113,900 64,200 48,900 45,300 46,300 26,600

Estimates of Net Migration

1985 stock 1979 1980 1981 1982 1983 1984 1985 1986 1987

167

Mass Emigration, Remittances, and Economic Adjustment

Calculations of Amount of Remittances The official source of remittances is the balance of payments accounting of the Central Bank. The Central Bank provides three estimates of remittances, all shown in columns 4-6 of table 5.1. First is the estimate of remittances exchanged through the banking sector. Second is the volume of private unrequited transfers recorded in the balance of payments accounts. Third is the estimate of total remittances through both the official and the black exchange markets. The third series is an estimate based on the number of emigrants abroad. The total value of remittances is calculated by multiplying remittances per emigrant in 1979, the last year for which remittances were channeled through the formal exchange market, by the estimate of the number of emigrants for each year. First, the Central Bank chose 1979, in which personal remittances were $49.2 million, as the base year. Second, the number of emigrants was determined to be 164,700 in 1979 and 500,000 in 1985. Between these two years, an interpolation adjusted slightly for the intensity of the internal conflict was applied to yield the estimation of emigration by year found in table 5A.6. The Central Bank projections of the number of emigrants over the period 1979-87, seen in table 5A.6, are 182,300 lower than the increase reported in exits and entrances of official migration statistics but consistent with the estimation of the emigrant population reported above. One possible calculation of remittances based on these estimated migration flows would be to apply the average remittance per migrant in 1979 ($299) or 1978 ($378, 120,000 emigrants) across the total number of emigrants estimated for a giGen year. The results of these calculations are shown in columns 1 and 2 of table 5A.7. The third column shows annual remittances based on average remittances of $907.40 per person abroad derived from the UCA Survey data. Table 5A.6

Central Bank Estimates of Emigration Adjusted Rate of Emigration (1)

1978 1979 1980 1981 1982 1983 1984 1985 1986 1987

... 30.6 26.6 22.6 18.6 14.6 10.6 6.6 2.6

No. of Salvadoran Emigrants (2)

Yearly Increase (3)

164,700 215,100 272,300 333,800 395,900 453,700 501,800 534,900 548,800

50,400 57,200 61.500 62,100 57,800 48,100 33,100 13,900

...

168

Edward Funkhouser

Table 5A.7

Remittance Projections Based on Aggregate Emigrant Projections (millions $US) Projections Based on Average Remittances

1978 (1)

I978 1979 1980 1981 1982 1983 I984 1985 1986 1987

45.3 62.3 81.3 103.0 126.2 149.7 171.5 189.7 202.2 207.5

1979 (2)

1987 (3)

...

108.7 149.6 195.2 247.3 295.7 359.4 411.7 455.4 485.4 498.1

49.2 62.9 79.7 97.7 115.9 132.8 146.9 156.6 160.6

Central Bank Proj. (4)

... 49.2 59.6 74.7 87.3 97.0 121.4 101.9 134.5 168.7

Note: Columns 1-4 are based on the Central Bank estimate of the number of emigrants in col. 2 of table 5A.6.

The method of the Central Bank, shown in column 4, is to calculate an effective net increase in remittances based not only on the increase in the number of migrants but also on the change in the exchange rate, the level of the domestic interest rate, and the rate of inflation. The overall effect of the Central Bank adjustment is to lower the projected level of remittances for the years prior to the rapid inflation following 1985. The second source for data on remittances is the survey conducted by Segundo Montes in which detailed questions were asked both about family members who sent dollars and about the total amount of dollars received from all sources. Multiplying the previously calculated number of one million Salvadorans in the United States by an average of $113.62 sent per month, Montes (1987) estimates total remittances at $1.4 billion dollars. With the data available, the proper method for making this calculation is again the household unit in El Salvador. The total number of households multiplied by the proportion with any relative abroad and the average remittances received from all sources yields the total value of remittances for 1987, shown in table 5A.8. In table 5A.9, high and low estimates of remittances for each of the three levels of migration variants are shown. For all, the following assumptions are used. (1) Remittance behavior depends on destination. (2) The number of Salvadorans in the United States is equal to the number found in the 1980 Census, adjusted upward by 50 percent to account for undersampling, or 141,670 persons. (3) Until 1979, 60 percent of emigrants ended up in the United States. After 1979, this proportion increased to 80 percent. (4)The shape of migra-

Mass Emigration, Remittances, and Economic Adjustment

169

Table SA.8

Estimates of Remittance per Emigrant Based on the 1987 UCA Survey

Total no. of households in El Salvador Proportion with family members abroad (9%) Proportion of families with members abroad who receive remittances (%) Average monthly remittance ($) per household from all sources (all households with family members abroad) Average monthly remittance ($) per household from all sources (households that receive remittances) Total remittances ($ million), 1987 = (1) X (2) X (4) X 12

1,079,245 35.6 62.3 98.4 158.3 453.7

Average ($) per emigrant, 1987: 348,095 emigrants 500,000 emigrants 700,000 emigrants

1,303 907 648

Nore: The average per emigrant is across all emigrants, independent of whether they are working.

Table 5A.9

1985 level

1980 1981 1982 1983 1984 1985 1986 1987

Author Estimates of Volume of Remittances Migration Low

Migration Medium

Migration High

300,000

450,000

650,000

Remittances

Remittances

Remittances

Low Variant

High Variant

Low Variant

High Variant

Low Variant

High Variant

94.4 111.5 135.5 152.6 168.0 183.4 199.7 212.7

121.0 146.9 182.6 210.7 237.1 264.1 292.7 317.2

94.4 125.2 170.5 201.4 228.6 225.8 284.9 307.1

132.7 176.4 240.9 289.1 333.9 379.9 429.5 470.2

94.4 143.5 217.2 266.4 309.3 352.4 398.6 433.1

148.3 215.8 318.5 393.8 463.0 534.4 611.9 674.1

tion flows over time is similar to that of the official statistics, with one adjustment; for 1981, it is assumed that net emigration was 90,000 to account for uncounted exits. The high variant assumes that remittances per emigrant from the United States increased from $600 to $1,200 between 1978 and 1987 and that remittances from other areas increased from $250 to $400. The low variant assumes that remittances from the United States increased from $600 to $900 and that there are no remittances from other areas.

Appendix B Estimates of Salvadoran Emigration Study

Type of Emigration

Period of Emigration

Principal Findings

Source country studies MIPLAN (1988)

Registered migration

1972-89

Official exits minus entrances: 1972-77 232.2 1978-83 394.9 1984-89 324.1 Total 1972-89 951.2 Total 1978-89 719.0

MIPLAN (1986)

Household members outside the country

October 1979-October 1984

193,096 family members outside the country; 72.5 % are between the ages of 20 and 39

Montes ( 1987)

Relatives in U.S. (compared with sample of Salvadorans in US.)

Residence in U.S. in 1987

Between 988,551 and 1,1042,340 Salvadorans lived in the U.S. in 1987. Emigrants tend to come from urban areas, to take unskilled jobs in the U.S., and not to want to return to El Salvador. A higher proportion are found to have emigrated for political reasons in the 1980s. On average, emigrants send $1 15 per month to relatives in El Salvador.

MIPLANi CELADE ( 1986)

Annual levels of net emigration

1980-2025

From 1970 to 1975, net migration of Salvadorans was 23,000 per year. From 1975 to 1980, net migration increased to 42,000 per year. For 1980-85, net migration was estimated to be 76,000 per year. And for 1985-90, net migration was projected to be 42,000 per Year

1975-85

355,000 Salvadorans working in the U.S. A conservative estimate of remittances is $250-$300 million per year

Webb et al. (1988)

702,900 Salvadorans living in the U.S. in 1980. Of these, 576,400 are working. Remittances are estimated to be between $286.3 and $629.9 million for 1986

Lopez and Seligson ( 1990) Recipient country studies

CIREFCA (1989a, 1989b)

Refugees

Peterson (1987) and Torres-Rivas and Jimenez ( 1985)

International migrants

Levels 1989

Levels 1987 (Peterson) and 1985 (Torres-Rivas)

3,000 13,269 20,000 70,000-1 20,000

Salvadorans in: Belize Costa Rica Guatemala Honduras Nicaragua Mexico Panama U.S.

Torres-Rivas: 3 ,000 10,000 70,000 19,000 17,500 120,000 1 ,000

Belize, 1980 Costa Rica, 1984 Guatemala, 1981 Nicaragua, I97 1 Mexico, 1980 Panama, 1980 U.S., 1980

Salvadorans counted in most recent national Census

U.S. INS, Sfatistical

Refugees in: Guatemala Honduras Nicaragua Mexico

Legal immigration to the U.S.

1975-86

Cohort 1975-80

Amnesty applicants

Continuous residence in U.S. since 1 January 1982

143,000 applicants

=

Yearbook (1986)”

U.S. INS (1989)

Peterson: 6,609 12,975 180,049 28,276 18,074 120,000 797 280,000

1,150 8,743 16,805 2,210 2,055 1,791 94,447 25,611; cohort 1981-86 = 53,785

Appendix C Data Sources Source

Sample

Coverage

Questions on Migration

Data sources in El Salvador National Household Survey, 1985'

UCA Survey, 1987b

National, except two rural areas; 42,156 persons in 9,103 households

Demographic characteristics, labor market activity, and fertility

Each household was asked if any members had left the country between October 1979 and October 1984. For each emigrant, age, sex, and occupation were asked

1,287 Salvadoran households that

Demographic characteristics, labor market performance, migration information, and remittances of the 2.1 12 family members in the U.S.

Familial relationship of emigrant, year of emigration, motive for emigration, labor market status and income in the U.S., and remittance behavior

Demographic characteristics, labor market activity, fertility. and residence in 1979, 1984, and 1986

Each household is asked if it receives any assistance. Those that receive money are asked whether the source is national or foreign

had at least one relative in the U.S.

National Household Survey, 1988'

25,184 persons in 5,563 households

in urban areas

Data sources in the United States U.S. Census, 1980

4,792 Salvadorans in 5% A Sample

Demographic characteristics, labor market status, and income in 1979

Bracket year of immigration, language ability, and naturalization status

CPS, March-April-June match, 1983

68 Salvadorans over the age of 14

Demographic characteristics and labor market activity

Year of immigration, naturalization status, country origin of parents, language spoken, and children born abroad

UCA Survey, 1987h

1,255 Salvadorans in U.S. drawn from consulates and Central American assistance groups

Labor market performance and living arrangements

Year of immigration, legal status, motive for migration, and remittance behavior

CPS, June 1988

124 Salvadorans over the age of 14

Demographic characteristics and labor market activity

Year of immigration, naturalization status, country origin of parents, language spoken, and children born abroad

'MIPLAN (1986). hMontes(1987). cMIPLAN (1989b).

174

Edward Funkhouser

References Bhagwati, J., and R. Brecher. 1981. Foreign Ownership and the Theory of Trade and Welfare. Journal of Political Economy 89(June):497-5 11. Conferecia Internacional Sobre Refugiados Centroamericanos (CIREFCA). 1989a. Documento de la Republica de El Salvador. Guatemala City, February. . 1989b. Documento de la Republica de Honduras. Guatemala City, February. Djajic, Slobodan. 1986. International Migration, Remittances, and Welfare in a Dependent Economy. Journal of Development Economics 2 1:229-34. Gallardo, Maria Eugenia, and Jose Roberto Lopez. 1986. Centroamerica: La crisis en cifras. San JosC, Costa Rica: FLACSO. Instituto Nicaraguense de Estadisticas y Censos (INEC). 1989. Diez anos en cifras. Managua. Instituto Salvadoreno del Seguro Social ( I S S S ) . 1978. Estadisticas: ISSS. San Salvador. . Estadisticas: ISSS. San Salvador. International Monetary Fund (IMF). Various issues. International Financial Statistics. Washington, D. C. Lopez, Roberto Lopez, and Mitchell Seligson. 1990. Small Business Development in El Salvador: The Impact of Remittances. Working Paper no. 44. Washington, D.C.: Commission for the Study of International Migration and Cooperative Economic Development, June. Ministerio de Planificacion (MIPLAN). 1975. Encuesta nacional de mano de obra y aspectos demograjcos, Abril-Julio, 1975. San Salvador: Direccion General de Estadistica y Censos, Seccion de Investigaciones Muestrales. . 1979. Encuesta de hogares de propositos multiples, Febrero-Septiembre, 1978. San Salvador: Direccion General de Estadistica y Censos, Unidad de Investigaciones Muestrales, October. . 1986. Encuesta de hogares de propositos multiples, 1985. San Salvador: Direccio? General de Estadistica y Censos, Unidad de Investigaciones Muestrales. . 1988. Indicadores economicos y sociales. Avance no. 2. San Salvador: Direccion General de Estadistica y Censos, August. . 1989a. El Salvador: Program economic0 y social para 1989. San Salvador: Direccion General de Estadistica y Censos, January. . 1989b. Encuesta de hogares de propositos multiples, Enero-Junio, 1988. San Salvador: Direccion General de Estadistica y Censos, Direccion General de Coordination, Unidad de Investigaciones Muestrales. Ministerio de Planificacion (M1PLAN)ICELADE. 1986. Estimaciones y proyecciones de poblacion, 1950-2025. San Salvador: Direccion General de Estadistica y Censos, Direccion de Poblacion, November. Montes, Segundo. 1985. El Salvador 1985: Desplazados y refugiados. San Salvador: Instituto de Investigaciones, Universidad Centroamericana Jose Simeon Canas. . 1986. El Salvador 1986: En busca de soluciones para 10s desplazados. San Salvador: Instituto de Investigaciones, Universidad Ceentroamericana Jose Simeon Canas. . 1987. El Salvador 1987: Salvadorenos rejiugiados en 10s Estados Unidos. San Salvador: Instituto de Investigaciones, Universidad Centroamericana Jose Simeon Canas. Peterson, Linda. 1987. Central American Migration: Past and Present. Washington, D.C.: Bureau of the Census, Center for International Research. Torres-Rivas, Edelberto, and Dina Jimenez. 1985. Informe sobre el estado de las migraciones en CentroAmerica. Anuario de Estudios Centroamericanos 11(2):25-66.

175

Mass Emigration, Remittances, and Economic Adjustment

U.N. Economic Commission for Latin America. Various issues. Economic Survey of Latin America and the Caribbean. Santiago, Chile. U.S. Immigration and Naturalization Service (INS). 1989. Provisional Legalization Application Statistics. Washington, D.C.: Statistical Analysis Branch, Office of Plans and Analysis, 12 May. . Various issues. Statistical Yearbook of the Immigration and Naturalization Service. Washington, D.C. Webb, Richard, Alain Thery, Emesto Kritz, and Eliane Karp. 1988. El Salvador: Income, Employment, and Social Indicators: changes over the Decade 1975-1985. Washington, D.C.: International Science and Technology Institute. Mimeo.

This Page Intentionally Left Blank

6

When the Minimum Wage Really Bites: The Effect of the U. S .-Level Minimum on Puerto Rico Alida J. Castillo-Freeman and Richard B. Freeman

Since the passage of the Fair Labor Standards Act in 1938, Puerto Rico has been subject to minimum wage regulations. For many years, industry boards set separate minima by industry and occupation that were markedly below the U.S. federal minimum wage. In 1974, the U.S. Congress, supported by the Puerto Rican government, initiated a policy to raise the level and coverage of federally mandated minimum wages on the island to U.S. standards. By 1983, the minimum wage in Puerto Rico reached the $3.35 per hour rate then prevailing in the United States, and coverage matched the U.S. rate of 60 percent or more of the work force. With hourly earnings on the island just two-thirds of mainland hourly earnings, the result was an extraordinarily high ratio of the minimum wage to average pay-producing a minimum wage with genuine economic bite. To what extent has the U. S.-level minimum reduced employment in Puerto Rico? How important has migration been in adjusting to the minimum wageinduced loss of jobs? Has the minimum contributed to the migration of lesseducated and less-skilled Puerto Ricans to the United States (Ramos, in this volume)? This paper seeks to answer these questions using diverse bodies of data on employment and earnings in Puerto Rico and on the employment and earnings of Puerto Rican migrants in the United States. It reports the following findings. (1) The U.S.-level minimum altered the distribution of earnings in Alida J. Castillo-Freeman is a research assistant at the National Bureau of Economic Research. Richard B. Freeman is professor of economics at Harvard University, director of the Labor Studies Program of the National Bureau of Economic Research, and senior research fellow at the Centre for Economic Performance, the London School of Economics. The authors benefited from discussion with Fernando Ramos, who guided them through much of the Census of Population data on Puerto Rico, and to Roland0 Vele, Antonio Ortiz Miranda, and Antonio Sotto for helping them with CPS data for the island. They also thank Dan Sweeney of the U.S. Department of Labor for furnishing industry-level minimum wage data.

177

178

Alida J. Castillo-Freeman and Richard B. Freeman

Puerto Rico to an extraordinary extent, creating marked spikes that dominate the earnings distribution. (2) Imposing the U.S.-level minimum reduced total island employment by 8-10 percent compared to the level that would have prevailed had the minimum been the same proportion of average wages as in the United States. In addition, it reallocated labor across industries, greatly reducing jobs in low-wage sectors that had to raise minima substantially to reach federal levels. (3) Migrants from Puerto Rico to the United States are drawn largely from persons jobless on the island, with characteristics that make them liable to have been disemployed by the minimum wage. As the Puerto Rican minimum rose toward U.S. levels, the education of migrants fell below that of nonmigrants. (4) Migration was critical in allowing Puerto Rico to institute U.S.-level minimum wages and played a major role in the longterm growth of real earnings in Puerto Rico by reducing the labor supply and raising the average qualifications of workers on the island. We present the evidence for these claims in four stages. First, we describe the minimum wage system in Pugrto Rico and show that it altered the island’s distribution of earnings. Second, we estimate the employment consequences of the minimum using time-series and cross-industry data. Third, we examine the volume and characteristics of migrants to the United States as the island moved toward the federal minimum. Finally, we consider the consequences of migration for the minimum wage system and outcomes in the Puerto Rican labor market.

6.1 The Minimum Wage in Puerto Rico The 1938 Fair Labor Standards Act (FLSA) established minimum wages in Puerto Rico as it did in the United States proper. At first, the law applied the mainland minimum rate ($0.35) to Puerto Rico, but Congress soon recognized that this would devastate the island’s economy and passed a separate amendment that established committees in some forty industries to set separate industry and occupational minima “that would not substantially curtail employment” while also not giving Puerto Rico “an unfair competitive advantage over mainland competitors” (U.S. Department of Commerce 1979, 2:633-42). From 1940 until 1974, amendments to the FLSA expanded coverage on the island but maintained the industry-committee mode of setting minima. With the 1974 and 1977 Amendments to the FLSA, however, Congress introduced a new policy, increasing coverage and enacting automatic increases in Puerto Rican minima to bring them to the U.S. level. The 1977 Amendment required industries with minima at U.S. levels to follow the scheduled mainland increases and those whose minima were below U.S. levels to increase wages by $0.30 per year until they reached the federal minimum. By 1983, Puerto Rico had effectively reached the mainland minimum. Table 6.1 records levels of the minimum wage and coverage on the island and, for purposes of comparison, in the United States, in each year that Con-

Effect of the U.S.-Level Minimum Wage on Puerto Rico

179

Minimum Wage, MinimumiHnurly Earnings in Manufacturing, and Coverage

Table 6.1

Puerto Rico

Year

1950 1956 1961 1963 1967 1968 1914 1915 1916 1978 1919 1980 1981 1987

Min. ($) Min./Mfg. Cov. (1) (2) (3)

.20 .45 .61 .12 .91 1.10 1.68 1.87 2.03 2.5 1 2.71 3.00 3.26 3.35

.41 .10 .62 .64 .70

.29 .29 .29 .29

.11

.44 .41 .66

.12 .13 .13 .75 .15 .15 .I4 .63

.44

.64 .64 .64 .64

.64 .64

United States (Min./Mfg.) (Min./Mfg.) X Cov. Min. ($) Min./Mfg. Cov. X Cov. (4) (5) (6) (7) (8)

.I4 .20 .I8 .I9 .31 .31 .34 .48 .47 .48 .48 .48 .41 .40

.15

1 .oo

1.15 1.25 I .40 1.60 2.00 2.10 2.30 2.65 2.90 3.10 3.35 3.35

.52 .51

SO

.5 1 .50 .53 .45 .44 .44 .43 .43 .43 .42 .34

.36 .38 .43 .44 .55 .54 .62 .60 .60 .62 .63 .63 .63 .64

.19 .19 .22 .22 .28 .29 .28 .26 .26 .21 .21 .21 .26 .22

Sources: h e r t o Rico: minimum calculated from U.S. Department of Labor’s “Minimum Wage Industry Studies”; average hourly earnings in manufacturing from the Yearbook of Labour Statistics; coverage based on unpublished estimates from the U.S. Department of Labor, Employment Standards Administration. United States: minimum wages from the 1990 Staristical Abstract of the United States. table 675; manufacturing earnings from the 1990 Economic Report of rhe President; coverage is estimated from Welch (1978) by multiplying by ratio of nonagricultural private employees to total employment. Note: Cov = number of covered nonsupervisory employees divided by civilian employment

gress changed the law and in 1987. As there was no single minimum in Puerto Rico until the 1980s, the pre-1983 “average minimum” for the island in column 1 is the employment-weighted average of forty-four separate industry minima (based in some cases on averages of occupational minima within industry, as described in Castillo [ 19831). Column 2 gives the ratio of the average minimum to average hourly earnings in manufacturing; it shows that industry boards set rates on the order of 60-70 percent of average manufacturing earnings through 1974. Column 3 presents estimates of the ratio of the number of workers covered by the minimum to civilian employment. Because agriculture, government, and much of the trade and service sector were not covered by the law, however, only 29 percent of the island work force was subject to the minimum from the 1950s though 1967, compared to much higher proportions of the U.S. work force. As a result, until 1967 the ratio of the coverage-weighted minimum to average earnings-a crude measure of the overall strength of the minimum wage-was lower in Puerto Rico than in the United States (col. 4 vs. col. 8). Hence, the effect of the minimum on the aggregate Puerto Rican labor market was modest. Not until 1975 did the

180

Alida J. Castillo-Freeman and Richard B. Freeman

coverage-weighted minimum on the island markedly exceed the U.S. level, after which it remained 80 percent or so higher through 1987. 6.1.1 Effect of the Minimum on Wages An effective minimum wage should produce spikes in the distribution of earnings in the area of the minimum, allowing the analyst to infer the level of the minimum from the shape of the distribution. Our investigation of three data sets of earnings shows such a pattern for the island, with remarkable spikes at the relevant minima in each distribution, implying that the minimum wage law is a major determinant of actual wages paid. The first data set consists of wage distributions for workers in covered industries obtained by the Bureau of Labor Statistics U.S. Department of Labor (“Minimum Wage Industry Studies”) as part of its assessment of separate industry minima through the early 1970s. We examined distributions for dozens of industries and found that, in all low-wage industries, an extraordinary proportion of workers was paid the industry minimum, with modal pay changing with changes in the minimum. For example, in 1964, the hourly minimum was $0.83 in shoes and related products, and 41 percent of workers received exactly $0.83; in 1968, when the industry minimum had increased to $1.17, 84 percent were paid $1.17. Similarly, in 1964, when the hourly minimum in the women’s and children’s underwear industry was $0.96, 49 percent of workers received $0.96, whereas in 1972, when the industry minimum was $1.45,41 percent received that wage. In rubber products, 46 percent were paid the $0.98 minimum in 1963 and 36 percent the $1.32 minimum in 1969. Comparable distributions for industries in the United States show no such bunching of wages around the minimum except for such sectors as sawmills in the south in the early years of the minimum wage law. Second, an examination of earnings from the 1980 Census of Population for Puerto Rico reveals spikes at pay levels where differing industry minima covered workers during the period of transition to the U.S. minimum. The Census asked individuals to report annual earnings, weeks worked, and usual hours worked per week in the preceding year; we used these data to calculate hourly pay by dividing annual earnings by weeks worked times hours worked per week.’ In contrast to the Department of Labor’s surveys, the Census encompasses the entire island, is obtained from individuals who have no incentive to report wages at the minimum, and covers the period when federal regulation rather than industry boards set the minima. Figure 6. l a displays the distribution of 1979 hourly pay for full-time workers in Puerto Rico from the Census files. Since not all industries had reached the federal minimum in 1. One disadvantage of the Census data is that they have not been cleaned for errors due to miscoding etc., as is typically done with CPS data. We deleted observations when weeks worked was less than twenty, when hours worked last week was less than ten or equal to ninety-nine, and when wage was less than $0.50 per hour.

181

Effect of the U.S.-Level Minimum Wage on Puerto Rico

1979, the law cannot be expected to create a single spike in the earnings distribution. In 1979, approximately 50 percent of covered workers were at the U.S. minimum of $2.90,. 13 percent were covered by a minimum within $0.10 of that value, and 25 percent were covered by the industry minimum in the $2.50-$2.60 range. An effective minimum would thus produce one spike at $2.90 and a smaller one around $2.50-$2.60. The figure shows such a pattern. A third source of data on the Puerto Rican earnings distribution is usual hourly earnings ( = usual weekly earnings/usual hours worked) from the Current Population Survey (CPS) in Puerto Rico. The CPS provides earnings information in the 1980s, when the Puerto Rican minimum reached the U.S. level throughout the island. The CPS usual weekly earnings and hours information is closer to wage rates than the census figures for annual earnings and hours/weeks worked and thus should be subject to smaller measurement error than the Census data. We estimated the distribution of hourly earnings for Puerto Rico using CPS files for 1983 and 1988. The distributions for both years show the dominance of the U.S. minimum on the island’s pattern of earnings. Figure 6. l b displays the distribution of hourly earnings from the CPS for 1983. It reveals a spike around the $3.35 U.S. minimum: 25 percent of the workers were paid between $3.30 and $3.40 in that year. The change in the shape of the earnings distribution from one centered on $2.50-$2.60 and $2.90 in 1979 to one centered on $3.35 in 1983 indicates that imposition of the U.S. minimum on Puerto Rico altered the distribution of pay on the island. Figure 6. l c gives the comparable distribution for 1988, at which time the U.S. minimum had been in effect for roughly five years. The spike in the $3.30$3.40 range is as pronounced as in 1983, with 28 percent paid in this range, the vast majority at exactly $3.35. That the $3.35 minimum continued to be the modal rate of pay despite five years of rising nominal average wages is striking evidence that the minimum did indeed constrain island wage setting. As a final test of how the minimum affected earnings in Puerto Rico, we regressed the log of average hourly earnings economy-wide and in manufacturing on the log of the minimum wage from 1951 to 1987, controlling for a general time trend, real GNP in Puerto Rico, and the GNP deflator in Puerto Rico. We used three related statistical specifications: ordinary least squares, a first-order autoregressive structure, and least squares with earnings lagged one year as an additional regressor. The minimum obtained a positive significant coefficient in each regression (table 6.2) with elasticities in the range of 0.20.4. The below-unit elasticities are consistent with the evidence in figure 6.1 that the minimum affects the shape of the earnings distribution by increasing wages at the lower end rather than by producing general wage inflation. In sum, imposition of the U.S.-level minimum wage in Puerto Rico altered hourly earnings on the island in ways that make Puerto Rico an excellent institutional setting for assessing the effects of the minimum on the job market. Does the minimum have the “economic bite” on employment in Puerto

182

Alida J. Castillo-Freemanand Richard B. Freeman A. 1979

n 1

I

I

1

2.352190

10

4

Wage B. 1983

1 .02 I3

Wage

Rico that textbook discussions of wage-fixing laws lead one to expect? Has the minimum wage affected migrant flows as well?

6.2 Employment Effects of the Minimum Wage To determine the employment effects of the minimum, we performed two related analyses. First, we used time-series data to estimate the effect of the minimum on the employment-population rate and unemployment rate on the

183

Effect of the U.S.-Level Minimum Wage on Puerto Rico C. 1988

c

0 .-

wage Fig. 6.1 Distribution of hourly earnings in Puerto Rico Source: a, 1980 Puerto Rican Census. b, 1983 Puerto Rican CPS. c, June 1988 Puerto Rican CPS. Nore: The Census data are based on annual earnings. We deleted observations when weeks worked was less than twenty, when hours worked last week was less than ten or equal to ninetynine, and when the wage was less than $0.50 per hour.

Table 6.2

.

The Effect of the Minimum Wage on Average Earnings in Puerto Rico, 1951-87 Average Earnings

Constant Log min. Trend Log PR def. Log PR GNP

Average Manufacturing Earnings

(1)

(2)

(3)

(4)

-4.34 .32

-3.69 .27

- .26 .31

(.01) .I9

(.@)

(.05)

-2.10 .I9 (.04)

.01 (.01)

.03 (.01)

(.01)

.39 (.12) .51

.45

.I5

(. 11)

(.W

(.07)

.02 ~03)

~07)

,999 ,028 ...

,999 .025 .61

(5) - .42

SE A N 1)

- .21 .20 (.04)

(.20)

.21 (.07) .05 (.02) .06 (.2% - .03 (. 16)

,999 .020

,998 ,031

,999 ,028

,999 ,022

...

...

.81

...

- .oo

.30

(.05) .04

.01

Lagged earnings R2

(6)

(. 14)

Source: Estimated by the authors from data in app. A. Nore: Standard errors are given in parentheses.

-

.oo

.09 (.09) .05 (.06) .I2 (.09)

184

Alida J. Castillo-Freeman and Richard B. Freeman

island. Our time-series regressions are based on models analogous to those in U.S. minimum wage studies (see Brown, Gilroy, and Kohen 1982) in which the key independent variable is an average of the ratio of the minimum to average earnings times coverage among industries. Second, exploiting the fact that industries in h e r t o Rico had different minima for many years and reached the U.S. level at different times, we estimated a cross-sectional timeseries model linking industry employment to industry minima. Table 6.3 presents the results of our time-series analysis for the period 1951-87. As employment and population data for Puerto Rico are given in calendar year and fiscal year terms and are periodically revised on the basis of the latest Census of Population, we present estimates using two related employment and population series. In columns 1-2, our dependent variable is based on calendar year employment and population data for persons aged 16 and over adjusted to the 1980 Census for 1963-87 and fiscal year data for persons aged 14 and over for earlier years.* In columns 3-4, our dependent variable uses a consistent fiscal year series for persons aged 14 and over, also adjusted to the 1980 census benchmark. In columns 5-6, the dependent variable is the In of the unemployment rate. For the minimum, we use two related measures. The first is a “Kaitz” employment-weighted average of each sector’s minimudaverage hourly earnings multiplied by its coverage: where a, is industry i’s share of island employment; m iis the minimum in industry i, w iis the average hourly earnings in industry i, and c, is the coverage in industry i. The second measure is the ratio of the average minimum (m) to averagelhourly earnings (w)in the economy multiplied by an economy-wide coverage figure obtained from a different source than the industry coverage figures (see app. A).3In all cases, the regressions are in In form and include Puerto Rican and U.S. GNPs (in In constant dollars) and a linear trend. Because ordinary least squares regressions showed considerable serial correlation, the calculations are based on an AR(1) model (OLS estimates yielded larger coefficients on the minimum, so our results are the more conservative ones) . 4 The regressions show that, however specified, the minimum had a significant effect on the employment-population rate, with estimated short-run elasticities that range from 0.15 (cols. I and 3) to 0.10 (col. 4). These elasticities

2. There is only a modest effect of changing from the age 14 and over to the age 16 and over data (see U.S. Department of Commerce 1979,2:600, table 16). 3. Because the minimum wage alters average earnings in h e r t o Rico, however, we also estimated a model in which the GDP deflator for Puerto Rico replaces average earnings as the deflator in the minimum variable. Our results are similar but weaker than those in the table. 4. The OLS specification yielded coefficients (standard errors) for the minimum wage variables in the columns in the table of -.24 (.05), - . I 9 (.04), - . I 8 (.05), and - . I 4 (.04) on the employment-population regressions and of .29 (.26) and .27 (.20) on the unemployment rate.

185

Effect of the U.S.-Level Minimum Wage on Puerto Rico

Table 6.3

Regression Coefficients (standard errors) for the Effect of the Minimum Wage (and other factors) on Employment-Populationand Unemployment Rates in Puerto Rico, 1951-87 Employment Rate

Minimum: Kaitz

-.15 (.07)

(MiniAvg.) X comp. Puerto Rican GNP

U.S. GNP Trend

.25 (.11) .38 (.22) - ,024 (.006) -5.51 .65

R2 SE

-.I1 (.05) .27 (.Ill .37 (.22) -.024 (.006) -5.54 .64 (. 16)

.92 .027

.92 ,027

Unemployment Rate

-.15 ~07)

.27 (.20)

.04

- .84 (.57) - 1.21 ~71) ,081 (.022)

- .87 (57) - 1.17 (.70)

12.56 .85 (. 12) .89 ,083

12.52 .86 (. 11) .90 ,084

(. 10)

.62 (.25) -.024 (.006) -5.70 .43 (. 19) .91 ,029

Source: Estimated using data in app. A. Employment rate: measure A uses the B the PREPOPF series.

PREFQP

.21 (. 14)

,080 (.023)

series, measure

are of comparable magnitude to those found on the effects of the minimum on teenage employment in the United States prior to the 1980s (Brown 1988; Brown, Gilroy, and Kohen 1982) but with lower standard errors, presumably because in Puerto Rico a much greater proportion of the work force is affected by the minimum, giving us the equivalent of a larger data set from which to draw statistical inferences. In addition to the calculations in table 6 . 3 , we did several other analyses of the time-series data to check on the robustness of our findings. In one set of calculations, we added additional lagged dependent variables: the coefficients on the minimum wage variable were similar in magnitude and significance to those in the table, and the coefficients on the lagged dependent variable were on the order of 1/2. If we interpret the lagged coefficients in a partial adjustment context, the responses of employment are roughly twice those in table 6.3.5In another set of calculations, we estimated ARIMA models of different orders. The results were comparable to those in 5. Because of the lagged term, we estimated regressions for 1952-87. Our coefficients (standard errors) on the minimum wage variable and the lagged dependent variable for the models in table 6.3 are - .17 ( .06) and .45 ( .19); - . I 3 ( .04) and .48 ( .17); - . 12 ( .OS) and .49 ( .17); - .09 (.04) and .50 (.l6); .30 (.13) and .65 (.16); and .41 (.19) and .41 (.19).

186

Alida J. Castillo-Freeman and Richard B. Freeman

table 6.3, indicating that the finding of a substantial minimum wage effect is robust to the precise model used to analyze the data. Rather than giving all our specifications, we present the time-series data in appendix A, which readers can manipulate as they desire.6 The short-run elasticities of employment to the minimum in table 6.3, while modest in magnitude, suggest that Puerto Rico experienced massive job losses as a result of the application of the U.S. minimum to the island. This is because the minimum is so high relative to average earnings in Puerto Rico. In 1987, for example, the coverage-weighted minimudmanufacturing earnings in table 6.1 were 0.63 In points higher in Puerto Rico than in the United States, and a Kaitz index of the minimum for Puerto Rico comparable to the U.S. Kaitz index was 0.64 In points higher in Puerto Rico than in the United States.’ A 0.64 In difference in the minimum implies that, even with relatively modest elasticities, island employment would have been 9 percent higher in 1987 than if the minimum was at the same level relative to pay as in the United States. For the period 1973-87, our analysis suggests that the increased minimum in Puerto Rico reduced the employment-population ratio by roughly 0.02 points, accounting for over one-third of the 0.052 point actual drop. * Turning to the unemployment results in table 6.3, economic theory has no prediction about the effects of the minimum wage on unemployment rates. Some workers displaced by the minimum may leave the work force (in the case of Puerto Rico, they may leave the island), while others may be attracted to the labor force by the higher pay-and our regressions show correspondingly weaker effects on unemployment rates than on employment-population rates. Still, given the magnitude of the minimum in Puerto Rico, the point estimates imply that the minimum raised the unemployment rate substantially. A 0.63 In point increase in the Kaitz measure of the minimum, for example, 6. In one analysis, we entered the In of the minimum, coverage, and average earnings separately and obtained the following coefficients (standard errors): - .12 (.09) on In minimum; - .10 (.07) on In coverage; and - .10 (.20) on In average earnings. That the In minimum and In coverage have similar coefficients supports the restriction of entering them together in the equation. The insignificant coefficient on average earnings suggests that it plays little role in the results. In another analysis, we estimated the effects of the minimum for the period prior to imposition of the U.S.minimum, 1951-73, and the period when island minima began moving toward the federal level, from 1974 to 1987. The result was a large significant coefficient in the latter period but an insignificant coefficient in the former. 7. The U.S. Kaitz index relates to nonagricultural private wage and salary employment, whereas our measure for Puerto Rico was based on total employment. Accordingly, we estimated a Kaitz index for Puerto Rico based solely on private nonagricultural wage and salary workers in Puerto Rico. 8. Specifically, we multiply the 0.64 difference in the Kaitz minimum wage variable by the 0.15 coefficient in col. 1 and get a 0.096 In point reduction in the employment-population rate. With a 1987 employment-population rate in Puerto Rico of 0.369, this produces a 0.406 employment-population rate, for a 10 percent increase in employment. Similarly, the - 0.11 coefficient in col. 2 yields an estimated reduction in the employment rate of 0.07 In points, for a 7 percent increase in employment.

187

Effect of the U.S.-Level Minimum Wage on Puerto Rico

would have raised the rate of unemployment by 3 percentage points, according to the coefficient in column 5. Note finally that, if workers displaced by the minimum were more likely than other workers to migrate to the United States, the table 6.3 employment and unemployment effects understate the full effect of the minimum in displacing labor. This is because they are based solely on the population in Puerto Rico rather than on the larger number of persons who would have been on the island and unemployed absent the migration option. 6.2.1 Cross-Industry/Time-Series Analyses The existence of separate minima by industry in Puerto Rico from the 1950s through 1983, and the corresponding differential increase of industry minima to the U.S. rate, offers an alternative to the standard time-series mode of estimating the effect of minimum wages on employment. It allows us to assess the effects of the minimum wage by contrasting changes in employment among industries as their minima moved toward the federal level. For this analysis, we created a cross-industry time-series data set for Puerto Rico from 1956 to 1987 by matching employment and earnings data for forty-two industries that cover the entire economy, save agriculture and g ~ v e r n m e n twith ,~ minimum wages from the industry reports of the U.S. Department of Labor (“Minimum Wage Industry Studies”) and with industry coverage from the U.S. Department of Commerce (1979, table 1 , 2:634). With forty-two industries and thirty-one years (we excluded 1982 as there was no Survey of Manufacturing conducted in Puerto Rico in that year), we have a sample of 1,302 observations that provides a stronger test of the hypothesis that the minimum affected employment than thirty-one time-series observations. We use the pooled cross-industrykime-series data set to estimate the effect of minimum wages on employment in an analysis of covariance framework: (1)

In EMP,, = a

+ b ln(c,, x

m,,/w,,)+ T,

+ IND, + u,,,

where T, is a vector of year dummy variables to control for cyclical or trend factors that influence employment, and IND, is an industry dummy variable to control for the scale of employment in an industry, and uI, is the error term. The pooled cross-industryhime-series analysis differs from the time-series analysis in several ways beyond sample size. It allows us to enter separate year and industry dummies and thus isolate the within-industry within-year variation in variables that is generally more difficult to explain than pure timeseries variation. It also permits us to separate our analysis between the period when minima were set by the industry councils and the period when the Congress mandated changes in the minimum and thus to test for the potentially 9. We use data for thirty-seven detailed manufacturing industries from the Puerto Rican Survey of Manufacturing and for five one-digit nonmanufacturing industries from the Departamento del Trabajo y Recursos Humanos.

188

Alida J. Castillo-Freeman and Richard B. Freeman

greater employment effects under the latter regime. With output terms excluded from the equation, moreover, the minimum will capture changes in employment due to minimum wage-induced changes in industry output as well as those due to minimum wage-induced changes in employment with output (level of demand) fixed. And by focusing on employment by industry, the analysis captures the movement of workers from industries with large increases in minimum wages to those with small increases as well as from employment to nonemployment. The latter two considerations suggest that employment effects from the cross-industry analysis will exceed those from the aggregate time-series analysis. Table 6.4 presents the regression coefficients and standard errors on the minimum wage variables from our industry analysis for the entire period and for the subperiods 1956-73 and 1974-87. The estimates strongly confirm the proposition that minimum wage substantially affected employment in Puerto Rico. In column 1, the minimudaverage times coverage variable obtains a negative significant coefficient of 0.54, more than four times its standard error. The estimates in columns 2 and 3 show that the effect of the minimum occurs entirely in the period after the 1974 Amendment imposed increases in minima toward the U.S. federal level. The elasticity of employment to the minimum is -0.91 after 1974, compared to an estimated 0.20 elasticity before 1974. This suggests that the industry councils took seriously the mandate to set minima so that they “would not substantially curtail employment,” whereas the congressionally mandated changes were, of course, exogenous to the economic conditions on the island. Underlying the sizable minimum wage effects in table 6.4 is a substantial Table 6.4

Regression Coefficients (standard errors) for the Effect of the Minimum Wage and Other Variables on Ln Employment-Population, 1951-87, and on Ln Employment by Industry, 1956-87 Cross Industry 1956-87 (1) Minimum Industry dummies Year dummies Sample size R2

- .54 ~13) 41 30 1,302 .87

1956-73 (2) .20 (.I21 41 17 756 .95

1974-87 (3) - .91

( ,241

41 12 546 .95

Source: Estimated by authors from a data set available on disk from the NBER. The detailed industry data for Puerto Rico are obtained from the U.S. Department of Labor’s “Minimum Wage Industry Studies” and the Departamento del Trabajo’s “Seria historica del empleo, despempleo y grupo trabajador in Puerto Rico.” Note: Minimum is the multiplicand of coverage for industry times minimudhourly earnings in industry. Figures exclude 1982 owing to the absence of the Puerto Rican Survey of Manufacturing. The regressions cover forty-two industries, with agriculture and government excluded.

189

Effect of the U.S.-Level Minimum Wage on Puerto Rico

reallocation of Puerto Rican workers from low-wage industries that increased their minima greatly to reach U.S. standards to high-wage industries that required only modestly higher minima to reach the federal level. In industries at the U.S. minimum in 1973, employment grew by 1 percent from 1974 to 1983. In industries whose minimum was within $0.10 of the U.S. minimum in 1973, employment increased by 2 percent over the same period. But in industries whose 1973 minimum was more than $0.10 below the U.S. minimum and that therefore had to raise minima substantially to reach the U.S. level, employment dropped by 32 percent from 1974 to 1983! These calculations indicate that the U. S.-level minimum reallocated employment on the island as well as reducing the total employment-population rate. They show that a major reason for the higher elasticities of employment to the minimum in table 6.4 than in table 6.3 is the movement of workers across industries.

6.3 The Minimum Wage and Migration Since Puerto Ricans face no legal restrictions migrating to the United States, migration depends on the incentives facing individuals, including those that result from the minimum wage. Changes in the minimum can affect both the volume and the composition of migration. Economic analysis has no clear prediction about how the volume of migration might respond to higher minimum wages. Workers will be more likely to migrate if they are disemployed by the minimum than if they hold jobs at below-minimum wages, but workers whose earnings are increased by the minimum will be less likely to migrate. The net effect of the minimum on the size of the migrant flow will depend on the number of workers in the two groups and their response to the minimum wage-induced change in their conditions. l o Economics does, however, predict how the minimum will affect the composition of migrants. Since employers are likely to lay off or forgo hiring the least skilled workers when wages rise, the minimum should induce greater migration of the less skilled. 6.3.1

Volume of Migration

Puerto Rican migration to the United States has been immense. In 1980, one-third of 20-64-year-old persons born in Puerto Rico resided in the United States (Ramos, in this volume, table 2.1). Although there are no administrative statistics on migration from Puerto Rico, data from the U.S. and Puerto Rican Censuses of Population and CPSs and passenger travel records show sizable changes in the volume of migration over time. In the 1950s, massive 10. If, in accord with our estimates in sec. 6.2, the elasticity of demand to the minimum is below one, expected earnings of all workers will be higher under the minimum. This does not, however, imply that the minimum reduces migrant flows. for workers who suffer the loss of jobs but do not gain the benefits of higher wages may respond more to their condition than those who obtain modest gains in wages. Moreover, the elasticity is based on total employment, including workers whose skills make the minimum irrelevant to them.

190

Alida J. Castillo-Freeman and Richard B. Freeman

migration from Puerto Rico more than doubled the share of the Puerto Rico born living in the United States (see table 6.5). For the next two decades, migration rates fell, seemingly stabilizing the share of the Puerto Rico born on the mainland, only to surge once more in the 1980s. Figure 6.2 graphs the only available measure of annual migrant flows-net passengers journeying from the island-relative to the population on Puerto Rico, revealing further fluctuation in flows from year to year. From 1951 to 1987, migration averaged 0.9 percent of the island population, with a standard deviation of 0.9. In the 1950s, 1.9 percent of the island population left annually; in the 1960s, 0.5 percent; in the 1970s, 0.3 percent; and in the 1980s, 1.0 percent. To see if the swing in migration is related to the minimum wage, we regressed the ratio of net migrants to the Puerto Rican population (MIG) on the Kaitz measure of the minimum, economic conditions in the United States and Puerto Rico, and the proportion of the Puerto Rico born living in the United States in the previous year (PRUS), using various measures of economic conditions and specifications. In nearly all cases, the minimum variable had a statistically insignificant effect on the volume of migration. For example, controlling for GNP in Puerto Rico and the United States, we obtained MIG=

- . 1 4 + . 0 7 log

USGNP-

(.02)

.05 log

PRGNP+

(.oa

,005 log

M I N - . ~ PRUS ~

(.05)

(.01)

R2

= .57 (standard errors in parentheses). Here, the coefficient on the minimum wage implies that the 0.64 In point lower minimum that would make it comparable to the minimum on the mainland relative to average earnings would reduce migration by 0.3 percent of the population, but the standard error is too high to place any confidence in this estimate. One interpretation of the high standard error is that the emigration-

Migration from Puerto Rico, 1950-87

Table 6.5

1950 I960 1970 198@ 1987

Puerto Rico. Total Population ( I ,000s) (1)

Puerto Ricans Living in U.S. ( I ,000s) (2)

’3 Puerto Ricans in U.S. (2)/ ( 1 +2)

2,203 2,339 2.659 2,889 3.294

226 617 810 93 I 1.155

,093 ,209 ,233 ,244 ,260

(3)

Sources: “Puerto Ricans in the U.S.” (1950, 1960, 1970); “Characteristics of the Population: Puerto Rico” (1950, 1960. 1970, 1980): “General Social and Economic Characteristics” (1980, table 167); Junta de Planificacion, “Informe economico al gobernador,” various eds.; Centro de Estudios de Puertoriquenos (1979, table 9, p. 187). Note: The number of Puerto Ricans living in the United States in 1987 IS estimated by summing annual net passenger flows from 198 1 to 1987 and adding half the 1980 flow. The flows are from Falcon (1990).

191

Effect of the U.S.-Level Minimum Wage on Puerto Rico 3.54

years Fig. 6.2 Percentage of Puerto Ricans who migrated to the United States

inducing effect of the minimum through disemploying workers and the emigration-reducing effect of the minimum through raising wages roughly balance out. Another interpretation is that the minimum has a nonnegligible effect on migration that cannot be reliably detected in the fluctuating timeseries data. While econometric manipulations that smooth the data might yield better estimates of the effect of the minimum,Il we believe that there is no convincing story about the volume of migration in the time series and turn instead to micro survey data on how the minimum may have influenced the characteristics of migrants. 6.3.2

Premigration Employment Experience of Migrants

Economic analysis predicts that workers displaced by the minimum should be more likely to leave the island than others and, given the higher capital/ labor ratios in the United States, should have greater success finding jobs in the United States than in Puerto Rico. Consistent with this, Puerto Rican male and female migrants have much higher employment-population rates than Puerto Rican residents with the same years of schooling (cols. 1-2 of table 6.6). However, because these data do not tell us how migrants fared when they were in Puerto Rico, they leave open the possibility that migrants actually came largely from the ranks of the employed. To determine the employment experience of migrants prior to emigration, we turned to the 1982-83 “Encuesta de migracion” of the Junta de Planificacion de Puerto Rico (1984). The encuesta asked migrants at the international airport their employment status during the previous three months. Because migrants are younger than typ11. Santiago (1990) estimates a complex model with monthly flows that suggests that the minimum raised migration, but he has no other control variables, and his raw correlations give the opposite result. We interpret this as indicating the difficulty of making inferences from these data.

192 Table 6.6

Alida J. Castillo-Freeman and Richard B. Freeman Percentage Employed of Puerto Rico-born Men and Women, by Year of Schooling and Migrant Status, 1980 and 1983 Males

1980 Years of Schooling

5 6 7-9 10-1 I 12 13-15 16+ All

1983 Migrants to U.S. 3 Months before Migration

In Puerto Rico

In U.S.

In Puerto Rico (adjusted for age)

39 49 55 67 71 87 57

64 68 72 81 84 85 71

32 40 52 60 71 79 55

30 33 36 38 48 50 38

7

5 11 5 21 32 55 20

Females

5 6 7-9 1c-I 1 12 13-15 16f All

10 17 21 38 55 72 28

22 26 31 46 56 61 33

14

13 34 46 67 29

Sources: 1980, tabulated from the Public Use Files of the U.S. and Puerto Rico Censuses. 1983, tabulated from the Puerto Rican CPS. 1983 migrants, tabulated from the Junta de Planificacion de Puerto Rico (October 1984), tables 12, 14, and A7. Note: The'1983 rates are weighted by the age distribution of migrants from the Junta de Planificacion de Puerto Rico (October 1984), using table A-5 for weights. The same age weights were used for both sexes.

ical island residents, we contrasted migrant employment rates with a weighted average of age-specific employment rates of residents, using the age distribution of migrants as weights. Further, to avoid problems due to changes in island employment over time or between surveys and Censuses,12 we contrasted these rates with 1983 employment rates from the Puerto Rican CPS. The results of these comparisons show that the percentage of migrants employed is markedly below the percentage of island residents employed in the same education-sex group (cols. 3 and 4 of table 6.6). In the aggregate, 38 percent of male migrants were employed, compared to 55 percent of male residents. Given the numbers of residents and migrants, this implies that approximately 4.4 percent of employed males and 7.5 percent of jobless males 12. The U.S. Department of Commerce (1979, 2:601) noted substantial differences between Census and CPS estimates of unemployment rates. The revisions in annual employment and population estimates after the 1980 Census benchmark was introduced also imply inconsistencies between the two sources.

193

Effect of the U.S.-Level Minimum Wage on Puerto Rico

migrated. Among women, the pattern is weaker, presumably because women are “tied movers” (Mincer 1976): 20 percent of migrants worked, compared to 29 percent of residents, giving rates of migration of 2.6 percent for the employed and 3.7 percent for the not employed.I3 The encuesra also asked why individuals were migrating. Consistent with the notion that lack of work induced considerable migration, the vast majority of migrants exclusive of military personnel and students said that they were migrating to work (33 percent) or to search for work (47 percent) (Junta de Planificacion 1984, table A- 18). Less-educated and younger persons were especially likely to be moving to search for work.

6.3.3 Years of Schooling of Migrants Ramos (in this volume) has shown that male Puerto Rican migrants to the United States had fewer years of schooling than similarly aged nonmigrants in 1980, which he attributes to the higher payoff to education in Puerto Rico. Since the minimum will disproportionately reduce the employment of lesseducated workers, this pattern is also consistent with the minimum wage influencing the characteristics of migrants. To try to identify a separate minimum wage effect, we compared the education of migrants to that of residents in Puerto Rico in earlier Censuses. If migrants had less schooling than nonmigrants in periods when the minimum was weak, we would be loathe to ascribe much of the 1980 census pattern to the minimum wage. By contrast, evidence that the migrant-nonmigrant schooling gap developed when the minimum increased would lend support to the hypothesis that the minimum influenced the educational composition of migrants. Published Census data on the median years of schooling of Puerto Rican migrants and of residents on the island aged 25 years and older in 1950, 1960, and 197014showed the opposite pattern of relative attainment to that for 1980. In 1950, when relatively few Puerto Ricans had migrated to the United States, migrants had 7.7 years of school completed, compared to 3.7 years for persons on the island. In 1960, migrants completed 7.9 years of schooling, nonmigrants 4 . 6 years. In 1970, migrants had 8.4 years and nonmigrants 6.9 years completed-a smaller but still substantial gap that implies that as late as 1970 the bulk of Puerto Rican migrants were more educated than nonmigrants. Not until the 1980 Census do migrants have less education than non13. The encuesfa estimated that there were 59,000 male and 37,000 female migrants aged 1664 in 1983. The Puerto Rican CPS suggests that there were approximately 501 ,000employed men and 479,000 not employed men on the island aged 16-64. There were 294,000 employed women and 806,000 not employed women. We used these data to estimate the rate of migration of employed and not employed men and women from Puerto Rico, unadjusted for the age of migrants. 14. The figures for the Puerto Rico born living in the United States are taken from the special Census volume on “Puerto Ricans Living in the U.S.” in 1950, 1960, and 1970. The figures for residents in Puerto Rico come from the Puerto Rico volumes of the Census. The Census canceled the 1980 volume on Puerto Ricans in the United States.

194

Alida J. Castillo-Freeman and Richard B. Freeman

migrants, consistent with the contention that movement toward the U.S. in the 1970s altered the selectivity of migration. To explore the 1970-80 change in migrant-nonmigrant education differences further, we estimated equations for years of schooling from the Public Use Files of the 1970 and 198015 U.S. and Puerto Rican Censuses and the June 1988 U.S. and Puerto Rican CPSs. Building on Ramos’s analysis of the 1980 Census, we pooled the records of 16-64-year-old persons born in Puerto Rico and residing in the United States with those of Puerto Ricans living in Puerto Rico. We regressed years of school completed on dummy variables for age (to control for the upward trend in schooling), migrant status, and (in 1970 and 1980) a “recent migrant” dummy that takes the value of one if the person migrated to the United States in the preceding five years. The estimated coefficients and standard errors on the migrant dummy variables are given in table 6.7. The coefficient in the 1980 regression for males corroborates Ramos’s finding that migrant men had fewer years of schooling than resident of Puerto Rico. The coefficient in the 1988 regression is smaller but still negative. By contrast, the coefficient in the 1970 regressions for males show migrants having fewer years of schooling than island residents. The regressions for females tell a similar story: migrants had less education than nonmigrants in 1980 and 1988 but more education than nonmigrants in 1970. In 1980 and 1988, moreover, migrant-nonmigrant educational differentials were greater for women than for men, possibly because the minimum affects the employment of female workers more than that of male workers owing to the lower wages of women. The regressions that distinguish between recent and earlier migrants tell a more complex story. The estimated coefficients on recent migrants are negative in the 1970 Census as well as in the 1980 Census and are absolutely larger in the 1970 Census as well. This implies that 1965-70 migrants were disproportionately drawn from less-educated Puerto Ricans, whereas those who came earlier were drawn from the more educated. If the 1965-70 migration of the less educated was due to the minimum wage, we would expect a sharp rise in the minimum in 1965-70. Table 6.1 above shows such a rise in the coverage-weighted minimum in Puerto Rico due to the 1965 Amendment to the Fair Labor Standards Act. While this does not prove that the increased migration of the less educated in 1965-70 is due to the minimum, it is consistent with the minimum wage contributing to the change. 15. For Puerto Ricans in the United States in 1970, we used the 1/100 sample of the U.S. Census; for Puerto Ricans in Puerto Rico in 1970, the l/lOO sample of the Puerto Rican Census was used; for Puerto Ricans in the United States in 1980, we used a sample derived by combining the lil00 sample with the 5/100 sample from selected states with many Puerto Ricans. The sample represents 90 percent of the total Puerto Rico born population in the United States. For details, see Ramos (in this volume). The sample of Puerto Ricans residing in Puerto Rico is extracted from the 5/1OO Census for Puerto Rico. We included all out-of-school persons not in the military in the age bracket 16-64.

195

Effect of the U.S.-Level Minimum Wage on Puerto Rico

Table 6.7

Estimated Coefficients (standard errors) for Differences in Years of Schooling Completed by Puerto Rican Migrants to the United States and Puerto Ricans on the Island, 1970-88 Males 1988186

Migrants

- .35 (.25)

1980

1970

- .53 (45)

(. 10)

Earlier migrants Dummies for age

N

1970

- .34 (. 14) - .55

- .66 ( .24) .32 (. 10)

.18

Recent migrants

R2

1980

(.W

9 .06 5,442

9 .07 56,809

9 .08 18,850

9 .07 55,809

9 .08 18,850

- .65 .05 - .70

.57 (.22) .37 (.09)

9 .I5 63,561

9 .I2 20,921

Females Migrants

- .76 ( ,201

- .70

(.W

.23 (.W

Recent migrants Earlier migrants Dummies for age R‘ N

9 .13 6,451

9 .15 63,561

9 .12 20,921

Sources: 1970 and 1980 tabulated from the Public Use Files of the U.S. and Puerto Rican Censuses. 1988 tabulated from the June 1988 Puerto Rican CPS, with migrants from June 1988 and 1986 from the U S . CPS.

We examined the changed selectivity of migrants by education between the 1970 and the 1980 Censuses in one additional way. We used the two Censuses to calculate the average years of schooling of the Puerto Rico born in Puerto Rico and in the United States in specified “pseudocohorts”-persons aged y years in 1970 and y + 10 years in 1980. Figure 6 . 3 displays the results of this analysis in terms of the change in years of schooling completed by pseudocohorts on the island and in the United States. It shows greater increases in years completed for cohorts in Puerto Rico than for cohorts in the United States, implying a substantial sorting of the Puerto Rico born in a given cohort, with the less educated moving to the United States and the more educated returning to Puerto Rico. This is what one would expect if the movement toward the U.S.-level minimum in the 1970s made it more difficult for the less educated to find work on the island. An alternative explanation of why migrants had less schooling than nonmigrants in the 1980s but not in the 1970s is that the rewards to education

196

Alida J. Castillo-Freeman and Richard B. Freeman

26-30 3-35

36-40

41-45

46-50

1970 age cohorts

Migrants

0 Living in Puerto Rico

Fig. 6.3 Change in mean years of schooling completed by cohort and residence, Puerto Rico-born men, 1970-80

increased from 1970 to 1980 in Puerto Rico relative to the United States. We examined this possibility for male workers by estimating log earnings equations for men in Puerto Rico and for migrants to the United States in 1970 and 1980 for both annual and hourly eamings.I6 Our model included years of potential work experience dummy variables and a linear years of schooling term (all the regressions are in app. B). For Puerto Rico-born men in the United States, the regressions for weekly earnings gave a coefficient on schooling of 0.048 in 1970 and 0.036 in 1980, for a 0.012 drop in the effect of schooling on earnings. By contrast, the regressions for Puerto Ricans on the island gave a coefficient on schooling of 0.071 in 1970 and 0.077 in 1980, for a 0.006 rise in the effect of schooling on earnings. These results suggest that changes in the returns to schooling contributed to the changed educational selectivity of migrants in the decade and thus buttress Ramos’s story. The regressions for yearly earnings showed a 0.010 fall in the effect of schooling on the earnings of migrants, compared to a 0.018 rise in the effect of schooling on the island. Since yearly and weekly earnings differ by weeks worked, the implication is that the effect of schooling on weeks worked increased by just 0.002 among migrants but by 0.012 among island residents. This is consistent with the notion that the minimum wage altered the selectivity of migrants by reducing the employment of the less educated more than of the educated. Thus, there is evidence that both joblessness due to the increased minimum wage and edu16. The 1970 Censuses did not ask for usual hours worked, so our analysis is based on information on hours worked last week. In addition, the Public Use tapes present time-worked data in categories rather than in actual values; we used category midoointg in our analysis.

197

Effect of the U.S.-Level Minimum Wage on Puerto Rico

cational pay differentials contributed to the greater immigration of the less educated to the United States in the 1970s.

6.3.4 Migration, Education, and Language Another way to examine how the minimum wage might affect migration is to contrast the likelihood of migration among workers more or less likely to be affected by the minimum. We have done this using education as an indicator of being affected by the minimum. For six education groups, we used 1980 Puerto Rican Census data to estimate the proportion of workers likely to be affected by the minimum had they resided in Puerto Rico; we also used pooled U.S. and Puerto Rican 1980 Census data to estimate the proportion of the Puerto Rico born residing in the United States. There is no easy way to assess the proportion of workers affected by the minimum (PMIN). The distribution of earnings from the 1980 Puerto Rican Census shows the proportion earning the minimum or less, but not the number of persons who lost their jobs or who migrated as a result of the minimum, and thus understates the proportion affected by the minimum. The understatement will, moreover, be greater for groups whose employment is most reduced and/or who migrate in large numbers, biasing the estimates against finding a minimum wage effect on those outcomes. Still, as the Census earnings distribution offers the only direct indicator of PMIN, we used it to compute the proportions of Puerto Ricans in different education groups with earnings at or below the $2.90 U.S. minimum. Column 1 of table 6.8 shows that these proportions decline with education steeply after high school. To estimate the proportion of the Puerto Rico born who were migrants in our six educatibn groups, we regressed a zero-one variable for U.S. residence on dummy variables for years of schooling groups and for ten age groups in our pooled U.S. and Puerto Rican census file. The coefficients on the schooling dummies, given in columns 2 and 3 of table 6.8, reveal a nonlinear relation between years of schooling and migration that was hidden in the comparison of mean education of migrants and nonmigrants: the probability of migration is smallest for the least educated and for the most educated and highest for those who graduate grade school but do not go beyond high school. The limited migration of the least educated-who are the most affected by the minimum, who have the lowest employment rates, and who have the smallest rewards from schooling-appears to be due to their lack of spoken English (and/or lack of information related to the possession of that language skill). Of men with zero to six years of schooling, only 13 percent of those in Puerto Rico could speak English, compared to 31 percent of those 17. As an alternative estimate of PMIN, we assumed that the 1969 earnings distribution given in the 1970 Census would have held in 1979 absent the minimum wage, inflated 1969 earnings by the rate of growth of average hourly earnings in manufacturing, and estimated the proportion of workers by education in the inflated distribution likely to be paid the 1979 U.S. minimum or less. The results were sufficiently similar to those in the text that we forgo presenting them.

AIida J. Castillo-Freeman and Richard B. Freeman

198 Table 6.8

Estimated Proportions of Workers Subject to the Minimum in Puerto Rico and Migrant Status Proportion Paid Minimum or Below or Not Employed

Education

Proportion of Migrants, 1980

Not Adjusted

Adjusted for Language

~~

0-6 7-8 9-1 I 12 13-15 16 +

.67 ,523 .54

.40 .25 .09

.09 .I6 .I7 .09 .08 .00

- .03 - .14

.I1 .I9 .I9 .I1 .07 .01

.04 .02 - .03 -.I4 - .25 - .37

.00

.oo - .23

- .36

Females 0-6 7- 8 9-1 1 12 13-1 5 16f

.73 .71 .64 .55 .38 .I4

Source: Column 1 estimated from regressions in appendix C, using 1980 Puerto Rican Census. Column 2 estimated from regressions in appendix C, using 1980 Puerto Rican and U.S.Censuses. Column 3 estimated from regressions in appendix C, using 1980 Puerto Rican and U.S. Censuses, with inclusion of dummy variable for English-speaking proficiency.

with seven to eight years of schooling, 59 percent of high school graduates, and 92 p'ercent of college graduates. Even among migrants there was a marked difference in the ability to speak English: 47 percent of migrants with zero to six years of schooling reported speaking English well, compared to 58 percent of migrants with seven to eight years of schooling, 96 percent of high school graduates, and 99 percent of college graduates.I8 Since persons with the least English-speaking ability are unlikely to migrate, language skill is an omitted variable associated with education that depresses the migration of the less educated. Adding dummy variables for ability to speak English to our migrant regression turns the inverse U-shaped education-migration relation into a monotonic relation in column 3 of table 6.8. As migrants improve their English, the estimated effect of English speaking on migration to the United States will be biased upward, which will in turn bias the estimated coefficients on education groups in the migration regression, but this bias is unlikely to 18. These calculations are based on questions on English-speaking ability asked in both the U.S. and the Puerto Rican Censuses. The U.S. Census had a fourfold categorization: speaking English very well, well, not well, and not at all. The Puerto Rican Census had a threefold categorization: speaking English easily, with difficulty, and not at all. We collapsed the speaking very well and the speaking well categorizations in the U.S. Census so that we had comparable threeway groupings.

199

Effect of the U.S.-Level Minimum Wage on Puerto Rico

account for the elimination of the inverse U-shaped education-migration relation on addition of the English-speaking variable to our regression. l9 The correlations between PMIN and the adjusted or unadjusted proportion migrating are high. For men, the correlations are 0.99 with the adjusted proportion migrating and 0.77 with the unadjusted proportion. For women, the correlations are .99 with the proportion of migrants corrected for language and .84 with the proportion of migrants absent the language adjustment. Still, since those most affected by the minimum have lower wages, it is possible that the correlation between PMIN and the proportion of migrants by education group is due not to the minimum but to differences in earnings by education group. To assess this possibility, we calculated employment rates and earnings by schooling group for workers on the island and for migrants to the United States. Figure 6.4 shows that in differences in weekly earnings between migrants and nonmigrants fell sharply with education, consistent with the earnings differential explanation of the greater migration of the less educated. It also shows that In differences in employment rates between migrants and nonmigrants fell with education, consistent with the minimum wage/joblessness interpretation of the greater migration of the less educated. With both factors working in the same direction, it is difficult to assess their relative importance. In any case, the result is that persons with the skills likely to make them affected by the minimum dominated the 1970s migrant

6.4 Effects of Migration on the Puerto Rican Job Market We argue next that, regardless of the causal effect of the minimum wage on Puerto Rican migration, migration has been a key “safety valve” in the Puerto Rican job market without which it would have been virtually impossible to impose the U.S .-level minimum on the island. Migration reduced joblessness, raised the average skills of workers and the marginal productivity of labor, and contributed to the growth of real earnings on the island. 19. In the underlying data, the zero to six years of schooling group has a lower proportion in the United States than the seven to eight years of schooling group, but it has a higher proportion in the United States for those who speak English well, who speak English somewhat, and for those who do not speak English. To see if this pattern might be due to those in the United States learning English, we estimated migration status equations comparable to those in app. C for recent migrants, who have less time to improve their English proficiency. In our regressions (which eliminated pre-1975 migrants from the sample), we obtained results similar to those for all migrants: additional of English-speaking dummies explained the low migration rate among the zero to six years of schooling group. 20. If the minimum wage induced the less qualified to migrate, we would expect migrants to do especially poorly in the U.S. labor market conditional on education. Ramos has shown that, in 1980, migrant males, particularly the most recent migrants, had lower earnings than otherwise comparable to U.S.-born F’uerto Ricans. We estimated log earnings equations for apooled sample of U.S.-born Puerto Ricans and migrants in 1970 and found a smaller effect for all migrants in 1970 than in 1980, but we found equally large gaps between the earnings of recent migrants and those of U.S.-born F’uerto Ricans in 1970 as in 1980.

200

Alida J. Castillo-Freeman and Richard B. Freeman

n ~

-

0-6

I

7-13

9-11

~

12

13-15

16+

grade

Men log dif mig/pr epop

log dif mig/pr earnings

-

I

0-6

1 9-11

~

7-8

n 12

13-15

16+

gr de

women rn log dif mig/pr epop 0 log dif mig/pr earnings Fig. 6.4 Log difference of employment to population and earnings by education group, migrants to Puerto Ricans 6.4.1

Migration and Joblessness in the Presence of the Minimum

What might the employment-population rate and unemployment rate have been in Puerto Rico with a U.S.-level minimum wage but no outlet for migration? Consider first the potential labor market effects of imposing the U.S. minimum in the 1970s and 1980s if no Puerto Rican migrated to the United States after 1974. In this case, there would have been approximately 232,000 additional persons of working age on the island,21increasing the working-age pop21. This estimate is the sum of approximately 90,000 net migrants from 1975 through early 1980 (based on the 1980 Census of Population) and 200.000 net migrants from passenger traffic

201

Effect of the U.S.-Level Minimum Wage on Puerto Rico

ulation by about 10 percent. If the number of jobs for the less skilled was fixed because the minimum disallowed employment-creating movements down the demand curve, all migrants whose characteristics would have earned them less than the minimum as well as those with characteristics that would have made them unemployed would have been jobless on the island. To find out how many migrants would fit in this set, we estimated the effect of age and schooling on the probability that someone in Puerto Rico would have been jobless or paid the minimum or less and applied the resultant equation to the characteristics of recent migrants. The calculation indicates that 77 percent of migrants would have been jobless on the island and just 23 percent employed.22 Hence, 199,000 of the 232,000 “return migrants” would have lacked jobs, and just 53,000 would have found work. Assuming that the “return migrants” participated in the labor force at the same rate as they did in the United States (56 percent), the number of additional unemployed workers would be 69,000. Because the island employment-population rate was already low (.359), however, the addition of these workers would have reduced the island employment-population rate by only .013, or 3 percent. But, because there would have been more unemployed than employed return migrants, the unemployment rate would have risen by 4 percentage points, from 17 to 21 percent. Similar calculations assuming that return migrants would have found employment at the 31 percent rate of employment of migrants in the three months preceding migration given in the “Encuesta de migracion” suggest a substantial 3 percentage point rise in the unemployment rate, although only a 0.5 percentage point fall in the employment-population rate on the island. 6.4.2

Long-Run Migration

How might the Puerto Rican labor market have fared in the entire postWorld War I1 era if Puerto Ricans had never had the option of migrating to the United States? For simplicity, we answer this question assuming that all Puerto Rican migrants to the United States had remained residents of the island and that other factors that determine economic well-being were unchanged on the island. As other factors would also have changed, our exercise should be viewed not as a prediction of what might actually have happened but rather as a way to demonstrate the magnitude of the economic effects of immigration on the island. A full analysis of what might have happened absent migration requires a complete model of the Puerto Rican economy that lies beyond the scope of this study. Absent migration from Puerto Rico to the United States, the one-third of data. This gives 290,000 migrants. On the basis of 1980 Census data, approximately 80 percent will be between age 16 and age 64, giving the 232,000 in the text. 22. We base this estimate on a two-stage analysis. First, we estimated separately for sex the relation between being employed at above-minimum wages in Puerto Rico in the 1980 Census and a set of education and age dummy variables. Then we predicted the proportion of recent migrants who would have been employed on the island, given that equation. The equations are given in app. C.

202

Alida J. Castillo-Freeman and Richard B. Freeman

the Puerto Rico born of working age living in the United States would be on the island, raising the working-age population by about 50 percent, or 0.40 In points. Assuming, as in our earlier calculation for the effects of migration after 1974, that 70-80 percent of these return migrants were jobless in the presence of a U.S .-level minimum wage even if island productivity remained unaffected (i.e., if capital also increased by one-third or so) and that 56 percent would have been in the work force, the rate of unemployment would have risen to 30-35 percent. This effect is so large as to call into question the assumption that the U.S.-level minimum could have been applied to the island absent migration. More likely than not, the result would have been similar to that in 1935, when Congress quickly rescinded the U.S.-level minimum and chose instead the industry-council mode of setting minima. Put differently, massive migration to the United States was a prerequisite for applying the U.S. minimum to the island in the 1970s. What might have been the level of wages on the island absent migration and absent the minimum wage? If nothing else changed, labor supply would have been 0.40 In points higher in 1980 than it was. Given an elasticity of substitution of u, a fixed capital stock, labor’s share of output of a, and market clearing, a 0.40 In point increase in the work force would reduce pay by [u/(1 - a)].40 In points.23At any plausible levels of a and u, the change in labor supply would have a devastating effect on real wages. For instance, with a Cobb-Douglas production function and labor’s share of two thirds of GNP,24a 0.40 In point increase in labor supply would reduce average earnings by 1.2 In points, or to 30 percent of their current level. Even with a relatively small elasticity of substitution of, say, 0.15, comparable to our estimated elasticity of employment to the minimum wage, wages on the island would have been cut by 0.18 In points, to 84 percent of their current level. Of course, the economy would have made other adjustments: the return to capital would have risen, inducing greater investment that would have partially restored the capital/labor ratio; fertility might have fallen; investments in human capital might have risen; and so on. But the first-order effect of massive return migration would clearly fall on real wages. Given the lower education of migrants, moreover, there would be an additional reduction in earnings and productivity due to the reduced qualifications of the work force. If all the Puerto Rico born had remained on the island, the average education of the Puerto Rican work force would be roughly 0.3 years 23. With market clearing, the elasticity of substitution (u) is (K‘ - L’) / ( w ’ - r ‘ ) . so that, with fixed capital stock, L’ = - u ( w ’ - r’). The factor price frontier equation is aw’ + (1 - a)r’ = 0, where the price of output is the numeraire. Substituting for r ’ , we obtain L = -a{w’ [a/(1 - a)]w’}, which simplifies toL = uw‘/(l - a), as in the text. 24. In 1980, compensation of employees was $7,202 million, national income was $9,722 million, and GNP was $11,031 million. Thus, labor’s share of national income was 0.74, or approximately three-quarters, while its share of GNP was .65, or approximately two-thirds (see Junta de Planificacion, “lnforme economico al gobernador,” 1981). In the 1950s. labor’s share of output was smaller owing to the greater importance of agriculture.

+

203

Effect of the U.S.-Level Minimum Wage on Puerto Rico

lower, implying that wages would have been some 2 percent less owing to educational qualifications. In addition, if we follow Ramos and interpret the within-education differential in earnings between migrants and U.S.-born Puerto Ricans as indicating the lower qualifications of migrants, we estimate that the earnings power of the return migrants would be 13 percent less than that of nonmigrants, reducing the productivity and earnings of the average Puerto Rican by 4 percent. All told, even with a 0.15 elasticity of substitution, we estimate that, absent migration (and further capital investments), real earnings in Puerto Rico would conservatively be on the order of 25 percent lower in 1988 than in fact they were. The implication is that, at the minimum, onequarter of the long-term trend in real earnings on the island (an increase of 174 percent from 1951 to 1988) can be attributed to migration to the United States. 6.5

Conclusion

This paper has shown that the imposition of the U.S.-level minimum wage to Puerto Rico distorted the Puerto Rican earnings distribution, substantially reduced employment on the island, reallocated labor across industries, and affected the characteristics of migrants to the United States. In addition, we argued that, absent migration of the less skilled, imposition of a U.S.-level minimum on the island would have raised unemployment so much as to call into question the viability of such a policy. Thus, migration was a prerequisite for the high minimum wage. Our estimates indicate further that migration was a major contributor to the growth of real earnings on the island. All told, the massive migration from Puerto Rico demonstrates both the interplay between domestic labor market policy-in this case, imposition of a minimum wage with a bite-and migration and the potential contribution of migration to the growth of real wages in a source economy.

Appendix A Documentation for the Puerto Rican Minimum Wage Time-SeriesData Set Minimum Wage-Related Variables Average minimum wages (AVEMIN) is a weighted average of forty-four industry minimums (thirty-seven three-digit manufacturing and seven one-digit industries). The data were gathered from the individual U.S. Department of Labor reports (the “Minimum Wage Industry Studies”) that record the industry minimums in the years when industry committees set minima. The reports usually give minima for very detailed occupations. To arrive at a single minimum wage for each industry, the data had to be amalgamated. Because em-

204

Alida J. Castillo-Freeman and Richard B. Freeman

ployment by occupation was unavailable, we took a simple average of the occupational minimum. Average coverage (AVECOV) is a weighted average of coverage for the eight one-digit industries, based on table 1 of U.S. Department of Commerce (1979, 2:634). All three-digit manufacturing industries are covered by the same figure. The Department of Commerce table gives the number of wage and salary workers covered by the changes in the minimum wage law (in 1966 and 1974). This number was divided by total employment in each industry to determine the effect of the minimum on the entire economy. Since the law changed in the middle of 1974, the coverage figure for that year is the average between the 1973 and the 1975 numbers. We also created average coverage excluding agriculture and government (COVAG), for wage and salary workers (AVENCOV), and for wage and salary workers excluding agriculture and government (NCOVAG). Economy-wide coverage (COVT) is based on coverage figures for 1962, 1964, 1965, 1966, 1969, 1970, 1971, 1972, 1975, and 1976 from U.S. Department of Labor, Employment Standards Administration (1977). For 1976, the figures are the same as in table 1 of U.S. Department of Commerce (1979, 2:634). We divided the figures by total employment to obtain the coverage number. Average wage (AVEWAG) is the weighted average of the forty-four industry average hourly earnings. The thirty-seven detailed three-digit manufacturing earnings are from the Departamento del Trabajo’s “Census of Manufacturing Industries” (1956-87). The Census was collected every October through 1981 and then not again until March 1983, so there are no figures for 1982. In the time-series analysis, the 1982 figure is the average between 1981 and 1983. For the years 1950-55, the Census was not conducted, so we applied the change in one-digit manufacturing hourly earnings from each year to 1956 to the 1956 three-digit earnings, on the assumption that earnings in each detailed sector changed at the same rate as the average in manufacturing. The one-digit industry data were obtained from the “Salario semanal mediano de 10s empleados asalariados por grupo industrial principal” (Departamento del Trabajo y Recursos Humanos). This source gives weekly earnings by month. To make hourly earnings, we divided the July weekly earnings by thirty-two hours. Average manufacturing wage (MFGWAG) comes from the Yearbook of Labour Statistics ( 1950-87). Kaitz minimum wage index (KAITZ) is the employment-weighted average of coverage times minimudhourly earnings: where a, is the share of employment in industry i, m, is the minimum in industry i, w, is average hourly earnings in industry i, and c, is the coverage in that industry. The index used the coverage, minimum, and hourly earnings figures described above. We also created a Kaitz index for wage and salary workers (NKAITZ).

205

Effect of the U.S.-Level Minimum Wage on Puerto Rico

The employment by industry numbers used in the weighting come from two sources. The individual manufacturing industry numbers are from the Departamento del Trabajo’s “Census of Manufacturing Industries” for the three-digit manufacturing industries (1956-87). To get the 1950-55 numbers, we took the ratio of employment in all manufacturing in each of these years to employment in all manufacturing in 1956 and multiplied this by the 1956 employment in the detailed industry. For the remaining seven one-digit industries, the employment numbers are from the Departamento del Trabajo’s “Seria historica del empleo, desempleo y grupo trabajador en Puerto Rico.” We used thirty-seven three-digit manufacturing industries: footwear, leather gloves, electrical, women’s and children’s clothing, children’s outerwear, corsets and brassieres, men’s and boy’s clothing, leather handbags, women’s outerwear, miscellaneous apparel, miscellaneous fabricated textiles, toys and athletic goods, jewels and jewelry, costume jewelry, office and art supplies, alcoholic beverages, cigars, tobacco, drugs, petroleum, chemicals, food, household furniture, other furniture, sawmills, paper and allied products, cement, cut stone and asbestos, portland cement and pottery, glass, sugar, textile mill products, plastics, rubber, footwear, professional instruments, and machinery and transportation equipment. We used seven one-digit industries: transportation, construction, services, trades, finance, agriculture, and public administration.

Macroeconomic Variables Puerto Rican deflator (PRDEF) is from the Junta de Planificacion’s “Informe economico a1 gobernador,” 1954 base year. Puerto Rican GNP (PRGNP)is from the Junta de Planificacion’s “Informe economica al gobernador,” 1954 constant dollars. These series Puerto Rican employment to population ratio (PREFQP and PREPOPF) come from the monthly household surveys done in Puerto Rico. The PREFQP series combines two series: 1950-63 uses the fiscal year fourteen years old and over, reported in the Departamento del Trabajo’s “Seria historica del empleo, desempleo y grupo trabajador en Puerto Rico” (table IV); 196487 uses the calendar year sixteen years old and over numbers from the Departmento del Trabajo’s “Empleo y disemployeo en Puerto Rico” (table 17), adjusted to the 1980 Census benchmark. The PREPOPF series uses the fiscal year fourteen years old and over numbers for the entire time period, also adjusted to a 1980 Census benchmark. Puerto Rican unemployment rate (PRUNEMP) is from the Departamento del Trabajo’s “Seria.historica del empleo, desempleo y grupo trabajador en Puerto Rico.” This series comes from the monthly household surveys done in Puerto Rico similar to our Current Population Survey. US.GNP (USGNP)is from the Economic Report of the President. Table 6A. 1 gives figures for both the minimum wage-related variables and the macroeconomic variables.

Data in the Time-Series Analysis

Table 6A.l

Minimum Wage-related Variables Year

1950 1951 1952 1953 1954 I955 I956 1957 1958 1959 I960 1961 I962 1963 1964 1965

AVEMIN

,198 ,209 ,225 .31 I ,313 ,369 .447 .488 ,555 .588 .616 .608 .707 .723 .809 ,834

AVEWAGE

,398 ,410 ,421 .480 ,508 .547 .601 ,685 ,716 .789 .840 .875 .933 1.036 1.097 1.176

KAITZ

,155 .I64 ,180 ,229 ,211 ,231 ,257 ,251 .258 ,266 ,268 ,251

.270 ,255 ,274 ,271

AVECOV

COVT

.201 .207 .226 ,231 ,224 ,236 ,245 .244 ,238 .260 ,270 ,269 ,279 ,279 ,294 ,302

.29 .29 .29 .29 .29 .29 .29 .29 .29 .29 .29 .29 .29 .29 .31 .31

Macroeconomic Variables MFGWAGE

.43 .45 .48 SO .52 .57 .64 .76 .83 .87 .92 .99 1.06 1.13 1.18 1.23

PRDEF

.859 .881

.953 ,970 1 .Ooo I .003 1.011 1.035 1.089 1.110 1.138 I . 173 1.216 1.247 I .298 1.327

PREPOP

PREPOPF

PRGNP

PRUNEMP

USGNP

.470 ,449 .434 ,428 ,415 ,419 ,412 .412 ,397 .394 ,403 ,397 ,385 ,395 ,396 ,401

.470 .449 ,434 ,428 ,415 .419 ,412 ,412 ,397 ,394 ,403 ,397 ,385 ,395 ,396 ,401

878.7 925.0 1,015.9

,154 ,160

1,203.7 1,328.2 1,380.0 1,435.3 1,416.2 1,494.9 1,525.6 1,551. I 1,539.2 1,629. I 1,665.3 1,708.7 1,799.4 1,873.3 1,973.3 2,087.6

1,081.3

1,104.4 1,138.5 1,185.1 1,221.8 1,258.4 1,363.6 1,473.2 1,562.8 1,683.9 1,820.7 1,916.8 2,083.0

.I48 .I45 .I53 ,132 .I33 ,128 ,142 .133 .I18 .127 .I28 ,110

. I 12 ,117

Year 1

2 3 4 5 6 7 8 9 10

11 12 13 14 15

16

Dummy 1974 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 0

1966 1967 1968 1969 1970 1971 1972 1973 I974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987

.854 ,971 1.104 1.149 1.209 1.224 1.257 1.262 1.681 1.871 2.034 2.198 2.509 2.768 2.997 3.264 3.305 3.350 3.350 3.350 3.350 3.350

1.288 1.371 1.512 1.667 1.856 1.990 2.144 2.281 2.452 2.562 2.681 3.023 3.323 3.589 3.883 4.181 4.318 4.456 4.498 4.565 4.725 4.879

.325 .365 ,381 ,360 .347 .331 .316 ,304 ,381 ,434 ,442 ,435 ,456 ,468 .461 ,467 ,461 ,454 .449 ,440 .426 .409

.444 .448 .455 ,455 ,458 ,457 ,449 .452 .544 ,594 338 S90 .596 ,594 ,589 .587 .585 .583 ,586 ,579 ,581 ,582

.44 .44 .44 .49 .47 .47 .47 .47 .60 .66 .64 .64 .64 .64

.64 .64 .64 .64 .64 .64 .64 .64

1.29 1.39

1.55 1.65 1.76 1.87 2.00 2.13 2.32 2.56 2.78 3.02 3.36 3.69 4.02 4.39 4.64 4.83 5.02 5.19 5.31 5.33

1.358 1.421 1.500 1.552 1.616 1.708 1.780 1.817 1.946 2.082 2.174 2.240 2.340 2.483 2.716 2.954 3.175 3.321 3.461 3.548 3.697 3.787

.399 ,399 .403 .399 ,428 .423 ,423 ,421 ,405 .368 ,364 ,358 ,362 ,360 ,359 ,343 ,318 .321 ,334 ,331 ,351 ,369

,399 .399 ,403 ,399 .395 ,398 ,393 ,389 ,358 .336 ,334 ,364 ,360 .360 ,355 ,328 .314 .329 ,332 ,338 ,362 ,377

2,223.2 2,328.4 2,455.3 2,684.0 2,901.4 3,075.6 3,215.9 3,450.3 3,493.6 3,424.7 3,461.6 3,623.5 3,817.4 4,025 .O 4,076.8 4,127.0 3,976.5 3,894.8 4,048.4 4,172.8 4,281.6 4,496.7

,123 ,116 ,103 ,103 .I07 ,116 ,119 ,116 .I32 ,181

.I95 .I99 ,181

.170 ,171 ,199 ,228 ,234 ,207 ,218 ,189 ,168

2,208.3 2,271.4 2,365.6 2,423.3 2,416.2 2,484.8 2,608.5 2,744.1 2,729.3 2,695.0 2,826.7 2,958.6 3,115.2 3,192.4 3,187.1 3,248.8 3,166.0 3,279.1 3,501.4 3,607.5 3,713.3 3,819.6

17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38

0 0 0 0 0 0

0 0 1

2 3 4 5 6 7 8 9 10 11

12 13 14

208

Alida J. Castillo-Freeman and Richard B. Freeman

Appendix B Log Earnings Equations for Men in Puerto Rico and for Migrants to the United States Log Wage

1980

Grade complete Experience: &5

6-10

PR

.036 (.003)

(.001)

(W 16-20

- .15 ~03) - .09

21-25

- .05

11-15

(.W

26

+

Constant R’ N

(.W ...

1.43 .04 6,247

1980

1970

Mig.

- .51 (.03 - .40

Log Annual Earnings

.077

- .47 (.02) - .31 (.02) - .21 (.02) -.lo (.02) - .07 (.02)

Mig.

1970

PR

Mig.

PR

Mig.

PR

.07 I (.002)

,048 (.003)

,127 (.001)

,048 (.004)

,071 (.002)

-.39 (.04)

-.29 (.03) -.16 (.02) -.06 (.03 -.07 (.03)

-.96 (.05) -.60 (.04)

-.24 (.04) -.I1

(.W -.04 (.04)

-.96 (.02) -.57 (.02) -.37 (.02) - .20 (.02) -.13 (.02)

... .51

.12 26,193

-.52 (.07) -.26 (.05) -.08 (.05) - .04 (.05) -.04 (.05)

- .40

.69 .07 1,912

- .07 .16 8,694

~04) - .29

(.W - .I6 (.02) - .06 (.02) - .08

(.W ...

-.07 .16 8,694

8.76 .08 6,215

7.36 .23 26,743

Source: 1980 and 1970 Puerto Rican and U.S. Censuses. Nore: “Mig.* is Puerto Rican men who have migrated to the United States; “PR” is men in Puerto Rico. Standard errors are given in parentheses.

Appendix C The Effects of Age, Grade, and English Ability on Employment in Puerto Rico and Emigration to the United States Probability Not Employed or Paid 5$2.95 in Puerto Rico Controls

Male

Female

Age: 1620

21-25 2630 31-35 3640 41-45 4650 51-55

56-60 61-64 Grade: 0-6 7-8

9-1 1 12 13-15 16+ English ability: Well Not well Not at all Constant R2 N

Source: 1980 Puerto Rican and U.S. Censuses. Note: Standard errors are given in parentheses.

Probability Migrant

Probability Migrant Adjusted for English

Male

Male

Female

Female

210

Alida J. Castillo-Freeman and Richard B. Freeman

References Brown, Charles. 1988. Minimum Wage Laws: Are They Overrated? Journal of Economic Perspectives (Summer), pp. 133-43. Brown, Charles, Curtis Gilroy, and Andrew Kohen. 1982. Effect of the Minimum Wage on Employment and Unemployment: A Survey. National Bureau of Economic Research Working Paper no. 846. Castillo, Alida Josefina. 1983. Jobless in the Sun: A Study of the Effect of The Federal Minimum Wage on Employment in Puerto Rico. Undergraduate honors thesis, Harvard University. Centro de Estudios Puertoriquenos. 1979. Labor Migration under Capitalism: The Puerto Rican Experience. London: Monthly Review Press. Characteristics of the Population: Puerto Rico. 1950, 1960, 1970, 1980. U.S. Census of Population. Washington, D.C.: U S . Bureau of the Census. Departamento del Trabajo y Recursos Humanos. Various years. Census of Manufacturing Industries of Puerto Rico. San Juan, P.R. . Various years. Empleo y desempleo en Puerto Rico. San Juan, P.R. . Various years. Empleo y salario promedio por hora en las industias manufactureras de Puerto Rico. San Juan, P.R. . Various years. Salario semanal mediano de 10s empleados asalariados por grupo industrial principal. San Juan, P.R. . Various years. Seria historica del empleo, desempleo y grupo trabajador en Puerto Rico. San Juan, P.R. Economic Report of the President. Various years. Washington, D.C.: U.S. Government Printing Office. Falcon, Luis. 1990. Migration and Development: The Case of Puerto Rico. Working Paper no. 18. Washington, D.C.: Commission for the Study of International Migration and Cooperative Economic Development, February. General Social and Economic Characteristics. 1980. Washington, D.C.: U.S. Bureau of the Census. Junta de Planificacion de Puerto Rico. 1984. Encuesta de migration, 1982-83. San Juan, P.R. . Various years. Informe economico a1 gobernador. San Juan, P.R. Mincer, J. 1976. Unemployment Effects of Minimum Wages. Journal of Political Economy 84:87-104. Puerto Ricans in the U.S. 1950, 1960, 1970. Special report, U.S. Census of Population. Washington, D.C.: U.S. Bureau of the Census. Santiago, C. 1990. Minimum Wages and Migration in Puerto Rico. Paper presented at the Inter-University Consortium for Latino Research and the Social Science Research Council, Miami, 11 March. Statistical Abstract of the United States. Various years. Washington, D.C.: U S . Department of Commerce, Bureau of the Census. U.S. Department of Commerce. 1979. Economic Study of Puerto Rico. 2 vols. Washington, D.C.: U.S. Government Printing Office. U.S. Department of Labor. 1977. Estimated Number of Nonsupervisory Employees Subject to the Minimum Wage Provisions of the Fair Labor Standards Act in Puerto Rico, 1962-1976. Washington, D.C.: Employment Standards Administration, 16 May. . 1978. Handy Reference Guide to the Fair Labor Standards Act. WH Publication no. 1282. Washington, D.C. . Various years. Minimum Wage Industry Studies. Washington, D.C.: U.S. Department of Labor Library.

211

Effect of the U.S.-Level Minimum Wage on Puerto Rico

Welch, Finis. 1978. Minimum Wages: Issues and Evidence. Washington, D.C.: American Enterprise for Public Policy Research. Yearbook of Labour Statistics. Various years. Geneva: International Labour Organisation.

This Page Intentionally Left Blank

7

On the Labor Market Effects of Immigration and Trade George J. Borjas, Richard B. Freeman, and Lawrence F. Katz

In the 1980s, the wages and employment-population rate of less-skilled Americans, particularly young men, fell relative to those of more-skilled workers. The real earnings of 25-34-year-old male high school graduates and dropouts declined, continuing a trend begun in 1973 that breaks with the historic pattern of rising real earnings for less-skilled American men.l Two widely suggested causes of this change are the inflow of less-skilled immigrants, including illegal immigrants, and the trade deficit, notably the increase of imports in industries that hire low-skill workers. How much did trade and immigration alter the labor skill endowments of the United States in the 1980s? Hovi great a contribution did they make to the decline in the relative earnings of the less skilled in the 1980s? We present a conceptual and empirical analysis of these questions. Using demographic data from the 1980 Census and from the Current Population Surveys as well as detailed data on exports, imports, output, and employment for a large number of manufacturing industries, we first estimate the magnitude and educational composition of the labor supply embodied in trade flows and legal and illegal immigrations during the 1980s. We then calculate the percentage growth in the ratio of highly educated to less-educated labor in the

George J. Borjas is professor of economics at the University of California, San Diego, and a research associate of the National Bureau of Economic Research. Richard B. Freeman is professor of economics at Harvard University and director of the Labor Studies Program at the National Bureau of Economic Research. Lawrence F. Katz is professor of economics at Harvard University and a research associate at the National Bureau of Economic Research. The authors are grateful to Kevin M. Murphy for helpful discussions and for providing data from the March Current Population Surveys and to Zadia Feliciano for expert research assistance. 1. Studies documenting changes in the U.S. wage structure during the 1980s include Blackbum,Bloom, and Freeman (1990). Bound and Johnson (1989). Juhn, Murphy, and Pierce (1992), Katz and Murphy (1992), Katz and Revenga (1989). and Murphy and Welch (1992).

213

214

G . J. Borjas, R. B. Freeman, and L. F. Katz

United States that can be attributed to these flows. Finally, within the context of a standard model of labor market equilibrium, we assess the potential effect of changes in these skill endowments on earnings differentials by education. We report the following findings. 1. The annual increase in implicit labor supply due to the mid- and late 1980s trade deficit in manufactures was on the order of 1.5 percent for the economy as a whole and 6 percent for the manufacturing sector. These labor supply shifts exceed the percentage increase in labor supply due to the annual flow of immigrants, which increased labor supply only by about 0.3 percent per year. However, unlike trade deficits that change the implicit labor supply only annually, immigration increases the nation’s work force permanently (as long as immigrants remain economically active). The 1980s immigrant flow raised the share of the U.S. work force that is foreign born from 6.9 percent in 1980 to 9.3 percent in 1988. 2. Both trade and immigration augmented the nation’s effective supply of less-skilled workers by more than they augmented the effective supply of more-skilled workers. The 1985 trade deficit raised the relative supply of high school dropouts to college graduates by 5-12 percent among men and by 1017 percent among women. Increasing numbers of immigrants with less than a high school degree and declining numbers of native high school dropouts meant that over 20 percent of the high school dropout work force was foreign born by 1988. 3. Because immigrants consist disproportionately of workers who lack a high school diploma, and because import industries tend to employ relatively low-skill workers (including many immigrants), the pattern of trade and immigration observed in the United States during the 1980s fits the HeckscherOhlin trade model, in which trade and immigration are alternative ways of increasing the factor that is relatively scarce in the United States compared to the rest of the world-in this case, the declining number of less-skilled native workers. By applying these supply shifts to the textbook model of labor market equilibrium, we estimate that from 15 to 25 percent of the 11 percentage point rise in the earnings of college graduates relative to high school graduates from 1980 to 1985 can be attributed to the massive increase in the trade deficit over the same period but that the effects of trade on the college/high school wage differential diminished with improvements in the trade balance during the late 1980s. In contrast, immigration had only a small effect on the supply of high school “equivalent” workers relative to college “equivalent” workers and consequently is likely to have had only a small effect on the college/high school wage differential. Nevertheless, the large share of new immigrants with less than a high school education and the concentration of the trade deficit on industries that intensively employ high school dropouts mean that both trade and immigration are likely to have contributed substantially to the declining earnings and employment opportunities of high school dropout workers. We estimate that

215

Labor Market Effects of Immigration and Trade

between 30 and 50 percent of the nine-log-point decline in the relative weekly wage of high school dropouts from 1980 to 1988 can be attributed to trade and immigration flows. The “explanatory power” of these factors is of similar magnitude to that of other variables that have been the focus of recent research, such as the declining unionization of the U.S. labor force. Our findings regarding the effect of trade on the U.S. labor market are consistent with those of studies that document the influence of trade on earnings and employment at the industry level (Freeman and Katz 1991; Revenga 1989; MacPherson and Stewart 1990). In contrast, our findings with respect to the labor market effects of immigration differ drastically from those reported in the existing literature. These studies typically find modest and imprecisely estimated differences in earnings and employment rates of workers in cities with greatedlesser immigrant flows (Altonji and Card 1991; Butcher and Card 1991; Card 1990; LaLonde and Tope1 1991). Our results probably differ because we focus on changes in economy-wide factor endowments while the existing literature focuses on differences in factor endowments across local labor markets. In the concluding section, we reflect on the causes and interpretation of the differences in the results obtained from differing modes of analysis.

7.1 lkade and Labor Supplies In the 1970s and 198Os, the U.S. economy became more connected with the rest of the world. The ratio of the sum of exports and imports to GNP increased from 16 percent in 1970 to 25 percent in 1990. The balance of trade turned substantially negative in the mid-l980s, with a trade deficit of some 3 percent of GNP. At the same time, immigration flows increased as upwards of 700,000-800,000 legal and illegal immigrants entered the country annually. How have these changes altered the implicit supply of labor in the country in total and among skill groups? To estimate the labor supply equivalents of trade, we transform trade flows into equivalent bodies on the basis of the labor inputs in the domestic manufacturing industries that constitute the bulk of the traded-goods sector. We do this by estimating the direct labor supply embodied in trade, ignoring indirect input-output effects. Formally, let T,, be the trade flow in industry i in year t , LJO,, be the labor input per unit of output in industry i in year t , and L, be the number of personhours needed to produce total traded output-the labor supply equivalent of the trade f l o ~ sThen, . ~ for any given year t , we have

2. It is irrelevant whether we treat trade as shifts in labor supply or as shifts in labor demand. This is obviously the case in terms of the likely effects of trade on wages since, in a marketclearing model, W’ = (D’ - S’)/(e + h ) , where W’ is the change in In wages, D ’and S’ are shifts in demand and supply, and e and h are the elasticities of supply and demand, respectively. Whether we treat imports as increasing labor supply (an increase in S’) or as reducing labor demand (a decrease in D’) is a matter of taste.

216

G. J. Borjas, R. B. Freeman, and L. F. Katz

When T,,refers to imports, L, will be positive, as imports are equivalent to an increase in labor supply. When T,,refers to exports, L, will be negative, as exports effectively reduce the labor supply for domestic production. When T,, refers to net imports (imports-exports), it can be positive or negative. To allocate the implicit labor supply in trade among groups of workers with different levels of skill, we apply a modified version of equation (1) to different skill groups, using where LJ,is the implicit labor supply embodied in trade of skill groupj in year t , and a,, is the average proportion of workers in industry i in thejth skill group

for the period 1967-87. In equation ( 2 ) , we assume that the effects of trade-induced changes in output on the employment of both production and nonproduction workers are identical to equivalent domestic-induced changes in output. It is also reasonable, however, to treat the labor market effects of the two flows differently. Whereas exports are likely to create employment for both kinds of workers, or possibly create greater employment for production workers than for nonproduction workers, imports have the potential for displacing production workers to a greater extent than nonproduction workers. This possibility is likely because the sales, finance, and related activities of nonproduction workers may be relatively complementary with production workers overseas. Given these considerations, we provide two estimates of the effects of trade on employment. In method I, we treat the labor market effects of exports and imports identically, as in equation ( 2 ) ; in method 11, we use equation ( 2 ) for exports but allocate the implicit labor supply contained in imports to production workers only. Empirically, the assumption that imports have a greater effect on production than on nonproduction labor implies that the percentage of production workers should drop in industries with increasing imports. This prediction is consistent with our data. Finally, we transform implicit labor supplies due to trade into “efficiency units” that correct personhours for shifts among industries with different qualities of labor. When trade affects highly skilled labor, an efficiency-unit measure translates this into a greater effect on aggregate labor supply than when trade affects less-skilled labor. To estimate efficiency units of labor by industry in a given year, we divide each industry’s labor force into sixty-four groups, based on sex (two groups), education (four groups), and experience (eight groups), and weight the proportion of persons in each group by the average 3. The net labor supply contained in imports for skill group j in year r using the productionwhere p,, is the average proporworkers-only allocation scheme is given by LP,, = X,p,jL,,(T,,/O,,), tion of production workers in industry i from the jth skill group for the period 1967-85. We classify as production workers those workers in the manufacturing sector in the following broad occupational categories: craft workers, handlers and laborers, operatives, transport operatives, and service workers.

217

Labor Market Effects of Immigration and Trade

weekly wage for full-time workers in the group from 1963 to 1987. We normalize the measure so that the total number of efficiency units in the economy in each year equals o n e 4 We then estimate implicit labor input in efficiency units from trade in year t for groupj (EL,) as

(3) where E, is the total number of labor efficiency units employed in the U.S. labor market in t , e,, is the average proportion of group j in total labor efficiency units in industry i for the period 1967-87, and EL,is the total labor efficiency units used in industry i in year t. The implicit labor input in trade measured in efficiency units as a fraction of aggregate labor input in t is then given by (4)

EL, = XjELj, = ~i(E,,lE,)(T,,lOi,).

7.1.1 Trade Data Estimates of implicit labor supplies using equations (1)-(4) require information on personhours, the skill mix of workers, and trade flows in the tradedgoods sector. Our data on personhours, employment, and wages for sixty-four skill groups are derived from the Annual Demographic Supplements to the March Current Population Surveys (CPS) for 1964-88.5 The March CPS provides information on earnings and weeks worked in the calendar year preceding the March survey that we use to compute total personhours and efficiency units by skill group by year from 1967 to 1985 for the aggregate labor force, the manufacturing sector, and for twenty-two detailed manufacturing industries. Our data on imports, exports, output (value of shipments), total employment, and production worker employment are from the NBER Immigration, Trade, and Labor Markets Data Files.'j The data cover four-digit SIC manufacturing industries for each year from 1967 to 1985. We aggregate these data into the twenty-one manufacturing industries for which we have estimates of personhours and efficiency units by labor skill group from the CPS. We also calculate the average share of personhours and efficiency units contributed by production workers for each skill group in each of the twenty-one industries. Table 7.1 presents our estimates of the implicit change in the supply of personhours due to trade flows in manufactures relative to total labor and relative to manufacturing labor for the period 1967-85. Columns 1-3 record the 4. This normalization focuses on the mix of labor in a sector relative to the economy. Alternatively, we could have normalized on efficiency units in a given year. This would have added a trend factor to the overall number of efficiency units without affecting industry differences in efficiency units. 5 . The CPS data set utilized is described in detail in Juhn, Murphy, and Pierce (1992) and Katz and Murphy (1992). 6. Abowd (1991) provides detailed discussions on this data set and the construction of trade data on a four-digit SIC industry basis. The data on output and employment in each industry given by the NBER data set are from the Annual Survey of Manufactures.

218

G. J. Borjas, R. B. Freeman, and L. F. Katz Estimates of Itade-Induced “Change” in Aggregate Labor Supply,

Table 7.1

1967-85

Implicit Labor Input in Manufactured-Goods Trade Flowsa As a % of

As a % of Total U.S. Labor

As a % of Total U.S. Labor

Input in Personhours

Input in Efficiency Units

(2) X

(3) Net

(4)

M

M

(5) X

(6) Net

1.32 1.67 1.86 2.11 2.37 2.57 2.70 3.36 3.69

1.50 1.62 2.11 2.18 2.66 2.49 2.18 2.06 2.06

-.I8 .05 - .25 - .07 - .30 .08 .52 I .30 1.63

1.31 1.64 1.82 2.06 2.31 2.46 2.58 3.16 3.49

1.57 1.69 2.17 2.26 2.75 2.58 2.26 2.13 2.15

- .26

(1)

Year

1967-69 1970-72 1973-75 I97678 1979-8 1 1982 1983 1984 1985

- .05

- .35 -.20 - .44 - .12 .32 1.03 1.34

U.S. Labor in Mfg. in Eff. Units, (7) Net

- .87 -.18 - 1.39 - .80 - 1.86

... ... ... 6.36

Sources: Data on trade flows are from the NBER Immigration, Trade, and Labor Markets Data Files. Data on labor input and wages are from the March CPS files. “Labor input is measured in either personhours or efficiency units on the basis of a sixty-four group decomposition of the U.S. labor force. The sixty-four groups arise from splitting the labor force into two sexes, four education groups, and eight experience classes. Efficiency units in year f for group j are the average wage for group j over the period 1963-87 times the total hours of labor input of group j in year t . M = imports, X = exports, and Net = M - X . Implicit labor and implicit labor input in exports in year t = C,E,, input in imports in year t = 1,E,, (M,,/S,,), (XJS,,), yhere i is industry, E,, is efficiency units (or personhours) used in industry i in year t , and S is the value of shipments of domestic producers.

labor supply equivalence in personhours of U.S. imports ( M ) , exports (X), and the net of the two (imports minus exports) as a percentage of the total U.S. labor force. Columns 4-6 give the labor supply equivalence of trade in efficiency units as a percentage of total efficiency units for the entire U.S. labor force (calculated using the same wage weights as we used to determine the labor supply implicit in trade flows). Column 7 gives the labor supply equivalence in efficiency units of net imports in the major traded-goods sector, manufacturing. Three things stand out in the table. First is the marked change in the implicit effects of trade on the labor market between the period of the 1960s and 1970s and the period since the early 1980s. In the 197Os, trade reduced slightly the net supply (increased slightly the net demand) of labor in the United States. In the mid-l980s, the trade deficit produced a much larger increase in the implicit supply of workers both in personhours and in efficiency units.’ 7. The figures for 1984 and 1985 are likely to be fairly representative of the rest of the decade, as the trade deficit was at comparable levels on average in the periods 1986-89 and 1984-85.

219

Labor Market Effects of Immigration and Trade

Second is the difference between the implicit labor input in personhours and efficiency units between exports and imports. Because export industries are relatively skill intensive, the implicit labor input is roughly 4 percent higher in efficiency units than in personhours. Imports, by contrast, come increasingly from industries with less-skilled labor. In 1967-69 the implicit labor input in imports was 1 percent lower in efficiency units than in personhours, whereas in 1985 the implicit labor input in imports was 5.5 percent lower in efficiency units than in personhours. As a result of these patterns, the implicit labor supply due to net imports is 18 percent lower in terms of efficiency units than of personhours in 1985. Third, the growth of the implicit labor input due to net trade flows in manufacturing shows a much more dramatic picture of the “first-stage” effect of trade on workers in that sector. In 1985, the trade imbalance was equivalent to a 6.4 percent increase in efficiency units of labor in manufacturing. This highlights the fact that the direct effects of trade-induced changes in implicit labor inputs fall on only some workers. Others are affected when those displaced or not hired in manufacturing seek work in other sectors of the economy.

7.1.2 Effects of Trade on Labor Skills The difference between the personhours and efficiency-units measures of implicit labor supply for imports and exports in table 7.1 indicates that import industries employ relatively less-skilled labor than export industries. Changes in the level of trade and, more important, in the trade balance are thus likely to have very different effects on labor of different skills. To assess the magnitude of these differential effects, we used equation (3) to estimate the implicit labor supply in imports, exports, and the net trade balance for four education groups: high school dropouts, high school graduates, persons with some college, and college graduates. We made estimates for all men and women and for those with the least potential labor market experience (up to ten years after school leaving). Figure 7.1 displays plots of the estimated implicit labor supply embodied in the net trade balance in manufactures relative to the entire domestic labor force in all industries for male workers by education (fig. 7. la) and for female workers by education (fig. 7. lb). The plots with the0 notation are made under assumption I, that imports and exports affect production and nonproduction workers in the same manner as an equivalent amount of domestic production. The plots with the notation are made with assumption 11, that imports affect only production workers while exports affect both groups in the same manner as domestic production. The figure tells a clear story. Until the trade deficit developed, the implicit change in relative labor supply due to trade was modest-dwarfed by the ongoing trend in supplies due to changes in the educational attainment of workers. Among males, for example, trade in 1980 increased the implicit supply of high school dropouts by - 0.1 percent (using our method I) to 1.5 percent (using method 11) while reducing the implicit supply of college graduates by

0

0

0

0

0

0

0

N Z 8 8 S E : 8 8 0

10

m

a r

m

m

fi

,f h

P r

h

m h

P

a

B

.m m

f

kz

r c r

h

m

2

B.

Female High School Dropouts

-

$ 0.12

Method I

: 0.10 f

M + I I

a

Female High School Graduates

g 0.08

9B

c g

0.06

0.04

C

0.02

4

I: 0.00 -0 - 4.02

e

m

67

71

73

75

7-1

79

81

83

a5

67

69

71

73

75

Y W Females with Some College

M

2

77

79

81

83

85

Year

5 0.12 n

l

: - 0.10 B

M 4 I I

Female College Graduates

-

Method I

M 4 I I

0.08 0

;;0.06 Q

2- 0.04

2 0.02 n

f: 0.w

-0 - -0.02

e

J

37

i

I

I

I

69

71

73

75 Year

I

I

I

I

I

77

79

81

83

85

,

I

,

,

37

69

71

73

I

75

l

l

,

77

79

81

Year

Fig. 7.1 Implicit supply of labor in net trade in manufactures as a share of total domestic labor supply

,

,

83

85

222

G . J. Borjas, R. B. Freeman, and L. F. Katz

0.8-2.1 percent.* As a result, trade in 1980 decreased the ratio of college to high school dropouts by 0.6 percent when imports are allocated to all workers and by 3.6 percent when imports are allocated to production workers only. Among females, trade increased the implicit supply of high school dropouts by 1.2-3.2 percent and reduced the ratio of college graduates to dropouts by 1.4-3.7 percent. These changes are quite modest in comparison to the increase in the ratio of the number of college graduates to high school dropouts in the nation’s nonimmigrant work force in the same period (e.g., an increase of 52 percent among men and 87 percent among women from 1975 to 1980). Balanced trade of the sort that predominated in the 1970s and early 1980s thus had only modest effects on the market for skills. By contrast, the figure also shows that the trade deficit that began in the 1980s produced a large increase in the implicit labor supply of less-educated workers, particularly high school dropouts, but had only a slight effect on the implicit supply of college graduates. In 1985, the implicit supply of male high school dropouts via trade was 4-8 percent of the number of male dropouts in the U.S. labor force (fig. 7.la), while the implicit supply of female high school dropouts via trade was 8-13 percent of the number of female dropouts in the U.S. labor force (fig. 7. Ib). Extrapolating to the end of the 1980s, the continued, although declining, trade deficit (U.S. Council of Economic Advisors 1990, p. 297) implies that the United States kept “importing” large numbers of less-educated workers through trade for the remainder of the decade. Figure 7.2 presents comparable plots for the implicit labor embodied in net trade relative to the domestic manufacturing labor force. Since the implicit labor embodied in net trade is the same as in figure 7.1 while the denominator is about one-fifth as large, the relative increases in supply due to trade are correspondingly larger: for high school dropouts, for instance, trade flows increased the implicit supply in manufacturing by 14-27 percent for males (fig. 7 . 2 ~ and ) by 24-40 percent for females (fig. 7.2b). These figures highlight the fact that the “first-order’’ effect of the trade deficit is for less-educated workers in manufacturing. In addition to the implicit labor supply calculations in figures 7.1 and 7.2, we also estimated the effects of trade on the implicit supply of workers with less than ten years of experience. We obtained results similar to those in figures 7. I and 7.2, reflecting the fact that less-experienced workers are distributed in roughly similar proportions among sectors as more experienced workers. If trade-induced changes in implicit labor supply help account for the exceptionally large decline in the position of younger less-skilled workers, it is because these workers are on the “active market,” unprotected by seniority, 8. We report a range of estimates for the effect of trade on implicit labor supplies. In each case, the smaller number in absolute value is obtained by our method I, in which we treat production and nonproduction workers the same. The larger number in absolute value is obtained by our method 11, in which we assume that imports d o not displace nonproduction workers.

223

Labor Market Effects of Immigration and Trade

internal labor markets, and firm-specific human capital, not because they are disproportionately concentrated in industries facing import competition compared to more experienced less-skilled workers. 7.1.3 Comparing Methods I and I1 for Estimating Trade Effects Our first method of estimating the effects of trade on the implicit labor supply of skills assumes that imports and exports affect production and nonproduction labor equally. Our second method assumes that imports have no effect on the employment of nonproduction labor. If imported goods reduce employment of production labor more than nonproduction labor, industries with increasing imports ought to be associated with declines in the production worker share of employment. Accordingly, we regressed the In change in the production worker share of employment on the change in the ratio of imports to imports plus sales, the change in exports/sales, and the change in In sales. Our regression for the period 1960-85 shows that changes in import ratios contributed to the decline in the share of production workers in employment while changes in export ratios essentially had no effect: d In(% production workers) = 0.040 - 0.159 d(import ratio) (0.049) - 0.005 d(export ratio) - 0.049 d(ln sales), (0.071) (0.009)

N = 427, R2 = 0.07, where the numbers in parentheses are standard errors. Over half the within-sector change in the percentage of workers in production labor occurred in the 1980s. For 1979-85, our regression yields d In(% production workers) = -0.029 - 0.088 d(import ratio) (0.061) - 0.007 d(export ratio) - 0.007 d(ln sales), (0.071) (0.015) = 427,R2 = 0.005. Because of the smaller size and significance of the coefficient on import ratios in the period when there was “most action,” we hesitate to draw any strong inference from the data. Although the data. are consistent with the notion that increased imports alter the skill mix of industries away from lesseducated production workers and that this process may be associated with the out-sourcing of production jobs to other countries, we conclude that the evidence favoring the out-sourcing hypothesis is far from overwhelming. It is for this reason that we present our implicit labor supply calculations assuming both that exports and imports affect production and nonproduction workers equally and that imports affect production workers only.

N

Lo

a 8

f

E:

0

i;

3

Lo

a 0

k >i I Q .

I.

I

1

Female High School Graduates 2

040

4 '

M

ln

4

I

1 : i + -0.10 MI

.e

g-0.10

71

E

74

77

80

83

85

I1

_E

68

71

74

Year Femaleswtth Some College 2

z

0.40

r" 0.30

M

4

77

80

e3

85

80

83

85

Year Female College Graduates

l

M 4 I I

0 .-

1

B-

o.20

;

m

p

0.10

!-

-c

$ 0.00 2 r

g -0.10 -E

J

58

,

71

; p

0.10

I-

0.00

1

I

74

77 Year

,

80

$ -0.10 I

83

85

-E

68

71

74

77 Year

Fig. 7.2 Implicit supply of labor in net trade in manufactures as a share of domestic labor supply in manufacturing

226

G. J. Borjas, R. B. Freeman, and L. F. Katz

7.2 The Immigrant Contribution to Labor Supply What about immigration-induced changes in labor supply? How has the flow of immigrants altered the nation’s endowment of more- and less-educated workers? Because immigrants are largely permanent additions to the country’s labor force, comparisons of the number of immigrants who enter the work force annually with the implicit labor input in trade can give a misleading picture of how the two flows affect the labor market. Most traded goods displace domestic production in the same period, motivating our calculation of implicit labor supply equivalents based on individual year trade jlows. By contrast, an immigrant arriving in the United States in any given year contributes to the economy in every subsequent year in which he or she is economically active in the United States. Therefore, the effect of immigration on labor supplies is best described by the stock of the work force who are immigrants relative to U.S. domestic labor supply, rather than by the flow of immigrants to labor supply. Calculating the stock of immigrants in the United States at any point in time is not easy. In 1980 and other Census years, the Decennial Census reports the number of foreign-born persons. In April 1983, June 1986, and June 1988, the Current Population Survey (CPS) contained questions on country of birth that can also be used to estimate the number of immigrants. Both the Census and the CPS numbers, however, miss many illegal alien immigrants, who are especially likely to be in the low-skill work force. For purposes of estimating labor supplies due to immigration, we need to know the number, labor force participation rate, and educational composition of uncounted immigrants as well as the same information for those counted in the Census and CPS surveys. 7.2.1

Estimating Uncounted Illegal Immigrants

Using the enumeration provided by the 1980 Census as a base, we adjusted the 1980 count upward to allow for uncounted illegal immigrants, by sex. First, we counted 6.47 million foreign-born persons aged 18-64 who are labor force participants using the lil000 file of the 1980 Census. Warren and Passel (1987) indicate that this Census enumeration “found’ many more foreign-born persons than would be expected given the legal flows reported by the U.S. Immigration and Naturalization Service. They estimate that the 1980 Census enumerated two million illegal aliens, of whom approximately 60 percent were of working-force age (aged 18-64). On the basis of vital statistics by country of birth, Borjas, Freeman, and Lang (1991) estimate that there were approximately 50 percent more illegal aliens in the United States in 1980 than were counted in the Census. This suggests that the Census count fell short of the number of illegal immigrants in the economy by roughly one million persons. Assuming that 60 percent of these aliens were of working age and that their distribution by sex is similar to that for counted illegal aliens

227

Labor Market Effects of Immigration and Trade

Table 7.2

Immigrants in the U.S. Labor Force, 1980-88: Labor Force Participants Aged 18-64 No. of Immigrants in the Labor Force (thousands) Counted

1980 1983 1986 1988

Male 3,771 4,312 5,294 5,719

Female 2,701 3,056 3,697 4,083

Counted Plus Uncounted Total 6,472 7,368 8,991 9,802

Male 4,025 4,601 5,647 6,102

Female 2,836 3,209 3,884 4,287

Total 6,861 7,810 9,530 10,390

Immigrant Share of Labor Force:' (Ctd + Unctd Imm)/Total Labor Force Male 7.0 1.9 9.2 9.9

Female 6.8 7.2 8.0 8.6

Total 6.9 1.6 8.6 9.3

Sources: Counted immigrants: 1980, tabulated from the 1980 U.S. Census of Population; 1983, tabulated from the April 1983 CPS, using sample weights; 1986-88, tabulated from June CPSs, using sample weights. Native labor force: tabulated from the same surveys as counted immigrants. Numbers of uncounted immigrants: In 1980, the number of uncounted immigrants is assumed to be equal to 50 percent of the number of counted illegal immigrants in the 1980 Census of Population. Estimates of counted illegal immigrants in the 1980 Census are from Warren and Passel (1987). In 1983, 1986, and 1988, the numbers are estimated by applying the ratio of uncounted illegals to counted immigrants in 1980 (0.06). 'Total labor force = native labor force plus counted immigrants plus uncounted immigrants.

(55 percent of counted illegal aliens are men, according to Warren and Passel), we estimate that there were approximately 330,000 uncounted illegal men and 270,000 uncounted illegal women in the U.S. labor force in 1980. Since most counted illegal immigrants came to the United States in 1975-80, suggesting that uncounted illegals came also in that period, we apply the 1980 labor force participation rate of immigrants who arrived in 1975-80 (76.9 percent for men, 50.0 percent for women) to these numbers to estimate the number of uncounted immigrants in the labor force in 1980. The resultant figures suggest that 253,800 uncounted illegal men and 135,000 uncounted illegal women were in the labor force in 1980. This increases our estimates of the number of immigrants in the 1980 labor force by roughly 6 percent (compare cols. 3 and 6 in table 7.2). To measure the stock of immigrants in ensuing years, we make use of the April 1983, June 1986, and June 1988 CPSs. Each of these surveys included a question on immigrant status. Because the April 1983 CPS had problems with estimating Hispanics, demographers have questioned the sample weights in the survey (Passel and Woodrow 1987). As a result, we pay greater attention to the results for 1986 and 1988.9 Our analytic procedure for estimating the immigrant labor force from each CPS is the same. First, we enumerate the number of foreign-born persons aged 18-64 in the labor force in the survey week. Second, we assume that the ratio of uncounted illegals to counted immigrants in each survey is constant 9. Woodrow, Passel, and Warren (1987) provide a detailed discussion of the quality of the immigration data from the June 1986 CPS, and Woodrow and Passel (1990) provide an analysis of the immigration data from the June 1988 CPS.

228

G. J. Borjas, R. B. Freeman, and L. F. Katz

over time at the 6 percent estimated for the 1980 Census. This leads us to adjust upward the 1983 CPS count by 422,100 workers, the 1986 CPS count by 539,500, and the 1988 CPS by 588,100. Table 7.2 shows the results of our calculations. It records the unadjusted and adjusted estimated number of immigrants in the U.S. labor force in total and by sex in 1980, 1983, 1986, and 1988 and gives the proportion of the total work force consisting of immigrants. Because immigrant stocks are the cumulation of annual net flows of immigrants in different years, the immigrant contribution to the nation’s labor supply exceeds the implicit contribution of trade to the nation’s labor supply by over fourfold even during the mid-1980s trade deficit period (compare the 8.6 percent in table 7.2 for 1986 with the 1.6 percent implicit labor supply in col. 3 of table 7.1 for 1985). In addition, the table shows a strong upward trend in the immigrant share of the work force among both sexes over the period. From 1980 to 1988, the immigrant share of the work force increased by approximately 35 percent! Growth in the number of immigrant workers accounted for over 25 percent of the growth of the U.S. work force in the 1980s. 7.2.2 Immigrants by Education

To determine how the inflow of immigrants altered the educational composition of the U.S. work force, we estimated the number of immigrants by education and sex in 1980, 1983, 1986, and 1988. For persons counted by the Census or the CPS, determining years of schooling is a direct matter because schooling is recorded for all respondents, including immigrants. As many immigrants are educated overseas under different educational systems and in different languages than prevail in the United States, however, their years of schooling may have a different value in the U.S. labor market than years of schooling in the United States. For simplicity, we ignore this problem and treat years of schooling of immigrants as equivalent to the schooling of natives. For illegal aliens who were not enumerated in the Census or the CPS, we estimate educational attainment from the educational distributions of 197580 immigrants reported in the 1980 Census for two national origin groups: Mexican immigrants and all other immigrants. We distinguish between these groups for two reasons. First, the educational distribution of immigrants differs markedly between Mexican and non-Mexican immigrants. As figure 7.3 shows, Mexican immigrants are disproportionately high school dropouts, whereas non-Mexican immigrants have a more even distribution among education categories. Hence, the proportion of uncounted immigrants who are Mexican will affect the estimated educational distribution of uncounted illegal workers. Second, a large proportion of counted illegals are in fact Mexicans (one-half, according to Warren and Passel [ 1987]), suggesting that many uncounted illegals are Mexican or are similar to illegal Mexican immigrants. On the basis of evidence that about three-quarters of the persons who ap-

229

Labor Market Effects of Immigration and Trade

cf LessthanHS HS Graduate

0 Some College College Graduate Mexican Immigrants

Non-Mexican Immigrants

Fig. 7.3 Education distributions of Mexican and non-Mexican labor force participants who entered the United States in 1975-80

plied for legalization of their immigration status as a result of the 1986 Immigration Reform and Control Act were of Mexican origin (U.S. Immigration and Naturalization Service 1990), we assume that 75 percent of uncounted illegal immigrants were Mexican and that 25 percent were of non-Mexican origin. We take a weighted average of the educational distribution of Mexican immigrants (0.75) and of the educational distribution reported by all other immigrants (0.25) to obtain an estimated educational distribution of uncounted illegal immigrants. l o Finally, we estimate the educational distribution of all immigrants by summing relevant numbers of counted and uncounted immigrants in each of our four educational categories. The resulting educational distributions are reported in table 7.3. This table gives our estimated distribution for the entire immigrant population (including uncounted immigrants), the distribution for the enumerated population (so that the reader can see the effect on the results of our assumptions about the uncounted population), and the comparable distribution for native workers. The distribution for native workers is obtained by enumerating the number of native-born workers in each education category from the relevant Census or CPS file. Consider first the column giving the educational distribution of the 1980 male immigrant work force. It indicates that 39.6 percent of the male immigrant labor force was composed of persons with less than a high school education. This statistic is notable because only 22.7 percent of native male workers lacked a high school diploma in 1980. By contrast, the proportion of immigrants with college degrees is similar to the proportion of native American workers with college degrees. The data for 1988 tell an even more striking story of disparity among dropouts: 36.0 percent of immigrant men compared to 15.3 percent of native men in the labor force lacked a high school diploma. The figures for female immigrant workers follow the same pattern: a dispro10. We also made estimates assuming that 50 percent of the uncounted illegal workers were Mexican, on the basis of the Warren and Passel estimate that approximately 50 percent of the counted illegals in the 1980 Census are of Mexican origin. The results differed only modestly from those reported here.

230

G. J. Borjas, R. B. Freeman, and L. F. Katz

Table 7.3

Educational Distributions of Immigrant and Native Workers, i9no-nn Percentage Shares ~

Males Completed Years of Schooling

1980

Females

1983

1986

1988

1980

1983

1986

1988

Counted plus uncounted immigrants Less than 12 39.6 34.1 12 22.4 22.5 18.1 13-15 16.3 I6 or more 21.7 25.3

36.9 23.1 16.3 23.1

36.0 22.6 16.2 25.1

35.7 30.3 17.9 16.0

29.7 30.3 18.1 22.1

21.0 30.8 20.5 21.7

28.3 21.6 21.8 22.3

Natives Less than 12 12 13-15 16 or more

22.7 37.9 19.2 20.2

19.0 35.8 22.6 22.1

16.5 36.5 23.6 23.3

15.3 36.0 24.6 24.0

18.2 44.9 20.9 16.0

14.2 42.8 24.8 18.2

12.1 41.6 26.6 19.6

11.0 40.6 21.6 20.1

Counted immigrants Less than 12 12 13-15 16 or more

37.5 23.0 16.9 22.6

31.6 23.1 18.8 26.5

34.7 24.4 16.9 24.0

33.7 23.2 16.8 26.2

34.0 31.2 18.4 16.4

21.7 30.9 18.6 22.8

24.8 31.7 21.1 22.4

26.2 28.3 22.5 23.0

Sources: Educational distributions for natives and for counted immigrants are based on the number of immigrant and native workers with the specificed education levels counted in the 1980 Census of Population, the April 1983 CPS, and the June 1986 and 1988 CPSs. The education distribution for uncounted immigrants is based on a weighted average of the education distributions of counted Mexican immigrant workers (75 percent) and all other counted immigrant workers (25 percent) who entered the United States from 1975 to 1980; these educational distributions were tabulrlted from the 1980 Census and are displayed in fig. 7.3. The total numbers of uncounted immigrants by sex and year are from table 7.1.

portionate number were high school dropouts compared to native female workers. Turning to changes over time, table 7 . 3 shows that the share of native workers with less than a high school degree fell sharply in the 1980s. Indeed, the underlying statistics reveal a massive 5.6 million drop in the number of U.S.born workers with less than a high school degree from 1980 (19.2 million) to 1988 (13.6 million). This is the result of the retirement of older less-educated workers and increased rates of high school graduation among younger cohorts. For immigrants, by contrast, the high school dropout share of workers fell modestly, and there was an actual growth in the number of dropout immigrant workers by some 0.8 million persons. The implication is that, during the 1980s, immigration became a major source of the supply of high school dropout workers whereas it was only a much smaller source of the supply of more-educated workers. Table 7.4 pursues this implication by reorganizing the data on the numbers of immigrant and native workers by education to show the relative increment

231

Labor Market Effects of Immigration and Trade Estimated Increments to the Supply of Labor by Education and Sex Arising from Immigration and 'lkade, 1980-88

Table 7.4

Males