P. 1
Immigration, Trade, and the Labor Market

Immigration, Trade, and the Labor Market

|Views: 287|Likes:
Published by University of Chicago Press an imprint of UChicagoPress
Are immigrants squeezing Americans out of the work force? Or is competition wth foreign products imported by the United States an even greater danger to those employed in some industries? How do wages and unions fare in foreign-owned firms? And are the media's claims about the number of illegal immigrants misleading?

Prompted by the growing internationalization of the U.S. labor market since the 1970s, contributors to Immigration, Trade, and the Labor Market provide an innovative and comprehensive analysis of the labor market impact of the international movements of people, goods, and capital. Their provocative findings are brought into perspective by studies of two other major immigrant-recipient countries, Canada and Australia. The differing experiences of each nation stress the degree to which labor market institutions and economic policies can condition the effect of immigration and trade on economic outcomes
Contributors trace the flow of immigrants by comparing the labor market and migration behavior of individual immigrants, explore the effects of immigration on wages and employment by comparing the composition of the work force in local labor markets, and analyze the impact of trade on labor markets in different industries. A unique data set was developed especially for this study—ranging from an effort to link exports/imports with wages and employment in manufacturing industries, to a survey of illegal Mexican immigrants in the San Diego area—which will prove enormously valuable for future research.
Are immigrants squeezing Americans out of the work force? Or is competition wth foreign products imported by the United States an even greater danger to those employed in some industries? How do wages and unions fare in foreign-owned firms? And are the media's claims about the number of illegal immigrants misleading?

Prompted by the growing internationalization of the U.S. labor market since the 1970s, contributors to Immigration, Trade, and the Labor Market provide an innovative and comprehensive analysis of the labor market impact of the international movements of people, goods, and capital. Their provocative findings are brought into perspective by studies of two other major immigrant-recipient countries, Canada and Australia. The differing experiences of each nation stress the degree to which labor market institutions and economic policies can condition the effect of immigration and trade on economic outcomes
Contributors trace the flow of immigrants by comparing the labor market and migration behavior of individual immigrants, explore the effects of immigration on wages and employment by comparing the composition of the work force in local labor markets, and analyze the impact of trade on labor markets in different industries. A unique data set was developed especially for this study—ranging from an effort to link exports/imports with wages and employment in manufacturing industries, to a survey of illegal Mexican immigrants in the San Diego area—which will prove enormously valuable for future research.

More info:

Publish date: Dec 1, 2007
Added to Scribd: Nov 19, 2009
Copyright:Traditional Copyright: All rights reservedISBN:9780226000961
List Price: $99.00

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
This book can be read on up to 6 mobile devices.
Full version available to members
See more
See less

02/05/2016

446

9780226000961

This Page Intentionally Left Blank

Immigration, Trade, and
the Labor Market
A National Bureau
of Economic Research
Project Report
Immigration,
Trade, and the
Labor Market
Edited by John M. Abowd and
Richard B. Freeman
The University of Chicago Press
Chicago and London
JOHN M. ABOWD is professor of labor economics and management,
Cornell University, and a research associate of the NBER. RICHARD B.
FREEMAN is professor of economics at Harvard University and director of
the Labor Studies program at NBER.
The University of Chicago Press, Chicago 60637
The University of Chicago Press, Ltd., London
0 1991 by the National Bureau of Economic Research
All rights reserved. Published 1991
Printed in the United States of America
00999897969594939291 5 4 3 2 1
Library of Congress Cataloging-in-Publication Data
Immigration, trade, and the labor market / edited by John M. Abowd and
Richard B. Freeman.
cm. - (National Bureau of Economic Research Project
“Papers presented at a conference held in Cambridge, Massachusetts,
Includes bibliographical references and indexes.
ISBN 0-226-00095-8 (acid-free paper)
1. Alien labor-United States-Congresses.
p.
paper)
11-12 September 198T-Pref.
2. Alien labor-Can-
5. Labor market-Can-
ada-Congresses. 3. Alien labor-Australia-Congresses.
4. Labor market-United States-Congresses.
ada-Congresses. 6. Labor market-Australia-Congresses.
7. Foreign trade and employment-United States-Congresses.
8. Foreign trade and employment-Canada-Congresses.
trade and employment-Australia-Congresses.
11. Freeman, Richard B. (Richard Barry).
HD8081.A5153 1991
331.6’2-dc20 90-24954
9. Foreign
I. Abowd, John M.
111. Series.
CIP
8 The paper used in this publication meets the minimum requirements of
the American National Standard for Information Sciences-Permanence
of Paper for Printed Library Materials, ANSI 239.48-1984.
National Bureau of Economic Research
Officers
George T. Conklin, Jr., chairman
Paul W. McCracken, vice chairman
Martin Feldstein, president and
chief executive oficer
Geoffrey Carliner, executive director
Charles A. Walworth, treasurer
Sam Parker, director offnance and
administration
Directors at Large
John H. Biggs
Andrew Brimmer
Carl F. Christ
George T. Conklin, Jr.
Kathleen B. Cooper
Jean A. Crockett
George C. Eads
Morton Erhlich
Martin Feldstein Robert T. Parry
George Hatsopoulos Peter G. Peterson
Lawrence R. Klein Robert V. Roosa
Franklin A. Lindsay Richard N. Rosett
Paul W. McCracken Bert Seidman
Leo Melamed Eli Shapiro
Michael H. Moskow
James J. O’Leary
Donald S. Wasserman
Directors by University Appointment
Jagdish Bhagwati, Columbia
William C. Brainard, Yale
Franklin Fisher, Massachusetts Institute of
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
Burton A. Weisbrod, Wisconsin
Michael Yoshino, Harvard
Arnold Zellner, Chicago
Technology Craig Swan, Minnesota
Directors by Appointment of Other Organizations
Richard A. Easterlin, Economic History
Gail Fosler, The Conference Board
A. Ronald Gallant, American Statistical
Bruce Gardner, American Agricultural
Robert S . Hamada, American Finance
Robert C. Holland, Committee for Economic
Association
Association
Economics Association
Association
Development
Directors Emeriti
Moses Abramovitz
Emilio G. Collado
Frank W. Fetter
David Kendrick, American Economic
Ben E. Laden, National Association of
Rudolph A. Oswald, American Federation of
Association
Business Economists
Labor and Congress of Industrial
Organizations
Association
Certified Public Accountants
Douglas D. Purvis, Canadian Economics
Charles A. Walworth, American Institute of
Thomas D. Flynn
Gottfried Haberler
Geoffrey H. Moore
George B. Roberts
Willard L. Thorp
William S. Vickrey
Relation of the Directors to the
Work and Publications of the
National Bureau of Economic Research
1. The object of the National Bureau of Economic Research is to ascertain and to present to
the public important economic facts and their interpretation in a scientific and impartial manner.
The Board of Directors is charged with the responsibility of ensuring that the work of the National
Bureau is carried on in strict conformity with this object.
2. The President of the National Bureau shall submit to the Board of Directors, or to its
Executive Committee, for their formal adoption all specific proposals for research to be instituted.
3. No research report shall be published by the National Bureau until the President has sent
each member of the Board a notice that a manuscript is recommended for publication and that in
the President’s opinion it is suitable for publication in accordance with the principles of the
National Bureau. Such notification will include an abstract or summary of the manuscript’s con-
tent 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 Presi-
dents), consisting of three Directors selected as nearly as may be one from each general division
of the Board. The names of the special manuscript committee shall be stated to each Director
when notice of the proposed publication is submitted to him. It shall be the duty of each member
of the special committee to read the manuscript. If each member of the manuscript committee
signifies his approval within thirty days of the transmittal of the manuscript, the report may be
published. If at the end of that period any member of the manuscript committee withholds his
approval, the President shall then notify each member of the Board, requesting approval or dis-
approval of publication, and thirty days additional shall be granted for this purpose. The manu-
script 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 manu-
script committee, until forty-five days have elapsed from the transmittal of the report in manu-
script form. The interval is allowed for the receipt of any memorandum of dissent or reservation,
together with a brief statement of his reasons, that any member may wish to express; and such
memorandum of dissent or reservation shall be published with the manuscript if he so desires.
Publication does not, however, imply that each member of the Board has read the manuscript, or
that either members of the Board in generahor the special committee have passed on its validity
in every detail.
6. Publications of the National Bureau issued for informational purposes concerning the work
of the Bureau and its staff, or issued to inform the public of activities of Bureau staff, and volumes
issued as a result of various conferences involving the National Bureau shall contain a specific
disclaimer noting that such publication has not passed through the normal review procedures
required in this resolution. The Executive Committee of the Board is charged with review of all
such publications from time to time to ensure that they do not take on the character of formal
research reports of the National Bureau, requiring formal Board approval.
7. Unless otherwise determined by the Board or exempted by the terms of paragraph 6, a copy
of this resolution shall be printed in each National Bureau publication.
(Resolution adopted October 25, 1926, as revised through Seprember 30, 1974)
Contents
Preface
Introduction and Summary
John M. Abowd and Richard B. Freeman
I. IMMIGRANT FLOWS AND PERFORMANCE IN THE UNITED STATES
1. Immigration and Self-Selection
George J. Borjas
2. Undocumented Mexican-born Workers in the
United States: How Many, How Permanent?
George J. Borjas, Richard B. Freeman,
and Kevin Lang
and Labor Flows in Mexico
Juan Diez-Canedo R.
3. The Effect of Policy Restrictions on Capital
4. Internal Migration of U.S. Immigrants
Ann P. Bartel and Marianne J. Koch
5. Migration, Ethnicity, and Labor
Force Activity
Marta Tienda and Franklin D. Wilson
11. IMPACT OF IMMIGRATION, TRADE, AND CAPITAL FLOWS ON THE
U.S. LABOR MARKET
6. Labor Market Adjustments to Increased
Immigration
Robert J. LaLonde and Robert H. Tope1
ix
1
29
77
101
121
135
167
viii Contents
7. The Effects of Immigration on the Labor
Market Outcomes of Less-skilled Natives
Joseph G. Altonji and David Card
8. Industrial Wage and Employment
Determination in an Open Economy
Richard B. Freeman and Lawrence F. Katz
United States 26 1
Jonathan S. Leonard and Rachel McCulloch
20 1
235
9. Foreign-Owned Businesses in the
10. Immigration, International Ikade, and the
Wages of Native Workers 285
Peter Kuhn and Ian Wooton
11. Immigrants, Labor Market Pressures, and
the Composition of the Aggregate Demand
Susan M. Collins
305
111. COMPARATIVE EXPERIENCES: CANADA AND AUSTRALIA
12. An Analysis of the Earnings of Canadian
Immigrants 321
David E. Bloom and Morley Gunderson
13. The Effects of International Competition
on Collective Bargaining Outcomes:
A Comparison of the United States
and Canada 343
John M. Abowd and Thomas Lemieux
Experience in Australia 369
John J. Beggs and Bruce J. Chapman
15. Why are Low-skilled Immigrants in the
United States Poorly Paid Relative to Their
Australian Counterparts? Some of the Issues
Illustrated in the Context of the Footwear,
Clothing, and Textile Industries 385
R. G. Gregory, R. Anstie, and E. Klug
Appendix: The NBER Immigration, Ikade,
and Labor Markets Data Files
John M. Abowd
List of Contributors 423
14. Male Immigrant Wage and Unemployment
407
Author Index
Subject Index
425
429
Preface
This volume consists of papers presented at a conference held in Cambridge,
Massachusetts, 11-12 September 1987, and is part of the National Bureau of
Economic Research Labor Studies program. Support for the project came
from the Ford Foundation. We are grateful to Jennifer Amadeo-Holl, Jane
Konkel, and Jean Brown for help in getting these papers ready for publication.
Any opinions expressed in this volume are those of the respective authors
and do not necessarily reflect the views of the National Bureau of Economic
Research or the sponsoring organization.
This Page Intentionally Left Blank
Introduction and Summary
John M. Abowd and Richard B. Freeman
During the 1970s and 1980s, immigration, trade, and foreign investment in
the United States became increasingly important in the U.S. labor market.
The number of legal and illegal immigrants to the country increased, altering
the size and composition of the work force and substantially raising the im-
migrant share of labor in “gateway” cities such as Miami, Los Angeles, and
New York. The national origins of immigrants changed from primarily Euro-
pean to Mexican, Latin American, and Asian. Foreign trade rose relative to
gross national product, and a massive trade deficit developed in the 1980s,
turning the United States into a substantial debtor nation.
Because the composition of employment shifted from manufacturing to
nontraded services, the immediate burden of adjusting to trade-induced
changes fell on a decreasing segment of the work force. As the flip side of the
trade deficit, foreign investment in the United States grew rapidly, with for-
eign direct investment increasing until 3% of American workers were em-
ployed in foreign-owned firms. While at one time labor market analysts could
look on the United States as a largely closed economy, the changes of the
1970s and 1980s brought about the internationalization of the U.S. labor
market.
What are the interrelations among the flows of foreign outputs and inputs
that have caused such a change in the way we look at the U.S. labor market?
How have the flows changed over time? Which industries or areas are most
John M. Abowd is professor of labor economics and management, Comell University, and a
research associate of the National Bureau of Economic Research. Richard B. Freeman is professor
of economics, Harvard University, and director of the Labor Studies Program of the National
Bureau of Economic Research.
The authors acknowledge financial support from the Ford Foundation and the National Science
Foundation (grant 88-13847 to Abowd). The authors thank George Bojas, Lawrence Katz, and
Robert Topel for their help in preparing this paper. Daniel Kessler, Laura Leete, and Ana Revenga
served as research assistants.
2 John M. Abowd and Richard B. Freeman
heavily affected by the movements in goods, the influx of immigrants, or for-
eign direct investment? What have we learned from studying the effects of
these increased flows in the U.S. labor market? This paper presents back-
ground information about the growing internationalization of the U.S. labor
market and summarizes the results of the NBER studies contained in this vol-
ume.
The paper highlights several aspects of the internationalization of the U. S.
work force.
1 . Although the number of immigrants relative to the population increased
from the 1950s to the 1980s, the immigrant share of the growth of the work
force was relatively moderate. The rapid growth of the labor force due to
increased female participation and entry of the baby boom generation to the
labor market kept pace with the influx of immigrant workers.
2. The trade content of the U. S. economy, measured by exports plus im-
ports relative to sales or GNP, has increased markedly, but the share of labor
in traded sectors, notably manufacturing, has fallen, so that a smaller fraction
of workers are directly affected by trade than in the past. Those workers are,
however, more closely tied to world markets than in the past.
3. Direct foreign investment was substantial in the 1980s, reaching 34% of
gross U.S. investment in 1988. Three percent of the private U. S. work force
was employed in foreign-owned enterprises by the mid- 1980s.
4. The immigrant share of the labor force differs largely across geographic
areas, whereas the trade share of product markets differs largely across indus-
tries. This motivates the research strategy for the studies of the effects of im-
migration and trade on the U.S. labor force: studies concerned with how im-
migration affects labor market outcomes contrast wages and employment in
local areas with different immigrant shares of the work force; studies con-
cerned with how trade affects labor market outcomes contrast wages and em-
ployment in industries with different trade shares of output.
5. Industries are related to the open economy in a variety of ways. Indus-
tries in which there are considerable imports employ a disproportionate share
of immigrants, whereas high export industries employ relatively few immi-
grants. Direct foreign investment is concentrated in manufacturing. Overall,
the first-order effects of the internationalization of the labor market fall on
manufacturing.
6. There are significant differences in the characteristics of workers be-
tween export-intensive, import-intensive, and immigrant-intensive sectors.
Women workers and lower-paid, less-skilled workers are highly concentrated
in sectors where imports are significant and where relatively many immigrants
work. Perhaps surprisingly, foreign-owned enterprises have a comparable
unionization rate to domestically owned enterprises: they are concentrated in
traded goods sectors and have higher wages than domestic producers.
The paper is divided into four parts. Section 1 deals with the aggregate
3 Introduction and Summary
flows of people, goods, and capital from overseas to the United States. Sec-
tion 2 describes the industrial pattern and regional dimensions of the labor,
goods, and capital flows as shown in the NBER data files developed for this
project.' Section 3 turns to the characteristics of workers in sectors most af-
fected by trade, immigration, and foreign direct investment. In sections 1-3,
we have compiled statistics from a wide variety of sources in order to present
comparable figures for 1960, 1970, 1980, and the most recent year available.
Section 4 summarizes the findings of the papers included in this volume.
1. The Aggregate Flows of People, Goods, and Capital
In this section, we review the basic data on each of the three flows under
study: labor, goods, and capital. In contrast to demographic studies that focus
on the immigrant share of the population and the increase in population, we
focus on the immigrant share of the labor force and the increase in the labor
force. In contrast to trade studies that focus on balance of payments issues,
we focus on the proportion of workers in traded sectors and the ratio of ex-
ports plus imports to output in those sectors. In contrast to financial studies
that consider international capital mobility broadly defined (and equal to im-
ports minus exports by definition), we focus on direct foreign investment in
plant and equipment.
Flows of Labor
Table 1 presents the basic data on the flows of immigrants entering the
country (pt. A) and the stock of immigrants in the United States (pt. B) from
the 1940s through the 1980s. The table provides figures for legal immigrant
flows and legal plus estimated illegal immigrant flows in absolute numbers
and relative to the population, labor force, and change in the labor force. The
data in the first two columns of the table show that the number of immigrants
coming into the United States and the number per one thousand inhabitants
rose in the 1970s and 1980s, consistent with the increased public concern
about immigration. When we consider the immigrant share of changes in the
population and labor force, however, a different story emerges. Because the
baby boom increase in the U.S. population occurs in the early postwar years
(through 1960), the legal immigrant share of population growth is relatively
small during this period. Because of the increased participation of women, the
influx of baby boomers into the labor markets, and the fact that many legal
immigrants enter for family unification reasons rather than for labor market
reasons, the estimated immigrant flow share of the growth of the labor force
1. For a description of these data, see Abowd (in this volume).
4 John M. Abowd and Richard B. Freeman
Table 1 Flows and Stocks of Immigrants Relative to the Population
and Labor Force
A. Flows of Immimants
DeCadal Flow of Inflow Per Immigrant Flow Share of Change:
Immigrants 1,Ooo U.S.
Period (thousands) inhabitants In Population (8) In Labor Force
Legal flows only:"
1941-50 1,035 .7 5.2 7.3
1951-60 2,515 1.5 8.9 14.5
196 1-70 3,322 1.7 13.6 11.1
1971-80 4,493 2.1 19.8 9.3
1981-90 5,900 2.5 26.8 16.2
Legal and illegal flowsb
1971-80 5,800 2.7 25.6 12.0
1981-90 8,400 3.6 38.2 23.1
B . Stocks of Immigrants
Number of Number of Foreign
Foreign Born Born in Civilian Immigrant
Counted % of Labor Force % of Labor
Census of Population (thousands) Population (thousands) Force
As reported:
1940 11,657 8.8
1950 10,431 6.9 4,838 8.2
1960 9,738 5.4 4,134 6.1
1970 9,619 4.7 4,223 5.2
1980 14,080 6.2 7,001 6.7
1980 15,380 6.8 7,647 7.3
Adjusted for undercount:c
Sources: Part A Flow of immigrants from U.S. Bureau of the Census, Statistical Abstract of the
United States, 1989, table 7 (from the Statistical Yearbook of the Immigration and Naturalization
Service), with the 1981-90 flow estimated by extrapolating the 1981-87 flows. Immigrant flow
shares of changes were obtained by dividing flows by changes in decadal population from the
relevant decades (tables in Council of Economic Advisers, Economic Report of the President,
1990). To obtain immigrants in the labor force, we assumed that the labor force participation rate
of the decadal flow of immigrants was the same as the ratio of foreign-born workers in the civilian
labor force to the foreign-born population (see pt. B). Part B: Foreign-born count and percentage
of population from Sruristical Abstracr of the UnitedStates. 1988, table 44 (from the U.S. Census
of Population). Foreign-born in the civilian labor force from various Censuses of Population.
Qfficial counts from the Immigration and Naturalization Service summed over the indicated
years.
bAdjusted for illegal flows using estimates from Borjas, Freeman, and Lang (in this volume) and
Warren and Passel (1987), as described in the text.
cAdjusted by adding the 1.3 million estimated uncounted illegal immigrants to the 1980 Census
counts.
5 Introduction and Summary
actually falls from the 1950s to the 1970s, raising serious doubts about the
labor market basis for concern over immigration until the 1980s. Then the
number of immigrants rises substantially, and the contribution to both popu-
lation and labor force growth reaches a postwar high.
The figures in part B for actual counts of the stock of immigrants (which
depend not only on inflows of immigrants but also on emigration and the death
or retirement of persons who immigrated decades earlier and which include
some illegal immigrants) tell a generally similar story. While declines in the
immigrant share of the population and labor force are reversed for the decade
1971-80, the immigrant proportion of the population or labor force in 1980
remains below the 1950 proportion.
What happens to this picture when adjustments are made for the widely
publicized illegal immigration into the United States? We have made adjust-
ments in the table based on the methods of Borjas, Freeman, and Lang (in this
volume) and earlier research on illegal immigration (Warren and Passel 1987).
The bases for our adjustments are Warren and Passel’s estimate that the 1980
Census included about two million illegal immigrants and Borjas, Freeman,
and Lang’s estimate that approximately 6 1 % of illegal (Mexican) immigrants
were counted in the Census. Taken together, these estimates suggest that there
were on the order of 3.3 million illegal aliens in the United States in 1980.
Warren and Passel estimate that 75% of the illegals counted in the Census
came in the 1970s. Assuming, conservatively, that 75% of the uncounted ille-
gal immigrants also came in the 1970s, we get 2.5 million as the estimated
flow of illegal immigrants in the 1970s. Adding this number to the number of
legal immigrants reported by the Immigration and Naturalization Service
(INS) in the rows giving “legal and illegal flows” changes greatly the picture
of immigrant flows given in part A. Immigrant flows now rise sharply in the
1970s compared to the 1960s. Similarly, adding 1.3 million uncounted immi-
grants to the 1980 Census count raises the immigrant’s share of population
and labor force in part B of the table to levels close to those of 1950.
If illegal flows proceeded in the 1980s at the same rate as in the 1970s,
then, given the INS estimates of legal immigrant flows, we estimate that some
8,400,000 immigrants came to the United States in the 1980s. This raises the
immigrant inflow per one thousand United States inhabitants and the immi-
grant share of the change in population and labor force above the levels of the
1970s. The 1970s and 1980s were periods of marked acceleration in immigra-
tion, in large part because of illegal flows.
Another aspect of the flow of immigrants to the United States deserves at-
tention. The change in the geographic origins of immigrants following the
1965 Immigration Act has produced a dramatic shift in immigrant origins
from Europe and Canada to Asia. Figure 1 illustrates this change. If we ad-
justed the proportions in the figure for illegal immigrants (largely Mexican),
the share from Latin America would also rise.
6 John M. Abowd and Richard B. Freeman
Fig. 1 Distribution of immigrant origins
Source: Immigration and Naturalization Service Statistical Yearbook, 1987, table 2,
“Immigration by Region and Selected Country of Last Residence.”
Flows of Goods
Figure 2 shows the widely heralded increase in the role of trade in the U.S.
economy in terms of two related measures-exports ( X ) plus imports (M)
relative to GNP, which we will call the trade content of the economy and the
ratio of the trade balance (exports minus imports) to GNP. In the 1950s and
1960s, the overall trade content of GNP was roughly 10%-11%, with U.S.
exports exceeding imports. In the 1970s, the trade content jumped, particu-
larly after 1978, reaching a peak in 1981, then hovered around this level for
the rest of the decade. The balance of trade diverged modestly from year to
year until 1983, when it became negative. Large negative trade balances char-
acterize the rest of the decade and are unlike any other postwar period.
While the trade content of the U. S. economy has risen sharply, the propor-
tion of workers employed in the traded goods sectors-manufacturing , min-
ing (including crude oil), and agriculture-has fallen, so that relatively fewer
workers are directly imported by foreign competition. Table 2 shows the ratio
of exports plus imports to sectorul GNP for traded goods (agriculture, mining,
and manufacturing), all other sectors, and the entire U.S. economy for the
7 Introduction and Summary
a,
Y 8 -
a 6 -
4 -
- 4 -
-6
+ (Exports - Irnports)/GNP
Fig. 2 Openness of the U.S. economy
Source: U.S. National Income and Product Accounts, July 1989.
, I I , , , , 1 , 1 1 1 1 1 1 1 , 1 , 1 1 1 , I , I , , r , , , , I ,
years 1960, 1970, 1980, and 1987. The table also shows the percentage of
GNP originating in the sector and the percentage of full-time equivalent em-
ployment in the sector. Exports plus imports as a percentage of sectoral GNP
rise sharply in the traded goods sectors, but the share of GNP and the share of
employment in the traded goods sectors fall. Whereas in 1960 33% of the
work force and 35% of GNP were in the traded goods sectors, by 1987 only
21% of employment and 23% of GNP were in those sectors. A smaller frac-
tion of the labor force is directly affected by foreign competition by 1987 than
in the earlier decades. The table also shows the employment-weighted exports
plus imports as a percentage of sectoral GNP (last row) and the comparable
ratio for the overall economy (“total all sectors”). The traded portion of the
entire U.S. economy (goods and services) rose from 10% in 1960 to 22% in
1987 by either overall measure (also shown in fig. 2). The economy-wide
trade ratios rise by much less than the ratios in the traded goods sectors. In
terms of direct competition from foreign-produced goods, a decreasing pro-
portion of the labor force faces the consequences of increased traded goods
flOWS.2
2. This assumes that exports plus imports is a good measure of trade dependence. Under some
circumstances it will be. Under others it may understate trade dependence: e.g., when prices are
determined by the world market but there are no trade flows.
8 John M. Abowd and Richard B. Freeman
Table 2 The Changing lkade Content of the U.S. Labor Market (%)
1960 1970 1980 1987
(Exports + Imports)/GNP in sector?
Agricultureb 28.6 27.8 53.5 33.8
Mining' 22.6 25.0 76.6 45.3
Manufacturing 17.5 26.6 56.8 64.4
Total traded goods' 19.2 26.6 59.3 60.0
All other sectorsf 5.8 6.9 11.0 11.2
Total all sectors' 10.5 12.7 24.5 22.3
Percentage of GNP in sector:
Agriculture 4.2 2.9 2.8 2.1
Mining 2.5 1.8 3.9 1.9
Manufacturing 28.0 24.8 21.3 18.9
Total traded goods 34.7 29.6 28.0 22.8
All other sectors 65.3 70.4 72.0 77.2
Agriculture 3.1 1.8 1.8 1.6
Mining 1.2 0.9 1.2 0.7
Manufacturing 28.6 26.6 22.8 19.0
Total Traded Goods 33.0 29.2 25.7 21.3
All other sectors 67.0 70.8 74.3 78.7
(employment weighted)' 10.1 12.7 22.9 21.8
Sources: Exports and imports 1960, 1970, and 1980 from Bureau of the Census, U.S. Commodiiy
Exports and Imports as Related to Output 1981/80 (1983), table A. Exports and imports 1987
from U.S. Department of Commerce online data base of official statistics. GNP in sector from
National Income and Product Accounts, table 6.1 (extracted from CITIBASE). Full-time employ-
ment in sector from National Income and Product Accounts, table 6.7B (extracted from CITI-
BASE).
*Exports plus imports as a percentage of GNP originating in the industry group.
bAgriculture is SIC industry groups 01-09.
cMining is SIC industry groups 10-14.
dManufacture is SIC industry groups 20-39.
Traded goods are agriculture, mining, and manufactures.
'All other sectors include SIC industry groups 15-17 and 40-99. Exports (imports) in all other
sectors are defined as the difference between total exports (imports) and traded goods exports
(imports).
SExports and imports from the National Income and Product Accounts. Traded goods sectors
consist of manufacturing (SIC 20-39), mining (SIC 10-14), and agriculture (SIC 01-09).
hFull-time equivalent employees from the National Income and Product Accounts.
'Exports plus imports as a percentage of GNP originating in the industry group weighted by
employment in the industry group.
Percentage of employment in sector:h
(Exports + Imports)/GNP in sector
Capital Flows
The flow of capital across international borders is the most difficult flow to
measure and analyze. Net capital flows should equal the balance on current
accounts (plus allocations of special drawing rights), but, in fact, the two
differ significantly, requiring a statistical discrepancy line to produce the defi-
nitional equality. In terms of the effects on labor markets, we want to distin-
9 Introduction and Summary
guish a foreign capital investment that is a long-term job creating flow from a
short-run financial flow. If all net capital flows were of the former kind, public
focus on the disemployment effects of an imbalance on the current account
would be erroneous. If all the net capital flows were of the latter kind, by
contrast, such concerns might be valid, although the imbalance would even-
tually alter the real exchange rate and, in principle, correct itself. It is not easy,
however, to determine the degree to which capital flows fall along a spectrum
from long-term job-creating to short-term financial flows. Presumably, direct
foreign investment is job creating, while currency transactions are likely to be
short run, though we still need to know the “motive” and likely holding period
of these intermediate investments. A foreigner who buys stocks, corporate
bonds, or U.S. Treasury obligations or even leaves money in a U.S. bank
account for a long time can, through the flow of funds, produce as much long-
term investment in the United States as a foreigner who builds a plant.
We distinguish in table 3 between direct foreign investments in plant and
equipment, likely to be long run, and other forms of capital flows. As can be
seen in the table, both direct and indirect capital flows increased dramatically
in recent years. Net U.S. investment abroad (the change in U.S. assets abroad
from the international transactions accounts) increased from $4,099 million in
1960 (shown as a negative number in the table to reflect a capital outflow) to
$82,110 million in 1988. Net foreign investment in the United States (the
change in foreign-owned assets from the international transactions accounts)
increased from $2,294 million in 1960 to $219,299 million in 1988. Direct
U.S. investment abroad and direct foreign investment in the United States also
increased dramatically since 1960. By 1988, over a quarter of foreign invest-
ment in the United States consisted of direct foreign investment.
Are the international capital flows sizable or negligible in the context of the
U.S. economy? Table 3 also compares net foreign investment in the United
States and direct foreign investment in the United States to GNP and U.S.
gross investment. Direct foreign investment in the United States rises from
. l % of GNP and .4% of gross investment in 1960 to 1.2% of GNP and 9.2%
of gross investment in 1988. While Japanese investment in the United States
has received the most public attention, the percentage of direct foreign invest-
ment by country of ultimate beneficial ownership in table 3 shows that Euro-
pean direct investment is quantitatively much larger, although Japan increased
its share dramatically in the late 1980s.
2. Industrial and Geographic Patterns
Flows of goods, people, and capital occur differently by sector and area of
the economy. Some industries produced traded goods, while others do not.
Immigrants are overrepresented in some sectors and underrepresented in oth-
ers, and immigrants go to some areas of the country, and not to others. For
some long-term general equilibrium purposes, the sectoral division of the
10 John M. Abowd and Richard B. Freeman
Table 3 Capital Market Flows between the United States and the Rest
of the World
Investments (millions of dollars)
1960 1970 1980 1988
Net U.S. investment abroad'
Direct investment abroad
Net foreign investment in the
United Statesb
Direct foreign investment
Investment outlaysc
- 4,099 - 9,337 - 86,118 -82,110
- 2,940 - 7,590 - 19,222 - 17,533
2,294 6,359 58,112 219,299
315 1,464 16,918 58,436
NA NA 12,172 65,019
Net foreign investment in the
Net foreign investment in the
United StatesiGNP
United States/gross
investmentd
Direct foreign investment'GNP
Direct foreign investment'gross
investment
Percentage of direct foreign
investment by country'
Canada
Japan
Europe
West Germany
The Netherlands
United Kingdom
Rest of the World
Relative Figures (% of base)
1960 1970 1980 1988
.4
2.8
.1
.4
100.0
NA
NA
NA
NA
NA
NA
NA
.6
4.1
. I
I .o
100.0
NA
NA
NA
NA
NA
NA
NA
2.1
12.9
.6
3.8
100.0
16.1
4.9
62.9
11.7
13.6
25.2
16.1
4.5
34.7
1.2
9.2
100.0
16.0
21.8
52.5
2.1
3.0
33.1
9.7
Sources: U.S. International Transactions accounts from the Survey of Current Business (June
1989). Percentage distribution by country of ownership from Survey of Current Business, U.S.
Business Enterprises Acquired or Established by Foreign Direct Investors, 1980 and 1988. Na-
tional Income and Product Account data extracted from CITIBASE.
Note: NA = not available on a comparable basis.
*From U.S. assets abroad, net (increaseicapital outflow [ -1). in the U.S. International Transac-
tions accounts. Negative numbers indicate a net outflow. Direct investments abroad is a subac-
count of U.S. private assets, net.
bFrom foreign assets in the United States, net (increaselcapital inflow [ +I ) , in the U.S. Interna-
tional Transactions accounts. Positive numbers indicate a net inflow. Direct foreign investments
is a subaccount of other foreign assets in the United States, net.
CInvestment outlays from the U.S. Department of Commerce, Bureau of Economic Analysis,
Survey of New Foreign Direct Investment in the United States (1983).
Cross investment series from the U.S. National Income and Product Accounts, annual data.
'Percentage of Bureau of Economic Analysis survey investment outlays by country of ultimate
beneficial owner. Figures for 1988 are preliminary.
11 Introduction and Summary
flows is unimportant. For many short- and intermediate-term questions, how-
ever, sectoral flows are critical. To deal with this issue, the NBER developed
the Immigration, Trade, and Labor Markets Data Files (see Abowd, in this
volume). These data allow us to examine the pattern of trade across industry
lines over time, to contrast the industrial distribution of trade and the employ-
ment of immigrants, to determine the characteristics of workers in industries
with more or less trade and with sizable or limited employment of immigrants,
and to compare the geographic and industrial patterns of trade and immigra-
tion effects.
Table 4 uses the NBER immigration and trade data files to assess the varia-
bility of trade ratios across manufacturing industries. The table records the
mean, standard deviation, and coefficient of variation of trade ratios among
the 450 four-digit SIC manufacturing industries and of immigration ratios
Table 4 Variation in Trade and Immigratino Ratios for Manufacturing
Industries (employment weighted)
(Exports + (Exports -
Export/ Import/ Imports)/ Imports)/ Immigrants/
Shipments New Supply Shipments Shipments Labor Force
(%) (%'.)" (%)
1960:
Mean
Standard deviation
Coefficient of
variation
1970:
Mean
Standard deviation
Coefficient of
variation
1980:
Mean
Standard deviation
Coefficient of
variation
1985:
Mean
Standard deviation
Coefficient of
variation
Change, 196W30:
Mean
Standard deviation
4.27
5.91
1.38
5.62
7.07
1.26
10.31
11.38
1.10
8.48
11.09
1.31
5.50
8.37
2.30
4.61
2.01
4.59
6.11
1.33
7.41
8.51
1.14
10.94
11.38
1.04
5.38
6.42
7.00
10.17
1.45
11.08
12.91
1.17
19.94
23.59
1.18
24.95
52.17
2.11
12.62
17.34
1.56 8.48
9.87 3.66
.43
.20 7.13
11.40 3.37
.47
.75 1.96
18.56 4.34
.55
- 7.95
50.78
- 1.58 - .40
14.03 2.82
Source: NBER Immigration, Trade, and Labor Markets Data Files (see Abowd, in this volume).
Nofe: All ratios are stated as percentages of the relevant base. The statistics are averages over
four-digit SIC industries using the annual employment in the industry as the weight. There are
450 SICs with valid immigrant ratio data and 430 SICs with valid import and export data.
*New supply is the sum of shipments and imports.
12 John M. Abowd and Richard B. Freeman
among those industries. The table shows considerable variation in ratios of
trade (functions of exports X and imports M) to shipments (S) across indus-
tries. The relatively stable coefficient of variation in the exports-to-shipments
ratio contrasts with a declining coefficient of variation in the imports-tc+new
supply ratio, implying an unchanged concentration of the former compared to
an increasing concentration of the latter. Among the sectors with the largest
increase in trade content are footwear except rubber, electrical equipment, and
electronic resistors. Imports grew especially rapidly in footwear, and exports
grew especially rapidly in electrical equipment. Both exports and imports in-
creased in electronic resistors. Comparing the trade and immigration ratios
across industries, one striking fact emerges: trade ratios are much more vari-
able among sectors than are immigration ratios. This has important conse-
quences for the way in which NBER and other researchers study the effects of
trade on the labor market: focusing on differences across industries.
A very different pattern emerges when we consider regional differences in
trade ratios, immigrant flows, and foreign direct investment in the United
States. Here we find exactly the opposite: immigration ratios vary much more
across regions than across industries. This fact leads NBER and other re-
searchers to study the effect of immigration on the labor market by focusing
on differences across areas.
The geographic concentration of immigration is documented in table 5. The
table shows INS figures on the number of immigrants declaring selected stan-
dard metropolitan statistical areas (SMSAs) as their intended residence from
1976 to 1979 and the contribution of that flow to the growth of the labor force.
Table 5 displays the ten SMSAs with the largest percentage of foreign born in
the area. These gateway cities absorbed a very substantial fraction of all im-
migrants who entered the United States in the four-year period illustrated
(comparable data were not collected for 1980). Table 6 shows Census of Pop-
ulation figures on the percentage of the labor force that are immigrants by
SMSAs in 1970 and 1980. What stands out in these tables is the substantial
concentration of immigrant flows by SMSA.
How might the flow of illegal immigrants into the United States change the
picture of geographic concentration shown in tables 5 and 6? Given the con-
centration of illegal aliens in California, where Warren and Passel estimate
that 50% of illegals counted in the Census are located, the concentration of
immigrants would become even more dramatic.
Table 7 presents data on all our flows by state. Columns 1-3 give data from
the Census of Population for immigrants as a percentage of the population in
1970 and 1980 and in 1980 adjusted for the likely undercount of illegals in the
Census on a state-by-state basis using the estimates in Passel and Woodrow
(1984). While the geographic diffusion of the stock of immigrants is lower
than the diffusion of the flow of new immigrants among SMSAs, there is still
considerable variation across areas.
Columns 4-6 of the table turn from immigrant to trade figures. As data on
13 Introduction and Summary
Table 5 Flows of Immigrants into Selected Standard Metropolitan Statistical
Areas (SMSAs)
Immigrants Declaring Change in Labor Estimated Immigrant
SMSA as Intended Force from Contribution to Labor
SMSA Place of Residence’ 1976 to 1979b Force Growth (%)
Miami
Los Angeles
New York City
El Paso
Newark
Washington, D.C.
Houston
Cleveland
Philadelphia
Dallas
79,099
74,515
247,052
13,053
8,879
8,359
23,868
3,800
10,571
10,735
54,233
254,000
38,000
8,836
40,738
166,193
255,367
38,108
85,016
220,331
73.3
14.7
326.8
74.3
11.0
2.5
4.7
5.0
6.3
2.4
Sources: Number of immigrants from the Statistical Yearbook of the Immigration and Naturali-
zation Service, 1976-79: table number varies; table title “Immigrants Admitted by Specified
Countries of Birth and Rural and Urban Area and City.” Change in the labor force from the
Bureau of the Labor Statistics, Employment and Earnings, various issues. The immigrant labor
force-tc+immigrant population ratio was estimated from the 1980 Census of Population Detailed
Population Characteristics U.S. Summary, sec. A-U.S. PCSO-1-D1-A. Total immigrants is from
table 254, “Citizenship and Year of Immigration for Foreign Born Persons by Country of Birth.”
Immigrants in the labor force is from table 255, “Selected Economic and Social Characteristics
by Nativity.”
‘Number of immigrants who declared the SMSA as the intended place of permanent residence
during the period from 1 October 1975 to 30 September 1979. SMSAs are listed in descending
order of percentage foreign born in the area.
Thange in the size of the labor force from 1976 to 1979, inclusive.
CEstimated as 50.3% of col. 1 divided by col. 2.
exports or imports by geographic location are unavailable, our estimates of
the trade content of a state’s industry mix are obtained by weighting industry
trade ratios according to the industrial distribution of state labor forces as
follows:
T, = CWi j Ti ,
i
where T = relevant state trade ratio, W, = proportion of workers in state j
who work in industry i, and Ti = national trade ratio in industry i.
In contrast to the wide variation in immigration ratios across states, the
trade ratios differ relatively moderately, except for the net export ratio
([X - M] / S) . For example, the five states whose industry structures have the
highest import ratios ( M/ [ S + MI ) have an average value of 8.8, compared to
an average figure of 3.8 for the five states with the lowest import ratios. While
there are surely individual localities that are greatly sensitive to trade, the
implication of the table is that trade flows are unlikely to have great effects on
local labor markets, except, possibly, where there is a substantial net export
ratio ([X - M] / S) , as in Alaska.
14 John M. Abowd and Richard B. Freeman
Table 6 Immigrants as a Percentage of the Labor Force Selected SMSAs
~ ~~ ~ ~~~
I970 1980
Atlanta 1.3 2.9
Baltimore 3.9 3.5
Boston 10.3 10.6
Chicago 10.2 11.6
Dallas-Fort Worth 2.4 4.8
Detroit 8.4 6. 3
Houston 3.3 8.3
Los Angeles 13.6 24.2
Miami 27.9 41.2
New York City 18.0 24.0
Philadelphia 6.1 5.0
St. Louis 2.3 2.3
San Francisco 13.7 16.2
Washington, D.C. 6.0 9. 0
Pittsburgh 4.3 2.7
Sources: Based on individual data from the 1970 Census of Population and Housing 1/100 Public
Use County Group Sample and the 1980 Census of Population and Housing Public Use Microdata
A Sample.
Note: The numerator is the number of immigrants in the labor force in the SMSA indicated. The
denominator is the number of individuals in the labor force in the SMSA. SMSA definitions in
the 1970 and 1980 Censuses of Population were made comparable by selecting the appropriate
area and subarea codes (1970) and SMSA codes (1980).
The popular and business press are filled with stories about the decision of
Japanese and other foreign investors to locate plants in certain regions of the
country as opposed to others. Column 7 of table 7 presents data from the 1980
Benchmark Survey of Direct Foreign Investment in the United States (U.S.
Department of Commerce 1983) on the proportion of the private work force
employed in foreign-owned enterprises among the states. It shows consider-
able variation in employment in foreign-owned affiliates, with a range far ex-
ceeding that for trade shares, and a regional pattern differing greatly from that
for immigrant employment.
3. Characteristics of Workers in Sectors Affected by
Internationalization
To evaluate the type of workers most likely to be affected by trade or immi-
gration, we have performed a two-part analysis. First, we tabulated the aver-
age characteristics of workers by employment. Second, we calculated corre-
lation coefficients between worker characteristics by industry and the relevant
trade or immigrant worker ratio.
Table 8 presents the results of the first analysis with sectors divided between
traded goods and nontraded goods, between export- and import-intensive
15 Introduction and Summary
Table 7 Geographic Distribution of Immigration, Bade, and Direct Foreign
Investment
Immigrant % % of U.S.
of Labor Force % % Net Affiliate
Export % Import Export Employment in
Adjusted Ratio Ratio Ratio Total Private
1970 1980 1988 XIS MIS + M X - M/ S Employmentb
Alabama
Alaska
Arizona
Arkansas
California
Colorado
Connecticut
Delaware
District of
Florida
Georgia
Hawaii
Idaho
Illinois
Indiana
Iowa
Kansas
Kentucky
Louisiana
Maine
Maryland
Massachusetts
Michigan
Minnesota
Mississippi
Missouri
Montana
Nebraska
Nevada
New Hampshire
New Jersey
New Mexico
New York
North Carolina
North Dakota
Ohio
Oklahoma
Oregon
Pennsylvania
Rhode Island
South Carolina
South Dakota
Tennessee
(conrinued)
Columbia
.7 1.1
5. 9 4.7
5.4 6.1
.5 1.1
11.4 16.3
3.8 4.0
10.6 8.9
2.7 3.1
8.0 6.9
10.0 11.5
1.2 1.9
11.4 15.3
1.8 3.0
7.3 8.0
2.2 1.9
1.8 1.5
1.3 2.1
.2 .9
1.4 2.3
5.8 3.5
4.4 5.0
10.6 8.5
6.7 4.3
3.1 2.3
.2 1.0
1.8 1.8
1.6 2.2
1.4 1.9
6.2 7.9
5.1 4. 3
10.6 11.2
3.3 4.1
13.9 14.8
.8 1.4
2.7 2.0
3.5 2.7
1.1 2.0
4.4 4.1
4.1 3.1
8.8 8.8
.7 1.6
1.2 1.3
1.1 1.2
1.2
4.9
6.9
1.2
19.3
4.5
9.0
3.2
8.7
12.2
2.1
15.2
3.5
8.9
2.0
1.6
2.3
1.1
2.4
3.4
5.7
8.8
4.4
2.4
1.1
1.9
2.2
2.1
8.6
4.3
11.6
4.9
15.8
1.5
2.1
2.8
2.3
4.5
3.1
9.0
1.7
1.3
1.3
8.6
14.9
13.8
8.4
11.8
10.4
14.8
12.3
2.5
10.2
9.2
5.5
9.4
11.5
10.3
13.1
14.6
9.9
9.1
8.8
8.2
12.1
11.3
12.5
8.1
10.1
11.4
9.8
8.6
13.4
9.6
7.5
10.7
7.7
8.2
10.9
11.6
11.0
9.1
11.1
9.5
9.4
9.0
7.1
2.9
6.6
7.5
6.4
6.7
7.2
6.6
1.6
6.2
7.2
7.1
5.4
6.8
8.2
6.6
6.0
7.2
6.1
9.4
6.6
7.6
10.4
6.0
7.6
7.5
4.8
6.6
7.6
7.8
6.4
7.6
1. 3
6.2
5.5
7.5
6.7
6.4
7.3
8.1
6.8
4.5
7.7
.8
11.7
6.5
-.2
4.8
2.9
6.8
4.9
.8
3.2
1.2
- 2.6
3.3
4.0
1. 1
5.8
7.9
1.9
2.4
-2.3
.8
3.5
- . 9
5.9
- .4
1.5
6.2
2.5
- .2
4.6
2.4
- 1.2
2.5
.9
2.1
2.4
4.2
3.9
1 .o
2.0
2.0
4.3
.3
2.0
6.9
1.7
2.2
2.5
1.9
2.7
3.9
.6
2.1
3.6
4.6
1.3
2.8
2.6
2.0
1.7
2.3
2.9
2.7
3.1
4.0
2.3
2.0
1.4
1.8
.8
1 .o
1.3
3.6
4.6
2.0
2.9
3.2
1.2
2.2
2.0
1.2
2.8
2.0
5.3
.6
2.9
16 John M. Abowd and Richard B. Freeman
Table 7 (continued)
Immigrant % %of U.S.
of Labor Force % % Net Affiliate
Export % Import Export Employment in
Adjusted Ratio Ratio Ratio Total Private
1970 1980 198W XIS MIS + M X - MIS Employmentb
Texas
Utah
Vermont
Virginia
Washington
West Virginia
Wisconsin
Wyoming
Summary:
Mean of
Mean of
Difference
top 5
bottom 5
3.5 6.5 7.6 12.0 6.6
2.9 4.0 4.5 11.8 6.6
7.4 4.5 4.5 14.2 6.5
2.4 3.7 4. 2 8.5 6.6
5.9 6.1 6.5 15.2 6.1
1.2 1.1 1.1 10.5 7.1
3.3 2.5 2.6 11.5 7.1
1.6 1.9 2.1 9.6 6.0
13.5 13.8 14.8 14.7 8.8
.5 1.0 1.1 6.3 3.8 -
13.0 12.8 13.7 8.4 5.0
4.7 2.7
4.5 2.2
7.1 3.6
1.3 2.2
8.5 1.5
2.5 3.6
3.5 3.4
3.0 1.7
8.4 5.1
1.5 1 .o
9.9 4.2
Sources: Immigrant, export, and import ratios are from the NBER Immigration, Trade, and Labor Mar-
kets Data Files (see Abowd, in this volume). Employment in U.S. affiliates is from U.S. Department of
Commerce (1983), Foreign Direct Investment in the United States, 1980.
The adjustment is based on Passel and Woodrow’s (1984) table 1, “Estimates of Undocumented Aliens
Counted in the 1980 Census and Legally Resident Aliens by State of Residence and Period of Entry.”
bFrom the 1980 Benchmark Survey of Foreign Direct Investment in the United States, “U.S. Affiliate
Employment by State.”
manufacturing (reported separately for high-import and high-export indus-
tries), and by high and low immigrant-worker ratios.
The differences between traded and nontraded sectors reflect basic eco-
nomic differences between characteristics of workers in goods and services
industries: workers in nontraded goods are better educated, younger, more
likely to be female, and less likely to be union than workers in traded goods.
In addition, wages and GNP per worker in this sector are lower than in traded
goods. While immigrant ratios are nonnegligible in nontraded sectors, they
are lower than in the traded goods sector, indicating that the traded goods
sector is more directly tied to the international economy by flows of people as
well as by flows of goods.
Decomposing manufacturing into high (top quintile) export and import to
shipments sectors, we find striking differences in the characteristics of the
work forces. These differences indicate which workers are more or less likely
to be directly positively or negatively affected by trade. The principal differ-
ences among workers revealed by the table are that export sectors have pro-
portionately more educated workers, fewer blacks, and strikingly fewer
female workers than import-intensive industries; that high-export manufactur-
17 Introduction and Summary
Table 8 Average Characteristics of the Labor Force in 1980 (industry averages
weighted by employment)
Production Value
With W o Workers Earnings Added
College (8) (%) (%) (%) (%) Union(%) Worker Worker
Years of Black 16-24 Female Immigrant Who Are per per
Traded goods 18.7 9.3 21.4 32.4 7.8 36.8 16.9 32.6
Manufacturing 19.3 10.2 20.2 35.9 8.2 43.7 17.4 28.6
High exports
High imports
(top 20) 25.5 1. 5 18.3 29.9 7.5 41.6
(top 20) 15.4
10.9 20.2 44.8 10.4 40.6
Nontraded goods
All industries 28.6 10.0 24.1 46.8 6.5 25.0
/labor force 36.2 9.4 17.4 41.2 3.4 33.0
/labor force 20.9 14.1 22.6 52.4 12.0 24.1
and services 31.2 10.2 24.9 50.6 6.1 21.8 13.0 21.8
Low immigrant
High immigrant
~
Source: Calculated from the NBER Immigration, Trade, and Labor Markets Data Files (see Abowd, in
this volume).
Thousands of dollars per worker.
ing sectors have lower proportions of immigrants than high-import manufac-
turing (though their ratio still exceeds the economy-wide average). The frac-
tion of blue-collar workers unionized in the sectors does not, by contrast,
show any noticeable differences.
Turning to the characteristics of workers by immigrant ratios, we find that
high-immigrant-ratio sectors tend to have less educated workers, relatively
more blacks, relatively more women, relatively more young workers, and rel-
atively fewer union workers. With the exception of the unionization pattern,
these differences mirror those between export- and import-intensive industries
in manufacturing.
Table 9 records the correlation coefficients between mean characteristics of
workers and export and import ratios and the net export ratio in manufacturing
and between the mean characteristics of workers and immigrant ratios in man-
ufacturing, nonmanufacturing , and all industries. The correlations confirm
the evidence given in table 8, revealing a strikingly high positive correlation
between the percentage of workers who are women and the percentage of
workers who are immigrants in manufacturing industries. The correlations
also show that female, black, and immigrant workers tend to be concentrated
in industries with negative net exports and that educated workers tended to be
in industries with positive net exports. These calculations suggest that both
trade and immigrant flows may have an especially large effect on the female
work force, especially in manufacturing.
18 John M. Abowd and Richard B. Freeman
Table 9 Correlation Coefficients for Immigration and lkade Ratios with
Various Labor Force Characteristics, 1980 (employment weighted)
~~~~
Nonmanu- All
Manufacturing Only facturing Industries
Immigrant/ Immigrant! Immigrant/
XIS Mi(S + M) Labor Force (X - M) / S Labor Force Labor Force
2 Years of college
% Black
% 16-24
% Female
% Immigrant
% Unionized
Eamingsiworker
S hipmentsiworker
.31 - . I 3
- .25 .08
-.29 -. 14
- .22 .I1
-. 16 .22
.05 .oo
.32 -. 06
. I 3 -.09
- .21
.06
. I6
.I1
1 .oo
- .36
- .53
- .29
.38 - .01 - .21
- .29 . I 2 .15
- .18 .29 .18
- .32 .20 .25
- .30
.05 - .35 -. 18
.35
. I9
Source: NBER Immigration, Trade, and Labor Markets Data Files (see Abowd, in this volume).
Note: The statistics are painvise correlation coefficients computed using the percentage of total
employment in the industry as the weight.
Finally, we consider the characteristics of workers in foreign- and U.S.:
owned businesses operating in the United States. Table 10 displays a collec-
tion of comparisons from the 1974, 1980, and 1987 Benchmark Surveys of
Foreign Direct Investment. The 1980 survey is the most detailed, and it re-
veals that employees of nonbank U.S. affiliates of foreign companies are
about as likely as employees of U.S.-owned companies to be unionized. Fur-
ther, 1980 hourly earnings levels are somewhat higher, though sales per em-
ployee levels are similar in affiliates. Because the benchmark surveys are not
comparable in the universe (banking affiliates are included in 1974 but ex-
cluded in 1980 and 1984) and in the summary data tables (employment is not
reported by industry in 1974), it is difficult to discern trends in the compari-
sons of foreign-owned to U.S.-owned businesses. It seems likely that the dif-
ferences are not substantial, and this conclusion is supported by other research
(Leonard and McCulloch, in this volume).
4. Findings of the NBER Project
Motivated by the internationalization of the American labor market de-
scribed in sections 1-3, the NBER undertook the research project whose re-
sults are given in this volume. The first part of the project studied the factors
that influence the number and characteristics of immigrants and their location
in the country, including the undocumented, largely Mexican aliens who have
aroused so much public concern. The second part of the project examined how
immigration and trade affect the wages and employment of American work-
ers. The third part of the project added an international comparative dimen-
19 Introduction and Summary
Table 10 Selected Business and Employment Characteristics of U.S. Affiliates
of Foreign Companies, by Industry
1974 1980 1987
Number of employees (thousands)’ 1,083 2,034 3,160
Percentage of civilian labor force 1.2 1.9 2.6
Percentage union in U.S. affiliates
Percentage union in all U.S. businesses
Average hourly earnings for production work:
29.3
25.2
Workers in U.S. affiliates
Workers in all U.S. businesses
(manufacturing) 7.85
(manufacturing) 7.27
Sales per employee (thousands) of dollars):b
U. S. manufacturing affiliates 88.86 137.65
U. S. manufacturing businesses 91.27 125.63
Traded goodsc 3.3 5.4 1.7
manufacturing 2.8 5.0 8.4
Selected nontraded goodsd 9.9 3.0 3.6
Sources: Survey of Current Business, “Benchmark Survey of Foreign Direct Investment in the
United States, 1974” (May 1976). U.S. Department of Commerce (1983), Foreign Direct Invest-
ment in the United States, 1980. Survey of Current Business, “U.S. Affiliates of Foreign Com-
panies: 1987 Benchmark Survey Results” (July 1989). U.S. shipments and employment data from
Survey of Current Business, various issues.
”11 U.S. affiliates in 1974; nonbank U.S. affiliates in 1980, 1987.
bManufacturing industries only, thousands of dollars per employee.
‘Agriculture, mining, and manufacturing only.
dWholesale trade, retail trade, finance (except banks), insurance, and real estate.
Percentage of total employment:
sion by studying immigration, trade, and the labor market in two other coun-
tries that are major immigrant recipients, Canada and Australia. Canada is of
particular interest for several reasons: Canadian immigration policies histori-
cally have stressed job skills as a condition for entry to a greater extent than
the United States does; Canada had a significant balance of trade surplus with
the United States in the 1980s; and Canada has long depended on foreign
(largely U.S.) capital to employ a large share of its work force. Australia is of
interest because immigrants constitute an exceptionally large proportion of its
work force, raising issues about how immigration affects the macroeconomy,
and because its protectionist trade policies contrast with the free trade policies
of the United States. The differing experiences among the United States, Can-
ada, and Australia indicate the degree to which different labor market institu-
tions and economic policies can condition the effect of immigration and trade
on economic outcomes.
The project researchers used different strategies to study the flow of immi-
grants, the effects of immigration on labor market outcomes, and the effects
of trade on those outcomes. The studies that focus on the flow of immigrants
20 John M. Abowd and Richard B. Freeman
compare the labor market and migration behavior of individual immigrants
since it is the immigrant decisions and performance in the labor market that
are at issue. The studies of the effects of immigration on wages and employ-
ment compare local labor markets that have different immigrant shares in the
work force. The principal reason for focusing on local markets is that immi-
grants are concentrated by geographic area, constituting large and increasing
proportions of the work force in gateway cities but negligible proportions
elsewhere, as indicated in tables 5 and 6. By contrast, the studies focusing on
trade examined the effect of trade on industry labor markets. This is because
the export and import components of economic activity vary and change
greatly among industries, suggesting that the first-order labor market effects
of trade occur at the industry level.
Most of the findings are based on data from government surveys such as the
Census of Population and the Census of Manufactures. To answer certain
questions, however, researchers developed new data sets, ranging from one
that links import prices to collective bargaining contracts in Canada (Abowd
and Lemieux) to a survey of illegal Mexican immigrants in the San Diego area
(Borjas, Freeman, and Lang). Because trade, immigration, and labor market
data are collected using different standards by diverse government surveys,
researchers developed the industry-based trade and labor markets data file for
U. S. manufacturing industries from the 1950s through the 1980s and the area-
based immigration and labor market data file for the 1960s, 1970s and 1980s
(Abowd; Altonji and Card; and LaLonde and Topel).
Studying immigration and trade by comparing outcomes across individu-
als, areas, or industries differs from most studies in international trade, where
researchers use general equilibrium models to make inferences about the eco-
nomic effects of immigration, trade, and capital flows. While there is no in-
herent conflict between these two types of research approaches (some of the
studies use input-output and trade models; (e.g., Kuhn and Wooton; Collins),
our decision to concentrate on individuals and markets was a conscious one
that conditions the issues we address and our major findings. Our approach
pins down the first-order effects of trade and immigration on the economic
well-being of the groups most affected by the internationalization of the U.S.
labor market but does not yield estimates of the broader benefits of trade or
immigration to the overall society. The approach has the advantage of basing
inferences on the great variation in the experiences of individuals, areas, or
industries and of requiring less formal structure than general equilibrium anal-
yses, at the cost of being unable to answer questions about how things may
work out for the society as a whole in the long run.
As a broad generalization, the American labor market adjusted well to im-
migrant flows, absorbing immigrants into local area work forces with little
redistributive losses to natives, but it had greater difficulty adjusting to the
surge of imports, which produced some noticeable losses to natives in affected
industries. Still, industry wages were as flexible to changes caused by trade as
21 Introduction and Summary
to changes caused by domestic factors, falling where imports reduced domes-
tic production and thus buffering employment to some extent. By contrast, in
Australia, industry wage responsiveness to imports was limited, and the gov-
ernment sought to protect labor through import restrictions. The research
highlights the supply responsiveness of immigrants to economic and political
conditions and to foreign as well as to American immigration policies in the
context of a “world market for immigrants.”
Immigrant Flows
1. The flow of illegal immigrants to the United States, while sizable, falls
far short of the huge numbers often reported in the media. NBER estimates of
Mexican-born illegal immigrants based on the number of deaths and births of
Mexican-born persons in the United States, Mexican surveys of returned mi-
grants, and analyses of apprehension statistics that take account of the fact
that apprehensions are determined by Border Patrol activity as well as by im-
migrant flows support the claim of Census Bureau demographers that the 1980
Census enumerated over half the illegal immigrants. The number of illegal
Mexican immigrants in 1980 was on the order of two million rather than ten
to twelve million. Moreover, most of the likely illegal Mexican immigrants
counted in the Census have a family composition and type. of employment
similar to those of legal immigrants (Borjas, Freeman, and Lang; Diez-
Canedo). Consistent with a factor endowment explanation of immigration,
most illegal aliens are unskilled.
2. The characteristics of immigrants are influenced significantly by the eco-
nomic and political situation in the home countries and by the attractiveness
of the United States in the “world market for immigrants,” where the United
States competes with other immigrant-recipient countries such as Canada and
Australia. All else the same, workers with a high earnings potential are espe-
cially likely to migrate to the United States from a country with an egalitarian
wage structure (where they cannot easily make high earnings), while workers
with a low earnings potential are especially likely to migrate from a country
with great wage inequality. The 1965 changes in U.S. immigration policy pro-
duced a wave of immigrants whose labor market skills were lower relative to
those of native Americans than was true of earlier waves of immigrants, who
did especially well in the labor market relative to natives (Borjas). Changes in
Canadian immigration laws produced a similar pattern of declining skills in
the late 1970s. Australia, by contrast, attracted immigrants who did well com-
pared to natives through 1980 (Borjas; Beggs and Chapman; Bloom and Gun-
derson) .
3. New immigrants to the United States are as mobile across geographic
areas as natives, on average, but their mobility has not led them to spread out
across the country. Instead, they move to cities where their fellow countrymen
reside in large numbers. The tendency of immigrants to cluster dominates
such economic incentives as differences in unemployment rates or welfare
22 John M. Abowd and Richard B. Freeman
benefits across areas in determining immigrant migration flows (Bartel and
Koch). Cuban, Mexican, and Puerto Rican immigrants and natives who move
from cities with a high proportion of persons of their ethnic background to
cities with a low proportion of persons with their ethnic background have
roughly similar earnings and employment experiences as their peers who
move from cities with a low proportion of persons of their ethnic background
to cities with a high proportion of persons of that ethnic background (Tienda
and Wilson). The direct advantages and costs of immigration thus continue to
be borne by gateway cities, while the persistent geographic concentration of
immigrants may reduce their economic progress and rate of long-run assimi-
lation into the broader society.
The Effects of Trade and Immigration on Labor Markets
4. Increased immigration has a modest adverse effect on the wages of the
immigrants themselves and on the wages of earlier waves of immigrants, but
it has only a modest effect on the wages of the young black and Hispanic
Americans who are likely to be the next closest substitutes (LaLonde and To-
pel). Neither the employment nor the wages of less educated black and white
natives worsened noticeably in cities where immigrant shares of the popula-
tion rose in the 1970s. On the positive side, there is some evidence that, in
cities with more immigrants, employment grew more rapidly or declined
more slowly in low-wage industries where immigrants tended to find jobs and
that less-skilled natives moved into better jobs (Altonji and Card). The broad
implication is that immigrants have been absorbed into the American labor
market with little adverse effect on natives.
5. “General equilibrium analysis” of the potential effects of immigration on
the labor market through changes in sectoral outputs and prices further sup-
ports the claim that immigration has not harmed American labor. Indeed, the
concentration of immigrants in import-intensive, traded goods manufacturing
industries and the distribution of capital and native labor among export, im-
port, and nontraded goods sectors suggests that increased immigration may
actually benefit native labor, at least in the short run (Collins; Kuhn and
Wooton).
6. Wages in industries where sales are adversely affected by trade tend to
decline relative to wages in other industries, just as do wages in industries
where sales are adversely affected by domestic market developments, buffer-
ing to some extent the loss of jobs in industries facing large increases in im-
ports. Unionized sectors make greater wage adjustments than nonunion sec-
tors, apparently because workers in those industries earn above-market wages
that can be reduced to save jobs whereas nonunion wages are closer to com-
petitive levels (Abowd and Lemieux; Freeman and Katz). Once workers are
dislocated by trade, however, they appear to have greater difficulty finding
work than workers displaced for other reasons (Kruse 1988).
7. Foreign-owned firms employ nearly 3% of American workers. Despite
23 Introduction and Summary
the concern about foreign ownership, wages of production workers appear to
be higher in foreign-owned enterprises, and rates of unionization are not dif-
ferent from domestic-owned companies. Moreover, notwithstanding all the
attention given to Japanese firms, the bulk of direct foreign-owned enterprises
in the United States are European. Foreign-owned firms use substantially
more highly educated research-and-development employees (Leonard and
McCulloch).
Comparative Experiences: Canada and Australia
8. Canadian and Australian immigration policies, traditionally based on la-
bor market skill considerations, have moved toward admitting immigrants for
reasons of family unification, as in the United States. Since 1974, Canada has
given preferential treatment to persons with close relatives in the country as
well as to those who fulfill certain labor-market criteria. As a consequence,
immigrants coming to Canada after the mid-1970s apparently do worse in the
labor market relative to natives than earlier immigrant cohorts (Bloom and
Gunderson). Australia admitted immigrants on the basis of a labor market
point system from the 1970s through the early 1980s, with the result that the
labor market skills of Australian immigrants did not drop in the 1970s relative
to those of native workers, as in the United States and Canada (Beggs and
Chapman). The implication is that immigration policies significantly affect
the type of immigrants and their labor market performance.
9. Low-skill immigrants are relatively more highly paid in Australia than in
the United States. There are three reasons for this: (1) wage differentials by
occupation are smaller in Australia than in the United States; (2) immigrants
are more highly unionized in Australia than in the United States; and (3) Aus-
tralia has enacted trade policies that protect industries employing low-skill
immigrants. Australian protection of immigrant-intensive industries produces
relatively higher prices for the outputs of those sectors and extracts a sizable
social cost on the order of 50% to 100% of the wage bill in footwear, clothing,
and textiles (Gregory, Anstie, and Klug).
10. In Canada, changes in import and export prices, which reflect the pres-
sure of the international economy on producers, have significant effects on the
employment of workers covered by collective bargaining agreements. In-
creases in import prices, which shift demand to domestic producers, and in
export prices, which reflect greater returns from increasing sales overseas, are
associated with increases in employment of sizable magnitudes. In both the
United States and Canada, unionized employment is more sensitive to import
competition than to a comparable reduction in domestic production (Abowd
and Lemieux) .
Concluding Remarks
Perhaps the most intriguing finding of the Immigration, Trade, and Labor
Markets studies is the apparently different direct effect of immigration and
24 John M. Abowd and Richard B. Freeman
trade on workers in the affected labor markets. Whereas immigration does not
discernibly reduce the wages and employment of less-skilled native workers
in immigrant-intensive localities, imports reduce the pay as well as employ-
ment of workers in heavily affected industries. Why? What might account for
this differential effect?
While we cannot give a conclusive and quantifiable answer, the general
factors likely to underlie the differences do seem clear.
First, differences in the concentration and magnitude of imports and immi-
gration in affected areas certainly have an influence. In the ten industries with
the largest growth of import shares of sales from the 1960s to the mid-l980s,
import shares rose by 14% of domestic sales to 73% of domestic sales on
average. By contrast, in the ten standard metropolitan areas with the greatest
1970-1980 growth of immigrants relative to the work force, new immigrants
averaged 20% of the 1970 work force. Employment fell by 56% in the trade-
affected industries, while employment of natives increased in all the
immigrant-affected localities save for New York City.
Second, immigration has potential offsetting effects on the demand for la-
bor in affected areas, while trade has no such effects on demand for labor in
affected industries. Immigrants purchase goods and services in the area in
which they work, raising demand for labor. Immigrant skills are also likely to
complement the skills of some native workers, raising demand for them. By
contrast, even with balanced trade, workers in an industry facing a surge of
imports are unlikely to benefit directly from offsetting export-created demand
for labor or from complementary demands for native labor in retail and whole-
sale trade.
Third, it is possible that the concentration of immigrants in gateway cities
did not increase the labor supply in those areas by as much as the immigration
numbers would suggest. This would be the case if natives adjusted their
choice of location of residence to take account of the immigrant flows. The
flow of immigrants to, say, Los Angeles could have deterred midwesterners or
southerners from migrating there or impelled natives to move elsewhere, so
that the labor force in the city was not all that different from what it would be
absent immigration. No such mitigating response exists for trade-affected in-
dustries.
All these considerations suggest that the 1980s import surge caused a
greater "shock" in affected labor markets than did the influx of immigrants
and, thus, created greater difficulties of labor market adjustment. Trade upset
the demand-supply balance in industry labor markets more than immigration
upset the demand-supply balance in local labor markets.
One additional factor may also contribute to the greater effect of imports
than immigration on affected workers. In some industries, worker skills and
earnings are industry specific, so that shocks cause greater economic losses to
the affected employees. Consequently, labor mobility may be easier for work-
25 Introduction and Summary
ers facing immigrant competition in a local labor market than for workers
facing import competition in a trade-affected industry.
In summary, while trade and immigration may have the same long-run eco-
nomic effects on an economy, there are good reasons (and, more compelling,
empirical evidence) that they have different transitional costs for affected
workers.
References
CITIBASE. 1978. Citibank Economic Database [machine-readable magnetic data file,
1946 to present]. New York: Citibank.
Council of Economic Advisers. 1990. Economic Report of the President. Washington,
D.C.: U.S. Government Printing Office, February.
Kruse, Douglas. 1988. International Trade and the Labor Market Experience of Dis-
placed Workers: Evidence from the Displaced Worker Survey. Industrial and Labor
Relations Review 41, no. 3 (April): 402-17.
Passel, Jeffrey S. , and Karen A. Woodrow. 1984. Geographic Distribution of Un-
documented Immigrants: Estimates of Undocumented Aliens Counted in the 1980
Census by State. Washington, D.C.: US . Bureau of the Census, Population Divi-
sion.
U.S. Department of Commerce. Bureau of the Census. 1983. US. Commodify Exports
and Imports as Related to Output 1981180; Washington, D.C.: U.S. Government
Printing Office, November.
. Annual. Statistical Abstract of the United States. Washington, D.C.: US .
Government Printing Office.
. Decennial. Census of Population Detailed Population Characteristics U.S.
Summary. Washington, D.C.: U.S. Government Printing Office.
U.S. Department of Commmerce. Bureau of Economic Analysis. 1983. Foreign Di-
rect Investment in the United States, 1980. Washington, D.C.: U.S. Government
Printing Office.
. Monthly. Survey of Current Business. Washington, D.C.: U.S. Government
Printing Office.
U.S. Department of Labor. Bureau of Labor Statistics. Monthly. Employment and
Earnings. Washington, D.C.: U.S. Government Printing Office.
U.S. Department of Justice. Immigration and Naturalization Service. Annual. Statis-
tical Yearbook of the Immigration and Naturalization Service. Washington, D.C.:
US . Government Printing Office.
Warren, Robert, and Jeffrey S. Passel. 1987. A Count of the Uncountable: Estimates
of the Undocumented Aliens Counted in the 1980 Census. Demography 24 (Au-
gust): 375-93.
This Page Intentionally Left Blank
Immigrant Flows and
Performance in the
United States
This Page Intentionally Left Blank
1 Immigration and Self-selection
George J. Borjas
The insight that migrants may be systematically different from persons who
do not choose to migrate has long played an important role in sociological and
historical studies of the immigration phenomenon (see, e.g., the studies con-
tained in Jackson 1969). The selectivity hypothesis has also played a major
role in the modem economic literature that analyzes how immigrants do in the
U.S. labor market. For example, the early studies of Chiswick (1978) and
Carliner (1980) invoke the assumption that immigrants are positively selected
from the population of the countries of origin to explain the cross-sectional
empirical finding that immigrant earnings (after a short time period) “over-
take” the earnings of natives with the same observed socioeconomic charac-
teristics, such as age and education.’
My recent work in this area (Borjas 1985, 1987) has addressed two related
questions raised by the early studies. Since most of the literature analyzing
immigrant earnings focuses on the study of single cross-sectional data sets,
my 1985 paper raised the possibility that the overtaking findings could be due
to the fact that cross-sectional regressions confound aging and cohort effects.2
The positive correlation between immigrant earnings and years of residence
in the United States observed in the cross section could arise because immi-
grants “adapt” rapidly to the U.S. labor market or because earlier waves of
immigrants differ in substantial ways (labor market productivities, unobserved
George J. Bojas is professor of economics at the University of California, San Diego, and a
research associate of the National Bureau of Economic Research.
The author is grateful to Richard Freeman for many helpful discussions of the ideas presented
in this paper, to Charles Brown for insightful suggestions and comments, and to Bemt Bratsberg
for excellent research assistance. He is also grateful to the National Science Foundation (grants
SES-8604973 and SES-8809281) and to the National Institute of Child Health and Human Devel-
opment (grant R01-HD22344) for financial support.
29
30 George J. Borjas
abilities or skills) from more recent waves. Borjas (1985) adapted well-known
techniques (see, e.g., Heckman and Robb 1983) to separately identify aging
and cohort effects using the 1970 and 1980 U.S. Censuses. This methodology,
which “tracks” synthetic cohorts of immigrants over time, showed that ( a)
immigrant assimilation was not as fast as the cross-sectional studies indicate,
( b) the more recent immigrant waves performed substantially worse in the
labor market than the early postwar waves, and ( c) there was little likelihood
that the most recent immigrant waves would ever earn substantially more than
natives of comparable age and education.
An important insight provided by the study of synthetic cohorts is that in-
voking the assumption of positive selection, though it may be correct for some
cohorts of immigrants, may be completely wrong for other cohorts of immi-
grants. This raises the important question of exactly which factors determine
whether immigrants are positively or negatively selected from the population
in the countries of origin. Borjas (1987) presents an initial attempt to address
this problem and derives a simple economic model of selection on the basis of
unobserved characteristics (which, after all, form the focus of much of the
literature on immigrant earnings). This model, which will be discussed in
detail below, shows that there is no general law stating that immigrants must
be positively selected. In fact, under a reasonable set of conditions, it is likely
that immigrants are negatively selected (i.e., persons who have below-average
earnings and productivities are the most likely to migrate to the United
States). My empirical analysis revealed that positive selection was more likely
to characterize immigrants from the advanced industrial countries and nega-
tive selection was more likely to characterize immigrants from the Third
World countries, who form the bulk of migration to the United States in the
post-1965 period.
This paper expands my earlier work in a number of significant ways. The
theoretical analysis below will argue that, although most of the literature has
focused on the role that selection in unobserved Characteristics plays in deter-
mining immigrant earnings, there is also selection in observed characteristics
such as education. The theoretical framework clearly shows that it is com-
pletely possible for the most educated persons to migrate to the United States
(i.e., positive selection in education) but for these persons to be the least pro-
ductive in the population of highly educated persons (i.e., negative selection
in unobserved characteristics). The analysis below presents a number of prop-
ositions that yield insights into the process that determines the selection of
immigrants in these separate dimensions of “quality.”
The empirical analysis in this paper expands my previous work in two
ways. First, it presents a detailed analysis of the U.S. earnings of immigrants
by focusing on the roles played by selection in both observed and unobserved
characteristics. It will be seen that a number of the theoretical predictions are
confirmed by the data. Second, it is clear that potential migrants can choose
from a number of potential host countries. The empirical analysis below will
31 Immigration and Self-Selection
present a systematic study of the selection biases generated by the sorting of
migrants among three potential countries of destination: Australia, Canada,
and the United States. The evidence indicates that both country-of-origin and
country-of-destination characteristics play an important role in determining
the performance of immigrants in any labor market.
1.1 Theory of Immigration
1.1.1 The Roy (1951) Model
Migration is assumed to flow from country 0, the country of origin or the
“home” country, to country 1, the country of destination or, for concreteness,
the United States. This simple framework ignores three potential complica-
tions. First, it is likely that persons born in the United States also consider the
possibility of migrating to other countries, and perhaps many of them do so.
Second, even persons choosing the United States as a country of destination
may find that things do not work out (or perhaps work out much better than
expected), and some return migration is thereby generated. Third, individuals
contemplating migration in a particular country of origin enter the “immigra-
tion market” in which a number of other host countries (such as Australia and
Canada) compete for the immigrant’s human and physical capital. Little is
known about the size and composition of the migrant flows from the United
States to other countries; hence, these possibilities are ignored in what fol-
lows. Much more, however, is known about the size and composition of the
flows from any given home country to each of three potential host countries
(Australia, Canada, and the United States), and the implications of the simpler
two-country model will be applied below to the more general framework
where potential migrants not only decide whether to migrate but also choose
a country of destination.
Residents of the home country face an earnings (w) distribution given by
(1)
In w, = X S , + E ~ ,
where X is a vector of socioeconomic characteristics with value 6, in country
0, and the disturbance E~ is independent of X and is normally distributed with
mean zero and variance @.
The earnings distribution facing individuals in the United States is given by
(2) In w, = (1 - M) XS, + MX6, + E,,
where M is a dummy variable indicating whether the individual is foreign born
or native. The vector 6, gives the value that the U.S. labor market attaches to
the socioeconomic characteristics X for natives. This valuation may differ be-
cause of discrimination or other unobserved factors from the value 6 , that the
labor market attaches to the characteristics brought in by potential migrants.
The disturbance E, is again independent of X (and M) and is normally distrib-
32 George J. Borjas
uted with mean zero and variance 0:. Finally, the random variables E~ and E,
have correlation coefficient p.
Equations (1) and (2) completely describe the earnings opportunities facing
a potential migrant (as well as U.S. natives). Three questions are raised by
this simple framework. First, what factors determine the size of the migration
flow generated by the income-maximization hypothesis? Second, what types
of selection in the unobserved characteristics E are created by the endogenous
migration decision? Third, what types of selection in the observed character-
istics X are created by the endogenous migration decision?
The migration decision is determined by the sign of the index function:
where C gives the level of mobility costs, and 7~ gives a “time-equivalent”
measure (T = C/ wo) of the costs of migrating to the United States.
The level of migration costs C is likely to vary among individuals for two
reasons. First, there are time costs associated with migration, and these time
costs are likely to be higher for persons with higher opportunity costs. Sec-
ond, there are transportation costs associated with migration. These direct
costs include not only the air fare (which is likely to be constant across indi-
viduals) but also moving expenses of family and household goods, and it is
reasonable to suppose that these expenses may also be a positive function of
w,. These assumptions give little hint as to how the time-equivalent measure
of mobility costs, T, varies across individuals. It is instructive to assume first
that T is constant across individuals since the main implications of the Roy
model are clearest in this special case. The analysis below will show that the
treatment of T as a random variable in the population does not substantially
alter the analysis and will, in some instances, reinforce the conclusions of the
simpler model.
Since migration to the United States occurs when I > 0, the emigration rate
from the country of origin for persons of given characteristics X is given by
(4)
P( X) = pr{v > - [ X( S , - So) - T] } = 1 - @(z),
where v = E, - E ~ , z = - [ X( S , - So) - T]/u,, and 0 is the standard
normal distribution function. If the characteristics X have a joint density func-
tion given byf(x), then the emigration rate from country 0 is given by
P = P(x)f(x)dx.
X C i l
Equations (4) and (5) summarize the (rather obvious) economic content of
the theory of migration proposed by Hicks (1932) and further developed in
Sjaastad (1962). In particular, the emigration rate is a negative function of
mean income in the home country (po = XS,), a positive function of mean
33 Immigration and Self-Selection
income in the United States (p, = XS,), and a negative function of migration
costs. Much of the literature on the internal migration of persons in the United
States is devoted to testing these theoretical predictions (see the survey in
Greenwood 1975).
The immigration literature, on the other hand, has historically focused on
explaining not the size of migration flows but their composition or labor mar-
ket quality. As far back as 1919, for example, Douglas was asking whether
the skill composition of immigrant cohorts was constant across successive
immigrant waves. The theory of migration contained in equations (l)-(5) has
important implications about the selection biases that characterize the pool of
migrants in terms of both unobserved and observed characteristics. Consider
initially the selection mechanism in the unobserved characteristics E. In
particular, consider the conditional expectations E(1n w, I X, I > 0) and
E(1n w, I X, I > 0). Note that these means condition on two dimensions: the
observed characteristics X and the decision to migrate. Under the normality
assumptions, these conditional means are given by
(7)
where A = +( z) / P( X) , and + is the density of the standard normal. The vari-
able A is inversely related to the emigration rate and will be positive as long
as some persons find it profitable to remain in the country of origin (i.e.,
P [ X ] < 1).
Let Q, = E( E, I X, I > 0), Q, = E( E, I X , I > 0), and k = U,/U,. The
variables Q, and Q, measure the “quality” (in terms of unobserved character-
istics) of the migrant pool. The Roy model identifies three cases of substantive
interest.
Positive Selection, Q, > 0 and Q, > 0.
This type of selection exists when migrants have above-average earnings in
the country of origin (for given characteristicsx) and also have U.S. earnings
that exceed the earnings of comparable U.S. natives (ignoring the possibility
that immigrant earnings may be reduced because of their ethnic or racial back-
ground). Inspection of equations (6) and (7) shows that the necessary and
sufficient conditions for this type of selection to occur are
If p is sufficiently high, and if income is more dispersed in the United States
than in the country of origin, immigrants arriving in the United States will be
selected from the upper tail of the home country’s income distribution and
34 George J. Borjas
will outperform comparable natives on arrival in the United States. Intuitively,
this occurs because the home country, in a sense, is “taxing” high-ability
workers and “insuring” low-ability workers against poor labor market out-
comes. Since high-income workers benefit relatively more than low-income
workers from migration to the United States (regardless of how much higher
mean incomes in the United States may be relative to the country of origin), a
brain drain is generated, and the United States, with its greater opportunities,
becomes a magnet for persons who are likely to do well in the labor market.
Negative Selection, Q, < 0 and Q, < 0
This type of selection is defined to exist when the United States draws per-
sons who have below-average incomes in the country of origin and who, hold-
ing characteristics constant, do poorly in the U.S. labor market. The neces-
sary and sufficient conditions for negative selection to occur are
(9) p > k , k < 1.
Negative selection also requires that p be “sufficiently” positive but that the
income distribution in the country of origin be more unequal than that in the
United States. Intuitively, negative selection is generated when the United
States “taxes” high-income workers relatively more than the country of origin
and provides better insurance for low-income workers against poor labor mar-
ket outcomes. This opportunity set leads to large incentives for low-ability
persons to migrate, since they can improve their situation in the United States,
and to decreased incentives for high-ability persons to migrate, since income
opportunities in the home country are more profitable.
Refugee Sorting, Q, < 0 and Q, > 0
This kind of selection occurs when the United States draws below-average
immigrants (in terms of the country of origin) but migrants have above-
average earnings in the U.S. labor market. The necessary and sufficient con-
dition is
p < min (k, k) .
In other words, if p is negative or “small,” the composition of the migrant
pool is likely to resemble a refugee population. For instance, it is likely that p
is negative for countries that have recently experienced a Communist take-
over. After all, the change from a market economy to a Communist system is
often accompanied by structural changes in the income distribution and by
confiscation of entrepreneurial assets and redistribution to other persons. The
Roy model suggests that immigrants from such systems will be in the lower
tail of the “revolutionary” income distribution but will outperform the average
U. S. native worker.
The basic Roy model thus provides a useful categorization of the factors
35 Immigration and Self-Selection
that determine the quality or composition (in terms of unobserved character-
istics) of the migrant pool. Even at this level, several important implications
are generated that give some insight into a number of empirical findings in the
literature. For example, many studies have documented the fact that refugee
populations perform quite well in the U.S. labor market when compared to
native workers of similar socioeconomic characteristics. These empirical re-
sults are explained by the income-maximization hypothesis and by the fact
that these refugee populations, prior to the political changes that led to a wors-
ening of their economic status, were relatively well off in the country of ori-
gin. It is, therefore, unnecessary to resort to the arbitrary distinctions between
“economic” and “noneconomic” migrants to explain the refugee experience.
The Roy model also provides an interesting explanation for the empirical
finding that the quality of migrants to the United States has declined in the
postwar period (where quality is defined by the wage differential between mi-
grants and natives of the same measured skills). Prior to the 1965 amendments
to the Immigration and Nationality Act, immigration to the United States was
regulated by numerical quotas. The distribution of the fixed number of quotas
across countries was based on the ethnic population of the United States in
1920 and thus encouraged migration from (some) Western European countries
and strongly discouraged immigration from other continents, particularly
Asia. The favored countries have one important characteristic: their income
distributions are probably much less dispersed than those of countries in Latin
America or Asia. The 1965 amendments abolished the discriminatory restric-
tions against immigration from non-European countries, established a twenty
thousand numerical limit for legal migration from any single country (subject
to both hemispheric and worldwide numerical limitations), and led to a sub-
stantial increase in the number of migrants from Asia and Latin America. The
new flow of migrants thus originates in countries that are much more likely to
have greater income inequality than the United state^.^ It would not be sur-
prising, therefore, if the quality of immigrants declined as a result of the 1965
amendments.
The theoretical analysis yields two equations that can guide empirical anal-
ysis. These equations are given by
(1 1) Q, = gb0, CL, , n, u,,, u,, P),
Equation (1 1) gives a “reduced-form’’ equation, where immigrant quality
in the United States (i.e., the wage differential between migrants and natives
of equal measured skills) is a function of all the primitive parameters of the
model (i.e., the parameters of the two income distributions and migration
costs). My earlier paper (Borjas 1987) provides a detailed analysis of the theo-
retical restrictions implied by the income-maximization hypothesis on the di-
rection of the effects of the various variables in the model. These effects are
36 George J. Borjas
usually ambiguous and can be categorized in terms of “composition effects”
and “scale effects.” In particular, a change in any variable OL will create incen-
tives for a different type of person to migrate (the composition effect) and for
a different number of persons to migrate (the scale effect).
Equation (12) is a “structural” equation and states that, if knowledge of A is
available, a subset of the parameters of the model enters multiplicatively
through the h function (see eq. [7]). By holding A constant, the structural
equation essentially nets out the scale effect and leads to more unambiguous
predictions of the effect of the exogenous variables on the quality of immi-
grants. It is important to note that the h function in (12) does not depend on
mean income levels in the countries of origin and the country of destination
or on the level of migration costs since these factors play a role only through
the selectivity variable A.
Three comparative statics results are implied by analysis of the A-constant
structural quality equation.
1. An increase in the variance of the income distribution in the home coun-
try leads to a decrease in the quality of migrants in the United States.
2. An increase in the variance of the income distribution in the United
States leads to an increase in the quality of migrants in the United States.’
3. An increase in the correlation coefficient between earnings in the home
country and earnings in the United States increases immigrant quality if there
is positive selection and decreases immigrant quality if there is negative selec-
tion. The ambiguity arises because, the larger the correlation coefficient, the
better the “match” between the two countries. The improvement in the match
increases the quality of the immigrant flow if there is positive selection and
decreases it if there is negative selection.
1.1.2 Random Mobility Costs
These insights have been derived from the simplest version of the Roy
model, which treats mobility costs (defined as a fraction of potential income
in the country of origin) as a constant in the population. This assumption may
be restrictive, and it is important to ascertain how its relaxation affects the
results of the model. Suppose that mobility costs are normally distributed in
the population and can be written as
(13) Tr= P, + E,,
where p, is the mean level of mobility costs in the population, and E, is a
normally distributed random variable with mean zero and variance u;. The
random variable E, may be correlated with E, and E, and the correlation coef-
ficients are given by pno and p,, , respectively. The conditional expectations of
migrant incomes in the home and destination countries are now given by
u u
(14)
E(ln w, I X, I > 0) = X 6 , +
[(P - : ) - P d 3,
Ud
37 Immigration and Self-Selection
(15)
E(ln w, I X, I > 0) = X 6 , +
where v’ = E , - E, - E,.
Equations (14) and (15) show that the addition of migration costs does not
affect any of the substantive results of the simplest version of the Roy model
if migration costs are uncorrelated with earnings opportunities. However, if
migration costs are correlated with earnings opportunities, the type of selec-
tion that is generated may change in either direction. Suppose, for example,
that migration costs are positively correlated with earnings opportunities. For
instance, high-ability persons may take longer to find appropriate jobs. This
positive correlation makes both Q, and Q, more negative and hence increases
the likelihood of negative selection. Conversely, if migration costs (measured
in time units) and earnings opportunities are negatively correlated, the likeli-
hood of positive selection is increased.
Two additional points about this more general model are worth stressing.
First, the importance of variable migration costs in the analysis will diminish
greatly if the variance in migration costs is relatively small compared to the
variance in the income distributions. Second, regardless of how important
migration costs are, the key result that negative selection is more likely from
countries with high levels of income inequality and positive selection is more
likely from countries with more equal income distributions is unaffected.
1.1.3 Selection in Observed Characteristics
Equation (4), the probit equation determining the migration rate, contains
an additional insight: the migration rate is a function of X through the param-
eter (6, - 8,). Hence, the migration of persons with larger levels ofX is more
likely if X has a higher return in the United States than in the country of
origin, and the migration of persons with lower levels of X is more likely if
the country of origin values the characteristic X more than the United States.
A complementary analysis to the Roy model can be derived if it is assumed
that the vector X consists of only one variable, say education (s), that this
variable is uncorrelated with the disturbances in the earnings functions, and
that this variable, too, is normally distributed in the population. The assump-
tion of only one variable in the vector X is irrelevant since the results can be
easily generalized to any number of variables. The assumption of normality,
though unrealistic for some socioeconomic characteristics, does simplify the
mathematics substantially and allows a useful extension of the Roy approach
to the study of selection in observed skills and the analysis of the actual wage
differential between immigrants and natives (as opposed to the standardized
wage differential).
Suppose the earnings functions in the two countries are given by
(16)
In w, = po + 6,s + E,.
38 George J. Borjas
(17)
In w, = p, + 6,s + E, ,
and that the education distribution in the population of the country of origin
can be written as
(18) s = + E, r
where E, is normally distributed with mean zero and variance US.
ulation in the country of origin is given by
(19)
where t = ( E, - E ~ ) + (6, - SO) &, , and z* = -[(p, - po) +
Two interesting questions can be addressed within this framework. First,
consider the conditional expectation of schooling of persons who do migrate.
It is easy to show that
Assuming that mobility costs are constant, the emigration rate for the pop-
p = pr{(El - Eo) + (4 - $)E, > - [(PI - Po)
+ (6, - 6,)p$ - TI} = 1 - @(z*),
(6, - 6JPs - TI/U,.
Hence, the mean schooling of migrants will be less than or greater than the
mean schooling of the population depending on which of the two countries
values schooling more. Positive selection in schooling will be observed
when (6, - 6,) > 0 so that the U.S. labor market attaches a higher value to
schooling, while negative selection in schooling will be observed when
(6, - 6,) < 0 so that highly educated individuals have little incentive to
leave the country of origin.
It is important to stress that these selection conditions seem to have little to
do with the conditions determining selection in unobserved characteristics.
Any permutation of selection mechanisms in unobserved and observed char-
acteristics is theoretically possible. Hence, negative selection in unobserved
characteristics (or ability) may be jointly occurring with positive selection in
education, or vice versa. Simply because the United States attracts highly ed-
ucated persons from some countries does not imply that these highly educated
persons are the most productive highly educated persons in that particular
country of origin.
At a more fundamental level, however, the determinants of the two types of
selection are not all that different. The sorting in observed characteristics is
guided by international differences in the prices So and 6, . In the case of un-
measured skills, the sorting is guided by the variances ui and a:. In a sense,
these variances measure the “prices” of unmeasured skills in the respective
countries since these abilities are more highly rewarded in countries with
higher levels of income inequality. The sorting in all the dimensions of skills,
39 Immigration and Self-Selection
therefore, is guided by the same basic process: skills flow to whichever coun-
try offers the highest price for them.
The actual mean earnings of the migrant pool in each of the two countries
are given by
(21) E m w, I I > 0) = I l . 0 + 6oPs
*OU1 (p -
)]A,
(6, - 6,)6, + -
u,
Mean earnings of migrants depend on the mean education of migrants, as
given by (20), and on the mean level of their unobserved characteristics. Since
the two kinds of selections are independent, nothing can be said about how
the average migrant performs in the host country unless the kinds of selections
that occurred in each of the two dimensions of quality are known. Neverthe-
less, it is of interest to document the net effect of the selection in all the various
dimensions of skills on immigrant earnings, and the empirical analysis below
presents a detailed study of the unstandardized earnings differential between
immigrants and natives in the host country.
Equations (21) and (22) show that generalizations about the quality of im-
migrants based solely on observed education levels (or other measures of X)
are extremely misleading. In addition, it is well known that observed charac-
teristics such as education, age, marital status, health, etc. explain a relatively
small fraction of earnings variation across individuals. It is not uncommon,
for example, to find that the observed characteristics explain much less than a
third of the variance in wage rates or weekly earnings. The selection in unob-
served characteristics, therefore, is likely to be much more important empiri-
cally than the selection in observed characteristics.
A number of comparative statics results can be generated by analysis of
equation (20). Perhaps the most interesting of these results is
That is, a one-year increase in the mean education level of the country of
origin will increase the mean education level of persons who actually migrate
to the United States, but this increase will be by less than one year. The intui-
tion for this result follows from the fact that an increase in kS will change the
size of the immigrant flow. Suppose, for concreteness, that (6, - 6,) > 0 so
that there is positive selection in education. The increase in p, makes it worth-
while for more persons to migrate and thus dilutes the mean education level
of the population of migrants. Hence, the increase in the conditional expecta-
tion is less than the increase in the population mean. An important implication
40 George J. Borjas
of this theoretical prediction is that the variance in education levels across
immigrants (from different countries) in the United States will be smaller than
the variance in education levels of the actual populations across countries in
the world. In other words, the population of migrants in the United States is
more homogeneous (in terms of education) than the populations of the differ-
ent sending countries.
In general, equation (20) implies the existence of observable quality equa-
tions analogous to (1 l ) and (12):
(24) QT = g* b0, CL.,, m, uo, ul, P, F,, us, (6, - 6,,)1,
(25) QT = h*[uo, ul, P, IJ-~, us, (8, - 6,)lL
where QT gives the mean level of the observed characteristics of immigrants
in the United States. The estimation of (24) and (25), of course, is likely to be
extremely difficult in practice since they introduce a number of primitive pa-
rameters (e.g., 6, - 6,) that are unobservable and likely to remain so.
1.2 Empirical Framework
Recent empirical research on the earnings of immigrants stresses the impor-
tance of disentangling the cohort and aging effects that are confounded by a
single cross section of data. In the analysis presented in this paper, two Cen-
suses in the country of destination will be pooled (e.g., the 1970 and 1980
U.S. Censuses), and the following regression model will be estimated:
(26) In wo = Xjai + al yj + azy,’ + 2 p,C, + yimj + E ~ ,
(27)
In wnl = X, S, + y,ml + E,~,
where wii is the wage rate of immigrantj, wnr is the wage rate of native person
1, X is a vector of socioeconomic characteristics (e.g., education, age, etc.), y
is a variable measuring the number of years that the immigrant has resided in
the country of destination, C is a vector of dummy variables indicating the
year in which migration occurred, and IT is a dummy variable set to unity if
the observation is drawn from the 1980 Census and zero otherwise.6 The vec-
tor of parameters (a,, az), along with the age coefficients in the vector X,
provides a measure of the assimilation effect (i.e., the rate at which the age-
earnings profile of migrants is converging to the age-earnings profile of na-
tives), while the vector of parameters p estimates the cohort effects. The pe-
riod effects are given by yi for immigrants and by y, for natives.
The model in equations (26) and (27) is underidentified. In particular, some
of the right-hand-side variables in the immigrant earnings function are per-
fectly collinear. Suppose, for example, that the immigrant arrived in calendar
year 8 so that C, = 1. Then
t
41 Immigration and Self-Selection
(28)
y = (T - k - 8) + nk,
where T is the calendar year in which the latest cross section is observed, and
k is the number of years separating the two cross sections. The variable cap-
turing the period effect, therefore, is a linear combination of the cohort vari-
able and of the years-since-migration variable. Obviously, two cross sections
cannot be used to identify three separate effects: period, cohort, and aging
effects.
In order to estimate the structural parameters describing the extent of im-
migrant assimilation and cohort quality change, a restriction must be imposed
on the size of the period effect in the migrant population. A reasonable,
though unverifiable, assumption is that the period effect experienced by im-
migrants (y,) is identical to the period effect experience by natives (y,). In
other words, changes in the wage rate due to shifts in aggregate economic
conditions affect the immigrant and native wage levels by the same relative
magnitude. It is easy to show that this restriction is sufficient to identify all the
structural parameters in equations (26) and (27) exactly. This theoretical re-
striction leaves some amplitude for its empirical implementation since the
choice of the native base is essentially arbitrary. The choice of a native base
for the various immigrant groups under study will be discussed in detail below.
There are two dimensions of migrant quality that can be calculated from the
estimated regressions in (26) and (27): (a) the entry wage of immigrants when
they arrive into the United States and (b) the rate at which this wage changes
over time. To simplify the empirical analysis, the two measures will be com-
bined into a single measure of immigrant quality. In particular, let G,(8) be the
entry wage of an immigrant cohort that arrives in the United States at age
twenty in calendar year 8, and let Gn be the entry wage of a comparable (in
terms of all observable economic variables) native person who enters the labor
market at age twenty. Similarly, let gi be the rate at which the earnings of
immigrants grow over their lifetime, and let g, be the growth rate for natives.
Finally, let r be the rate of discount (assumed to be the same for migrants and
natives). If persons are infinitely lived, the present values associated with the
earnings streams of migrants and natives are given by
vj(e) =
Gi(8)e-@-gJ' df = wi(8) / ( r - gi ) ,
(29) I
0
= @ ( r - g ) .
(30)
The percentage difference in present values between immigrants of cohort
8 and natives is defined by
42 George J. Borjas
and a first-order approximation (using the assumption that earnings growth
rates are small relative to the discount rate) yields
gi -g,
In [Vi (e) / V,] = [In ai(8) - In an] + - .
r
Hence, the percentage difference in the present value of the earnings streams
faced by immigrants and natives is an additive function of the wage differen-
tial at the time of entry and of the difference in earnings growth rates over the
life cycle.7
The present value differential in (32) can be easily evaluated from the esti-
mates of equation (26) and (27) if two assumptions are made. First, the rate
of discount is assumed to be 5 percent. Clearly, the assumption of any higher
discount rate would lead to a worsening of relative immigrant earnings since
the latter part of the working life cycle (where immigrants tend to do better)
would be more heavily discounted. Second, the growth rates g, and g, must
be evaluated from the age and years-since-migration coefficients in the earn-
ings functions in (26) and (27). The quadratic specification for age and years
since migration in the earnings functions implies that the growth rate is not con-
stant over time. The empirical analysis below will define the growth rate g,
2, = ( Y, [ X, 50, 30, 81- Y , [X, 20, 0, 8])/30,
2, = ( Y, [ X, 501 - Y,[ X , 20])/30,
and g, by
(33)
(34)
where Y, [ X, A, y, 81 is the predicted (In) earnings for an immigrant with char-
acteristics X , at age A, with y years of residence in the United States, and who
migrated in cohort 8. Similarly, YJX, A] gives the predicted earnings for a
native with characteristics X at age A. In other words, the average growth rate
experienced by immigrants and natives between ages twenty and fifty (evalu-
ated at the mean characteristics of the migrant population, X) is used for esti-
mation of the growth rate in the present value expressions.
This approach has the useful property that the growth rates (for both immi-
grants and natives) are a linear function of regression coefficients, and, since
the entry wages are given by Y, [ X, 20,0, 01 for immigrants and YJX, 201 for
natives, the present value expressions in (33) and (34) are also linear functions
of regressions coefficients; hence, a standard error can be easily evaluated.
This approach makes a departure from the tradition in the empirical litera-
ture that analyzes immigrant earnings. The entire literature essentially focuses
on the estimation of entry wage levels and on the calculation of “overtaking”
points (if they exist). This type of analysis is not useful if overtaking points
occur rather late in the life cycle (or if they do not occur at all), as some recent
evidence suggests. The empirical use of the present value of earnings is much
more consistent with the theoretical content of the theory of migration and
43 Immigration and Self-Selection
deemphasizes the somewhat misleading concept of overtaking points. The
analysis of the success of migrant groups in the United States, to borrow from
the human capital theory that guided much early research on immigrant earn-
ings, should be based not on the calculation of wage differentials at given ages
but on the life-cycle wealth accumulated by migrants and natives. Hence, the
present value approach used in the empirical sections of this paper is much
more in the tradition of the human capital literature and of the Roy model of
immigration developed in the previous section.
1.3 Earnings of Immigrants in the United States
1.3.1 Data and Descriptive Statistics
This section analyzes the relative earnings of immigrants in the U.S. labor
market. The data are drawn from the 1970 2/100 U.S. Census (obtained by
pooling the 5% SMSA and County Group Sample and the 5% State Sample)
and the 1980 5/100 A sample. The complete samples are used in the creation
of the immigrant extracts, but random samples are drawn for the native “base-
line’’ populations.8 The analysis is restricted to men aged 25-64 who satisfied
five sample selection rules: (1) the individual was employed in the calendar
year prior to the Census; (2) the individual was not self-employed or working
without pay; (3) the individual was not in the armed forces (as of the survey
week); (4) the individual did not reside in group quarters; and (5) the individ-
ual reported annual earnings exceeding $1,000. Throughout this section, the
dependent variable is the logarithm of the individual’s wage rate in the calen-
dar year prior to the Census. The individual’s wage rate is defined as the ratio
of annual earnings to annual hours worked. In the 1970 Census, annual hours
worked is given by the product of weeks worked in 1969 and hours worked in
the Census week, while, in the 1980 Census, annual hours is the product
of weeks worked in 1979 and usual hours worked per week in that calendar
year.
Forty-one countries were chosen for analysis. These countries were se-
lected on the basis that both the 1970 and the 1980 Censuses contained a
substantial number of migrants from that country. In particular, it is necessary
to have at least eighty observations of persons born in a particular foreign
country in the pooled UlOO 1970 Census to enter the sample of forty-one
countries. The countries thus chosen account for over 90 percent of all immi-
gration to the United States between 1951 and 1980. It must be noted, how-
ever, that this restriction omits some countries that during the late 1970s be-
came important source countries (e.g., Vietnam). Since two Censuses are
required for the complete identification of the parameters of the model pre-
sented in section 1.2, however, a systematic analysis of the relative earnings
of these migrants will have to await the 1990 Census.
Table 1.1 begins the empirical analysis by presenting the unstandardized
44 George J. Borjas
differential between the log wage rate of the various migrant groups and ‘ha-
tives.” In these statistics, the native population is defined as the group of U.S.-
born white, non-Hispanic, non-Asian men aged 25-64. Perhaps the most
striking finding in the table is the fact that migrants from European countries
tend to have wage rates that often exceed the wages of white natives, while
migrants from Asian or Latin American countries tend to have wage rates that
are substantially below those of white natives.
Table 1.1 also presents the relative earnings of the 1965-69 cohort of mi-
grants as of 1970, the relative earnings of the same cohort in 1980, and the
relative earnings of the 1975-79 cohort as of 1980. These statistics yield im-
portant insights into the process of assimilation (the rate at which the earnings
of migrants and natives are converging) and into the extent of productivity
differences across successive cohorts. The “tracking” of the 1965-69 cohort
across Censuses shows that the relative earnings of this cohort of migrants
improved over time for most national groups. At the same time, the compari-
son of successive immigrant cohorts (i.e., the comparison of the 1965-69
cohort as of 1969 and the 1975-79 cohort as of 1979) shows that for some
countries the relative earnings of migrants increased while for other countries
the relative earnings of migrants decreased substantially. For example, the
most recent migrant from France in 1970 was earning about 8 percent less
than natives at the time of entry, while the most recent migrant from France in
1980 was earning about 22 percent more than natives at the time of entry.
Conversely, the most recent migrant from India in 1970 earned about 4 percent
more than white natives at the time of entry, but the most recent migrant from
India in 1980 was earning 21 percent less than white natives at the time of
entry.
Table 1.2 continues the descriptive analysis by presenting the mean (com-
pleted) education level of four different cohorts of immigrants that arrived in
the 1960-80 period. Since the education data available in the Census does not
differentiate between education obtained prior to immigration and education
obtained in the United States after immigration, the mean education levels for
the 1970-74 and 1975-79 cohorts are obtained from the 1980 Census, and
the mean education levels for the 1960-64 and 1965-69 cohorts are obtained
from the 1970 Census. This use of the available data is designed to minimize
the contamination of the education variable by postmigration schooling.
The statistics in table 1.2 are consistent with the well-known secular in-
crease in education levels over time for practically all migrant cohorts. It is
worth noting, however, that for some countries the increase in education has
been quite small (e.g., Portugal) while for others (e.g., Norway) it has been
amazingly large. As the theoretical analysis in section 1.1 shows, these trun-
cated education means can be understood only in terms of the population
means of the education distribution in the countries of origin. To provide some
insights into the extent of self-selection on the basis of education, table 1.2
also presents mean education levels calculated for the population in the coun-
tries of origin. The mean education level for the 1960s is calculated using
45 Immigration and Self-Selection
Table 1.1 Unstandardized U.S. Earnings of Immigrants Relative to White Natives
1970 1980
All 1965-69 Sample All 1965-69 1975-79 Sample
Country Immigrants Cohort Size Immigrants Cohort Cohort Size
Europe:
Austria
Czechoslovakia
Denmark
France
Germany
Greece
HwwY
Ireland
Italy
Netherlands
Norway
Poland
Portugal
Romania
Soviet Union
Spain
Sweden
Switzerland
,1969
,1229
,1208
,1109
.I600
- ,1722
.1304
- .0369
- ,0150
.OW3
,1653
.0392
- ,1913
.1153
,0813
,1485
,2424
- ,1572
United Kingdom .I669
Yugoslavia ,0353
China - ,1543
Egypt - ,0073
India ,1667
Iran - ,0116
Israel ,0707
Japan ,0535
Korea - ,0781
Philippines - ,1920
Asia and Africa:
Americas:
Argentina ,0319
Brazil .0212
Canada ,1072
Colombia - .I452
Cuba - ,2822
Dominican
Republic - .3576
Ecuador - ,2343
Guatemala -.I940
Haiti - ,3041
Jamaica - .I645
Mexico - ,4094
Panama - ,0187
Trinidad
and Tobago - ,1561
,2182
,0466
,1803
- ,0766
,1095
- .3704
- ,0631
- ,0260
- ,1707
.1412
,2629
- ,0952
- ,2406
- .I915
- .I048
- ,3480
.2573
.0095
.1902
- .I382
- .3459
- .2127
,0413
- ,3556
-.I951
.0519
- .2183
- ,2389
- ,1644
- ,0993
,1084
- ,2337
- .4461
- ,5157
- ,451 1
- ,5372
- ,3061
- .2462
- .6021
- .1899
- .2909
380
398
141
317
2,399
634
650
754
3,068
430
243
1,629
349
259
907
210
22 1
177
2,231
646
880
136
363
121
141
228
142
816
218
101
3,430
254
1,960
210
174
82
130
263
3,122
101
86
,2108
,1483
.2387
,1071
,1577
- .I874
,1059
,0688
- ,0124
,1717
26.96
,0165
- .2104
.0551
- .0533
- .0417
,2392
,3307
,2111
,0546
- ,2212
,0737
,1221
- .0545
- ,0274
,1362
- ,0881
- ,0707
- ,0096
,0485
,1258
-.2313
- ,1828
- ,4768
- .2473
- .3425
- ,3726
- ,2132
- .3975
- ,0761
- .1488
,3598
,1141
,3570
,1158
,2350
- ,2556
,1027
,0737
- ,0790
,2179
.4183
.0207
- .1949
.0928
- .0578
- ,0184
.4570
,2121
,3188
- ,0191
- .1324
.3222
,4050
,1375
- ,0392
,1492
,2409
,0694
.0086
,1407
,1440
- .2027
- .2698
- ,4319
- ,2858
- ,2182
- ,2296
-.I245
- .3431
- ,1263
- ,0685
-.1258
.0273
,4241
,2237
,2646
- ,3392
- ,1805
- ,1421
- ,1616
,2824
,2444
- .3698
- ,3240
- ,2913
- ,2856
- ,2143
,1617
.4735
.1924
- ,1706
- S372
- ,2892
- .2085
- ,2237
- .2483
.2020
- .3007
- ,3143
- .1428
,0481
,1739
- ,4464
- .5392
- ,6785
- ,5229
- s977
- ,6536
- .3604
- .6402
- ,3663
- ,4150
746
872
29 1
952
6,499
2,328
1,356
1,580
7,236
1,161
408
3,278
2,213
614
2,104
730
335
397
5,475
1,967
3,875
6%
3,629
1,027
789
1,634
2,013
4,955
834
345
7,083
1,760
6,837
1,605
1,097
723
1,133
2,061
24,955
584
782
46 George J. Borjas
Table 1.2 Completed Years of Schooling in Immigrant Cohorts
Mean Education
Year of Arrival in Population
Country 1975-79 1970-74 1965-69 1960-64 1970s 1960s
Europe:
Austria
Czechoslovakia
Denmark
France
Germany
Greece
Ireland
Italy
Netherlands
Norway
Poland
Portugal
Romania
Soviet Union
Spain
Sweden
Switzerland
United Kingdom
Yugoslavia
Asia and Africa:
China
India
Iran
Israel
Japan
Korea
Philippines
Americas:
Argentina
Brazil
Canada
Colombia
Cuba
Dominican Republic
Ecuador
Guatemala
Haiti
Jamaica
Mexico
Panama
Trinidad and Tobago
Hungary
Egypt
14.8
15.4
15.5
15.6
15.2
11.1
13.6
13.8
10.6
15.9
15.2
12.7
6.6
13.7
14.3
13.2
15.4
15.4
15.1
11.0
11.3
15.9
16.1
15.2
14.2
15.7
14.0
14.2
13.6
15.4
14.6
11.9
11.3
8.9
10.9
9.0
10.2
11.3
6.5
13.1
11.7
13.9
14.5
13.6
14.8
14.2
9.9
13.5
13.1
8.5
15.1
15.6
11.9
6.7
14.5
13.5
11.3
15.8
15.4
14.7
10.6
12.8
16.2
17.6
16.3
13.8
14.7
14.9
14.9
12.1
13.1
13.7
11.3
9.9
9.1
11.0
9.7
12. I
10.9
6.8
12.7
12.0
13.4
14.1
16.1
14.5
13.3
8.8
12.3
12.9
6.8
14.1
14.0
10.7
5.2
11.6
10.5
10.3
15.5
14.5
13.7
10.7
12.8
15.5
16.7
15.3
13.5
15.4
15.8
14.8
12.0
12.6
12.9
10.6
9.5
8.4
10.4
9.9
12.0
10.7
6.1
12.4
11.0
12.8
12.5
11.6
12.8
12.0
10.9
12.6
11.3
7.5
12.3
11.7
9.5
5.8
11.9
11.3
9.9
14.4
13.6
13.1
9.4
13.2
15.1
17.0
15.5
14.0
15.0
16.5
13.9
12.6
12.8
11.4
11.5
11.9
7.9
11.3
12.0
11.2
10.6
6.0
11. 1
14.4
8.7
10.2
11.2
11.1
10.7
9.2
10.6
9.1
9.1
10.4
9.9
11.2
8.2
9.5
11.4
8.0
10.3
8.7
10.8
9.7
8.4
5.7
4.9
3.6
9.8
11.2
8.0
8.2
8.7
8.6
10.3
5.0
8.3
6.2
6.2
2.9
3.2
9.5
6.1
10.1
7.9
6.7
9.1
8.5
7.0
10.1
6.2
7.2
8.1
5.6
8.8
7.2
7.0
3.5
5.3
8.1
4.4
8.7
6.7
9.9
3.5
4.3
4.0
2.2
1.3
7.0
9.2
5.0
5. 1
6.3
2.8
8.5
2.2
4.1
3.6
3.4
1.5
1.7
4.5
2.9
5.9
7. 1
47 Immigration and Self-Selection
enrollment data in the various countries of origin during the 1950s, while the
mean education level for the 1970s is calculated using enrollment data in the
various countries of origin during the 1960s. The “lagged” construction of
the variable giving mean education levels in the country of origin is designed
to account for the fact that, in the samples used here, the average person mi-
grated at about age 20. The relevant education distribution, therefore, is given
by that of persons enrolled in school a few years earliec9
The means in table 1.2 present a remarkable picture. Even after allowing
for the substantial errors involved in calculating the population means for each
country of origin, the truncated means are almost always much greater than
the population means. For example, the mean of education in Haiti is about
three years, but the most recent Haitian immigrants report ten years of educa-
tion in the 1980 Census. Surprisingly, the two statistics are most similar for
Mexico, where both immigrants and the Mexican population have 6-7 years
of education. Overall, table 1.2 suggests that immigrants are positively se-
lected on the basis of education. The model presented earlier implies that this
result is consistent with the hypothesis that the “rate of return” to education is
greater in the United States than in most countries of origin. However, it is
also consistent with the hypothesis that migration costs are lower for persons
with higher education levels. This conjecture has received intensive study in
the internal migration literature (Schwartz 1968).
1.3.2 Basic Regression Results
The regression model in equations (26) and (27) was estimated on each of
the 41 countries under analysis using the pooled 1970 and 1980 Census data.
As noted earlier, the choice of the native baseline is an important step in the
estimation procedure. In this section, the reference group is chosen according
to the racial/ethnic background of the population of each country of origin.
The estimation uses the white, non-Hispanic, non-Asian sample of native men
as the reference group for migrants from Europe, Canada, and the Middle
East. The group of Asian natives is the reference group for migrants from all
other Asian countries. The group of Mexican natives is the reference group
for Mexican migrants, and the group of “other Hispanic” men is the reference
group for persons from all other Spanish-speaking countries in the American
continent. Finally, the group of black natives is the reference group for mi-
grants from countries with predominantly black populations (i.e., Haiti, Ja-
maica, and Trinidad and Tobago).
The definition of the reference group in terms of the racial/ethnic back-
ground of the immigrant population is a simple way of specifying different
period effects for the various immigrant groups. Presumably, the effect of
changes in aggregate economic conditions on immigrant earnings is likely to
be better approximated by the period effects experienced by populations that
closely resemble the immigrant group. It is important to note, however, that,
although the baseline populations differ across the 41 countries, the calcula-
48 George J. Borjas
tion of the present value differentials defined in equation (32) will always be
relative to white, non-Hispanic, non-Asian natives (as in table 1. 1) . In other
words, the use of alternative reference groups is simply used to “net out” the
period effect in the 1980 Census, and, after controlling for period effects, all
comparisons between migrants and natives are conducted with respect to the
“white” population.
Initially, the regression model is estimated using a detailed list of demo-
graphic controls. These include education, marital status, health status, and
metropolitan residence (as well as age and age squared). The calculated pres-
ent value differentials estimated from the 41 runs of the model are presented
in table 1. 3 for each of the 6 cohorts identifiable in the Census data. It is worth
stressing that these present value differentials measure the differences in earn-
ings among migrants and white natives of equal measured skills and hence are
empirical counterparts to the quality measure Q, defined in terms of unob-
served characteristics.
Table 1. 3 shows that there are substantial differences in the “abilities” of
migrant groups across the 41 countries of origin. Immigrants from European
countries (particularly Western European countries) tend to do quite well rel-
ative to white natives of comparable socioeconomic characteristics. Recent
immigrants from the United Kingdom, for example, can expected about 10
percent larger earnings over their lifetime than comparable white natives, re-
cent immigrants from France will earn about 8-19 percent more than compa-
rable white natives, and recent immigrants from Sweden will earn about 10-
20 percent more than white natives over their lifetime.
On the other hand, immigrants from most Asian and Latin American coun-
tries do not perform well in comparison to white natives of equal observable
skills. Recent immigrants from Taiwan, for example, will earn about 16-34
percent less over their lifetime than comparable white natives, immigrants
from Israel will earn about 20-30 percent less, immigrants from Argentina
about 20 percent less, and immigrants from Colombia about 24-38 percent
less. An immigrant’s birthplace plays an important role in determining the
type of selection that characterizes the migrant flow.
In addition, table 1. 3 shows that, even within a given country of origin,
there are sizable differences in the unobserved quality of immigrants across
the various cohorts. The quality of immigrants from some countries has been
increasing rapidly, while the quality of immigrants from other countries has
been declining rapidly. For instance, the most recent French immigrants have
a higher earnings potential than earlier cohorts (particularly those arriving
before 1970), while the most recent Polish migrants have much lower earnings
potential than migrants of earlier cohorts. Similarly, the most recent Canadian
immigrants earn about 8-15 percent more than most of the earlier cohorts,
while the most recent Mexican immigrants earn about 9-13 percent less than
the earlier Mexican cohorts.I0
49 Immigration and Self-Selection
lhble 1.3 Present Value Differentials between Immigrants and Natives
Year of Arrival
Country 1975-79 1970-74 1965-69 1960-64 1950-59 < 1950
Europe:
Austria
Czechoslovakia
Denmark
France
Germany
Greece
Hungary
Ireland
Italy
Netherlands
Norway
Poland
Portugal
Romania
Soviet Union
Spain
Sweden
Switzerland
United Kingdom
Yugoslavia
- ,0841
( - .74)
- ,0141
(-.14)
,4432
(2.76)
. I879
(2.29)
,0733
(1.69)
-.lo60
(-2.00)
,1542
(- 1.94)
,1267
(1.58)
.0498
(1.30)
.2815
(3.66)
,1880
(1.45)
- .1926
(-4.11)
.0293
( . 5u
- ,2030
(-2.12)
- ,2641
(-4.42)
,1047
(1.17)
,1141
(1.01)
,2395
(2.15)
,1052
(3.11)
,0602
(.93)
,1344
(1.25)
- ,0546
( - .73)
,0623
(.35)
,0829
(1.15)
.0638
(1.50)
- .I818
( - 5.08)
-.I132
(- 1.81)
,0817
(1.39)
,0424
(1.75)
- .O917
(-1.11)
.2468
(1.56)
,0727
(1.95)
,0348
(.82)
.0911
(1.21)
- .03W
( - .55)
,1287
(2.01)
,1621
(1.15)
.lo71
(.93)
.0910
(2.88)
.0746
(1.84)
,1945
(2.84)
- .0036
(-. l o)
,1522
(1.45)
- .0415
( - .74)
,0385
(1 .a)
-.I344
(-4.40)
- ,0128
( - .26)
.1758
(3.75)
.0693
(3.77)
.0936
(1.69)
.1757
(1.74)
.0784
(2.65)
,0785
(2.31)
- .0050
(-. l o)
- .0332
( - .57)
.0518
(.96)
,2205
(1.97)
.I407
(1.69)
,0948
(4.19)
.0625
(1.94)
,0707
(1.33)
.m
(1.01)
- ,0010
( - .01)
-.I179
(-2.57)
,0115
- .0402
(-1.10)
- ,0389
( - .86)
.0676
(2.14)
.0839
(5.04)
,0264
,2017
(2.55)
.0387
(1.66)
.0954
(2.44)
- ,0253
( - .39)
- ,0456
(-1.06)
- .0022
( - .01)
,0721
(.78)
,0967
(1.41)
,0449
(2.32)
,1389
(3.90)
~ 7 0 )
- .o004
( - .01)
,0182
- ,0434
( - .65)
- ,0626
(-1.56)
,0150
- .0381
(-1.39)
.0380
(1.45)
- .0252
~ 4 2 )
(.97)
( - .84)
.0695
(5.10)
- ,0422
(-1.40)
,1437
(2.48)
,0526
(2.44)
,0871
(2.18)
,0534
(1 .04)
,0203
(.68)
- ,1186
(-2.22)
.o001
(.01)
,0594
(1.05)
,0098
,1089
(4.15)
(.%I
- .0312
( -
.05%
(.W
. I 105
(.94)
.0539
(.81)
,1174
(4.26)
- ,1230
(-2.28)
,1441
(2.55)
- ,2171
(-4.82)
,0627
(2.48)
-.I736
( - 2.77)
- .0290
( - .35)
.0764
(2.31)
.1746
(2.11)
- ,0041
( - .01)
,0322
(.67)
- ,1001
( - .94)
,0153
(.14)
.0264
(.26)
- ,0432
( - 1.47)
.0237
(.39)
(continued)
50 George J. Borjas
Table 1.3 (continued)
Year of Anival
Country 1975-79 1970-74 1965-69 1960-64 1950-59 <I950
Asia and Africa:
China
Egypt
India
Iran
Israel
Japan
Korea
Philippines
Americas:
Argentina
Brazil
Canada
Colombia
Cuba
Dominican Republic
Ecuador
Guatemala
Haiti
Jamaica
Mexico
Panama
Trinidad and
Tobacco
- ,3662
(-7.58)
- ,1597
( - 1.70)
- .3365
(-5.28)
.075 1
(.77)
- ,3304
( - 3.50)
,0741
(.95)
- ,1840
( - 2.00)
-.I884
(-4.01)
- .2537
(-2.97)
.0679
(.54)
.1440
(4.07)
- ,3764
(-4.99)
- ,2711
(-5.89)
- ,1566
( - 5.73)
- ,2965
(-3.16)
- ,3163
( - 2.68)
- ,4721
(-4.81)
- ,2958
( - 4.48)
- ,1566
( - 5.73)
- ,2717
( - 2.20)
- ,2433
( - 2.15)
- ,3362
( - 10.26)
-.I186
( - 1.46)
- .1635
(-2.94)
- .0084
(-.lo)
- ,2346
( - 3.32)
- ,0145
( - .22)
-.I162
( - 1.51)
- ,0778
( - 2.53)
- .2723
(-3.69)
- ,0944
( - .91)
,0497
(1.51)
- .2372
( - 3.62)
- .0850
( - 2.92)
- ,0628
( - 3.52)
-.1742
( - 2.08)
- ,2695
(-3.17)
- ,2447
( - 2.85)
- ,1505
( - 3.24)
- ,0628
( - 3.52)
- ,0221
- .0774
( - .95)
( - .24)
- ,2274
( - 8.62)
- ,0588
( - .82)
- ,0497
( - .98)
.0215
(. 32)
- ,2766
( - 4.18)
- ,0496
( - .81)
- ,0966
( - 1.31)
- ,0689
( - 2.75)
- ,1908
( - 3.10)
.0623
(.75)
,1149
(5.27)
-.I562
( - 2.65)
- .I366
( - 6.08)
- .0399
(-2.75)
- .2348
(-3.07)
- ,2551
(-3.32)
- .1227
( - 1.56)
- ,1078
(-2.72)
- .0399
( - 2.75)
- ,1267
( - 1.80)
- .0438
( - . W
- ,1842
(-6.88)
- ,0980
,039 1
- ,0470
(-1.34)
(.78)
( - .71)
- ,1978
(-3.10)
- .0401
- ,0738
(-1.04)
- ,1075
( - 3.26)
( - .71)
- ,0822
(-1.47)
.0782
(.88)
,068 1
(4.24)
- ,0614
( - 1.11)
- ,0687
(-3.52)
- ,0338
(-2.39)
- ,0657
( - .89)
- ,2085
(-2.33)
- ,0189
( - .22)
- ,2182
(-4.51)
- ,0338
(-2.39)
- ,0972
(-1.47)
- ,0002
(- .01)
- .1228
(-4.20)
- ,0900
( - 1.33)
,0371
(.73)
- ,0207
( - .32)
,0060
(. 10)
- ,0799
( - 1.67)
- .2214
(-3.55)
-.I856
(-5.94)
- ,0497
( - .83)
- .0006
- .01)
,0359
(2.45)
.0920
(1.69)
- .1752
(-7.87)
- ,031 1
( - 1.94)
- ,0810
( - 1.20)
- ,0959
(-1.13)
- ,1056
(-1.29)
- ,0780
( - 1.67)
-.0311
( - 1.94)
-.I131
(-1.74)
.098 1
(1.04)
- ,0944
( - I .88)
,1511
(.84)
,1138
(.94)
- .0143
(-. l o)
.2476
(1.72)
- .1887
(-1.39)
- .0144
( - .01)
-.1108
( - 2.15)
,0221
(1.44)
- ,3063
(-1.59)
- ,0065
( - .26)
- ,2959
( - 1.77)
- ,0870
(-1.61)
- ,0904
( - 2.99)
,0581
(.28)
,3290
(1.85)
- ,4107
(-2.05)
- ,0451
( - .51)
- ,0904
( - 2.99)
,0544
(.42)
- ,1023
( - .57)
Note: The t-ratios are presented in parentheses
51 Immigration and Self-Selection
1.3.3
The Roy model suggests that the quality differentials documented in table
1.3 can be “explained” by economic and political characteristics of both the
various countries of origin and the United States at the time of migration.
Because it is easier to obtain such data for the post-1960 period, and also to
maintain comparability with the analysis that will be conducted in the next
section across host countries, the empirical study in this section focuses on
explaining the variation in quality across the four cohorts that arrived in the
post-1960 period. Hence, there are 164 observations (41 countries times 4
cohorts per country) in the data set analyzed here. The aggregate variables
used in the analysis are, for the most part, obtained from my earlier study
(Borjas 1987) and are described in table 1.4. They include measures of polit-
ical conditions in the country of origin, mobility costs, and characteristics of
the income distribution (the mean and the variance).
The empirical analysis of the differences in the present value differentials
between immigrants and natives in the 164-observation data set is presented
in table 1.5.” The first column of the table presents estimates of the reduced-
form equation derived in (1 1). This regression reveals that a relatively small
number of country-specific variables explains a large fraction of the inter- and
Determinants of Selection in Unobserved Characteristics
Table 1.4 Definition of Aggregate Variables
Mean (Standard
Variable Definition Deviation)
FREE
COMMUNIST
LOSTFREE
INEQUALITY
UNEMPLOYMENT
USLAW
ENGLISH
DISTANCE
AGNP
= 1 if the country had a competitive party system at
the time of migration, 0 otherwise
= 1 if the country had a Communist government at the
time of migration, 0 otherwise
= 1 if the country lost a competitive party system
within the last 10 years, 0 otherwise
= ratio of household income of the top 10 percent of
the households to the income of the bottom 20 per-
cent of the households ca. 1970
Unemployment rate in the United States at the time of
migration
= 1 if migration occurred after 1970, 0 otherwise
Fraction of 1975-80 cohort of immigrants who speak
English well or very well
Number of air miles (in thousands) between the coun-
try’s capital and the nearest U.S. gateway (Los An-
geles, Miami, and New York)
Difference in (In) GNP per capita between the country
of origin and the United States at the time of migra-
tion
.48
(.SO)
.I7
(.38)
.08
(.W
7.50
(6.08)
6.25
(1.64)
.5
.74
(.25)
3.73
(1.98)
~50)
- 1.39
(1.05)
Note: For additional details on the creation of these variables, see Borjas (1987).
52 George J. Borjas
Table 1.5 Determinants of Differences in Unobserved Characteristics
Reduced-form Equation Structural Equation'
Variable Coefficient t Coefficient t
CONSTANT
FREE
COMMUNIST
LOSTFREE
INEQUALITY
UNEMPLOYMENT
USLAW
ENGLISH
DISTANCE
AGNP
R2
h
~ ( US LAW = 0)
/@SLAW = 1)
- ,1574
.0410
,0113
- ,0333
- ,0040
.0334
- ,1593
,0797
,0003
.0495
,402
. . .
( - 1.61)
(1.45)
C.37)
( - .93)
(-1.79)
(1.81)
(1.70)
(4.02)
(-2.61)
~ 0 5 )
. . .
- ,0537
,0336
,0072
- ,0106
- ,0029
,0108
- ,0505
. . .
. . .
,382
.0085
- .0167
- .O419
( - 1.70)
(4.31)
(.69)
( - .86)
(-4.56)
(1.70)
( - 2.42)
. . .
. . .
(-5.24)
( ,791
( - 3.79)
'All the variables in the structural equation are interacted with A, the selection variable. For
details, see eq. (12).
intracountry variance in the unobserved quality of immigrants. Many of the
aggregate characteristics are statistically significant. Consider, for instance,
the variable measuring the extent of income inequality in the country of ori-
gin. The coefficient of this variable is negative and marginally significant, as
predicted by the theory. Similarly, the difference between mean GNP in the
country of origin and mean GNP in the United States is positive and signifi-
cant, indicating the fact that migrants from countries with advanced econo-
mies are characterized by larger levels of unobserved abilities or productivi-
ties.
It is worth stressing that the measure of income inequality not only is statis-
tically significant but also has a sizable numerical effect on the quality of the
immigrant flow. This point is best illustrated by considering two countries: the
United Kingdom and Mexico. The inequality measure takes on a value of 4.0
for the United Kingdom and of 12.3 for Mexico. The regression coefficient in
table 1.5 suggests that, holding all other factors constant, Mexican immi-
grants earn 3-4 percent less than British immigrants simply because of the
selectivity effects of higher levels of income inequality.
Three other variables seem to be quite important in the regression. The first
measures the English proficiency of the immigrant pool. Immigrants from
countries where English is prevalent do much better in the United States than
immigrants from non-English-speaking countries. Second, the unemployment
rate in the United States is an important determinant of immigrant quality: the
higher the unemployment rate at the time of migration, the better the quality
of the migrant pool. This result is consistent with the Roy model if unemploy-
ment particularly effects the earnings opportunities of less-skilled workers.
53 Immigration and Self-selection
For instance, an increase in the unemployment rate will worsen the opportu-
nities for persons in the lower end of the ability (i.e., income) distribution and
hence will lead to reduced incentives for these persons to migrate. The quality
of the self-selected immigrant pool increases as a result of the withdrawal of
these persons from the immigration market.
Finally, the reduced-form regression in table 1.5 introduces a dummy vari-
able signaling whether the cohort arrived in the post-1970 period. Recall that
U.S. immigration policy was changed drastically by the 1965 amendments
(which became fully effective in 1968). Hence, post-1968 cohorts, holding
constant characteristics of the country of origin, should differ from earlier
cohorts. This is precisely what the results in table 1.5 indicate. In particular,
post-1970 cohorts have nearly 16 percent lower (relative) earnings over the
life cycle than immigrants who arrived prior to the change in U.S. policy. This
result provides striking evidence of a significant structural shift that occurred
in the unobserved quality of U.S. immigrants in the last two decades.
As noted earlier, since data exist on the emigration rate of immigrants from
any given country of origin (i.e., the number of immigrants in a particular
cohort as a fraction of the population of the country of origin at the time of
migration), the selectivity variable A can be calculated, and the structural
equation in (12) can be estimated. The structural equation is written as
Q, = hX, and the h-function can be approximated by h = PZ, where Z is the
vector of variables proxying for the relevant primitive parameters. Hence, the
empirical counterpart to (12) is Q, = P(Zh). This structural equation is pre-
sented in the second column of table 1.5. The selectivity variable directly
controls for changes in mobility costs and means of income distributions, and
these variables are omitted from the structural regression. Remarkably, the
structural equation leads to estimates that are highly significant and very sup-
portive of the Roy model. In particular, the inequality variable becomes neg-
ative and very significant, the unemployment variable remains positive and
significant, and the dummy variable indexing post-1970 cohorts remains neg-
ative and strong.
The estimated regression parameters can be used to calculate h = 6Z. The
estimates of h are presented at the bottom of column 2 in table 1.5. Three
estimates are presented: one evaluated at the mean of all the variables, a sec-
ond one evaluated at the same means but letting the dummy variable USLAW
index pre-1970 cohorts, and a third evaluated at the same means but letting
the dummy variable USLAW index post-1970 cohorts. These simulations
show that there seemed to be weak positive selection prior to 1970 but very
strong negative selection in the post-1970 period.
1.3.4
As noted earlier, self-selection occurs not only on the basis of unobserved
ability but also on the basis of observed characteristics such as education.
Table 1.2 documented the strong differences in educational attainment across
immigrant groups from different countries. In addition, it was seen that the
Determinants of Selection in Education
54 George J. Borjas
observed educational attainment of immigrant groups differed from the mean
educational attainment of the population in the country of origin. It is of inter-
est, therefore, to analyze whether the same conceptual framework that ex-
plains the differences in unobserved characteristics also explains the differ-
ences in educational attainment.
The key implication of the Roy model is that highly educated persons are
most likely to originate in countries that have a low rate of return to education
(relative to that found in the United States). Put another way, holding the mean
educational attainment in the population of the source country constant, there
should be a negative relation between the mean educational attainment of im-
migrant national origin groups and the rate of return to schooling in the source
country.
A detailed study by Psacharopoulos (1973) reports the private rate of return
to higher education for a number of countries. Unfortunately, there are only
15 countries in common between his sample and the sample of countries that
are important sources of immigration. Nevertheless, using the data presented
in table 1.2 for each of the four post-1960 cohorts (so that there are 60 obser-
vations), the following regressions were estimated:
E( s 1 I > 0) = 13.02 + .23 p, -9.76r, R2 = .236,
(9.78) (1.90) ( - 2.44)
E(s I I > 0) = 8.01 + .66p,, -9.82r - 1.44 AGNP, R2 = .421
(4.81) (4.51) (-2.79) (-4.24)
where the dependent variable is the mean educational attainment of the immi-
grant group in the United States, ps is the mean educational attainment in the
source country, r is the rate of return to education in the source country, AGNP
is the percentage difference in per capita GNP between the source country and
the United States at the time of migration, and the t-statistics are presented in
parentheses.
The most important finding in these regressions is the significant negative
effect of the rate of return to schooling in the source country on the educa-
tional attainment of the immigrant pool. A 10 percentage point increase in the
rate of return to schooling decreases the mean educational attainment of the
immigrant group by almost one year. The key implication of the Roy model,
therefore, is confirmed by the data. The educational composition of the
sample of immigrants is determined by the relative payoff to schooling in the
source country.
The regressions also indicate that the mean level of educational attainment
in the country of origin has a positive effect on the mean educational attain-
ment of immigrants and that the coefficient, as predicted by the theory, is
between zero and one. I should add, however, that this confirmation of the
theory-like the results regarding the rate of return to schooling-should be
treated with some caution because the data on international differences in
mean education levels and rates of return are measured with substantial error.
55 Immigration and Self-Selection
1.3.5 The Unstandardized Wage Differential between Immigrants
and Natives
Up to this point, the empirical analysis has focused on ascertaining the
types of selection that occur on the basis of unmeasured skills and on the basis
of educational attainment. As noted in the theoretical section, selection occurs
along every single dimension of skills that is valued by the labor market, with
particular skills flowing to countries that value those skills the most. As a
result of these selection processes, the actual earnings of the typical immi-
grant in the United States are likely to differ substantially from the earnings of
the typical white native.
To determine the effect of self-selection in the various dimensions of skills
on the earnings gap between immigrants and natives, the 41 regression models
were reestimated without including any demographic characteristics in the
vector X (except, of course, age and age squared so as to trace the age-
earnings profile over the working life). The resulting present value differen-
tials are presented in table 1.6. As can be seen by comparing tables 1.3 and
1.6, the relative earnings of some immigrant groups are lowered significantly
when no standardization for demographic variables is conducted. For in-
stance, the most recent immigrant cohort from Mexico earns about 47.8 per-
cent less than white natives, but they earn only about 15.7 percent less than
demographically comparable white natives. On the other hand, the most re-
cent cohort of immigrants from the United Kingdom earns about 18.8 percent
more than white natives, but only about 10.5 percent more than demographi-
cally comparable white natives. In other words, some immigrant groups have
demographic characteristics that are much less valuable than those of natives,
while other immigrant groups have demographic characteristics that are much
more valuable than those of the typical native.
It turns out that the same country-specific variables that explain the varia-
tion in standardized earnings differentials among source countries also explain
the variation in the unstandardized differentials. Table 1.7 analyzes the deter-
minants of the intercountry variation in unstandardized immigrant earnings
(analogous to the second-stage regressions on standardized earnings presented
in table 1.5). It is worth noting that, as before, immigrants originating in
countries with high levels of income inequality have lower earnings than im-
migrants originating in other countries. This is not surprising since the level
of income inequality can be interpreted as a summary index of skill prices.
Thus, the key prediction of the Roy model is confirmed by the analysis of the
actual earnings (as opposed to the standardized earnings) of immigrants.
1.4 Immigrant Sorting among Host Countries
The last section showed that the labor market performance of immigrants
currently living in the United States is strongly influenced by the economic
and political characteristics of the country of origin at the time of migration.
56 George J. Borjas
Table 1.6 Unstandardized Present Value Differentials between Immigrants and
Natives
Year of Arrival
Country 1975-79 1970-74 1965-69 1960-64 1950-59 <1950
Europe:
Austria
Czechoslovakia
Denmark
France
Germany
Greece
HwzwY
Ireland
Italy
Netherlands
Norway
Poland
Portugal
Romania
Soviet Union
Spain
Sweden
Switzerland
United Kingdom
Yugoslavia
,0142
,0561
,6004
(3.54)
.2658
(3.05)
,1673
(3.64)
- .I990
( - 3.52)
-.I137
(-1.34)
,0148
(. 17)
- ,0645
(-1.61)
.4170
(5.17)
,3004
(2.20)
- ,2091
(-4.22)
- .2826
(-4.65)
-.I151
(-1.13)
- .2086
( - 3.29)
.I182
(1.24)
.1826
,2868
(2.42)
,1875
(5.22)
(. 10)
( . 50)
(1.20)
- ,0115
(-.17)
.1987
(1.72)
,0045
.0933
(.49)
,1464
(1.89)
,1054
(2.32)
- ,2996
(-7.83)
- ,0799
( - 1.19)
,0306
( .49)
- .0947
( - 3.72)
~ 0 3 )
,0036
,4233
(2.50)
.0479
(1.21)
- ,2518
( - 5.66)
,2340
(2.94)
- ,0061
(-.lo)
,0601
(.87)
,2013
(1.36)
.1716
(1.39)
.1705
(5.05)
. w 3
(.lo)
~ 0 3 )
,3104
(4.20)
,0805
(1.54)
.3028
(2.73)
.0173
(.28)
.w37
(3.27)
- ,2326
(-7.13)
.0442
(.82)
.I284
(2.57)
- ,0526
( - 2.73)
,2121
(3.57)
,3319
(3.12)
,0736
(2.33)
- ,1962
( - 5.46)
.1141
(1.61)
- ,0293
( - .47)
,0221
( .39)
,3170
(2.64)
,2385
(2.67)
,1898
(7.88)
,0123
(.36)
,1157
(2.03)
,1289
(1.99)
,0409
(.50)
- ,0847
(-1.73)
,0359
(1.78)
- ,0534
( - 1.37)
,0412
(35)
,0676
(2.00)
( - .76)
,1008
(2.50)
,2751
(3.24)
,0186
(.75)
- ,1538
( - 3.70)
,0808
(1.15)
- ,0118
( - .26)
- ,0084
(-.14)
.1426
(1.44)
.I904
(2.60)
,1198
(5.80)
,1105
(2.90)
- ,0132
,0535
(1.27)
.I191
(2.58)
.0312
(.W
,0116
( .26)
.0526
(3.29)
- ,0370
( - 1.26)
,0925
(3.36)
,0299
(.94)
.oO07
,0295
,2018
(3.26)
,0563
(2.45)
- ,1270
( - 3.03)
,0724
(1.35)
,0429
(1.36)
- ,1335
( - 2.35)
,0867
(1.18)
.I589
(2.66)
,0905
(5.20)
.0938
(3.36)
c.03)
~ 9 0 )
,1135
(1.49)
.1745
(2.46)
,1331
(1.07)
,1896
(2.68)
.2464
(8.44)
- ,0627
( - 1.09)
.3162
(5.20)
-.I139
( - 2.40)
,0344
(1.29)
- ,0980
(-1.48)
,0117
,1322
(3.75)
,0159
,0895
(33)
. I402
(2.74)
- ,0793
( - .69)
.1199
(.98)
.2197
(1.99)
.0490
(1.58)
,0549
(.83)
(. 14)
(. 17)
57 Immigration and Self-selection
Table 1.6 (continued)
Year of Arrival
Country 1975-79 1970-74 1965-69 1960-64 1950-59 <I950
Asia and Africa:
China - ,3703
(-7.24)
Egypt ,0101
(.lo)
India - ,0705
( - 1.04)
Iran ,101 1
(.94)
Israel - ,2692
(-2.66)
Japan ,1857
(2.24)
Korea - ,0846
( - .87)
Philippines - . I415
(-2.84)
Americas:
Argentina - ,2239
Brazil .I575
Canada .I862
Colombia - ,4451
(-2.47)
(1.19)
(4.95)
(-5.59)
Cuba - ,3450
( - 7.13)
( - 8.37)
(-4.06)
Dominican Republic - ,6794
Ecuador - ,4016
Guatemala - ,4927
( - 3.94)
( - 5.64)
Jamaica - .3938
Haiti - ,5907
( - 5.62)
( - 17.14)
(- 2.49)
Tobago ( - 2.73)
Mexico - .4780
Panama - ,3263
Trinidad and - .3313
- ,2659
(-7.59)
,0583
( .66)
,1339
(2.28)
.lo77
(1.39)
-.I891
( - 2.50)
,0683
( I .02)
,0047
,0236
(.73)
~ 0 3 )
- ,3123
( - 3.98)
- ,0816
( - .74)
,0593
(1.69)
- .3141
(-4.53)
- ,1693
(-5.52)
- ,4560
(-6.89)
- .2428
( - 2.75)
- ,4041
( - 4.48)
- ,3037
(-3.31)
- .2388
(-4.85)
- ,3518
( - 19.94)
- ,0653
(-.71)
-.I167
( - 1.35)
-.I115
( - 3.97)
,1559
(2.03)
,2617
(4.87)
,1501
(2.09)
-.I761
( - 2.49)
,0565
(37)
.0826
(1.06)
.0245
~ 9 2 )
-. I801
(-2.76)
,0717
(.82)
.I069
(4.58)
- ,1903
(-3.04)
-.1877
(-7.93)
- ,3963
(-6.81)
- ,2789
(-3.44)
- ,3604
(-4.41)
- .I307
(-1.55)
- ,1406
(-3.35)
- .3074
( - 21.97)
- ,1376
(-1.85)
- ,0435
(e.60)
- .0379
(-1.33)
,0914
(1.17)
,3240
(6.1 1)
,1339
(1.89)
- ,0764
(-1.12)
.0927
(1 S5)
,1141
(1.52)
- ,0109
( - .32)
- ,0122
( - .20)
.0807
(.87)
.0654
(3.80)
- .0522
( - .89)
- .0234
(-1.15)
- .3034
(-5.44)
(e.44)
- ,2532
( - 2.68)
,0069
(. 10)
- ,2124
(-4.12)
- ,2798
( - 20.59)
- .0905
(-1.29)
,0693
(.75)
- ,0339
,0157
. I147
(1.59)
.2784
(5.24)
,2055
(2.90)
,1744
(2.99)
,0500
(1.00)
- .0169
( - .26)
- ,1089
(-3.31)
( 5 1 )
,0736
(1.17)
.0615
(.74)
.0685
(4.37)
,1791
(3.10)
- .I658
(-7.08)
- ,1546
( - 2.37)
,0186
(.26)
- ,1017
(-1.14)
- .0527
( - .61)
( - .94)
- ,2350
( - 15.06)
- ,0433
( - .63)
,1741
(1.73)
- .0471
- ,0562
( - 1.05)
.3073
(1.59)
,2143
(1.68)
,2627
(1.68)
,3943
(2.56)
- .0139
(-.lo)
.3015
(1.30)
- ,1758
(-3.31)
,4027
(2.47)
- .2328
(-1.14)
,0537
(2.06)
( - .65)
- ,0256
( - .45)
- ,0256
( - .33)
,1277
(.58)
,4213
(2.23)
- ,2614
(-1.22)
,0514
(.55)
- ,2918
(-9.42)
,2219
(1.64)
.0340
-.I150
(. 17)
Note: The r-ratios are presented in parentheses.
58 George J. Borjas
Table 1.7 Determinants of Differences in Unstandardized Earnings Differentials
Reduced-form Equation Structural Equationa
Variable Coefficient t Coefficient t
CONSTANT
FREE
COMMUNIST
LOSTFREE
INEQUALITY
UNEMPLOYMENT
USLAW
ENGLISH
DISTANCE
AGNP
R2
h
h(USLAW = 0)
h(USLAW = 1)
- ,4069
,0372
,1056
- ,0383
- ,0061
,0188
-.I322
,3917
.0145
.0133
.754
(-4.39)
(1.39)
(3.71)
( - 1.34)
( - 2.30)
( 1 .w
(-2.13)
(9.79)
(3.38)
(1.06)
- ,0505
,0692
,0344
- ,0047
- ,0041
,0054
- .0354
. . .
3 2
- ,0248
- ,0071
- ,0425
(-1.24)
(7.50)
(2.73)
( - .34)
(-4.00)
( .66)
( - I .27)
(-6.42)
( - .52)
( - 2.77)
"All the variables in the structural equation are interacted with A, the selection variable. For details,
see eq. (12).
Potential emigrants in the source countries, however, chose to come to the
United States instead of migrating to other potential countries of destination.
In a sense, the observed pool of immigrants in the United States is the out-
come of competition in the immigration market among various countries of
destination. Different countries, by offering different immigration policies and
different income distributions, will attract different kinds of immigrants.
Three countries, Australia, Canada, and the United States, have been the
main countries of destination for permanent migrants in recent years. U.N.
statistics, for example, report that, in the period 1975-80, nearly five million
persons migrated to a different country, with nearly two-thirds of these indi-
viduals migrating to one of these three countries.
Each of these countries, of course, is characterized by a long history of
immigration. The size of the recent flows generated by the self-selection of
immigrants into each of the three potential countries of destination is illus-
trated in table 1.8. Over the period 1959-81, about 14.7 million persons le-
gally left the various countries of origin and migrated to either Australia, Can-
ada, or the United States. Sixty percent of these migrants chose the United
States as their destination, and the remainder were evenly split between Aus-
tralia and Canada. Table 1.8 also shows that these statistics vary significantly
between the early part of the period (1959-70) and the later (197 1-8 1). Re-
cent migrants are disproportionately more likely to select the United States as
their destination (nearly two-thirds of migrants in the 1970s chose to do so)
and disproportionately less likely to choose Australia as their destination
(only 14 percent did so).
Table 1.8 Migration Flows to the United States, Canada, and Australia
Origin
Period of Migration
1959-70 I97 1-8 1 1959-81
Number % to % to % to Number %t o %t o % to Number 9%to %t o % to
(1,OOOs) U.S. Canada Australia (1,OOOs) U.S. Canada Australia (1,OOOs) U.S. Canada Australia
Africa
America
Asia
United Kingdom
Europe (excluding U.K.)
Oceania
Total
~~
115.1 37.5 29.6 32.8
2,111.6 84.9 13.4 1.7
708.3 69.5 19.2 11.3
1,322.9 20.3 28.8 50.9
2,583.4 47.5 28.9 23.6
123.7 18.9 32.5 48.6
6,965.0 55.2 23.3 21.5
~~
220.5 48.3 32.4 19.3 335.5 44.6 31.5 23.9
2,687.7 81.0 15.9 3.1 4,799.3 82.7 14.8 2.5
2,580.8 73.5 17.7 8.7 3,289.0 72.7 18.0 9.3
751.1 18.4 31.7 49.9 2,074.0 19.6 29.8 50.5
1,309.2 55.7 26.0 18.3 3,892.6 50.3 27.9 21.8
176.9 23.5 19.4 57.2 300.5 21.6 24.8 53.6
7,726.2 65.9 20.3 13.8 14,690.9 60.8 21.7 17.5
Sources: U.S. Department of Commerce, Statistical Abstract of the United States; U.S. Immigration and Naturalization Service, Statistical Yearbook of the
Immigration and Naturalization Service; Canada Statistics, Historical Statistics of Canada and Canada Yearbook; Australian Department of Immigration and
Ethnic Affairs, Australian Immigration (1982).
60 George J. Borjas
These aggregate statistics mask important country-of-origin differences.
During the period 1971-81, the United States was less likely to attract immi-
grants from Africa, the United Kingdom, and Europe and significantly more
likely to attract immigrants from Asia or North and South America. Canada,
on the other hand, seemed a relatively attractive destination for immigrants
from Africa, the United Kingdom, and Europe, while Australia was the des-
tination of choice for persons emigrating from the United Kingdom: nearly
half the two million persons who left the United Kingdom in the period 1959-
8 1 migrated to Australia. l 3
1.4.1
One important constraint on the size and the composition of the flow of
migrants to potential host countries is the set of statutes and policies used by
the various countries to screen the applicant pool. U.S. immigration policy,
prior to the 1965 amendments to the Immigration and Nationality Act, was
guided by the objective of restricting the migration of persons whose national
origin did not resemble the national origin and ethnic composition of the
United States population in 1920. The 1965 amendments abolished the “dis-
criminatory” national origin quota system and instituted the goal of family
reunification as the main objective of U.S. immigrant policy. These changes,
as we saw above, may have been responsible for a very large decline in the
unobserved skills of immigrants admitted by the United States.
Canadian immigration policy, until 1961, also had a preferential treatment
of immigrants originating in Western European countries. The 1962 Immigra-
tion Act (and further relatively minor changes in the statutes and regulations
through the 1970s) removed the country-of-origin and racial restrictions and
shifted emphasis toward skill requirements. Under the new regulations, poten-
tial migrants who were not relatives of Canadian citizens or residents could
enter Canada if they passed a “test.” Applicants were graded and given up to
100 points according to a “point system,” and 50 points were necessary
to obtain permission to migrate to Canada. These points were given according
to the applicant’s education (a point per year of schooling, up to 20 points),
occupational demand (10 points if the applicant’s occupation was in strong
demand in Canada), age (up to 10 points for applicants under the age of thirty-
five, minus 1 point for each year over age thirty-five), a “personal assessment”
by the immigration officer that was valued up to 15 points, etc. In 1976, the
Canadian Immigration Act was amended to incorporate the goal of family
reunification as an important policy objective.
Australian immigration policy has a long history of restricting the migra-
tion of persons who are not of British origin. These restrictions, known as the
“White Australia Policy,” operated both in terms of denying entry to persons
of non-British or non-Northern European origin and in terms of denying fi-
nancial assistance (to cover transportation and resettlement expenses) to un-
desirable migrants.
Migration Policies in Host CountriesI4
61 Immigration and Self-Selection
World War I1 raised doubts among Australian officials about the feasibility
of defending a large continent with a small population, and a series of govern-
ments pursued a national policy of substantially increasing the number of im-
migrants who chose Australia as their destination. This objective, however,
could not be achieved by allowing only British citizens to immigrate, and thus
Australia began looking elsewhere for migrants (e.g., Germany, the Nether-
lands, Malta, Italy, and Greece all signed formal arrangements with the Aus-
tralian government to recruit and assist persons from these countries in their
migration to Australia). Further political changes in Australia led to the abol-
ishment of the White Australia Policy in 1972. An immigration policy devoid
of discrimination by national origin and race was announced, and a point sys-
tem based on the Canadian system was instituted. During the early 1980s,
Australia began to stress the concept of family reunification in its migration
policy (see Birrell 1984). It is unlikely, however, that this shift in policy will
have much effect on the 1981 Australian Census data that will be analyzed
below.
The effect of these changes in immigration policy on the national origin
composition of the immigrant pool in each of the countries is documented in
table 1.9. In all host countries, the national origin of the immigrant population
has changed drastically over time. For example, in both Canada and the
United States, the share of migrants originating in European countries de-
clined drastically between the 1960s and the 1970s. During the 1960s, 23.5
percent of immigrants to Canada originated in the United Kingdom, and an
additional 46.0 percent originated in other European countries. During the
197Os, the fractions had fallen to 15.2 and 21.7 percent, respectively. Con-
versely, the fraction of immigrants originating in Asia was only 8.4 percent
during the 1960s, and this fraction had increased to 29.1 percent during the
1970s.
Table 1.9 shows that the United Kingdom accounted for nearly half the
migrants to Australia during the 1960s but for only a third of the migrants
during the 1970s. A similar decline is observed in the fraction of Australian
immigrants originating in other European countries: from 40.8 to 22.4 per-
cent. On the other hand, the fraction of immigrants from Asia increased from
5.3 to 21.1 percent, a fourfold increase in a ten-year period.
1.4.2 Data and Descriptive Statistics
The data are drawn from Public Use Samples of the Censuses available for
each of the three destination countries. The U.S. data are identical to that used
in the previous section and require no further description.
The Canadian Censuses were conducted in 197 1 and 198 1. Both these Cen-
suses have the important characteristic that they report the year in which
foreign-born persons arrived in Canada. Hence, the agingkohort decomposi-
tion described in section 1.2 can be carried out. The 1971 data for both im-
migrants and natives residing in Canada are a 1/100 random sample of the
Table 1.9 Migration Flows into the United States, Canada, and Australia, 1959-81
Period of Migration
United States Canada Australia
1959-70 1971-81 1959-81 1959-70 1971-81 1959-8 1 1959-70 197 1-81 1959-81
Number %of Number %of Number %of Number %of Number %of Number %of Number %of Number %of Number %of
(1,OOOs) Total (1,OOOs) Total (1,OOOs) Total (1,OOOs) Total (1,OOOs) Total (1,ooOs) Total (1,OOOs) Total (1,OOOs) Total (1,OOOs) Total Origin
Africa
America
Asia
United
Europe
Oceania
Total
Kingdom
and other
43.2 1.1
1,792.0 46.6
492.2 12.8
106.5 2.0 149.7
2,175.7 42.7 3,967.7
1,898.1 37.2 2,390.3
1.7 34.1 2.1 71.5 4.6 105.6 3.3 37.8 2.5 42.5 4.0 80.2 3.1
44.3 283.5 17.5 427.9 27.3 711.4 22.3 36.1 2.4 84.1 7.9 120.2 4.7
26.7 136.3 8.4 457.3 29.1 593.6 18.6 79.8 5.3 225.4 21.1 305.1 11.9
268.8 7.0
1,228.2 31.9
23.4 .6
3,847.8
138.5 2.7 407.3
729.5 14.3 1,957.7
41.5 .8 64.9
5,089.9 8,937.7
4.6 381.2 23.5 237.8 15.2 619.0 19.4 672.9 44.9 374.8 35.1 1,047.7 40.9
21.9 745.4 46.0 340.1 21.7 1,085.5 34.0 609.8 40.8 239.6 22.4 849.4 33.1
.7 40.2 2.5 34.3 2.2 74.5 2.3 60.1 4.0 101.1 9.5 161.1 6.3
1,620.7 1,568.9 3,189.6 1,496.3 1,067.5 2,563.7
Sources: U.S. Department of Commerce, Statistical Abstract of the United States; U.S. Immigration and Naturalization Service, Statistical Yearbook of the
Immigration and Naturalization Service; Canada Statistics, Historical Statistics of Canada and Canada Yearbook; Australian Department of Immigration and
Ethnic Affairs, Australian Immigration (1982).
63 Immigration and Self-Selection
Canadian population, while the 1981 micro file is a 2/100 random sample of
the Canadian population. All observations that satisfy the sample restriction
of being prime-age men (aged 25-64), not self-employed, not residing in
group quarters, and whose records report positive annual earnings in the year
prior to the Census are used in the analysis.
The Australian data used in this paper are drawn from the 1981 Census of
Population and Housing, the only micro Australian Census file available at
present. This Census file is a 1 / 1 0 random sample of the Australian popula-
tion, and the entire sample (for both immigrants and natives) that satisfies the
sample restrictions listed above is used.
Three important problems are raised by the Australian data. First, only one
Census is available; therefore, the aginglcohort decomposition cannot be con-
ducted. The Australian results, therefore, are not directly comparable to those
for the other two countries. Nevertheless, a simple solution that allows some
rough comparisons will be proposed below. Second, the Australian Census
does not report annual earnings but instead reports annual incomes (which
include nonsalary receipts). This problem may not be very serious since the
analysis focuses on nativehmmigrant earnings differences, and self-employed
persons are omitted from the study. Finally, the Australian Census (unlike the
U. S. or Canadian data) does not contain good measures of labor supply.
Hence, a wage rate for the year prior to the Census cannot be calculated. The
empirical analysis in this section, therefore, will be conducted on the loga-
rithm of annual earnings.
Table 1.10 presents summary statistics (mean earnings and education) as
well as sample sizes for the various samples that will be used in the analysis.Is
In addition, table 1.10 decomposes the immigrant population in each of the
host countries according to the continent of origin. This decomposition by
continent (rather than country) is mandated by the fact that, in both the Aus-
tralian and the Canadian Censuses, the decomposition by country leads to a
very small number of observations for most countries. In addition, the Cana-
dian Censuses identify the country of origin only for a select group of Western
European immigrants.
The results for the United States, as expected, show a downward trend in
the earnings of immigrants (relative to natives) over the decade. The average
immigrant in 1970 earned, on average, about as much as the typical native
worker. By 1980, however, immigrant earnings were about 15 percent below
the native wage. The Canadian data show little change in the relative earnings
of immigrants between 1971 and 1981. In both Censuses, the average immi-
grant had slightly higher earnings than the typical native worker. The excep-
tion seems to be immigrants originating in Latin America: their earnings are
about 10 percent lower than those of Canadian natives in 1971 but 19 percent
lower in 198 1. The Australian Census shows that the typical immigrant in
1981 had roughly the same earnings as the typical native person and that the
differential varied somewhat by country of origin.
64 George J. Borjas
Table 1.10 Summary Statistics
Country of Destination: United States
1970 1980
Country of Origin In ( w) EDUC N In(w) EDUC N
Natives
Asia
Africa
Europe
Latin America
All immigrants
Natives
Asia
Africa
Europe
Latin America
All immigrants
Natives
Asia
Africa
Europe
Latin America
All immigrants
8.99 11.5 28,978 9.61 12.7 15,071
8.88 13.3 3,495 9.47 14.6 25,288
8.88 13.9 172 9.40 15.3 2,622
9.06 10.8 16,922 9.69 12.1 42,734
8.67 9.2 7,507 9.23 9.4 48,929
8.95 10.8 32,491 9.46 11.7 134,252
Canada
1971 1981
8.82 9.9 28,049 9.79 11.3 61,205
8.72 13.2 409 9.66 13.6 2,372
8.86 14.1 119 9.74 14.0 504
8.86 10.0 6,633 9.86 10.9 12,193
8.72 12.0 223 9.60 12.1 1,229
8.86 10.5 8,018 9.81 11.7 17,417
Australia, 1981
9.39 11.6 23,086
9.34 12.9 1,074
9.45 13.1 267
9.34 11.4 7,799
9.35 12.1 102
9.36 11.7 9,936
It is instructive to compare the Australian statistics with the relevant num-
bers for Canada and the United States. For instance, European immigrants in
Australia actually have the lowest education levels of any of the migrant
groups in Australia and have a wage disadvantage of only 5 percent. In Can-
ada, European immigrants also tend to have slightly lower educational levels
but higher earnings than natives, while in the United States European immi-
grants outperform all other immigrant groups despite the fact that they have
lower educational levels than the native population. This comparison (as well
as similar comparisons for other regions of origin) reveals the nonrandom
sorting of migrants across the various host countries.
An important insight is provided by these statistics: generalizations about
the productivity or earnings capacities of ethnic or national groups are mis-
leading since they ignore the self-selectivity that generated the composition of
the migrant pool in each of the host countries. In other words, there is no such
thing as the effect of Asian ethnicity or race on immigrant earnings. The value
65 Immigration and Self-selection
attached by the host country’s labor market to ethnichacia1 characteristics de-
pends greatly on the kinds of selections that generated the particular flow of
migrants.
1.4.3 1980-81 Cross-sectional Regressions
Since the agingkohort decomposition cannot be conducted for the Austra-
lian data, it is instructive to begin the empirical analysis by focusing on the
1980-81 cross section. Table 1.1 1 presents the cross-sectional earnings func-
tion estimated separately in the samples of immigrants and natives in each of
the three countries of destination. The regressions in the native samples are of
interest mainly because they are so similar across the destination countries.
The coefficients of age, marital status, and urbanization status all have the
expected signs and are of similar magnitudes whether the labor market is in
Australia, Canada, or the United States. The only coefficient that seems to be
an outlier in the native samples is that of education in Australia, where the
coefficient is almost twice as large as that in the United States or Canada.
The regressions in the immigrant samples are interesting because they illus-
trate the general result that practically all socioeconomic variables have a
smaller effect among immigrants than among natives regardless of the country
of destination. The earnings of immigrants are much less responsive to socio-
economic characteristics than the earnings of natives in these economies.
The immigrant regressions in table 1.11 also include a vector of variables
indicating the time of migration.I6 An important use of these coefficients (and
of the socioeconomic variables) is to predict the size of the wage differentials
between immigrants and natives for each of the cohorts. These predictions are
calculated using the mean socioeconomic characteristics of the immigrant
sample in each of the host countries. In addition, these predictions are ob-
tained by holding the age of immigration constant at 20 for all cohorts.
Hence, the typical immigrant in the 1975-1980 cohort is 23 years old when
the prediction is calculated, the typical immigrant in 1970-74 is 28 years old,
etc. The predicted age-earnings profile, therefore, incorporates both aging and
cohort effects. These profiles are presented in table 1.12.
The U.S. and Canadian profiles resemble the ones usually reported in the
literature: the earlier cohorts, either because they are older and have been in
the country longer or because there are vintage or cohort effects, do much
better in the labor market than more recent cohorts. Table 1.12, however,
shows that the Australian experience is very different. The Australian cross-
sectional age-earnings profile for immigrants is essentially flat! In fact, it is
impossible to find any statistical difference in the relative earnings of immi-
grants among the cohorts that arrived in Australia after 1950. Their relative
earnings hover around 7-8 percent less than natives, and there is no discem-
ible trend over time. This result implies that, if there is any assimilation effect
in Australia, the quality of immigrants to Australia must have increased dur-
ing the period 1960-80. Hence, a simple comparison of the cross-sectional
66 George J. Borjas
Table 1.11 1980-81 Cross-sectional regressions
Country of Destination
United States Canada Australia
Sample Coeff. t Coeff. t Coeff. t
Natives:
CONSTANT
EDUC
AGE
AGE'
MAR
HLTH
URBAN
R'
All immigrants:
C 0 N S T A N T
EDUC
AGE
AGE^
MAR
HLTH
URBAN
1970-74Wave
196549 Wave
1960-64 Wave
1950-59 Wave
Pre-1950 Wave
RZ
6.6488
.0587
,0841
- ,0009
,3151
- ,3337
,1545
,193
6.6378
,0497
,0802
- ,0009
,2325
- ,3502
,0574
.2107
,3141
,3750
,4436
,4752
,226
(76.33)
(33.92)
(20.17)
( - 18.00)
(23.53)
( - 15.15)
( 12.07)
(223.77)
(133.61)
(55.39)
( - 51.35)
(50.52)
( - 34.48)
(9.43)
(36.81)
(51.89)
(56.74)
(74.88)
(64.63)
7.0465
.0510
,0873
- .0009
,2973
,1036
. . .
1.71
7.3415
,0415
,0710
- .oO08
,2190
. . .
- ,0016
,1609
,2816
,2825
,3679
.4287
.I63
( 193.01)
(76.26)
(49.42)
(-45.21)
(51.10)
(22.78)
. . .
(95.72)
(40.97)
(19.31)
(18.42)
(-.16)
(9.73)
( 18.03)
(1 5.39)
(25.59)
(17.50)
(- 18.44)
. . .
6.3522
.0908
,0886
- ,001 1
.2727
,1605
,245
. . .
6.7307
,0748
.0779
- ,0010
,2013
,1079
.0444
,0491
.0810
.0811
.1159
. . .
1.88
(104.68)
(58.77)
(32.01)
( - 34.61)
(31.31)
(16.61)
. . .
(66.17)
( 16.86)
( - 18.70)
(14.16)
(5.41)
(2.11)
(2.36)
(3.68)
(4.18)
(4.63)
(35.59)
. . .
regressions across the destination countries leads to an important finding
about the trends that mark the self-selection of immigrant flows to the host
countries over the last two decades.
1.4.4 Present Value Differentials
Since two Censuses are required to identify aging and cohort effects, the
analysis of equations (26) and (27) is initially restricted to the U.S. and Ca-
nadian Censuses. Within each country of destination, five immigrant samples
will be analyzed: the pooled sample and the subsamples of immigrants origi-
nating in Africa, Asia, Europe, and Latin America. The regression will con-
tain a vector of demographic characteristics including education, marital
status, health status (when available), and metropolitan residence. These re-
gressions are used to calculate the present value differential between immi-
grants and demographically comparable natives for each of the cohorts. These
present value differentials are presented in table 1.13. (The data presented in
table 1.13 for Australia will be discussed in detail below.)
67 Immigration and Self-Selection
Table 1.12 Earnings Differentials between Immigrants and Comparable Natives
in 1980-81 Cross Sections
Immigrant Cohort
Origin and
Destination 1975-80 1970-74 1965-69 1960-64 1950-59 < 1950
All immigrants in:
United States - .3460 - ,1534 - .0676 - ,0239 ,0177 ,0045
(-14.48) (-.10.42) (-6.91) (-2.58) (1.79) (.39)
Canada -.2271 -.1118 -.0286 -.0571 -.0020 ,0558
(-9.52) (-6.61) (-2.35) (-3.99) (-.22) (2.78)
Australia -.Of310 -.0642 -.0814 -.0656 -.0796 -.0342
(-2.51) (-2.87) (-4.98) (-4.05) (-6.06) (-1.82)
Note: The r-ratios are presented in parentheses.
Consider initially the pooled sample of immigrants. Table 1.13 documents
the systematic decline in the quality of immigrants arriving in the United
States over the last two decades. For instance, the typical immigrant arriving
in 1960-64 in the United States had only a slight wage disadvantage relative
to a comparable native, while the typical immigrant arriving in the United
States in 1975-79 has a wage disadvantage of nearly 27 percent over the life
cycle as compared to the native baseline. Remarkably, the Canadian Censuses
reveal very similar patterns: the 1960-64 migrant to Canada had a 6 percent
wage disadvantage over the life cycle (relative to natives), while the disadvan-
tage for the most recent migrants (1975-80) has increased to nearly 23 per-
cent.
The American and Canadian trends are less similar when the analysis is
restricted to men from a specific country of origin. For example, among Eu-
ropean immigrants, the U.S. Census reveals a substantial decline in quality
(from a 4 percent advantage to an 11 percent disadvantage) over the last
twenty years, while the Canadian Census reveals a roughly stable wage differ-
ential between immigrants and natives over the post- 1960 cohorts. Similarly,
among Asian immigrants, the Canadian data reveal that the 1960-64 and the
1975-80 cohorts had essentially the same relative standing, while the U.S.
data reveal a decline in quality from a 15 to a 27 percent disadvantage. These
results, therefore, imply that at least part of the similarity between the United
States and Canada at the aggregate level is due to the fact that the national
origin composition of the cohorts shifted over time, away from European im-
migrants (who tend to do quite well in the labor market) to Asian and Latin
American immigrants (who do much worse in the labor market).
As noted earlier, the Australian Census is available only for 198 1. Since
cohort and aging effects cannot be identified, the present value differentials
cannot be calculated directly. Recall, however, that the 198 1 cross-sectional
regressions estimated in the Australian data showed that immigrants in
68 George J. Borjas
nbl e 1.13 Present Value Differentials between Immigrants and
Comparable Natives
Year of Arrival
Group 1975-80 1970-74 1965-69 1960-64 1950-59 < 1950
All immi-
grants in:
States ( - 18.99)
Canada - .2297
Australia .0149
(.46)
United - ,2656
( - 13.25)
African immi-
grants in:
United - ,3779
States (-5.11)
Canada -.4092
(-3.00)
Australia - ,1688
(- 1.01)
Asian immi-
grants in:
States ( - 11.47)
Canada -.3930
United - ,2692
(-6.88)
(- .84)
Australia - .0634
European
immigrants in:
United - .lo68
States ( - 6.06)
Canada -.0516
( - 2.22)
Australia .0745
(1.68)
Latin American
immigrants in:
States ( - 14.62)
Canada -.3312
( - 3.77)
Australia ,1671
(.61)
United - ,2716
-.1228 -.0827 -.0453 -.0260 -.0451
(-12.20) (-10.40) (-6.88) (-4.37) (-4.38)
-.1306 -.0449 -.0632 -.0344 ,0212
(-8.57) (-3.75) (-4.63) (-3.57) (1.10)
,0136 -.0570 -.0740 -.1330 -.IN14
(.61) (-3.49) (-4.57) (-10.12 ) (-4.86)
-.3097 -.1425 -.1577 -.I997 - . I806
(-6.08) (-3.21) (-3.62) (-4.28) (-1.69)
-.4555 -.2690 -.3297 -.2595 .2108
(-3.23) (-2.03) (-2.55) (-2.65) (.61)
-.2197 -.I191 -.I317 -.3413 -.4481
(-1.90) (-1.42) (-1.26) (-.88) (-3.26)
-.4117 -.1565 -.I495 -.2551 -.2487
( - 8.33) ( - 10.53) ( - 9.89) (- 17.54) (-9.08)
- .3658 - ,2534 - .3651 - .3868 .0637
(-6.56) (-4.86) (-6.38) (-10.19) (34)
.0022 -.2348 -.2367 -.3817 ,0141
(.04) (-4.75) (-3.63) (-7.42) (.20)
- ,0167 ,0218 ,0436 ,0307 ,0219
( - 1.25) (2.14) (5.07) (4.44) (1.79)
,0113 .0022 -.0290 .0116 .0423
( 5 5 ) (.14) (-1.92) 1. 04) (2.04)
,0350 -.0524 -.0732 -.1121 -.0833
(1.33) (-2.87) (-4.26) (-8.15) (-4.18)
-.1273 -.1243 -.0841 -.1282 -.1629
(-9.53) (-11.42) (-8.91) (-13.56) (-8.18)
- ,2820 - ,1693 - ,1230 - ,1757 ,1788
(-3.25) (-2.10) (-1.46) (-3.07) (.91)
- ,0677 - .3991 - ,2721 ,0827 -.2868
(-.38) (-2.45) (-1.15) (.15) (-.70)
Note: The r-ratios are presented in parentheses.
69 Immigration and Self-Selection
Australia face significantly different age-earnings profiles than their counter-
parts in the United States and Canada. In particular, in the cross section, there
seems to be little relation between the earnings of immigrants in Australia and
the length of residence in Australia. If there is any assimilation effect in Aus-
tralia, therefore, this result must imply that the quality of immigrants to Aus-
tralia has increased over the last two or three decades.
A rough estimate of this increase can be obtained if it is assumed that the
unobserved assimilation effect experienced by immigrants in Australia re-
sembles the assimilation effect of similar persons (i.e., persons from the same
country of origin) in Canada or the United States. Given this approximation,
the assimilation effects can then be subtracted from the Australian cross-
sectional coefficients (thus netting out the role played by pure aging in the
generation of the cross-sectional results), and the present value differentials
can be computed for each of the cohorts. Since there are two sets of assimila-
tion parameters (one for Canada and one for the United States), a number of
different approximations can be calculated. In general, these experiments led
to similar qualitative findings. In this paper, therefore, the assimilation rate
used is the average of the two assimilation rates (i.e., the U.S. and Canada
aging effects) experienced by immigrants from the same continent of origin.
Given these assimilation rates and the cross-sectional regressions estimated
in the Australian Census for each region of origin, it is a simple matter to
calculate the predicted present value differential between the various cohorts
of immigrants and comparable natives in Australia. These predictions are also
presented in table 1.13. Two substantive results are worth noting. As implied
by the flat earnings profiles found in the (pooled) Australian cross section, the
quality of immigrants to Australia increased slightly over the last twenty to
thirty years. The typical immigrant in 1960-64 could expect a 7 percent wage
disadvantage over his life cycle, while the typical immigrant in 1975-80 has
no wage disadvantage relative to natives over his life cycle. Second, this in-
crease in immigrant quality can essentially be found in every one of the na-
tional origin groups under analysis. For example, the typical European immi-
grant in the early 1960s had a 7 percent wage disadvantage, while the typical
European immigrant in the late 1970s has a 7 percent wage advantage over
natives. Similarly, the average Asian immigrant in the early 1960s had 24
percent lower earnings over his life cycle than natives, while the differential is
only 6 percent (and insignificant) for the most recent migrants.
The data presented in table 1.13 provide a unique descriptive analysis of an
important question. Which host countries are the “winners” and which are the
“losers” in the immigration market? This comparison, of course, depends on
the assumption that the native base across countries has a similar level of pro-
ductivity and skills. This assumption makes the relative wage of immigrants
across host countries directly comparable as an index of immigrant quality.
The assumption that natives among the three host countries are roughly simi-
70 George J. Borjas
lar is not empirically verifiable. However, it does not seem unreasonable since
all three countries share a common language and culture, have similar politi-
cal and economic systems, and are at similar stages of economic development.
Given this assumption, the statistics presented in table 1.13 present an in-
teresting story of the extent of self-selection in the generation of the foreign-
born population in each of the countries. Consider the trends for the pooled
sample. During the 1940s and 1950s, Australia was attracting immigrants
who had lower productivities than the immigrants attracted by Canada and the
United States. This type of selection, however, was drastically reversed during
the 1960s, as both Canada and the United States began to attract persons who
did not perform as well in the labor market and Australia began to attract
immigrants with relatively high levels of unobserved skills.
1.4.5 Determinants of Immigrant Quality
Consider the following regression model:
(35)
where Q,,(t) is the present value differential between immigrants and compa-
rable natives of a cohort migrating from country i to country j at time t , X, ( t )
is a vector of variables describing conditions in the country of origin i at time
t , and XJ ( t ) is a vector of variables describing conditions in the country of
destinationj at time t .
The specification of (35) builds in a very important (and restrictive) as-
sumption. In particular, the relative earnings of a person from country i in
country j at time t are independent of events in other periods t’ ( t # t ’ ) , and,
more important, they are also independent of conditions in other countries
(particularly, they are independent of conditions in other potential countries of
destination). This empirical framework, in a sense, introduces an “indepen-
dence of irrelevant alternatives” assumption into the study. Although this as-
sumption is not likely to be strictly satisfied, it does simplify the empirical
analysis greatly. If the assumption was invalid, for instance, the right-hand
side of (35) would have to be expanded to include the characteristics of all
other potential countries of destination, and the increase in the number of
variables would rapidly drive the number of degrees of freedom to zero.
Table 1.14 presents the estimates of the reduced-form equation in (35). The
sample consists of 48 observations (4 continents of origin times 4 post-1960
cohorts times 3 countries of destination). The regression in table 1.14 reveals
that a small number of characteristics of the countries of origin and the coun-
tries of destination do “explain” a very large fraction of the variance in the
unobserved quality of immigrants. The variables in the reduced-form equa-
tion, for example, explain over 80 percent of the variance in the quality mea-
sures presented in table l . 13. Despite this success, however, it must be noted
that, because the countries of origin are defined in terms of continents, the
71 Immigration and Self-Selection
Table 1.14 Determinants of Immigrant Quality across Host Countries
Variable Coefficient t
CONSTANT . 1 2 5 2 (-2. 77)
USLAW - . 0 5 1 1 ( - 1. 79)
UNEMPLOYMENT . 0 0 1 1
INEQUALITY( ~) - ,0044 ( - 1 . 8 9 )
INEQUALITY( 1) , 0 4 3 1 ( 4 . 3 5 )
AGNP .0903 ( 8. 78)
RZ ,801
Note: Key to additional variables: UNEMPLOYMENT = unemployment rate in the host country at
the time of migration; INEQUALITY(O) = average income inequality (as defined in table 1. 4) in
selected countries from continent of origin in decade of migration; INEQUALITY( ~) = inequality
measure for destination countries in decade of migration; AGNP = difference in (1n)GNP per
capita between sending and host countries at time of migration.
two variables measuring country-of-origin characteristics (the relative GNP
level and the extent of income inequality) are, in effect, averaged over a large
and diverse number of countries.” It is unclear what biases are caused by this
aggregation, but it is important to remember that the coefficients in table 1.14
are, at best, suggestive of the underlying economic behavior.
Both the GNP of the continent of origin (relative to GNP per capita in the
country of destination) and the inequality measure for the continent of origin
affect the quality of migrants significantly. Migrants from wealthier regions do
better no matter where they go, and migrants from regions with large levels of
income inequality do worse than other migrants. Similarly, the inequality
measure for the country of destination has a positive and significant effect on
relative immigrant earnings, as predicted by the Roy model. Finally, the
change in U. S. immigration policy (as measured by USLAW) has a negative
and marginally significant effect and thus helps identify the effect of this major
change in policy relative to other countries. The change in U.S. immigration
policy lowered the earnings of migrants by 5 percent relative to the earnings
of migrants who chose other countries of destination.
1.4.6 The Point System and Immigrant Quality
It is somewhat surprising that the cohort quality trends in Canada and the
United States are so similar despite the major differences in immigration poli-
cies between the two countries. Immigration policies, however, can screen
applicants only on the basis of observed demographic characteristics such as
education, occupation, and age. The results summarized in table 1.13 show
that even a stringent point system, such as that used by Canada, was unable to
prevent a decline in immigrant skills in that country similar to that experi-
enced in the United States.
On the other hand, however, the point system clearly affects the observable
skills of the incoming immigrant flow. A clear way of ascertaining the impor-
72 George J. Borjas
tance of this direct effect is to reestimate the regression model on the pooled
sample of immigrants without controlling for any demographic characteris-
tics. The unstandardized differentials in the present value of lifetime earnings
are presented in table 1.15. It is evident that, during the 1970s, immigrants
admitted to the United States were substantially less skilled (relative to U.S.
natives) than immigrants admitted to other host countries (relative to the na-
tives of those countries).
It is also evident that the point system generated an immigrant flow into
Canada that had relatively favorable socioeconomic characteristics. For in-
stance, even though the most recent immigrants into Canada (the post-1975
arrivals) earned 23 percent less than comparable Canadian natives, they
earned only 12.2 percent less than the typical Canadian native. These immi-
grants, therefore, have more favorable demographic characteristics than Ca-
nadian natives. Although the point system was unable to prevent the decline
in immigrant skills, it greatly tempered the extent of the drop.
This fact has interesting implications. If a host country decides that it
wishes to attract more skilled immigrants, a point system seems to be a very
direct and simple way of doing so. At the same time, however, the point sys-
tem has the major limitation that it cannot screen for unobservables, and these
unobservables are major determinants of individual earnings. Hence, as long
as economic and political conditions motivate relatively unskilled persons to
emigrate, the point system may restrict entry to only those who pass the test,
but the immigrant pool will be composed mostly of relatively unskilled per-
sons with the acceptable demographic characteristics.
Table 1.15 Standardized and Unstandardized Present Value Differentials in Host
Countries
All Immigrants in:
United States Canada Australia
Year of -
Arrival Standardized Unstandardized Standardized Unstandardized Standardized Unstandardized
1975-79 - ,2656
( - 18.99)
1970-74 - ,1228
( - 12.20)
1965-69 - ,0827
( - 10.40)
(-6.88)
1950-59 - ,0260
(-4.37)
1960-64 -.0453
< 1950 -.0451
(-4.38)
- ,3706 - ,2297
(-23.91) ( - 13.25)
- ,2443 - ,1306
(-21.94) (-8.57)
- .1710 - ,0449
( - 19.52) (-3.75)
- ,1086 - ,0632
(-15.18) (-4.63)
- ,0712 - .0344
(-11.12) (-3.57)
- ,0562 .0212
(-5.05) (1.10)
-.I222 ,0149
( - 6.65) (.46)
- .0271 ,0136
(-1.67) (.61)
,0568 - ,0570
(4.47) ( - 3.49)
- ,0079 - ,0740
(S4) (-4.57)
,0094 - ,1330
(.91) (-10.12)
,0286 - ,0914
(1.40) (-4.86)
- ,0020
( - .23)
- .0377
( - .41)
- ,0976
( - 2.10)
- ,0856
( - 1.96)
-.1803
(-5.56)
- ,1997
( - 7.21)
Nore: The r-ratios are presented in parentheses.
73 Immigration and Self-Selection
1.5. Summary
Self-selection plays a dominant role in immigration (as it does in all other
forms of turnover). There is selection in the determination of the composition
of the persons who leave any given country, in terms of both observable char-
acteristics (such as education) and unobservable characteristics (such as abili-
ties and productivities). In addition, this nonrandom sample is then sorted
across various possible host countries in a nonrandom way. Hence, the pool
of immigrants in any host country is, in a sense, doubly self-selected: the pool
of immigrants in the host country is composed of persons who found it prof-
itable to leave the country of origin and who did not find it profitable to go
anywhere else.
This paper attempts to use the economic theory of self-selection as a guide
to understanding how immigrants perform in the labor market. The assump-
tion of wealth-maximizing behavior provides important insights into the me-
chanics that guide the selection process.
The empirical analysis studied the role played by self-selection in the earn-
ings of immigrants in the United States, and compared these migrants to the
pool of migrants who chose to reside in other countries (Australia or Canada).
The study of the various Censuses revealed that the United States, as a result
of major changes in immigration policy, began to attract relatively less skilled
persons in the 1970s. In a sense, the United States became less competitive in
the international marketplace that determines the migration decision and the
sorting of migrants across host countries.
Notes
1. A recent survey of this literature is given in Greenwood and McDowell(l986).
2. Jasso and Rosenzweig (1985) also stress this important technical point in their
work.
3. A fourth case where Q, > 0 and Q, < 0 is theoretically impossible since it re-
quires p > 1.
4. Data on international differences in income inequality are published by the World
Bank (1986). These data, however, do not correspond directly to the variances that lie
at the heart of the Roy model. In particular, a; and a: measure the dispersion in “op-
portunities” (for given X) rather than the variance in incomes across households in a
given country.
5. There is a slight technical problem that must be taken into account in the deriva-
tion of this result. An increase in crl “stretches” the income distribution of the United
States and will lead to a different mean wage level in the pool of migrants even if this
pool is restricted to include the same persons. A simple solution to this problem is to
define quality in terms of “standardized units,” or Q,/a,. The prediction in the text can
then be easily derived.
6. Because the year of immigration is not precisely available in the U.S. Census, it
is approximated as the midpoint of the available intervals. In the 1980 Census, y takes
74 George J. Borjas
on a value of 2.5 if the immigrant arrived in 1975-79, 8 if arrived in 1970-74, 13 if
arrived in 1965-69, 18 if arrived in 1960-64, 25.5 if arrived in 1950-59, and 45 if
arrived prior to 1950. In the 1970 Census, y is 2.5 if arrived in 1965-69, 8 if arrived
in 1960-64, 15.3 if arrived in 1950-59, and 39.8 if arrived before 1950. The esti-
mated y for these last two intervals are calculated using the more precise year of migra-
tion data available in the 1970 Census.
7. There is an implicit assumption in (25) that is directly responsible for this simple
framework. In particular, growth rates for immigrants are independent of the year of
migration 0. The model can be generalized to allow for these types of interactions.
However, the estimating equations would include higher order polynomials, and the
estimation of the underlying structural parameter may become quite sensitive to the
very high correlation among the right-hand-side variables.
8. The construction of the data sets is described in detail in Bojas (1987).
9. The enrollment data are available in Unesco (1969, 1980). Enrollments are avail-
able for each “level” of education. The data sources also give the number of years of
education associated with that “level” for each country. The means presented in table
1.2 are calculated using both these statistics.
10. It is important to note that many of these differences in quality across cohorts
from a given country of origin are statistically significant at conventional levels. For
some evidence on this point, see Borjas (1987).
11. Since the dependent variable in the “second-stage” regressions is a linear func-
tion of regression coefficients, the regressions are weighted to account for heteroske-
dasticity. For details, see Borjas (1987).
12. These statistics are available in the United Nations (1982, 44). The calculations
ignore the large (and presumably) temporary flows from Ethiopia to Somalia in the late
1970s as well as the movement of guest workers to oil-producing countries in the
Middle East.
13. A number of previous studies (e.g., Tandon 1978; Chiswick and Miller 1985;
and Chiswick 1988) analyze the labor market performance of immigrants in Australia
and Canada. These studies, however, do not study the nonrandom sorting of migrants
across host countries.
14. This section is based on the excellent descriptions and summaries of immigra-
tion policies given by Boyd (1976), Keely (1979), and Keely and Elwell (1981), Kubat
(1979), and Price (1979).
15. Throughout this section, the native base in each of the host countries is the entire
population of native persons (regardless of ethnic or racial origin). This differs from
the native baselines chosen in the previous section but makes the comparisons among
host countries less arbitrary.
16. There are some differences in the calendar years bracketed by these dummy
variables across the countries of destination. The brackets reported in the table are
those that apply to U.S. data. The Canadian and Australian brackets are quite similar
for post-1960 migrants but differ for pre- 1960 migrants.
17. The average was calculated over the two or three source countries that formed
the bulk of immigration from that continent to the particular host country.
References
Australian Department of Immigration and Ethnic Affairs. 1982. Australian Immigra-
tion, Consolidated Statistics no. 12. Canberra: Australian Government Publishing
Service.
75 Immigration and Self-Selection
Birrell, R. 1984. A New Era in Australian Migration Policy. International Migration
Review 18:65-84.
Borjas, George J. 1985. Assimilation, Changes in Cohort Quality, and the Earnings of
Immigrants. Journal of Labor Economics 3:463-89.
. 1987. Self-Selection and the Earnings of Immigrants. American Economic
Review 77531-53.
Boyd, Monica. 1976. Immigration Policies and Trends: A Comparison of Canada and
the United States. Demography 18533-104.
Carliner, Geoffrey. 1980. Wages, Earnings, and Hours of First, Second, and Third
Generation American Males. Economic Inquiry 18:87-102.
Chiswick, Barry R. 1978. The Effect of Americanization on the Earnings of Foreign-
born Men. Journal of Political Economy 86:897-921.
. 1988. Immigration Policy, Source Countries, and Immigrant Skills: Austra-
lia, Canada, and the United States. In The Economics of Immigration, ed. Lyle
Baker and Paul Miller, 163-206. Canberra: Australian Government Printing Ser-
vice.
Chiswick, Barry R., and Paul W. Miller. 1985. Immigrant Generation and Income in
Australia. Economic Record 61 540-53.
Douglas, Paul H. 1919. Is the New Immigration More Unskilled than the Old? Journal
of the American Statistical Association 16:393-403.
Greenwood, Michael J. 1975. Research on Internal Migration in the United States: A
Survey. Journal of Economic Literature 13:397-433.
Greenwood, Michael J., and John M. McDowell. 1986. The Factor Market Conse-
quences of U.S. Immigration. Journal of Economic Literature 24: 1738-72.
Heckman, James J., and Richard Robb. 1983. Using Longitudinal Data to Estimate
Age, Period, and Cohort Effects in Earnings Equations. In Analyzing Longitudinal
Data for Age, Period, and Cohort Effects, ed. H. Winsborough and 0. Duncan,
173-50. New York: Academic.
Hicks, John R. 1932. The Theory of Wages. London: Macmillan.
Jackson, J. A. 1969. Migration. Cambridge: Cambridge University Press.
Jasso, Guillermina, and Mark R. Rosenzweig. 1985. How Well Do U.S. Immigrants
Do? Vintage Effects, Emigration Selectivity, and the Occupational Mobility of Im-
migrants. University of Minnesota. Mimeo.
Keely, Charles B. 1979. The United States of America. In The Politics of Migration
Policies, ed. D. Kubat, 51-64. New York: Center for Migration Studies.
Keely, Charles B., and Patricia J. Elwell. 1981. International Migration: Canada and
the United States. In Global Trends in Migration, ed. M. Kritz, C. Keely, and S.
Tomasi, 181-207. New York: Center for Migration Studies.
Kubat, Daniel. 1979. Canada. In The Politics ofMigration Policies, ed. D. Kubat, 19-
36. New York: Center for Migration Studies.
Price, Charles. 1979. Australia. In The Politics of Migration Policies, ed. D. Kubat,
3-18. New York: Center for Migration Studies.
Psacharopoulos, George. 1973. Returns to Education: An International Comparison.
Amsterdam: Elsevier.
Roy, A. D. 1951. Some Thoughts on the Distribution of Earnings. Oxford Economic
Papers 3:135-46.
Schwartz, Aba. 1968. Migration and Lifespan Earnings in the U.S. Ph.D. diss., Uni-
versity of Chicago.
Sjaastad, Larry A. 1962. The Costs and Returns of Human Migration. Journal of Po-
litical Economy 70:80-93.
Tandon, B. B. 1978. Earnings Differentials among Native Born and Foreign Born
Residents of Canada. International Migration Review 12906-10.
Unesco. 1969. Statistical Yearbook. 1968. Paris: United Nations.
76 George J. Borjas
. 1980. Statistical Yearbook, 1978-1979. Paris: United Nations.
United Nations. 1982. Demographic Indicators of Countries. New York: United Na-
tions.
U.S. Department of Commerce. Various Issues. Statistical Abstract of the United
States. Washington, D.C.: U.S. Government Printing Office.
U.S. Immigration and Naturalization Service. Various Issues. Statistical Yearbook of
the Immigration and Naturalization Service. Washington, D.C.: U.S. Government
Printing Office.
World Bank. 1986. World Development Report. New York: Oxford University Press.
2 Undocumented Mexican-born
Workers in the United States:
How Many, How Permanent?
George J. Borjas, Richard B. Freeman, and Kevin Lang
Few issues in the area of immigration to the United States generate as much
concern and confusion as the influx of illegal aliens. Estimates of the number
of illegal immigrants vary widely. Some observers, noting the explosive
growth of Border Patrol apprehensions of aliens to over a million a year, have
suggested that the country has harbored five to ten million or more undocu-
mented residents.’ Others, relying on 1980 Census of Population and related
demographic data, put the numbers on the order of two to three million in that
year2 Some think that illegal aliens are largely transient agricultural workers,
slipping across the Mexican border for seasonal work. Others stress the per-
manence of many illegal aliens, who have sufficiently long stays in the United
States to be eligible under the 1986 Immigration Reform and Control Act to
attain legal resident status. In light of the difficulties in analyzing illegal im-
migration, one recent reviewer has written that the issue is “inaccessible to
accurate measurement” with no “firm evidence” ever likely to become avail-
able (Teitelbaum 1986, 153).
In this paper, we take a more positive approach, analyzing three govern-
ment data sets and a small survey of illegal aliens in the San Diego area in an
effort to evaluate conflicting claims about the illegal Mexican-born migrant
population.
Among the government data sets we analyze is, first, the 1980 U.S. Census
George J. Borjas is professor of economics at the University of California, San Diego, and a
research associate of the National Bureau of Economic Research. Richard B. Freeman is professor
of economics at Harvard University and director of the Labor Studies program of the National
Bureau of Economic Research. Kevin Lang is professor of economics at Boston University and a
faculty research fellow of the National Bureau of Economic Research.
The authors wish to thank Betty Smith of Statistical Resources at Public Health Service and
Don Neeler, Rod Field, and Blanke Shanks of the Immigration and Naturalization Service for
valuable assistance in providing data. For research assistance, they are grateful to Rachel Fried-
berg and Steven Rader.
77
78 G. J. BorjadR. B. FreemadK. Lang
of Population figures on the number of Mexican-born residents of the United
States. As pointed out by Warren and Passel (1987), the Mexican-born popu-
lation in the 1980 Census exceeds by over one million the number of legal
immigrants expected in the Census on the basis of Immigration and Naturali-
zation Service (INS) records. We build on the Census analysis by tabulating
the family composition and economic characteristics of the nonnaturalized
Mexican born most likely to be illegal. Second, we analyze vital statistics on
the number of deaths of Mexican-born persons and of births to Mexican-born
mothers in the United States. Under specified assumptions, these data provide
us with independent estimates of the size and growth of the illegal Mexican
population. Third, we analyze INS figures on apprehensions of illegal aliens.
We relate these figures to Border Patrol expenditures and measures of eco-
nomic incentives to migrate and examine the seasonality of apprehensions to
make inferences about the growth and nature of border crossings. In addition,
we examine the consistency between the level of apprehensions and Census/
Vital Statistics-based estimates of the size of the illegal Mexican population.
Our original data consist of a survey of 289 illegal Mexican male aliens in
fourteen locations around San Diego obtained during the summer of 1986. We
use these data to estimate time spent in the United States and apprehensions
by the Border Patrol per crossing as well as to determine the characteristics of
this part of the illegal alien population.
Our principal findings are as follows.
1. The number of illegal Mexican aliens in the United States in 1980 was
on the order of 1.8 million. In addition to the approximately 1.1 million ille-
gal Mexican immigrants counted in the Census, there were perhaps 600,000-
700,000 illegal Mexican immigrants not counted in the Census, giving a total
illegal Mexican migrant population of less than two million. Births to
Mexican-born women and deaths of Mexican-born persons suggest that the
illegal population may have increased to 2.0-2.3 million by 1984.
2. The explosive growth of Border Patrol apprehensions is due in substan-
tial part to increased Border Patrol activity, making the growth of apprehen-
sions an upwardly biased estimate of the growth of illegal border crossings
and a poor indicator of the growth of the illegal Mexican migrant population.
Moreover, the seeming inconsistency between the 750,000-1,000,000 plus
annual apprehensions and the estimate of fewer than two million illegals in
1980 appears to be due partly to short-term migrants who cross the border and
generate apprehensions frequently but whose spells in the United States con-
tribute less than a person-year to the stock of illegals.
3. The bulk of illegal Mexican aliens in the Census of Population live with
their families, are engaged in work activities beyond agricultural labor, and
have other characteristics suggesting that they are relatively permanent resi-
dents in the country. The earnings of these persons are considerably below
those of other Mexican-born residents, but measures of work experience are
similar.
79 Undocumented Mexican-born Workers in the U.S.
The difference between the relatively permanent illegal immigrants who
seem to constitute the bulk of those counted in the Census and the more itin-
erant who cross the border frequently (and are covered in our survey in San
Diego) may reflect a life-cycle change to illegal alien migration, as new young
migrants come first without their families, go back often at holidays or at
breaks in seasonal jobs, but later bring their families to the United States and
move toward permanent residence. On the other hand, it may also reflect a
substantial difference between permanent immigrant and sojourner popula-
tions.
2.1 Estimating the Size and Growth of the Illegal Mexican Population
One of the more surprising facts about the 1980 Census of Population doc-
umented by Warren and Passel (1987) is that the Census counted a sizable
number of illegal Mexican-born migrants (see table 2.1). In part, this results
from the fact that, “for the 1980 Census, the Bureau of the Census made extra
efforts to count difficult-to-enumerate groups such as undocumented aliens”
and had sufficient success that the Bureau’s evaluation of population estimates
turned up “a large national error of closure between the 1970 and 1980 Cen-
suses” (U.S. Bureau of the Census, Population Growth and Distribution, 3,
7). Our tabulations of the 1980 Census in table 2.1 show that 2.18 million
Mexican-born persons (exclusive of native Americans born in Mexico) were
counted, of whom 1.67 million reported that they were not citizens as of the
date of the survey. According to Warren and Passel (1987), who used INS
Table 2.1 Size of the Mexican-born Population in the 1980 Census
of Population
Group N %
Reporting as born in Mexico:
1. Total
2. Naturalized citizens
3. As percentage of total
4. Not a citizen
5. As percentage of total
Illegal immigrants from Mexico:
6. Estimated number
7. As percentage of total
8. As percentage of noncitizens
9. As percentage of noncitizens
arriving after 1970
2,182,900 100.0
23.3
509,400
1,673,500
16.1
1,130,OOO
51.8
67.5
103.6
Sources: Lines 1-5 tabulated from the Public Use Sample of the U.S. Census of Population, the
A Sample (5%). Line 6 from Warren and Passel (1987).
80 G. J. Borjas/R. B. FreemadK. Lang
figures on legal alien entry to the United States (modified in various ways) to
estimate the number of legal migrants and, as a residual, the number of ille-
gals, the 1980 Census counted 1.13 million illegal Mexican-born aliens. This
is more than half the total reported (line 1) and two-thirds of those who were
not citizens (line 4), implying that the Census data can be used to make infer-
ences about the characteristics of a large number of illegal Mexican-born
aliens. Going a step further, note that the figure of 1.09 million Mexican-born
noncitizens in the Census who came after 1970 is approximately equal to War-
ren and Passel’s estimated number of illegal aliens. Given that relatively few
illegal Mexican aliens are likely to have come to the United States prior to
1970 and that many legal immigrants are likely to have become citizens, one
can reasonably treat Mexican-born noncitizens who came after 1970 as a pop-
ulation dominated by likely illegals, as we do in section 2.3.
These conclusions are, of course, affected by the fact that the counts pro-
vided by the 1980 U.S. Census of Mexican nationals and of naturalized Mex-
icans are measured with error. It is well known, for instance, that naturaliza-
tion rates calculated from individual responses to the Census questionnaire
greatly overstate the naturalization rates recorded by the official INS docu-
ments. The careful study of the 1980 Census data by Warren and Passel, how-
ever, attempts to correct the counts for errors in misreporting both of country
of birth and of citizenship status. Using these corrected population counts in
our analysis does not alter the qualitative nature of any of our conclusions
and does not greatly affect the order of magnitude of the statistics reported in
table 2.1.
How many illegal Mexican migrants might be missed in the Census
count-one million, two million, ten million?
To provide an answer to this important question, we make use of two pieces
of data from the Vital Statistics of the United States: the number of deaths of
persons of Mexican birth and numbers of births to Mexican-born women.
Assuming, as seems reasonable, that mortality and birth data are more com-
plete than the Census count of the population, we expect to find more deaths/
births for the Mexican born than the counted population could plausibly gen-
erate.3 Given assumptions about true death and birth rates, the “excess” deaths
or births will yield estimates of Census undercounts and thus of the true pop-
ulation. Algebraically, the structure of our analysis can be most simply seen
in the following accounting equation:
(1)
R, = r,(POP,) +r; (HPOP,),
where R, = number of events, deaths or births, in Vital Statistics, r, = the
true rate of occurrence of events to the measured population, I.: = the rate of
Occurrence of events to the hidden population, POPt = measured population,
HPOP, = hidden population, and i indexes an agehex group. Then HPOP, =
(R, - r,POPt)/r: provides an estimate of the undocumented population of
age i .
81 Undocumented Mexican-born Workers in the U. S .
The accuracy of estimates of the hidden or undocumented population based
on ( 1) depends on two factors: the extent to which funeral directors and hos-
pitals accurately record country of origin on death and birth forms and the
accuracy of the postulated “true” death or birth rates for the measured popu-
lation and the hidden population. While there may be some tendency for
friends and relatives of illegal aliens to disguise country of origin, the head of
Registration of Methods of the Public Health Service informed us that in his
opinion these records are no less accurate for illegal aliens than for other
groups, leading us to discount this potential source of The problem,
then, is to estimate “true” birth or death rates for the undocumented popula-
tion. Our approach is to assume that the rates for the undocumented are the
same as those for the population as a whole (death rates) or for the docu-
mented migrant population (birth rates). This approach yields the following
formula for estimating the hidden population:
(2)
HPOP; =R, / rj - POP,.
If, as seems reasonable, Mexican-born immigrants have higher death rates
than native Americans (rr > r j ) , estimates of the uncounted population based
on national death rates will provide an upper bound to the population. Simi-
larly, if undocumented Mexican-born women have higher birth rates than doc-
umented immigrants, estimates of the hidden population based on national
birth rates will also provide an upper bound to the number of undocumented
Mexican-born female migrants .5
2.1.1
Since 1979, Viral Statistics has published data on the number of deaths of
the Mexican-born persons in the United States. Beginning in 1984, data are
also available by age, allowing us to use equation ( 1) above to estimate the
likely number of such persons by age. Columns 1-2 of table 2.2 record the
basic data for this analysis: the number of deaths to Mexican-born persons in
the United States by age (col. 1) and the 1984 mortality rates for all Ameri-
cans by age (col. 2). Assuming that the U.S. mortality rate represents the true
death rate for the Mexican born, we obtain the estimated 1984 population of
Mexican-born persons by age in column 3. Summing down the column gives
an estimated total population of 2.97 million persons. As we expect the mor-
tality of the Mexican born to be higher than that of native Americans, this is
likely to be an upper bound to the true number.
To use our 1984 estimates to obtain estimates for 1980 when the Census
was conducted, we must adjust them for potential changes in the Mexican-
born population from 1980 to 1984. According to Vital Statistics, the number
of deaths of Mexican-born persons rose from 13,180 in 1980 to 14,050 in
1984, suggesting that the population may have been 6.6% higher in 1984 than
in 1980. By this calculation, the Mexican-born population in 1980 was about
2.8 million persons, which implies that the Census missed 600,000-700,000
Mortality of the Mexican Born
82 G. J. Borjask. B. Freeman/K. Lang
Table 2.2 Mortality and Undocumented Mexican-born Population Estimated
from Deaths of the Mexican Born
1984 U.S. 1984 “Uncounted”
Age 1984
Mortality Rate/ Mexican-born 1980 Mexican
Group Deaths (1) 100,000 (2) Population (3) Population (4) Born ( 5 )
< 1
1-4
5-14
15-24
25-34
3 5 4
45-54
5544
65-74
75-84
85 +
Total
16
39
70
922
1,093
742
819
1,252
2,308
4,150
2,627
14,038
1,085.6
51.9
26.7
96.8
121.1
204.8
521.1
1,287.8
2,848.1
6,399.3
15,233.6
1,500
75,100
262,000
952,000
902,600
362,000
157,000
97,200
8 1,000
64,900
17,300
2,972,600
7,800
52,300
271,000
516,000
551,800
310,000
186,000
126,000
94,900
53,600
13,900
2,183,000
- 6,300
22,800
-9,000
436,000
350,800
52,000
- 29,000
- 28,800
- 13,900
11,300
3,400
789,600
~~
Sources: Columns I and 2 from Vital Statistics of the United States. Column 3 = (column 1)/
(column 2). Column 4 tabulated from 1980 U.S. Census of Population tapes. Column 5 =
column 3 - column 4.
persons-roughly 25% of the Mexican-born population. Adding the
600,000-700,000 undocumented persons to Warren and Passel’s estimated
1.1 million in the Census yields a total illegal Mexican population in 1980 on
the order of 1.8 million persons.
As a check on the plausibility of our calculations, we record in column 4 of
table 2.2 the number of Mexican-born persons by age in the 1980 Census and
in column 5 the difference between these numbers and the 1984 numbers im-
plied by the mortality data. The calculations show the biggest divergence to
be among 15- to 24-year-olds and 25- to 34-year-olds, which seems plausible
in terms of the likely age distribution of transient illegal aliens. The implica-
tion is that the mortality data are, indeed, giving us a reasonable handle on the
order of magnitude of the missing population.
2.1.2 Births
The number of births to Mexican-born women provides another source of
information on the potential size of the population not counted in the Census.
To estimate the size of the undocumented Mexican-born female population,
we return to equation (1): as our measure of the number of events R, we take
the number of births reported for Mexican-born women by age from Vital
Statistics (col. 1 of table 2.3); as our measure of the true event rate, we take
birth rates for Mexican-born women estimated by Bachu and O’Connell
(1984) from the April 1983 Current Population Survey (col. 2 of table 2.3).6
We then divide the number of births by the birth rates to estimate the Mexican
immigrant female population in 1980 (col. 3). For the group aged 18-39,
83 Undocumented Mexican-born Workers in the U.S.
Table 2.3 Births and Undocumented Mexican-born Female Population
Estimated from Births to Mexican-born Women
(1) 1980 Births to (2) 1983
Mexican-born Birth Rates (3) Estimated (4) 1980 Census (5) Hidden Female
Age Women per 1 ,OOO Population Population Population
18-24 52,464 173 303,300 174,700 128,600
25-29 3 1,730 144 220,300 133,500 86,800
30-34 17,322 110 157,500 112,100 45,400
35-39 1,276 78 93,300 84,400 8,900
Total 774,400 504,700 269,700
Sources: Column 1 from Viral Sratistics of the United Stares. Column 2 provided by Martin
O’Connell from April 1983 CPS tapes. Column 3 = (column I)/(column 2). Column 4 tabulated
from 1980 Census of Population tapes. Column 5 = column 3 - column 4.
which accounts for most births, this sums to 774,000. The difference between
these figures and those in the Census given in column 4 is our estimate of the
unrecorded Mexican-born female population in the child-bearing years:
269,000. If we take 700,000 as our estimate of uncounted Mexicans from
mortality statistics and assume no illegal alien women outside the 18-to-39
age bracket, then women would constitute slightly less than 40% of the un-
counted illegal Mexican migrant population. As the Census shows that 46%
of Mexican-born nonnaturalized immigrants who came after 1970 were
women, the implication is that the undocumented population contains more
males than females, but not by as much as one would think on the basis of
apprehensions of illegals, some 80%-85% of whom are males. One possible
reason may be that Mexican women are more permanent migrants than men
so that they contribute more to the total stock of persons in the country than
to apprehensions at the border.’
All told, both the death and the birth figures support the growing consensus
among demographers (see Passel 1986) that the number of illegal Mexican
immigrants is on the order of two to three million.
Finally, we also note that the estimated size of the illegal alien population
revealed by the demographic data is roughly consistent with the number of
persons who applied for legalization under the provisions of the Immigration
Reform and Control Act of 1986 (IRCA). This legislation provides amnesty
to illegal aliens under one of two provisions. First, amnesty is granted to
aliens who have been present in the country illegally and continuously since
before 1 January 1982 and who applied for amnesty in the year ending on 4
May 1988. Second, amnesty is granted to agricultural workers through the
Special Agricultural Worker (SAW) program if the illegal alien worked in
perishable crop agriculture in the United States for at least ninety days in the
year ending on 1 May 1986 (Immigration and Naturalization Service 1989).
Approximately 1.2 million Mexican-born persons applied for amnesty under
the regular program, and an additional 1.1 million Mexicans applied under
84 G. J. Borjask. B. FreemadK. Lang
the SAW program. Because of widespread fraud in applications to the SAW
program (“A Million Late Arrivals” 1988), the number of undocumented
Mexicans who qualified for amnesty was probably under two million, a num-
ber roughly of the same order of magnitude as the estimates reported in this
paper.
2.1.3 Trends in Mexican Immigration
In addition to providing estimates of the numbers of illegal aliens in the
United States, the Vital Statistics data can be used to estimate the growth rates
of the underlying population. For the brief period in which we have deaths for
the Mexican-born population, there is no dramatic trend upward, as the fol-
lowing numbers of deaths indicate: 1979: 12,288; 1980: 13,180; 1981:
13,135; 1982: 13,078; 1983: 13,066; 1984: 14,050. Taking the end periods,
we have a rate of growth of 14% over the five years, or 2.7% per year com-
pounded. For the longer period over which we have birth rates, the figures are
more dramatic: in 1980, there were 117,126 births to Mexican-born women,
compared to 48,796 in 1970. This implies a near two-and-a-half-fold increase
in the Mexican-born female population, given constant birth rates. However,
comparisons of the number of Mexican-born women counted in the 1970 and
1980 Censuses show that this is entirely consistent with the measured growth
of the number of Mexican-born women-an approximate two-and-a-half-fold
increase from 419,754 in 1970 to 1,038,700 in 1980 and a larger increase in
the number in prime child-bearing years.8 Overall, there were 8 17,000
Mexican-born persons in the 1970 Census, compared to the 2,182,000 in the
1980 Census. If we take 700,000 as the number of Mexican immigrants miss-
ing from the Census, the true population of Mexican-born immigrants in-
creased by some 260% over the period to nearly 2.9 million. This, in turn,
implies a net immigration of 2.1 million persons to the United States in the
1970s-210,000 persons per year-of whom roughly three-quarters were il-
legal. This is an enormous increase in the Mexican-born population due to
illegal immigration, but far below the growth in the number of apprehensions
of Mexicans at the border that underlies much alarmist concern. We turn next
to the apprehension data.
2.2 The Number and Growth of Apprehensions
Apprehensions of illegal Mexican immigrants have increased at truly ex-
traordinary rates since the late 1960s. In 1967, 100,OOO persons were appre-
hended for trying to cross the border; in 1986, nearly 1.7 million persons were
apprehended-a seventeenfold increase that dwarfs our estimated growth of
the Mexican-born pop~l at i on. ~ The level of apprehensions as well as the
growth also seems exceptionally large relative to the estimated size of the
population: from 1970 to 1979 there were some 7.5 million apprehensions
(three-quarters of a million per year), while from 1980 to 1986 there were 6.2
85 Undocumented Mexican-bom Workers in the U.S.
million apprehensions (nearly a million per year). If the number of persons
who successfully crossed the border was, say, four times as large as the num-
ber of apprehensions, and if those who crossed successfully averaged a two-
to three-year stay in the United States, these figures would indicate an illegal
Mexican alien population of about six to nine million persons in 1980 and
perhaps eight to twelve million in the mid- 1980s.I0
Do the apprehension data really imply such a large and rapidly growing
illegal alien population? What caused the explosion in apprehensions? Is there
a way to interpret apprehensions that would make this consistent with Census-
and Vital Statistics-based estimates of the size and growth of the illegal im-
migrant population, or are the different data incommensurate?
We suggest in this section that the growth of the number of illegal aliens
trying to enter the United States is much less than indicated by the apprehen-
sion figures because a sizable part of the increase in apprehensions is due to
the increased efficacy of the Border Patrol. We further suggest, on the basis of
estimates of durations of time in the United States and of the ratio of appre-
hensions to successful crossings for a sample of relatively transient illegal
migrants, that a population of illegals of the magnitude suggested by Census
and Vital Statistics data together with a modest population of Mexicans who
fail to cross the border could have generated the bulk of the apprehensions.
2.2.1
The first factor that suggests that increased Border Patrol activity is a sub-
stantial determinant of the growth of apprehensions is the sharp increase in
real expenditure on the Border Patrol in the 1970s and 1980s. In 1967, the
Border Patrol spent twenty million dollars; in 1986, they spent forty-eight
million in 1967 dollars-a 140% increase. If there were no change in the
productivity of a dollar of resources in apprehending illegal border crossers,
this increase in expenditures could by itself explain over half the growth of
apprehensions. Indeed, as figure 2.1 shows, from the late 1970s to 1986,
when so much concern was expressed about the explosion of apprehensions,
apprehensions per dollar of Border Patrol expenditures rose only modestly,
implying that the trend in apprehensions could have resulted largely from in-
creased Border Patrol resources. From 1967 to 1976, on the other hand, the
number of apprehensions per real dollar expenditure on the Border Patrol in-
creased greatly.
There are two possible explanations for the sharp pre-1976 increase in ap-
prehensions shown in figure 2.1. The first is that the number of attempted
illegal border crossings rose, presumably in response to economic incentives
to migrate illegally to the United States in the wake of the termination of the
Bracero Program. Increased real hourly earnings in the United States in the
late 1960s and early 1970s may have made working in the United States more
attractive, while the growth of real earnings in Mexico may have eased pos-
sible credit constraints in risking an illegal trip to the United States. On the
The Effect of Border Patrol Activity on Apprehensions
86 G. J. Borjash. B. FreernanlK. Lang
35 -
30 -
25 -
20 -
15 -
10 -
5
68 70 72 74 76 78 80 82 84 I 3
Fig. 2.1 Ratio of apprehensions to real Border Patrol expenditure (in constant
dollars)
other hand, the number of legal Mexican immigrants exempt for family rea-
sons from the INS quota limits fell in the period, which is the opposite of what
one would expect given greater economic incentive and capital to enter the
United States. Perhaps more important in terms of long-term immigration, the
Census of Population data seem inconsistent with an explosion of permanent
illegal migration in the period. According to the Census, the number of
Mexican-born persons who immigrated in 1970-74 is just about twice the
number who immigrated during 1965-69, whereas the number of apprehen-
sions in 1970-74 is four times the number in 1965-69. Similarly, the number
of Mexican-born persons who arrived during 1975-80 is about 40% higher
than the number who arrived during 1970-74. The apprehension figures for
1975-80 are more than twice those for 1970-74.” Given that a higher pro-
portion of individuals who immigrated in the earlier period are likely to have
returned to Mexico, these figures raise serious doubts about interpreting the
increased number of apprehensions as reflecting economically induced in-
creases in the number of long-term illegal immigrants.
The second possible interpretation of the 1967-76 spurt in apprehensions is
that it represents a “learning curve” for the Border Patrol following the end of
the Bracero Program. There is scattered evidence that the effectiveness of the
Border Patrol increased over the period. Cornelius (1977) reports that the use
of “coyotes” (smugglers of illegal aliens) increased over time in his sample,
which suggests greater difficulty in crossing over time. And the Border Patrol
introduced more capital intensive and modem technologies to detect illegal
87 Undocumented Mexican-born Workers in the U.S.
aliens, ranging from helicopters to sophisticated electronic detection devices
planted along the border.
As both increased border crossings and increased Border Patrol resources
and effectiveness are likely to have contributed to the observed growth of ap-
prehensions, a quantitative analysis is needed to evaluate the potential magni-
tude of each. Accordingly, we have regressed the log of the number of appre-
hensions on the log of the Border Patrol budget measured in real dollars; the
log of average hourly earnings in the United States and the log of GDP per
capita in Mexico, as indicators of the relative incentive to come to the United
States; and a trend term. Our analysis covers the period 1967-84, when appre-
hensions skyrocketed following the end of the Bracero Program. Table 2.4
presents the regression results. As a base for judging the effect of the Border
Patrol budget and other factors on apprehensions, column 1 records a regres-
sion estimate of the annual compound growth in apprehensions-approxi-
mately 12% per year. Column 2 includes the effect of real Border Patrol ex-
penditure. The estimated elasticity of apprehensions with respect to Border
Patrol expenditure exceeds two, and the annual growth rate in apprehensions
falls to 5%, indicating that over half the observed increase in apprehensions
can be attributed to the growth of Border Patrol spending. Column 3 gives the
regression coefficients and standard errors for the estimated effect of the log
of real Border Patrol expenditures, average hourly earnings in the United
States and GDP per capita in Mexico and the trend variable. Here, the esti-
mated effect of Border Patrol resources on apprehensions has a near unit elas-
ticity, and economic factors also appear to affect apprehensions, with U.S.
Table 2.4 Determinants of Apprehensions of Illegal Mexican Aliens (time-series
estimates, 1967-84)
(1) (2) (3) (4) ( 5)
Budget
Budget ( - 1)
U.S. wage
Mexican GDP per
capita
Time trend .12
(8.8)
Constant 12.05
(78.7)
-
2.22 .94
(7.1) (2.3)
5.17
(3.5)
.66
(1.5)
.05 .08
(4.1) (3.3)
(3.3) (1.3)
10.14 - 4.07
.04
(4.3)
-9.48
(4.0)
.99
.29
(.4)
3.44
(2.3)
(1.3)
.48
(1.0)
.07
(2.0)
-5.22
(1.2)
Sources: Apprehensions data from INS Budget, Central Border Patrol Office. U.S. nonagricul-
tural average hourly wages from the ILO Yearbook (Geneva: ILO). Mexican GDP, population,
and deflator from Intermtional Financial Statistics. U.S. CIP from Economic Report of the Pres-
ident (Washington, D.C.: U.S. Government Printing Office).
88 G. J. Borjas/R. B. FreemadK. Lang
earnings increasing apprehensions and Mexican GDP per capita also increas-
ing them, perhaps as a result of the greater ability of poor Mexicans to raise
capital for migration to the United States. As real wages in the United States
and per capita income in Mexico declined in the late 1970s and early 1980s,
we can treat the coefficient on trend as our estimate of the increase in illegal
border crossings independent of the level of Border Patrol activity and with
economic factors held fixed. It implies that one-quarter of the observed
growth of apprehensions can be attributed to the increased border Patrol
spending. Columns 4 and 5 include lagged Border Patrol expenditure in the
regressions of columns 2 and 3 to capture potential learning effects. The re-
sults show a continued effect for Border Patrol spending, with residual trend
effects ranging from one-third to over half the .12 coefficient in column 1.
While we are leery of crude time-series regressions with just seventeen ob-
servations, it seems reasonable to conclude that Border Patrol activity has
influenced the trend in apprehensions and thus that the trend uncorrected for
Border Patrol activity exaggerates the growth of the flow of illegal Mexican
immigrants. Quantitatively, the regressions in table 2.4 indicate that some-
thing on the order of half the increase in apprehensions is due to increased
Border Patrol expenditure and thus that the growth of illegal crossings was
perhaps half as great as the growth of apprehensions.
2.2.2 Relating Apprehensions to the Stock of Illegal Immigrants
To analyze the relation between the level of apprehensions and the size of
the stock of illegal immigrants, we decompose apprehensions into three cate-
gories: those that result from the apprehensions of new illegal aliens who
eventually cross the border for a first trip to the United States ( P, ) ; those gen-
erated by experienced illegal aliens who make repeated crossings and are liv-
ing in the United States ( P, ) ; and those generated by persons who fail to cross
successfully ( P, ) . The annual number of apprehensions per successful new
crosser we denote as a, ; the number of apprehensions per successful repeat
crosser we denote as a,; the number of apprehensions per failed crosser we
denote as a,. Then total apprehensions ( A) will be
( 3)
This equation shows that apprehensions depend not only on the number of
successful border crossers resident in the United States and the number of
times they are apprehended in a year but also on the number of unsuccessful
crossers and their rate of apprehension.
To obtain information on the number of apprehensions per successful bor-
der crosser in the United States, we use the survey described in section 2.3. In
that survey, we asked the illegal aliens in the United States the number of
times they were apprehended and the number of visits they made to the United
States.Iz The ninety-one persons in the sample who were on their first visit
reported that they had been apprehended by the Border Patrol at least 95
A = a, ( P, ) + a,(P,) + a,(P,).
89 Undocumented Mexican-born Workers in the U. S .
times, giving a ratio of apprehensions to successes of slightly more than one,
which we will use to estimate a, in (3). The 132 individuals who were on a
second or later trip had been apprehended at least 242 times in the course of
356 reported trips.I3 Assuming that they were apprehended an average of once
on their first trip, the implication is that these illegal migrants were appre-
hended about 1 10 times in the course of 224 second or later trips. The ratio of
“successes” to apprehensions thus seems to be about two to one for “experi-
enced” border crossers.
While estimates of successful border crossings to apprehensions range all
over the ballpark, the one-to-one and two-to-one ratios in our data are in line
with the views of some informed observers. Alan Eliason, the chief Border
Patrol agent in San Diego, estimates that “we’re locating, at best, about half
the flow of illegal entrants” (Eliason 1986). “Official” estimates reported in
the newspapers have been in the range of two to three to one.I4
The ratio of successes per apprehension for repeat crossers does not, how-
ever, give us the u2 parameter in equation (3). This is because repeat crossers
may have made more than one trip to the United States in a year, generating
more than one apprehension a year. To estimate the number of annual appre-
hensions generated by repeat crossers residing in the United States, we use
the following steady-state condition:
(4)
u2 = (apprehensions per successful trip)/
(average length of trip measured as a fraction of year).
According to (4), shorter trips generate more apprehensions per person in
the United States because they imply that each person makes multiple trips
per year. Put differently, we must “blow up” apprehensions per trip to obtain
u2 because more than one successful trip is required to make up a full “person-
year” in the United States. In our sample of illegal aliens, the average com-
pleted duration of the most recent trip of persons who were on at least their
second trip was six months-.5 years-which, together with the estimated
number of apprehensions per trip, yields a value of a, of about one.
This estimate, however, may be biased. First, there is sample bias in our
survey group that is due to the greater likelihood of our reaching those with
longer spells. Put differently, failure to interview persons who made success-
ful crossings but are now in Mexico biases our results. Data from CENIET’s
Encuesta Nacional de Emigraci6n a la Frontera Norte del Pais y a 10s Estados
Unidos suggests, however, that our estimates may not be that far off. The
CENIET figures show that, of the nearly one million persons considered by
their families to be living in Mexico who had migrated to the United States in
the 1978-79 period, half were in Mexico at the time of the study (see Diez-
Canedo in this volume). This is consistent with the notion that these migrants
average about half their time in each country. Another problem with our
sample is that it may not be representative of persons uncounted in the Census
or of the sojourner population. Still, for want of better data, we shall use the
90 G. J. BorjadR. B. FreemadK. Lang
estimate that both new and repeated border crossers in the United States are
likely to have generated one apprehension. Finally, while we have no infor-
mation on how many times unsuccessful crossers were apprehended, they had
to have at least one apprehension so that a, > 1.
Given our estimate of one apprehension generated per successful crosser,
what is the likely size of the relevant populations that made a successful cross-
ing (PI + PJ in a year? As a crude estimate of the number who may cross
frequently, we will take persons who were sufficiently transient to have been
missed by the Census and those in the Census who have come after 1970 and
are without their families. Estimates in the previous section suggest that there
were 600,000-700,000 persons in the first category, while calculations given
in the next section suggest that there were about 275,000 persons in the Cen-
sus who came after 1970 and were living without their immediate families,
giving a 1980 population of prospective new and repeated border crossers
close to one million. Given our estimates of a, and a2, this is large enough to
have generated apprehensions of the same magnitude. As additional appre-
hensions were undoubtedly generated by persons who failed to cross the bor-
der, we conclude that the observed number of apprehensions is more consist-
ent with the estimated population of illegal Mexican migrants in the United
States than first appears to be the case.
2.2.3 Seasonality of Apprehensions
The notion that there is a significant population of Mexican migrants who
cross the border often, generating apprehensions, can be checked further by
examining the seasonal pattern of migration. While permanent illegal immi-
gration to the United States might be seasonal, it seems more plausible that
seasonality in apprehensions would reflect sojourner migration, with the same
or different persons crossing regularly after short trips that do not add greatly
to the stock of illegals in the United States at a moment in time.I5
To estimate the seasonality of apprehensions, we obtained monthly appre-
hension data from the INS. For each month, we took the ratio of apprehen-
sions in that month to a twelve-month moving average centered on that month
and averaged the fraction over the sample period. Figure 2. 2 gives the
monthly seasonality factors for the period 1957-64, when the Bracero Pro-
gram was operative, and for 1967-85, when apprehensions skyrocketed.I6
The figure shows substantial seasonality in both periods, but of quite a differ-
ent kind. In the post-Bracero period, apprehensions peak in March and bottom
out in December, consistent with the view that many immigrants return to
Mexico for extended Christmadwinter vacations. In contrast, in the earlier
period, February has the lowest apprehension level, while there is a strong
peak from July through October, which would appear to indicate workers
heading to the U.S. harvests. While a more detailed analysis of the likely
causes of seasonality in apprehensions is needed, the marked seasonality from
1967 to 1985 is consistent with the argument that illegal migrants with short
91 Undocumented Mexican-born Workers in the U.S.
-I
1.300
Jan Peb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
1- 1958-1964 _ _ _ _ _ 1967-19851
Fig. 2.2 Seasonality of alien apprehensions
stays in the United States may have generated a large proportion of the appre-
hensions. Had we found no seasonal pattern, we would feel less comfortable
with our argument that large numbers of apprehensions may be generated by
short-term migrants.
2.3 Characteristics of Illegal Mexican Aliens
In this section, we turn from counts of illegal Mexican aliens to the char-
acteristics of those Mexican-born persons in the 1980 Census of Population
who are likely to be illegal aliens and the characteristics of the illegal male
Mexican-born migrants in our survey in the San Diego area.
2.3.1
As the Census does not contain direct information on whether a Mexican-
born migrant entered the country legally or illegally, we exploit the fact, noted
earlier, that the number of Mexican-born noncitizens in the Census who came
after 1970 was approximately equal in size to Warren and Passel’s (1987) es-
timate of illegal Mexican aliens in the Census. We define this group of persons
as “likely illegal aliens” and compare them to naturalized Mexican-born per-
sons for the purpose of making inferences about the correlates of illegal alien
Likely Illegal Immigrants in the Census
92 G. J. Borjask. B. FreemadK. Lang
status.” Errors of classification in this (or any similar) scheme are likely to
bias downward estimates differences between the groups.
Tables 2.5-2.7 present our analysis of the likely illegal alien population in
the Census. To begin with, table 2.5 records information on the distribution
of the Mexican born by family status. Column 1 gives the percentage distri-
bution of Mexican-born citizens; column 2 gives the distribution for nonciti-
zens; column 3 focuses on noncitizens arriving after 1970 (our likely illegal
migrant group); while column 4 records distributions for male noncitizens
who arrived after 1970. Family status in these data is divided into household-
ers (primarily adult males) and persons (primarily females and children), with
subdivisions to reflect whether the individual is living with his immediate
family (spouse, children, parents), other relatives (primarily siblings), or un-
related persons. Rows 8 and 9 at the bottom of the table show the proportion
of the entire group living with close relatives and the proportion in all other
categories.
What stands out in the data is the large number of Mexican-born nonciti-
zens who live with their families: 77% of all noncitizens, 73% of the recent
immigrants, and even 65% of recently arrived male noncitizens live with their
immediate families. Given that the bulk of these populations consist of ille-
gals, the implication is that the majority of illegal Mexican-born residents in
the 1980 Census reside here with their families. To the extent that immigrants
living with their families should be viewed as relatively permanent migrants,
table 2.5 suggests that most illegal Mexican migrants counted by the U.S.
Census are permanent migrants.
Table 2.5 Percentage Distribution of Mexican Born by Family Status
Family Status
Noncitizens Arrived Male Non-
All Citizens Before 1970 After 1970 after 1970
citizens Arrived
Householders living with:
1. Closely related persons 36.2
2. “Other relatives” 2.0
4. Single householders 8.0
5. Closely related house-
holder 45.1
6. Other relative house-
holder 4.9
7. Unrelated householder 2.7
Total householders and persons living with:
8. Close relatives 81.3
9. All other 18.7
Number (OOOs) 509.0
3. Unrelated persons 1.1
Persons living with:
26.6 20.3
2.1 2.1
1.2 1.3
4.1 3.1
49.9 52.2
10.5 13.7
5.6 7.3
76.5 72.5
23.5 27.5
1,674.0 1,090.0
32.9
3.1
2.2
4.9
31.6
15.6
9.6
64.5
35.5
593.0
Sources: Tabulated from the U.S. Bureau of the Census.
93 Undocumented Mexican-bom Workers in the U.S.
Table 2.6 Means of Socioeconomic Variables
Noncitizens
Citizens Likely Legal Likely Illegal
Variable Male Female Male Female Male Female
Education 7.41
LFP .77
FXlIl . l l
Maid .oo
LWKS 3.70
LWeekly 5.27
Size 4.51
Year of migration:
Age 37.64
1975-80 .16
1910-74 .18
1965-69 .14
1960-64 . l l
1950-59 .I7
< 1950 .24
Sample size 2,623
7.17
39.40
.43
.04
.01
3.41
4.88
4.51
.I4
.18
.14
.10
.17
.27
2,471
7.02
40.62
.80
.15
.oo
3.12
5.35
4.83
.oo
.oo
.41
.22
.23
.22
2,886
6.88
42.68
.41
.07
.02
3.46
4.83
4.72
.oo
.oo
.37
.23
.23
.17
2,945
5.92
23.77
.88
.16
.oo
3.65
5.04
5.33
.59
.41
.oo
.oo
.oo
.oo
5,933
5.51
23..96
.46
.06
.02
3.39
4.76
5.71
.56
.44
.oo
.oo
.oo
.oo
4,971
Nore: LFP = 1 if participating in the labor force; 0 otherwise. Farm = 1 if employed in the
agricultural sector, 0 otherwise. Maid = 1 if employed in the personal services industry, 0 oth-
erwise. LWKS = log of weeks worked in 1979. LWeekly = log of weekly earnings in 1979.
Size = household size. The labor force variables are calculated among persons aged 16 or older.
Table 2.7 Determinants of Labor Market Outcomes
Dependent Variable
Variable LI T LWKS LWeekly Farm
Likely illegal ,026 - ,014 - ,179 - ,009
(3.09) ( - .93) (7.90) ( - .83)
Citizen - ,003 - ,003 - ,073 - ,005
( - .50) (-.16) ( - 2.89) ( - .43)
Education ,0003 .004 ,029 - ,017
(.38) (3.09) ( 13.40) ( - 17.14)
Age ,047 ,049 .075 - ,015
(28.78) (15.71) ( 1 5 SO) (-6.48)
Age squared - ,0006 - ,0006 - .0008 ,0002
( - 27.88) ( - 14.03) (-13.29) (6.66)
R2 .09 .04 .09 .05
Note: The r-ratios are presented in parenthesis. The regressions are restricted to men aged
16-64.
94 G. J. Borjask. B. FreemadK. Lang
Table 2.6 gives the means of selected socioeconomic characteristics for the
likely illegal alien group, for citizens, and for noncitizen Mexican-born per-
sons who are likely to be legal. The differences between the groups shown in
the table suggest that our classification succeeds in capturing important as-
pects of the likely illegal alien group. For example, the average age of a likely
illegal alien man is 23.8 years, while the average age of other Mexican-born
groups (citizens and noncitizens who arrived before 1970) is between 38 and
41 years. Similarly, the mean years of schooling of a likely illegal alien man
is between 1 and 1.5 years below that of other Mexican-born men. The means
in the table also reveal differences in labor force participation rates (higher for
likely illegals) and in earnings (lower for likely illegals). These differences
notwithstanding, table 2.6 is also remarkable for what it does not show. It does
not show the likely illegal population to be primarily male: only 54.4% of the
group are men, compared to 50.4% of the remainder of the Mexican-born
population. It does not show the likely illegal alien population to be heavily
concentrated in agriculture; only 16% of men are employed in agriculture, as
contrasted to 11%-15% of other Mexican-born men.
What about the labor market experience of likely illegals? Table 2.7 exam-
ines the effect of likely illegal status on four aspects of market performance
for men aged 16-64: labor force participation (LFP); In weeks worked over
the year (LWKS); In weekly earnings (LWKLY); and the probability of agri-
cultural employment (FARM). It shows that, with other variables held fixed,
the likely illegals have somewhat higher rates of participation in the work
force than other Mexican-born persons, which makes sense if they migrate to
obtain work and leave when they are out of work, and have much lower earn-
ings than other Mexican-born persons, which also makes sense given their
likely lower level of skill and lack of recourse to legal protections.
The equations in table 2.7 constrain the coefficients of various socioeco-
nomic characteristics to be the same for likely illegal aliens and other
Mexican-born groups. This constraint is implausible given that illegals are
likely to have less incentive to invest in human capital than legal migrants
because of the likely shorter periods of time that they spend in the United
States and are likely to have education that is less suitable to the job market.
Accordingly, we estimated earnings equations separately for likely illegals
and likely legal migrants, obtaining substantial differences in the effect of age
(A) and education (E) and citizen status ( C) on log weekly earnings, as the
following regressions show for likely legals:
LWKLY = .034*E + .105*A - .001*A2 - .063*C, R2 = .09
(11.6) (15.0) (13.3) (2.4)
and for likely illegals:
LWKLY = .023*E + .061*A - .001*A2, R2 = .04,
(7.6) (8.1) (7.0)
95 Undocumented Mexican-born Workers in the U.S.
where both regressions include a constant term, and the t-statistics are in pa-
rentheses.
The finding that the earnings of legal migrants are more responsive to tra-
ditional human capital variables than the earnings of illegal migrants is con-
sistent with evidence provided by Chiswick (1986a, 1986b) using a survey of
undocumented workers in Chicago. The similarity between our results and
those for the undocumented workers in Chicago suggests that problems of
misclassification are not overly serious for our sample. Moreover, it suggests
that Chiswick’s results for Chicago generalize to the broader population of
illegal Mexican immigrants.
2.3.2 Results from Our Survey
As noted earlier, we conducted a small survey (289 observations) of illegal
Mexican male migrants in the San Diego area; the participants were chosen to
cast light on aliens unlikely to be counted in the Census.ls Interviews took
place at downtown “shape-ups,” in agricultural “residences,” or wherever
community contacts led us to illegal Mexican workers. The sample therefore
captures the least stable and lowest-paid segment of the illegal immigrant
community. While this sample of male illegal immigrants is by no means ran-
dom, it is still instructive to look at their characteristics.
As can be seen in table 2. 8, the men in the sample are lower paid and more
Table 2.8 Sample Characteristics in Survey of Illegal Aliens
Education
Age
Farm
LWeekly
Year of first trip (%):
1986
1985
1984
1983
1982
1981
1980
Before 1980
Family status: (number):
With wives living in U.S.
With children but without wives
Potential immigrant status:
Intend to stay in U.S. permanently
Will remain in U.S. indefinitely
Will return to Mexico and not come back to U.S.
N
6.31
28.36
0.26
4.90
25
26
17
1
10
3
6
6
39
3
31
87
50
289
Sources: 1986 Summer Survey of Illegal Aliens in the San Diego Area.
96 G. J. Borjask. B. FreemadK. Lang
Table 2.9 Determinants of Log-weekly Wages (survey sample)
Education - ,002
(1.2)
(3.1)
(1.3)
R2 .12
Age ,045
Age squared - ,005
Note: The t-ratios are presented in parentheses.
heavily concentrated in agriculture than the likely illegal group in the Census
(compare table 2.6). Adjusting for the fact that the survey was conducted six
years after the Census, the survey sample arrived more recently than the ille-
gal immigrants in the Census, implying that the sample does indeed reach the
group we intended. Even so, a substantial number of persons in the sample
appear to be on their way to becoming permanent immigrants: almost one in
seven had wives living in the United States, a figure below that for likely
illegals in the Census but still nonnegligible; thirty-seven said they intended
to remain in the United States permanently, and eighty-seven intended to re-
main indefinitely, at least if they find employment. Only fifty said that they
intended to return to Mexico and not come back to the United States.
Finally, table 2.9 gives the results of the estimation of the weekly earnings
equation for our sample. Relative to the estimates for the likely illegals in the
Census, the effects of age on earnings are attenuated, and the effect of educa-
tion is negative and statistically insignificant, further suggesting that human
capital variables do not do much for the pay of illegal aliens. Whether these
results reflect the differential experience of relatively recent and temporary
immigrants or our sample design, which includes largely low-wage workers,
is an open question. They do, however, confirm that we have indeed identified
a very different set of illegal immigrants than those in the Census.
2.4 Conclusions
Given the difficulties in trying to measure any illegal activity, conclusions
about the size and socioeconomic characteristics of the illegal Mexican popu-
lation in the United States must inevitably be subject to numerous caveats.
Our response to the problem of “inaccessibility” of the population of illegals
has been to examine several different data sets, to search for consistencies
among them that would allow for firm conclusions, and to make “strong”
assumptions to obtain bounds on critical statistics. We found that the bulk of
the data are consistent with the existence of an illegal Mexican-born popula-
tion on the order of 1.8 million in the 1980s, that this population has grown
rapidly over the decade but at a rate far below the growth of apprehensions,
that a large portion of the illegal Mexican migrant population consists of “per-
97 Undocumented Mexican-born Workers in the U.S.
manent” migrants, but that border crossings by the transient part of the illegal
migrant population may underlie a large portion of apprehensions. Despite
the diverse data problems that we encountered, the consistency in our results
across data sets lends some credence to our conclusions.Ig
Notes
1. The estimates from top officials in the 1970s included four to seven million (for-
mer Attorney General William Saxbe) and four to twelve million (Commissioner
Chapman of the Immigration and Naturalization Service [INS]). Lesko and Associates
estimated through a Delphic technique that there were 8.2 million illegal aliens in
1975, of which 5.2 million were Mexicans. The INS used a similar consensus method
and came up with 5.5-6 million as of late 1975.
2. Census demographers have consistently come up with estimates below those in n.
1 above. See Lancaster and Sceuren (1977), Heer (1979), Robinson (1980), and War-
ren and Passel (1987). Other studies of the illegal alien population include Bean, King,
and Passel (1983), Brown and Shue (1983), Corwin (1982), Cuthbert and Sterens
(1981), Fogel (1978), Heer and Passel (1985), Jones (1984), North and Houstoun
(1976), Passel and Woodrow (1984), Reichert and Massey (1979), and Siegal, Passel,
and Robinson (1980).
3. The insight that death statistics can be used to measure the hidden population can
be attributed to Robinson (1980). Death rate statistics by country of origin were not
available at the time he conducted his research. Instead, he used changes in death rates
in states such as California that were expected to have large illegal immigrant popula-
tions compared with changes for the United States as a whole. Since Robinson’s tech-
nique is not subject to bias if country of origin is misrecorded, we attempted to use it
to get estimates of the “missing” illegal population. Unfortunately, our experience sug-
gests that the technique is not robust. Robinson implicitly assumes that the death rate
in California (or in other states with large illegal alien populations) changes by the
same amount as death rates in other states unlikely to have many illegal aliens. The
assumed unit coefficient linking death rates is, however, inconsistent with the actual
pattern of death rates for the period 1960-70, when changes in the size of the illegal
immigrant population are expected to be small: regression analysis shows very little
connection between changes in state death rates and national totals. The problem is
that state death rates are very “noisy.” When we “smoothed’ the data, we found strik-
ingly different results depending on whether we started our analysis in 1969, 1970, or
1971 and whether we ended it in 1980 or 1984. Without smoothing, the results are
even more sensitive to the choice of base year. Hence, we have eschewed use of Rob-
inson’s technique in this paper.
4. Telephone interview, 10 August 1987.
5. We are less certain about the bias in the birth rate calculations since some of the
Mexican-born women not counted in the Census may be temporary sojourners who are
unlikely to have children. Others, however, may resemble the Mexican-born women
in the Census, be living with their families, and have birth rates more like those in
Mexico than like those of immigrants permanently established in the United States.
6. In their published article, Bachu and O’Connell (1984) do not report birth rates
for Mexican-born women by detailed age. They kindly provided us with the relevant
numbers from their computer printouts.
7. There is an important conceptual problem with the use of birth data to estimate
98 G. J. Borjas/R. B. FreemadK. Lang
the size of the illegal alien population. Because fertility decisions are endogenous,
many Mexican women may temporarily migrate to the United States simply to bear
their children. This ensures that their offspring are American citizens and thus have the
option, on reaching adulthood, of migrating legally to the United States. To the extent
that this type of migration is common among Mexican-born women, the illegal alien
estimates provided by the birth data are biased upward.
8. The 1980 Census has 474,000 Mexican-born women in the age group 15-34,
whereas the 1970 Census reports 156,000 in the age group 14-34. This is a threefold
increase.
9. There was also a sizable number of apprehensions of illegal aliens in the 1950s.
We have not contrasted the situation then to that in the 1970s and 1980s.
10. These estimates are obtained by multiplying the number of apprehensions per
year by the postulated ratio of successful crossings to apprehensions by the postulated
duration under steady-state assumptions.
11. Our tabulations of the 1980 Census show that 13% of migrants came in the
1965-69 period, compared to 25% in the 1970-74 period, while 33% of migrants
came in 1975-80. Dividing these percentages gives the figures in the text.
12. There is some problem with interpretation of the questions. We have added
together the total number of times individuals report having been refused entry to the
United States and the number of people who report having been caught by the Border
Patrol within the United States. The interviewer who conducted the survey believes
that respondents interpreted the questions as meaning that they had been caught by the
Border Patrol entering the United States (refused entry) or having made significant
progress into the United States (apprehended by the Border Patrol within the United
States) but before having reached their destination. Assuming that this interpretation is
correct, we underestimate the number of apprehensions because we do not know how
many times an individual was apprehended after having made significant progress into
the United States.
13. Because not all illegal immigrants answered all questions, the number of re-
sponses we use to generate various statistics does not necessarily sum to the total in the
survey.
14. New York Times, 21 February 1986; Newsweek, 17 March 1986. On the other
hand, we note that our interviews with Border Patrol agents produced noticeably
higher estimates, averaging around four to one.
15. In particular, there are no strong economic reasons for permanent migrants to
move at one time in the year rather than another, given that any seasonal differences in
returns to moving would be amortized over a long period.
16. We allowed a two-year gap for adjustment to the end of the Bracero Program.
17. For a comparable analysis, see Bean, Browning, and Frisbie (1984).
18. The survey was conducted by Eric Waggoner, a student at Harvard University.
19. Our conclusions are also consistent with the number of illegal Mexican immi-
grants who sought amnesty under the new immigration law and with the apparent re-
luctance of many of those eligible to apply for amnesty because of fears that other
members of their families might not be eligible. Our analysis also suggests that the
decline in apprehensions that began in about October 1986 may reflect in part not only
fears by potential new illegal immigrants that employer sanctions will destroy their
chances to obtain employment but also the fact that the new law made it potentially
costly for current immigrants to return to Mexico for brief visits since being caught
might jeopardize their claim to being continuously resident in the United States. More-
over, the “grandfather” clause exempting existing employees from the employer sanc-
tions may also have the unintended consequence of turning sojourner laborers into
permanent residents owing to the increased cost of giving up their jobs to return to
Mexico.
99 Undocumented Mexican-born Workers in the U.S.
References
Bachu, Amara, and Martin O’Connell. 1984. Developing Current Fertility Indicators
for Foreign-born Women from the Current Population Survey. Review of Public
Data Use 12:185-95.
Bean, Frank D., Harley L. Browning, and W. Farker Frisbie. 1984. The Socio-
demographic Characteristics of Mexican Immigrant Status Groups: Implications
for Studying Undocumented Mexicans. International Migration Review 18(3):672-
91.
Bean, Frank D. , Allan G. King, and Jeffrey S. Passel. 1983. The Number of Illegal
Migrants: Sex Ratio Based Estimates for 1980. Demography 20, no. l(Feb-
mary):99-109.
Brown, Peter G. , and Henry Shue, eds. 1983. The Border That Joins: Mexican Mi -
grants and US . Responsibility. Totowa, N.J.: Rowman & Littlefield.
Chiswick, Barry R. 1986a. Illegal Aliens: A Preliminary Report on an Employee-
Employer Survey. American Economic Review 76, no. 2(May):253-57.
. 1986b. Mexican Immigrants: The Economic Dimension. Annuls of the Amer-
ican Association of Political and Social Science 487(September):438-53.
Cornelius, Wayne A. 1977. Illegal Mexican Migration to the United States: Re-
cent Research Findings and Policy Implications. Congressional Record (July 13):
Corwin, Arthur F. 1982. The Numbers Game: Estimates of Illegal Aliens in the United
States, 1970-1981. Law and Contemporary Problems 45, no. 2(Spring):223-97.
Cuthbert, Richard W., and Joe B. Sterens. 1981. The Net Economic Incentive for
Illegal Mexican Migration: A Case Study. Znternational Migration Review 15, no.
3(Fall):54 1-49.
22726-32.
Eliason, Alan. 1986. Letter to the Editor. New York Times, 29 April.
Fogel, Walter. 1978. Mexican Illegal Alien Workers in the United States. Los Angeles:
University of California Press.
Heer, David M. 1979. What Is the Annual Net Flow of Undocumented Mexican Im-
migrants to the United States? Demography 16, no. 13(August):417-23.
Heer, David M., and Jeffrey S. Passel. 1985. Comparison of Two Different Methods
for Computing the Number of Undocumented Mexican Adults in the Los Angeles
PMSA. Paper presented at the annual meeting of the Population Association of
America, Boston.
Immigration and Naturalization Service. 1989. Provisional Legalization Application
Statistics. Statistical Branch Analysis. 12 May.
Jones, Richard C., ed. 1984. Patterns of Undocumented Migration: Mexico and the
United States. Totowa, N.J.: Rowman & Allanheld.
Lancaster, Clarice, and Fredrick J. Sceuren. 1977. Counting the Uncountable Illegals:
Some Initial Statistical Speculations Employing Capture-Recapture Techniques.
Proceedings of the American Statistical Association, Social Statistics Section, pt. 1,
pp. 530-35.
“A Million Late Arrivals.” 1988. Time, 12 December.
North, D. S., and M. F. Houstoun. 1976. The Characteristics and Role of Illegal
Aliens in the U S . Labor Market: An Exploratory Study. Washington, D.C.: Linton.
Passel, Jeffrey. 1986. Undocumented Immigration. Annals of the American Associa-
tion of Political and Social Science 487(September): 18 1-200.
Passel, Jeffrey S., and Karen A. Woodrow. 1984. Geographic Distribution of Un-
documented Immigrants: Estimates of Undocumented Aliens Counted in the 1980
Census by State. International Migration Review 18(Fall):642-7 1.
Reichert, J. S., and D. S. Massey. 1979. Patterns of U.S. Migration from a Mexican
100 G. J. BorjadR. B. FreemadK. Lang
Sending Community: A Comparison of Legal and Illegal Migrants. International
Migration Review 13, no. 4(Winter):599-624.
Robinson, J. G. 1980. Estimating the Approximate Size of the Illegal Alien Popula-
tion in the United States by the Comparative Trend Analysis of Age-specific Death
Rates. Demography 17, no. 2(May):159-76.
Siegal, Jacob S., Jeffrey S. Passel, and J. Gregory Robinson. 1980. Preliminary Re-
view of Existing Studies of the Number of Illegal Residents in the United States. In
U. S. Immigration Policy and the National Interest. app. E, Papers on Illegal Immi-
gration to the United States, 13-40. Staff Report of the Select Commission on Im-
migration and Refugee Policy. Washington, D.C.: U.S. Government Printing Of-
fice.
Teitelbaum, Michael S. 1986. Immigration and Demographic Change. In World Pop-
ulation and U.S. Policy, ed. Jane Menken. New York: Norton.
Warren, Robert, and Jeffrey S. Passel. 1987. A Count of the Uncountable: Estimates
of Undocumented Aliens Counted in the 1980 Census. Demography 24 (Au-
gust):375-93.
3 The Effect of Policy
Restrictions on Capital and
Labor Flows in Mexico
Juan Diez-Canedo R.
This study analyzes the magnitude and characteristics of migratory flows to
Mexican border states and to and from the United States using the Mexican
Encuesta Nacional de Emigraci6n a la Frontera Norte del Pais y a 10s Estados
Unidos, a special 1978 household survey designed to obtain data on such
flows. The study focuses mainly on those aspects of the migratory flows that
can be traced to the distinct free trade commercial regime in the border area.
The findings of the study are as follows. (1) At the time of the 1978 survey,
1.5 percent of the Mexican working population-about 520,000 people-
were working in the United States, a number far below the millions often cited
in the press, while an additional 1.3 percent were return migrants. (2) The
border states enjoyed an economic boom when Mexico was in a prolonged
recession, with capita inflows and a migrant stream exceeding that to the
United States. (3) Other characteristics held fixed, residence in the border area
has become nearly as important a determinant of migration to the United
States as residence in the historical source regions of the country. (4) Migrants
to the border differed in their personal characteristics and regional source from
migrants to the United States, for reasons seemingly related to the commercial
policies in the border area and suggestive of a family migration pattern. Wives
work at the border area, while husbands migrate back and forth to the United
States.
3.1 Migration Patterns
Because of the scarcity of migration data in general and the poor quality
and inherent biases of the available information on illegal migration, in 1978
Juan Diez-Canedo R. is Deputy Director General, Investment Banking, Banco Intemacional,
S.N.C.
The author is grateful to Enriqueta Lopezlira, Fiorella Tapia, and Gisela Pineda for excellent
research assistance and to Kenneth Flamm for comments made on a previous draft of this paper.
Of course, he alone is responsible for the contents of this paper.
101
102 Juan Diez-Canedo R.
the Mexican Labor Department (Secretaria del Trabajo) conducted a national
household survey in order to measure migratory flows to and from the United
States (people who worked in the United States from January 1974 to Novem-
ber 1978 but were living in Mexico at the time of the survey), migratory flows
to the Mexican border adjoining the United States, and internal migrati0n.l
This national household survey was carried out from December 1978 to Janu-
ary 1979 and covered 115 locations and 62,500 homes selected on a probabi-
listic basis. A census was done for each household, and four different ques-
tionnaires were used depending on the different migratory characteristics.
These questionnaires covered 300,000 individuals. The information was
grouped according to region (see fig. 3.1) and according to variables such as
age, sex, marital status, rural or urban origin, job characteristics, education,
and whether respondents were employed, unemployed, or not in the labor
force.* Only the most important results pertaining to migration to and return
migration from the United States were analyzed in a study by Garcia y Griego
and Zazueta (1982). The data on migration to the border and internal migra-
tion have not been analyzed.
An overall view and the relative importance of the different migratory pat-
terns can be seen in table 3.1. First, at the time of the interview, 1.5 percent
of the working-age population were in the United States, 1.3 percent were
return migrants, and 1.6 percent had migrated to a border county. In terms of
absolute numbers, the figure of 520,000 migrants to the United States shown
in table 3.1 is broadly consistent with the figures of Borjas, Freeman, and
Lang (in this volume) and inconsistent with the millions of illegal immigrants
bandied about in the popular American press. Using information from the
Agua Prieta
Fig. 3.1 The border area and other regions of Mexico
Table 3.1 Migratory Characteristic and Region (population fateen years and older)
Region
I I1 I11 IV V
Total
Total 35,622,489 (100.0) 1,688,098 (100.0) 6,945,926 (100.0) 3,153,942 (100.0) 5,152,855 (100.0) 18,681,668
(100.0) (4.7) (19.5) (8.9) (14.5) (52.4)
Migration to the United States 519,406 (1.5) 50,044 (3.0) 205,740 (3.0) 140,010 (4.4) 77,336 (1.5) 46,276
Return migration 474,888 (1.3) 58,194 (3.4) 140,861 (2.0) 119,120 (3.8) 96,491 (1.9) 60,222
(100.0) (12.3) (29.7) (25.1) (20.3) (12.7)
Migration to the border 557,966 (1.6) 317,972 (18.8) 75,929 (1.1) 37,838 (1.2) 89,446 (1.7) 36,781
(100.0) (57.0) (13.6) (6.8) (16.0) (6.6)
Internal migration 12,312,267 (34.6) 637,723 (37.8) 2,660,613 (38.3) 784,485 (24.9) 1,868,745 (36.3) 6,360,701
(100.0) (5.2) (21.6) (6.4) (15.2) (51.7)
Nonmigrants 21,484,101 (60.3) 614,551 (36.4) 3,797,756 (54.7) 2,051,931 (65.1) 2,987,434 (58.0) 12,032,429
(100.0) (2.9) (17.7) (9.6) (13.9) (56.0)
Nonspecified 273,861 (.8) 9,614 (.6) 65,027 (.9) 20,558 (.7) 33,403 (.6) 145,259
(100.0) (3.5) (23.7) (7.5) (12.2) (53.0)
(100.0) (9.6) (39.6) (27.0) (14.9) (8.9)
Source: Encuesta Nacional de Emigraci6n a la Frontera Norte del Phis y a 10s Estados Unidos, 1978-79 (CENIET 1984).
Nore: For definitions of regions, see fig. 3.1. Numbers in parentheses are percentages.
104 Juan Diez-Canedo R.
survey, Garcia y Griego and Zazueta (1982, 50) estimated that 750,000 Mex-
ican workers were working in the United States at some point during 1978.
Second, most of those workers either migrating to or returning from the
United States lived in regions I1 and I11 (66.6 and 54.8 percent of total migra-
tion from such regions, respectively), which together comprise the states of
the center of Mexico (see the Appendix). Third, almost 60 percent of migrants
living in the border area had come from another border county, while 16 per-
cent came from an adjoining state (see fig. 3.1). Fourth, as could be expected,
the bulk of the population does not migrate (60.3 percent), and the most im-
portant migratory flows are internal (34.6 percent).
While there are several studies on Mexican internal migration (see Diez-
Canedo 1980; Isoard 1976; Greenwood and Ladman 1977; Greenwood 1978)
and on Mexican migration to the United States, few analyze migration to the
border areas, although the border cities were among the ones with the highest
rates of growth in Mexico. From 1970 to 1980, these cities had annual average
rates of growth as high as 7 percent for Tijuana or 6.5 percent for Matamoros
(see table 3.2), placing them probably among the cities with the highest rates
of growth by world standards. Such growth, however, may be explained in
part by two important factors. First, the nearly two-thousand mile Mexico-
U.S. border is probably a unique case-within a few yards, the going mini-
mum wage leaps from around $.40 (Mexico) to $3.35 (the United States) an
Table 3.2 Population Change in Major Mexican Cities Bordering the United
States and in U.S. SMSA’s Bordering Mexico
Population (thousands)
Percentage Change,
1970 1980 1970-80
Tijuana
Mexicali
Ciudad Jufirez
Nuevo Laredo
Reynosa
Matamoros
United Mexican States
San Diego
Tucson
Las Cruces
El Paso
Laredo
McAllen
Brownsville
South
West
United States
227
267
407
149
137
138
50,695
1,357
352
70
359
73
182
140
63,000
35,000
203,000
542
495
680
272
240
258
69,393
1,861
531
96
480
99
283
210
75 ,000
43 ,000
227,000
96
85
67
83
75
87
37
51
38
34
36
56
49
19
23
11
Source: Hansen (1985).
105 The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
hour (see table 3.3). Second, also at the time of the study, the northern border
of Mexico operated under a different trade regime (which basically allows free
trade) from the one applied in the rest of the country.
The wage differential, along with the U.S. labor structure, which condi-
tions a high demand for migrant workers, has created a tremendous magnetic
force that attracts millions of Mexican workers. Also, an interesting phenom-
enon of capital and labor attraction has been happening at the border. Through
special trade and foreign-ownership laws and rapidly increasing employment
creation, large migratory flows have been seeking permanent residence in the
border counties.
On the other side of the border, the U.S. Standard Metropolitan Statistical
Areas (SMSAs) bordering Mexico have also experienced very rapid rates of
growth (see table 3.2). El Paso, which had the lowest 1970-80 border SMSA
population growth rate (34 percent), grew at over three times the correspond-
ing U.S. national rate of 11 percent and well above the rates of the South (19
percent) and the West (23 percent). Also, the increase in personal income in
those SMSAs was, except for the Las Cruces and El Paso, higher than the
Table 3.3 Minimum Hourly Wages (annual average in dollars)
Mexican Minimum Wage
as a Percentage of
Mexico (1) United States (2) U.S. Minimum Wage (2)/(1)
1966 .24 1 .oo 24.0
1967 .24 1 .oo 24.0
1968 .28 1.14 24.5
1969 .28 1.29 21.7
1970 .33 1.44 22.9
1971 .33 1.59 20.7
1972 .39 1.60 24.3
1973 .41 1.60 25.6
1974 .55 1.80 30.5
1975 .65 2.00 32.5
1976 .67 2.20 30.4
1977 .59 2.30 25.6
1978 .66 2.65 24.9
1979 .77 2.90 26.5
1980 .89 3.10 28.7
1981 1.09 3.35 32.5
1982 .78 3.35 23.2
1983 .52 3.35 15.5
1984 .58 3.35 17.3
1985 .59 3.35 17.6
1986 .42 3.35 12.6
Source: For Mexico, Salarios Minimos 1987, Comision Nacional de 10s Salarios Minimos (Sa-
lario Minimo General Promedio Nacional, in dollars using the average controlled rate). For the
United States, Staristical Abstract of the United States (federal minimum hourly wage rate for
nonfarm workers).
106 Juan Diez-Canedo R.
increases in the United States as a whole and in the South and the West (Han-
sen 1985).
3.2 The Border Commercial Zone
Mexico’s Border Industrialization Program was created in the mid- 1960s
and aimed in part to absorb what was perceived as growing unemployment in
the border areas due to the termination of the Bracero Program in 1964. Its
most important element was the creation of a different trade regime for the
border areas through the Muquilu Program. Since 1965, duty-free imports of
machinery, equipment, and components for processing and assembly within a
twenty-kilometer strip along the border were allowed, provided that all im-
ported products were reexported. The assembly plants are called maquila-
doras; they allow for 100 percent foreign ownership as well as exemption of
export Also, in some cases, firms may sell up to 20 percent of their
production in Mexico. Correspondingly, U.S. tariff regulations 806.30 and
807.00 permit the return of the U.S. component portion duty free, taxing only
the value added in Mexico.
The development of the maquiladoras has been surprising. At the start of
this program, 806.30/807.00 data show that Mexico was less than five times
as important as Hong Kong and about as important as Taiwan in the process
of industry exports to the United States (see Grunwald and Flamm 1985). In
1983, total 806.30/807.00 imports from Hong Kong and Taiwan were only 12
and 15 percent, respectively, of the Mexican 806.30/807.00 imports. In that
year, the main imports under 806.301807.00 came from Japan (30.0 percent),
Mexico (18 percent), and West Germany (13 percent).
In the last few years, employment in the Mexican manufacturing sector has
actually declined, and the gross capital formation in the economy has dropped
by 28.5 percent in real terms from 1982 to 1986. While in that period total
employment in the Mexican manufacturing industry diminished 6.7 percent,
employment in the border assembly plants grew by 96.5 percent, and average
work hours increased 88 percent. Thus, maquiladora employment as a per-
centage of total manufacturing employment increased from 3.4 percent in
1975 to almost 10.8 percent in 1986 (see table 3.4).
Although offshore investment has been considered to be “footloose” (Piore
1979, 35-43), especially in the semiconductor assembly operations, and has
been found to be highly dependent on the U.S. economic cycle (Bolin 1964),
it has also been shown that the U.S. economic cycle has had greater effects on
multinational corporations inside the United States than in their offshore op-
erations, for they tend to cut the most costly operations first (Grunwald and
Flamm) .
The importance of labor to assembly costs, proximity, and the relatively
unskilled but highly productive nature of the segment working in assembly
plants which, as may be seen in table 3.4, is composed mostly (although in-
107 The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
Table 3.4 Total Employment and Participation of Women in the Muquila Industry,
1975-86 (in thousands)
Total Employment Maquiladoras
Blue Collar Female Participation
Industry Maquiladoras" Total Women Blue Collar Total
Manufacturing
(1) (2) (3) (4) (5) = (4)/(3) (6) = (4)/(2)
1975 2002 67.2 57.9 45.3 78.3 67.4
1976 2046 74.5 64.7 51.0 78.8 68.4
1977 205 1 78.4 68.2 53.2 78.0 67.8
1978 2133 90.7 78.6 60.4 76.8 66.6
1979 2291 111.4 95.8 73.8 77.1 66.3
1980 2417 119.6 102.0 78.9 77.3 66.0
1981 2543 131.0 110.7 85.7 77.4 65.4
1982 2485 127.1 105.4 81.4 77.2 64.1
1983 2340 150.9 125.3 93.3 74.5 61.8
1984 2361 199.7 165.5 117.3 70.9 58.7
1985 2415' 212.0 173.9 120.0 69.0 56.6
1986 231gb 249.8 203.9 139.1 68.2 55.7
Sources: Estadfstica de la Industria Maquiladora de Exportacidn, Subdireccidn de Estadisticas Eco-
nbmicas, Secretm'a de Programacidn y Presupuesto, Mexico D.F. (1987). Sistema de Cuentas Naci-
onales de Mdxico, Secretm'a de Programacidn y Presupuesto, Mexico D.F. (1987). Indicadores Eco-
ndmicos, Banco de M&xico, Mexico D.F. (1987).
'Includes blue collar and white collar.
bEstimation based on annual variations reported in the Encuesta Industrial Mensual, Secretaria de Pro-
gramacidn y Presupuesto.
creasingly less so) of women, helps to explain the relatively steady growth of
the maquiladoras.
In the last ten years the number of assembly plans has practically doubled
and, with the exception of 1982 (the year of the Mexican debt crisis), its
growth has been steady. Except for 1982, during which the peso was clearly
overvalued and there were signs of political instability and no economic
growth, investment in the assembly plants has been not only stable but grow-
ing significantly, while as of the 1982 crisis and up to 1989 the opposite hap-
pened to investment in the manufacturing sector.
Figure 3.2 helps to explain in part why such a phenomenon happened.
After 1970 and up to 1986 the real exchange rate, estimated using the con-
sumer price indexes, was considerably favorable for exports, with the excep-
tion of only three years. If the real exchange rate is calculated using, instead
of consumer prices, wage indexes which are more relevant from the point of
view of exporters, the Mexican real exchange rate has a considerable increase
of 113.3 percent as of December 1985, from the base year of 1970. So the
global competitiveness of manufacturing in the period under study was con-
siderable, and even more so at the border where, on top of that, there have
been no import restrictions whatsoever.
108 Juan Diez-Canedo R.
225
175 t
0.
’ 0 .
. .
0
150
125
100
65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86
Fig. 3.2 Real exchange rate index (1970 = 100)
Source: Gerencia del Sector Real, Banco de Mexico.
0 Mean-Consumer Prices: Estimated in base to a divisia index of consumer price index of
133 countries, and a divisia index of the exchange rate of each country, weighted both by the
share of each country’s GDP on total GDP. 0 Mean-Wages: Estimated in base to a divisia
index of the wage indices of 19 countries, and a divisia index of the exchange rate of each
country, weighted both by the share of each country’s GDP on total GDP.
The migratory flows both to the border and to the United States must be
following better economic opportunities. However, it is interesting to see
which factors separate migrants to the border from migrants to the United
States or whether in fact migration to the border is just a step in a process of
migration to the United States. After all, it would appear that factor comple-
mentarity should be relatively similar at the border, given its special trade
regime, to that in the United States.
Piore’s (1979, 35-43) dual labor market hypothesis for explaining the
functioning of the labor market and the logical role of the migrants in the
secondary sector serves to explain the role of migrants in the U.S. labor mar-
ket and may also serve to explain the specific case of the Mexican assembly
plants. These plants grew out of an external shock, which in this case was the
end of the Bracero Program. This event may have been perceived by some
U.S. plants as a new need for exporting at least part of the secondary labor
requirements to Mexico and by Mexico as a need of capturing those same
jobs.
For these reasons, foreign technicians and managerial personnel (primary
labor market types) are allowed to reside in Mexico. Also, foreign entrepre-
neurs, originally only from the United States, established twin plants, (Bolin
1964) with capital-intensive processes in the United States and labor-intensive
ones in Mexico, thus minimizing costs and maximizing managerial functions
(a sort of intraindustry Heckscher-Ohlin production scheme). Although there
are a number of twin plants, in many cases the U.S. home office is actually a
109 The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
long distance away from the border (in 1978, forty-eight of the Fortune 500
companies had maquiladoras in Mexico; see Grunwald and Flamm 1985), but
there is an actual trend, which includes growing numbers of Japanese compa-
nies in the United States (like Sony), toward the twin plant concept of produc-
tion.
Table 3.5 gives an idea of how labor markets could be complementary, both
through Mexican migration to the United States and at the border, and tends
to confirm a dual labor market hypothesis type of relation. Some differences,
however, are apparent. While farm workers are the most important segment
of migrants to the United States and white-collar workers represent only 4.5
percent of these workers, the most important segment of border migrants was
blue collar, followed by service workers, and the percentage of white-collar
workers was more than twice that of return migrants.
Additional differences between flows to the United States and border areas
are shown in table 3.6, which compares the sex and education composition of
migrants to the United States, migrants returning from the United States, and
migrants to Northern Mexico. While migrants going to and returning from the
United States were men in 83 and 80 percent of the cases, respectively, mi-
grants to the border were mainly women (5 l .7 percent). The fact that migrants
to the border are primarily females may reflect the particular demands of the
maquiladora industry, which employs mostly women.
The different patterns of migration also reveal different levels of education.
Although the level of education is in general very poor in all cases-at least
50 percent of male and female migrants are virtually illiterate-a higher pro-
portion of migrants to the border have junior high and high school educations.
This fact could be conditioned by the relatively higher employment require-
ments that exist in the commercial and assembly plant sectors as well as by
the presence of a higher proportion of white-collar workers (see table 3.5).
Table 3.5 Occupational Distribution of Border and Return Migrants Compared
to U.S. Workers
Population Return Migrants
of Mexican from the Border
Occupation United States Southwest Origin United States Migrants
Total 100.0 100.0 100. 1 100.0 100.0
White-collar workers 50.0 53.8 28.4 4.5 9.4
Blue-collar workers 33.4 30.6 50.6 35.8 39.7
Service workers 13.6 13.2 16.3 23.5 28.4
Farm workers 3.0 2.4 4.8 36.1 22.5
~
Sources: First four columns taken from Garcia y Griego and Zazueta (1982, 81). Column 1: U.S.
Bureau of the Census, Statistical Abstract of the United States, 1979. Column 2: U.S. Bureau of
Labor Statistics, Geographic Profile of Employment and Unemployment: States 1978, Metropol-
itan Areas 1977-78, September 1979. Column 3: U.S. Bureau of the Census, Current Population
Reports, Series p. 20, no. 339, June 1979. Columns 4 and 5: Encuesta Nacional de Emigracdn
a la Frontera Norte del Pais y a 10s Estados Unidos, 1978-79 (CENIET 1984).
110 Juan Diez-Canedo R.
Table 3.6 Migrants in the United States and Return Migrants from the United
States, by Sex and Education
Without
Formal Junior High
Total Education Elementary High School College
Migrants in the United
States
Percentage by sex
Percentage male
Percentage female
Percentage by sex
and education
Percentage male
Percentage female
Return migrants
Percentage by sex
Percentage male
Percentage female
Percentage by sex
and education
Percentage male
Percentage female
Migrants to the
northern border
Percentage by sex
Percentage male
Percentage female
Percentage by sex
and education
Percentage male
Percentage female
503,803
100.0
83.0
17.0
100.0
100.0
100.0
465,766
100.0
80.0
20.0
100.0
100.0
100.0
532,802
100.0
48.3
51.7
100.0
100.0
100.0
302,936
100.0
85.5
14.5
60.1
61.9
51.4
308,672
100.0
83.3
16.7
66.3
69.0
55.3
266,623
100.0
48.7
51.3
50.0
50.5
49.6
148,862
100.0
79.2
20.8
29.5
28.2
36.1
107,415
100.0
75.8
24.2
23.1
21.9
27.8
138,749
100.0
46.4
53.6
26.0
25.0
27.0
41,770 9,434
100.0 100.0
17.2 87.4
22.8 12.6
8.3 1.9
7.7 2.0
11.1 1.4
33,598 13,907
100.0 100.0
67.1 76.4
32.9 23.6
1.2 3.0
6.1 2.9
11.8 3.5
91,203 30,833
100.0 100.0
47.5 52.0
52.5 48.0
17.1 5.8
16.8 6.2
17.4 5.4
80 1
100.0
95.5
4.5
.2
.2
.o
2,174
100.0
32.8
67.2
.5
.2
1.6
5,394
100.0
61.2
32.8
1 .o
1.4
.6
Nore: Some migrants did not answer the schooling question.
All the figures given above seem to confirm the secondary labor market char-
acteristics of the three patterns of migration.
3.3 The Determinants of Migration to the United States and the
Northern Mexican Border
In the Survey of Migration to the United States and to the Northern Border
(Garcia y Griego and Zazueta 1982), data for the population aged fifteen years
old and over and for the employed labor force are recorded according to dif-
ferent migratory patterns: migration to the United States (USM); return migra-
tion from the United States (RMUS); migration to the north border area
(MNB); internal migration (IM); and nonmigration (NM). For each migratory
pattern, the information was available aggregated in relation to socioeconomic
111 The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
and demographic factors such as region, age, sex, marital status, education,
origin of the population, employment status, occupational status, and the eco-
nomic sector in which the individual was occupied (see the Appendix).
The available information did not make it possible to analyze all the popu-
lation characteristics through one econometric model. Because information
was available only through specific tabulations, the data had to be analyzed
through six different models for each migratory pattern. Information was
grouped in tables that contained three nonvarying characteristics and a fourth
one that varied. For instance, one group included region, age, and sex as the
nonvarying characteristics and marital status, schooling, or whether the ori-
gin was urban or rural, for example, as the fourth, variable, characteristic.
Furthermore, the way in which the information was classified made it virtu-
ally impossible to include additional explanatory variables such as income
differentials, differences in financial resources, investment, distance, etc. be-
cause the regions defined in the survey included groupings of states and, in
some cases, one region included only one part of one state.
Ideally, information at an individual level should have been used for the
analysis. Unfortunately, however, such data disaggregation could not be ob-
tained. As was mentioned before, available data were grouped and cross-
classified in tables. In such cases, a generalized linear model can be defined
for categorical data in which the observations consist of counts of frequencies
in the cells of a contingency table formed by the cross-classification of depen-
dent (or response) and explanatory (or independent) variables. In this contin-
gency table method, a log-linear model was specified and fitted to the different
migratory patterns (USM, RMUS, MNB).
For the purpose of estimation, a Poisson distribution was assumed. Since
this involves unrestricted independent random variables with distributions
from the exponential family, the Newton-Raphson estimation procedures im-
plemented in GLIM (Numerical Algorithms Group, Oxford) were used to es-
timate the parameters (see Nelder 1974; and Dobson 1983, 99). (The GLIM
estimates are maximum likelihood estimates.)
The responsive variables of the log-linear models were standardized in the
following form: each number in a contingency table cell was divided by
the total number of related cells. The following examples illustrate this point.
The number of 15- to 29-year-old single males who migrated to the United
States from region I was divided by the total number of 15- to 29-year-old
single males of region I, and the number of 30- to 44-year-old married women
who migrated to the northern border from region I1 was also divided by the
total number of 30- to 44-year-old married women in region 11.
For each migratory characteristic, the following log-linear models were
specified:
(1) mrlasc = K, + R, + A, + S, + CS,,
(2) mzJasn = K , + R, +Aa + S, + U,,
112 Juan Diez-Canedo R.
(3)
(4)
( 5)
(6)
where, for example,
mrJaro = K , + R, + A, + S, + OP,,
m,,,sp = K, + R, + A, + S, + Sch,,
mlJOsq = K, + R, + A, + S, + ES,,
mllasr = K , + R, + Aa + S, + OC,,
h= I
where
M =
i =
K =
R =
A =
S =
cs =
U =
OP =
Sch =
ES =
oc =
a =
j =
s =
c =
n =
P =
4 =
r =
o =
understandardized information;
the different migratory patterns: USM, RMUS, MNB, and internal
and nonmigration (which have been pooled);
mean or constant;
region;
age;
sex;
marital status;
type or origin;
occupational status;
schooling;
economic sector;
type of job;
region I, 11, 111, IV, or V;
age groups: 15-29, 30-44,45-49, 60 and over;
male, female;
single, married;
rural, urban;
employed, unemployed, not in the labor force;
schooling: less than elementary, elementary, high school, college;
economic sector: primary, secondary, tertiary;
self employed, blue collar, day laborer, unpaid family member,
ejidutario, l andh~l der. ~
3.4 Results
For simplicity, tables 3.7 and 3.8 summarize the results of these estimates
in terms of the estimated coefficients for analyses that treat age, region, sex,
and either rural origin or occupational status, with appropriate interactions.
The total population fifteen and over is the base for the calculations when rural
origin is included, while the economically active population is the base when
113 The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
Table 3.7 Estimates of Log-Linear Main and Interaction Effect Parameters for
Migration to the United States (Ml)
Coefficient SE T
Mean (K)
Region ( R) :
I
I1
I11
IV
V
Age (A):
15-29
3044
45-59
Sex (S):
Male
Female
Origin (0:
Urban
Rural
Region I Urban
Region I1 Urban
Region 111Urban
Region IV Urban
Interaction, Region-Origin:
Mean ( K)
Region ( R) :
I
I1
I11
IV
V
Age (A):
15-29
3044
45-59
Sex (S):
Male
Female
Self-employed
Blue-collar
Day laborer
Unpaid family
member
Ejidatario
Landholder
Occupation status (OP):
By Region, Age, Sex, and Origin
-1.145
2.804
2.888
3.161
2.064
. . .
1.684
1.612
1.372
1.619
. . .
.593
- ,6263
- 1.633
- 3.24
- ,9778
,5918
,5563
,5550
,5514
.5739
,235 1
,2364
,2417
,1487
,675
,7018
,7208
,8041
,7413
-7.0*
5.0*
5.2*
5.7*
3.6*
7.2*
6.8*
5.7*
10.9*
.9
- .9
-2.3*
-4.0*
- 1.3
By Region, Age, Sex, and Occupation
- 1.754 ,6228 - 2.8*
2.797 .526 5.3*
1.859 .549 3.4*
2.531 ,5306 4.8*
1.438 ,568 2.5*
. . .
1.695 .3342 5.1*
1.755 ,3327 5.3*
,8985 ,3644 2.5*
- 1.774 s605 -3.2*
. . .
- ,7971 ,3829 -2.1*
- ,583 ,3566 -1.6*
,302 .2815 1. 1
- 6.087 4.166 - 1.5*
-1.334 . ,4668 -2.9*
(continued)
114 Juan Diez-Canedo R.
Table 3.7 (continued)
Coefficient SE T
Interaction, occupational
status-sex:
Self-employed, male 2.572 ,6788 3. 8*
Blue-collar, male 2.4 .6629 3.6*
Day laborer, male 1.542 ,6247 2.5*
Unpaid family 5.313 4.267 1.2
Ejiduturio, male ,7819 ,9761 .8
member, male
Note: Goodness-of-fit deviance = 29.1, x 2 = 90.53. The model explains 87.3 percent of the
mean deviance of the null model.
*Significant at the 5 percent level.
occupation is included. Computations for other classifications gave similar
results.
Table 3.7 records estimates of the effect of the various factors on migration
to the United States. It shows that region is an important determinant of mi-
gration to the United States, with residence in the border region having nearly
as significant an effect in increasing the probability of migration to the United
States as residence in region 111, which comprises the states that are com-
monly reported as the source of migration to the United States since the turn
of the century. The implication is that the border has become an important
stepping-stone for migration to the United States, controlling for other differ-
ences. Note further that, in the calculations with urban (as opposed to rural)
origin included, that factor does not enter significantly for the border area but
does enter for regions I1 and 111. With respect to occupation, the surprising
result is the relative weakness of the occupational variables, which produced
a poorer fit than did other classifications of the data. While male day laborers
and blue-collar workers are especially likely to migrate to the United States,
the stereotypes of the Mexican immigrant as an unskilled farm worker seems
exaggerated on the basis of this calculation. The fact that male agricultural
migrants work in areas close to the Mexican border, where about 90 percent
of the Border Patrol is located, and that workers from regions I1 and I11 are
more likely to be of rural origin makes it especially likely that they are cap-
tured by the Border Patrol, leading to the view that the vast bulk of migrants
are agricultural laborers. In fact, while migrants to the United States are less
educated than other Mexicans (see table 3.6), they are not overwhelmingly
farm laborers (our results are consistent with Borjas, Freeman, and Lang’s [in
this volume] findings from the U.S. Census). Finally, age coefficients show
that the population group 30 to 44 years old was about as significant for ex-
plaining migration to the United States as the 15-29 age group.
Table 3.8 presents my estimates of the effect of the factors on being a return
migrant from the United States and on being a migrant to the Mexican border.
115 The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
Table 3.8 Estimates of Log-Linear Main and Interaction Effect Parameters for
Migration Models
Coefficient SE T
Return Migration from the United States
Mean (K)
Region (R):
I
I1
I11
IV
V
Age (A):
15-29
30-44
45-59
Sex (S):
Male
Female
Single
Manied
Marital status (CS):
- 2.736 ,4286 - 6.4*
2.301 ,3760 6.1*
1.867 ,3850 4.8*
2.251 .3770 6.0*
1.502 ,3960 3.8*
1.155 ,2290 5.0*
1.115 .23 4.8*
,871 .238 3.7*
1.424 ,1630 8.7*
. . .
- .419 ,132 -3.2*
. . .
Migration to the Border Areab
Mean (K)
Region (R):
I
I1
111
IV
V
Age (A):
15-29
30-44
45-59
Sex (S):
Male
Female
Single
Married
Marital status (CS):
-1.162 ,3633 - 3.2*
4.492 ,359 12.5*
1.727 ,387 4.5*
1.613 ,391 4.1*
1.987 .38 5.2*
- ,471 ,103 -4.6*
-.186 ,095 - 2.0*
- .01 .09 - . l
- ,087 .069 - 1.3
- ,223 ,069 -3.2*
*Significant at the 5 percent level.
a Goodness-of-fit deviance = 48.6, x2 = 90.5. The model explains 76.3 percent of the mean
deviance of the null model.
b Goodness-of-fit deviance = 33.3, x2 = 90.5. The model explains 95.8 percent of the mean
deviance of the null model.
116 Juan Diez-Canedo R.
The return migrant calculations show that the probability of being a return
migrant is highest for the border region, again indicating that residence in the
border has become an important factor in explaining migration to the United
States. In general, the coefficients for the determinants of return migrants are
similar to those for migrants still in the United States, implying that I am
identifying roughly comparable populations. The highest proportion of return
migrants were aged 15 to 29 years, followed by the groups aged 30 to 44 and
45 to 49. Note also that being male raises the probability of having migrated
to and returned from the United States. Models that included rural or urban
origin, schooling level, occupational status, and economic sector of activity
were not useful for explaining this migratory pattern.
By contrast, the patterns of migration to the northern border differ consid-
erably from those observed for migration to the United States. First, sex is not
relevant for explaining this type of flow, therefore indicating that this migra-
tion is mostly of a permanent nature, as opposed to the temporary one shown
in the results for migrants to the United States and return migrants. Second,
the most important regions for this migratory flow are I (the border itself) and
IV (the region adjoining the border). The increasing flow of migration from
the adjoining counties happens first as a daily trip to the border (the transpor-
tation business is booming), which apparently becomes permanent after a
while. The older age group (60 and over) was the most important in relation
to this migratory flow. Another interesting fact is that, although not signifi-
cant, the proportion of women is more important than men in this migratory
process, a fact that is explained by the existence of a higher proportion of
women in the assembly plants.
Examined together, tables 3.7 and 3.8 suggest an interesting pattern of
interrelated migration to the border and to the United States. The different
effect of sex on migration in the calculations can explain where males go when
their wives work at the border. Since an important part of the work force, at
least in maquiladoras, is female, the ideal overall migration strategy may be
accomplished by a joint family decision where males migrate to the United
States and females migrant to the border. This hypothesis is backed by the fact
that married males who had migrated to the United States had the highest
probability of being return migrants.
In conclusion, it can be said that the 1964 termination of the Bracero Pro-
gram conditioned a policy response that gave birth to a very successful border
industrial venture. This venture has created an important number of jobs of a
secondary market type, has attracted foreign and local capital, and has condi-
tioned, through forward and backward linkages, growth in the border area
while the rest of the country was, in the period under study, in the middle of a
protracted recession. However, another important fact that should be noted is
that migrants, whether to or from the United States or to the border, compose
only a negligible portion of the total Mexican work force. Evidence found in
this research also suggests that migration to the northern Mexican border may
117 The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
indeed be a step toward migration to the United States. Given that migration
to the border seems to entail complete family units, that assembly plants em-
ploy mostly women, and that married males are an important element for
explaining migration to the United States from the northern border, this pat-
tern of migration may indeed be an optimal labor market decision for family
units. From this perspective, one cannot understand Mexican migration to the
United States separately from migration to the border.
Appendix
The population aged 15 years old and over was recorded for each migration
pattern, in various contingency tables formed by the cross-classification of
four different independent variables each with two or more categories:
Contingency Table 11.1. Region, age, sex, and education.
Contingency Table 11.2. Region, age, sex, and origin.
Contingency Table 11.3. Region, age, sex, and marital status.
Contingency Table 11.4. Region, age, sex, and employment status.
Similarly, the employed labor force for each migratory pattern was cross-
classified in accordance with four independent variables, each with two or
more scales:
Contingency Table 11.5. Region, age, sex, and occupation status.
Contingency Table 11.6. Region, age, sex, and economic sector.
The survey divided the territory into five different zones according to the
Region I. The northern border area.
Region 11. The states of Jalisco, Colima, Guanajuato, Michoach, part of
Guerrero and the state of Mexico, and Ensenada in Baja Cali-
fornia Norte.
Region I l l . The states of Aguascalientes, Durango, Nayarit, Zacatecas,
Querktaro, San Luis Potosi, and part of the state of Hidalgo.
Region IR The states of Tamaulipas, Nuevo Le h , Coahuila, and Sonora
(exempting the border area), plus the state of Sinaloa.
Region K The states of Baja California Sur, Campeche, Chiapas, the
Federal District, Oaxaca, Quintana Roo, Tabasco, Veracruz,
Yucathn, Morelos, Puebla, Tlaxcala, and part of the states of
Hidalgo, Mexico, and Guerrero.
density flow of migrants (see fig. 3.1):
The dichotomous variables available are
a) Male or female.
b) Urban or rural population.
118 Juan Diez-Canedo R.
c) Marital status-single (includes single, widowed, divorced and sepa-
rated) or married (including those who live in free union).
The age variable is divided into four ranks:
a) fifteen to twenty-nine years old.
b) thirty to forty-four years old.
c) forty-five to fifty-nine years old.
d ) sixty years old and over.
The education variable is divided into four parts:
Less than elementary. Those who are illiterate or did not finish elemen-
Elementary. Those who finished elementary school but not junior high
Junior high school, Those who finished junior high school but not senior
Senior high school. Those who finished senior high school but not col-
College.
tary school.
school or the equivalent.
high school or the equivalent.
lege.
The employment status variable included three categories:
a) Employed labor force.
b) Unemployed labor force.
c) Not in the labor force.
The occupational status is divided into six categories:
a) Self-employed.
b) Blue collar.
c) Day labor.
d ) Unpaid family member.
e) Ejidatario.
f ) Landholder.
The economic sector of occupation is divided into the following categories:
a) Primary sector.
b) Secondary sector.
c) Tertiary sector.
Notes
1. For a detailed explanation of available data on migration and biases in that data,
see Diez-Canedo (1980). The Secretm’a del Trabajo conducted the household survey
through the Centro Nacional de InformaciBn y Estudios del Trabajo (CENIET).
119 The Effect of Policy Restrictions on Capital and Labor Flows in Mexico
2. See the Appendix. For a detailed explanation of the methodological aspects of
this survey, see Garcia y Griego and Zazueta (1982).
3. At the time of the survey, except for the border areas, 100 percent ownership was
allowed only a very few cases; normally, a 51 percent Mexican share was required.
Recently, however, this restriction on foreign investment has been relaxed drastically.
4. An ejidatario is an ejido’s “landowner,” although the ejidos cannot be sold or
mortgaged. The ejido is a plot of land owned by the nation through a community of
“ejidatarios. ”
5. Variables in contingency table analyses are described that are not reported in
tables.
Bibliography
Amemiya, Takeshi. 1981. Qualitative Response Models: A Survey. Journal of Eco-
nomic Literature 19: 1483-1536.
Bolin, R. 1964. Industrial Opportunities for Ciudad Juhrez. Report prepared by Arthur
D. Little Consulting Firm, 1 August.
Diez-Canedo, J. 1980. A New View of Mexican Migration to the United States. Ph.D.
diss., Massachusetts Institute of Technology.
Dobson, Annette J. 1983. An Introduction to Statistical Modeling. London and New
York: Chapman & Hall.
Flamm, K. 1984. The Volatility of Offshore Investment. Journal of Development Eco-
nomics, 16-34.
Garcia y Griego, M., and C. Zazueta. 1982. Los Trabajadores Mexicanos en Estados
Unidos. Mexico City: Secretaria del Trabajo, CENIET.
Greenwood, M. J. 1978. An Econometric Model of Internal Migration and Regional
Economic Growth in Mexico. Journal of Regional Science 18( 1).
Greenwood, M. J. , and J. Ladman. 1977. Economia de la Movilidad Geogr&ca de la
Mano de Obra en MCxico. Demografia y Economi’a 1 l(2).
Grunwald, J., and Kenneth Flamm. 1985. The Global Factory. Washington, D.C.:
Brookings.
Hansen, Niles. 1985. The Nature and Significance of Border Development Patterns. In
The U S . and Mexico: Borderland Development and the National Economies, ed. L.
Gibson and A. Corona, 3-17. Boulder, Colo.: Westview.
Isoard, C. 1976. Inter-regional Migration in Mexico. University of Chicago, Depart-
ment of Economics. Typescript.
Nelder, J. A. 1974. Log-linear Models for Contingency Tables: A Generalization of
Classical Least Squares. Applied Statistics 23:323-29.
Piore, M. J. 1979. Birds of Passage. Cambridge: Cambridge University Press.
Tamayo, J. 1981. Algunas Ideas Inexactas Acerca de la Frontera Norte de Mtxico.
Tellez, L. 1986. Essays on the Financial and Real Aspects of an Open Economy: The
Mexico City: Centro de Investigacibn y Docencia Econhi ca. Mimeo.
Case of Mexico. Ph.D. diss., Massachusetts Institute of echnology.
This Page Intentionally Left Blank
4 Internal Migration of
U. S . Immigrants
Ann P. Bartel and Marianne J. Koch
In a recent study of Indochinese refugees who arrived in the United States in
1975, after the fall of Saigon, Baker and North (1984) observed that 45 per-
cent of these individuals lived in a different state in 1980 than in 1975, com-
pared to 9 percent of the U.S. population. This finding is interesting not only
because of the sheer magnitude of the migration but also because of its effect;
the result of the relocation was an increased concentration of the refugees in a
small number of states. This particular group of immigrants constitutes a very
special case in that they all came as refugees from the same part of the world
and in that they did not choose their 1975 locations but were placed there
under a program operated by the Inter-Agency Task Force for Indochinese
Refugees. Whether these findings generalize to other groups of immigrants
that have recently arrived in the United States is the subject of this paper.
Specifically, we ask how dispersed U.S. immigrants are on their arrival to this
country and whether they change locations as time in the United States
elapses. Further, we explore the determinants of the internal migration of U.S.
immigrants and consider the effect of mobility on their earnings.
The study of the internal migration of immigrants is important in light of
the expanding numbers of immigrants to this country. The ability to predict
which areas of the country will receive immigrants will aid in planning for the
provision of services to local populations. Further, if it is the case, as it was
shown to be with the Indochinese refugees, that immigrants tend to cluster
where their countrymen are located, planners in these cities may need to de-
sign bilingual or even multilingual programs to implement the delivery of ser-
vices; this may, in turn, lead to a more permanently segregated U.S. society.
Ann P. Bartel is professor of business at Columbia University’s Graduate School of Business
and a research associate of the National Bureau of Economic Research. Marianne J. Koch is
assistant professor of management in the College of Business Administration at the University of
Oregon.
121
122 Ann P. Bartel and Marianne J. Koch
In this paper, we investigate the internal mobility patterns of Asian, Central
and South American, and European immigrants who arrived in the United
States after 1964. In section 4.1, we study the extent to which they disperse
throughout the country as time in the United States elapses and compare their
inter-SMSA (standard metropolitan statistical area) mobility rates to those of
the native population. Section 4.2 presents the results of estimating an econ-
ometric model of the determinants of changing SMSAs. A brief analysis of
the effect of internal migration on earnings is discussed in section 4.3, and
conclusions and policy implications are presented in section 4.4
4.1 Mobility Patterns
In this section, we present statistics that can be used to describe the internal
mobility patterns of recent immigrants to the United States. We restrict our
analysis to SMSAs that have at least ten male immigrants aged 22-64 in the
Public Use B Sample of the 1980 Census of Population. This resulted in fifty-
three SMSAs. Since 90% of all the immigrants in the Census files (and 43
percent of the U.S. population) live in one of these SMSAs, this is not a very
restrictive selection rule.’
The “new immigrant” population-or immigrants arriving in the United
States after 1964-is stratified into three cohorts based on year of arrival in
the United States: arrivals between 1965 and 1969, 1970 and 1974, and 1975
and 1979. The individuals in each of these cohorts are restricted to be of work-
ing age at the time of their arrival in the United States; hence, the most recent
arrivals, those who came between 1975 and 1979, are aged 22-54 in 1980,
while those who came between 1970 and 1974 are aged 27-59 in 1980, and
the 1965-69 arrivals are aged 32-64 in 1980.
One way to describe the internal mobility patterns of recent immigrants to
the United States is to calculate a Herfindahl index for the various samples of
immigrants that we wish to study; this statistic provides a summary measure
of the extent to which each of the groups is geographically dispersed through-
out the United States.Z Higher values of the index represent greater geographic
concentration. In table 4.1, Herfindahl indices are shown for three main ethnic
groups-Asians, Central and South Americans, and Europeans-and for sev-
eral subcategories in each of these groups. The Herfindahl indices for the most
recent cohort (i.e., the 1975-79 arrivals) are shown in column 1, for the
middle cohort (i.e., the 1970-74 arrivals) in column 2, and for the earliest
group (i.e., the 1965-69 arrivals) in column 3. In order to appreciate the
magnitude of these indices, we also calculated the Herfindahl index for
the total population in the fifty-three SMSAs in 1980; as shown in the note to
the table, this statistic is .04. Numbers in excess of .04 indicate that the group
under study is more concentrated than the total population in this sample of
cities.
123 Internal Migration of U.S. Immigrants
Table 4.1 Herlindahl Indices for 1980 Geographic Distributions of Various
Immigrant Cohorts
(1) (2) (3)
Amved 1975-79 Amved 1970-74 Arrived 1965-69
(aged 27-59 in 1980) (aged 22-54 in 1980) (aged 32-64 in 1980)
Asians:
China
India
Japan
Korea
Philippines
Vietnam
Central and
South Americans:
Cuba
Mexico
Other
Europeans:
England
Italy
Greece
.09
.17
.09
.15
. I 8
.14
.08
.13
.40
.20
.17
.07
.05
.09
.14
.09
.I6
.08
.18
.I1
.I1
.I6
.14
.34
.24
.24
.10
.05
.19
.I2
.09
. I5
.07
.I9
.I1
.I2
.50
.I4
.36
.26
.26
.08
.06
.15
.14
Nore: The Herfindahl Index for the total U.S. population in the fifty-three cities in the sample
is .04.
The data in table 4.1 can be used to answer the question whether immi-
grants become more geographically dispersed as they acquire experience in
this country. The degree of geographical dispersion is one indicator of assim-
ilation into the new country (Massey 1981). To the extent that immigrants are
able to learn about opportunities in other parts of the country as time in the
United States elapses, we would expect to observe greater dispersion over
time or evidence of assimilation. In terms of table 4.1, we would expect to
observer smaller values in column 3 as compared to column 2 and, certainly,
as compared to column 1. A possible offsetting factor is that immigrants may
move from their initial destinations in the United States only to discover that
they are unable to live without the support of their ethnic enclaves and then
return to their SMSAs of initial destination. If this happens, we would observe
that the degree of dispersion of the three cohorts in the 1980 cross section
would be very similar.
We begin our discussion of table 4.1 by looking at the three main ethnic
groups. Two important findings emerge. First, the immigrants, and especially
the Central and South American immigrants, are more geographically concen-
trated than the U.S. population in the fifty-three cities. Second, for each of the
main ethnic groups, there is no evidence of increased dispersion over time.
Since the country composition of each of the main ethnic groups is likely to
124 Ann P. Bartel and Marianne J. Koch
have changed over time, the constancy of the Herfindahl indices across co-
horts need not imply that there has been no dispersion for a given subcategory.
For example, subgroups that are more dispersed on arrival to the United States
may account for a larger proportion of the main ethnic group that has recently
arrived; in table 4.1, this could mask the dispersion over time of other
subgroups that represented a large share of the early cohorts. Hence, we also
study the trend in the Herfindahl index for selected subgroups. There are only
two cases, namely, the Koreans and the Cubans, in which we see greater geo-
graphic dispersion of the cohorts that arrived earlier.3 In sum, the evidence of
table 4.1 gives only limited support for the hypothesis that, as time elapses in
the United States, the immigrants will become more dispersed throughout the
country.
Of course, one of the problems with table 4.1 is that it is calculated from
cross-sectional data on immigrants who were living in the United States in
1980. We have attempted to interpret these data as a pseudo-panel in order to
draw a conclusion about changes in geographic dispersion over time. Actu-
ally, this conclusion can be based only on data for a given sample of immi-
grants who are observed at more than one point in time. Since 50% of the
individuals in the Public Use Sample of the 1980 Census of Population were
asked where they lived in 1975, it is possible to create a panel for this group.
We calculated the Herfindahl indices for these individuals first for their 1975
locations and then for their locations in 1980. If the dispersion-with-time hy-
pothesis is correct, then we should observe a decrease in the Herfindahl index
between 1975 and 1980.
Table 4.2 reports these results for the 1965-69 and 1970-74 cohorts. When
the main ethnic groups are not disaggregated, the Herfindahl indices hardly
change between 1975 and 1980. For certain subgroups, however, we do find
evidence of increased geographic dispersion between 1975 and 1980, and, to
provide an arbitrary benchmark, we have indicated with an asterisk those
groups for which there is at least a 20% fall in the Herfindahl index over the
five-year period. These groups are entirely from the Asian category, namely,
the immigrants from India, Japan, and Korea. We had inferred from table 4.1
that the Cubans dispersed over time, but, as table 4.2 clearly shows, this con-
clusion was erroneously based on our interpretation of a cross section of co-
horts as a pseudo-panel. Relying on table 4.2 as the more correct picture of
dispersion, we conclude that geographic dispersion over time is not a typical
characteristic of the post- 1964 immigrants to the United States.
Next we ask, although the new immigrants are not dispersing throughout
the country, are they moving at all, or are they remaining in their original
destination SMSAs? In other words, it is possible that these individuals are
moving between SMSAs but that the degree of dispersion of the group is not
changing; that is, person A is moving from city 1 to city 2 while person B is
moving from city 2 to city 1. The percentages of various immigrant and ethnic
125 Internal Migration of U.S. Immigrants
Table 4.2 Herfindahl Indices in 1975 and 1980
Arrived 1970-74 Arrived 1964-69
(aged 27-59 in 1980) (aged 32-64 in 1980)
1975 1980 1975 1980
Asians:
China
India
Japan
Korea
Philippines
Vietnam
Central and
South Americans:
Cuba
Mexico
Other
Europeans:
England
Italy
Greece
. l l
.22
.13
.14
. I4
. I2
.22
.16
.32
.27
.30
. I1
.08
.14
.15
.10
.19
. lo*
.11*
.lo*
. I 1
.22
.15
.36
.26
.28
. I 0
.07
. I 3
.13
.09
. I6
.10
.25
.I5
.I4
. . .
. I3
.32
.27
.25
.10
.07
.22
. I4
.09
.I5
.08*
.19*
.12*
.14
. . .
.13
.36
.27
.23
.G9
.08
.18
.14
Nore: Herfindahl indices are calculated for the sample of immigrants that resided in one of the
fifty-three SMSAs in both 1975 and 1980. See text.
*The Herfindahl index decreased by at least 20% between 1975 and 1980.
native groups who moved between one of the fifty-three SMSAs in our sample
during the period 1975-80 are shown in table 4.34 As before, the results are
specific to the particular ethnic group.
We find that, although Asian immigrants are between two and three times
as likely as natives of Asian ethnicity to change SMSAs, the Central and
South American and European immigrants are less mobile than their native
counterparts. Among the Asians, the Indians and Koreans stand out as the
most mobile, and these were two of the three groups for which dispersion over
time was observed.
The major conclusion from our analysis of the data in tables 4.1-4.3 is that
there is little systematic evidence of dispersion of the immigrants throughout
our sample of cities. This is true even for the Asian ethnic group whose inter-
SMSA mobility rate greatly exceeds that of any of the native groups. In other
words, immigrants do move between SMSAs during their first five or ten
years in this country, but this mobility does not substantially affect the index
of dispersion, except in a few isolated cases. In the next section of the paper,
we describe and estimate a logit model that can explain why the immigrants
in our sample did change SMSAs between 1975 and 1980.
126 Ann P. Bartel and Marianne J. Koch
Table 4.3 Percentage of Immigrants and Natives Changing SMSAs between
1975 and 1980 (sample sizes in parentheses)
(1) (2) (3)
Arrived 1970-74 Arrived 1965-69 Natives
(aged 27-59 in 1980) (aged 32-64 in 1980) (aged 27-64 in 1980)’
Asians:
China
India
Japan
Korea
Philippines
Vietnam
Central and
South
Americans:
Cuba
Mexico
Other
Europeans:
England
Italy
Greece
7.4 (838) 10.6 (1,965)
4.0 (253)
9.5 (380)
8.4 (439) 14.0 (1,275)
12.1 (33)
7.8 (102)
5.1 (59)
7.8 (205)
Nore: Percentages are. calculated for the sample of immigrants that resided in one of the fifty-three
SMSAs in both 1975 and 1980.
aThe natives are disaggregated into three categories: (1) Asian ethnicity; (2) Central and South
American ethnicity; and (3) non-Hispanic whites or European ethnicity.
4.2 Determinants of Internal Migration
In this section, we describe and estimate an econometric model of the de-
terminants of the internal migration of recent immigrants to the United States.
In section 4.2.1, the model is described and the variables we utilize defined.
Section 4.2.2 presents the results for the immigrants and compares them to
the results for natives of similar ethnicity as well as native whites.
4.2.1 Econometric Framework
Beginning with Sjaastad (1962), economists have argued that an individual
changes location within a country if the discounted net gain from moving is
positive. In other words,
(1) PM, = f ( G, ) ,
where PM, is the probability that the individual moves in time period t, and
G, is the discounted net gain from moving. G, can be written as follows:
G, = Y,! - Y, - C,,
127 Internal Migration of U.S. Immigrants
where YT is the present value of the expected real income stream if the indi-
vidual migrates in time period t, Y, is the present value of the expected real
income stream in the current location calculated at time t , and C, are the costs
of migration. In order to study the determinants of the probability of moving,
those variables that measure the discounted net return from moving must be
identified. Greenwood’s 1975 survey of the literature on geographic mobility
and articles on this subject published after 1975 (e.g., Bartel 1979; and Fields
1979) show that economists have used information on the individual’s char-
acteristics and the characteristics of the area in which he or she resides at the
beginning of the period under study as proxies for the components of G, .
We follow this procedure in specifying a model of the determinants of the
1975-80 inter-SMSA movement of immigrants who arrived in the United
States between 1965 and 1974. Since our sample is obtained from the Public
Use Sample of the 1980 Census of Population, information on personal char-
acteristics as of 1975 is rather limited. We do have three variables in this
category: (1) the individual’s age in 1975 (AGE) (2) his education in 1980
(EDUC); and (3) whether he reports the ability to speak English well or very
well in 1980 (SPEAKS). Although education is measured in 1980, since the
individuals in our sample were at least 22 years old at the time of immigration,
it is very unlikely that educational attainment changed in any systematic way
between 1975 and 1980. Similarly, SPEAKS is measured in 1980 and, if any-
thing, is an overestimate of some individuals’ true ability to speak English in
1975, so that its estimated effect on migration is an understatement of the true
effect.
We expect age to have a negative effect on the probability of changing
SMSAs since the time period over which to capture the discounted returns
from migration shrinks as the individual ages. Education should have a posi-
tive effect since, as it has been argued in previous work, the more educated
individual is better able to adapt to new locations and is more efficient in
searching for jobs in other locations. SPEAKS should have a positive effect on
internal migration since the immigrants who are more facile with the English
language will have better information on opportunities throughout the country
and will also be better able to adapt to a new location.
Five variables are used to describe the SMSA in which the individual re-
sided in 1975. They are (1) total population of the SMSA in 1975 (TPOP), (2)
the 1975 unemployment rate in the SMSA (UNRATE), (3) the logarithm of the
average wage in the city (LNWAGE), (4) the level of welfare benefits in the
SMSA in 1975 as measured by the logarithm of the average monthly general
assistance payment per recipient (LNGEN), and (5) the proportion of the
SMSA’s population in 1975 that is foreign born and of the same ethnicity as
the immigrant being studied (PFOR) (in other words, for an Asian immigrant,
PFOR is coded as the percentage of the SMSA’s 1975 population that was born
in Asia, for a Mexican immigrant, PFOR is the percentage of the SMSA’s 1975
population that was born in Mexico, etc.).
128 Ann P. Bartel and Marianne J. Koch
TPOP is obtained from the Statistical Abstract of the United States, UNRATE
is from the State and Metropolitan Area Databook, LNWAGE is calculated
from the data in the Public Use Sample, LNGEN is from the Public Assistance
Statistics, and PFOR is calculated from the published volumes of the 1980 Cen-
sus of Popul at i ~n. ~ TPOP is expected to have a negative sign since population
acts as a measure of job opportunities and general economic activity. UNRATE
should have a positive sign because, as the probability of finding a job in the
SMSA decreases, out-migration should occur. LNWAGE and LNGEN are pre-
dicted to have negative signs since they measure attractive characteristics of
the city. (LNGEN is likely to be correlated with the level of social services in
the city.) Finally, the location of fellow countrymen has been shown to be a
significant determinant of the settlement patterns of U.S. immigrants (see Bar-
tel 1989; and Dunlevy 1980). We would then expect to see a negative effect of
PFOR on the probability of leaving the 1975 SMSA; this effect should be
smaller for immigrants who arrived between 1965 and 1969 than for those
arriving later since time spent in the United States should weaken the attach-
ment to fellow countrymen.
4.2.2 Logit Results
Equation (1) is estimated on the sample of male immigrants who arrived in
the United States between 1965 and 1974, who reported their 1975 and 1980
U.S. residences as one of the fifty-three SMSAs that were defined in section
4.1, and who were between the ages of 22 and 54 when they arrived in the
United States.
Columns 1-4 in table 4. 4 report the results of estimating equation ( 1) on
the two immigrant cohorts. The logit technique is used to estimate these equa-
tions. Among the personal characteristics, EDUC and AGE have the hypothe-
sized signs, with the more educated and the younger immigrants being signif-
icantly more likely to change SMSAs between 1975 and 1980. Two ethnic
dummy variables, ASIAN and CSA (Central and South Americans), are used to
compare the migration rates of these groups to the excluded group, the Euro-
peans. In columns 2 and 4, these dummy variables are interacted with PFOR to
capture differences in responsiveness to the location of fellow countrymen.
We find that, even when the other personal characteristics are included in the
equation, the Asians who arrived in the 1970-74 wave are still significantly
more mobile than the other immigrants who arrived during that time interval.
The variables that measure relative economic opportunities in the cities
(UNRATE, LNWAGE, and LNGEN) do not have the effects that we had predicted.
In only one case, UNRATE for the 1970-74 cohort, do we find a significant
effect. The result that stands out in table 4. 4 is that for both cohorts PFOR is
negative and significant. As predicted, the immigrants are most likely to move
from those cities where their fellow countrymen constitute a very small per-
centage of the total population. This is precisely what Baker and North ob-
served for the Indochinese refugees who did not choose their 1975 locations
129 Internal Migration of U.S. Immigrants
lsble 4.4 Probability of Changing SMSAs between 1 9 5 and 1980 (&values in
parentheses)
(1) (2) (3) (4) ( 5) (6) (7)
White Ethnic
1965-69 Immigrants 1970-74 Immigrants Natives Natives Natives
(aged 32-64 in 1980)
(aged 27- (aged 27- (aged 27-
64 in 1980) 64 in 1980) 64 in 1980)
(aged 27-59 in 1980)
EDUC
AGE
SPEAKS
UNRATE
LNWAGE
LNGEN
TPOP
PFOR
ASIAN
CSA
.09
(4.35)
- .03
(-2.57)
- .01
( - .05)
.02
(.49)
- 1.53
(-1.17)
.21
(.62)
- .37
(-1.06)
- 6.45
( - 2.27)
.13
(.49)
- .03
(-.14)
PFOR*ASIAN
PFOR*CSA
N 1,611
.10
(4.64)
- .03
( - 2.63)
- .08
( - .40)
.02
(.45)
- 1.67
(-1.26)
.03
(.W
- .37
(-1.04)
13.33
(1.43)
.82
(1.54)
.10
(1.95)
- 10.71
(-.86)
-23.38
( - 2.38)
1.61 1
.09
(5.07)
- .02
(-2.34)
- . l l
( - .65)
.06
(1.62)
- 1.03
( - .94)
.20
(.67)
- .55
( - 1.91)
-6.94
( - 2.41)
.48
(2.10)
.02
(. 12)
2,049
.09
(5.11)
- .02
(-2.30)
- .I2
( - .71)
.of5
(1.53)
-1.14
(-1.00)
.15
- .56
(.51)
(-1.88)
4.52
(.50)
.97
(1.95)
.61
(1.25)
- 9.42
( - .80)
( - 1.40)
2,049
- 13.25
.15
(6.09)
- .03
(-4.24)
.17
(.98)
.13
(2.95)
2.65
(2.68)
- 1.12
( - 3.65)
.12
(4.34)
- .45
( - .23)
- .57
( - 2.52)
- .30
(-1.30)
2,824
.21
(4.58)
- .01
( - .55)
.97
(2.04)
.10
(1.17)
2.72
(1.48)
- 1.05
( - 2.22)
.02
6.42
033)
(2.11)
676
. l l
(3.93)
- .05
(-4.79)
.07
(.39)
.09
(1.51)
1.96
(1.59)
- .86
( - 1.90)
.20
(4.52)
-7.01
(-1.95)
- .23
(-1.11)
2,148
on their own. We had expected to see a weaker effect of this variable for the
immigrants who had spent more time in the United States as of 1975 (i.e., the
1965-69 arrivals), but their attachment to their fellow countrymen is just as
strong. In columns 2 and 4, we interact PFOR with the ethnic dummy variables,
ASIAN and CSA, in order to see if our ethnic groups are equally responsive to
this variable. For both cohorts, the results show that only the Central and
South Americans changed SMSAs in response to the presence of fellow coun-
trymen.
If the Central and South American immigrants are indeed moving from
those cities where PFOR is low, then we would expect to observe an increase
between 1975 and 1980 in the mean values of PFOR for those individuals who
move. In rows 1, 3, and 5 of table 4.5, we show the ratio of mean PFOR in
1980 to its mean in 1975 separately for the movers and the stayers in our
sample. The table also reports, for the movers only, the ratio of the 1980 mean
130 Ann P. Bartel and Marianne J. Koch
value of PFOR based on their 1975 locations, divided by the 1975 mean value.
In other words, this ratio, shown in rows 2 , 4, and 6, describes how the 1975
locations changed between 1975 and 1980. Table 4.5 confirms that PFOR in-
creased more between 1975 and 1980 for the Central and South American
movers than the stayers; the immigrants in this ethnic group who changed
locations in the United States experienced an increase in the concentration of
fellow countrymen, and, as shown by comparing rows 3 and 4, this increase
could only have occurred by changing SMSAs. Although these movers are
initially in cities with lower values of PFOR compared to the stayers, by 1980
the gap has narrowed.6 Consistent with the regressions in table 4.4, the results
for the other ethnic groups in table 4.5 show that a change of SMSA does not
increase the concentration of fellow countrymen; in fact, for the Europeans,
there is a decrease for both cohorts.
In columns 5-7 of table 4.4, we report the results of estimating equation
(1) on a sample of natives aged 27-65 in 1980. Column 5 includes white
natives and natives who report themselves of either Asian, Cuban, Mexican,
or other Hispanic ancestry. The purpose of this analysis is to compare the
behavior of the immigrants with that of individuals who were born and raised
in the United States and presumably have better information about the coun-
try.’ Although EDUC and AGE have the same effects for the natives as they did
for the immigrants, we do observe different effects for the city characteristics.
LNGEN is now negative and significant, indicating a greater sensitivity (or
awareness?) of the natives to welfare benefits and social services. But
LNWAGE actually has the wrong sign! The most interesting finding is that the
white natives (col. 6) are more likely to move from those cities with a high
concentration of foreign born (in col. 6, PFOR refers to alf ethnic groups),
while the ethnic natives, like the immigrants, prefer to stay in cities with a
high concentration of individuals from their country of ancestry.
Table 4.5 A Comparison of 1975 and 1980 Mean Values of PFOR for Movers
and Stayers
~
1965-69 Cohort 1970-74 Cohort
Movers Stayers Movers Stayers
Asians:
1. PFOR8O/PFOR75 1.46 1.67 1.76 1.72
2. PFOR~O (in 75 c i t y ) / ~ ~ 0 ~ 7 5 1.64 1.69
Central and South Americans:
3. PFOR80iPFOR75 1.63 1.19 1 .S3 1.32
4. PFOR80 (in 75 CitY)/PFOR75 1.28 1.27
5 . PFOR80/PFOR75 .83 1.10 .89 1.10
6. PFOR80 (in 75 CitY)/PFOR 75 .97 1.10
Europeans:
131 Internal Migration of U. S. Immigrants
4.3 The Effect of Internal Migration on Wages
In this section of the paper, we study the effects of a change in SMSA
between 1975 and 1980 on the 1980 wage rates of the immigrants who arrived
between 1965 and 1974. We also compare the results to similar wage equa-
tions that are estimated on our sample of natives. Previous work on migration
has not conclusively established that migrants necessarily experience an in-
crease in wages following a change in location.8
We examine the effect of internal migration on wages by estimating a
log(wage) equation on 1980 data that includes the standard variables such as
education, experience, experience squared, marital status, health status, and
a vector of ethnicity dummies. We then add a variable called MOVE, which
equals one if the individual changed SMSAs between 1975 and 1980. Since
the various ethnic groups are pooled in our sample, we also interact MOVE
with our two ethnic dummy variables and create MOVEASN and MOVECSA; the
comparison group is, therefore, the Europeans. We also distinguish two cate-
gories of moves, moving to an SMSA where PFOR is higher than it was in the
origin SMSA (MOVEUP) and moving to an SMSA where PFOR is lower than it
was in the origin SMSA (MOVEDOWN). We expect the coefficient on MOVEUP
to be less positive than the coefficient on MOVEDOWN for two reasons. First,
since immigrants prefer to live with fellow countrymen, compensating differ-
entials will reduce the return to MOVEUP. Second, crowding effects should
depress the wages of immigrants in cities with large foreign-born popula-
tions.
The results of estimating the wage equation are shown in table 4.6. Only
the coefficients on the mobility variables are shown; all the other variables had
the usual signs. The equations were estimated separately for the two immi-
grant cohorts in order to see if time spent in the United States influences the
returns to internal migration. In columns 1 and 4, we see that MOVE is in-
significant for both cohorts but that it is positive fro the earlier immigrants.
Distinguishing the returns to migration for the different ethnic groups is
important, as shown in column 2. Among the 1965-69 arrivals, only the Euro-
peans experienced a significant increase in wages when they changed SMSAs
between 1975 and 1980. The hypothesis regarding the relative effects of
MOVEUP and MOVEDOWN is confirmed for this cohort. MOVEDOWN has a posi-
tive and significant coefficient, while the coefficient on MOVEUP is insignifi-
cant. The fact that only 9% of the movers are in the MOVEDOWN category
suggests that this small group is unique in its ability to leave ethnic networks
and move to areas where their skills are more highly rewarded. In the case of
the 1970-74 arrivals, however, ethnicity plays no role; MOVE is insignificant
for all the groups. Finally, columns 7 and 8 show the effect of internal migra-
tion on the wages of natives. The results here conform to the findings of pre-
vious research, namely, an insignificant effect of geographic mobility on wage
rates.
132 Ann P. Bartel and Marianne J. Koch
Table 4.6 Effects of Changing SMSA between 1975 and 1980 on Log (1980
wage rate) of Immigrants and Natives
1965-69 Arrivals 1970-74 Arrivals Natives
(1) (2) (3) (4) (5) (6) (7) (8)
MOVE .06 .34
(.94) (2.52)
MOVECSA - .38
( - 2.30)
MOVEASN - .34
( - 1.92)
MOVEUP
MOVEDOWN
N 1,400 1,400
- .01 .08
(-.26) (.60)
- . I 6
( - .99)
- .07
( - .46)
.02
(.31)
.29
(1.80)
1,400 1,809 1,809
.03 .03
(S9) (.27)
. I 1
(.87)
- . I 8
(-1.26)
- .01
( - .24)
- .02
( - . ] I )
1,809 2,387 2,387
Note: Other variables included in these equations-are education, experience, experience squared,
marital status, health status, and ethnicity.
4.4 Summary
The major finding of this paper is that, although recent immigrants to the
United States move between SMSAs at a rate that is comparable to or in some
cases exceeds that of ethnic natives, there is little systematic evidence that this
immigrant population becomes more geographically dispersed as time in the
United States elapses. The only groups for whom we observed evidence of
dispersion between 1975 and 1980 were the Indians, Japanese, and Koreans.
The logit analysis of the determinants of migration between SMSAs
showed that the more educated and younger immigrants are most likely to
move. The variables that measure relative economic opportunities in the cities
(unemployment rate, area wages, and welfare benefits) did not have signifi-
cant effects on the probability that an immigrant changes SMSAs. For the
Central and South American immigrants, we found that the concentration of
fellow countrymen in the city was an important determinant of migration.
These immigrants changed SMSAs in order to move to cities with higher con-
centrations of Central and South Americans. On the other hand, the high mo-
bility rates of the Asian immigrants were unrelated to the percentages of
Asians in the various cities, while the Europeans who moved actually experi-
enced a decrease in the concentration of fellow countrymen. Obviously, there
are important differences in the characteristics of these ethnic groups that
could explain their different behavior, but with our data we have been unable
to measure these factors.
While it is difficult to explain satisfactorily why immigrants change loca-
tions in the United States, we can conclude that whatever migration does oc-
133 Internal Migration of U.S. Immigrants
cur is unlikely to lead to a substantial increase in the geographic dispersion of
newer immigrants in the United States. It is important to note that this conclu-
sion presumes that, if geographic dispersion occurs at all, it takes place within
the first fifteen years of experience in the United States. At best, treating our
cross section as a pseudo-panel, we have been able to observe the new immi-
grants only fifteen years after arrival in the United States. If we rely on our
actual panel data, then our conclusion is extrapolated from only five years of
data. Hence, it is possible that dispersion may occur but that we have been
unable to observe it. An even larger question, of course, remains unanswered
by our study. Is the lack of geographic dispersion a problem? On the one hand,
some sociologists have argued that this will inhibit the process of assimila-
tion. But an equally valid argument could be that ethnic enclaves provide the
financial and emotional support necessary for immigrants to achieve success
in their new country.
Notes
1. Adding more SMSAs increases the costs of data collection because our logit
analysis reported in the next section requires information on city characteristics. The
rate at which these costs rise far exceeds the rate at which the size of the immigrant
sample increases.
2. The index is defined as 2 S;, where S, is the proportion of individuals in the ith
SMSA.
3. The 1975-79 arrivals from Vietnam are more dispersed than the earlier cohorts
because of the special placement program described earlier.
4. The samples of natives that are of Asian or Central and South American ethnicity
each represent 25% of the actual number of observations in the Public Use B Sample.
The non-Hispanic whites, or those of European ethnicity, are 1% of the actual number
in the sample.
5. Note that LNWAGE is measured as of 1980, so we are implicitly assuming that
high-wage cities in 1980 were also high-wage cities in 1975; in other words, that there
was no systematic change in the ranking of cities according to wage level during this
time period.
6. The 1975 ratio of PFOR for Central and South American movers relative to PFOR
for Central and South American stayers is .55 for the 1965-69 cohort and .73 for the
1970-74 cohort. The 1980 ratio is .76 for the 1965-69 group and .85 for the 1970-74
group.
r = l
7. Regarding the construction of the native sample, see n. 4 above.
8. Greenwood (1975) shows that, while many studies have found a positive return
to migration, others have been unable to support this conclusion. Bartel (1979) found
that the wage gains from migration are dependent on the nature of the move (i.e.,
whether it is accompanied by a quit, a layoff, or an internal transfer within a company)
and the age of the migrant. In particular, only those who were transferred by their
companies experienced an increase in wages, which was significant for individuals
under age 45.
134 Ann P. Bartel and Marianne J. Koch
References
Baker, R. P., and D. S. North. 1984. The 1975 refugees: Theirjrstjveyears in Amer-
ica. Washington, D.C.: New Transcentury Foundation.
Bartel, A. P. 1979. The migration decision: What role does job mobility play?Ameri-
can Economic Review 69(December):775-86.
. 1989. Where do the new U.S. immigrants live? Journal of Labor Economics
7(0ctober):371-91.
Dunlevy, J. A. 1980. Nineteenth-century European immigration to the United States:
Intended versus lifetime settlement patterns. Economic Development and Cultural
Change 29(0ctober):77-90.
Fields, G. 1979. Place to place migration. Review of Economics and Statistics
6 1 (February):2 1-32.
Greenwood, M. J. 1975. Research on internal migration in the United States. Journal
of Economic Liferafure 13(June):397-433.
Massey, D. S. 1981. Dimensions of the new immigration to the United States and the
prospects for assimilation. Annual Review of Sociology 757-85.
Sjaastad, L. A. 1962. The costs and returns of human migration. Journal of Political
Economy 70, suppl. (October):80-93.
5 Migration, Ethnicity, and
Labor Force Activity
Marta Tienda and Franklin D. Wilson
5.1 Introduction
In this paper, we investigate how internal geographic mobility and the eth-
nic segmentation of jobs influence the employment and earnings of black,
Mexican, Puerto Rican, Cuban, and American Indian men. Many of the men
in our sample from the 1980 Census of Population are foreign immigrants,
and the migration of minority men has parallels with the immigration of for-
eigners. Specific questions that guide our analysis are, (1) Do the labor force
participation and unemployment of minority workers depend on whether they
move between or within high- or low-ethnic-density labor markets? (2) Do
ethnic job queues influence labor force behavior and unemployment risks? (3)
Does migration within or between high- and low-ethnic-density markets and
ethnic job queues also influence remuneration patterns? Evidence that hiring
queues and remuneration depend either on spatial concentration of minority
groups or on preferential hiring according to race and national origin is crucial
for establishing direct links between the declining labor market status of
Puerto Ricans (Bean and Tienda 1987; Tienda 1989); the economic bifurca-
tion of the black (Wilson 1987), Puerto Rican (Tienda and Jensen 1988), and
Native American (Sandefur 1986) populations; the slow economic progress of
Marta Tienda is professor of sociology, University of Chicago, and associate director, Popula-
tion Research Center of NORC and University of Chicago. Franklin Wilson is professor and chair
of the department of sociology at the University of Wisconsin.
This research was supported by grants from the Ford Foundation and the Department of Health
and Human Services to the Institute for Research on Poverty. Computational support was fur-
nished by a grant from the Center for Population Research of the National Institute for Child
Health and Human Services and Human Development to the Center for Demography and Ecology
(HD-05876). The authors gratefully acknowledge technical assistance from Sarah Rudolph and
Gary Geisserer, programming assistance from Cheryl Knobeloch, and research assistance from
Elaine Fielding, Jaewoong Shim, Brett Brown, and Albert0 Martini. The opinions expressed here
are those of the authors, not of the sponsoring institutions.
135
136 Marta Tienda and Franklin D. Wilson
Mexican men (Portes and Bach 1985; Bean and Tienda 1987); and the rapid
economic progress of Cuban men (Portes and Bach 1985).
We focus on the relation between the economic outcomes of migration and
two new variables, area ethnic density, which indexes the concentration and
distribution of an ethnic group in a particular labor market, and the ethnic type
of an individual’s industry and occupation. We find no evidence that spatial
assimilation will necessarily promote socioeconomic integration and thereby
reduce ethnic labor market inequities, and we find some indication that migra-
tion is associated with lower labor force activity, possibly because of the dis-
ruptive effects of the migration process per se.
5.1.1 Migration Types
Depending on their direction, composition, and social underpinnings, mi-
grant streams can promote ethnic consolidation or disintegration. For ex-
ample, migration may strengthen ethnic solidarity in work and school do-
mains by changing the racial/ethnic density of communities or institutional
settings. Such outcomes also depend on the existence of ethnic labor market
niches, the extent of school and neighborhood segregation, and the existence
of ethnic power bases. Thus, it is conceivable that the benefits accruing to
migrants who participate in flows leading to greater ethnic spatial concentra-
tion differ from those that produce ethnic spatial dispersion.
Geographic moves involving dispersion could improve the employment
and earnings prospects of migrants if market factors (i.e., the demand for
migrants’ skills) rather than cultural and social factors (i.e., ethnic markers)
dominate decisions to move and also influence choice of destination. Con-
versely, concentrated migration flows, which frequently are motivated by non-
economic considerations (such as the desire to reside in closer proximity to
relatives and friends of like ethnicity), may render migrants less satisfactory
employment outcomes, at least over the short run. This would follow espe-
cially if the reinforcement of cultural and ethnic bonds through concentrated
flows involves a trade-off between psychic and economic rewards. However,
if proximity to friends and relatives enables migrants to secure employment,
then the gains from participation in concentrated flows might be greater than
the economic penalties associated with ethnic crowding. Similarly, if the abil-
ity of dispersed migrants to secure a better-paying job is jeopardized by the
absence of social networks in low-ethnic-density markets, then the potential
economic gains from dispersed migration flows will be reduced.
5.1.2 Ethnic Hiring Queues
The significance of geographic mobility for the labor market stratification
of minority workers depends not only on the employment opportunities af-
forded movers but also on how individual ethnic traits circumscribe choices,
are evaluated in the market place, and are used to organize the labor market.
Specifically, if national origin is used as a criterion to define and maintain job
137 Migration, Ethnicity, and Labor Force Activity
queues-as demonstrated by previous research (Lieberson 1980; Hechter
1978)-then the economic costs and benefits of migration will derive not only
from opportunities to interact with members of like ethnicity but also from the
role of national origin in channeling minority workers to particular categories
of jobs.'
The viability of ethnic hiring queues, however, is related to ethnic spatial
concentration patterns. Lieberson (1980, chap. 10) argues that the connection
between the ethnic composition of labor markets and ethnic job queues
reflects differences in opportunities over time and the force of history in
stratifying the U.S. labor force according to race and national origin. He
claims, furthermore, that a discriminatory hiring queue results when, given
the existence of a queue, employers preferentially hire workers on the basis of
ethnic traits rather than market skills. Two aspects of Lieberson's queuing
premises have implications for our concerns. First, he argues that the job con-
figuration of groups will vary in accordance with its share of the labor force in
a given market.2 Second, he claims that, because of the existence of ethnic
hiring queues, shifts in unemployment would be highest for the group(s) at
the bottom of the queue during periods of rising ~nemployment.~ Applying
these arguments about ethnic residential and ethnic occupational segmenta-
tion to the relation between migration and employment, we hypothesize that
the market experiences of Mexican, Puerto Rican, Cuban, black, and Ameri-
can Indian men are influenced by their differential participation in concen-
trated versus dispersed migration streams and by their unequal placement in a
job queue.
5.2 Data and Methods
Our statistical analysis uses the 5% Public Use Microdata Samples (PUMS)
of the 1980 Census. We limited our sample to men aged 25-64 who had valid
responses to the migration questions and who self-reported their race or na-
tional origin as Mexican, Puerto Rican, Cuban, black, or American Indian.4
Restricting the lower end of the age distribution to 25 rather than 16 ensures
that our sample of respondents was eligible to participate in the labor force
prior to the beginning of the migration interval (1975). Additional sample
restrictions purged our results from status changes that are systematically as-
sociated with migration probabilities. For this purpose, we excluded individ-
uals who (1) never worked or were out of the labor force continuously during
the migration interval, (2) were enrolled in school or in the military in either
1975 or 1980, or (3) resided outside the United States in 1975.5
Combined, these restrictions and additional random sampling of the Mexi-
can and black populations (for computational efficiency) yielded the following
subsample N' s: 6,076 Mexicans, 6,630 Puerto Ricans, 4,134 Cubans, 5,827
blacks, and 5,810 American Indians.6 As can be seen in table 5.1, sizable
proportions of our samples are foreign born: 36 percent of Mexicans, 93
138 Marta Tienda and Franklin D. Wilson
Table 5.1 Means and Standard Deviations of Selected Variables Included in
Regression Analysis (standard deviations in parentheses)
Puerto American
Mexican Rican Cuban Black Indian
Individual characteristics:
Education:
% Finishing high school
% Finishing less than
high school
Experience (years)
% Good English ability
% Foreign born
% Work disabled
Weeks workeda
Average hours’
Family status:
% Household head
% Children under 6
% Married
Labor market characteristics:
Region:
% North Central
% south
%West
% Metro residence
Area wage rate ($)
Area unemployment rate
32.5
(46.9)
63.3
(48.2)
24.8
(12.5)
78.0
(41.5)
36.0
(48.0)
6.8
(25.1)
46.2
(11.0)
41.7
(10.2)
85.2
(35.5)
34.3
(47.5)
81.3
(39.0)
10.6
(30.8)
35.0
(47.7)
53.6
(49.9)
86.0
(34.7)
7.22
( .97)
6.19
( I .98)
34.2
(47.5)
62.0
(48.5)
24.3
(11.7)
81.3
(39.0)
77.5
(41.8)
8.7
(28. I )
46.7
(10.8)
40.0
(10.0)
79.6
(40.3)
26.9
(44.4)
71.7
(45.0)
10.8
(3 1 .O)
8.2
(27.4)
6.6
(24.9)
97.0
(16.9)
7.88
6.69
(1.23)
~ 7 2 )
40.7
(49.1)
42.1
(49.4)
29.2
(11.6)
62.6
(48.4)
93.2
(25.1)
5.2
(22.2)
48.0
42.6
(11.1)
89.6
(30.6)
14.7
(35.4)
81.8
(38.6)
(9.2)
4.1
(19.8)
61.8
(48.6)
8.5
(27.8)
98.2
(13.3)
7.38
(.61)
5.66
(1.30)
45.4
(49.8)
47.8
(50.0)
25.1
(12.9)
99.8
(4.3)
3.6
(18.7)
10.7
(30.9)
45.6
(11.9)
40.0
(11.0)
77.8
(41.5)
20.0
(40.0)
64. 5
(47.9)
20.8
(40.6)
52.3
(50.0)
7.9
(27.0)
82.4
(38.1)
7.29
(1.01)
6.53
(1.92)
51.2
(50.0)
41.4
(49.3)
23.1
(11.9)
97.1
(16.8)
2.5
(15.5)
14.5
(35.2)
43.2
(14.0)
42.0
(12.1)
81.6
(38.7)
25.1
(43.4)
72.0
(44.9)
16.9
(37.5)
29.1
(45.4)
47.7
(50.0)
55.9
(49.7)
7.14
(1.39)
7.02
(2.51)
Source: 1980 5% A Sample Public Use Micro-data Samples, migrant subsample.
a Based on subsample with positive annual earnings in 1979.
139 Migration, Ethnicity, and Labor Force Activity
percent of Cubans, and 78 percent of Puerto Ricans (where by “foreign born”
we mean born on the island). Thus, for these minority groups, we analyze
secondary choice of location of large numbers of immigrants. Restricting
these samples to individuals with some wage and salary income in 1979 for
the earnings analyses further reduced the population samples from 6 (Mexi-
cans and Cubans) to 11 percent (Puerto Ricans).
5.2.1 Variables
Our analyses focus on the relation among three variables: migration type
(i.e., whether moves took place within or between high- and low-ethnic-
density types); ethnic j ob segmentation (i.e., whether jobs were ethnic typed,
Anglo typed, or not ethnically differentiated); and labor market outcomes
(specifically, whether respondents were in the labor force or unemployed in
1980 and their 1979 [logged] annual earnings).
Migrants are defined as persons who changed residence during the five
years prior to the census. We chose standard metropolitan statistical areas
(SMSAs) and nonmetropolitan county groups (rather than states) to define
migration status. Our distinction between high- and low-ethnic-density labor
markets is derived from an analysis of both the ethnic composition of labor
markets and the distribution of each ethnic group among them. Procedures
used to classify labor markets ( N = 414) into high- and low-ethnic-density
areas are detailed in Appendix A. Briefly, a labor market area was defined as
high ethnic density for a given reference group if the group was overrepre-
sented relative to its share of the total population based on standardized (z)
scores.
Our hypotheses about the influence of geographic movement in altering the
social environments and economic opportunities of migrants emphasize the
direction of the flows. Ethnic residential dispersion involves moves from
high- to low-ethnic-density labor market areas; flows from low to high ethnic
density produce concentration; and flows within low- or high-ethnic-density
areas, labeled intradensity moves, involve no changes in the ethnic composi-
tion of labor markets from the perspective of individual migrants.’ However,
these moves usually alter economic opportunities. Since previous research
(Tienda and Lii 1987) has shown that ethnic spatial concentration directly
influences socioeconomic outcomes above and beyond productivity character-
istics, we differentiate between moves within high- and low-concentration
areas to detect the effects of concentration among intradensity movers. All
totaled, we classified individuals into five categories according to whether
they migrated and subsequently distinguished among those who participate in
dispersed, concentrated, and intradensity moves within high- andlor within
low-ethnic-concentration areas.
Operationalizing our notion of ethnic job segmentation was more compli-
cated than the coding of migration types. Because the statistical procedures
we used are detailed in Appendix B, we only highlight the logic used in distin-
140 Marta Tienda and Franklin D. Wilson
guishing among workers classified in ethnic-typed, Anglo-typed, and non-
typed jobs. We began with a thirty-cell-matrix representing a two-way classi-
fication of six industry sectors by five occupation groups using 1970 Census
data.8 Sector-by-occupation matrices were computed for each of the five eth-
nic groups and non-Hispanic whites. Based on the results of a log-linear anal-
ysis, we classified job cells according to whether each ethnic group was over-
represented (ethnic typed), underrepresented (Anglo typed), or approximately
equally represented (nontyped) relative to non-Hispanic whites and net of
group differences in education and age composition. These results, summa-
rized in Appendix tables 5B. 1 and 5B.2 below, are substantively informative,
but we do not dwell on them in the interest of brevity.
5.2.2 Modeling
Our conceptualization of the employment experiences of minority men in-
tegrates two structural attributes of labor markets-the ethnic segmentation
of jobs and the ethnic composition of markets-and assesses their influence
on labor force participation, unemployment, and (logged) annual earnings.
The segmentation of jobs along ethnic lines requires a critical mass of minor-
ity workers; hence, we hypothesize that labor market experiences of minority
men may differ in high- and low-density labor markets. Accordingly, our em-
pirical model, which assumes that both geographic mobility and the ethnic
labeling of jobs influence the labor force participation and unemployment
prospects of minority men, takes the form:9
(1)
where
LF, =
Mj =
P, =
z, =
e, =
PR(LF), = CL + PI MI + ya P, + Z, + e, ,
labor force status of individual i; 1 = in, 0 = out for the participa-
tion equation; 1 = unemployed, 0 = employed for the unemploy-
ment equation.
migration type, and k = 4, 3, 2, 1, and 0 represent whether individ-
uals participated in dispersed, concentrated, intradensity low, or intra-
density high flows, or were nonmigrants, respectively;
ethnic job segments, and k = 2, 1, and 0 for ethnic-typed, Anglo-
typed, and nontyped segments, respectively;
a vector of controls;
random disturbances.
The predicted effects of ethnic job segmentation are informed by economic
logic as well as sociological insights about the significance of race and ethnic-
ity in demarcating boundaries for social interaction. If the existence of ethnic
hiring queues “reserves” jobs for minority workers (as in ethnic niches or
enclaves), then y2 > 0 in the participation equation, and y2 < 0 in the unem-
ployment equation. This result would show the influence of social (ethnicity)
forces in defining paths of labor market activity for minority workers (Portes
141 Migration, Ethnicity, and Labor Force Activity
and Bach 1985).1° However, if workers destined for Anglo-typed jobs are
more likely to be in the labor force than their (statistical) counterparts identi-
fied with nondifferentiated job categories, the y1 > 0 in the participation
equation, and the y, < 0 in the unemployment equation. In this instance, fac-
tors other than ethnicity will govern the employment prospects of minor-
ity men.
Our predictions about the influence of migration types on employment out-
comes are informed theoretically by research and writing on the socioeco-
nomic significance of ethnic spatial concentration (Tienda and Lii 1987). Ex-
tending ideas about ethnic density to geographic movement, we expect
nonzero effects associated with participation in concentrated (p, # 0) and
dispersed (p, # 0) migration flows, but the direction of these effects is an
empirical question. For example, if p, > 0 in the participation equation
(p, < 0 in the unemployment equation), then dispersed migration would ap-
pear to promote the labor market assimilation of minority men. A negative
value of p, in the participation equation (positive in the unemployment model)
would indicate that the investment properties of the migration decision either
require a long time to mature or else may depend on the ethnic hiring queues
at destination. Alternatively, a positive coefficient for p, in the participation
equation (negative in the unemployment equation) would indicate that con-
centrated labor flows promote socioeconomic assimilation in the context of
increasing ethnic pluralism. While not denying the importance of supply fac-
tors in determining minority labor market outcomes, this result is consistent
with the premises of queuing and overflow perspectives of labor market dy-
namics (see nn. 3, 4). Negative returns to concentrated migration (p, < 0 in
the participation equation and p, > 0 in the unemployment equation) suggest
that, over the period considered, the disruptive aspects of the investment de-
cision offset the investment gains from the decision. Intradensity moves pre-
sumably represent investment decisions in response to better employment
prospects; hence, we expect p, and p2 > 0. Our assessment of the earnings
consequences of migration in the context of ethnic residential concentration
and ethnic job segmentation assumes the following form:
(2) Y, = OL + pj Mj + ye P, + Z, + e, .
This model is analogous to our participation and unemployment equations,
although the vector, Zi is not identical (see the definitions following eq. [ 11) .
Interpretations of the ethnic queuing effects reflect our hypotheses about the
underlying stratifying mechanisms. If minority workers destined for ethni-
cally typed jobs gain financially compared to those destined for nondifferen-
tiated jobs, then y2 > 0. This result is highly plausible in contexts where
minorities are overrepresented among the self-employed (Lieberson 1980;
Portes and Bach 1985). However, if minority workers destined for Anglo-
typed jobs earn more than their (statistical) counterparts holding nontyped
jobs, then y1 > 0. These findings would indicate that the more desirable jobs,
142 Marta Tienda and Franklin D. Wilson
usually dominated by whites, hold the key to reducing wage disparities be-
tween minority and nonminority men. Our interpretation of migration effects
will be similar to those discussed above.
We introduce in all models a set of controls for individual and labor market
characteristics known to influence labor market outcomes. Appendix C sum-
marizes all variables included in the vector Z,, providing a brief operational
description of the controls as well as the key dependent and independent vari-
ables. I 1 It also indicates whether the control variables were included in the
labor supply equations, the earnings equations, or both.
5.2.3 Techniques
Because the dependent variables in the labor supply equation, labor force
participation and unemployment, are dichotomous, we use a maximum like-
lihood logistic regression to estimate the models. For ease of interpretation,
we report only the partial derivatives of the probabilities using the procedure
derived by Petersen (1985). The (log) earnings equations are estimated using
OLS regression.
5.2.4 Evidence
Before presenting results from the regressions, table 5.1 provides back-
ground information about our samples of minority men. The disadvantaged
labor market status of Mexican and Puerto Rican men reflects their low stocks
of human capital. Fully two-thirds of mature Mexican and Puerto Rican men
had not completed high school, compared to 48 percent of black men and 41
and 42 percent of Indian and Cuban men, respectively. At the other extreme,
17 percent of mature Cuban men and approximately 7 percent of black and
American Indian men were college graduates, compared to roughly 4 percent
of Mexican and Puerto Rican men.
Lack of proficiency in English is a problem confined largely to Hispanics.
That only 63 percent of Cubans reported good to excellent proficiency in En-
glish reflects the predominantly foreign origins of our sample of mature men.
Although roughly three-fourths of Puerto Rican men were born on the island
of Puerto Rico, where Spanish is the predominant language, English is taught
in the schools, and bilingualism is quite pervasive. Only 78 percent of Mexi-
can men reported good to excellent proficiency in English, yet only 36 percent
were immigrants. This results partly because of the higher rates of Spanish
retention among the native born (Nelson and Tienda 1985) and partly because
of the lower levels of schooling completed by mature Mexican men.
The incidence of work limiting disability is highest among black and Amer-
ican Indian men and lowest for Cuban and Mexican men. Poor health is a
corollary of the high poverty rates characteristic of these groups, particularly
those isolated on remote reservations (Sandefur 1986) and inner-city ghettos
(Wilson 1987). Not surprisingly, average weeks worked in 1979 were lowest
143 Migration, Ethnicity, and Labor Force Activity
for these two groups. Cuban men reported the highest average weeks worked
in 1979 and the longest average work week.
Our samples are further differentiated by family and household character-
istics. Nearly four-fifths of Mexican and Cuban men were married, compared
to less than 65 percent of black men and only 72 percent of Puerto Rican and
American Indian men. Cuban and Mexican men were most likely, and black
and Puerto Rican men least likely, to identify as households heads. Reflecting
their younger age composition and higher fertility, approximately one-third of
our Mexican sample had young children at home, compared to one-fourth of
Puerto Rican and black men and less than 15 percent of Cuban men.
Finally, minority men are differentiated by geographic characteristics. In
1980, less than 1 percent of mature Mexican men and 6 percent of American
Indian men resided in the Northeast, compared to nearly three-fourths of ma-
ture Puerto Rican men. Mexicans were disproportionately concentrated in the
South and West, Cubans in the South (Florida) and Northeast, while blacks
were underrepresented in the West. Hispanics (Cubans and Puerto Ricans in
particular) were largely a metropolitan population; in contrast, less than 60
percent of American Indian men reported living in metropolitan areas in 1980.
These residential profiles have direct implications for employment and in-
come opportunities. American Indians resided in labor markets characterized
by the highest unemployment rates, followed by Puerto Rican men, while
Cuban men confronted the lowest average unemployment rates in 1980.
American Indian men faced the lowest average wage rates, largely because
they were disproportionately located in nonmetropolitan and reservation
areas, while the highest wages corresponded to labor markets where Puerto
Ricans live. That high unemployment also characterized these high-wage
markets is a key piece of information for decoding the declining economic
status of Puerto Ricans.
Table 5.2 shows that American Indians were the most mobile minority
group during the late 1970s, as nearly one in five reported having changed
county groups between 1975 and 1980. At the other extreme, only 8 percent
of mature black men migrated between 1975 and 1980. Among Hispanics,
Cubans were the most mobile, and Puerto Ricans were slightly less mobile
than Mexicans, with 10-12 percent changing labor markets during the time
period. More interesting are the differences in the direction of the migrant
flows. Geographic movement between high-ethnic-concentration labor mar-
kets was the modal migration type for all groups, accounting for 50-65 per-
cent of intermarket moves by Hispanic men and 35 and 43 percent of moves
by American Indian and black men, respectively. Flows between labor mar-
kets with low levels of ethnic concentration were prominent only among
American Indians. In the main, these moves capture the residential mobility
of the nonreservation Indian population.
Although the numbers of migrants engaged in dispersed and concentrated
144 Marta Tienda and Franklin D. Wilson
Table 5.2 Descriptive Statistics for Migration Qpe and
Ethnic Job Segmentation
Mexican
Puerto
Rican Cuban Black
American
Indian
Migration type:
Intradensity high
Intradensity low
Concentrated
Dispersed
Migrants (totals)
Ethnic typed
Anglo typed
Nontyped
Ethnic job segmentation?
6.5
1.2
1.2
1.7
10.6
13.0
4.1
82.9
4.9
1.4
1 .o
2.7
10.0
18.6
6.4
75 .O
6.8 3.4
1.3 1.2
2.3 1.5
1. 8 1. 8
12.3 7.9
13.1 36.6
14.1 5.6
72.8 57.8
6.7
5. 8
3.3
3.0
18.8
19.9
15.0
65.1
N 6,076 6,630 4,134 5,827 5,810
Source: 1980 5% A Sample Public Use Micro-data Samples, migrant subsample.
Nore: All tabulations exclude recent immigrants.
a Compares each group to Anglos. See App. B.
migration flows did not differ greatly within groups (Puerto Ricans being a
notable exception), concentrated moves were more pervasive among Cubans
and American Indians, while Mexicans and blacks became slightly more dis-
persed. Puerto Ricans experienced the greatest residential dispersion during
the late 1970s as a result of internal migration. Having always been a metro-
politan population on the U.S. mainland, for them dispersion involved moves
out of New York and the Northeast in general (Bean and Tienda 1987).
The last three rows of table 5.2 support the view that blacks were especially
likely to be in ethnic-typed jobs. Over one-third of all mature black men in
the experienced civilian labor force identified with jobs where blacks were
disproportionately concentrated. Appendix table 5B. 1 below shows that in
1970 blacks were uniformly overrepresented in the lower nonmanual and in
upper manual jobs in the producer, social, and personal services sector. Aux-
iliary tabulations (available from the authors) reveal that these patterns per-
sisted until 1980. Based on their disproportionate concentration in manual
jobs, roughly 20 percent of Puerto Rican and American Indian men identified
with ethnic-typed jobs. Finally, Mexican and Cuban men were least likely to
be ghettoized in “ethnic jobs” in 1980. For Mexicans, ethnic job typing
largely involved agricultural activities but also included lower manual jobs in
the distribution and personal service sectors. Cubans, on the other hand, were
overrepresented in some nonmanual as well as manual activities, but princi-
pally in the producer and personal services sectors. l2
At the other extreme, Cubans and American Indians were the only groups
for whom the share engaged in "Anglo-typed" jobs exceeded 10 percent. For
both groups, these jobs involved nonmanual activities in the transformative
145 Migration, Ethnicity, and Labor Force Activity
and distributive services sectors. Clearly, employment in nontyped jobs was
the model form of labor market insertion for the overwhelming share of mi-
nority men, with blacks least likely and Mexicans most likely to be so situated
as of 1980. However, these are very conservative measures of ethnic job seg-
mentation (see App. B). Had our method to establish over- and under-
representation not included statistical adjustments for age and educational
differentials between the groups, the proportion of workers allocated to
“ethnic-typed” jobs would have been considerably greater. Also, had we used
more liberal cut points for stipulating the “tolerable” limits of ethnic job seg-
mentation, the numbers of industry-occupation cells designated as ethnic
typed and the share of workers so-allocated would have been higher.
The three dependent variables analyzed are summarized in table 5.3. Even
though our sample is limited to men aged 25 to 64, the labor participation
rates range from a high of 96 percent for Cubans to a low of 86 percent for
American Indians. The labor force participation rate for blacks was slightly
higher than that of American Indians, but this does not necessarily mean that
the latter group is situated at the bottom of the employment hierarchy. Rather,
this reflects the extremely limited employment opportunities on reservations
(Sandefur 1986; Snipp and Sandefur 1988). Auxiliary tabulations revealed
that American Indian labor force participation was significantly lower and un-
employment significantly higher in areas of high ethnic concentration, while
the reverse was true for blacks. Differential participation and unemployment
rates among groups reflect variation in employment opportunities by resi-
dence and differential placement of groups in the hiring queue. Among His-
panics, Puerto Ricans have especially low average participation rates and high
unemployment (see also Tienda 1989).
The ranking of minority groups based on 1979 annual earnings reaffirms
the placement of black men at the bottom, where they are accompanied by
Puerto Ricans rather than American Indians. At the opposite extreme stand
Table 5.3 1980 Labor Force Participation and Unemployment Rates and la79
Average Annual Earnings of Minority Men Aged lbenty-Five to
Sixty-Four (standard deviation in parentheses)
Puerto American
Mexican Rican Cuban Black Indian
Labor force participation 93.6
(24.4)
Unemployment 6.6
(24.3)
Logged annual earnings 9.22
( .89)
Average annual 13,342
earnings ($) (9,414)
91.4 95.6
(28.0) (20.6)
7.5 3.8
(26.3) (19.2)
9.18 9.43
(37) (.82)
12,587 16,368
(8,647) (13,068)
88.9
(31.4)
8.6
(28.1)
9.14
(.97)
12,585
(8,485)
86.3
(34.4)
10.9
(31.2)
9.16
(1.06)
13,938
( 1 1,269)
~~ ~
Source: 1980 5% A Sample Public Use Micro-data Samples, migrant subsample.
146 Marta Tienda and Franklin D. Wilson
Cubans, with average annual earnings of $16,400 in 1979. Mexican and
American Indian men are situated in an intermediate position relative to
blacks and Cubans. Although the average annual earnings of American Indi-
ans may seem high, recall that this figure is based on the subset with positive
earnings in 1979. Since a lower share of this group participates in the labor
market, the subset with earnings is a highly selected segment of this popula-
tion.
Overall these descriptive statistics provide useful background information
for interpreting our multivariate results, which analyze, in sequence, the prob-
ability of being in the labor force or unemployed and logged annual earnings
as a function of migration type, ethnic job typing, and a vector of control
variables.
5. 2. 5 Labor Force Participation and Unemployment
The transformed logit coefficients in table 5.4 reinforce a picture of diver-
sity in the determination of participation decisions and unemployment risks
among minority men. Considering first the results for labor force participa-
tion, one general inference is that minority men who migrated during the latter
part of the 1970s were consistently less likely, or no more likely than their
nonmigrant counterparts to be in the labor force. Although only nine of the
twenty estimated coefficients attained statistical significance, the reliable
coefficients were uniformly negative. These results appear to contradict con-
ventional wisdom about the investment properties of migration decisions.
However, it is conceivable that, for minority men, migration may require a
substantial period of adjustment before the returns accrue. There is no way for
us to control for the exact timing of migration during the five-year interval;
hence, it is not possible to ascertain to what extent our results may be captur-
ing the initial disorganization following migration. l 3
The diverse effects on labor force decisions of migration type warrant more
extensive discussion. At one extreme stand Native Americans, whose labor
force behavior was relatively unaffected by geographic movement. These
findings are consistent with those of Sandefur (1986), who found no signifi-
cant differences in the effects of interstate migration on the labor force partic-
ipation of American Indians. l4 Although Sandefur did not speculate why this
result emerged, it is conceivable that the resettlement assistance provided
movers does not stipulate that recipients secure employment prior to moving.
Nonmigrants chose to stay either because they have employment or because,
in its absence, they can rely on other forms of income maintenance and infor-
mal social supports to subsist.
For blacks and Mexicans, the pattern of migration effects on labor force
behavior is roughly similar. Moves within high-density labor markets lowered
the probability of participation by 10 percent for mature black men and by
approximately 5 percent for mature Mexican men. That dispersed migration
substantially lowered rather than increased the probability of labor force par-
147 Migration, Ethnicity, and Labor Force Activity
ticipation for these groups illustrates both the disruptive aspect of moves and
the employment difficulties experienced by men of color in sociocultural en-
vironments where members of like ethnicity are less available.
The results for Cubans and Puerto Ricans were surprising for different rea-
sons. Because the vast majority of concentrated flows of Cubans involve in-
dividuals moving to Miami, Florida (Bean and Tienda 1987), where a thriving
enclave economy shields workers from general competition (Portes and Bach
1985), we expected that participation in concentrated flows would increase
the probability of labor force participation for Cubans. Instead, the estimated
effect is strongly negative, indicating participation probabilities 8.6 points be-
low those of nonmigrants, on average. This result may indicate that the Miami
labor market is already saturated with Cuban workers or that these migrants
can afford to spend longer periods in job search precisely because the availa-
bility of ethnic compatriots (usually relatives) provides necessary supports
during the interim period.
The results for Puerto Rican men are cause for concern because of the sharp
decline in the participation of mature men experienced during the 1960s and
through the 1970s-a pattern less pronounced for other Hispanic populations
(Tienda 1989). For Puerto Ricans, internal migration consistently lowered the
employment prospects from 7 percent (for intradensity high moves) to 15 per-
cent (for dispersed moves). This finding is all the more disturbing because
Puerto Ricans have become more residentially dispersed than other Hispanic
populations. Thus, the declining participation rates of mature Puerto Rican
men result from many sources, including the disruptive aspects of moves in
general but dispersed moves in particular, the greater prevalence of dispersed
compared to concentrated flows, and the sharply declining employment op-
portunities in labor markets where Puerto Ricans traditionally have concen-
trated.
The unemployment effects of migration were generally weaker than those
on labor force participation in several senses. First, only three of the twenty
estimated coefficients (for migration types) attained statistical significance;
second, these coefficients were confined to Mexican and Native American
men. Stated differently, migration does not appear to influence the unemploy-
ment prospects of Puerto Rican, Cuban, or black men, irrespective of the
ethnic density of origin and destination labor markets. This is not the case for
Mexican and American Indians, for whom migration increased the probability
of unemployment.
For Mexicans, the statistically significant effects corresponded to the mi-
gration types that generated reliable coefficients in the participation equation,
except that the coefficients were oppositely signed (as predicted). Specifically,
Mexicans who participated in dispersed migration flows experienced unem-
ployment rates almost 10 percent higher than (statistically) equivalent nonmi-
grants. In contrast, migrants who participated in moves designated as concen-
trated were as likely as nonmigrants to be unemployed in 1980. Mexican
Table 5.4 Effects of Geographic Mobility and Ethnic Job Segmentation on Labor Force Participation and Unemployment
(transformed logit coefficients)
Mexican Puerto Rican Cuban Black American Indian
LFP UNEM LFP UNEM LIT UNEM LFP UNEM LFP UNEM
Migration type:
Intradensity high
lntradensity low
Concentrated
Dispersed
Ethnic typed
Anglo typed
Human capitaVfamily status:
If < high school complete
If high school complete
Experience
Experience squared
If English good
If foreign born
If work disability
If married
Ethnic job segmentation:
-.047** ,029 -.068**
- ,014 .029 -.091*
- ,011 -.035 -.118*
- .068* .097*** - . l52***
. . . -.024** ,007
,015 -.050** ,011
- ,027
,001
.003*
. . .
- ,008
.016*
- a***
.007
.216*** - .199***
.115* -.097*
- ,002 .005**
-.020** ,011
- ,009 ,005
- .025** .019*
. . . . . .
.037* - .446***
.009
.003
- ,014
,023
- .010
- ,010
.loo**
.050
- ,002
- .020**
- .017*
- .022**
. . .
.070***
-.013 -.012 -.103*** ,045
,005 ,010 -.040 .025
-.086** -.001 - ,042 ,028
,006 -.025 -.143*** .020
-.008 -.008 -.021* ,016
- ,003 .003 ,020 -.050**
- .028 .102*** - ,045
- ,014 ,029 -.036
- ,001 . . . . . .
,003 -.001 .008***
.009 ,004 .05 1
.015 ,004 ,048
- .408*** .069*** - .429***
.014 -.014* .025*
.147***
.072*
- ,003
- ,029
.009
.041*
- .025**
. . .
- .008 .058**
-.011 ,039
- .019 ,053
- .063 .060*
-.040** ,016
.037* - .051***
- .123*** ,309***
- .094** .240***
.006*** - .005**
. . . . . .
-.043 -.027
- ,034 ,025
- .352*** .050**
.033** - .034***
If household head
If children < 6
If region = Northcentral
If region = South
If region = West
Area unemployment rate
Area wage rate
If metro residence
% Durable manufacturing
% Nondurable manufact-
Labor market:
uring
X2
df
N
.045*** - .024**
.020* ,018
,019 -.013
,015 -.025
... - ,011
- ,002 ,011
.004 -.002
,001 ,016
.Ooo . . .
-.001 -.001
2,265 2,576
6,051 5,663
6,076 5,688
.049***
.024*
,025
.035*
.038**
- ,012
,008
,026
,001
.007***
3,056
6,605
6,630
- ,015
,002
- ,003
- ,041
- ,001
- .013
.009**
.074*
. . .
- ,001
3,08 1
6,035
6,060
.024***
,017
- ,003
- .094*
,015
- ,005
- .025
- ,001
- ,001
1,152
4,109
4,134
.039***
- .015*
- .011
,019
- ,022
,015
- ,006
- ,009
- .026*
- .002*
,001
1,205
3,925
3,950
.069***
.057***
,020
.038**
,013
- .009**
.015*
.025
.001
.003**
3,159
5,802
5,827
- .037*** .088***
-.006 -.009
.03 1
- ,002
.045*
- .015***
,011
- ,016
.001
,001
2,765
5,157
5,182
- ,030
,007
- .005
- ,003
- .013**
.033**
.005***
,002
3,744
5,785
5,810
- .036***
- ,002
,029
- .029
- .017
.OlO***
,006
- ,019
. . .
. . .
3,179
4,989
5,014
Note: Evaluated at the following means: Mexicans, LFP = ,936, UNEM = ,066; he r t o Ricans, LFF’ = ,914, UNEM = .075; Cubans, LFP = .956, UNEM =
,038; American Indians, LFF’ = .863, UNEM = ,109; blacks, LFP = ,889, UNEM = ,086.
a Coefficient less than ,001.
+ p 5 .I0 (two-tailed test).
*p 5 .05 (two-tailed test).
**p 5 .01 (two-tailed test).
***p 5 ,001 (two-tailed test).
150 Marta Tienda and Franklin D. Wilson
migrants who moved between labor markets where Mexicans were dispropor-
tionately represented experienced unemployment rates only 3 percent higher
than nonmigrants, on average, but this effect was statistically trivial. Ameri-
can Indian unemployment rates were most affected by residential mobility, as
all migrants except those who moved between low-Indian-density labor mar-
kets and from low- to high-density labor markets experienced higher unem-
ployment than their nonmigrant counterparts. The risks of unemployment for
these American Indian migrants averaged approximately 6 percent more than
those of (statistically) comparable nonmigrants.
The influence of job segmentation on labor force participation and unem-
ployment probabilities revealed very limited evidence that the ethnic typing of
jobs operated to reserve positions for minority men or to shield them from
unemployment. Only for blacks and American Indians, two groups identified
as occupying the lower rungs of the hiring queue, did statistically reliable
effects emerge in the participation equation. These results indicate that mature
black and American Indian men destined for ethnic-typed jobs have lower
rates of labor force participation than their (statistical) counterparts destined
for nontyped jobs. l 5 However, Native Americans associated with Anglo-typed
jobs participated in the labor force at a rate 4 percent higher than their simi-
larly skilled counterparts associated with nontyped jobs. This finding,
coupled with evidence that American Indian men associated with Anglo-typed
jobs experience significantly lower unemployment than their counterparts
destined for nontyped jobs, possibly reflects the results of affirmative hiring
of American Indians within the Bureau of Indian Affairs and in other nonman-
ual government jobs.
The only shred of evidence that ethnic segmentation of jobs “reserves” slots
for minority men emerges from the Mexican unemployment equation, which
shows that adult men destined for “Mexican jobs” were slightly less likely to
be unemployed in 1980 than their counterparts associated with nontyped jobs.
But, as the results for Mexicans, blacks, and American Indians show, the
lowest unemployment risks are associated with Anglo-typed jobs-predomi-
nantly nonmanual and skilled manual jobs. Apparently, the desirability of jobs
rather than the ethnicity of incumbents influences the unemployment risks of
minority men. Moreover, because the dominant group generally occupies the
more desirable jobs, ethnic-typed jobs exhibit greater employment instability
and hence expose their incumbents to greater risks of unemployment.
With respect to labor market characteristics, none emerged as significant
predictors of Mexican men’s labor supply, in contrast to results for the remain-
ing groups. Participation rates of Puerto Ricans were lowest in the Northeast,
the region where this group is disproportionately concentrated and where the
dislocation effects of industrial restructuring have been particularly severe.
That Puerto Rican unemployment risks were not responsive to our regional
dummies reveals that the massive numbers of F’uerto Rican workers displaced
151 Migration, Ethnicity, and Labor Force Activity
from jobs in the Northeast have not been absorbed by other regions. The sharp
impoverishment of the population during the 1970s and 1980s (Tienda and
Jensen 1988) can largely be understood in these terms. Further evidence for
this argument is found in the higher unemployment probabilities associated
with residence in metropolitan areas and high-unemployment areas. Cubans
residing in the South had lower participation rates than their counterparts re-
siding elsewhere, but they also had lower unemployment rates. Black men
residing in the South and in high-wage areas participated more in the labor
force than other black men, and highest unemployment risks were associated
with residence in the West, where blacks are less concentrated. Finally, partic-
ipation rates of American Indians were slightly higher and unemployment
rates lower in metropolitan areas.
Among the set of human capital and family status variables included in our
models, work disability produced consistently strong negative effects on labor
force participation and strong positive effects on unemployment risks. English
proficiency lowered unemployment probabilities only for Mexican and Puerto
Rican men. The influence on the labor force participation and unemployment
of minority men’s educational credentials was as expected for Mexicans and
Puerto Ricans, but, surprisingly, we detected no schooling effects on the par-
ticipation decisions of black and Cuban men. This result is all the more puz-
zling since, according to our previous discussion, these men occupy the top
and bottom of the hiring queue and educational credentials presumably should
not function identically for both. That they do highlights the importance of
ascription over achievement (i.e., race over skill) in differentiating labor mar-
ket experiences.
To summarize, the evidence from our analyses of labor force participation
challenges the widespread assertion that dispersed migration streams poten-
tially can promote socioeconomic assimilation along with spatial assimila-
tion. In fact, for three of the five groups analyzed (Mexicans, Puerto Ricans,
and blacks), the effects on labor force participation of dispersed migration
flows were negative rather than positive; for the remaining two groups (Cu-
bans and American Indians), dispersed migrants were neither more nor less
likely to participate in the labor force than nonmigrants. Also, identification
with ethnic- or Anglo-typed jobs differentiated the participation decisions of
American Indian men and to a lesser extent black men, but it had little to do
with the participation decisions of Hispanic men. However, Mexican, black,
and American Indian men associated with Anglo-typed jobs experienced
lower unemployment risks compared to those associated with nontyped jobs.
5.2.6 Annual Earnings
By and large, and in contradistinction to the results reported for the labor
supply models, the earnings consequences of migration types are generally
weak (see table 5.5). For Cubans and Puerto Ricans, the economic returns to
Table 5.5 Effects of Geographic Mobility and Ethnic Job Segmentation on 1979
(logged) Annual Earnings: Migrant and Nonmigrant Men Aged lkenty-Five
to Sixty-Four (standard errors in parentheses)
Puerto American
Mexican Rican Cuban Black Indian
Migration type:
Intradensity high
Intradensity low
Concentrated
Dispersed
Ethnic job
segmentation:
Ethnic typed
Anglo typed
Human capitaVfamily
status :
If < high school
complete
If high school
complete
Experience
Experience
squared
If English good
If foreign born
If work disability
If manied
Weeks worked
Usual hours
A
Constant
R2
N
- .078*
(.039)
.001
(.087)
- ,102
(.092)
,027
(.073)
- . I lo***
(.029)
.195***
- .361***
(.105)
- .637***
(.114)
,025
(.ow
- 0@)4***
(.W1)
.176***
(.028)
- .071**
(.024)
- .120**
(.046)
.152***
.033***
(.001)
.007***
(.001)
.o001
(.030)
6.489
,342
5,726
,036
( . a51
- .061
(.081)
- ,120
(.094)
- . 118+
(.063)
- .145***
(.025)
.077*
(.039)
- .287***
(.069)
(.077)
,015
(.ow
- .0002**
(.W1)
.116***
(.026)
- . I 1 I***
(.024)
- .154***
(.042)
.142***
(.025)
.035***
(.@)I)
.007***
(.001)
.011*
- .456***
6.448
.315
5,908
,036
(. 045)
,071
(.098)
,020
(.073)
-.150+
(.082)
- .249***
.117***
(.033)
(.031)
- .268***
(.039)
- .331***
(. 045)
.010*
(.004)
- .0002*
(.OOO1)
(.028)
- ,083 +
( . 045)
- .159**
(.061)
.140***
(.032)
.034***
(.001)
.007***
(.001)
.019***
(. 006)
6.656
,340
3,895
.167***
,081
(.ow
-.184+
(.101)
,145
(.091)
- ,022
( ,085)
- .185***
(.023)
,001
(.047)
- .243***
(.061)
- .373***
(.078)
- .020***
- .0003***
(.OOO1)
- .I82
(.272)
(.059)
- .205***
(.049)
.119***
(.027)
.034***
(.@)I)
o@J***
(.001)
.057*
(.023)
6.566
,376
5,235
- ,036
- .057
.056
- .060
(.046)
(.063)
- .138*
(.067)
- ,017
(.029)
(.032)
.097**
- .298***
(.044)
- .518***
(.054)
.022***
(.W)
- @)04***
(.OOO1)
.316***
(.071)
(.074)
(.042)
(.034)
- ,028
- .234***
.186***
.036***
(.001)
.011***
(.001)
- ,004
(.017)
5.590
,428
5,324
Source: 1980 5% A Sample Public Use Micro-data Samples, migrant subsample.
Note: Net of vector of labor market characteristics: region; area wage rate; metropolitan residence.
+ +p 5 . I0 (two-tailed test).
*p 5 .05 (two-tailed test).
**p 5 .01 (two-tailed test).
***p 5 ,001 (two-tailed test).
153 Migration, Ethnicity, and Labor Force Activity
different migratory destination are statistically trivial (p,-p, = 0; see eq.
[ 2 ] ) , although the income-depressing effects of dispersed streams border on
statistical significance. l6 However, for each of the remaining groups, at least
one migrant type emerged as a significant determinant of annual earnings.
Participation in dispersed migration streams rendered American Indians a loss
of 15 percent. Apparently, spatial dispersion does not automatically translate
into economic gains for American Indian men. This questions the wisdom
underlying efforts to relocate Native Americans as a strategy promoting social
integration. For Cubans and Puerto Ricans, it appears that spatial dispersion
is not associated with economic success, but other types of movers earn the
same as equivalent nonmigrants. The (logged) annual earnings of migrant and
nonmigrant black men were approximately equal, and our disaggregated mi-
grant typology yields no further insights into the economic consequences of
specific flows. However, Mexicans who participate in intradensity streams re-
ceived negative returns on the migration decision.
In contrast to our previous results, the regressions of (logged) annual earn-
ings show large negative retums from incumbency in ethnic-typed jobs (with
the exception of American Indians) and substantial positive returns from in-
cumbency in Anglo-typed jobs (with the sole exception of blacks). Specifi-
cally, the earnings penalty from incumbency in ethnic-typed jobs ranged from
a low of 11 percent for Mexicans to a high of 25 percent for Cubans, while
the bonus from incumbency in Anglo-typed jobs ranged from a low of 8 and
10 percent for Puerto Rican and American Indian men, respectively, to a
high of 20 percent for Mexican men. That blacks did not benefit financially
from incumbency in Anglo-typed jobs reflects the legacy of discrimination
in excluding them from these jobs and denying them equal returns for equal
work.
In sum, our findings on earnings challenge arguments about the economic
benefits of ethnic segmentation for minority workers. Because the jobs in
which minority men are concentrated disproportionately involve manual work
in extractive, manufacturing, and personal services, their remuneration is
lower, on average, compared to that of nonmanual jobs, particularly profes-
sional and managerial positions in the producer and social services sectors.
The few minority men who do manage to enter Anglo-dominated jobs do re-
markably well, financially. Nevertheless, as long as ethnicity continues to
have exchange value in the labor market, not only will the earnings of minor-
ity men remain highly unequal, but so also will the financial rewards asso-
ciated with ethnically segmented jobs.
5.3 Conclusion
On balance, our results provide some evidence about how and why the
paths of labor market insertion differ among mature minority men, but our
154 Marta Tienda and Franklin D. Wilson
story about the role of migration and ethnic job segmentation in stratifying the
minority work force is complex. First, when significant effects of migration
on labor force participation emerged, they were uniformly negative (positive
in the unemployment equation). This implies that the higher unemployment
experiences and lower labor force activity rates of migrants may reflect the
disruptive effects of the migration process per se. These effects, if they are
associated with the process of movement per se, might disappear as migrants
acquire experience and familiarity with their destination labor markets.
Unfortunately, our cross-sectional Census data did not permit us to investi-
gate whether migrants who had moved earlier were more likely to participate
in the labor force than later migrants. This certainly is an important research
priority for additional work on the economic consequences of residential mo-
bility for minority populations.
That the effects of migration were most pronounced for labor force behavior
but not economic returns further challenges the premises of microeconomic
theory, which presumes that decisions to move represent rational choices
geared to improve economic well-being. But whether ethnic alliances are in-
volved in explaining the prevalence of concentrated flows, and those within
areas of high ethnic concentration, or in ameliorating the disruptive aspects of
residential mobility is not clearly evident from our results. On the basis of the
theoretical arguments presented at the outset, evidence for such claims should
derive from the nature of ethnic segmentation and the existence of ethnic hir-
ing queues. Although our ethnic segmentation effects on labor force decisions
and unemployment risks were trivial at best, their effects were far more pro-
nounced on (logged) annual earnings.
One general implication from our results is that there is no evidence that
spatial assimilation will necessarily promote socioeconomic integration and
thereby reduce ethnic labor market inequities. Only Cuban men benefited
from dispersed migration streams, while American Indians sustained substan-
tial losses from disrupting their social ties and participating in dispersed mi-
gration streams. Certainly, this does not justify mandatory relocation pro-
grams as a strategy for promoting social equity.
Appendix A
Analytical Procedures to Determine High-Density
Labor Market Areas
To determine which labor market areas contain an above-average concentra-
tion of a particular racial or ethnic group, we examined two relevant variables,
the raciallethnic composition of each labor market area and the distribution of
each group across the 414 labor market areas. These labor market areas were
155 Migration, Ethnicity, and Labor Force Activity
derived from the Census-defined county groups and consist of SMSAs or
groups of nonmetropolitan counties within states. Population counts from
the 1980 1/100 PUMSA were used to calculate these variables for the follow-
ing groups: blacks, American Indians, Mexicans, Puerto Ricans, Cubans, and
other Hispanics. The total population was divided in mutually exclusive cate-
gories as follows: anyone identifying himself or herself as “American Indian”
on the race question was considered American Indian; non-Indian Hispanics
were identified on the “Spanish origin” question, which contained separate
spaces for Mexican, Puerto Rican, Cuban, and other; the remainder were
placed into either the white, Asian, or black categories on the basis of their
answer to the race question.
A labor market area was defined as high density if the reference group was
overrepresented in terms of both composition and distribution. Overrepresen-
tation was determined by calculating a set of standardized scores for the two
variables. For the compositional z-score, the group’s percentage for the coun-
try as a whole (the weighted mean across areas, e.g., 11.58 percent for
blacks) was used to represent the value expected if that group was evenly
distributed across labor market areas relative to all other groups. The simple
mean, which is the same for all groups (.24 percent, or %M), was used for the
distributional z-score. A labor market area was classified as concentrated if
both these standardized scores were greater than zero. Therefore, a concen-
trated black labor market area would be one containing more blacks than the
total U.S. average and a higher than average share of blacks. If only one of
these conditions were met, the labor market area was not classified as high
black density. Details about the classification of specific labor market areas
are available from the authors.
The results of this analytical procedure are available from the authors.
Blacks are the most dispersed groups, with seventy-three concentrated labor
market areas containing 75 percent of all blacks, and Cubans are the least
dispersed, with 83 percent living in just seventeen areas. The percentage of
each group living in concentrated labor market areas is fairly similar, ranging
from a low of 68 percent for American Indians (sixty-two areas) to 85 percent
for Mexicans (forty-nine areas). There were thirty-five concentrated labor
market areas with 82 percent of the Puerto Ricans and forty areas with 72
percent of the other Hispanics.
Appendix B
Estimation of Ethnic Job Queues
For the estimation of the ethnic job typology, we used a sample from the 1970
Public Use Microdata Files of Hispanic, black, American Indian, and non-
156 Marta Tienda and Franklin D. Wilson
Hispanic white men aged sixteen to twenty-four who worked within the last
five years preceding the Census and had nonmissing industry and occupation
in 1970. First, we arrayed the data into a 6 X 6 X 5 X 3 X 3 matrix rep-
resenting six ethnic groups, six industry sectors, five occupational groups,
three education groups, and three age groups. Table 5B.2 presents the indus-
try by occupation distribution for the five minority groups.
To establish whether nonwhites are over- or underrepresented relative to
non-Hispanic whites, we computed a log-linear analysis that establishes
the associations among ethnicity, occupation, industry, education, and age
groups. We estimated a saturated log-linear model of the form:
where
A = ethnicity (i = 1, . . . , 6);
B = industry sector ( j = 1 , . . . ,6);
C = occupation group ( k = 1, . . . 3;
D = education ( I = 1, . . . ,3);
E = agegroup(rn = 1, . . . ,3).
In its abbreviated notation, the hierarchical model is:
[ABCDE]
From the 154 estimated parameters,I9 we computed the ethnic typing pa-
rameters for twenty-five occupation X industry cells for Mexicans, Puerto
Ricans, Cubans, blacks, American Indians, and non-Hispanic whites using
the following formula:
fori = 1 , . . . , 6.
The first three tau’s indicate the net effects of occupation by industry, and
the second three tau’s denote the interaction effects of occupation by industry
with ethnicity. All these parameters have been purged of the main and inter-
action effects of age and education. Ethnic-job-typing effects were calculated
as the ratio of the total effects of occupation by industry for each ethnic mi-
nority group relative to that of non-Hispanic whites. Since the net effects
of occupation by industry are uniform for all ethnic groups, our ethnic-
typing effects represent the ratio of interaction effects of occupa-
tion X industry x ethnicity.
The results of these computations, reported in table 5B. 1, provided the ba-
157 Migration, Ethnicity, and Labor Force Activity
Table 5B.l Ethnic Job ppi ng Effects by Ethnicity
Distributive Producer Social Personal
Occupational Groups Extractive Transformative Services Services Services Services
Mexican:
Upper nonmanual
Lower nonmanual
Upper manual
Lower manual
Farmer
Puerto Rican:
Upper nonmanual
Lower nonmanual
Upper manual
Lower manual
Farmer
Upper nonmanual
Lower nonmanual
Upper manual
Lower manual
Fanner
Upper nonmanual
Lower nonmanual
Upper manual
Lower manual
Farmer
American Indian:
Upper nonmanual
Lower nonmanual
Upper manual
Lower manual
Fanner
Cuban:
Black:
,971
2.040
2.143
2.158
1.146
,930
3.442
,932
1.946
,359
1.118
4.137
,992
2.367
,329
,399
.738
,938
1.636
.542
2.202
5.170
3.219
3.847
.920
,432
,615
,706
1.255
. . .
,217
,649
,618
1.021
. . .
,456
,600
,657
,883
. . .
,324
,717
1.191
2.408
1
. . .
.245
.405
.565
1.585
. . .
,552
,550
1.117
1.500
. . .
,361
,725
,595
1.524
. . .
,488
.693
.697
1.264
. . .
.289
,529
1.146
2.494
. . .
.334
,307
.554
1.490
. . .
.627
.430
1.817
1.135
. . .
.588
.722
2.734
3.571
. . .
.771
,848
1.883
1.812
. . .
,449
,525
1.676
2.729
. . .
,393
,388
2.353
2.210
. . .
,626
,772
1.433
1.061
. . .
,513
1 . w2
,983
1.173
. . .
,513
,676
1.030
,694
. . .
1.080
1.794
1.506
1.940
. . .
,626
,903
1.438
1.068
. . .
,722
1.282
1.379
1.528
. . .
,801
2.728
1.785
2.173
. . .
1.034
3.068
1.618
2.080
. . .
,823
,890
1.622
1.931
. . .
,613
1.667
1.646
,884
. . .
Source: 1970 PUMS Files. All men who worked between 1965 and 1970 and had valid industry and
occupation codes.
Note: Effects are ratios of Tau ( 7) parameters for each ethnic group relative to whites, as described in
App. B. Cut points for over, under, or equally represented are established at .5 and 1.5.
*These cells are structurally impossible because farm laborers occur only in the extractive sector.
sis for the trichotomous representation of ethnic job queues analyzed in the
text. Specifically, ethnic-typed jobs included those with tau ratios less than or
equal to 1.5; Anglo-typed jobs included those with tau ratios less than or equal
to .5; and nondifferentially preferred jobs categories obtained scores between
.5 and 1.5 exclusive. These cut points provide very conservative profiles of
ethnic job queues. A more liberal definition of the tolerable limits of ethnic
job typing would use .33 and 1.67 as the relative cut points. However, we
used the more conservative measures in our analysis.
158 Marta Tienda and Franklin D. Wilson
Table 5B.2 1980 Employment Classification of Men Aged Ikrenty-Five
to Sixty-Four
National Origin
Sector and Puerto American
Occupation Mexican Rican Cuban Black Indian
Extractive
Upper nonmanual
Lower nonmanual
Upper manual
Lower manual
Farmer
Upper nonmanual
Lower nonmanual
Upper manual
Lower manual
Farmer
Distributive services
Upper nonmanual
Lower nonmanual
Upper manual
Lower manual
Farmer
Producer services
Upper nonmanual
Lower nonmanual
Upper manual
Lower manual
Fanner
Upper nonmanual
Lower nonmanual
Upper manual
Lower manual
Fanner
Personal services
Upper nonmanual
Lower nonmanual
Upper manual
Lower manual
Farmer
Transformative
Social services
Totals
12.6
.4
.o
2.8
2.0
7.4
44.8
3.2
1.5
32.5
7.6
NA
18.2
2.0
2.9
10.0
3.2
NA
3.8
1 .o
.9
.7
1.1
NA
11.9
3.6
1.6
2.5
4.3
NA
8.8
1.1
. I
3.2
4.3
NA
100.0
2.3
.2
.o
.3
.8
1.1
42.4
2.4
2.6
32.5
4.8
NA
20.2
2.9
4.2
9.6
3.5
NA
9.7
2.7
I .8
1.2
3.9
NA
15.0
3.5
2.9
1.7
6.9
NA
10.4
1.3
. 5
2.9
5.7
NA
100.0
2.0
.2
.o
.2
.8
.6
38.0
6.7
2.5
25.0
3.8
NA
26.3
6.7
7.1
9.9
2.6
NA
10.6
4.8
2.9
.7
2.2
NA
11.5
6.0
1.4
1.1
3.0
NA
11.5
2.3
.6
3.4
5.2
NA
100.0
3.8
.2
. I
.5
.9
2.0
44.3
2.4
2.4
30.3
9.3
NA
20.1
1.8
3.6
10.6
4.2
NA
5.8
1.2
1.4
.8
2.5
NA
19.1
5.7
3.2
2.8
7.4
NA
6.9
1 .o
.2
2.2
3.4
NA
100.0
9.2
. 6
.1
2.8
2.0
3.6
41.9
5.0
1.4
28.1
1. 5
NA
17.0
2.8
2.9
8.7
2.7
NA
4.9
1.8
1 .o
.8
1.3
NA
21.2
7.0
2.3
4.3
7.6
NA
5.8
1 .o
.2
2.0
2.5
NA
100.0
Source: 1980 5% A Sample Public Use Micro-data Samples, migrant subsample.
Note: Upper nonmanual includes professionals, semiprofessionals, and managers. Lower non-
manual includes clericals and sales. Upper manual includes crafts and operatives. Lower manual
includes service workers and laborers. Farmer includes farmers and farm laborers.
159 Migration, Ethnicity, and Labor Force Activity
Appendix C
Variables Included in Logit, Probit and OLG
Regressions
Migration Type
Categorical variables coded as dummies for four types of density:
Intradensity High Moves within high-ethnic-density SMSAs or
nonmetropolitan county groups
Intradensity Low Moves within low-ethnic-density SMSAs or
nonmetropolitan county groups
Concentrated Moves from low- to high-ethnic-density
SMSAs or nonmetropolitan county groups
Dispersed Moves from high- to low-ethnic-density
SMSAs or nonmetropolitan groups
Nonmigrants No residence changes across SMSA bounda-
ries
Ethnic Job Typing
sures calculated separately for each national origin group (see App. B):
Ethnic Typed
Categorical variables coded as dummies for two preferential statuses; mea-
Denotes job cells in which ethnic workers
were overrepresented relative to non-
Hispanic whites in 1970, unique for each
group
Denotes job cells in which ethnic workers
were underrepresented relative to non-
Hispanic whites in 1970, unique for each
group
Denotes job cells in which ethnic workers
were approximately equally represented
relative to whites in 1970, unique for each
group
Anglo Typed
Nontyped
Density
tion:
Categorical variables coded as dummies for two levels of ethnic concentra-
High
Low
If 1980 labor market areas meet criteria for
high concentration of Mexican, Puerto Ri-
can, Cuban, black, or American Indian
groups (see App. A)
Remaining labor market areas
160 Marta Tienda and Franklin D. Wilson
Human CapitallFamily Status
Weeks and Hours for earnings model only:
Household Head and Young Children for labor supply equations only;
Education
Experience
Experience Squared
English Ability
Foreign Birth
Married
Household Head
Young Children
Weeks
Hours
Dummy variables designating completion of
college, high school or some college or
less than high school
Labor market experience proxy derived as
(age-education-6)
Square of experience
Dummy variable denoting good or excellent
Dummy variable denoting foreign birth
Dummy variable denoting if respondent was
Dummy variable denoting if respondent was
Dummy variable denoting presence of chil-
Weeks worked last year
Usual hours worked
proficiency in English
married
household head
dren less than 6 years old
Labor Market
Definition of labor market in Appendix A; Area Unemployment Rate and
Dummy variable designating four regions of
residence: West, South, Northeast, and
North Central
Unemployment rate for SMSAs or nonmet-
ropolitan county groups
Mean wage rate for SMSAs or nonmetropol-
itan county groups
Dummy variable coded 1 if metropolitan res-
idence
Share of area labor force engaged in durable
and nondurable manufacturing
Industry for earnings model only:
Region
Area Unemployment Rate
Area Wage Rate
Metro
Industry
Notes
1. Affirmative action hiring on the basis of race, gender, and national origin is one
mechanism through which minority populations may be designated as preferred hires
for a specific set of jobs. The changes in minority employment patterns as a result of
161 Migration, Ethnicity, and Labor Force Activity
affirmative action laws and their enforcement are documented by Leonard (1984) and
Smith and Welch (1984).
2. As an example, Lieberson explains that, if the most favored group is 10 percent
of the population in one city and 20 percent in another city, the second-ranked group
can begin to fill jobs ranked at the ninetieth percentile in the first city and the eightieth
percentile in the second city. In other words, if the lowest-ranked group is large, then
members of this group will find jobs higher in the occupational queue because the
higher-ranked groups do not push as far down.
3. Alternatively, during periods of labor shortage, groups at the bottom would ex-
perience broadened employment opportunities because the preferred groups would not
fill all the traditional employment opportunities. Thus, employment shifts usually will
be most radical for groups at the bottom of the employment queue.
4. To save money, migration variables were coded for roughly half of all persons
aged 5 and over. Because the A file is based on a 5 percent sample of the total popula-
tion, we did not encounter sample size restrictions with the minority populations.
5. Restrictions pertaining to work status ensure that all individuals in the sample had
valid occupation and industry codes, which are needed to derive our ethnic job cate-
gories. The restriction on U.S. residence in 1975 was necessary for computing the
migration types. Although the ethnic concentration of origin countries of recent immi-
grants is uniformly high, it was not pertinent for the comparisons in our migration
typology, which focuses on internal moves. Finally, men in the military or enrolled in
college either in 1975 or 1980 were excluded because the groups have higher migration
propensities de facto, independent of the social and economic motivations underlying
these decisions.
6. The stringency of our sample restrictions prompted additional diagnostics of the
social and demographic characteristics of the excluded population. These analyses re-
vealed that men who never worked or who were in the military or enrolled in college
in 1980 tended to be younger and were more apt to be unmarried than their respective
subpopulations. Individuals who were not in the labor force in 1980 and who had last
worked before 1975 were older, on average, than their source subpopulation. Also,
since recent immigrants tend to be younger and less often married, the final Mexican
and he r t o Rican samples contained fewer men under age 30 and fewer unmarried men
than would be true if recent immigrants were not excluded from the analysis. This was
less true of Cuban immigrants, who tend to be older than all immigrants (on average).
7. Whether migration streams alter the ethnic composition of specific labor markets
depends on the differential net migration of minority and nonminority groups. Thus,
the ethnic density of a particular labor market may increase either because of net out-
migration of non-Hispanic whites or net in-migration of minorities. Ethnic density may
similarly decrease because of a net out-migration of minorities or net in-migration of
whites. Our focus, however, is not on the changing residential configuration per se but
rather on the significance of ethnic concentration for stratification processes or, more
specifically, the direction of migrant streams relative to the ethnic density of origin and
destination areas.
8. Our analysis of preference status categories was based on 1970-prior to the
migration interval studied-rather than 1980 industry by occupation classifications to
avoid endogeneity problems from confounding the causal ordering or migration and
resulting ethnic job queues. However, the actual classification of individuals among
the three categories is based on industry and occupation categories reported as of 1980,
at the end of the migration interval.
9. Since the models used to predict unemployment probabilities are identical, we do
not repeat them; we expect the effects of our key independent variables to be opposite
those produced for the labor force participation models.
10. This implies that demand for workers can be ethnicity specific, especially in
162 Marta Tienda and Franklin D. Wilson
contexts where recruitment is governed by an ethnic job queue or where the assump-
tions of homogeneous labor and perfect competition do not hold.
1 I . For the earnings models, we corrected for selection bias resulting from differ-
ences in the probability of being in the wage sample. Results are available on request.
12. This overrepresentation in producer and personal services reflects the large num-
bers of Cubans involved in the banking industry in Miami as well as the large share of
Cuban-owned and -operated enterprises employing a highly ethnic (Cuban) labor
force. See the discussion in Portes and Bach (1985).
13. A competing explanation is that the selection process governing the joint migra-
tion and employment decisions has not been adequately modeled. Elsewhere, we have
modeled employment and migration outcomes using a bivariate probit specification
that takes account of the selection processes and reached essentially the same conclu-
sions.
14. Sandefur analyzed the effects of migration on labor force participation sepa-
rately for endogamous, exogamous, and intermarried American Indian couples and
found uniformly nonsignificant effects for all three groups. This finding obtained de-
spite the very different rates of interstate migration for each of these three groups.
15. Although the result is not statistically reliable, association with ethnic-typed jobs
appears to increase unemployment for these two groups. These results are consistent
with evidence presented in Appendix table 5B. 1 showing that blacks and American
Indians were differentially concentrated in nonmanual jobs even after adjusting for
differences in their age and education composition. The alternative of not working may
be more attractive than holding such low-status jobs, but this conclusion cannot be
confirmed from these results.
16. To verify the robustness of these results, we substituted a single dichotomous
variable indicating nonmigrant status in the earnings equation and confirmed that the
logged annual earnings of migrants and nonmigrants did not differ significantly for any
of the minority groups analyzed. However, by differentiating among migration types,
we were able to detect some significant ethnic differences in the influence of migration
on earnings.
17. For American Indians, dispersed flows were about as pervasive as concentrated
flows and amounted to about 16 percent of all moves.
18. Determining boundaries of labor markets was a complicated process. The basic
unit is the SMSA, which we reconstructed from county group codes. Then nonmetro-
politan areas within states were divided up into two or three areas. The result was 414
labor market areas: 310 SMSAs and 104 nonmetropolitan areas. Individual area codes
were determined not by the SMSA code but by a combination of the state and county
group codes. This caused problems when county groups spread across two or more
SMSAs or when SMSAs crossed state boundaries. The decision rules used to allocate
county groups are available from the authors.
19. Because farm occupations are found only in the extractive sector, the industry
occupation interaction yields twenty five rather than thirty cells. By deleting the five
structurally impossible cells for all groups, we estimated thirty fewer parameters than
model 1 implies.
References
Bean, F. D., and M. Tienda. 1987. The Hispanic Population of the United States. New
York: Russell Sage.
163 Migration, Ethnicity, and Labor Force Activity
Hechter, M. 1978. Group formation and the cultural division of labor. American Jour-
nal of Sociology 84(2):293-317.
Leonard, J. 1984. The impact of affirmative action on employment. Journal of Labor
Economics 2(4):439-63.
Lieberson, S. 1980. A Piece ofthe Pie. Berkeley and Los Angeles: University of Cali-
fornia Press.
Nelson, C. , and M. Tienda. 1985. The structuring of Hispanic ethnicity: Historical
and contemporary perspectives. Ethnic and Racial Studies 8( 1):49-74.
Petersen, T. 1985. A comment on presenting results from logit and probit. American
Sociological Review 50( 1): 130-3 1.
Portes, A., and R. L. Bach. 1985. Latin Journey: Cuban andMexican Immigrants in
the United States. Berkeley and Los Angeles: University of California Press.
Sandefur, G. 1986. American Indian migration and economic opportunities. Interna-
tional Migration Review 20( l): 55-68.
Smith, J. P., and F. Welch. 1984. Affirmative action and labor markets. Journal of
Labor Economics 2(2): 269-301.
Snipp, C. M., and G. Sandefur. 1988. Earnings of American Indians and Alaskan
Natives: The Effects of Residence and Migration. Social Forces 66:994-1008.
Tienda, M. 1989. Puerto Ricans and the underclass debate. Annals of the American
Academy of Political and Social Sciences SOl(January): 105-19.
Tienda, M., and L. Jensen. 1988. Poverty and minorities: A quarter-century profile of
color and socioeconomic disadvantage. In Divided Opportunities: Minorities, Pov-
erty and Social Policy, ed. G. Sandefur and M. Tienda, 23-61. New York: Plenum.
Tienda, M., and D. T. Lii. 1987. Minority concentration and earnings inequal-
ity: Blacks, Hispanics and Asians compared. American Journal of Sociology
93(July): 141-65.
Wilson, W. J. 1987. The Truly Disadvantaged. Chicago: University of Chicago Press.
This Page Intentionally Left Blank
Impact of Immigration, Trade,
and Capital Flows on the U.S.
Labor Market
This Page Intentionally Left Blank
6 Labor Market Adjustments to
Increased Immigration
Robert J. LaLonde and Robert H. Topel
During the 1970s, immigration to the United States was higher than in any
decade since the 1920s, raising the number of immigrants in the U.S. labor
market by 45 percent. The flow of new immigrants has actually increased
during the 1980s, and in many areas immigration is a major component of
labor force growth. These facts are central to the current debate over immigra-
tion policy since it is widely believed that new immigrants have deleterious
effects on the labor market opportunities of native Americans.’ For example,
if the main costs of immigration are borne by less-skilled natives through re-
duced earnings and employment opportunities, the case for immigration con-
trols and redistributive policies is strengthened.2 In contrast, if the labor mar-
ket easily absorbs new immigrants without serious distributional effects, these
policy options are less attractive.
Increased labor supply due to immigration enhances the welfare of the typ-
ical consumer, but it also creates adverse distributional effects among workers
whose skills compete with those of immigrants. Yet it is difficult to argue that
even the large flow of immigrants in the 1970s could have had a substantial
Robert J. LaLonde is associate professor of industrial relations at the Graduate School of Busi-
ness, University of Chicago, and a faculty research associate of the National Bureau of Economic
Research. Robert H. Topel is professor of business economics at the Graduate School of Business,
University of Chicago, and a faculty research associate of the National Bureau of Economic Re-
search.
This research was supported by the Olin Foundation through a grant to the Center for the Study
of the Economy and the State and by the Ford Foundation through a grant to NBER. Topel also
acknowledges support from the National Science Foundation and the William Fishman Research
Fund. LaLonde also acknowledges support from the Graduate School of Business at the Univer-
sity of Chicago. The authors thank Dan Hamermesh, Richard Freeman, Arlene Holen, Kevin
Murphy, Finis Welch, and workshop participants at the Board of Governors of the Federal Re-
serve, Chicago, Georgetown, Maryland, NBER, Northwestern, and the University of California,
Santa Barbara and Los Angeles, for helpful comments. William Anderson assisted with the cal-
culations. Errors are the authors’.
167
168 Robert J. LaLonde and Robert H. Topel
effect on the U. S. labor market. New immigrants of all ages contributed only
about 2.5 million extra persons to the labor force over this period, compared
to the concomitant increase of twenty million among workers aged 32 or less
that was caused by the baby boom and increased labor force participation by
young women.3 At this level of aggregation, immigration would have only a
second-order effect on the labor market. However, nearly half of all new im-
migrants live in six metropolitan areas, so that the potential effects of in-
creased immigration may be similarly concentrated in local labor markets. In
fact, this feature of immigration is the focus of the current policy debate.
Those who believe that immigration has important effects are concerned not
with the earnings and employment of the typical worker, who probably gains,
but instead with the prospects for certain groups who reside in specific areas,
such as young blacks in Miami or native Hispanics in Los Angeles.
Our analysis exploits this geographic diversity to study the effect of immi-
gration on local labor markets. In our view, the empirical issue is how in-
creased immigration affects labor market opportunities for workers who are
close substitutes for immigrants. Since theory offers little guidance about
which groups these are, our strategy is to analyze the effect of immigration on
labor market outcomes for workers who are a priori similar to new immi-
grants-other members of current and past immigrant cohorts of similar eth-
nicity. Substitution effects for these workers will generally dominate those for
nonimmigrant labor, so estimates of these effects will serve as upper bounds
for the effect of immigration on labor market outcomes for native^.^ We also
test these bounds by estimating corresponding substitution effects for young
blacks and native Hispanics.
The empirical analysis uses earnings and employment data for immigrants
and native-born workers from the 1970 and 1980 Censuses. To estimate the
effect of immigration on labor market outcomes, we rely on three distinct
sources of variation in the relative importance of immigrants in local labor
markets: (i) the share of all immigrants within a locale; (ii) the share of new
immigrants; and (iii) the changes in these immigrant shares between 1970 and
1980. The first source of variation, the immigrants’ labor force shares, will
generate corresponding differences in immigrant and nonimmigrant earnings
if the geographic location of new immigration is exogenous and if nonimmi-
grant or other factor mobility does not fully arbitrage wage differences in mar-
ket equilibrium. In other words, labor supply to a locale must be inelastic, at
least in the short run. We find that earnings of both new and old immigrant
cohorts are lower in areas where immigrants-especially new immigrants-
form a large or growing portion of the local labor force. This relation suggests
important effects of immigration on earnings, although an alternative expla-
nation is that less-skilled immigrants locate in areas where immigrants form a
large share of the labor force.
The second, and more important, source of variation in immigration across
areas is the labor force shares of different immigrant cohorts within locales.
169 Labor Market Adjustments to Increased Immigration
Variations over time in the rate and location of immigration have generated
substantial differences between locales in average arrival times of immigrants.
We find that an increase in the relative share of an immigrant cohort within an
area (e.g., immigrants who arrived between 1970 and 1974) causes a corre-
sponding decline in the wages and earnings of members of that cohort. Our
best estimate is that a doubling of new immigration to a locale would reduce
new immigrant annual earnings by less than 3 percent. This modest earnings
disadvantage for members of large immigrant cohorts dissipates with time in
the United States. It appears that immigrants assimilate into the broader labor
market as they accumulate skills that are appropriate to the U.S. labor market.
Our evidence also indicates that new immigration reduces the earnings of ear-
lier immigrant cohorts. Thus, “new” and “old” immigrants are substitutes.
However, as theory predicts, these substitution effects on wages are found to
be smaller for older immigrant cohorts, which is also consistent with the as-
similation of immigrants. We regard these results as evidence for the existence
of within-market substitution effects of immigration on wages.
In light of these results, it is not surprising that the effect of immigration on
natives appears to be minor. For young (aged 16-34) blacks, we find a small
negative effect of immigration on relative earnings. Our largest estimate is that
a long-term doubling of immigration to an area may reduce the annual earn-
ings of young blacks by about 4 percent, with much smaller effects on young
Hispanics. Since market outcomes for young blacks and Hispanics are likely
to be the most sensitive to changes in the supply of immigrants, we think this
evidence weakens the case for serious distributional effects of immigration.
These conclusions are reinforced by estimates derived from the third source
of variation: within-market changes in the labor force shares of immigrant
workers generated by new immigration. Interarea mobility will arbitrage geo-
graphic wage differentials in the long run, but we view the accelerated pace of
immigration during the 1970s as an exogenous increase in supply that in the
short run will generate relative wage adjustments in areas of unusually heavy
immigration. Thus, we expect a decline in the relative earnings of immigrants
(and close substitutes) between 1970 and 1980, and we expect that this de-
cline will be concentrated in areas with unusually heavy immigration as well
as among more recent immigrant cohorts. We find evidence for these effects:
our best estimate from this experiment is that a doubling of the number of
recent immigrants within a locale would reduce their relative earnings by
about 3 percent. The strong correspondence between these panel estimates
and those generated from a single cross section increases our confidence in
the results.
Our broad assessment of this evidence is that immigration flows do affect
earnings and employment of immigrants and nonimmigrants. Members of
large immigrant cohorts suffer slightly reduced earnings, especially on first
arriving in the United States. But it appears that immigrants assimilate rapidly,
and important effects on nonimmigrants are difficult to find. We conclude that
170 Robert J. LaLonde and Robert H. Topel
recent increases in the pace of immigration have been easily absorbed by the
labor market so that distributional consequences are not a firm basis for poli-
cies that would further restrict immigration to the United States.
6.1 The Empirical Setting
The geographic distribution of immigrants, especially new immigrants, is
central to our analysis. Table 6.1 illustrates the geographic concentration of
both new and old immigrants, showing the arrival date of the stock of immi-
grants in six “gateway” metropolitan areas in 1970 and 1980. These areas
account for about 40 percent of all immigrants in both years and nearly half
(47 percent) of all recent immigrants (those arriving within ten years of the
survey date). Reflecting the increased flow of immigration, the population
Table 6.1 Immigrants in the United States and Six Gateway Cities,
1970 and 1980
United Los New San
States Chicago Houston Angeles Miami York Francisco
Foreign born as
% of population:
1970 4.8
1980 6.2
Immigrants in SMSA
as % of immigrants
in U.S.:
1970
1980
SMSA as % of all recent
immigrants in U.S.:
1970
1980
Proportion of immi-
grants in SMSA
arriving in past:
0-10 years:
Recent immigrants in
1970 29.3
1980 39.5
1970 18.1
1980 22.2
10-20 years:
8.1
10.5
5.8
5.3
5.5
5.6
37.9
41.9
22.8
19.8
2.6
1.6
.5
1.6
.8
2.6
41.7
64.5
19.9
18.3
11.2
22.3
8.1
11.8
11.6
17.1
42.0
57.0
21.8
21.8
24.4 15.0
35.6 21.3
3.2 17.8
4.1 13.8
6.9 18.4
3.7 13.6
63.9 30.4
35.6 38.7
10.7 16.0
45.2 24.6
11.0
15.7
3.5
3.6
4.5
3.9
38.2
43.1
20. I
24. I
Sources: U.S. Census of Population, 1980, General Social and Economic Characteristics, United
States Summary and State Summaries, table 99; and U.S. Census of Population, 1970, Charac-
teristics of the Population, table 144. The six cities in the table accounted for 40 percent of all
immigrants in the United States. Note that these statistics, which report the importance of immi-
grants in the total population and in different SMSAs, understate the importance of immigrants
in the work force of these cities since a larger share of immigrants is in the labor force than is the
case for the native population as a whole.
171 Labor Market Adjustments to Increased Immigration
share of immigrants increased by 30 percent during the 1970s; by 1980, about
40 percent of all immigrants in the United States had arrived during the pre-
vious decade, up from 30 percent in 1970. This estimate is widely distributed
across cities: in Los Angeles, more than half of all immigrants arrived during
the 1970s, and the population share of immigrants doubled over the decade.
Currently, immigrants account for nearly a quarter of the male labor force in
the Los Angeles standard metropolitan statistical area (SMSA) and more than
a third in Miami.
Table 6.2 offers a more detailed picture of the geographic distribution of
immigrants and their importance as a source of labor force growth. The first
two columns report the distributions of “new” and “old’ male immigrants for
SMSAs that account for at least 1 percent of all foreign-born persons in the
work force. The remarkable correspondence between the flow distribution for
new immigrants (col. 1) and the distribution of the stock that arrived before
1970 illustrates the importance of immigrant “enclaves”: new immigrants go
where previous ones went. Separate distributions for persons of European,
Mexican, and Asian origin confirm the relation and show that enclaves are
primarily ethnic in origin. Because of this factor, the geographic distribution
of immigrants tends to replicate itself through time. Thus, there is little evi-
dence of wide swings in the geographic distribution of immigrants over time,
which partly justifies our assumption, exploited below, that the locational de-
cisions of new immigrants are exogenous.
The last two columns of table 6.2 show the importance of immigration as a
source of labor force growth in these areas. Though immigration is a minor
factor in economy-wide labor force growth, it is the most important factor
contributing to the growth of some markets. For example, in Los Angeles,
immigration during the 1970s would in itself have caused a 31 percent in-
crease in the local labor force, and new immigrants accounted for nearly two-
thirds of the actual increase in the labor force during this period. Of course,
these estimates would be even larger if the base population were restricted to
those with skills that are similar to those of new immigrants.
Our econometric analysis will treat immigrants from different arrival co-
horts as imperfect substitutes in production. This assumption will hold if
either (i) immigrants “assimilate” with time in the United States in the sense
of acquiring skills relevant to the American market or (ii) different arrival
cohorts bring qualitatively different skills to the United States. Table 6.3 ex-
amines these possibilities, presenting differences between the mean log
weekly wages of immigrants and white natives of the same age, by Census
year. The table demonstrates three important facts. First, within a Census
year, relative earnings profiles appear to reflect assimilation in the sense that
earlier arrivals earn more (Chiswick 1978). Second, however, assimilation
appears to be much less important if an arrival cohort is followed through time
(Borjas 1985). For example, workers who were 25 to 34 years old in 1970
earned about 29 percent less than their native white counterparts, and by 1980
Table 6.2 New Immigration Flows and Stocks (the distribution of new and old male immigrants)
1970-80 Immigrants as
All Immigrants Europeans Mexicans Asians a Proportion of:
Share of Share of Share of Share of Share of Share of Share of Share of 1970 Labor 1970-80 Labor
1970-80 pre-1970 1970-80 pre-1970 1970-80 pre-1970 1970-80 pre-1970 Force Force Growth
Anaheim 3.4 2.4 1.4 2.3 6.9 4.3 3.0 4.4 23.3 14.1
Boston 1.8 2.3 4.6 3.5 . . . .1 1.5 2.3 7.3 17.6
Chicago 8.3 7.7 8.8 9.0 11.9 10.3 7.8 6.2 10.5 21.5
Cleveland .6 1.2 1.7 2.0 . . . . . . 1 .o .6 2.8 23.6
DallaslFort Worth 2.1 1 .o 1.1 .6 4.3 2.8 1.6 .8 7.5 6.5
Detroit 1.4 3.1 3.2 4.9 .1 .7 1.5 1.8 4.0 8.8
Houston 4.4 2.0 1.4 .8 8.5 6.1 4.1 2.6 17.3 9.5
Los Angeles 22.1 12.7 6.8 7.6 42.2 34.9 19.0 14.7 31.1 65 .O
Miami 3.9 6.1 1.6 1 .o .2 .1 .5 .2 38.0 37.7
New York 17.4 19.7 22.0 22.4 .5 .6 16.3 14.9 5.7 -56.4
Newark 2.2 2.3 4.1 2.9 . I .1 1.6 1.5 10.8 22.4
Philadelphia 1.5 1.9 3.6 2.6 .2 .1 2.3 1.5 3.4 16.2
Seattle .7 1.2 1.1 1.7 .2 .2 1.7 1.8 6.9 8.1
San Diego 1.9 1.8 .9 1.4 3.6 5.6 1.6 1.5 18.9 14.7
San Francisco 4.1 4.1 2.3 3.6 2.3 4.1 10.2 16.1 15.9 30.4
San Jose I .8 1.6 I .4 1.3 1.7 2.6 3.3 5.6 19.2 14.7
Washington, D.C. 2.2 1.6 3.0 1.4 . . . .2 2.7 3.3 10.4 16.8
Source: Census of Population, Public Use Sample B (see the Appendix for selection criteria used to create the sample for the table).
Note: This table lists all SMSAs with at least a 1 percent share of either new or old immigrants in their work forces. All other U.S. SMSAs have a smaller than 1
percent share of the immigrant population.
Jersey City 1.2 1.4 1.6 1.1 . . . .1 .6 .4 . . . . . .
173 Labor Market Adjustments to Increased Immigration
Table 6.3 Relative Weekly Wages of Male Immigrants by Place of Origin and
Years Since Immigration, 1970 and 1980
Years Since Immigration
Place of Origin 0-5 6-10 11-15 16-20
All immigrants:
25-34
35-44
3 5 4
45-54
1970:
1980:
EurOpe:
1970
25-34
3 5 4
35-44
45-54
1980:
Asia:
1970:
25-34
3 5 4
35-44
45-54
1980:
Mideast:
1970:
25-34
3 5 4
3 5 4
45-54
1980:
Mexico:
1970:
25-34
3 5 4
3 5 4
45-54
1980:
Other Latin American:
1970:
25-34
35-44
35-44
45-54
1980:
- .292
- .409
- ,473
- ,582
-.133
- ,229
- . I 4 4
- ,356
- ,206
- .400
- .472
- .549
- ,244
- .081
- .274
- ,262
- .659
- ,856
- ,983
- ,927
- ,436
- ,559
- .609
- ,895
-.I19
- .244
- ,349
- ,529
,017
- ,090
- .210
- ,259
- ,016
-.I52
- ,065
- .360
- .040
- .171
- ,001
- .243
-.411
- ,623
- .764
- ,930
- .196
- .354
- ,492
- .740
- .075
- .095
- .220
- ,385
- ,022
- .008
- .032
- .214
.061
,103
.058
- .288
.147
.123
,087
,009
- ,353
- ,398
- ,582
- ,605
- ,150
- ,332
- ,446
- .535
0
- ,055
-.122
- .206
,085
.004
- ,015
- .050
- ,486
- ,039
,036
. O H
,122
- ,209
.123
,129
- .322
- ,351
- ,399
- ,669
- ,083
- ,034
- .193
- ,355
Sources: Public Use Files from the 1970 and 1980 Censuses. For sample selection criteria, see
the Appendix.
Note: Estimates are differences between mean log weekly earnings of immigrants and of white
natives in the indicated age category.
174 Robert J. LaLonde and Robert H. Tope1
they still earned 22 percent less. This estimate of relative assimilation is much
smaller than what either cross section would imply, and it means that the av-
erage skills of successive arrival cohorts may have declined through time. The
third point is that relative immigrant wages declined during the 1970s. This
holds in virtually all age categories in the table. Again, one possibility is
declining immigrant quality, but another is simple price adjustments in re-
sponse to market forces, driven perhaps by the increased supply of new im-
migrants.
Do wages respond to immigration flows? Table 6.4 reports relative weekly
wages in 119 SMSAs for various immigrant groups in 1970 and 1980. The
estimates are tabulated by the proportion of the local labor force in 1980 that
is accounted for by immigrants who arrived during the previous decade. “Top
third” in the table refers to the set of SMSAs with the largest immigration
rates, which account for one-third of all immigration over the decade.
“Middle third” refers to the next most immigrant-intensive SMSAs, which
account for another third of total immigration, and so on. Relative weekly
wages are calculated as the difference between the mean log wages of immi-
grants and of white males of the same age, within each SMSA. The data show
that relative immigrant wages were dramatically lower in labor markets where
new immigration flows were largest. For example, row 6 of the table shows
that weekly wages of recent immigrants (those who arrived in the last ten
years) fell by over 20 percent relative to white natives in cities with the highest
immigration rates. The comparable estimate in cities with the lowest immigra-
tion rates is only 7.5 percent. One explanation for this pattern is that wages
adjust to increases in supply, at least in the short run. An alternative explana-
tion with much different implications is that less-skilled immigrants locate in
immigrant enclaves, so that large immigrant populations are less skilled, on
average. Our econometric approach seeks to isolate the first of these effects.
6.2 Theoretical Framework
In what follows, we view immigration flows as exogenous shifts in the
supply of labor to geographically defined labor markets. So long as immi-
grants form labor aggregates that substitute imperfectly for others in local
production, these supply shifts will have their largest effect on immigrant
earnings, with declining effects on other input aggregates as substitution pos-
sibilities decline. Thus, for example, an increase in new immigration to a
locale will reduce the relative earnings of new immigrants and to a lesser
extent the earnings of others for whom immigrants are substitutes. The empir-
ical issue is the magnitude of these effect^.^
If immigrants have important effects on other market participants, it must
be via the substitution effects just mentioned. An a priori restriction that we
find reasonable is that the best substitute for the representative immigrant is
another immigrant. At the other extreme, the representative native may be a
175 Labor Market Adjustments to Increased Immigration
Table 6.4 Relative Weekly Wages of Male Immigrants and White Natives in
SMSAs Ranked by Shares of Immigrants in Work Forces
Top Third Middle Third Bottom Third
1, Immigrants arriving after
1970, in 1980
25-34
35-44
45-54
2. All immigrants in 1980
3. Pre-1970 immigrants in 1980
4. All immigrants in 1970
5. Immigrants arriving after
6. Change in relative wages of
7. Change in relative wages of
1960, in 1970
recent immigrants (1-5)
all immigrants (2-4)
- ,642
- ,500
- ,640
- .765
- ,422
- ,210
- .278
- ,438
- .204
- . I 4 4
- ,491
- ,395
- ,537
- .662
- ,291
- ,159
- ,184
- ,352
- .I39
-. 113
- ,284
- ,244
- ,257
- ,419
- ,078
- .046
- ,036
- ,208
- ,075
- ,042
Sources: Public Use Files from the 1970 and 1980 Censuses. For sample selection criteria, see
the Appendix.
Note: Estimates in the table are differences between geometric mean wages for immigrants and
white males within each SMSA. SMSAs are ranked by the share of the male labor force ac-
counted for by immigrants who arrived between 1970 and 1980. “Top third” refers to SMSAs
with the largest immigration rates that together account for one-third of all post-1970 immigrants.
“Middle third‘’ refers to those SMSAs with the next largest rates of immigration that together
account for another third of all new immigrants. Finally, the column labeled “bottom third” refers
to SMSAs with the lowest rates of immigration that together account for the remaining one-third
of new immigrants.
very poor substitute for new immigrants, who enter the United States with
skills (e.g., language and institutional knowledge) that typically are less val-
ued in the American market. Yet over time the immigrants assimilate. In our
analysis, this assimilation entails greater ease of substitution between an im-
migrant cohort and native workers as their time in the United States accumu-
lates.6 Thus, substitution between old and new immigrant cohorts is also im-
perfect. We further expect these intercohort substitution effects to dominate
those between new immigrants and (most) native workers. The following
model formalizes these ideas and serves to guide the subsequent empirical
work.
We assume the existence of a large number of geographically distinct labor
markets.’ Immigrant and nonimmigrant labor are combined in a concave local
production function represented by
In equation (l), Y, refers to total output produced in locale c (empirically an
SMSA), and M,. is total human capital supplied by labor aggregatej in locale
c. In our discussion, we assume that the labor aggregates in g(.) include im-
migrant arrival cohorts ( j = 1, . . . , k - 1) plus nonimmigrant labor ( j =
176 Robert J. LaLonde and Robert H. Topel
k) as inputs, though empirical implementation requires further judgments
about substitution possibilities. Thus, some natives who are thought to be
close substitutes for immigrants-young Hispanics or blacks, for example-
can be included as separate factors. Another possibility is to allow immigrant
groups of different ethnicities to form separate inputs in the production func-
tion. The (weak) separability assumption in equation (1) is maintained
throughout. Given our specification of g(.), h( . ) contains capital and other
resources that are incidental to the analysis. The parameters 8, and ac are
locale-specific factor-neutral shifters of the effective quantities of labor and
other factors. For example, these shifters may represent forces that shift the
local demand for labor. Given varying sizes of cities to which the model may
be applied, a plausible assumption about g(.) is that it has constant returns, so
that doubling all labor quantities leaves relative wages unchanged within any
locale.8
Assume for the moment that each member of arrival cohort j supplies one
unit of relevant human capital. Then the marginal product (wage) of group j
workers at locale c is
(2)
where subscripts to functions denote partial derivatives with respect to the
indicated argument. The separability assumption in equation (1) implies that
other inputs enter the marginal product of labor only through F, ( . ) . Thus,
shifts in 8,, a,, or other nonlabor components leave relative wages for labor
inputs unchanged. Given (2), the log wage of groupj workers in locale c is
w, = F1(.)8,gj(Mc1, . . f 9 MC,h
(3) wcj = In [F,(.)8,1 + In gj (Mc, , . . . , Mck).
The first term on the right-hand side of equation (3) is an area-specific term
and is independent ofj; it is fixed for all labor inputs within a locale. Equation
(3) is the basis for our empirical analysis.
The first step toward an empirical specification of (3) is to replace In
[ F, ( . ) 8, ] with an area-specific fixed effect, p, and to expand In gj (Mc, , . . . ,
Mck) to first order in logs:
(4) wcj = p, + &yji In Mci .
In (4), the parameters yji (i = 1, 2, . . . , k) are “elasticities of complemen-
tarity” (8 In W, / 8 In MCi ) that satisfy Xi yji = 0 if there are constant returns.
Aside from this homogeneity condition, the only restriction implied by theory
is y, < 0-an increase in the supply of groupj workers reduces their wage.9
However, if j indexes cohorts by their time in the United States, we expect
yi j < 0 (for i # j ) with effects that dissipate as I j - il increases. In the lan-
guage of demand theory, adjacent immigrant cohorts should be q-substitutes
(see Hamermesh 1986). In other words, recent immigrants offer the greatest
substitution possibilities for new immigrants, so the yij will trace out an as-
similation profile of wage adjustments. Complementarity is also a possibility
177 Labor Market Adjustments to Increased Immigration
( yi j > 0). For example, a large enclave of past immigrants may improve mar-
ket opportunities for new immigrants, especially when language and cultural
ties are important. These restrictions are tested below.
To complete the empirical specification, we drop the assumption that each
individual contributes a single unit of human capital to the stock Mj . We as-
sume that an individual’s stock of human capital, m, depends on his character-
istics, X, so that, for person 1 in cohortj and city c, mcjl = exp(X,G + Pj +
Po + E, ~~}. The cohort effects Pj allow for both assimilation (earlier cohorts
have acquired skills relevant to the U.S. market) and differences in the quality
of immigrants over time. Similarly, the “origin effects” Po control for differ-
ences in average immigrant characteristics between broadly defined places of
origin. With this assumption, the log wage of individual 1 is
wcjl = P, + Pj + Po + X,G + Ziyji In Mci + E,~,.
Three points about (5) are noteworthy. First, the appearance of locale ef-
fects (p,) in (5) implies that the yji’s capture shifts in the relative earnings of
different immigrant groups within a locale that are induced by changes in the
relative shares of immigrants. More precisely, with fixed city effects, the es-
timable substitution parameters are yji - yki , j = 1 , . . . , k - 1. Since k
refers to native workers as an aggregate, our maintained assumption is that
yki = 0; changes in the stocks of immigrants do not affect the wages of the
typical native worker, so relative wage adjustments capture the effects of in-
terest.I0 For example, yI1 < 0 implies that, in a locale where the market share
of new immigrants is large, wages of new immigrants will be low relative to
the earnings of other workers in that area. Thus, our analysis examines the
effect of immigration on rotations of the assimilation profile of immigrant
wages within locales. Sample selection due to unobservable differences across
areas in immigrant “quality” will not affect our results. For the same reason,
controlling for locale effects implies that our results are not affected by differ-
ences in demand conditions, local amenities, or the cost of living across mar-
kets, so long as these conditions have factor-neutral effects on the wages of
separate labor categories within a locale. Therefore, demand-induced shifts in
immigration to a locale are not an issue unless they have differential effects on
certain immigrant cohorts.
A second noteworthy point about equation (5) is that, while the estimated
area effects, pC, subsume wage adjustments for each locale, it is still true that
an increase in the total supply of immigrants will normally reduce immigrant
earnings relative to those of nonimmigrants. These relative wage adjustments
can be evaluated from ( 5) . Because the cross-substitution effects yji (for j #
i ) typically will be nonzero, an increase in the supply of all immigrants may
have a larger negative effect on immigrant earnings than would be implied by
own substitution effects (y,) alone.”
Finally, equation (5) controls for cohort (time of arrival) effects directly, so
that differences in immigrant quality over time do not influence the estimates
178 Robert J. LaLonde and Robert H. Topel
of yji. For example, if recent immigrant cohorts are less skilled than their
predecessors, a model like (5) that did not control for time in the United States
might attribute the entire decline of relative earnings among recent immigrant
cohorts to the increased relative supply of new immigrants. Model ( 5) is
not subject to this bias as long as within-cohort average quality is neutral with
respect to locale. Similar arguments apply to the presence of place of origin
effects, Po.
6.2.1
Despite these controls, model ( 5) arguably is inappropriate since mobility
of either natives or other factors may arbitrage geographic wage differentials
in the long run. Differences in immigrant shares can persist in equilibrium,
but, if factors are mobile, these differences have no implications for wage
differentials. This argument is less persuasive when applied to short-run ad-
justments to changes in the flow of immigrants. Our evidence in section 6.1
documented the large increase in the flow of immigrants during the 1970s and
showed that the direction of this flow was mainly toward existing immigrant
enclaves. We assume that these facts represent an exogenous increase in the
supply of immigrants to these areas and that the effects of this supply shift on
wages cannot be arbitraged in the short run by mobility of other factors. This
suggests a comparison of within-area wage changes between the 1970 and the
1980 Censuses in response to changes in the stock of immigrants. More for-
mally, we effectively difference ( 5) within areas by including city by cohort
effects in the model:’*
Relative Wage Adjustments within Areas
where t indexes Census year (1970, 1980). In equation (6), differences in
immigrant earnings across areas are subsumed in the P,’s, which vary by area
and entry cohort (but not by year). In this model, parameters yi, are identified
from within-area changes in relative immigrant shares over time. For ex-
ample, yI , < 0 implies that areas experiencing an increase in the share of new
immigrants over the decade will also show declining wages of new immi-
grants relative to other workers in those locales.
6.3 Empirical Results: The Effects of Immigration on Wages
and Earnings
6.3.1 Results from the 1980 Census
In this section, we report parameter estimates for versions of models (5)
and (6). The basic sample consists of 26,681 immigrants derived from the 1%
Sample of the 1980 Census. Immigrant arrival cohorts in these data are de-
179 Labor Market Adjustments to Increased Immigration
fined by date of arrival in the United States as recorded by the Census. The
six identifiable cohorts are immigrants with zero to five, six to ten, eleven to
fifteen, sixteen to twenty, twenty-one to thirty, and more than thirty years in
the United States. All our results are for men between the ages of 16 and 64
who were labor force participants (employed or unemployed) at the time of
the Census survey (roughly, April 1980) and who had positive earnings during
the previous calendar year. These men resided in 119 SMSAs (listed in the
Appendix). Details of selection criteria, variable definitions, and summary
statistics are also appended. The dependent variable in all the models esti-
mated below is the natural logarithm of the average weekly wage (annual
earnings divided by weeks worked) for each individual. l 3
Judgments about which labor aggregates to include in the model determine
the base group against which relative earnings adjustments of immigrants are
measured. We have tried several aggregation schemes, with very similar re-
sults, of which two are noteworthy. First, when the base group is defined to
be natives as an aggregate, the estimable substitution matrix r = [y,,] has
forty-two indcpendent elements. Using this model, the substitution effects
that we estimate are quite small. Further, we were unable to reject (either
jointly or individually) the hypothesis that y6, = 0, i = 1, , . . , 6, in this
case, which indicates that the wages of immigrants with more than thirty years
in the United States are fully assimilated insofar as substitution effects are
concerned. This suggests a second aggregation scheme that restricts attention
to immigrants only and where the normalizing group is immigrants with more
than thirty years in the United States. In this case, r contains thirty indepen-
dent elements. This sample produced slightly larger estimates of the effect of
immigration on relative wages. Since our main finding is that these effects are
small in all relevant cases, we report only the results using the second ap-
proach. l4
Column 1 of table 6.5 reports the estimated diagonal elements of r = [y,,]
from a completely unrestricted model. Taken literally, these “own” effects im-
ply that a 10 percent increase in the flow of new immigrants to an area would
reduce new immigrant weekly wages by about 1 percent ( - .098 x . l ) rela-
tive to immigrants in the United States for more than thirty years.I5 This esti-
mate is not very precise, and, in the unrestricted models of columns 1-3, it is
the only effect that is larger than its standard error. Off-diagonal terms in r
(not reported) are also imprecisely estimated. In part, this imprecision reflects
a vain effort to estimate the thirty free parameters of r from immigrant shares
in only 119 SMSAs. The problem is colinearity. One way to impose further
structure and summarize the overall effect of immigration on wages is to esti-
mate the effect of a proportional increase in the size of all immigrant groups.
Since d In w = Td In M, the estimated effect on each cohort is simply the row
sum from the substitution matrix, r. Estimates of these effects for the unre-
stricted models of columns 1-3 of table 6.5 are shown in table 6.6. In general,
these estimates imply that immigration reduces wages, especially among
180 Robert J. LaLonde and Robert H. Topel
Table 6.5 The Effects of Immigration on the Wages of Immigrants (dependent variable
is log average weekly earnings of males in 1979)
Unrestricted Version: All Restricted Version: Cross Effects
Parameters Free Constrained to Zero
(3) (4) (7)
Own effects: y, :
Years in the U.S.:
0-5 - ,098
C.043)
- ,052
- ,001
.018
(.045)
.050
(.065)
(.053)
- ,099
(.M)
- ,064
(.054)
- ,001
(.049)
.017
(.ow
,061
(.066)
- .099
- ,072
(.055)
.006
(.049)
,031
(.046)
,064
(.OW
- .045
(.014)
- .045
(.014)
- .031
(.014)
- .002
(.015)
.020
(.020)
- .047
(.015)
- ,051
(.014)
- ,036
(.014)
- .002
(.015)
,015
(.021)
- ,045
(.015)
- ,047
(.014)
- ,033
(.015)
,003
(.015)
,021
(.021)
- ,045
(.015)
- ,050
(.014)
- ,037
(.015)
- ,005
(.015)
.014
(.021)
6-10
11-15
16-20
21-30
Arrival cohort effects:
Years in the U.S.:
0-5 - ,519
(.106)
- .292
(.107)
- ,294
(.109)
- ,180
(.118)
,013
(. 105)
- ,567
- .375
(. log)
- ,346
(.111)
- .216
(.119)
(.107)
(.107)
- ,032
- ,579
(.108)
- ,375
(.109)
- ,333
(.112)
-.189
(.120)
- ,012
(. 108)
- .376
(.050)
- .267
(.049)
-.164
(.052)
- ,006
(.058)
.076
(.071)
- ,412
(.05 1)
- ,316
(.050)
- .201
(.053)
- ,019
(.059)
,042
(.072)
- .414
(.051)
(.050)
- ,186
(.053)
- .003
(.059)
,065
(.073)
- ,312
- ,443
(.051)
- ,347
(.050)
- .231
(.053)
- .041
(.059)
.039
(.072)
6-10
11-15
16-20
21-30
Regression includes:
Cross effects: years
in U. S. interacted
with other city co-
hort shares
Occupation controls
Industry controls
Place of origin con-
trols
no no no no
no
yes
yes
no
no
Yes
no
no
Yes
no
Yes
no
Nore: Regressions control for years of schooling, potential experience and experience squared, race, the
presence of children, two marital status dummy variables, and a dummy variable for a disability that
limits a person’s work. There are nine occupation controls, eighteen industry controls, and six place of
origin controls: Europe, Canada, Australia, and New Zealand; Asia; the Middle East; Mexico; other
Latin America; and other immigrants. Standard errors are in parentheses. The cross-substitution esti-
mates associated with columns 1-3 are not reported in the table.
more recent arrivals: a sustained increase in immigration of the indicated mag-
nitude would reduce the wage of new arrivals by about 9 percent, with smaller
but still substantial effects on the wages of earlier cohorts.I6 Only the earliest
arrivals (twenty-one to thirty years) are insulated from relative wage adjust-
ments.
181 Labor Market Adjustments to Increased Immigration
Table 6.6 Estimated Effects on Log Weekly Wages of a Proportional Increase
(d In Mi = 1) in All Immigrant Cohorts, Unrestricted Substitution
Effects, 1980
Years Since Immigration
Model 0 -5 6-10 11-15 16-20 2 1-30
1 - ,091 - ,046 - ,054 - ,040 ,006
(.028) (.029) (.030) (.032) (.028)
2 - ,096 - .059 - ,065 - ,046 .o
(.029) (.029) (.030) (.032) (.029)
3 - ,098 - ,057 - ,060 - ,039 ,004
(.029) (.030) (.031) (.033) (.029)
Nore: Calculated from estimated substitution matrix for the unrestricted models in columns 1-3
of table 6.5. For each immigrant group, the estimated effect of a proportional increase in all
immigrants in a local labor market is the sum of coefficients in the corresponding row of the
substitution matrix. For other controls in each model, see table 6.5. Standard errors are in paren-
theses.
Columns 4-7 of table 6.5 report more parsimonious specifications that con-
strain the off-diagonal terms of r to zero. In these specifications, arrival co-
horts are assumed to be independent inputs in local production, so there are
no intercohort crowding effects on wages. In these models, a larger own labor
force share for an immigrant cohort tends to reduce wages for that cohort but
has no effect on other cohorts. These own-substitution effects also tend to die
out as time in the United States accumulates. Thus, there appears to be signif-
icant crowding among recent arrivals, but the effects of own cohort size dissi-
pate over time.
The parsimony of the specification in columns 4-7 of table 6.5 was pur-
chased with a substantial loss of generality: cross-cohort substitution was as-
sumed away. We next reintroduce these substitution effects with additional
structure. We hypothesize that, for each immigrant cohort, cross effects are
smaller than own effects (yj j < yji for i f 1) and that these substitution effects
dissipate as [ i - j ] increases. That is, members of adjacent arrival cohorts are
better substitutes than are members of distant ones. Under this hypothesis, in
each row of r the largest negative element is along the diagonal, while other
effects should be smaller moving away from the diagonal in either direction.
To test this hypothesis, we allow
(7) yji = 7, + AAi - j l ,
j = 1 , 2, . . . , 5 .
If adjacent cohorts are imperfect substitutes, then we expect yjj < 0 and
Aj > 0, with [Aj I < ly,l. The linear restrictions (7) reduce the number of esti-
mated substitution parameters from thirty to ten while retaining substantial
flexibility. In fact, the restrictions imposed in equation (7) cannot be re-
jected-either individually or jointly-in any form of the model that we have
estimated. Estimates based on (7) are shown in table 6.7 for various combi-
nations of other controls.
182 Robert J. LaLonde and Robert H. Topel
Table 6.7 The Effects of Immigration on Wages, Linear Restrictions on
Intercohort Substitution (dependent variable: log average weekly
earnings in 1979)
(1) (2) (3) (4) (5)
Own Cross Effects:
0-5 years:
Years in the U. S. :
Own effect (y,,)
Cross effect ( A, )
6 1 0 years:
Own effect (y,,)
Cross effect (A,)
11-15 years:
Own effect (yaJ
Cross effect (A?)
1620 years:
Own effect (y4)
Cross effect (A,)
21-30 years:
Own effect (y,,)
Cross effect (A,)
Regression includes:
Occupation
controls
Industry controls
Place of origin
controls
Arrival cohort
controls
- .032
( . al l )
,011
(.005)
- ,037
(.012)
,018
(.@37)
- ,010
(.015)
,003
(.011)
,015
(.019)
- ,011
(.012)
,027
(.014)
- .013
(.007)
Yes
Yes
Yes
Yes
- .035
(.012)
.012
(.005)
- ,038
(.012)
,017
(.@37)
- ,013
(.015)
,004
(.011)
,015
(.019)
- ,012
(.012)
,028
(.015)
- ,011
(.007)
no
Yes
Yes
Yes
- ,032
(.012)
.011
(.@35)
- .036
(.012)
,016
(.007)
- ,009
(.016)
,002
(.01)
.01
- .01
(.02)
(.01)
.03
(.01)
- ,013
(.007)
no
no
Yes
Yes
- ,030
(.012)
,010
(.@35)
- .036
(.012)
,016
(307)
- .012
(.016)
,003
(.011)
.013
(.019)
- .010
(.013)
.027
(.015)
- ,013
(.@37)
no
no
no
Yes
- ,033
(.012)
.011
(.005)
- ,037
(.012)
,017
(.007)
- ,015
(.015)
,005
(.011)
.011
(.019)
- .010
(.012)
,026
(.015)
- ,013
(.@37)
no
Yes
no
Yes
Nore: See note to table 6.5. The own effects (y,) for each cohort j are unrestricted. The cross-
substitution effects are restricted to follow y,, = y, , + A, ( i - j l , where i indexes the time of
arrival of cohort i relative to cohort j .
The key finding in table 6.7 is that own effects of cohort size ( yj j ) are nega-
tive and significant for recent arrivals, while cross-cohort substitution effects
die out with the difference in time of entry to the United States (A > 0). Es-
pecially for recent arrivals, we find that adjacent cohorts are q-substitutes.”
Both the own and the cross effects of cohort size tend to diminish with years
since entry. We take this finding as evidence of assimilation: the effect on
183 Labor Market Adjustments to Increased Immigration
wages of a large cohort is diluted as immigrants melt into the broader market
of native workers. To pursue this point, we add additional structure to (7) by
assuming that
Together, restrictions (7) and (8) express the form of the substitution matrix in
terms of just four parameters. If assimilation means increasing substitution
between past immigrants and the labor market as a whole, as time in the
United States accumulates, we expect y < 0, p > 0, A > 0, and + < 0.
Furthermore, the parameters should also satisfy 141< IyI and (A1 < IyI if there
is (imperfect) substitution among immigrant cohorts.
Estimates of this parameterization of the substitution matrix are shown in
table 6.8 for three illustrative sets of other controls. Other specifications differ
trivially from these. All parameters are of the anticipated signs and relative
magnitudes, and they are significantly different from zero by conventional
standards. The reported F-statistics in the table test the four-parameter struc-
ture given by (8) against the unrestricted, thirty-parameter model of r. Re-
markably in a sample of this size, the restrictions are never rejected.'* Thus,
(8) offers a good summary of the data. The estimates imply that an increase of
roughly 170 percent (d In M = 1) in the stock of new immigrants would
reduce the relative weekly wages of new immigrants by about 3 percent
(y < 0). The immediate effect on earlier immigrant cohorts of this increase
would be smaller (A > 0). As time in the United States accumulates for an
arrival cohort, the earnings disadvantage caused by being a member of a large
cohort evaporates (p. > 0), as do the cross effects of cohort size on adjacent
Table 6.8 The Effects of Immigration on Wages: Linear Restrictions on Own
and Intercohort Substitution (dependent variable: log average weekly
earnings in 1979)
Parameter
F-Statistic
Model Y A CL 4 for Restrictions
Nore: Model 1 controls include cohort, origin, occupation, and industry effects in addition to the
demographic controls listed in table 6.5. Model 2 drops occupation from the set of controls, and
model 3 drops industry and occupation. The reported F-statistics test the restricted four-parameter
model relative to the completely unrestricted model with thirty parameters. Dependent variable
is log weekly wages; standard errors are in parentheses. For definitions of parameters, see the
text.
184 Robert J. LaLonde and Robert H. Tope1
arrivals (4 < 0).I9 All these estimates are consistent with immigrant crowding
in local markets, tempered by assimilation and imperfect substitution.
6.3.2 Annual Earnings versus Wages: Do Quantities Matter?
The analysis to this point has focused only on market clearing price adjust-
ments with inelastic labor supply. However, if immigration also causes quan-
tity adjustments in terms of unemployment, hours, or weeks worked, then
annual earnings may be a more appropriate measure of welfare. A detailed
analysis of adjustments on each of these margins is beyond the scope of this
paper (see Altonji and Card, in this volume). Yet quantity and price adjust-
ments are likely to be correlated, so the effects of immigration on annual earn-
ings may be larger than on wages. The estimates in tables 6.5-6.8 would then
underestimate the distributive effects of immigration. To explore this possibil-
ity, table 6.9 reproduces the estimates in tables 6.6 and 6.8 when log annual
earnings in 1979 instead of log average weekly earnings is used as the depen-
dent variable.
The estimates from the unrestricted model of the effect of a proportional
increase in all immigrant groups, in part A of the table, are slightly larger than
the corresponding estimates in table 6.6 (the most recent arrival cohort is an
exception). For the specification in row 2 of table 6.9, part A, the effects on
earnings exceed those on wages, on average, by about a third, though the
standard errors are large enough that equality of effects cannot be rejected.
Thus, it appears that the main distributive effects of immigration operate
through price flexibility rather than through adjustments in unemployment or
participation. This conclusion is reinforced by a comparison of the estimates
in part B with those of table 6.8, which report restricted estimates of substi-
tution parameters. The estimates for wages and annual earnings differ only
trivially. On this evidence, we conclude that the main actor in market adjust-
ments to immigration must be wage flexibility. Adjustments in unemployment
or participation are negligible.
6.3.3
To this point, we have treated all nonimmigrants as a single aggregate,
while focusing on substitution possibilities among immigrants. For these
groups, the effect of immigration on measures of welfare are quite small. Even
so, some groups of native Americans may be more sensitive to the crowding
effects of immigration than others, and for them the implied redistributive
effects are of some concern. Here, we focus on two identifiable groups who
may face the most important crowding effects of immigration: young (aged
16-34) blacks and Hispanics.
We treat young blacks and Hispanics as separate inputs that interact with
immigrants in local production (see eq. [ 13 above). The unrestricted matrix of
estimated substitution effects now contains fifty-six parameters, and it is not
very informative. As above, we may calculate the effect of a scale (d In Mi =
The Effects of Immigration on Young Native Blacks and Hispanics
185 Labor Market Adjustments to Increased Immigration
Table 6.9
A. Effects on Log Annual Earnings of a Proportional Increase (d In Mi = 1) in All Immigrant
Cohorts, Unrestricted Substitution Effects (dependent variable: log annual earnings in 1979)
Years Since Immigration
Model 0 -5 6-10 11-15 16-20 2 1-30
~~ ~
1 - .089 - .064 - .077 - ,066 - ,008
(.032) (.032) (.034) (.036) (.032)
(.033) (.033) (.034) (.037) (.033)
(.033) (.033) (.034) (.037) (.033)
2 - .093 - .079 - ,085 - .071 - .014
3 - ,091 - ,074 - .078 - ,062 - ,008
B. The Effects of Immigration on Earnings: Linear Restrictions on Own and lntercohort
Substitution (dependent variable: log annual earnings)
Parameter
F-Statistic
Model Y A CL 4 for Restrictions
1 - .026 ,008 .009 - ,003 ,721
2 - ,030 ,009 .011 - ,004 ,756
3 - ,028 ,009 .011 - ,004 ,822
(.008) (.ow (.ow
(.009) (.ow (.ow
(.ow (.ow (.ow
Note: See notes to tables 6.6 and 6.8. Standard errors are in parentheses.
1) increase in all immigrant cohorts on the wages or earnings of blacks and
Hispanics. These estimated effects are shown in part A of table 6.10 for two
specifications of the model.20 Overall, there is only weak evidence that immi-
gration reduces the wages and earnings of these natives. The largest estimates
that we obtained are shown in row 1: the point estimate of the effect of a 170
percent increase in the size of all immigrant cohorts on black wages is only
2.4 percent, though the estimate is smaller than its standard error. The corre-
sponding estimate for Hispanics is less than 1 percent. Surprisingly, in light
of our previous results, the effects on earnings are slightly larger than on
wages. Thus, there is some evidence of reinforcing adjustments on time
worked, especially among young blacks. Again, however, these effects are
not precisely estimated.
An alternative strategy for examining these effects is to impose the restric-
tions given by (7) and (8) on the matrix of intercohort substitution terms
among immigrants, while leaving own and cross effects for blacks and His-
panics as free parameters. To impose some structure, we allow black and His-
panic wages to be affected separately by immigrant cohorts that arrived before
and after 1965. The hypothesis is that crowding effects of immigration are
186 Robert J. LaLonde and Robert H. Tope1
Table 6.10 The Effects of Immigration on Wages and Earnings of Young Blacks
and Hispanics
A. The Effects of a Proportional Increase in All Immigrant Cohorts (unrestricted models)
Effect On:
Model Black Wages Black Earnings Hispanic Wages Hispanic Earnings
1 - .024 - ,059 - .009 - ,015
(.030) (.035) (.032) (.037)
2 - .020 - ,046 - ,008 - .012
(.030) (. 036) (.032) (.038)
B. Estimated Cross Effects of Immigrants on Blacks and Hispanics (linear restrictions imposed)
Effect on Blacks of an Increase in: Effect on Hispanics of an Increase in:
Native Post-1965 Pre-1965 Native Post-1965 Pre-1965
Model Blacks Immigrants Immigrants Hispanics Immigrants Immigrants
Earnings:
1 - .042
(.018)
(.018)
1 - ,042
(.015)
2 - ,031
(.015)
2 - .028
Wages:
- ,005
(.012)
- ,008
(.012)
,008
(.010)
,005
(.010)
- ,014 ,015
(.018) (.010)
- ,008 ,018
(.017) (.OlO)
- ,020 .007
(.015) ( . Ol O)
- ,014 ,010
(.015) (.009)
- ,025
(.015)
- ,030
(.015)
- ,013
(.013)
- ,016
(.013)
Note: Part A parameter estimates refer to the effect of a unit change in log employment of all
immigrant cohorts (d In M, = 1 for all i) on log wages or earnings of blacks and Hispanics. Part
B estimates represent the effect of a unit change in log employment of the indicated group. Model
1 contains all demographic controls listed in table 6.8. Model 2 adds industry and occupation
controls. Standard errors are in parentheses. Part B models constrain intercohort substitution
matrix for immigrants to follow eqq. (7) and (8). Black and Hispanic effects are free parameters.
concentrated on these demographic groups and that recent immigration is the
most important factor.
We report (in pt. B of table 6.10) the own effects for both blacks and His-
panics as well as estimated cross effects with immigrants. In each case, we
find crowding effects of blacks and Hispanics on their own wages; increases
in the labor force shares of these groups reduce their wages, though only the
estimate for blacks is significant. We also find that recent immigrants are sub-
stitutes for young blacks, though the effect is small ( - .01 is the largest esti-
mate we obtained) and imprecisely estimated. It is substantially smaller than
the own effect on black wages ( - .042). The estimates for Hispanics are more
mixed. Finally, for neither group do we find important differences between the
wage and the earnings estimates, suggesting that employment and hours ad-
justments are also minor concerns. Overall, these estimates do not suggest to
187 Labor Market Adjustments to Increased Immigration
us that immigration is a prime factor affecting labor market outcomes for these
young natives.
6.3.4 Results from the 1970 Census
According to Census data, immigration to the United States in the 1970s
was roughly double its level in the 1960s. Because this sharp increase in the
flow of new immigrants was highly geographically concentrated (see sec.
6. l ) , it is plausible that short-run labor market adjustments would generate a
stronger relation between immigration and relative wages in the 1980 Census
than in the 1970 Census. We examine this point in table 6.11, which summa-
rizes estimates of the substitution effects from the 1970 Census. Because the
story is not much different in these data, we report only the substitution effects
of a proportional increase in all immigrant cohorts from the unrestricted
model (eq. [5]) in part A of table 6.11 and the restricted form of intercohort
substitution effects (eq. [S]) in part B.
Table 6.11
A. Effects on Relative Log Weekly Wages of a Proportional Increase in All Immigrant Cohorts
(unrestricted substitution effects, 1970)
Years Since Immigration
Model 0 -5 6-10 11-15 16-20 2 1-30
1 ,012 - ,024 - ,007 - ,017 .ow
(.022) (.024) (.027) (.028) (.024)
2 .008 - ,025 - ,003 - ,024 ,012
(.023) (.024) (.028) (.029) (.025)
3 ,009 - ,028 - ,006 - .017 ,015
(.023) (.024) (.028) (.029) (.025)
B. Estimated Substitution and Assimilation Parameters for Log Weekly Wages of Immigrants
(linear restrictions imposed, 1970)
Parameter
F-Statistic
Model Y h P 4J for Restrictions
1 - ,019 ,008 ,001 - .0005 ,861
2 - ,020 .008 ,001 - .o004 ,834
3 - ,021 ,008 ,001 - .oO04 ,814
(.003) (.001)
(. 007) (.003) (.001)
(.007) (.003) (.001)
Nore: Part A calculated from estimated substitution matrix for unrestricted models analogous to
those in columns 1-3 of table 6.5. These results are comparable to those in table 6.6. For other
controls in each model, see table 6.5. Calculations are based on a sample of 17,158 immigrants
in 119 large SMSAs from the 1970 Census. Standard errors in parentheses. For part B, see notes
to table 6.8. Dependent variable is log weekly wages; standard errors are in parentheses. The
results when the dependent variable is log annual earnings are similar.
188 Robert J. LaLonde and Robert H. Tope1
The estimates in part A should be compared to the corresponding estimates
for 1980 in table 6.6. Whereas the 1980 estimates implied sharply lower earn-
ings among new immigrants, the corresponding estimates for 1970 are negli-
gible. For earlier arrivals, the estimates are negative though generally smaller
than in 1980, and none are significant by conventional standards. These points
are also apparent in part B; while all the substitution relations take the antici-
pated sign, only y is significant. The key point is that all these effects are
substantially smaller than in the 1980 data (see table 6.8).
The relation between the estimates generated by the 1970 and 1980 cross
sections raises an important issue. Did the increased immigration of the 1970s
generate the substantial crowding effects that seem to show up in the 1980
cross section? To answer this question, we create a pseudo-panel from the
combined 1970 and 1980 Census files and analyze within-market changes in
immigration, wages, and earnings.
6.3.5 Panel Estimates: Relative Wage Adjustments within SMSAs,
1970-80
The preceding econometric results rely on cross-sectional differences in la-
bor force shares to generate price adjustments. Since labor is mobile in the
long run, the existence of these wage differentials appears inconsistent with
spatial equilibrium, so our interpretation of these results may be suspect. In
light of this problem, we estimate equation (6), which pools the data from the
1970 and 1980 Censuses. We add to the model six hundred fixed effects that
control for entry cohort (time in the United States) within each SMSA. Thus,
the variation used to estimate substitution effects occurs over time and within
SMSA-cohort cells. In effect, we ask whether areas that experienced un-
usually rapid immigration over the decade also experienced falling relative
wages and earnings of recent immigrants and whether there were spillover
effects of these changes on other groups.21
Results are summarized in tables 6.12 and 6.13. In table 6.12, we report
models for the determination of log weekly wages and annual earnings that
constrain intercohort substitution effects to follow (7). Each row of the substi-
tution matrix is summarized by two parameters: an “own” effect of increasing
cohort size on members of the cohort and a cross-cohort substitution effect
that allows each cohort to have the largest effects on adjacent arrival cohorts.
As above, we expect the former effect to be negative and the latter to be posi-
tive.
The results are surprisingly similar to the cross-sectional estimates (see
table 6.7), though standard errors are somewhat larger. In four of five cases,
the point estimate of the own effect of cohort size is negative, with smaller
effects on adjacent cohorts. Differences between the estimates for log weekly
wages and annual earnings are small, which indicates again that the main
effects of immigration are on wages rather than employment (weeks worked).
Furthermore, the estimates show a tendency to “die out” as time in the United
189 Labor Market Adjustments to Increased Immigration
Table 6.12 Wages Changes within Locales: The Effect of Immigration on
Changes in Wages and Earnings within SMSAs, Linear Restrictions
on Intercohort Substitution, 1970-80
Dependent Variable
Log Weekly Wage Log Earnings
Cohort: Years
Since Immigration (1) (2) (3) (4)
0-5:
Own effect
Cross effects
6-10:
Own effect
Cross effects
11-15:
Own effect
Cross effects
16-20:
Own effect
Cross effects
21-30:
Own effect
Cross effects
Origin effects
Cohort x SMSA
effects
Industry effects
Occupation
effects
R2
- ,039
(.014)
,018
- .041
(.014)
,020
- .007
(.013)
,006
(.009)
,032
(.018)
- ,011
(.011)
- .010
(.015)
,003
Yes
Yes
no
no
,258
- ,034
(.014)
,016
(.006)
- ,036
(.014)
,018
- .009
(.013)
,007
(.009)
,027
(.018)
- ,008
(.011)
- .010
(.014)
.001
Yes
Yes
Yes
no
,272
- ,049
(.016)
.021
(.ow
- ,065
(.015)
.036
(.009)
- ,002
(.014)
,002
(.010)
,047
(.021)
- .019
(.013)
- ,006
(.016)
,002
(.007)
Yes
Yes
no
no
.243
- .045
(.015)
,020
(.006)
- ,061
(.015)
,020
- ,003
(.014)
.003
(.010)
,042
(.021)
- ,015
(.012)
- .005
(.016)
.001
(.007)
Yes
Yes
Yes
no
,257
Note: For other regressors, see note to table 6.5. The models include a dummy variable for 1980.
Standard errors are in parentheses.
States accumulates: effects of within-city changes in shares are stronger for
more recent arrivals.
In light of the last point, table 6.13 shows estimates for the most parsimon-
ious specification, which restricts substitution terms to follow (8). These
“panel” estimates should be compared to the cross-sectional results reported
190 Robert J. LaLonde and Robert H. Tope1
Table 6.13 Wage Changes within Locales: The Effect of Immigration on Changes
in Log Wages and Earnings within SMSAs: Linear Restrictions on
Own and Cross-Substitution Effects, 1970-80
Log Weekly Wage Log Earnings
(1) (2) (3) (4)
- ,020
(.009)
,012
(.004)
,003
(.002)
- ,002
(.001)
- ,029
(.014)
,016
(.004)
,005
- ,002
(.001)
- ,027
(.010)
,015
,005
- ,002
(.001)
Origin effects Y e5 Ye5 Yes Yes
Cohort x SMSAeffects Yes Yes Yes Yes
Industry effects no Yes no Yes
Occupation effects no no no no
R2 .257 ,272 ,243 .256
Note: See note to table 6. 8. Standard errors are in parentheses. N = 44,004
in tables 6.8 and part B of table 6.9. In light of our previously stated concerns,
we are surprised that the panel and cross-sectional results are almost identical.
All parameters are of the anticipated signs, with relative magnitudes that ac-
cord with theory. Our point estimates imply that a rough tripling (d In M = 1)
of the rate of new immigration to an area would reduce the relative wages and
earnings of new immigrants by 2-3 percent. Again, this crowding effect of
membership in a large cohort dies out as U. S. experience accumulates, which
indicates assimilation. Effects of new immigration on previous immigrants are
smaller than the direct effects, which is indicative of imperfect substitution.
6.4 Conclusion
This paper has examined the effect of immigration on the labor market. Our
basic finding is that increased immigration reduces the wages and earnings of
immigrants and their close substitutes, though in our view the effects are not
large. For immigrants themselves, a sustained doubling of the rate of new
immigration may reduce relative earnings of new immigrants by about 3 per-
cent, but even this effect tends to die out over time as immigrants assimilate
to the American market. Labor market effects on nonimmigrants appear to be
quantitatively unimportant: the wages and earnings of young blacks and His-
panics are not very sensitive to immigration. In short, our estimates imply that
immigrants are rather easily absorbed into the American labor market. There
191 Labor Market Adjustments to Increased Immigration
is little here to indicate that the redistributive effects of immigration should be
a major policy concern.
These conclusions are tempered by at least two points. First, our analysis
has relied heavily on differences in wages across geographic areas. These dif-
ferentials are difficult to rationalize as an element of a long-run equilibrium of
the labor market. We argued that the upsurge of immigration in the 1970s was
a change in labor supply that generated short-run wage adjustments among
areas, and comparison of time-series and cross-sectional results tended to sup-
port this assumption. Second, our analysis mainly treated immigrants as a
homogeneous group, and so we ignored the effect that specific immigrant
groups may have. For example, in light of our results, it is plausible that
illegal immigration from Mexico affects mainly young Hispanics. These
points deserve attention, but we defer them to later research.
Data Appendix
Selection and Construction of Variables
The data used in this study were drawn from the 1970 and the 1980 U.S.
Census of Population and Housing, Public Use Samples (see U.S. Bureau of
the Census 1973, 1983). The samples include males sixteen to sixty-four
years old, who were not attending school, who were currently in the labor
force at the time of the Census, who had worked for pay during 1979, who
were not institutionalized, and who were living in SMSAs identified on both
the 1970 and the 1980 Public Use Samples. For 1980, we used the 1%-B
Public Use Sample. For 1970, we used the l%5% questionnaire-county
Group Public Use Sample.
SMSA Definitions
During the 1970s, the Office of Management and Budget changed the defi-
nitions of many SMSAs based on population and commuting patterns in the
1970 Census. These changes are published in “Standard Metropolitan Statis-
tical Areas” (Office of Management and Budget 1976). We used this in-
formation to make the SMSA definitions in the 1980 and 1970 samples as
comparable as possible. In principle, there are two ways to make these adjust-
ments: (i) the SMSA definitions in the 1980 sample can be adjusted so that
they conform to the 1970 definitions (see Altonji and Card, in this volume);
(ii) the SMSA definitions in the 1970 sample can be adjusted so that they
conform to those in 1980. Neither Public Use Sample provides enough infor-
mation so that a user can redefine the SMSA definitions to make the two years
exactly comparable. However, for most SMSAs, the changes do not add or
192 Robert J. LaLonde and Robert H. Topel
subtract many persons from the sample. The first procedure (i) is a little more
precise, although it leads to a smaller sample size, while the second procedure
(ii) is less precise but leads to a larger sample size. We tried both procedures
and found that the results were robust to either method. All the results re-
ported in the paper are based on the second procedure, where we redefined the
1970 SMSA definitions to make them comparable to the 1980 definitions.
Adjusting the SMSA definitions is difficult because the Public Use Samples
do not provide enough information on a household’s county group. Therefore,
a user often does not know for sure whether some households are in a partic-
ular SMSA after a county (or a portion of a county) has been either added or
subtracted between two Census years. In many cases, we drew a random
sample of persons from a particular county (or group of counties if this was
the finest level of identification) that corresponded to the share of persons in
the area that was actually added or subtracted from the SMSA definition. This
task is particularly difficult in New England and eastern Virginia. In a few
cases, it was simpler and more precise to use the 1970 SMSA definitions as
opposed to the 1980 definitions as the standard. This poses no problems for
the analysis as the important thing is to have comparable SMSA definitions
for the two years.
Table 6A.1 presents a list of the 119 SMSAs used in the analysis, along
with the shares of all immigrants and recent immigrants in both 1970 and 1980
and the 1980 shares of young (16 to 34 years old) blacks and Hispanics in
each SMSA’s employed labor force. Note that the share of employed young
blacks seems small in large SMSAs. This fact, however, is due to the concen-
tration of blacks in the central cities. For example, in Chicago, blacks are
concentrated in the city, whereas there are fewer blacks in heavily populated
suburban Cook, Lake, and DuPage counties. In southern SMSAs, a much
larger share of the outlying population is black.
Variable Definitions
We used two measures of earnings as dependent variables, weekly wages
and annual earnings. Annual earnings is the sum of wage and salary income
and self-employment income. We excluded persons who reported that their
self-employment earnings where negative. Weekly wages are defined as an-
nual earnings divided by weeks worked in 1969 and 1979.
Two potential problems with these earnings data are that (i) earnings are
reported up to a maximum of $50,000 in 1970 and $75,000 in 1980 and (ii)
in 1970 weeks worked is reported in discrete intervals. For practical purposes,
the “top coding” problem seems to be minor. In 1980, 1.2 percent of the im-
migrants, , 1 percent of the young black males, . 1 percent of the young His-
panic males, and 1.2 percent of all other native workers had either wage or
salary income or self-employment income that was greater than $75,000. In
1970, .6 immigrants had wage or salary income or self-employment income
that was greater than $50,000. To resolve the problem in the weeks worked
data for 1970, we inputted weeks worked for each person based on the mean
193 Labor Market Adjustments to Increased Immigration
Table 6A.1 Share of Immigrants and Young Native Blacks and Hispanics in
Large SMSAs
Proportion of Employed Male Labor Force
SMSA
~~
Immigrants in Immigrants in
1970 1980
Natives 16-34 Years
All Arrivals All Arrivals Blacks Hispanics
Recent Recent
AKRON,OH
ALBANY-SCHEN-TROY ,NY
ALBUQUERQUE,NM
ALLENTOWN-BETH-
EASTON ,PA-NJ
ANAHEIM-SANTA
ANA-GRDN GVE,CA
APPLETON-OSHKOSH ,WI
ATLANTA,GA
AUGUSTA,GA-SC
AUSTIN,TX
BAKERSRELD,CA
BALTIMORE ,MD
BATON ROUGE,LA
BEAUMONT-FT ARTHUR-
ORANGE,TX
BINGHAMPTON ,NY-PA
BIRMINGHAM,AL
BOSTON,MA
BRIDGEPORT,CT
BUFFAL0,NY
CANTON,OH
CHARLESTON,SC
CHARLOTTE,NC
CHATTANOOGA,TN
CHICAGOJL
CINCINNATI ,OH-KY-IN
CLEVELAND,OH
COLUMBIA,SC
COLUMBUS,OH
CORPUS CHRIST1,TX
DALLAS-FORT WORTH ,TX
DAVENPT-ROCK I S
MOLINEJA-IL
DAYTON'OH
DENVER,CO
DES MOINES,IA
DETROIT,MI
DULUTH-SUPERIOR,MI-WI
EL PAS0,TX
ERIE,PA
FLINT,MI
,045
.046
,027
,034
,082
,023
,009
,006
,021
,066
,034
.008
,014
,041
,003
,089
,100
.059
,023
,014
,015
.004
,096
,018
,067
.003
,012
,031
,020
,019
.014
,036
,021
,074
,033
,196
.025
.045
,013
,010
,012
,005
,033
,007
,004
,002
.005
,018
.010
,003
,001
,013
,002
,029
.033
,011
,006
.002
.007
.001
.034
,005
.017
.Ooo
.003
,006
,006
.004
,006
,013
.004
,016
.002
.049
,005
,017
.030
.040
,043
,036
,161
,022
.025
,017
,036
,097
,033
,018
,023
,034
,010
,095
,086
,064
.018
,015
,022
,010
,132
,024
,059
.014
.021
.063
.054
,027
,018
,046
,027
,065
,023
,233
.017
.014
,003
,012
,021
,010
,096
,009
,011
.006
,014
.047
.012
,006
,013
,012
,003
.037
.026
.017
,004
,004
,010
,007
.067
,007
.018
.006
.009
.026
,035
,012
,004
,020
,020
,017
,005
,127
.005
,004
.03 1
.012
,005
,008
,007
.Ooo
,102
,126
,041
,016
,089
,123
.077
,001
,099
,016
,020
,023
.020
,122
.092
,056
,060
.049
,055
.143
,047
.013
.058
.019
.035
.02 1
,021
,068
.Ooo
,006
.019
,061
,002
,003
.183
,012
.036
.Ooo
,002
,006
,078
,077
,002
,003
,012
,001
,002
,005
.036
,005
,003
,007
,001
,001
.019
.002
.007
,006
.002
.228
.029
,014
.001
.040
.004
,005
.Ooo
.209
,002
,010
(continued)
194 Robert J. LaLonde and Robert H. Topel
~ ~____ ~
Table 6A.1 (continued)
Proportion of Employed Male Labor Force
SMSA
Immigrants in Immigrants in
1970 1980
Natives 16-34 Years
All Arrivals All Arrivals Blacks Hispanics
Recent Recent
FTLAUDERDALE-
HOLLYWOOD,FL
FRESN0,CA
GARY-HAMMONSEAST
CHICAGOJN
GRAND RAPIDS,MI
GREENSBOReWSTN-SLM-
HIGH PT, NC
GREENVILLE,SC
HARRISBURG,PA
HARTFORD,CT
HOUSTON,TX
HUNTINGTON-
ASHLAND, WV- KY4H
JNDIANAPOLIS,IN
JACKSON,MS
JACKSONVILLE,FL
JERSEY CITY,NJ
JOHNSTOWN,PA
KANSAS CJTY,M*KA
KNOXVILLE,TN
LANCASTER,PA
LANSING,MI
LAS VEGAS,NV
LITTLE ROCK-N LITTLE
ROCK, AR
LORAIN-ELYRIA,OH
LOS ANGELES-LONG
BEACH,CA
LOUISVILLE,KY-IN
MADISON ,WI
MEMPHIS,TN-AR
MIAM1,FL
MILWAUKEE,WI
MNPLS-ST PAUL,MN
MOBILE,AL
NSHVL-DAVIDSON,TN
NEW HAVEN,CT
NEW ORLEANS,LA
NEW YORK,NY
NEWARK,NI
NWPTNWS-HAMPTON,VA
NFOLK-F"TSMTH,VA
OKLAHOMA CITY,OK
ORLAND0,FL
,053
,080
.062
,032
,005
.016
,019
,114
,032
,007
.013
,006
,020
,193
.009
,016
.005
,011
,023
,040
,009
,027
.I35
,008
,038
,013
,264
.039
,027
.011
.006
,092
,032
,158
,098
,010
,009
.009
.032
.026
,019
,012
.004
.001
,005
.001
,040
,014
.002
,003
,002
.006
,128
.Ooo
.006
.Ooo
.003
.Ooo
,020
,002
,005
,067
.001
,014
,004
.249
,006
,008
.Ooo
.003
,029
,016
,065
,036
,002
,001
,003
.016
.084
,137
,055
.03 1
.016
,019
.019
,095
.I07
.012
,013
,015
.019
,294
,009
.022
,018
.023
,025
.085
.010
,033
,266
.008
,033
,010
.420
,039
,027
,008
.013
,072
,044
,205
,137
.024
,023
,019
.044
.034
,069
,016
.007
.007
.007
.011
.028
,074
.003
,004
,006
,005
,153
,004
.009
,007
,010
,009
,036
,004
,005
,202
,004
,011
,003
.230
,009
,011
,003
.006
.020
,020
.I01
,064
.007
.009
.011
.021
,044
,015
,071
.026
,017
.055
,026
,020
.075
,009
,051
. I 6 4
,074
,037
.004
,044
.017
,003
,019
,046
,102
,026
.044
,041
,006
,168
,062
,034
,009
,120
,060
,043
. I13
,038
,059
,110
,134
,043
,042
.013
.099
.026
,007
.002
,001
,005
.014
.043
.004
,002
,001
.007
,059
.002
.008
.002
.012
,012
,024
,003
.020
,055
.Ooo
,006
,001
.025
,010
.004
.006
,001
,008
,013
,032
.015
,003
,005
,009
,011
195 Labor Market Adjustments to Increased Immigration
Table 6A.1 ( cont i nued)
Proportion of Employed Male Labor Force
Immigrants in Immigrants in
1970 1980
Natives 16-34 Years
SMSA All Arrivals All Amvals Blacks Hispanics
Recent Recent
OXNAR>VENTURA,CA
PATERSON-CLIFTON-
PASSAIC,NI
PEORIAJL
PHILADELPHIA,PA-NI
PHOENIX,AZ
PITTSBURGH ,PA
PORTLAND,OR-WA
PROVIDENCE, RI
READING,PA
RICHMOND,VA
ROCHESTER,NY
ROCKFORD,IL
SACRAMENT0,CA
ST LOUIS,MC+IL
SALINASMONTEREY ,CA
SALT LAKE CITY,UT
SAN ANTONI0,TX
SAN DIEG0,CA
SAN FRANCISCO-
OAKLAND,CA
SAN JOSE,CA
SANTA BARBARA,CA
SEATTLE-EVERETT,WA
SHREVEPORT,LA
SPOKANE,WA
SPRI NGRELHHCPEE-
HLYKE,MA<T
STOCKTON,CA
SYRACUSE,NY
TACOMA ,WA
TAMPA-ST PETE,FL
TOLED0,OH-MI
TRENTON,NJ
TUCSON,AZ
TULSA,OK
UTICA-ROME,NY
WASHlNGTON,DC-MD-VA
WEST PALM BEACH,FL
WICHITA,KA
WLMNGTN ,DEL-NJ-MD
WORCESTER,MA
YORK,PA
YNGSTWN-WRN ,OH
,105
,127
,009
,049
.041
,036
.038
,072
.026
.015
,072
,040
.064
,016
,121
.037
,061
.079
.126
,098
,104
,066
,007
,045
,069
,121
,040
.049
,038
,026
.076
.057
.008
.045
.054
.066
.012
,037
.079
,013
.027
,053
,046
,001
,011
,010
,006
,008
,025
,008
,005
,027
.006
,018
,004
,038
,008
,014
,033
,054
,044
.042
,020
.003
.012
,028
,041
,012
,011
,012
,005
.02 I
,016
,002
.011
,024
.029
.001
,015
,021
.Ooo
,003
,161
,177
,021
,048
,059
,025
,048
,085
,027
.023
,056
,031
,068
,019
,204
,032
,085
,138
,164
,142
,107
,065
.013
.042
.058
,123
.040
,036
,056
,022
,078
,058
,011
.02 1
,086
,090
,026
,033
,049
,019
,036
,078
,084
,012
,018
,025
,006
,019
.041
,012
,007
,014
.009
,021
,006
,101
,005
.036
,065
.076
.076
.038
.021
.001
,006
,021
,064
,007
,015
,016
,007
,011
,016
,004
,002
,045
,034
,010
,007
,012
,010
,006
,011
.033
,020
,054
.010
,022
,011
,007
.006
,128
,028
,018
,020
,055
,010
,007
,023
.02 1
,040
,017
.016
,015
,140
,007
.010
,011
,020
,019
,038
.022
.040
.011
,030
.013
,108
. a 9
,030
,052
,003
,011
,028
.053
.044
.Ooo
,008
.055
,001
,006
,002
.009
,007
,005
,006
.037
,003
,058
.027
,200
.040
.032
,063
.072
,007
,008
.006
.014
,066
,001
.010
,017
,009
,016
.088
,005
.Ooo
,006
.014
,008
,004
,007
,001
.003
Table 6A.2 Coefficients from Wage Equation Reported in Table 6.5, Column 1
(dependent variable is log average weekly earnings)
(Standard
Coefficient Error)
Education
Experience
Experience squared
Married
Divorced
Children
Disability
Black
Hispanic
Place of origin:
Europe, USSR, Canada, New Zealand, Australia
India, South and East Asia
Pakistan, Mideast, North Africa
Mexico
Other Latin America and Caribbean
All other areas
Occupation:
Professionals and technical workers
Managers and administrators
Sales workers
Clerical workers
Services (nonhouse)
Craft and repair
Nontransportation operatives
Transport operatives
Laborers, handlers
All others, including farm workers
Agriculture
Mining
Construction
Food, tobacco, textile, apparel, leather
Chemicals, petroleum products, rubber, and plastics
Paper, lumber, stone, glass, or clay products
Primary and fabricated metals
Electrical and nonelectrical machines
Transportation equipment
Other manufacturing
Transportation
Printing and publishing, communications, utility
Wholesale trade
Retail trade
Finance, insurance, real estate
Business repair
Personal and entertainment
Professionaligovemment administration
Industry:
Standard errors of regression
,035
,027
- .oO04
,167
,112
,058
- ,141
- ,155
- ,089
. . .
-.130
- ,045
- .095
- ,092
-.I17
. . .
,059
- .069
- .337
- ,386
- ,194
- ,323
- ,221
- .333
- ,360
. . .
,350
,152
,056
,145
.023
,185
, 151
.247
,100
.184
.I96
,121
- ,077
,075
,017
- ,059
,087
,684
(.001)
(.001)
(.oooo3)
(.014)
(.020)
(.010)
(.025)
(.024)
(.020)
. . .
(.015)
(.023)
(.023)
(.021)
(.021)
. . .
(.018)
(.023)
(.023)
(.019)
(.017)
(.019)
(.026)
(.023)
(.045)
. . .
(.077)
(.046)
(.048)
(.050)
(.050)
(.048)
(.047)
(.049)
(.051)
(.048)
(349)
(.048)
( . 045)
(.048)
(.047)
(.047)
(.046)
Note: N = 26,844.
197 Labor Market Adjustments to Increased Immigration
number of weeks worked for persons in the same interval in 1980. This pro-
cedure potentially affects only estimates where weekly wage, not annual earn-
ings, was used as the dependent variable.
The Public Use Samples allow a user to identify whether a person is an
immigrant and when he arrived in the United States. In 1980, immigrants are
classified into six cohorts based on when they arrived in the United States:
1975-80, 1970-74, 1965-69, 1960-64, 1950-59, and before 1950. In
1970, immigrants are classified into ten cohorts based on when they arrived
in the United States: 1965-70, 1960-64, 1955-59, 1950-54, 1945-49,
1935-44, 1924-34, 1915-24, before 1915, and a category for those who do
not report when they arrived in the United States.
The Public Use Samples record in both Census years the highest year of
schooling; age (which we used to construct potential experience as age minus
schooling minus 6); marital status (which we use to construct dummy vari-
ables for those who are married and those who are separated, widowed, or
divorced); whether there are children in the household; whether a person has
a disability that limits his work (which is defined differently in 1970 than in
1980); race (which we use to construct our samples of young blacks and His-
panics); and, finally, place of origin, occupation, and industry. All these vari-
ables are used in the regressions reported in tables 6.6-6.10, unless the table
indicates otherwise.
In table 6A.2, we present estimates of the coefficients that are not reported
in table 6.5, column 1 . The estimates of these coefficients in other tables are
similar. These estimated coefficients are of some interest in themselves when
comparing our findings to other research on the economic effects of immigra-
tion. Note that the returns to education and experience tend to be lower for
immigrants than the returns that are estimated for natives, that the effects of
marital status, children, and disability are consistent with other studies of the
wages of natives, and that an individual’s place of origin has an effect on
earnings even after controlling for all other observable variables. Immigrants
from Europe and the Mideast have the highest earnings, while immigrants
from Asia have the lowest earnings. An immigrant’s occupation and industry
also had a significant effect on earnings.
Notes
1 . Several recent papers examine the changing skills of different immigrant cohorts
(see Chiswick 1986; and Borjas 1985, 1987b). But the issues addressed in those papers
are not new. Between 1900 and 1910, new immigrants accounted for 10 percent of the
labor force, which sparked a similar policy debate at that time (see Douglas 1919).
This debate culminated in the Immigration and Naturalization Act of 1923, which
attempted to control the flow of immigrants through country-specific quotas. In 1965,
amendments to the immigration laws changed the principal criteria used to control
entry into the United States from the “national origin” quotas to a system based on
198 Robert J. LaLonde and Robert H. Topel
kinship with an American citizen or resident. The most recent legislation is the Immi-
gration Reform Act of 1986, which attempts to regulate the flow of illegal aliens into
the United States.
2. For a discussion of the potential distributional implications of immigration, see
Johnson (1980) and Greenwood and McDowell(l986).
3. The estimate of 2.5 million assumes a labor force participation rate of .45 among
all foreign-born persons in the 1980 Census who reported that they arrived in the
United States between 1970 and 1979.
4. In this sense, our analysis is similar to earlier studies of the effect of the baby
boom on wages, e.g., Welch (1979) and Bloom and Freeman (1987). These studies
estimate the “own effect” of increased cohort size on earnings and employment of
members of large cohorts.
5. Borjas (1987a) estimates a model of wage determination based on differences in
demographic group labor force shares across geographic areas. He finds, as do we, that
immigrant earnings are lowest in large immigrant enclaves.
6. Greater ease of substitution need not imply that immigrant wage levels converge
to those of natives. The long-run stock of skills for the representative immigrant may
remain below that of natives, so that immigrants earn less than natives, yet immigrant
and native skills may substitute perfectly. In fact, in 1980 immigrants who arrived in
the United States before 1950 earned more than natives.
7. Equilibrium among these markets is largely ignored in what follows, so we im-
plicitly assume that mobility costs form a significant barrier to intermarket arbitrage in
the short run. Topel (1986) contains an alternative approach that allows for migration
among geographic markets.
8. Later in the paper, we examine the effect of a doubling of all immigrant cohorts
on relative wages. Since native labor is being held constant for those calculations, the
relative marginal products of immigrant labor can change.
9. Symmetry of cross-substitution effects is not implied; i.e., y,l # yl,. Symmetry
of signs is a restriction of the theory; i.e., sign (yl,), as is negative definiteness of
the full matrix of substitution effects.
10. Symmetry of effects implies dWcJltMc, = dWJMcJ,, so y k, is proportional to Y , ~ ,
i = 1, . . . , k - 1. Thus, under the assumption that ykk = 0-an increase in the
number of white natives at a locale has a negligible effect on wages-we may test
yki = 0 if symmetry is assumed. Our tests indicate ytt = 0 in all cases.
11. This assumes that the matrix [ yi j ] is negative definite.
12. Cohort refers to years in the United States.
13. Estimates for the determination of hourly wages differ trivially from those re-
14. Estimates when natives are included in the data are available on request.
15. We remind the reader that our estimates measure relative wage adjustments,
where the normalizing group is persons with more than thirty years in the United
States.
16. This roughly corresponds to a tripling of the stock of immigrants. For example,
a d In M = 1 change in all immigrant quantities would roughly correspond to an SMSA
whose shares of the six immigrant arrival cohorts relative to native workers increased
from .033, .022, .022, .022, ,011, ,011, and ,011 to .107, ,067, .067, ,034, ,034,
and ,034, respectively. Multiply these estimates by .1 for the effect of a 10 percent
change in the immigrant stock.
17. Though not reported separately, corresponding estimates for hourly wages show
larger effects than for weekly wages. The implication is that adjustments in hours
worked plays a minor and unsystematic role: all the effects reported here are due to
price adjustments. We reach a similar conclusion with regard to weeks worked below.
18. We also computed F-statistics to test the four-parameter structure against the
ten-parameter structure given by (7). These additional restrictions are also not rejected.
ported here.
199 Labor Market Adjustments to Increased Immigration
19. For comparison with the unrestricted estimates in table 6.6, these parameters
imply that a proportional increase in the size of all immigrant cohorts would reduce the
earnings of the most recent immigrants by 6.5 percent. This effect dies out by 1 per-
centage point for each prior entry cohort; e.g., - .055 for those with six to ten years in
the United States.
20. Effects on immigrants themselves are nearly identical to those reported in table
6.6.
21. Many SMSAs in our sample experienced a significant increase in the share of
new immigrants in their work forces during the 1970s. Miami and Jersey City are
exceptions. Although the flow of new immigration into these cities was substantial
during the 1970s, this influx represented the continuation of a trend begun the decade
before.
References
Bloom, David, and Richard Freeman. 1987. The labour-market consequences of gen-
erational crowding. European Journal of Population 3: 13 1-76.
Borjas, George. 1985. Assimilation, changes in cohort quality, and earnings of immi-
grants. Journal ofLabor Economics 3, no. 4:463-89.
. 1987a. Immigrants, minorities, and labor market competition. Industrial and
Labor Relations Review 40 (April):382-92.
. 1987b. Self-selection and the earnings of immigrants. American Economic
Review 77(September):53 1-53.
Chiswick, Bany R. 1978. The effect of Americanization on the earnings of foreign
born men. Journal of Political Economy 36(0ctober):897-922.
. 1986. Is the new immigration less skilled than the old? Journal of Labor
Economics 4(April): 168-92.
Douglas, Paul. 1919. Is the new immigration more unskilled than the old? Journal of
the American Statistical Association 15, no. 126(June):393-403.
Greenwood, Michael J., and John M. McDowell. 1986. The factor market con-
sequences of U. s. immigration. Journal of Economic Literature 24(December):
Hamermesh, Daniel S. 1986. The demand for labor in the long run. In Handbook of
Labor Economics, vol. 1, ed. 0. Ashenfelter and R. Layard, 429-71. Amsterdam:
Elsevier.
1738-76.
Hicks, John R. 1963. The Theory of Wages (1932). 2d ed. New York: St. Martin’s.
Johnson, George. 1980. The labor market effects of immigrants. Industrial and Labor
Relations Review 33(April):33 1-41.
Office of Management and Budget. 1976. Standard metropolitan statistical areas.
Washington, D.C.: U.S. Government Printing Office.
Topel, Robert H. 1986. Local labor markets. Journal of Political Economy 94, no. 3,
pt. 2(June):Slll-S143.
U.S. Bureau of the Census. 1973. Technical documentation for the 1970 census of
population and housing. Public Use Samples. Washington, D.C.: U.S. Government
Printing Office.
. 1983. Technical documentation for the 1980 census of population and hous-
ing. Public Use Samples. Washington, D.C.: U.S. Government Printing Office.
Welch, Finis. 1979. The effect of cohort size on earnings: The baby boom babies’
financial bust. Journal of Political Economy 87, no. 5 , pt. 2(0ctober):S65497.
This Page Intentionally Left Blank
7 The Effects of Immigration on
the Labor Market Outcomes of
Less-skilled Natives
Joseph G. Altonji and David Card
One of the most controversial aspects of immigration policy is the extent to
which the arrival of immigrants helps or harms less-skilled natives. Although
economists have developed a variety of theoretical models to analyze this
question (see, e.g., Johnson 1980a, 1980b; Chiswick 1982; or Borjas 1987),
relatively little empirical evidence is available. In this paper, we use variation
in the fraction of immigrants across different cities to measure the effects of
immigration on the labor market outcomes of less-skilled natives. We as-
semble information from the 1970 and 1980 Censuses on labor market out-
comes of natives in 120 major cities. Information from consecutive Censuses
allows us to correlate changes in immigrant fractions with changes in native
outcomes within cities-thereby abstracting from differences across cities
that might bias a simpler cross-sectional analysis. We also provide a variety
of information on the industry distributions of natives and immigrants and
analyze the changes in these distributions that have occurred in cities with
higher and lower immigrant shares.
In the first section of the paper, we present a simple theoretical model that
describes the effects of immigration on the domestic labor market. We assume
that the labor market within each city consists of skilled and unskilled workers
and that immigration adds workers to both sectors, with relative additions
David Card is professor of economics at Princeton University and a research associate of the
National Bureau of Economic Research. Joseph G. Altonji is professor of economics at North-
western University and a faculty research fellow of the National Bureau of Economic Research.
The authors are grateful to Brain McCall and Sarah Turner for assistance with this research and
to the National Bureau of Economic Research, the Center for Urban Affairs and Policy Research,
Northwestern University, and the Industrial Relations Section, Princeton University, for research
funding. They thank John Abowd, Francine Blau, George Borjas, Gregory DeFreitas, Richard
Freeman, Peter Kuhn, and participants in seminars at Columbia University, the University of
Minnesota, Princeton University, and the National Bureau of Economic Research for comments
on earlier drafts.
201
202 Joseph G. Altonji and David Card
depending on the nature of immigrant inflows to the city in question. Our
theoretical framework departs from earlier models in two ways. On the one
hand, we disaggregate labor along skill lines rather than along the lines of
national origin. On the other hand, we allow for demand-side effects asso-
ciated with increases in the local population and for supply-side effects asso-
ciated with the possible crowding out of native workers in response to lower
wage rates. The model leads to a simple empirical specification in which wage
and employment outcomes of less-skilled natives (either in cross section or
within cities over time) vary with the share and skill composition of immi-
grants in the local labor market.
In the second section of the paper, we address the question of whether im-
migrants and natives within the same city compete in the same labor market.
Given the size of immigrant flows during the last two decades, our theoretical
analysis implies that large adverse effects on less-skilled natives are unlikely
unless increases in immigration lead to proportionately larger increases in the
supply of labor to less-skilled jobs. We focus on industry-specific labor mar-
kets within cities. We develop a simple index that measures the effect of a
given inflow of immigrants on the labor market of natives. We find that a 1
percentage point increase in the share of immigrants in a city generates ap-
proximately a 1 percent increase in the supply of labor to industries in which
less-skilled natives are employed. The degree of competition between immi-
grants and less-skilled natives varies somewhat by race and sex group, being
highest for black females and lowest for black males. Overall, however, the
results suggest that immigrants are not sufficiently concentrated in the indus-
tries that employ less-skilled natives to have large effects on the less-skilled
native groups.
We go on to investigate whether immigrant inflows have displaced less-
skilled natives from certain industries. Here, we compare the industry distri-
butions of less-skilled natives in cities with relatively high and relatively low
immigrant densities. We find some evidence that less-skilled natives in high-
immigrant cities have moved out of immigrant-intensive industries. We also
find that the nationwide trend of falling employment in these industries has
been slower in high-immigrant cities, suggesting that the availability of im-
migrant labor has enabled certain low-wage industries to survive in high-
immigrant cities.2
In the third section of the paper, we turn to a regression analysis of the
relation between immigrant shares (or the change in immigrant shares) and
employment outcomes of natives (or the change in these outcomes) across
major cities. The results vary somewhat between the cross-sectional and first-
difference analyses. We argue, however, that the first-difference analysis is
less likely to be contaminated by city-specific factors that affect immigrant
densities and native outcomes. The analysis of changes shows no effect of
increased immigration on participation or employment rates of less-skilled
natives. It does reveal a systematically negative effect on native wages, al-
203 Immigration and the Labor Market Outcomes of Less-skilled Natives
though the specific estimates depend on the group and on whether we use an
instrumental variables procedure to account for the fact that immigration in-
flows may depend on local labor market conditions. For the four racehex
groups that we consider, the instrumental variables estimates (which we pre-
fer) imply that an inflow of immigrants equal to 1 percent of the population of
a standard metropolitan statistical area (SMSA) reduces average weekly earn-
ings of less-skilled natives by about 1.2 per ~ent . ~ The least squares estimates,
by comparison, imply a more modest . 3 percent reduction.
7.1 Analytical Framework
Our framework for analyzing the effect of immigration on the labor market
outcomes of less-skilled natives is to view the inflow of immigrants to each
city (or, more precisely, SMSA) as an outward shift in the supply of labor.
Since we are specifically interested in the effects of immigration on less-
skilled natives, we consider a two-sector labor market consisting of skilled
and unskilled labor. Within skill categories, we make no distinction between
native and immigrant labor or between earlier and later cohorts of immigrants.
We assume that the demands for skilled and unskilled labor in each city are
decreasing functions of their respective wage rates and that prices of capital
and other inputs are exogenous to the local labor market.
This framework contrasts with the one adopted by Borjas (1987), for ex-
ample, who treats immigrants and natives as separate factors of production
and assumes that locally produced output is sold at an exogenous price. In this
case, the conventional elasticities of labor demand are undefined since an in-
crease in the wage rate of one type of labor with other factor prices held con-
stant leads to an increase in marginal cost that drives local firms out of busi-
n e ~ s . ~ Given that many of the goods produced within a city are nontraded
services, however, and that many others enjoy some degree of imperfect sub-
stitutability due to transportation costs, we believe that it is more reasonable
to posit the existence of downward-sloping labor demand functions at the lo-
cal level.
The observation that the demand for labor within a local economy arises in
part from the demand for location-specific goods and services implies that a
partial equilibrium model of the labor market is potentially misleading. In the
extreme case, if all output is locally consumed, and if new immigrants arrive
in the same skill proportions as the existing labor force, then an influx of
immigrants leads to a new equilibrium at the original wage rates, with propor-
tionately higher levels of employment, output, and cons~mption.~ More gen-
erally, the arrival of new immigrants shifts the demand for city output and
hence the demand functions for skilled and unskilled labor. The size of this
effect depends on the share of output consumed locally and on the relative
skill composition of the existing and immigrating labor forces.
To illustrate these propositions and establish a framework for our empirical
204 Joseph G. Altonji and David Card
analysis, consider an urban economy with two goods: a locally produced good
(or service), Y, that is consumed locally and exported to other cities and an
imported national good.6 Assume that Y is produced by a competitive industry
with a constant-returns-to-scale technology using skilled labor, unskilled la-
bor, and other inputs (capital apd/or raw materials) whose prices are exoge-
nous and fixed.' Under these conditions, total industry cost (in units of the
imported good) is described by a function of the form
C(W,, w,, Y ) = Yc(ws, WJ,
where w, and w, represent the real wages of unskilled and skilled labor (in
units of the imported good), and c(.) is a unit cost function.* Let q represent
the unit price of local output (denoted in units of the imported good).
The assumptions of constant returns and perfect competition imply that
Demand for Y arises from three sources: local demand from skilled work-
ers, Y,; local demand from unskilled workers, Y,; and export demand from
the rest of the economy, Y,. Let D,(q, w,) and D,(q, wu) represent the per
capita demand functions of skilled and unskilled workers, respectively, and
let D,(q) represent the demand function for locally produced output from the
rest of the economy. Let P, and P, represent the populations of skilled and
unskilled workers in the city, and denote the total population by P = P, +
P,. Product market equilibrium requires
(1)
q = c(w,, W").
y = p, . D, (q, w,) + P" * D, (4, w,) + D,(q).
Let Ls(ws, q) and L,(w,, q) represent the per capita labor supply functions
of skilled and unskilled workers, respectively. Equilibrium in the local labor
market requires
(2a)
and
(2b)
P, - Ls(w,, 4) = y * c, (ws, w,)
P" * L,(w,, 4) = y . cz(w,, W"),
where c, (. ) and c,(.) denote the partial derivatives of the unit cost function
with respect to unskilled and skilled wage rates, respectively.
Suppose that in an initial equilibrium the fraction of unskilled workers in
the local population is a = PJP We wish to analyze the effect of an inflow of
immigrants of size AZ. Let ci represent the share of unskilled workers in the
new group. The effects of an immigrant inflow can be obtained by differen-
tiating equations (l ), (2a), and (2b) and making use of the fact that the pro-
portional change in the price of output, Aq/q, equals the share-weighted sum
of the proportional changes in all factor prices.
For simplicity, assume that the cross-elasticities of the output demand and
labor supply are zero.9 Then the proportional changes in skilled and unskilled
wage rates satisfy the following pair of equations:
205 Immigration and the Labor Market Outcomes of Less-skilled Natives
(3a) A, ( da) AIIP = (quu - €, ) A log W, + q, A log wS,
(3b) A, [(l - or)/(l -a)] AUP = q,,,A log wu + (qSs - E*) A log w, ,
where q,, is the elasticity of labor demand for skill group i with respect to the
wage of group j, E, is the elasticity of labor supply of group i, and A, and A,
are a pair of numbers between zero and one:
A, = (Y - Y, - k, * Y, ) I Y ,
A, = (Y - k, * Y , - Y, ) / Y ,
k, = ~ ( l - a)/[ ~~( l - a ) ] ,
k2 = a(l-a)/[ ~( 1- a) ] .
The labor demand elasticities in equations (3a) and (3b) are determined by
the conventional Marshall-Hicks formulas:
%, = e,cw,, - Y)?
where 8, is the share of the value of output paid as wages to skill group i, u,]
is the partial elasticity of substitution of skill group i with respect to group j,
and y is the elasticity of demand for Y with respect to its relative price q (a
weighted average of the elasticities of demand exhibited by consumers in the
local market and those elsewhere in the economy).
The expressions A, ( a/ a) AI/ P and A, [(l I a)/(l - a) ] AI/ P in equations
(3a) and (3b) give the effective percentage increases in unskilled and skilled
labor resulting from an inflow of immigrants AI. The increases in skilled and
unskilled populations are aAl and (1 - a) Al , respectively. The proportional
increases in the populations of unskilled and skilled workers are therefore
( a/ u) Al / P and [(l - a)/(l - u) ] Al / P, respectively. The factors Xu and A,
adjust the gross increases in labor supply for the net increases in demand gen-
erated by the new immigrants. If local output is consumed entirely within the
city and immigration is balanced in the sense that a = a, then Xu = A, = 0.
Otherwise, the effective increases in labor supply depend on the fraction of
local output sold outside the city and on the imbalance of skill ratios between
the existing and the newly arriving population. In the simple case where
newly arriving immigrants have the same skills as the existing population, A,
= A$ = YJ X the fraction of output exported. If newly arriving immigrants
are less skilled, however, A, > YJY > A,, accentuating the effective increase
in unskilled labor supply.
Using equations (3a) and (3b), changes in wages rates can be related to
changes in the fraction of immigrants in the local population ( f ) by noting
that Af = A(I/P) = (1 - f ) Al / P. In the special case that the demand for
unskilled labor is independent of the wage rate of skilled labor (i.e., qu, = 0),
equation (3a) can be simplified to
(4)
A log W, = ~ ( a/ u) Al / P,
&u - rl",
206 Joseph G. Aitonji and David Card
which specializes to the formula derived by Johnson (1980a) when Xu = 1
and (Y = a. l o Our model extends Johnson's earlier analysis in two directions:
by allowing for skilled and unskilled workers in the existing and immigrating
populations and by accounting in a very simple manner for the effect of added
population on the demand for local output.
If the demand for unskilled workers depends on the wage rate of skilled
labor (i.e., qu, # 0), then the expression for the change in unskilled wage
rates takes the more general form
( 5)
A log w,, = B,,A l i p,
where
Using the labor supply function, the change in the per capita labor supply of
unskilled natives can then be written as
(6)
A log Lu = E, . B,A lip.
To get some idea of the magnitude of the coefficient BY relating wage
changes to immigrant inflows, suppose that (Y = a, so that Xu = A,. In this
case, equation (5) can be rewritten as
A log w, = AbUA l i p,
where the coefficient bU (b, < 0) is a function only of the supply and demand
elasticities for skilled and unskilled labor, and A equals the fraction of local
production exported to other cities. Values of the coefficient b, corresponding
to alternative values of the supply and demand parameters of the model are
displayed in table 7.1. The rows of the table present alternative choices for the
ratio between the partial elasticity of unskilled labor with respect to nonlabor
inputs (uUk) and the partial elasticity of skilled labor with respect to nonlabor
inputs (a,,). The share-weighted average of these two elasticities is con-
strained to equal .6." The columns of the table present alternative choices for
the partial elasticity of substitution between skilled and unskilled labor (us").
For each choice of the technological parameters, two values of bu are reported,
corresponding to alternative choices for the elasticities of labor supply: . 1 and
1 .O. Other parameters in the model are set as follows: the share of skilled labor
(0,) = .4, the share of unskilled labor (0,) = . 3, and the elasticity of demand
for city output (y) = - 2.5.
The first row of the table presents calculated values of bu under the assump-
tion that capital is a substitute for unskilled labor and a complement for skilled
labor.I2 As Hamermesh (1986, 460-62) has noted in his review of the litera-
ture on labor demand, many empirical studies based on the distinction be-
207 Immigration and the Labor Market Outcomes of Less-skilled Natives
Table 7.1 Predicted Effect of an Increase in Immigration on Unskilled
Wage Rates
Partial Elasticity of Substitution of Skilled
for Unskilled Labor (u,J
Ratio of Partial Elasticities
of Substitution with Labor Supply
Capital (u&J Elasticity ( E ) ~ .25 1 .o 3.0
1. -. 25 .1 . . . - .31 - .42
1 .o . . . - .21 - .30
2. 0 .1 - .21 - .39 - .45
I .o - .29 - .30 -.31
3. . 5 . 1 - .42 - .46 - .48
I .o - .32 - .33 - .33
4. 1.0 . 1 - .49 - .49 - .49
1 .o - .34 - .34 - .34
Note: For notation and assumptions, see the text.
a Share-weighted average of substitution elasticities of skilled and unskilled labor with capital is
constrained to equal .6.
Labor supply elasticities of skilled and unskilled workers are constrained to be equal.
tween blue-collar and white-collar workers in manufacturing have confirmed
this hypothesis. In contrast, the last row of the table presents values of b,
under the assumption that skilled and unskilled labor are equally substitutable
with capital.I3 Despite the wide variation in demand and supply parameters
represented in the table, the range of the coefficient b, is relatively modest:
from - .49 to - .27.14 Under the assumption that immigrants add nothing to
the demand for locally produced output (i.e., A = l), these coefficients imply
that a 1 percent increase in the population of a city due to an influx of immi-
grants with the same skill composition as the existing labor force reduces un-
skilled wages by .3-.5 percent. The implied reduction in the per capita labor
supply of natives (and existing immigrants) is proportional to this reduction
in wages, multiplied by the elasticity of labor supply. If the elasticity of labor
supply is in the range of zero to one, the implied reduction in per capita labor
supply of natives is 0-.5 percent.
The magnitude of these predicted effects is dampened by any expansionary
effect that immigrants have on the demand for locally produced goods. For
example, if one-third of output is consumed locally, then the implied wage
effects of a given immigrant inflow are reduced by approximately one-third.I5
Any imbalance in the skill distribution of arriving immigrants, on the other
hand, accentuates their effect on the local labor market. In the most extreme
case, if newly arriving immigrants are all unskilled and the proportion of
skilled workers in the existing labor force is . 5, then the predicted value of b,
ranges from - 2.0 to - 1 .O, implying roughly two to three times larger effects
on unskilled wage rates.
Our empirical strategy in section 7.3 below is to correlate variation in the
208 Joseph G. Altonji and David Card
share of immigrants in the local labor market with variation in the employ-
ment and wage outcomes of less-skilled natives. We interpret the coefficient
relating wages to immigrant shares as an estimate of the expression B, in equa-
tion ( 5) and the coefficient relating employment rates (or participation rates)
to immigrant shares as an estimate of the product of B, and the elasticity of
labor supply of unskilled native workers. As the previous discussion makes
clear, the value of B, depends on the nature of immigrant flows to each city
and on the characteristics of the demand for output produced in each city.
Even ignoring these issues (as we do), it is important to keep in mind the
potential endogeneity of immigrant inflows to different cities. If the supply of
immigrants is wage elastic, then the covariation across cities between the la-
bor market outcomes of natives and the share of immigrants in the labor mar-
ket will be a positively biased estimate of the expression BU. In our analysis,
we address this issue with an instrumental variables scheme that isolates the
component of immigrant inflows associated with the predetermined character-
istics of each city.
Before turning to the empirical work, two limitations of the model deserve
discussion. First, the model assumes that the existing native population is
immobile. However, one might loosely interpret the supply elasticity of na-
tives to reflect both labor supply changes of the current population of the city
and out-migration (or in-migration) of natives to (or from) other cities.16 If
one interprets the intercity mobility of natives as raising the long-run elasticity
of labor supply, then one would conclude that migration by natives in response
to immigrant inflows would lower the effect of immigration on wages. It
would also lower the effect on per capita labor supply of natives, as measured
by a variable such as the employmenVpopulation rati0.I’ However, intercity
migration would imply spillover effects on wages and employment/population
ratios in other cities, which we ignore in our empirical work.
Second, the model assumes that the local labor market clears. Within the
model, unemployment can be viewed as depending on the wage rate relative
to the benefits of being unemployed. This view is most sensible in the long
run. Barriers to wage adjustment (such as binding minimum wage levels or
fixed welfare benefits) might be expected to strengthen the effect of an in-
crease in immigrants on the employment and unemployment outcomes of na-
tives while weakening the effects on wage levels relative to those implied by
equations (6) and (7). The employment effects for natives could be especially
large if employers of immigrants are less likely to comply with minimum
wage laws or to be unionized.’”
7.2 Industry Distributions of Natives and Immigrants
Our empirical analysis is based on the labor market outcomes of less-skilled
natives in 120 major SMSAs in the 1970 and 1980 Censuses. We consider
four groups of “less-skilled” natives: white males with less than twelve years
209 Immigration and the Labor Market Outcomes of Less-skilled Natives
of completed education; white females with less than thirteen years of com-
pleted education; black males with less than thirteen years of completed edu-
cation; and black females with less than thirteen years of completed educa-
tion. Our data base consists of samples of each race/sex group drawn from the
1/100 Public Use Sample of the 1970 Census and the 5/100 “A” sample of the
1980 Census. A description of our sampling procedures and information on
our procedures for matching SMSA definitions between the 1970 and the
1980 Censuses are provided in Appendices A and B.
Table 7.2 provides an overview of our samples of less-skilled natives. The
samples are restricted to individuals between the ages of nineteen and sixty-
four who report themselves as not in school during the Census week.I9 Be-
cause of the age and education requirements, the average age of our less-
skilled native groups is close to 40. The average years of complete schooling
is less than eight for white male high school dropouts and between ten and
eleven for the other groups.
The labor market outcomes that we consider are the labor force participa-
tion rate during the Census week; the employment rate during the Census
week (measured for those in the labor force in the Census week); the
Table 7.2 Descriptive Statistics for Native Samples
Demographic
and Economic
Characteristics:
1. Age
2. Education
3. Labor force participa-
tion rate ( X 100)
White Male White Females Black Males Black Females
Dropouts No College No College No College
1970 1980 1970 1980 1970 1980 1970 1980
44.3 43.5 40.9 40.8 39.1 37.4 38.7 38.3
8.5 8.8 10.6 11.0 9.2 10.2 9.6 10.4
88.8 81.0 47.3 56.5 83.6 78.4 55.1 59.1
4. Employment rate
( X 100) 96.0 91.1 95.6 94.0 94.4 86.9 92.6 87.9
5. Employment pop-
ulation rate Census
week(X 100) 85.2 73.7 45.2 53.3 78.9 68.3 51.1 52.1
population rate last
year ( x 100) 91.6 82.9 54.5 61.1 86.7 78.0 60.8 60.1
7. Logarithm of weeks
worked last year 3.81 3.75 3.57 3.60 3.77 3.69 3.58 3.60
8. Logarithm of weekly
earnings last year
(current $) 4.95 5.52 4.26 4.96 4.61 5.29 4.03 4.90
9. Sample size 84,068 24,925 99,488 81,151 27,779 29,723 34,013 34,540
6. Employment
Nore: Samples consist of individuals age 16-64 in 120 major SMSAs. Individuals enrolled in school in
Census week are excluded. White male dropouts sample includes individuals with less than 12 years of
completed education. Samples for other groups include individuals with less than 13 years of completed
education. For further information, see App. A.
210 Joseph G. Altonji and David Card
employment-population ratio in the Census week; the fraction of people who
reported working at any time in the previous year (for simplicity, we refer to
this as the employment-population ratio last year); and the logarithms of
weeks worked and average weekly earnings during the previous year (mea-
sured for those individuals who report positive weeks of work and positive
earnings in the previous year). Precise definitions of these outcomes are pre-
sented in Appendix A.
The model of the previous section treats the market for less-skilled workers
within each city as homogeneous. Even within a particular city, however, the
market for less-skilled workers may be segmented along industry lines. If
immigrants and natives tend to work in different industries, then the first-
round effects of new immigration will be mainly concentrated among existing
immigrants. If immigrants tend to work in the same industries as a particular
subgroup of natives, however, then the effects of immigration on this
subgroup of less-skilled natives will be magnified.
Some simple evidence on the correspondence between industry distribu-
tions of natives and immigrants is presented in table 7.3. For the ten two-digit
industries with the highest immigrant employment shares and the ten indus-
tries with the lowest immigrant shares, this table shows the fraction of each of
the four less-skilled native groups in the industry in 1980.20 High-immigrant-
share industries include several low-wage manufacturing industries (apparel,
leather, furniture, miscellaneous manufacturing, and textiles) as well as
low-wage service industries (private household services, hotels and motels,
restaurants and bars, and transportation services) and agriculture. Low-
immigrant-share industries include the government sector as well as railroads,
communications, and several regionally based industries (tobacco, pipelines,
coal mining, and oil and gas extraction). A comparison of the second and third
columns of the table shows that industries with high or low immigrant shares
in 1980 exhibited the same characteristic in 1970, although the immigrant
fractions in many industries increased sharply between 1970 and 1980.21 The
immigrant share of total employment in all industries in our sample of 120
cities increased from 6.0 percent in 1970 to 9.6 percent in 1980.22
The data in table 7.3 suggest that immigrants are most directly competitive
with native women-particularly black women. In fact, the proportion of
black females in the ten highest-immigrant-share industries in 1980 was al-
most as high as the fraction of immigrants in those industries. By comparison,
black males are the least concentrated in high-immigrant-share industries and
the most heavily concentrated in low-immigrant-share industries.
One way to evaluate the effect of immigration on a particular native group
is to calculate the overlap in the industry distribution of the group with the
industry distribution of immigrants. Assuming that interindustry mobility
costs are large, the effects of immigration on native wages will be directly
proportional to the average increase in labor supply to industries in which
natives are employed. To formalize this measure, let S,, represent the share of
the native group in the ith industry, let E, represent the initial level of total
211 Immigration and the Labor Market Outcomes of Less-skilled Natives
Table 7.3 Distributions of Natives in High- and Low-Immigrant-Share Industries,
1980
~-
% of Natives in Industry
% % % of All
Immigrant Immigrant Immigrants White White Black Black
Industry 1980 1970 InIndustry All Males Females Males Females
High immigrant share:
1. Apparel
2. Leather
3. Agriculture, crops
4. Furniture
5 . Miscellaneous
manufacturing
6. Private household
services
7. Hotels and motels
8. Transportation ser-
9. Restaurants and
vices
bars
10. Textile mills
Total: 10 industries
Low immigrant share:
1. Pipelines
2. Gov’t.: justice and
public safety
3. Gov’t.: revenue
and taxation
4. Coal mining
5. Railroads
6. Tobacco
7. U.S. Post Office
8. Oil and gas extrac-
9. Communications
10. Gov’t.: economic
tion
programs
Total: 10 industries
38.4
21.3
25.8
21 .o
20.9
20.2
18.2
15.8
15.6
15.6
. . .
1.5
2.8
2.8
3.5
3.8
3.9
4.1
4.2
4.4
4.5
. . .
21.1
14.4
10.0
11.0
10.6
9.5
10.6
11.2
9.3
8. 8
. . .
1.9
2.3
3.4
2.4
3.5
1.8
2.4
2.0
3. I
2.9
5. 1 1.3 . 6 2.0 .5 2.3
. 6 .2 .3 .3 .1 .3
1.5 .6 1.2 .4 .5 .3
1 .O .4 .7 .4 .6 .4
2.3 1.1 1.2 1.3 1.0 1.4
1.4 .7 .2 . 8 .2 6.0
2.2 1.2 .7 1.7 1.2 3.5
.5 . 3 . I .4 . I .1
6.4 3.9 2.5 7.6 3.1 5. 5
.8 .5 .I .I .6 . 8
21.8 10.1 8.2 15.6 7.9 20.6
.o .o .o
.4 1.4 .9
.1 .4 .o
.O .1 . I
.3 .6 1.4
.o .1 . I
.4 1.0 .9
.2 .4 .4
. 8 1.7 .5
. 3 .6 .2
.o .o
. 8 2.0
.5 .2
.o . I
. I 1.1
. I .2
.4 2.6
.2 .2
2.1 1.1
.5 .9
.o
1 .o
.5
.o
.2
.2
1.3
.1
1.9
. 8
. . . 2.5 6.1 4.5 4.7 8.4 6.0
Note: Based on the industry distributions of 19- to 64-year-olds in 120 major SMSAs in the 1980 Census,
a All natives include all education groups. Other groups are defined in the note to table 7.2.
employment in industry i , and let AE, represent the increase in labor supply
to the ith industry associated with the amval of a fixed number of new immi-
grants AE. The average proportional increase in labor supply experienced by
the native group is
212 Joseph G. Altonji and David Card
Suppose that new immigrants sort themselves into industries in the same pro-
portions as existing immigrants. Then AE, = S,,AE, where s,, is the share of
existing immigrants employed in industry i. Finally, E, = S, E, where S, is the
share of all workers in industry i, and E is level of total employment in the
labor market. Thus, the average proportional increase in labor supply experi-
enced by the native group is PAEIE, where
This expression reduces to one in the case of a homogeneous labor market, in
which S,, = S,, = S, . In a heterogeneous labor market, however, the average
proportional increase in labor supply experienced by a particular native group
may be more or less than AEIE, depending on the degree of similarity be-
tween the industry distributions of immigrants and the native group.
Estimates of this index of labor market competition are presented in table
7.4 for the four groups of less-skilled natives. We have calculated the index
separately using the 1970 and 1980 industry distributions of natives and im-
migrants. We have also calculated the index separately over two subsets of
cities: the twenty cities with the highest fraction of less-skilled immigrants in
1980 and the forty cities with the lowest fraction of less-skilled immigrants
in 1980. These cities are identified in Appendix D.
Estimates of the index of labor market competition are very similar using
the 1970 and 1980 industry distributions. The values of the index range from
a low of .85 in 1980 for white males in low-immigrant cities to 1.28 in 1970
for black females and are consistently below one for black males. The results
confirm the impression that black females are in most direct competition with
immigrants, whereas black males are most isolated from immigrant competi-
tion. Nevertheless, the values of the index are not far from one for any of the
groups, suggesting that increases in the Share of immigrants in the labor mar-
ket have roughly proportional effects on the labor markets of unskilled na-
t i ve~. *~ The differences in the index between high- and low-immigrant cities
are positive for males and negative for females, suggesting that immigrants
and native males are in more direct contact in high-immigrant cities while
immigrants and native females are in less direct contact. One interpretation of
this finding is that less-skilled native females have been displaced from
immigrant-intensive industries in high-immigrant cities. We explore this hy-
pothesis next.
Evidence on the extent of industry displacement is presented in tables 7.5
and 7.6, which give the cross-sectional and time-series patterns of differences
in the industry distributions of less-skilled natives in high-immigrant and low-
immigrant cities. For ten high-immigrant-share industries and ten major
immigrant-employing industries, table 7.5 displays the relative share of un-
skilled natives in high- versus low-immigrant cities. Specifically, let EE, and
Ef;, represent the employment of native group N in industry i in high-
immigrant and low-immigrant cities, respectively. Let EY and Ef represent
213 Immigration and the Labor Market Outcomes of Less-skilled Natives
Table 7.4
Estimated Index of Labor Market Competition between
Immigrants and Natives
High-Immigrant Low-Immigrant
All Cities Cities Cities
Native Group 1970 1980 1970 1980 1970 1980
1. White male
2. White female no
3. Black males no
4. Black females no
dropouts 1.06 1 .oo 1.09 1.03 .99 .85
college 1.09 1.08 1.05 1.03 1.10 1.12
college .94 .94 .91 .93 .91 .91
college 1.24 1.15 1.28 1.06 1.20 1.16
Nore: For definition of index, see the text. High-immigrant cities include 20 SMSAs with highest
fraction of less-skilled immigrants. Low-immigrant cities include 40 SMSAs with lowest fraction
of less-skilled immigrants.
total employment in industry i in these cities, and let E; and Ek represent total
employment of the native group in these cities. For each industry and native
group, table 7.5 displays the ratio
E; , l EB E, H/ EH
Ek t l Ef . ’ E k / E L ’
which represents the relative employment share of natives in the ith industry
in high- versus low-immigrant cities, divided by the relative shares of natives
in total employment in those cities. A value of unity indicates that natives
have equal shares of employment in the industry in the two groups of cities,
controlling for their relative shares in total employment. A value of less than
unity, on the other hand, indicates relative displacement in the high-
immigrant-fraction cities.
For most of the high-immigrant-share industries, there is evidence of dis-
placement of natives in the high-immigrant-share cities. The displacement ef-
fects are less apparent for white males, with ratios in excess of unity for four
industries .24 For the other three groups, however, relative employment shares
in the set of high-immigrant cities are generally less than unity. By compari-
son, the evidence of displacement of less-skilled natives from the major
immigrant-employing industries in the lower panel of table 7.5 is mixed. On
balance, these data suggest that the industry displacement of natives is re-
stricted to low-wage service and manufacturing industries and agriculture. As
the ratios in the right-hand column of table 7.5 suggest, these industries are
generally more important in high-immigrant than low-immigrant cities, al-
though in cross section it is difficult to distinguish alternative explanations for
this effect.25
Table 7.6 repeats the analysis in table 7.5, taking the ratio of the relative
_ _ _ - ~
214 Joseph G. Altonji and David Card
Table 7.5 Relative Industry Distributions of Natives in High- and Low-Immigrant
Cities, 1980
Relative Share of Native Group:
High- vs. Low-Immigrant Cities' High-
% of All vs. Low-
Immigrants White White Black Black Immigrant
Industry in Industry Males Females Males Females Citiesb
High immigrant share:
I . Apparel
2. Leather
3. Agriculture,
4. Furniture
5. Miscellaneous
manufacturing
6. Private household
services
7. Hotels and
motels
8. Transportation
services
9. Restaurants and
bars
crops
10. Textile mills
Other major immigrant
employers:
1. Hospitals and
health services
2. Construction
3. Education
4. Business services
5. Electrical equip-
6. Machinery
7. Transportation
equipment
8. Grocery stores
9. Wholesale trade:
ment
nondurables
10. Food products
5.1
.6
1.5
1 .o
2.3
1.4
2.2
.5
6.4
.8
8.4
5.7
4.5
3.3
3.3
3.2
2.7
2.6
2.5
2.1
1.43
1.33
.56
.64
.83
...
1.42
.59
1.32
.73
1.71
.97
.94
1.51
.75
.91
.78
1.61
1.27
.81
.49
.71
.86
.68
1.04
.65
.91
1.12
.80
.77
.89
1.04
1.15
.81
1.13
1.62
1.52
.89
.94
1.35
1.29
.62
.84
.68
.65
.35
.67
.09
.95
1.22
1.48
.83
1.07
1.18
.61
.84
.74
1.89
.96
.65
.44
.97
.74
.36
.66
.79
.54
1.33
.50
.65
I .07
.81
I .oo
.99
.82
1.32
.72
.98
I .33
.70
2.64
1.40
1.71
.94
1.89
1.25
1.25
2.29
1.01
.51
.91
1 .oo
.89
1.51
1.17
.68
.74
1.03
1.17
.79
Note: Based on the industry distributions of 19- to 64-year-olds in 120 SMSAs in the 1980 Census.
High-immigrant cities include 20 SMSAs with the highest fraction of less-skilled immigrants. Low-
immigrant cities include 40 SMSAs with the lowest fraction of less-skilled immigrants.
a For each industry and native group, the relative share is the proportion of industry employment contrib-
uted by the native group in high-immigrant cities, divided by the same proportion in low-immigrant
cities. This ratio is then divided by the ratio of the shares of the native group in total employment in the
two groups of cities.
Ratio of industry share of total employment in high-immigrant cities to industry share of total employ-
ment in low-immigrant cities.
215 Immigration and the Labor Market Outcomes of Less-skilled Natives
Table 7.6 Relative Growth of Employment Shares of Natives in High- and Low-
Immigrant Cities, 1970-80
Relative Growth of Native Group:
High- vs. Low-Immigrant Citiesa of Total Growth
Employment: of Total
White White Black Black High- vs. Low- Employment
Relative Growth
Industry Males Females Males Females Immigrant Citiesb All Citiesc
High immigrant share:
1. Apparel
2. Leather
3. Agriculture,
4. Furniture
5. Miscellaneous
manufacturing
6. Private household
services
7. Hotels and
motels
8. Transportation
services
9. Restaurants and
bars
crops
10. Textile mills
Other major immigrant
employers:
1. Hospitals and
health services
2. Construction
3. Education
4. Business
services
5. Electrical
equipment
6. Machinery
7. Transportation
equipment
8. Grocery stores
9. Wholesale trade:
nondurablesd
10. Food products
1.73
1.33
.43
.77
.67
. .
1.47
.61
1.36
.94
1.75
.89
1 .oo
1.28
.60
.79
.83
1.33
. .
.75
.85
1.72
.72
.88
.91
.72
1.15
2.23
.97
.88
1.04
.77
1.15
.89
1.08
1.05
1.54
I .04
. .
1.09
.82
. I9
1.29
1.26
.75
.38
.71
.04
.89
2.01
1 .oo
.77
1.27
.56
.66
.65
1.07
1.15
. .
.89
.39
.43
1.45
1.59
.33
.83
.72
2.16
.98
.95
1.08
.72
I .52
.67
.64
.52
.87
1.14
. .
.72
I .30
3.10
1.88
1.06
1.11
1.55
.93
.68
.67
.62
.95
.85
.96
.52
. I6
.39
.94 1.05
.82 .56
.91 1.17
1.13 1.03
.82 .89
.97 I .32
1.38 .75
1.40 .93
1.11 .78
I .07 .92
.93 .78
Note: For definitions of high-immigrant and low-immigrant cities, see the note to table 7.5.
a For formula, see the text.
cities.
Relative ratio of 1980 to 1970 employment totals for industry in high-immigrant vs. low-immigrant
Ratio of 1980 to 1970 employment totals for industry in all cities.
Data for wholesale trade nondurables industry not available.
216 Joseph G. Altonji and David Card
employment share of natives in 1980 to the relative employment share in
1970. A value of unity for this ratio suggests that natives have maintained
their relative share of industry employment, controlling for the relative growth
of total employment of natives in the two sets of cities. A value of less than
unity, on the other hand, suggests that natives have lost relative share in the
industry in high-immigrant versus low-immigrant cities.26
The results in table 7.6 are generally consistent with those in table 7.5 and
suggest some movement of less-skilled natives out of high-immigrant-share
industries in the high-immigrant cities between 1970 and 1980. The fifth col-
umn of the table indicates the relative growth of total employment by industry
in high- versus low-immigrant-share industries, while the sixth column gives
the ratio of total employment in the industry in 1980 in all cities to total em-
ployment in all cities in 1980. Although several high-immigrant industries
were declining relatively quickly between 1970 and 1980, in most cases the
relative decline was slower in high-immigrant cities. This suggests that the
availability of immigrant labor may allow certain industries to survive in high-
immigrant cities even at the same time as natives continue to exit from these
industries.
Our analysis of the industry distributions of immigrants and 1ess:skilled
natives suggests three conclusions. First, a 1 percentage point increase in the
share of immigrants generates approximately a 1 percent increase in the sup-
ply of labor to industries in which less-skilled natives are employed. There is
no indication that immigrants and less-skilled natives are concentrated in par-
ticular industries in a manner that would greatly accentuate the labor market
competition between them or, on the other hand, substantially reduce the de-
gree of labor market competition between them. Second, among the four na-
tive groups that we consider, immigrants are most directly competitive with
black females and least competitive with black men. Third, differences in in-
dustry distributions between high- and low-immigrant cities suggest that na-
tives have been displaced from some low-wage service and manufacturing
industries and that these industries have declined less quickly in cities with
more immigrants.
7.3 An Analysis of the Effects of Immigration on Less-skilled Natives
In this section, we examine the correlation across cities between the labor
market outcomes of less-skilled natives and the fraction of immigrants in the
city. We present cross-sectional analyses for 1970 and 1980 as well as a first-
differenced analysis of changes between 1970 and 1980. Our basic approach
is very simple. We regress SMSA averages of the labor market outcome vari-
ables for our four racehex groups against measures of the immigrant fraction
in the SMSA and a variety of controls for the characteristics of each city.
Before turning to the results of the analysis, however, we first discuss the
construction of SMSA means for the outcome variables. We then briefly dis-
217 Immigration and the Labor Market Outcomes of Less-skilled Natives
cuss potential econometric problems with the cross-sectional and first-
differenced analyses and offer some comments on the interpretation of our
estimates.
7.3.1 Construction of SMSA-Level Outcome Measures and
Control Variables
The first step in our analysis is to construct SMSA-specific means of the
outcome variables that are purged of differences in the observable character-
istics of the native population across different cities. Given the limited infor-
mation collected in the Census, this step amounts to regression adjusting the
outcome variables for differences in age and education. Such an adjustment
has two potential advantages. First, it should reduce the sampling variation
associated with the means of the outcome variables across different cities.
Second, it should eliminate any bias arising from correlations between the
fraction of immigrants in a city and the age and educational attainment of
natives.
For each racehex group in each of the two Censuses, we regress each of the
outcome variables against a full set of SMSA dummies and a flexible function
of age and education. Specifically, we include a cubic polynomial in age, a
detailed set of dummy variables for different education levels, and a full set of
interactions of age and education up to the second order. We then use the
estimated SMSA dummies as our regression-adjusted outcome measures.27
The explanatory variables in the second step of our analysis include the
fraction of immigrants in each SMSA and three additional control variables:
the logarithm of SMSA population and SMSA-specific means of age and edu-
cation for the particular racehex group under consideration. Although the out-
come variables are adjusted for age and education, we found in preliminary
work that the mean of adjusted weekly earnings is correlated across cities with
the mean of education, particularly for blacks. We have no explanation for
this phenomenon, although it may indicate a correlation across cities between
the quality and the quantity of education among blacks or possibly a market
externality associated with higher levels of education among the less-skilled
black population. In any case, we include SMSA-specific means of age and
education for the particular racehex group in all our SMSA-level regressions.
These means are calculated directly from our native extracts.
Our measure of the fraction of immigrants in each SMSA is the fraction of
foreign-born residents, taken from published tabulations of the 1970 and 1980
Censuses. From the standpoint of the theoretical model, it would be prefer-
able to use the fraction of immigrants in the local labor force. Since our
sample sizes for 1970 are too small to provide reliable estimates of the fraction
of immigrants in many of the smaller cities, we have relied instead on the
published population data. Provided that changes in the immigrant labor force
are proportional to changes in the population of immigrants, the use of frac-
tion of immigrants in the population will not affect our results.
218 Joseph G. Altonji and David Card
7.3.2 Econometric Issues
We next turn to a brief discussion of our estimating equations. We focus on
three issues: possible sources of bias in the estimating equations; the interpre-
tation of differences between cross-sectional and first-differenced estimates of
the effects of immigration; and the use of weighted least squares in the esti-
mation.
Our cross-sectional estimating equations have the form
(7)
where 9, is the adjusted labor market outcome for native group N in city j ,
X, is a vector of control variables for the racehex group and city (the mean of
age and education for the group and the logarithm of SMSA population), f , is
the fraction of immigrants in the city, and e,, is a residual term. Similarly, our
first-differenced estimating equations have the form
AFNJ = AXNj b + Af, c + Ae,,,,,
where AZj refers to the change in the variable Z in ci t yj between 1970 and
1980.
Depending on the choice of outcome measure I: these equations have the
form of equations ( 5) or (6) derived from our theoretical model. The interpre-
tation of estimates of the coefficient c obtained from equation (7) or (8), how-
ever, depends on the nature of the residual terms in these equations. These
residuals can be decomposed into two conceptually distinct components: (1) a
market-level SMSA effect due to factors other than immigration (e.g., unmea-
sured characteristics of natives or demand shocks affecting the local economy)
and (2) sampling variation arising from the fact that we observe only a sample
of natives in each SMSA. Let Y, represent the true population value of the
outcome variable for natives in city j . Then we may decompose eNj as
where a,,,, represents the SMSA effect due to factors other than immigration,
and 9, - Y, is the component of e, attributable to sampling variability. Only
if a,, is orthogonal to the fraction of immigrants in the city will estimates of
the coefficient c from the cross-sectional regression (7) yield unbiased esti-
mates of BU or E * B, , as described by equation ( 5 ) or ( 6) . In the first-differ-
enced specification, the corresponding requirement is that changes in the un-
measured SMSA effects be uncorrelated with changes in the fraction of
immigrants in the city between 1970 and 1980.
Clearly, the main advantage of the first-differenced analysis is that it elimi-
nates any bias introduced by city-specific fixed effects that are correlated with
the fraction of immigrants in a city and the labor market outcomes of natives.
Transitory effects (associated with transitory fluctuations in the demand for
the output of specific cities, e.g.) will still lead to biases in the differenced
219 Immigration and the Labor Market Outcomes of Less-skilled Natives
analysis if they influence the inflow rate of immigrants. Bartel’s (1989) recent
analysis suggests that economic conditions have a relatively small effect on
the destination city chosen by immigrants. Instead, Bartel’s findings suggest
that immigrants are mainly attracted to cities with large concentrations of pre-
vious immigrants from the same country (see also Greenwood and McDowell
1986). Nevertheless, her research leaves open the possibility that the timing
and size of immigrant inflows are affected by economic conditions in particu-
lar cities.
We attempt to control for any potential correlation between immigrant in-
flows and local economic conditions in our first-differenced analysis by an
instrumental variables procedure. As suggested by Bartel’s (1989) work, we
use the fraction of immigrants in a city in 1970 to predict the change in the
fraction of immigrants over the following decade.28 Immigrant inflows are
strongly correlated with the initial fraction of immigrants in a city, and these
variables are reasonably strong predictors of the change in immigrant fraction.
In comparing the cross-sectional and first-difference results, one should
also keep in mind that the first-difference analysis is more likely to capture the
short-run effects of immigration, in which the capital stock and the industry/
skill composition of labor demand have not had time to adjust fully. The ef-
fects of immigration on per capita employment rates and wages may weaken
over time as natives move to other cities or to labor market sectors that are
less affected by immigrant competition. Dynamic issues are not addressed in
our formal model, but we suspect that the short-run effects of immigration on
employment of less-skilled natives will be larger than the long-run effects.
The relative magnitude of the short-run and long-run effects on wages depend
on whether there are barriers to wage adjustments in the short run. In fact, we
find that the cross-sectional estimates of the effect of immigration on employ-
ment outcomes of natives are larger than the differenced estimates, whereas
the opposite is true of the estimated effects on wages. This leads us to suspect
that the differences between the cross-sectional and the differenced results are
primarily due to correlations between city-specific effects and immigrant
shares that are eliminated in first-differences rather than to a distinction be-
tween long-run and short-run effects.
A final econometric issue arises from the relatively small samples of black
natives in many cities, particularly in our 1970 sample. We restrict our cross-
sectional and differenced analysis of each racekex group to the set of cities
for which we have at least thirty group members in both 1970 and 1980. Con-
sequently, we work with a set of ninety-one cities for black males, a set of
ninety-four cities for black females, and a full set of 120 cities for white men
and women. We also use weighted least squares methods to estimate our equa-
tions, using the square root of the number of observations for the race/sex
group in the city as a weight. In our first-differenced specifications, we use as
a weight (N,’ + N G * ) - ~ ’ ~ , where N, , and N, , are the number of observations
for the native subgroup in the SMSA in 1970 and 1980, respe~tively.~~ This
220 Joseph G. Altonji and David Card
weighting scheme assumes that the residual eN arises mainly from sampling
variability associated with the estimated outcome measure. Even controlling
for the covariates in our models, however, the labor market outcomes of dif-
ferent racehex groups are correlated across cities, suggesting the presence of
omitted city-specific effects. We have not adjusted our standard errors or esti-
mation procedures to take account of such error components.
7.3.3 Empirical Results
To provide an introduction and overview of our results, table 7.7 presents
weighted least squares estimates of the effects of immigration on the labor
market outcomes of the pooled set of four race/sex groups. The estimated
equations include unrestricted intercepts for the four groups as well as group-
specific coefficients on the means of age and education. The coefficients on
the immigrant share variable and the population variable, however, are re-
stricted to be the same across the four native subgroups.
The cross-sectional results for 1970 show significantly negative effects of
an increase in immigrant shares on the labor force participation rates and em-
ployment rates of less-skilled natives. The results imply that a 10 percentage
point increase in the fraction of immigrants in an SMSA would lead to a re-
duction in the employment/population ratio of less-skilled natives of roughly
2 percent. The employment rate would also fall by 1 percent, implying an
increase in unemployment rates of about 1 percent. Among those who work,
average weeks per year would fall by about 2 percent.
Table 7.7 Effects of Immigration on Four Groups of Less-Skilled Natives,
Pooled Sample (standard errors in parentheses)
Cross-sectional First-Differenced
Outcome Variable 1970 1980 1980-70 1980-70 IV’
1. Labor force/
population
2. Employment/
population
3. Employment/labor
force
4. Fraction worked last
Ye=
5 . Log weeks worked
6. Log earningslweek
- ,173
(.066)
- ,240
( ,074)
(.036)
(.063)
-.191
(.078)
.467
(.165)
- ,109
-.161
- .083
(.049)
(.ow
- .054
,019
- .I58
(.050)
- ,088
(.061)
,018
(.112)
,080
(.083)
,404
(.097)
.461
(.077)
.090
(.084)
.232
(.132)
- ,262
(.228)
- ,102
(. 122)
,085
(. 144)
,231
(.113)
- ,246
(.125)
,142
(.193)
- 1.205
(.342)
Note: All equations included subgroup-specific intercepts, the total population in the SMSA, and
the average education and age of the subgroup in the SMSA (with subgroup-specific coefficients).
The sample size is 424.
a Estimated by instrumental variables. The change in the fraction of immigrants in the SMSA is
instrumented with the fraction of immigrants in 1970 and its square.
221 Immigration and the Labor Market Outcomes of Less-skilled Natives
These negative employment effects contrast sharply with the finding that
immigration has a positive effect on weekly wages. The estimated coefficient
in row 6 implies that a 10 percentage point increase in the immigrant share
would lead to a 4.7 percent increase in weekly earnings. Within the context
of our model, these results can be reconciled only if the labor supply elasticity
of less-skilled natives is negative.3o
The 1980 cross-sectional results for the various employment outcomes also
indicate a negative effect of immigration, although the estimated coefficients
are smaller in magnitude than those for 1970. In the 1980 data, however, the
estimated effect of immigrant densities on the average weekly earnings of na-
tives is essentially zero. This gives further reason for caution in the interpre-
tation of the 1970 results.
Weighted least squares estimates of the first-differenced specification are
presented in the third column of table 7.7. In contrast to the cross-sectional
results, these estimates suggest a modest positive effect of the fraction of im-
migrants on the employment outcomes of natives. The estimated effect on
earnings per week is negative ( - .267) but not statistically different from
zero.
Instrumental-variables estimates of the first-differenced specification are
presented in column 4. These estimates give an ambiguous picture of the ef-
fect of immigration on the employment outcomes of natives. A marginally
significant positive effect on the employment rate in the Census week is coun-
terbalanced by a marginally significant negative effect on the employment-
population ratio last year. Nevertheless, the instrumented first-differenced re-
sults indicate a significantly negative effect of immigration on wages. The
coefficient is - 1.2 with a standard error of .242. The more negative effect
associated with the instrumental variables estimation scheme is consistent
with the hypothesis that the least squares estimate is positively biased by en-
dogenous immigration inflows.
On balance, the pooled data suggest that the effect of immigrant densities
on the employment and participation rates of natives is small and potentially
zero. If the instrumented first-differenced specification is taken at face value,
however, the effect on wages is apparently negative. For the most part, these
conclusions carry over to the detailed results for the four subgroups, to which
we now turn.
Results for Individual RacelSex Groups
Estimates of the relation between immigrant fractions and the labor market
outcomes of black males are presented in table 7.8, which has the same format
as table 7.7. As in the pooled analysis, the cross-sectional results for black
men suggest a negative correlation between the fraction of immigrants and
employment outcomes. In the differenced analysis, however, the relation is
much less consistent. Likewise, although the 1970 cross-sectional analysis
suggests a positive effect of immigration on black male wages, the 1980 cross-
sectional results and the differenced results indicate a negative effect.
222 Joseph G. Altonji and David Card
Table 7.8 Effects of Immigration on Black Males with Less than Thirteen Years
of Education (standard errors in parentheses)
Cross-sectional First-Differenced
Outcome Variable 1970 1980 1980-70 1980-70 IVa
1. Labor force/population - ,145
(.126)
2. Employment/population - ,264
(.156)
3. EmploymenVlabor force - ,165
(. 090)
4. Fraction worked last year - .183
(.loo)
5. Log weeks worked - .I54
(.121)
6. Log earningsiweek ,736
(. 346)
- ,136
(.084)
(.115)
,046
(.098)
- .214
(.081)
( . I l l )
- .153
(. 248)
- ,068
- ,051
- ,040
(.170)
.658
(.234)
,864
(.210)
.I01
(.168)
- ,447
(.252)
- .806
(.494)
- ,273
(.240)
,285
(.234)
,623
( .294)
- ,268
(.168)
,272
(.351)
- 1.910
( . 7W
Note: All equations include average age and education in the SMSA as well as total population.
The sample size is 91.
a Estimated by instrumental variables. See the note to table 7.7.
Table 7.9 Effects of Immigration on White Males with Less than 'helve Years
of Education (standard errors in parentheses)
Cross-sectional First-Differenced
Outcome Variable 1970 1980
1. Labor force/population - ,193
(.075)
2. EmploymenVpopulation - .279
(.101)
3. EmploymenVlabor force - ,107
(.053)
4. Fraction worked last year -.151
(.070)
5. Log weeks worked
(.074)
6. Log earningsiweek - ,264
(.201)
- ,223
- ,079
(.083)
- ,159
(.112)
-. I10
(.074)
- ,215
(.078)
- ,312
(.I%)
-. I78
(.212)
1980-70
,066
(.149)
,349
(.186)
.343
(. 134)
- .145
(.136)
- ,018
(.211)
- ,356
(.406)
1980-70 IV'
,036
(.231)
,109
(.289)
.086
(.211)
- ,609
(.211)
(.328)
(.637)
- ,190
-1.103
Note: All equations include average age and education in the SMSA as well as total population.
The sample size is 120.
a Estimated by instrumental variables. See the note to table 7.7.
The results for white male dropouts are presented in table 7.9. These results
are very similar to those for black males, although the point estimates of the
effects of immigration on wages are somewhat smaller in magnitude. Again,
the differenced specifications in particular suggest a negative effect of immi-
grant densities on native wage rates, while the effects on employment and
223 Immigration and the Labor Market Outcomes of Less-skilled Natives
participation rates are smaller and vary with the precise measure of employ-
ment.
The regression results for black females in table 7.10 are of particular inter-
est, given the evidence in section 7.2 that black women are in closer compe-
tition with immigrants than the other three groups. Nevertheless, the esti-
mated coefficients for this group are not much different than those for the other
groups. The cross-sectional results suggest a small negative effect of immi-
grant shares on employment outcomes and a modest positive effect on weekly
wages. These conclusions are reversed, however, in the first-differenced anal-
ysis, which suggests a generally positive effect on employment rates and a
negative effect on wage rates. The differenced results for black females are not
particularly sensitive to choice of least squares or instrumental variables esti-
mation, although as in previous tables the strongest negative wage effect is
obtained by the instrumental variables procedure.
Table 7.11 presents our results for white females. Again, the cross-
sectional results for 1970 indicate a negative relation between immigrant
shares and employment outcomes, while the differenced analysis indicates
much weaker effects. The cross-sectional and first-differenced specifications
fit by least squares suggest a positive effect of immigrant shares on wage rates.
When the change in immigrant share is instrumented, however, the estimated
wage coefficient is negative and consistent with the results for the other native
groups.
A check on the wage effects reported for the different native groups in tables
7.7-7.11 is contained in table 7.12. Here, we estimate the same specifications
using the wage outcomes of immigrant workers as the dependent variable. We
Table 7.10 Effects of Immigration on Black Females with Less than Thirteen
Year of Education (standard errors in parentheses)
Cross-sectional First-Differenced
Outcome Variable 1970 1980 1980-70 1980-70 IVa
1. Labor force/population
2. Employment/
population
3. Employmentflabor
force
4. Fraction worked last
Yea
5. Log weeks worked
6. Log eamingsiweek
- ,216
- ,221
- ,037
(.179)
(.192)
(.105)
(.169)
- ,247
(.232)
1.213
(.402)
- ,165
- .063
(.119)
,003
(.128)
.073
(.086)
- ,127
(.120)
.143
(. 143)
.533
(.236)
- .154
(.256)
.149
( ,269)
.457
(.186)
,054
(.272)
.735
(.387)
- .838
( ,609)
~
- .221
.032
,320
(.259)
- ,219
(.379)
.217
(.542)
- 1.369
(. 848)
(.357)
(.374)
Note: All equations include average age and education in the SMSA as well as total population.
The sample size is 94.
a Estimated by instrumental variables. See the note to table 7.7.
224 Joseph G. AItonji and David Card
Table 7.11 Effects of Immigration on White Females with Less than Thirteen
Years of Education (standard errors in parentheses)
Cross-sectional First-Differenced
Outcome Variable 1970 1980 1980-70 1980-70 IV"
I . Labor force/population
2. Employment'population
3. Employment'labor
4. Fraction worked last
5. Log weeks worked
6. Log eamingslweek
force
Year
- .037
(.144)
(. 150)
(.058)
(.145)
- .094
(.170)
,667
( ,245)
- ,095
-.132
- ,047
,058
(.097)
.027
(.105)
- .045
( . 045)
.005
(.098)
-. I18
(.110)
,397
(.132)
,273
(.137)
.420
(.154)
,306
(.125)
. I89
(.146)
,133
(.270)
.309
(.430)
- ,044
(.207)
- ,089
( ,240)
- ,017
(.190)
- ,162
(.222)
,335
( .399)
- ,955
(.663)
Note: All equations include average age and education in the SMSA as well as total population.
The sample size is 120.
a Estimated by instrumental variables. See the note to table 7.7.
Table 7.12 Effects of Immigration on Male Immigrant Wages (standard errors in
parentheses)
~ ~ ~~
Cross-sectional First-Differenced
Outcome Variable 1970 1980 1980-70 1980-70 IV'
1. Log eamings/week - ,459 - ,741 - ,504 - ,823
2. Log earningdweek ,116 - ,499 - ,958 - 1.492
(unadjusted) (.357) (.181) (.381) (S12)
(adjusted) (.302) (.167) (.354) (.481)
Nare: Immigrant group includes males age 16-64 not in school in Census week. All equa-
tions include average age and education in the SMSA as well as total population. The sample size
is 74.
a Estimated by instrumental variables. See the note to table 7.7
use two measures of immigrant wages: the mean of actual log weekly earnings
for male immigrants and an adjusted mean that controls for the average levels
of age and education of immigrants in each city. The results reveal three find-
ings. First, unadjusted mean earnings of immigrants are more strongly corre-
lated in cross section with the fraction of immigrants than mean earnings that
have been adjusted for measured skill attributes. This suggests a negative cor-
relation between the skill level of immigrants and their fraction in the popula-
tion. Second, as we found for the native groups, the instrumental variables
estimate of the first-differenced specification leads to the largest negative es-
timate of the effect of immigrant densities on wages. Finally, the instrumental
variables estimates of the effect of immigrant shares on immigrant wages is
very similar to the corresponding estimate for native wages. There is no evi-
225 Immigration and the Labor Market Outcomes of Less-skilled Natives
dence that immigrants have a stronger negative effect on their own wages than
on those of less-skilled natives.
Other Results
We estimated many of our least squares models for the 1970, 1980, and
differenced samples with a control for the fraction of blacks in the SMSA
population. This addition made little difference to the results.
We also reestimated many of our specifications using the fraction of “less-
skilled” immigrants in the SMSA population in place of the overall fraction of
immigrants in the SMSA population. We defined the fraction of “less-skilled”
immigrants as the product of the fraction of immigrants in the SMSA popula-
tion and the fraction of male immigrants in the SMSA whose predicted e m-
ings are less than the national median for male immigrants (see App. D). The
(unweighted) correlation across 120 cities between the “less-skilled” immi-
grant fraction and the total immigrant fraction is .94 in 1970 and .95 in 1980.
The correlation of changes in the two immigrant measures is .82. Perhaps as
a result, least squares results using the fraction of less-skilled immigrants are
similar to those reported in tables 7.7-7.11. The regression coefficients typi-
cally increase in absolute value, reflecting the fact that the scale of the less-
skilled immigrant variable is compressed relative to the other variable. It is
worth noting that instrumental variables estimates (using the fraction of im-
migrants in the SMSA in 1970 and its square as instruments) point to a some-
what larger negative effect of the fraction of less-skilled immigrants on the
weekly earnings of natives. The coefficients for black males, white males,
black females, and white females are -7.0, -4.8, - 12.9, and - 12.3, re-
spectively. These estimates are very imprecise, however, perhaps because the
correlation between fraction of immigrants in 1970 and the change in fraction
of less-skilled immigrants in the SMSA is only .27.31
Finally, we reestimated the 1980 cross-sectional specifications and the first-
differenced specifications for each of our labor market outcome variables
using the SMSA-specific mean of the corresponding labor market outcome for
white males age 31-64 with thirteen or more years of schooling as a control
variable. We view this approach, which uses the labor market outcomes of
highly skilled workers to control for general labor market conditions within
each city, as an alternative to our instrumental variables procedure. It is
strictly correct only if, in contrast to the implications of our model, immigra-
tion has no effect on more highly educated whitemales. The results from this
alternative procedure are generally similar to our ordinary least squares esti-
mates and suggest smaller negative effects of immigration on less-skilled na-
tive wages than the instrumental variables procedure.
7.4 Conclusions
This paper presents a variety of evidence on the effects of immigration on
the labor market outcomes of less-skilled natives. Working from a simple
226 Joseph G. Altonji and David Card
theoretical model of a local labor market, we show that the effects of immigra-
tion can be estimated from the correlations between the fraction of immigrants
in a city and the employment and wage outcomes of natives. We go on to
compute these correlations using city-specific outcomes for individuals in 120
major SMSAs in the 1970 and 1980 Censuses. We also use the relative indus-
try distributions of immigrants and natives to provide a direct assessment of
the degree of labor market competition between them.
Our empirical findings indicate a modest degree of competition between
immigrants and less-skilled natives. A comparison of industry distributions
shows that an increase in the fraction of immigrants in the labor force trans-
lates to an approximately equivalent percentage increase in the supply of labor
to industries in which less-skilled natives are employed. Based on this calcu-
lation, immigrant inflows of the magnitude observed between 1970 and 1980
generated 1-2 percent increases in labor supply to these industries in most
cities. A comparison of the industry distributions of less-skilled natives in
high- and low-immigrant-share cities between 1970 and 1980 shows some
displacement of natives out of low-wage immigrant-intensive industries.
We find little evidence that inflows of immigrants are associated with large
or systematic effects on the employment or unemployment rates of less-skilled
natives. Our estimates of the effect of immigration on native wage rates are
sensitive to the choice of specification and estimation procedure. When we
consider first-differences between 1980 and 1970 and use an instrumental var-
iables estimation procedure to control for endogeneity of immigrant inflows,
we find that a 1 percentage point increase in the fraction of immigrants in an
SMSA reduces less-skilled native wages by roughly 1.2 percent. The least
squares estimates imply a wage reduction of . 3 percent. We point out a num-
ber of reasons to prefer the instrumental variables procedure, but additional
research, perhaps with the 1990 Census, will be required before one can draw
strong conclusions about the response of wages to immigration.
Appendix A
Sampling Procedures and Variable DeJnitions
Sampling Procedures
Our 1970 samples are drawn from the 1/100 County Group Public Use
Sample based on the 5% version of the 1970 Census questionnaire. The
sample universe consists of all individuals age 19-64 currently residing in
one of 120 SMSAs. (The samples actually contain 121 SMSAs, but, for com-
parability with the 1980 Census, Dallas and Fort Worth are considered as one
SMSA). As described in the text, our analysis is limited to individuals not
227 Immigration and the Labor Market Outcomes of Less-skilled Natives
currently enrolled in school and in specific race/sex/education and national
origin groups from this universe.
Our 1980 samples are drawn from the 9100 Public Use “A” Sample of the
1980 Census. The sample universe consists of all individuals age 19-64 cur-
rently residing in one of 120 SMSAs (adjusted to 1970 boundaries: see App.
B). To limit the size of the samples, stratified random samples of individuals
meeting the above requirements were drawn by SMSA. Samples of native-
born nonblacks (i.e., race coded as white, American Indian, Asian, or other)
were drawn to generate approximately twenty-three hundred observations per
SMSA for all agelsexleducation levels. The samples were then further re-
stricted to two subsets of observations: females with twelve or fewer years of
completed education and males with eleven or fewer years of completed edu-
cation. Samples of native-born blacks were drawn to generate a maximum of
500 observations per SMSA for black females with twelve or fewer years of
completed education and 500 observations per SMSA for black males with
twelve or fewer years of completed education. One hundred percent samples
of foreign-born individuals were taken for all but five large SMSAs, which
were sampled with the following probabilities: Chicago, .400; Los Angeles,
.170; Miami, .500; New York, .137; and San Francisco, 3 0 .
Labor Market Outcome Variable Definitions
The following labor market outcome variables are defined for all individu-
als in the sample universe:
employed in the previous year (P35 = 0 in 1970; P94 = 1 in 1980);
in the labor force in the Census week (P31 = 1, 2, 4, 5 in 1970; P81 =
1, 2, 4, 5 in 1980);
employed in the Census reference week (P31 = 1, 2, 4, 5 in 1970; P81
= 1, 2, 4, 5 in 1980).
For individuals in the labor force in the Census week, a fourth variable is
defined to be one if the individual was employed in the Census week and zero
otherwise.
For individuals who worked in the previous year and who reported strictly
positive values for the number of weeks worked in the previous year (P36 =
0-5 in 1970; P95 > 0 in 1980) and earnings in the previous year (P37 =
0-500 in 1970; PlOl > 0 in 1980), two additional variables are defined:
weeks worked in the previous year and earnings per week in the previous year.
For 1980, these variables are constructed directly: weeks worked is measured
by variable P95; and earnings per week is measured by P101/P35. (These
calculations make no adjustments for allocated responses or truncation of the
reported earnings figure.) For 1970, only interval measures of weeks worked
and total annual earnings are available. We assigned midpoints of the intervals
to the weeks and earnings figures and then constructed earnings per week as
the ratio of the assigned values.
228 Joseph G. Altonji and David Card
Appendix B
Matching SMSA Dejinitions between I970 and 1980
The Public Use Samples of the 1970 Census identify 125 individual SMSAs
(see pp. 123-26 of the Description and Technical Documentation for the Pub-
lic Use Samples of Basic Records from the 1970 Census). A total of 120 of
these are used in our statistical analysis. Four SMSAs were deleted because
of difficulty matching between 1970 and 1980 or because of too small sample
sizes: Lorain-Elyria, Ohio; Johnstown, Pennsylvania; San BernadinWRiver-
side, California; and Wilkes Barre-Hazelton, Pennsylvania. The Fort Worth
SMSA was merged with Dallas (see below).
The Census Bureau publication Geographic Ident$cation Code Scheme
(1983, 11-17) gives a detailed list of changes in the county-level definitions
of SMSAs between 1970 and 1980. In most cases, these changes involve the
addition of surrounding counties or parts of these counties to the SMSA. The
major exceptions are (1) the combination of Dallas and Fort Worth into a
single SMSA; (2) the creation of a separate SMSA consisting of Nassau and
Suffolk counties of New York State (formerly part of the New York SMSA);
and (3) the reclassification of Bergen County, New Jersey, from the Paterson-
Clifton-Passaic SMSA to the New York SMSA.
Our general matching strategy was to redefine 1980 SMSA boundaries to
the 1970 boundaries. With only a few exceptions, this involved deleting indi-
viduals from the 1980 Census file who resided in counties that were classified
as part of the SMSA in 1980 but not in 1970. For example, Montgomery
County, New York, was added to the Albany-Schenectady-Troy SMSA in
1973. Individuals in this county were therefore deleted from the 1980 file.
County-level information for each household is coded in the variable COGRP
(location 6-8 of the household record) of the Public Use “A” Sample of the
1980 Census. County group codes are obtained from the 1980 County Group
Equivalence File (1980 Census of Population and Housing, Public Use Micro
Data Sample, part 77) and Appendix M of the 1980 Census Public Use Micro-
data Samples Technical Documentation. In most cases, individual counties
are identified by one or more county group codes. For these cases, the deletion
is accomplished by specifying the county group code(s) of those counties
added to the SMSA after 1970.
In some cases, only parts of a surrounding county group were added to the
SMSA. In these cases, we randomly deleted a fraction of individuals from the
added county or county group. The fraction of individuals deleted was set
equal to the relative population of the part of the county added to the SMSA.
Estimates of population for county subgroups were obtained from the 1980
County Group Equivalency File.
In all, a total of forty-nine counties or county subgroups were deleted from
the definitions of the 120 SMSAs. Another forty counties or county subgroups
229 Immigration and the Labor Market Outcomes of Less-skilled Natives
were partially deleted. The number of individual records actually affected by
these deletion procedures is small. For example, of 244,941 immigrants iden-
tified on the 1980 Public Use A Sample using the 1980 SMSA definitions,
2,609 (1.07 percent) were deleted in the change to the 1970 definitions. A
copy of the computer instructions that performed the deletions is available
from David Card on request.
To account for changes in the classification of Nassau and Suffolk counties
in New York State, we added individuals in the Nassau-Suffolk SMSA in 1980
to the New York SMSA sample. To account for the changes in definition of
the Paterson-Clifton-Passaic SMSA, we added individuals in the 1980 sample
living in Bergen County, New Jersey (classified as part of the New York
SMSA in 1980), to the Paterson-Clifton-Passaic SMSA sample and deleted
them from the New York SMSA sample. To account for the reclassification of
Dallas and Fort Worth into a single SMSA, we combined individuals from the
Dallas and Fort Worth SMSAs in the 1970 Census file into a single Dallas-
Fort Worth sample. No attempt was made to deal with minor reclassifications
affecting the Boston and Providence SMSAs and the Detroit and Flint
SMSAs.
Appendix C
Industry Definitions
Matching of 1970 and 1980 Three-Digit Codes
Our procedure was to reclassify the three-digit industry codes of individuals
in the 1970 Census to 1980 industry codes. The Census Bureau provided us
with cross-tabulations of 1970 and 1980 three-digit industry codes for
samples of males and females who had been coded under both systems. These
cross-tabulations were used to estimate the probability that an individual with
a given 1970 code would be classified in a particular industry under the 1980
coding scheme. Using these probabilities, a computer program was developed
that reclassifies individuals probabilistically from their 1970 three-digit indus-
try to a particular 1980 three-digit industry. The computer program processes
males and females separately. A copy of the program is available from David
Card.
Industry Classifications Used in Tables 7.3-7.6
Using the three-digit industry titles in Appendix H of the Public Use Micro-
data Samples Technical Documentation, we developed a “two-digit” classifi-
cation consisting of seventy-six individual industries. (There are 23 1 separate
industries in the 1980 Census industry coding system.) This classification
combines many smaller three-digit industries: for example, “agricultural ser-
230 Joseph G. Altonji and David Card
vices except horticulture” (industry 020) and “horticultural services” (indus-
try 021). A listing of the computer instructions used to classify three-digit
industries into this two-digit system is available from David Card.
Appendix D
Classification of High- and Low-Immigrant Cities
In order to determine average immigrant skill levels by SMSA, a regression
equation was fit to the log of average weekly earnings for the 1980 sample of
male immigrants. The equation included the same flexible function of age and
education used to regression adjust native outcomes (see the text description)
as well as a set of forty-six countryhegion dummy variables and their interac-
tions with an indicator variable for having entered the United States after 1970
and a variable representing years in the United States. (Chiswick [ 19781, Bor-
jas (1985, 19871, and others have shown that country of origin, immigration
cohort, and years since immigration affect earnings in the United States.) This
equation was then used to assign a predicted wage to each male immigrant.
Immigrants with a predicted wage less than the median predicted wage for the
entire United States were classified as “low skill.’’ Finally, the fraction of low-
Table 7D.1 ’lkenty Cities with Highest Fraction of Low-Skill Immigrants
Fraction Fraction Low-Skill
City Immigrants Immigrants
Miami
El Paso
Los Angeles
Salinas
Jersey City
Oxnard-Ventura
New York
Honolulu
Paterson
Fresno
San Diego
Anaheim
Bakersfield
Stockton
Santa Barbara
San Francisco
San Jose
Houston
San Antonio
Providence
.36
.21
.22
.19
.24
.13
.21
.15
.15
. l l
.13
.13
.09
. l l
.12
.16
.14
.08
.07
.09
.20
.20
.16
.16
.15
.10
.10
.10
.09
.09
.08
.08
.08
.08
.07
.07
.07
.06
.06
.06
231 Immigration and the Labor Market Outcomes of Less-skilled Natives
Table 7D.2 Forty Cities with Lowest Fraction of Low-Skill Immigrants
City
Fraction Fraction Low-Skill
Immigrants Immigrants
Huntington-Ashland, KY .01 .oo
Chattanooga .01 .oo
B irmingham .01 .oo
Knoxville .01 .oo
York, PA .01 .oo
Canton .02 .oo
Jackson, MS .01 .oo
Cincinnati .02 .01
Dayton .02 .01
Flint .03 .01
Appleton .02 .01
Louisville .01 .01
St. Louis .02 .01
Nashville .01 .01
Indianapolis .02 .01
Richmond .02 .01
Duluth .03 .01
Memphis .01 .01
Akron .03 .01
Greensboro .01 .01
South Bend .03 .01
Utica-Rome, NY .04 .01
Erie, PA .03 .01
Pittsburgh .03 .01
Harrisburg .02 .01
Binghampton .04 .01
Greenville .02 .01
Peoria .02 .01
Wilmington .03 .01
Fort Wayne .02 .01
Mobile .01 .01
Madison .03 .01
Lancaster .02 .01
Toledo .03 .01
Youngstown .04 .01
Lansing .03 .01
Columbus .02 .01
Atlanta .02 .01
Minneapolis .03 .01
Shreveport .02 .01
skill immigrants in each SMSA was determined by multiplying the fraction of
immigrants in the SMSA by the fraction of immigrants who are classified as
low skill. Table 7D.1 lists the twenty cities with the highest fraction of low-
skill immigrants. Table 7D.2 lists the forty cities with the lowest fraction of
low-skill immigrants.
232 Joseph G. Altonji and David Card
Notes
1. Most of the available evidence is summarized by Greenwood and McDowell
(1986), General Accounting Office (1988), and Papademetriou et al. (1989). Two stud-
ies of particular relevance to ours are Grossman (1982) and Borjas (1987). Lalonde
and Topel (in this volume) provide a parallel study to ours, focusing on the effects of
recent immigrants on the labor market outcomes of earlier immigrants. Muller and
Espenshade (1985) analyze the effect of immigrants on various California cities.
2. A similar conclusion is reached by Kuhn and Wooton (in this volume) and Papa-
demetriou et al. (1989, ch. 4).
3. The average change in the percentage of immigrants between 1970 and 1980 in
the 120 SMSAs in our sample is 1.4 and ranges between 0 and 11.4 percent.
4. If the price of output is exogenous, it is more convenient to work with the elastic-
ities of factor prices with respect to factor quantities, holding constant marginal cost.
These are usually known as elasticities of complementarity (see, e.g., Hamermesh
1986).
5. This depends, of course, on constant returns to scale and on perfectly elastic
supplies of capital and other inputs.
6. In order to avoid the theoretical prediction of factor price equalization across
cities, it is necessary to assume that the number of goods produced within a city is less
than the number of locally supplied factors. For further discussion of this point, see
Kuhn and Wooton (in this volume).
7. We ignore land or any other locally supplied factors.
8. For notational simplicity, we suppress the dependence of c(. ) on the prices of
nonlabor inputs.
9. In the notation of eqq. (1) and (2), dD,(9, w,)/dw, = 0, and dL,(w,, 9)/d9 = 0, for
j = ( u, s).
10. Johnson (1980a) makes the further assumption that the elasticity of labor supply
among existing immigrants is zero, so that the effective supply elasticity in the market
for unskilled labor is ( I - f,)~, wheref, is the fraction of immigrants in the existing
pool of unskilled workers, and E is the labor supply elasticity of natives.
11. That is, O p , + Osust = .6(O, + Os ) , where 8, represents the value share of
labor in thejth skill group.
12. No entries are included in the first row under the column for a,, = .25. In this
row of the table, a,, is strongly negative ( - S25). Thus, skilled and unskilled labor
must be relatively strong substitutes (i.e., a,, > .8) to satisfy the restrictions on the
matrix of partial elasticities.
13. If a,* = ask, eq. (5) implies that the value of the coefficient b, is independent of
the substitutability between skilled and unskilled labor.
14. The elasticities of demand for unskilled labor with respect to its own wage rate
(qJ implied by the parameter choices in table 7.1 range from - 1 .O (in the lower-left-
hand entries of the table) to - 2.6 (in the upper-right-hand entries of the table).
15. Estimates of the fraction of output produced in a city that is consumed locally
are not easily obtained. Roughly 35 percent of consumer e tures are allocated to
personal, health, business, and education services, public s, transportation ser-
vices, and other goods with a high local content.
16. If the immigrants are primarily unskilled, then one might expect out-migration
of unskilled natives and in-migration of skilled natives.
17. Filer (1988) shows that the net migration rate of natives to an SMSA between
1975 and 1980 is negatively related to the migration rate of immigrants into the SMSA
between 1970 and 1974 and to the migration rate of immigrants into the SMSA be-
233 Immigration and the Labor Market Outcomes of Less-skilled Natives
tween 1975 and 1980. The negative relation appears to be strongest for low-skilled and
less-educated natives.
18. Papademetriou et al. (1989, chap. 4) summarize evidence from a few industry
studies suggesting that in some cases immigrant labor has been used to undercut union
firms paying higher wages and employing native workers.
19. By “Census week” we mean the week immediately preceding the administration
of the Census, for which individuals report their major activity. The Census is admin-
istered on 1. April.
20. Our two-digit industry classification is explained in App. C.
21. Of the ten highest-immigrant-share industries in 1980, seven were in the top ten
industries by immigrant share in 1970. The rank-order correlation across industries
between the 1970 and 1980 immigrant shares is .86.
22. The average fraction of immigrants in the total population in our sample of cities
in 1970 was .044 and ranged from .003 to .242. The average fraction of immigrants in
the total population in 1980 was .058 and ranged from .008 to .357.
23. It should be pointed out that the index is computed from the industry distribution
of existing immigrants and cannot be used to assess the effects of an inflow of immi-
grants that are much different from the existing stock.
24. The number of white males in private household services is so low that the index
cannot be calculated.
25. For example, many high-immigrant-share cities are also major transportation
centers (New York, Los Angeles, Miami). This fact may partially explain the relatively
high share of the transportation services industry in the high-immigrant-share cities.
26. It is interesting to note that total employment growth rates between 1970 and
1980 for the twenty high-immigrant-share cities and the forty low-immigrant-share
cities were virtually identical-the ratio of 1980 to 1970 employment was .92 for the
high-immigrant-share cities and .91 for the low-immigrant-share cities. The relative
growth rates of less-skilled native employment, however, were somewhat different in
the two sets of cities. The relative ratios of 1980 to 1970 employment totals in high-
versus low-immigrant cities were .96 for white males, .90 for white females, 1.02 for
black males, and .87 for black females.
27. A similar approach is used by Borjas (1987).
28. An alternative strategy is to study the effect of immigrant flows to particular
SMSAs that one can identify as exogenous. For example, Card (1990) examines the
effect of the Maria1 boat lift on the Miami labor market and finds little effect on the
wages and unemployment rates of less-skilled blacks and other non-Cuban groups. His
results for wages are somewhat at variance with the instrumental variables estimates
we report below.
29. The instrumental variables estimation of the first-difference equation also uses
these weights.
30. The implied per capita labor supply elasticity is roughly minus one. An alterna-
tive explanation, which might be consistent with an extended version of the model
allowing for heterogeneity within the population of less-skilled natives, is that a down-
ward shift in the wage distribution induced by immigration results in the exit from the
labor force of natives with the lowest skill levels. However, given that the decline in
the employment population ratio is small, a compositional shift cannot explain the
results even if the wages of those who left employment were essentially zero prior to
their departure.
31. In contrast, the correlation between the fraction of immigrants in 1970 and the
change in fraction of all immigrants in the SMSA is .60. These correlations refer to the
unweighted sample of 120 SMSAs.
234 Joseph G. Altonji and David Card
References
Bartel, Ann. 1989. Where do the new U.S. immigrants live? Journal of Labor Eco-
nomics 7 (October):371-91.
Borjas, George. 1985. Assimilation, changes in cohort quality, and earnings of immi-
grants. Journal of Labor Economics 3(0ctober):463-89.
. 1987. Immigrants, minorities, and labor market competition. Industrial and
Labor Relations Review 40(April):382-93.
Card, David. 1990. The impact of the Mariel boatlift on the Miami labor market.
Industrial and Labor Relations Review 43(January):245-57.
Chiswick, Barry. 1982. The impact of immigration on the level and distribution of
economic well-being. In The gateway: US. immigration issues and policies, ed.
Barry Chiswick. Washington, D.C.: American Enterprise Institute.
Filer, Randall. 1988. The impact of immigrant arrivals on migratory patterns of native
workers. Typescript, Department of Economics, Hunter College-CUNY.
General Accounting Office. 1988. Illegal aliens: Influence of illegal workers on wages
and working conditions of legal workers. Washington, D.C.: U.S. Government
Printing Office.
Greenwood, Michael, and John McDowell. 1986. The factor market consequences of
US. immigration. Journal of Economic Literature 24(December): 1738-72.
Grossman, Jean. 1982. The substitutability of natives and immigrants in production.
Review of Economics and Statistics 64(November):596-603.
Hamermesh, Daniel. 1986. The demand for labor in the long run. In Handbook'of
labor economics, ed. Orley Ashenfelter and Richard Layard. Amsterdam: North-
Holland.
Johnson, George. 1980a. The labor market effects of immigration. Industrial and La-
bor Relations Review 33(April):33 1-41.
. 1980b. The theory of labor market intervention. Economica 47(August):
Muller, Thomas, and Thomas Espenshade. 1985. The fourth wave: California's newest
immigrants. Washington, D.C.: Urban Institute Press.
Papademetriou, Demetrios, et al. 1989. The effects of immigration on the U.S. econ-
omy and labor market. U.S. Department of Labor Bureau of International Labor
Affairs Immigration Policy and Research Report no. 1, May.
309-30.
8 Industrial Wage and
Employment Determination
in an Open Economy
Richard B. Freeman and Lawrence F. Katz
The increasing internationalization of the U.S. economy, evinced in the
growth of trade, immigration, and (post- 1982) trade imbalance-induced cap-
ital flows, raises questions about the responsiveness of the labor market to
shocks produced by open economy developments. How do trade-induced
changes in product demand and immigration-induced changes in labor supply
affect relative wages and employment? Do industrial labor markets respond to
shocks generated by international flows of goods and labor as they do to those
generated by domestic developments? Does a decline in demand due to inter-
national trade (and other factors) reduce wages in an industry relative to
those elsewhere? To what extent do wages respond differently in union than in
nonunion settings? To what extent do wages respond differently to increases
as opposed to decreases in relative demand?
To answer these questions, we analyze cross-section time-series data on
imports, exports, immigrant shares of employment, annual and hourly earn-
ings, and employment for detailed U. S. manufacturing industries over the pe-
riod 1958-84 and contrast the responsiveness of the industry earnings in more
and less highly unionized industries and between industries facing greater and
lesser shocks in sales. In contrast to studies that focus on the direct and indi-
rect effects of the trade balance or immigrant flows on the aggregate economy
(using general equilibrium models or input-output analysis), our concern is
with direct trade effects on disaggregated industries. I
The principal finding is that the industry wage structure responds to
Richard B. Freeman is professor of economics at Harvard University and director of the Labor
Studies program at the National Bureau of Economic Research. Lawrence F. Katz is associate
professor of economics at Harvard University and a research associate of the National Bureau of
Economic Research.
The authors are extremely grateful to Dan Kessler for expert research assistance and to Law-
rence Summers for helpful comments.
235
236 Richard B. Freeman and Lawrence F. Katz
changes in product market sales, with trade-induced changes in sales having
approximately the same effect on earnings as sales due to domestic market
developments: on average, a 10% annual change in relative industry revenues
resulting from trade or other factors alters relative earnings by about .5% over
the long run. In addition, we find enough variation in changes in wages, sales
held fixed, to trace out a demand curve “trade-off’ between wages and em-
ployment across industries. Surprisingly, perhaps, we also find that wages
respond more to sales in unionized industries than in nonunionized industries
and more to relative declines in sales than to relative increases in sales. Fi-
nally, industries with growing or large immigrant shares of employment tend to
fall in the industrial wage structure, apparently for reasons beyond any
immigrant-native pay differential within industries.
8.1 Potential Labor Market Responses to Ikade-induced Changes in
Product Demand
When product demand changes in an industry, one expects employment to
change in the same direction, with wage adjustments “buffering” the magni-
tude of job losses or gains. The extent of wage responses to shifts in demand
is likely to depend on the mechanisms for wage setting-in particular, on
whether wages are set in a decentralized fashion in industry labor markets by
supply and demand to clear labor markets; by collective bargaining/adminis-
tered wage setting that produces premium wages in some industries; or on a
national basis with little scope for industry variation.
In competitive decentralized wage setting, the extent of wage response to
shifts in demand or supply of labor depends on elasticities of demand and
supply.* Formally, write the industry demand for labor curve in first-difference
form as
(1)
dE = -kdW + dX,
where E = In employment, W = In wage, X = In shift in the derived
labor demand curve due to shifts in product demand, and d = the difference
operator. Let the industry labor supply curve be represented as
(2)
where S = In factors that shift supply.
dE = edW + dS,
Market clearing produces the following reduced-form relations:
dW = (dX - dS)/(k + e)
(3b)
dE = (edX - kdS)/(k + e)
Since elasticities are likely to be greater in the long run as factor mobility
increases and as firms move in and out of industries, wage responses will be
smaller and employment responses greater to any exogenous shock as time
237 Industrial Wage and Employment Determination in an Open Economy
proceeds. In the extreme, when the elasticity of labor supply to an industry
approaches infinity, there is no wage response to shifts in either schedule, and
the sole change is in employment.
Under collective bargainedladministered decentralized wage setting, wages
in particular industries diverge from market-clearing rates for any of a number
of possible reasons (efficiency wages, rent sharing, collective bargaining),
producing a queue of workers at going rates and an effective infinitehear infi-
nite elasticity of labor supply to the ~ect or . ~ Since neither employers nor
unions are directly constrained by labor supply conditions, there are several
possible wage responses to shifts in demand. Some argue that, in markets
where senior workers have a disproportionate influence on wage setting,
wages are less responsive to trade-induced changes in demand than competi-
tively determined wages. Grossman (1984) analyzes the conditions for this to
be true in a model in which wages are set (subject to a labor demand con-
straint) by a majority-voting union with a seniority layoff rule and free entry
into the union. Modeling an increase in international competition as a exoge-
nous decline in the world price of the product produced by the unionized sec-
tor, he shows that international competition has two offsetting effects on the
wage. For a union of a given size, a lower wage will be desired because
greater international competition increases the risk of layoffs. On the other
hand, declines in union membership will raise the average seniority level,
which, he argues, produces a median member who wants higher wages. The
net effect is ambiguous in general, but, in the case of a constant elasticity labor
demand schedule, his model predicts that the union wage will be completely
unresponsive to the international price.4 Others note the possibility (and exis-
tence in some industries in some periods) of “endgame bargaining” in which
unions, seeing little future to an industry, seek to extract as much as they can
in a short period (Lawrence and Lawrence 1985). Our analysis emphasizes
the possibility of greater-than-competitive downward wage adjustments when
wages exceed outside alternatives in a unionized-administered wage sector, so
that we expect larger downward wage adjustment for union than for nonunion
workers. The view that wage responses to increased product market competi-
tion may be greater under unionism is consistent with research on the effects
of trucking deregulation on wages that finds substantial relative wage reduc-
tions for union truckers and much less wage response for nonunion truckers
following deregulation (Rose 1987).
Consider, finally, industry wage responses under a centralized system in
which wages are set nationally (corporatist economies) or in which there is
considerable “spillover” or “flow on” of changes in wages across industries.
In these settings, we would anticipate shifts in demand to have little effect on
wages but substantial effects on employment. The existence of such wage-
setting systems outside the United States provides potentially fruitful controls
for evaluating the effect of decentralized wage setting in the United States on
wage responsiveness.
238 Richard B. Freeman and Lawrence F. Katz
8.1.1 Modeling Union Behavior
Consider first the wage policy of a union concerned with both wages and
employment that is subject to a labor demand constraint. In simplest form, its
decision making can be viewed as maximizing a utility function U ( W, E)
subject to a labor demand constraint E = E (W). Then the maximizing con-
dition is to set wages so that the ratio of the marginal value of employment to
the marginal value of wages equals the elasticity of labor demand. If trade (or
other factors) increases the elasticity of demand, wages are likely to drop, as
the wage-employment trade-off facing the union is worsened. Huizinga
(1987) shows that, in imperfectly competitive product markets, an increase in
international competition is likely to increase the elasticity of the product de-
mand elasticity facing domestic firms and lead to wage concessions by a mo-
nopoly union. If trade (or other factors) leaves the elasticity unchanged but
shifts the demand curve downward, the union is also likely to lower wages
when demand declines and raise them when demand increases.
Similar implications can be derived for the standard utilitarian union model
(McDonald and Solow 198 1) in which a union with a fixed membership max-
imizes the welfare of the representative member subject to the labor demand
~ons t r ai nt . ~ The union maximizes
where N is the fixed membership, u(. ) is the utility function of the represent-
ative member, e is the level of employment, w* is the opportunity (or alterna-
tive) wage, and w is the wage level. The maximand can be rewritten as
e(w)[u(w) - u(w*)] dropping a constant term. This formulation yields the
familiar optimizing condition in which the elasticity of the gain from employ-
ment is equated to the elasticity of labor demand:
-u’(w)w/[u(w) - u(w*)] = e’(w)w/e.
Here, the union wage depends only on the elasticity of labor demand and the
degree of risk aversion of the representative union member. Changes in inter-
national competition that affect the elasticity of labor demand will affect
wages in the same direction, while shifts in labor demand not changing the
elasticity will not affect wages. In the case of efficient bargains between the
union and firms, the level of labor demand will affect wages as well as
the elasticity.6
When senior workers play an especially important role in the union, there
is a strong possibility of a more complex response pattern. Faced with a posi-
tive shock in demand, existing union members are likely to weigh wage gains
highly relative to employment gains, producing sizable increases in wages.
Faced with modest negative shocks, they are less likely to sacrifice rents to
save the jobs of marginal employees, producing wage inertia. Faced with siz-
239 Industrial Wage and Employment Determination in an Open Economy
able negative shocks and threats of plant shutdowns, on the other hand, exist-
ing unionists may be willing to offer large wage concessions.
As there is reasonable a priori logic for expecting unions to respond less,
more, or even “perversely” to shocks due to trade or other factors, the ques-
tion of which response pattern dominates actual wage setting in the United
States is an empirical one.
8. 1. 2
The models described thus far relate changes in wages and employment to
exogenous shifts in demand and supply. To apply them to data, it is necessary
to measure the exogenous shifts, which, given our focus on trade, requires
that we obtain appropriate indicators of shifts in product demand and the con-
tribution of trade to such shifts. In this study, our primary indicator of shifts
are industry sales and its price and quantity components and sales decom-
posed into domestic market sales, exports, and imports, appropriately
weighted to take account of their relative magnitudes. As sales depend on
industry supply as well as demand conditions, however, simply replacing the
X terms in (3a)-(3b) with sales and regressing wages/employment on sales
does not yield the desired response parameters. There is a potentially impor-
tant simultaneity bias due to the effect of wages on industry prices and output.
As a first cut at the simultaneity problem, assume that the supply curve of
industry output is flat, so that prices depend solely on costs of production.
Then we can model the simultaneous relation between wages and sales with
the following simple market model:
Sales and Shifts in Demand
a) product demand:
(4)
dQ = -hdP + dX,
where Q = In output, P = In price of output, and X = In shifts in
demand, as in (3a);
b) the effect of wages on cost of productiodproduct price:
dP = adW,
where a is a labor’s share of cost;
c) wage determination equation:
dW = qdX,
where q is the parameter of interest to us ( = l/[k + el in eq. [3a]).
Note that this equation makes wage changes depend not on observable
changes in prices or quantities (which are affected by wages) but rather on the
unobserved exogenous shift in product market conditions.
Substituting (5) into (4) yields a relation between output and wages:
(7)
dQ = -hadW + dX,
240 Richard B. Freeman and Lawrence F. Katz
which, in turn, yields a relation between sales (dS = dQ + dP) and wages:
dS = (1 - h)adW + dX.
Solving for dX in (8)' and substituting into (6) yields an equation between
changes in wages and changes in sales?
(9)
dW = {q/[l + q(1 - h ) ~ ] } d S = AdS.
Adding an error term to (9) with the usual properties, we can estimate the
parameter A by least squares regression of observables on observables. We are
interested, however, not in A but in q, the response of wages to changes
in product market conditions. Rearranging terms we see that, for any esti-
mate of A:
(10)
= A/[1 - Aa(1 - h)].
Equation (10) shows us that the estimated parameter of wages on sales
yields the correct response coefficient only if the product demand elasticity is
unity. If h is less than unity, A will understate q, while, if h is greater than
unity, it will overstate q, with the magnitude of the difference between A and
q dependent on the magnitude of Aa(1 - h). For reasonable values of the
parameters, however, it turns out that the difference between A and q will be
small. For example, with the mean value of a (labor's share of cost of sales)
in our data of .25 and estimated values of A below .lo, the bias is modest for
anything short of huge elasticities of product demand.9 In the context of the
model of equations (3a)-(3b), moreover, the difference between the parameter
relating employment to shifts in demand and the regression coefficient of em-
ployment on sales will also be small.'O
The econometrics gets more complicated, however, if, rather than adding
an error term to equation (9), we allow for error terms in each of the underly-
ing equations as well. As we substituted for dX to get (9), the error terms in
the price and wage determination equations become part of the error structure
in (9), with the result that dS is correlated with the error.II In this case, it is
necessary to instrument dS to obtain a consistent estimate of A.I2
As an alternative way of modeling the relation between wages, prices, and
output in an open economy, consider the situation when prices are determined
on world markets so that an industry in a given country can sell as much as it
produces at the going world market price. Here, there is still likely to be a
feedback of wages on sales, as increases in wages increase costs of production
and reduce output, thereby reducing sales. We model this market pattern by
assuming an upward-sloping industry supply curve with a fixed elasticity. Fol-
lowing a logic analogous to that in (4)-(10) above, we can show that the re-
gression of wages on sales leads to an understatement of the parameter of
wage responsiveness of exogenous shifts in market demand, essentially be-
cause the reverse causality is negative.
241 Industrial Wage and Employment Determination in an Open Economy
8.1.3 The Trade Component of Sales
Turning to the effect of changes in trade, we decompose sales into its
component parts-the size of the domestic market (DOM = sales - ex-
ports + imports); exports; and the import share of domestic market sales
(MSHR = imports/DOM)-and use a first-order approximation to obtain
(1 1) dS = w,d In (DOM) + w,d In exports - w,d(MSHR),
where w, = (sales - exports)/sales, wz = exports/sales and, w, =
DOWsales. The weights are obtained by considering the effect of small
changes in domestic-generated revenues, export-generated revenues, and the
import share of revenues on changes in total revenues in a decomposition that
ignores interaction terms, and will, accordingly, be more accurate for small
than for large changes. The purpose of the weighting is to adjust the relevant
changes for the difference in absolute magnitude of sales generated by domes-
tic demand (90% or so of sales) and trade. When there is trade balance in an
industry, the weight on changes in the import share of the domestic market
becomes unity [DOWsales = (sales + imports - exports)/sales = sales/
sales]. When, as in the 1980% imports exceed exports, the weight placed on
this term exceeds unity.
Substituting (1 1) into (9), we obtain a relation between wages (or employ-
ment) and weighted In changes in the domestic and foreign components of
revenues:
dW = Aw, d In (DOM) + Aw,(d In exports)
(12)
- Aw,[d(MSHR)] + other factors.
Note that this model makes a strong implicit assumption about market be-
havior: it postulates that the labor market responds similarly to (weighted)
changes in sales due to trade-related factors as to those due to domestic fac-
tors. While in the short run there may be some differences in market responses
to trade-generated as opposed to domestic market-generated changes due,
say, to differing assessments of whether changes will persist over time (e.g.,
because foreign competition depends on highly volatile exchange rates), in
the long run we see no compelling argument to expect industry labor markets
to react any differently to changes in revenues from different sources: a 10%
shift in demand is a 10% shift in demand. If the assumption that foreign- and
domestic-based changes have the same effects on the labor market is valid,
then the coefficients on the trade and domestic revenue terms will be similar
in regression analysis.
8.1.4 Shifts in Supply and Immigrant Labor
The impact of immigrant labor on industry wage levels is twofold. To the
extent that immigrants are paid differently than otherwise comparable native-
242 Richard B. Freeman and Lawrence F. Katz
born workers, average wages in an industry will depend on the immigrant
share of labor with a coefficient equal to the wage-differential between immi-
grants and native workers. Changes in the immigrant share of labor in an
industry will, accordingly, be associated with changes in industry wages:
(13) dW = bd(IMS),
where IMS = immigrant share of the work force and b = wage differential
between native and immigrant labor.
In addition, however, if immigrant labor is a good substitute for native labor
in immigrant-intensive industries, supply-induced changes in immigrant
shares will alter the wages of natives in the industry as well, further reducing
the position of the industry in the industry wage structure.
8.2 Cross-Industry Analysis for U.S. Manufacturing
We estimate wage-sales and employment-sales equations using the NBER
Trade and Immigration Industry data set for manufacturing production work-
ers.I4 The data set provides information on the wages and employment of
production and all workers, trade flows and immigrant shares of employment,
as well as other control variables in 428 four-digit SIC manufacturing indus-
tries from 1958 to 1984.15 We examine the data in three ways. First, we ana-
lyze changes over the twenty-six-year period from 1958 to 1984, which can
be viewed as reflecting changes in long-run comparative statics for the “aver-
age” industry. Second, we relate annual changes in wages and employment
across industries to changes in revenues and the part of those changes due to
trade and domestic demand, with individual year dummy variables entered in
our regressions to capture economy-wide cyclic-type phenomena. I 6 Third, we
explore responses over different time periods to see whether responses in the
1980s period of large trade imbalances differ from those in earlier periods.
As a check on the results from our establishment-based analysis, we also
estimate wage change equations utilizing industry wage differentials esti-
mated from the 1974 and 1984 Current Population Surveys (CPSs) for the
fifty-eight three-digit 1980 Census industries (CICs) that can accurately be
matched to the 1970 CIC system used in the 1974 CPS and to the NBER trade
and immigration figures.
Our pooled cross-section time-series industry analysis differs, it should be
noted, from the time-series analyses for particular industries that other re-
searchers have used to investigate the effect of trade on the labor market.” We
examine the relative responsiveness of industries to the particular shocks that
face them, exploiting the differential patterns of change among industries
rather than the time-series patterns of change for a particular industry. As a
consequence, our estimated response parameters are average elasticities of
response across industries. Formally, in terms of the model of (12), if each
industry has its own response parameter a + ai, where a is the mean of the
243 Industrial Wage and Employment Determination in an Open Economy
industry response parameters, the form of our basic equation can be writ-
ten as
(14) dW,, = (a + a,)dS,, = ads,, + a@,,,
Where the latter term becomes part of the error structure. This term reflecting
the heterogeneity in industry responses creates heteroskedasticity in the errors
but does not bias estimation of the average response coefficient as long as the
individual industry component of the response (a,) is independent of other
variables in the equation.I8 In section 8.3, we consider potential differences in
response coefficients among industries.
8.2.1 Data Description
Table 8.1 gives the 1984 level of variables, 1958-84 changes in variables,
and standard deviations for variables of concern to us (pt. A) and selected
correlations of the changes for the period 1958-84 (pt. B).
The descriptive statistics reveal several characteristics of industry labor and
product markets that underlie the ensuing econometric results:
The shares of imports and exports relative to the size of the domestic
market are relatively modest even in 1984 after two or so decades of rapid
growth of trade, with imports averaging 14% of domestic demand and
exports averaging 8% of sales. Immigrant shares of labor are also modest,
averaging 8% in 1984.19
The principal dependent variables of concern to us-changes in In annual
earnings (obtained by dividing payroll by employment) and in In hourly
earnings (obtained by dividing payroll by person-hours)-show nearly
identical industry variation over the period 1958-84, indicating virtually
identical patterns of change in hours per employee. The correlation in the
1954-84 change in In annual earnings and change in log hourly earnings
is .95. When, by contrast, we examine short-run year-to-year changes, we
find considerable variation in hours per worker across industries and thus
differences in changes in annual and hourly earnings.
The standard deviation of In changes of industry employment exceeds the
standard deviation of In changes in hourly earnings by a factor of 3.9 (.67/
.17), documenting the fact that quantity adjustments dominate industrial
labor markets, possibly because workers are good substitutes across in-
dustry lines. In the goods market, the standard deviation of In changes in
physical output (deflated sales) exceeds the standard deviation of In
changes in prices by a similar proportion (3.0 = .73/.24).
The standard deviations of our major independent variables-weighted
changes in domestic demand, foreign demand, and the import share-
show considerable interindustry variation, as is necessary if we are to es-
timate their effects on the labor market with any precision.
Turning to the correlations in part B of the table, note first the .42 positive
correlation between changes in the level of imports and changes in In employ-
244 Richard B. Freeman and Lawrence F. Katz
Table 8.1 Descriptive Statistics for NBER 'kade and Immigration Data Set, 428 U.S.
Manufacturing Industries
A. Major Variables
Mean SD
Levels, 1984:
Imports/domestic demand
Exports/sales
ImmigrantslEmployment
Changes, 1958-84:
DLN(hour1y wages for production workers)
DLN(annua1 wages for production workers)
DLN(production employment)
ImmigrantsiEmplo yment
DLN( sales)
DLN(price)
DLN(output)
Weighted changes, 1 95 8-84: *
DLN(domestic demand)
DLN(foreign demand)
Import share
.14
.08
.08
1.43
1.43
- .02
- .OO
1.74
1.06
.67
1.71
.16
.I5
.I6
.10
.05
.I1
.18
.67
.03
.72
.24
.73
.65
.23
.43
B. Correlations, 1958-84: Log Changes
Weighted Changes
Production
Hourly Employ- %I&- Domestic Foreign Import
Wages ment grants Sales Imports Exports Demand Demand Share
Hourly wages 1.00 .04 -.12 .26 -.08 . I 0 .13 .I5 -.23
employment .04 1.00 .06 .90 .42 .44 .83 .23 - . 18
% Immigrantsb - . I 2 .06 1.00 .04 .13 .05 .10 - .04 .06
Production
Note: DLN(.) = change in logarithm of the variable.
a Weights utilized are the average of the 1958 and 1984 weights.
Absolute change in percent immigrant.
ment. If industries with increasing imports expand employment, why is there
such public concern over the effect of imports on jobs? The reason for the
seeming paradox is that imports and domestic production tend to increase in
the same industries-namely, those where domestic demand is growing. This
highlights the need to control for total demand or, alternatively, to focus on
the import share of the domestic market, in estimating the effect of imports on
the labor market. Indeed, the (weighted) change in the import share of domes-
tic sales is negatively correlated with changes in employment at a highly sig-
nificant - .18. Similarly, with respect to earnings, while the correlation be-
tween changes in imports and earnings is a modest - .08, the correlation
between changes in earnings and the (weighted) change in import share is a
245 Industrial Wage and Employment Determination in an Open Economy
hefty - .23. Note finally that changes in the immigrant share of employment
in an industry are negatively related to changes in wages and positively related
to employment changes, indicating that immigrants have gone into industries
with growing employment but declining wages.*O
8.2.2 Basic Regression Results
Table 8.2 presents coefficients and standard errors on the major variables of
concern from our regression analyses of long-term (1958-84) changes in
hourly earnings and personhours worked. All the regressions contain the set
of controls listed at the bottom of the table, including two-digit SIC industry
dummies (to allow for broad industry differences in responses). Columns 1-3
show the estimated effect on hourly earnings of sales, prices, and physical
output taken separately and weighted trade and domestic revenues; columns
4-6 show the estimated effects on hours of sales, price, and quantities and of
weighted trade and domestic revenues. In addition, we report the coefficients
and standard errors on the change and base year level of immigrant shares in
each regression. We do not report the results of comparable regressions for
Table 8.2
Coefficients (standard errors) for Effects of Changes in Sales, Trade
Variables, and Immigration Ratios on Wages and Employment of
Production Workers in U.S. Manufacturing Industries: Long Period
Log Changes (N = 428), 1958-84
Hourly Wages Annual Hours
Variables (1) (2) (3) (4) ( 5) (6)
~
Change in log sales
Change in log output
Change in log price
Change in % immigrant
% Immigrant in 1960
Weighted log change in
domestic demand
Weighted log change in
foreign demand
Weighted change in
import share
RZ
,049
(.011)
,049
(.011)
,108
(.035)
(.286) (.287)
- 1. 164 -1.198
(441) ( 4 1 )
-.445 -.389
.38 .39
- ,432
(.286)
- 1. 164
( ,440)
,040
(.011)
.076
(.034)
- ,064
(.017)
.39
,886
(.017)
,886
(.017)
.884
(.057)
1.082 1.080
(.462) (.466)
,900 ,901
(.713) (.714)
.89 .89
1.002
(.497)
,545
(.765)
,894
(.020)
,710
(.059)
- .479
(.030)
.88
~~
Nore: The reported regressions include two-digit SIC dummies, change in percent union, change
in percent production workers, and the initial (1958) values of the following variables: percent
union, percent production workers, and log of value added per worker. Weights utilized in
weighted log changes are means of 1958 and 1984 weights.
246 Richard B. Freeman and Lawrence F. Katz
annual earnings and production worker employment as they yielded virtually
identical coefficients to those in the table because of the lack of industry vari-
ation in hours per employee over the long run pointed out on page 243.21
There are three principal findings. First, the calculations show that changes
in revenues significantly affect relative wages, implying that the industry wage
structure is “flexible” with respect to changes in the market conditions in par-
ticular industries. Using the reduced-form model of (3a)-(3b) to interpret the
results, the ratio of the estimated effect of sales on hours to the estimated effect
of sales on earnings provides a measure of the magnitude of the elasticity of
labor supply across industries: from columns 4 and 1, it is 18.0, which implies
considerable labor mobility across industry lines in response to changes in
wages.
Second, despite the long period under study, changes in the (weighted) im-
port share of the domestic market and in exports have roughly comparable
effects on‘wages, as do weighted changes in domestic shipments, indicating
that, as a first approximation, the industrial wage structure in the United States
responds to open economy developments to the same extent as to domestic
developments. In addition, the trade-generated and domestic market-gener-
ated changes in sales have significant effects on person-hours, though with
noticeably different estimated coefficients.
Third, the long period change estimates show that both the change and the
initial level of the immigrant ratio are negatively related to changes in hourly
earnings and positively related to changes in employment. The magnitudes of
the coefficients on immigrant shares are, however, too large to be attributed to
pure immigrant-native wage differentials given the small proportion of immi-
grant workers and likely modest differentials correcting for worker skill (Bor-
jas 1985; Chiswick 1978). They are more likely to reflect the concentration
and movement of immigrants into low- and declining-wage industries. That
immigrants find jobs in industries that are increasing employment but falling
in the wage structure is consistent with the basic fact that employment in the
United States has been growing in industries with low and relatively declining
wages.
8.2.3 Annual Changes
Next we examine the effect of domestic- and foreign-generated changes in
sales on earnings using annual rather than long period changes in the vari-
ables. Because annual earnings and hourly earnings differ in the short run,
owing to short-run variations in person hours worked relative to employees,
we report results for both earnings variables. Because our focus is on interin-
dustry responses, we include year dummies in these regressions to control for
general cyclic phenomena.
As can be seen in table 8.3, we obtained somewhat different results be-
tween annual and hourly earnings. First, changes in sales had much larger
effects on annual earnings than on hourly earnings, owing to the implicit re-
247 Industrial Wage and Employment Determination in an Open Economy
Table 8.3 Coefficients (standard errors) for Effects of Changes in Sales and
Wade Variables on Wages in U.S. Manufacturing Industries:
Annual Log Changes (N = 11,165), 1959-84
Annual Earnings Hourly Earnings
Variables (1) (2) (3) (4) ( 5) (6)
Change in log sales ,068
(.004)
Change in log output
Change in log price
Weighted log change in
domestic demand
Weighted log change in
foreign demand
Weighted change in
import share
R2 .22
,069
(.004)
,056
(.007)
,066
(.004)
,071
(.015)
- ,059
(.011)
.22 .22
,027
(.004)
,026
(.@J7)
.032
(.007)
.029
.008
(.016)
- .01 I
(.012)
.19 .20 .20
Note: The reported regressions include year dummies, two-digit SIC dummies, change in percent
union, and change in percent production workers as well as the variables listed above.
sponse of hours per employee to changes in sales (the difference in the effect
of a variable on annual and hourly earnings is its effect on hours per worker)
as firms responded to relative declines in sales by reducing work hours and/or
temporarily laying off some workers and responded to increases in sales by
increasing work hours, including over time. Comparing the effect of sales on
hourly earnings in table 8.3 with the effect of sales on hourly earnings in table
8.2, we find that hourly earnings adjustments are greater in the long run, con-
trary to the purely competitive model of (3a)-(3b). This can be taken as evi-
dence for collectively bargained/administrated wage settlements as opposed
to spot market settlements. Second, and more disturbing to our analysis, table
8.3 shows that, while annual changes in sales due to open economy develop-
ments have substantial and well-defined effects on changes in annual earnings,
they have statistically insignificant effects on changes in hourly earnings, con-
trary to the findings of table 8.2.
To reconcile the findings on the magnitude of the effect of sales and trade-
induced changes in sales in the short run and long run, we made two further
calculations. First, we examined the determination of hourly earnings over
three intermediate periods: 1958-70, which covers the 1960s strong job mar-
ket; 1970-80, when the economy was sluggish; and 1980-84, when the coun-
try developed an extraordinary trade imbalance. The results of this analysis
are given in table 8.4. Consistent with the results for the entire period, they
show substantial and significant effects for import-induced changes in sales on
248 Richard B. Freeman and Lawrence F. Katz
Table 8.4 Long Period Log Hourly Wage Change Regressions by Time Period,
U.S. Manufacturing Industries, 1958-70, 1970-80, 1980-84
Time Period
Variables 1958-70 1970-80 1980-84 1958-70 1970-80 1980-84
Change in log sales .022
(.ow
Weighted log change in
domestic demand
Weighted log change in
foreign demand
Weighted change in
import share
Change in % immigrant - ,263
(. 269)
.035 .086
(.015) (.016)
,017
(.009)
,095
(.039)
- ,141
(.055)
( .262) (.271)
-.325 . . . - .148
.03 1 ,084
(.017) (.018)
,043 ,040
(.033) (.066)
-.076 -.067
(.057) (.018)
- ,342 . . .
( 3 4 )
R2 .27 .26 .20 .28 .26 .19
N 428 428 428 428 428 428
Note: The reported regressions include two-digit SIC dummies, change in percent union, change
in percent production workers, and the initial values of the following variables: percent union,
percent immigrant, percent production workers, and log of value added per worker. Weights
utilized in weighted log changes are mean of initial and final period weights. The 1958-70
regressions do not include change in percent union, and the 1980-84 regressions do not includz
change in percent immigrant because the required data are not available for these variables over
these periods.
hourly earnings but weaker effects for export-induced changes. The period
regressions also show a marked pattern of differences in wage responsiveness
among the periods, with the effect of sales on hourly earnings greatest in the
1980s, as might be expected given the wage concessions of that period.22 Sec-
ond, we have explored the timing of the effect of sales on hourly and annual
earnings by including lagged sales variables in our regressions of annual
changes in hourly earnings on annual changes in sales. The results of these
calculations (see table 8.5) suggest that within three years the effect of
changes in sales on hourly earnings rises to the long-run level and is roughly
equal to the effect of changes in sales on annual earnings.
The differences in timing of the effect of sales on hourly and annual earn-
ings suggest that hours worked (which vary because of both layoffs and over-
time or short time) may be an important indicator to workers of the need to
adjust hourly pay in the face of demand shocks. For our purposes, what mat-
ters is that these calculations show that the long period hourly earnings results
are the valid ones for assessing adjustments beyond a year or so.
8.2.4 Current Population Survey Data
As a check on our findings from establishment data, we have also estimated
the effect of trade and immigration on industry wages using household data
from the CPS tapes. These calculations have the advantage of letting us con-
249 Industrial Wage and Employment Determination in an Open Economy
Table 8.5 Regression coefficients (standard errors) for the Effect of dS on dw
Hourly Earnings Annual Earnings
dS .029 ,069
&(-I) - .001 - ,027
(.ow (.ow
(.ow (.ow
ds( - 2) .017 .005
(.ow (.ow
Sum ,045 . ,047
Nore: The coefficients are based on the same specifications as in table 8.3, with the addition of
the lagged sales variables.
Table 8.6 Long Period CPS Industry Wage Differential Change Regressions,
1974-84: Fifty-Eight Three-Digit 1980 CIC U.S. Manufacturing
Industries
Variables
All Union Nonunion
(3) (4)
Change in log sales
Weighted log change
in domestic demand
Weighted log change
in foreign demand
Weighted change in
import share
Change in %
immigrant
R2
N
,057
(.025)
,010
(.029)
,019
( . 095)
- ,228
(.050)
(.409) (.383)
-.869 -.523
.I31
(.047)
.097
(.055)
-.180
(.181)
(496)
- ,367
- 1.053 - .737
(.751) (.731)
- .ow
(.031)
- ,089
(.034)
,067
(.112)
- ,221
-.612 -.071
(.059)
(.501) (.451)
.21 .39 .18 .31 .09 .35
58 58 58 58 58 58
Note: Reported regressions include change in percent union and change in percent production
workers.
trol for individual characteristics that affect earnings at the cost of limiting the
sample to fewer and more aggregated industries. We proceeded in a two-part
analysis. In step 1, we estimated industry wage effects by regressing the In of
average hourly earnings of individuals on their characteristics and dummy
variables for the In step 2, we regressed the change in the estimated
industry effect on changes in sales, immigrant ratios, and sales decomposed
between trade and domestic factors. The basic results, shown in columns 1
and 2 of table 8.6 (we will discuss the findings in cols. 3-6 shortly), confirm
the table 8.2 finding that, over an extended period of time, changes in reve-
nues due to trade substantially aEect industry hourly earnings and also con-
firm the finding that industries with growing immigrant shares of the labor
250 Richard B. Freeman and Lawrence F. Katz
force fall in the wage structure. Indeed, in these calculations, the dominant
factor in changes in wages in the period 1974-84 is the import part of sales.
8.2.5 The Wage-Employment Trade-off
While our analysis shows that wages respond to changes in sales resulting
from trade and other factors, there is also a significant independent or unex-
plained component to changes in wages as well. Does this component of the
change in wages affect the quantity of labor used, sales held fixed? To what
extent do cross-industry data show a trade-off between wage responsiveness
and the employment of labor? To answer these questions, we have performed
the regression calculations summarized in table 8. 7, in which we relate annual
and long-run changes in In annual production hours to changes in wages and
sales (cols. 1, 4), product wages and deflated output (cols. 2, 5), and wages
and sales decomposed into trade and domestic market determinants (cols. 3,
6), using both annual change and long period change data. The results offer
strong support for the notion that, sales held fixed, there is a significant wage-
employment trade-off across industries in the U. S. labor market. Employment
Table 8.7 The Wage-Employment 'kade-off in U.S. Manufacturing Industries:
Dependent Variable = Change in Log Annual Hours of Production
Workers
Annual Changes, 1959-84 Long Period Changes, 1958-84
Variables (1) (2) (3) (4) ( 5) (6)
Change in log wage
Change in log product
Change in log sales ,670
Change in log output
Weighted log change
in domestic demand
Weighted log change
in foreign demand
Weighted change in
import share
R' .71
N 11,165
- ,628
(.013)
wage
(.om
- ,629
(.013)
- ,545
(.008)
,699
(.005)
.649
(.006)
,676
(.022)
- ,545
(.017)
.72 .69
11,165 11,165
- ,683
(.071)
- .887
(.045)
.921
(.016)
,921
(.016)
.9 1 .91
428 428
- ,612
(.079)
,922
(.018)
,754
(.055)
- ,524
(.028)
.89
428
Nure: The reported annual change regressions include year dummies, two-digit SIC dummies,
change in percent union, and change in percent production workers as well as the variables listed
above. The reported long period change regressions include the same variables except for the
time dummies and in addition include the change in percent immigrant. The wage variable uti-
lized in all the regressions is the hourly wage of production workers. The change in the log
product wage is given by the difference in the change in the log wage and the change in the log
shipments deflator.
251 Industrial Wage and Employment Determination in an Open Economy
growth is lower, sales fixed, by roughly 5%-9% in industries where wages
rise by lo%, relative to other industries. While we recognize that one cannot
interpret the estimated relation as a demand curve, particularly in light of our
analysis of wages as dependent on sales, the inverse relation is nonetheless
impressive. Unfortunately, we lack good measures of shifts in labor supply to
industries and of factors that lead to different union/employer wage-setting
policies to estimate a structural demand equation given our model, in which
wages are endogenous.
8.3 Differences in Wage Responses among Sectors
Are the wage responses found in section 8. 2 the same across all industries,
or do different wage-setting institutions or economic conditions produce dif-
ferent responses to changes in sales? Do the more heavily unionized industries
respond more, or less, to changes in sales due to trade and other factors than
do the less unionized industries? Is there evidence of asymmetric wage re-
sponses to increases and decreases in sales?
8.3.1 Union and Nonunion Responsiveness
To evaluate the effect of trade unionism on the degree of wage responsive-
ness, we have performed two related analyses. First, we estimated earnings
and employment response equations separately for industries whose union
density made them high (upper third), medium (middle third), or low (lower
third) in the distribution of union density as of 1973-75. Second, we added
interaction terms between changes in sales and dummy variables for high,
medium, and low union status to our basic regressions. As the findings for the
two analyses were quite similar, we present for purposes of parsimony the
separate union-density category regression results in table 8. 8. Row 1 of
the table records the estimated effects of sales on In hourly earnings by union
class from annual change regressions. It shows that earnings responses tend
to be higher in the more highly unionized Given that the scope
for union wage responsiveness is likely to be greater the greater the gap be-
tween union and other wages, we take the analysis a step further in rows 2-5,
by estimating a single change in In hourly earnings regression with more com-
plex interaction terms that distinguish not only between high, medium, and
low union density but also between industries with high (upper third), me-
dium (middle third), and low (lower third) hourly earnings in 1958. These
calculations show that it is the responsiveness of union wages to changes in
sales in high-wage industries that underlies the greater elasticity of union
wages to sales in row 1. To the extent that wages are high in these industries
because of large union wage effects, this finding supports the notion that wage
adjustments are greater where wages exceed competitive market levels. Fi-
nally, we note that regressions comparable to those in table 8. 8 with employ-
ment as the dependent variable show that the pattern of change in employment
252 Richard B. Freeman and Lawrence F. Katz
Table 8.8 Wage Responsiveness to Changes in Sales by Union Density: Annual
Log Wage Change Regressions, Hourly Wages, 1959-84
Wage Class
Union Density
High Medium Low
1. All industries ,041
(.ow
2. High initial wage ,088
(.025)
3. Medium initial wage .035
(.026)
4. Low initial wage ,027
(.022)
,021 .019
(.007) (.007)
,066 - ,010
(.019) (.013)
,003 ,032
(.022) (.014)
.023 ,031
(.012)
Note: The union density and initial wage classes are derived by dividing the industries into thirds
on the basis of initial union density and initial hourly wage. Row 1 presents the coefficients
(standard errors) on change in long sales for separate log wage change regressions by union
density class, which also include change in percent production workers, change in percent union,
and time dummies. The regression used for rows 2, 3, and 4 included the same controls plus
initial wage class and union class dummies, two-digit SIC dummies, and a full set of interactions
of the union class and wage class dummies with change in log sales. Numbers in parentheses in
rows 2, 3, and 4 are standard errors for difference between reported coefficient and the coefficient
on the base group, low union, low initial wage.
responsiveness is opposite that for wage responsiveness, with employment
responsiveness declining with union density, as would be expected if the wage
adjustments serve to “buffer” employment in the market. 25
Because the establishment-based data set does not permit us to differentiate
between union and nonunion firms or workers within an industry, it is possible
that the differences found in table 8.8 are not due to genuine differences in
behavior between union and nonunion firms.26 Accordingly, we have also es-
timated response parameters for union and nonunion parts of industries using
our CPS data set, where it is possible to distinguish between union and non-
union workers. Here, we estimated industry wage effects by regressing In
wages on worker characteristics and industry dummy variables for union and
nonunion workers taken separately and then regressed the change in the indus-
try effects on changes in industry sales, immigrant ratios, etc. These results,
given in the regression coefficients in columns 3-6 of table 8.6, support the
finding that unionization increases rather than reduces wage responsiveness:
the industry differentials for unionized workers are significantly influenced by
changes in sales, and the import component of changes in sales has an espe-
cially large effect in the union sector. By contrast, there is no noticeable effect
of changes in sales or the import component of those changes on the wages of
nonunion workers.
8.3.2 Responses of Industries at the Extremes
The finding that wages as well as employment respond to open market
shocks does not mean that those responses are a major element in industrial
253 Industrial Wage and Employment Determination in an Open Economy
wage and employment determination. Since the bulk of revenues are gener-
ated in domestic markets and most workers are native born, changes in trade
and immigrant flows cannot possibly be a dominant force in altering the in-
dustrial wage structure or the composition of employment. Still, the wage
responses to trade flows can have nonnegligible effects on wages and employ-
ment particularly at the extremes. A one standard deviation change in the
import share (.43 by table 8.1), for example, induces a .028 change in wages
and a .21 change in industry personhours, according to the coefficients in table
8.2. More strikingly, industries that faced massive changes in sales for either
trade or domestic market reasons experienced large changes in wages as well
as in employment. Figure 8.1 documents this by contrasting wage and em-
ployment changes from 1958 to 1984 between the ten industries experiencing
the greatest positive and negative weighted changes in import shares, exports,
and domestic market sales and changes in In total sales. The figure shows a
wide spread in wage responsiveness between the extremes. For imports, the
industries with the greatest increase in import share had wage increases some
nineteen In points below the average for all industries and some thirty-three In
points below wage increases for the industries with the largest decrease in
import shares. Industries with the most/least rapid growth of domestic market
sales show a smaller though still pronounced range of variation, while indus-
tries with the most/least rapid growth of exports show the least pronounced
range. As for employment, the figure shows declines of .81 In points in indus-
tries with the most rapid rises in import shares, but it also shows above aver-
age declines in employment in industries where import shares fell-a seeming
paradox that is due to the fact that import shares dropped most in industries
with falling domestic market sales.27 By contrast, the figure shows a mono-
tonic relation between extreme changes in exports and in domestic market
sales and employment: here, employment rises at rates far above average in
the ten industries where exports or domestic sales increase most while falling
at rates far below average in the ten industries where exports or domestic sales
increase the least. Finally, putting all the components of change in sales to-
gether, we see sizable differences in changes in wages as well as in employ-
ment between industries experiencing the extremes of the change in sales.
8.3.3 Asymmetric Responses to Changes in Sales
An important issue in decentralized labor markets where wages are flexible
to industry conditions is whether wage responsiveness is symmetrical to de-
clines in demand and increases in demand. In an economy in which wages are
above the reservation wages of unemployed workers, greater responsiveness
of wages to declines in demand than to increases in demand can increase em-
ployment.28 To examine the symmetry of response, we have divided the
change in the In sales variable into two parts: changes in excess of the mean
change and changes below the mean change. We then regressed changes in In
hourly wages over the twenty-six-year period 1958-84 on changes in sales
interacted with a dummy for above average changes and a dummy for below
254 Richard B. Freeman and Lawrence F. Katz
-
-
....
.....
....
- ....'
.' .*.*.*.
..... -
Low
-'I3' ' Sales ' 'Domestic' ' Foreign ' ' Import '
Demand Demand Share
-0.15
1
0,4
- 02
-0,8
- 1,4
- 2
....
..... ....
..... ....
.....
....
Low
-
-
High
'.:.:.:.'
....
7
Deviations from average LN change in production employment
Low
Demand Demand Share
Fig. 8.1
changes in each category relative to the overall mean change-1958-84
Mean changes for the ten industries with the highest and lowest
255 Industrial Wage and Employment Determination in an Open Economy
average changes. The coefficients on the two sale interaction terms (standard
errors) given below show greater responses to the below-mean than to above-
mean changes on changes in In hourly earnings:
effect of below-mean changes: .080 (.017);
effect of above-mean changes: .011 (.020).
This finding suggests that relative wage flexibility in U.S. manufacturing
has taken the form of “concessions” in industries doing more poorly to a
greater extent than it has of large relative wage gains in industries doing better
and thus may have contributed to the job growth in the country. We did not
find any evidence that industries paid exceptionally large wage gains in boom-
ing markets: most of the adjustment is on the down side.
8.4 Conclusion
This paper has documented with two different data sets that, in detailed
manufacturing industries in the United States, wages respond significantly to
changes in industry sales, whether generated by domestic market or trade-
related developments, and found that changes in immigrant shares are also
related to industry wages. It has also found that the wage-setting institutions
in the labor market condition wage responsiveness, with unionized high-wage
industries showing the greatest response to changes in sales. While far from
the dominant force in altering the industrial wage structure, shifts in product
demand due to changes in sales from trade as well as domestic market devel-
opments have contributed to changes in earnings by industry. As the observed
inverse relation between changes in the immigrant share of the work force and
changes in industry wages cannot be readily explained by compositional fac-
tors, this finding evidently merits more detailed analysis of the market for
immigrant labor.
Notes
1. Dickens et al. (1985) provide a detailed review of aggregate studies of the em-
ployment effect of trade in the United States. Lawrence (1984) provides a good ex-
ample of a study utilizing an input-output framework to analyze the effect of trade on
employment in U.S. manufacturing.
2. We define competitive decentralized wage setting to be a system in which wages
are set to equate labor supply and labor demand, as in a textbook perfectly competitive
labor market. We defer until later in this section a discussion of the possibility that
efficiency wage considerations may yield above market-clearing wages even with de-
centralized wage setting in competitive labor markets.
3. Katz (1986) surveys efficiency wage and rent-sharing models of wage determi-
nation and discusses some of the implications of these models for interindustry wage
differences. Dickens and Katz (1987), Katz and Summers (1989), Krueger and Sum-
256 Richard B. Freeman and Lawrence F. Katz
mers (1988), and Murphy and Topel (1987) provide evidence on the role of efficiency
wage, rent-sharing, and standard competitive factors in explaining interindustry wage
differentials. Lewis (1986) and Freeman and Medoff (1984) survey the vast literature
on the effect of collective bargaining on relative wages.
4. Grossman assumes that the unionized sector faces an infinitely elastic product
demand schedule at the exogenous international price. Labor demand elasticity de-
pends only on the production technology and factor substitution possibilities. Interna-
tional competition is not viewed as changing product market structure and/or the elas-
ticity of product demand.
5. A key assumption implicitly made in this model is that the membership is greater
than the employment level that would prevail at the competitive wage rate w*.
6. In the case of Nash bargains with a utilitarian union and strongy efficient bar-
gains, the wage is the mean of the average product of labor and the opportunity wage
(McDonald and Solow 1981). In this case, a decrease in rents from increased interna-
tional competition will lead to wage decreases in the union sector.
7. We can also substitute for dQ using (7) to obtain a relation between wages and
output, dW = [ q/ ( l + haq)]dQ.
8. While we have data on prices and quantities, the likelihood that price indices are
inaccurate leads us to focus on sales. In empirical work, however, we also examine the
separate effect of measured quantities and prices.
9. Consider the value of q when A is . I0 and n is .25. By eq. ( l l ) , we have
q = .10/[1 - . l o . .25 . (1 - h) = .10/[1 - .025(1 - h)]. For the minimum
value of h of zero, q differs from A by a bare 2.5%. For a high elasticity of, say, five,
the difference is less than 10%.
10. To see this, we substitute dS - (1 - h)ndW for dX in eq. (3b) and then substi-
tute (1 1) for dW. This yields the following:
dE = {v/[l + q(1 - h)a]}dS,
where v is the desired parameter.
11. Rewriting eq. (5) and (6) as dP = udW + u and dW = qdx + v, where u
and v are error terms, one can easily derive a relation between wages and sales analo-
gous to eq. (9):
dW = AdS + [V - q(1 - h)u]/[l + q(1 - h ) ~ ] .
12. Revenga (1989) uses an instrumental variables technique with the change in the
(import-source-weighted) industry exchange rate used as an instrument for dS to reana-
lyze the data set that we examine in sec. 8.2. Her findings are qualitatively quite simi-
lar to those we report in sec. 8.2 and suggest that our least squares estimates may
understate the response of industry wages to changes in industry sales.
13. This decomposition of change in log sales can be derived by writing sales as
S = [(S - x)/DOM]DOM + x, where x equals exports. The application of the dif-
ference operator to this decomposition of sales yields
dS = [S - x)/DOM]d(DOM)
- DOM . [d(MSHR)] - d(D0M) . d(MSHR) + dx.
Equation ( I 1) in the text can then be derived from the above expression by dividing
through by S to yield an expression in percentage changes, approximating percentage
changes as In changes, and dropping the interaction term. This approximation is almost
exact for annual changes. We have experimented with the exact decomposition using
percentage changes and including the interaction term in several of our specifications
and have found results in all cases quite similar to those obtained with our In change
approximation.
257 Industrial Wage and Employment Determination in an Open Economy
14. We have also examined all workers and nonproduction workers. As production
workers are the majority of workers, the results for all workers are quite similar to
those for production workers. For nonproduction workers, there are some modest dif-
ferences, but nothing substantial enough to change the tone of the findings.
15. Abowd (in this volume) provides detailed descriptions of the data set and its
construction.
16. We have also estimated annual change equations replacing the time dummies
with observed cyclical variables, such as the aggregate unemployment rate, and al-
lowed different industries to have different cyclical sensitivities. The results are quite
similar to the reported estimates based on equations with time dummies.
17. Branson and Love (1988) analyze the effect of time-series changes in the real
exchange rate on employment in U.S. manufacturing industries. Eichengreen (1988)
performs a similar time-series analysis of the effect of the real exchange rate on em-
ployment in four U.S. basic industries with the industry-specific wage treated as ex-
ogenous and included as an explanatory variable. Grossman (1987) takes changes in
the price of import substitutes as exogenous and analyzes the effect of time-series
changes in import prices on wages and employment in nine trade-affected U.S. manu-
facturing industries. The lack of import price data for a large number of industries at
the level of aggregation of available wage and employment data prevents us from fol-
lowing a similar strategy. Grossman (1986) also uses a similar methodology to analyze
the effect of international competition on employment in the steel industry but treats
the wage as exogenous.
18. We have computed White (1980) heteroskedasticity-consistent standard errors
for several of our specifications. The White standard errors typically differ from the
reported standard errors by less than 5%.
19. The immigrant share data for 1984 are based 1980 Census of Population data.
We utilize 1960, 1970, and 1980 Census of Population data on immigrant shares of
population in the ensuing empirical analysis.
20. The change in immigrant share from 1960 to 1980 has a correlation of - .25
with the initial (1958) industry In hourly wage level, and the initial (1960) immigrant
share has a correlation of - .35 with the change in the In hourly wages over the period
1958-84. Immigrants have moved into low-wage industries, and initially immigrant
intensive industries have experienced relatively low wage growth.
21. For example, the estimated effect of sales on annual earnings was ,045 with a
standard error of .011, while the effect of sales on employment was ,890 with a stan-
dard error of ,018.
22. We have also analyzed annual changes within the periods and found the same
result: greater responses of hourly earnings to sales in the 1980s than in earlier periods.
23. These regressions utilized samples from the May 1974 CPS and the full-year
1984 CPS for workers in the fifty-eight three-digit CIC manufacturing industries with
consistent industry classifications over this period. The controls included in the In earn-
ings regressions in addition to industry dummies were education and education
squared; experience and experience squared; nonwhite, female, SMSA, region, part
time, marital status, married times female, and occupation dummies; and interactions
of the education and experience variables with the female dummy. The regressions for
all workers (both union and nonunion) included a union dummy. Industry differentials
for all workers from regressions without a union dummy yield quite similar results.
24. The same pattern of larger wage responses to sales of the high-union density
class is also apparent when long changes for 1958 to 1984 are analyzed rather than
annual changes.
25. In separate annual change regressions by union category for 1959-84 of
changes in In annual production worker hours on changes in In sales and our usual set
258 Richard B. Freeman and Lawrence F. Katz
of controls, the coefficients (standard errors) on change in In sales were .63 (.01) for
high union density, .61 (.01) for medium union density, and .71 (.01) for low union
density.
26. For a discussion of the potential pitfalls in interpreting estimated coefficients on
union density in industry regressions, see Lewis (1986).
27. The industries with the lowest growth in import shares also experienced well
above average wage growth, suggesting possible endgame behavior of the type mod-
eled by Lawrence and Lawrence (1985). Still, fig. 8.1 does show that the industries
with the biggest declines in domestic market size had below average wage increases,
indicating that endgame is far from the norm for declining industries.
28. This assumes no difference in the size of sectors with increasing/decreasing
demand.
References
Borjas, George J. 1985. Assimilation, Changes in Cohort Quality, and the Earnings of
Immigrants. Journal of Labor Economics 3 (October): 463-89.
Branson, William H., and James P. Love. 1988. U.S. Manufacturing and the Real
Exchange Rate. Misalignment of Exchange Rates: Effects on Trade and Industry, ed.
Richard C. Marston. Chicago: University of Chicago Press.
Chiswick, Barry R. 1978. The Effect of Americanization on the Eamings of Foreign-
born Men. Journal of Political Economy 86 (October): 897-921.
Dickens, William T., and Lawrence F. Katz. 1987 Inter-industry Wage Differences
and Theories of Wage Determination. NBER Working Paper no. 2271. Cambridge,
Mass.: National Bureau of Economic Research, June.
Dickens, William T., Phillip Shapira, Laura Tyson, and John Zysman. 1985. The Em-
ployment Effects of International Trade: A Review of the Literature. University of
California, Berkeley, February. Mimeo.
Eichengreen, Barry. 1988. International Competition in the Products of U.S. Basic
Industries. In The United States in the World Economy, ed. Martin Feldstein. Chi-
cago: University of Chicago Press.
Freeman, Richard B., and James L. Medoff. 1984. What Do Unions Do? New York:
Basic.
Grossman, Gene M. 1984. International Competition and the Unionized Sector. Ca-
nadian Journal of Economics 17 (August): 541-56.
. 1986. Imports as a Cause of Injury: The Case of the U.S. Steel Industry.
Journal of International Economics 20: 201-23.
. 1987. The Employment and Wage Effects on Import Competition in the
United States. Journal of International Economic Integration 2: 1-23.
Huizinga, Harry. 1987. Union Wage Bargaining in the International Economy. Har-
vard University, April. Mimeo.
Katz, Lawrence F. 1986. Efficiency Wage Theories: A Partial Evaluation. In NBER
Macroeconomics Annual 1986, vol. 1, ed. S. Fischer, 235-75. Cambridge, Mass.:
MIT Press.
Katz, Lawrence F., and Lawrence H. Summers. 1989. Industry Rents: Evidence and
Implications. Brookings Papers on Economic Activity: Microeconomics, 209-75.
Krueger, Alan B., and Lawrence H. Summers. 1988. Efficiency Wages and the Inter-
industry Wage Structure. Econometrica 56 (March): 259-93.
Lawrence, Colin, and Robert Z. Lawrence. 1985. Relative Wages in U.S. Manufactur-
259 Industrial Wage and Employment Determination in an Open Economy
ing: An Endgame Interpretation. Brookings Papers on Economic Activity, no. 1: 47-
106.
Lawrence, Robert 2. 1984. Can American Compete? Washington, D.C.: Brookings.
Lewis, H. G. 1986. Union Relative Wage Effects: A Survey. Chicago: University of
Chicago Press.
McDonald, Ian, and Robert Solow. 198 1. Wage Bargaining and Employment. Ameri-
can Economic Review 7 1 (December): 896-908.
Murphy, Kevin M., and Robert Topel. 1987. Unemployment, Risk, and Earnings. In
Unemployment and the Structure of Labor Markets, ed. K. Lang and J. Leonard.
London: Blackwell.
Revenga, Ana L. 1989. Wage Determination in an Open Economy: International Trade
and U.S. Manufacturing Wages. Harvard University, December. Mimeo.
Rose, Nancy L. 1987. Labor Rent-Sharing and Regulation: Evidence from the Truck-
ing Industry. Journal of Political Economy 95 (December): 1146-78.
White, Halbert. 1980. A Heteroskedasticity-consistent Covariance Matrix Estimator
and a Direct Test for Heteroskedasticity. Econometrica 48 (July): 8 17-38.
This Page Intentionally Left Blank
9 Foreign-Owned Businesses in
the United States
Jonathan S. Leonard and Rachel McCulloch
For the United States, concern about foreign direct investment (FDI) has his-
torically centered on the costs and benefits to the nation from the establish-
ment of subsidiaries abroad by U.S. multinational firms. Since the mid-l970s,
however, the United States has emerged as the world’s leading host to inward
direct investment. Along with record purchases of U.S. securities, individual
and institutional investors across the globe have purchased U.S. farmland,
department stores, and luxury hotels. Foreign manufacturing firms have ex-
panded U.S. distribution and service facilities and local production capacity.
Japanese and European banks have opened branch offices in New York, Chi-
cago, and Los Angeles. By 1984, the total value of foreign direct investments
in the United States amounted to about 10 percent of the value of all New York
Stock Exchange stocks. Moreover, these holdings had increased at a dramatic
rate, from $35 billion in 1977 to $160 billion (nominal) in 1984.
This flood of inward foreign direct investment represents a dramatic shift
from the established pattern of the earlier postwar period. Until the 1970s,
FDI globally was dominated by the outward thrusts of U.S. firms: the multi-
national corporations doing the investing were viewed by many writers as a
peculiarly American phenomenon. But now foreign direct investment appears
to be one more area in which the nation’s industrial competitors have been
catching up to the United States. After years of relative stability, the ratio of
inward to outward direct investments rose from less than one-quarter in 1977
to more than three-quarters by 1985 (Lipsey 1987).
Jonathan S. Leonard is associate professor in the Organizational Behavior and Industrial Rela-
tions Group of the Haas School of Business Administration, University of California, Berkeley, a
research associate of the Institute of Industrial Relations at Berkeley, and a faculty research fellow
of the National Bureau of Economic Research. Rachel McCulloch is Rosen Family Professor of
Economics and director of the Lemberg Program in International Economics and Finance at Bran-
deis University, and a research associate of the National Bureau of Economic Research.
261
262 Jonathan S. Leonard and Rachel McCulloch
As the foreign presence grows, U.S. policy concerns shift accordingly. Pol-
icymakers in the United States once focused primarily on the effects of direct
investments abroad by U.S.-based corporations. The central issue in that pol-
icy debate was the relationship of outward investments to U.S. trade and do-
mestic employment. Researchers assessed the degree of complementarity or
substitution between U.S. exports and host-country production by U. S. sub-
sidiaries and, to a lesser extent, between U.S. production for domestic mar-
kets and imports from subsidiaries abroad (e.g., Musgrave 1975; Dewald et
al., 1978; Bergsten, Horst, and Moran 1978, chaps. 3, 4).
From a theoretical perspective, the key question was whether U.S. invest-
ments abroad were “defensive”-necessary to secure markets that would oth-
erwise be lost to foreign rivals. Sympathetic observers like Vernon (1971)
inferred from case studies that investments abroad by U.S. multinationals
were largely defensive, but most American labor unions and some academic
researchers (e.g., Frank and Freeman 1978) took a less optimistic view. Em-
pirical testing was complicated by the product-cycle character of most U.S.
direct investments abroad; the industries and firms with above-average pro-
pensities to invest abroad were also those with above-average propensities to
export.
Today, while the potential loss of U.S. jobs associated with establishment
of foreign subsidiaries by U.S. firms remains an important policy concern,
attention has moved to issues raised by inward direct investment-issues that
are novel from the U.S. perspective but widely discussed during the postwar
period in other major host countries, both industrialized and developing. The
fundamental concern is the extent to which the investing firm, rather than the
host country, derives the lion’s share of economic benefits from controlled
local production.
The distinguishing feature of foreign direct investment is an ownership
stake sufficient to permit a local management role for the investor.’ In dollar
terms, direct investment inflows from abroad remain small relative to the na-
tion’s total foreign borrowings. However, this measure may understate the
potential economic, political, and social impact. By definition, direct invest-
ments represent the extension of foreign firms’ managerial control into the
U.S. economy, just as U.S. direct investments abroad have allowed American
firms to control significant parts of the domestic economies of other nations.2
The recent pattern is controversial in the United States for much the same
reasons that the former one has been controversial in host nations abroad.
What accounts for the rapid increase in foreign direct investments in the
United States? Is the flood of capital from abroad linked to U.S. trade prob-
lems? Should the United States welcome foreign investors, or are there rea-
sons to limit the sale of U.S. assets?
This paper considers the theoretical explanations for the growing foreign
direct investment in the United States and explores some of the empirical reg-
ularities associated with foreign-owned companies. Section 9.1 discusses the
263 Foreign-Owned Businesses in the United States
many motivations underlying FDI in general and in the United States specifi-
cally. It provides a synthesis of theoretical explanations for the effects of these
investments on the domestic labor market. Section 9.2 presents our data
sources and methods. Section 9.3 analyzes the differences and similarities
between U.S. and foreign-owned businesses in the United States. Section 9.4
offers some conclusions from the analysis.
9.1 The Motivations for Foreign Direct Investment
9.1.1 Foreign Investment and Internationalization
Once largely insulated from developments abroad by its size and distance
from other industrial powers, the United States has in recent years been drawn
into increased economic intimacy with other nations. Foreign direct invest-
ments, first outward and more recently inward, have played a central role in
establishing these linkages, bringing production and sales of enterprises in the
United States and other nations under the control of a single management.
The internationalization of the U. S. economy can be measured in a number
of ways. The most obvious is that U.S. markets for goods and services are far
more open than in the past. Almost every manufacturing industry has experi-
enced a dramatic rise in the ratio of imports to domestic production-perhaps
no surprise at a time when the nation has run trade deficits of record propor-
tions. Less well known is that almost every U.S. manufacturing industry has
also experienced a rise, albeit not as dramatic, in its exports. The same is true
for agriculture, for the extractive industries, and for many of the service activ-
ities that now dominate U.S. employment.
Even more striking than the increased flow of goods across U.S. borders is
the growth in the volume and variety of asset transactions with other nation^.^
U.S. investors have long dominated international financial markets as lenders.
In the 1980s, however, the United States became a major international bor-
rower. Indeed, as a result of increased foreign borrowing and reduced foreign
lending, the nation is today billed as the world’s leading debtor nation-an
unfamiliar role, and one that many Americans find troubling. While the na-
tion’s overall dependence on capital from abroad is itself womsome, the pri-
mary focus of concern is direct investment and the associated control by for-
eign enterprises over U. S. productive activity.
The increased trade in financial assets reflects several independent devel-
opments. The United States and most other industrial nations have greatly
reduced legal barriers to both inward and outward financial flows, part of a
more general trend toward deregulation of financial markets. The recycling of
large petrodollar surpluses in the 1970s contributed to the growth of institu-
tions capable of handling huge international capital flows. Revolutionary
changes in global communications have facilitated the integration of national
financial markets into a single worldwide network of lenders and borrowers.
264 Jonathan S. Leonard and Rachel McCulloch
Together with expanded options for international communication, reduced
costs of transporting goods and people have promoted a shift by many firms
from national to global management of innovation, production, and distribu-
tion.
Although the recent cries of alarm suggest otherwise, U.S. borrowings
abroad are themselves nothing new. The United States was a borrower and net
debtor for most of the period between the nation’s founding and World War I.
What i s new is the emergence of the United States as the major host country
for inward foreign direct investments by multinational corporations based in
other countries. From 1961 through 1967, less than 3 percent of the world’s
flow of new FDI came to the United States. That fraction rose gradually dur-
ing the 1970s and peaked in 1981, when the United States attracted nearly
half of total direct investment inflows worldwide and two-thirds of total in-
flows going to developed countries (Lipsey 1987). Around 1980, the United
States displaced Canada as the world’s leading host to foreign subsidiaries.
The growth in U.S. trade and the growth in foreign direct investment have
been linked developments. Transactions between multinational firms and their
foreign subsidiaries have accounted for a major share of the overall rise in the
volume of U.S. trade.4 Coordinated by a single global management, trade in
intermediate inputs as well as completed goods allows international compara-
tive advantage to operate not only at the level of individual products but also
in determining the location of different steps in the production of a single
product.
As of 1985, about 15 percent of all foreign direct investments in the United
States (and nearly two-thirds of all Japanese investments) were in the whole-
sale trade sector. The main task of such investments is to promote the parent’s
imports to or exports from the United States. Even when local production by
a manufacturing subsidiary replaces goods previously imported, as has oc-
curred in autos and electronics, the resulting fall in imports of finished goods
is typically offset by an associated increase in imports of component^.^
9.1.2 Why Firms Go Abroad
To understand why foreign firms have been increasing their ownership
stakes in the U.S. economy, it is useful to review the basic conditions required
for profitable foreign direct investment. These conditions provide a context
for identifying specific changes in the global environment that may underlie
the recent surge in establishment of U.S. subsidiaries by firms abroad.
In one sense, the central puzzle concerning FDI is why it takes place at all,
given the significant competitive disadvantages faced by a firm entering a for-
eign market. To make the strategy viable, the investing firm must possess an
“advantage” in terms of product, process, or management sufficient to out-
weigh its obvious disadvantages relative to actual or potential domestic com-
petitors in the host country. The existence of additional hurdles and risks fac-
265 Foreign-Owned Businesses in the United States
ing a foreign entrant suggests that foreign subsidiaries should on average yield
higher profits than domestic operations in the same industry. These conditions
imply that direct investments will be concentrated in markets that do not con-
form to the paradigm of perfect competition-markets characterized by in-
complete information, barriers to entry, or other “imperfections.”
A rent-generating competitive advantage is still only a necessary condition
for the viability of foreign direct investment. Since the firm’s competitive ad-
vantage could in some circumstances be better exploited by exporting from
the home country, an additional requirement for setting up foreign operations
is a locational advantage. This could reflect the usual considerations of pro-
duction and transport costs as well as advantages arising from national policy
at home and abroad, for example, taxation, regulation, and barriers to trade.
In the absence of a significant locational advantage, the potential investor is
likely to choose exporting over the more costly and risky option of establish-
ing a local subsidiary.
Like the competitive advantage, the locational advantage is necessary but
not sufficient. Even with a locational advantage, there must be an organiza-
tional advantage of FDI over alternative strategies such as licensing or other
long-term contractual arrangements with firms in the host country. In the lan-
guage of industrial organization, there must be an internalization advantage-
an advantage to substituting internal modes of resource coordination within a
single firm for an external market-based arrangement between independent
firms. In other words, there must be an advantage of integrated global man-
agement .’
9.1.3 Internalization and Internationalization
Foreign direct investment is precisely a firm’s internalization of economic
activity across a national boundary-internalization of management. The
underlying motives are essentially the same ones that promote expansion of a
firm’s activities within a single domestic market, but with a larger anticipated
benefit required to offset the larger costs of international expansion.
This perspective is supported by the empirical finding that firms investing
abroad are on average larger in their domestic operations than other firms in
the same industry. Multinational firms based in small countries are also typi-
cally smaller in absolute size than their counterparts from large countries.
Both observations are consistent with the hypothesis that firms should exhaust
the gains from domestic internalization before going abroad.
Since the investments abroad of U.S. multinationals dominated the global
picture for several decades after World War 11, most of the empirical research
has focused on these. Studies of I31 by U.S. manufacturing firms have veri-
fied that investment activity is clustered in the industries where research and
development and advertising expenditures are important. Such expenditures
presumably create the competitive advantage necessary for a U.S. firm to op-
erate profitably in a foreign environment. Competitive advantages interact
266 Jonathan S. Leonard and Rachel McCulloch
with potential gains from internalization. For example, high-technology firms
tend to exploit their newest technologies via subsidiaries, while older prod-
ucts and processes are licensed to independent foreign producers.
Evidence on locational advantage is less compelling, except for resource-
based industries. Predictable factors such as availability of suitable labor at
lower cost, a large and protected domestic market, favorable tax or regulatory
treatment, and stable political environment all appear to have some influence
on location decisions.
Other locational considerations cannot be separated from the benefits of a
single global management structure (internalization). Some FDI is primarily
for the purpose of enhancing local sales of goods imported from a firm’s pro-
duction facilities elsewhere, as with distribution and service facilities.* Local
operations can also enhance exports by the parent to the host market by pro-
viding the parent with up-to-date market information when conditions are
changing rapidly, thus keeping the parent in close touch with market trends.
In a concentrated industry, the establishment of local production capacity may
be central to the investing firm’s competitive strategy (e.g., Graham 1978).
The subsidiary represents the firm’s precommitment to a substantial presence
in the local market.
Other advantages of multinational activity are associated with being multi-
national rather than with any specific host location. A global production net-
work permits the firm to diversify risk and, more generally, increases its op-
tions when conditions are volatile (e.g., Kogut 1983). The risk-handling
motive may be relevant in explaining not only investments in the post-Bretton-
Woods era of volatile exchange rates but also the classic foreign direct invest-
ments in extractive industries. Enhanced opportunities for tax avoidance are
another widely cited benefit of multinational operations. The modest reported
financial success (and correspondingly low tax burdens) of many foreign-
owned U.S. plants may reflect such accounting manipulation.
9.1.4 Exchange Rates and Direct Investment
Other things equal, a lower dollar makes U. S. products a better buy in
world markets. Should the same hold for assets? If a U. S. asset is seen as a
claim to a future stream of dollar-denominated profits, and if profits will be
converted back into the domestic currency of the investor at the same ex-
change rate, the level of the exchange rate does not affect the present dis-
counted value of the investment. Thus, dollar undervaluation (or overvalua-
tion) is irrelevant unless a major motive for the investment is speculation on
future movements in exchange rates. Speculative motives may influence port-
folio investments but are unlikely to play an important role in direct invest-
ments, where the planning horizon usually extends over years or even dec-
ades.
A more relevant consideration is that a weak dollar makes the United States
more attractive as a production site. By lowering U.S. production costs rela-
267 Foreign-Owned Businesses in the United States
tive to those in Europe or Japan, a fall in the dollar might shift locational
preference for direct investors toward the United States. Even so, advantages
of internalization would be required to make direct investment a profitable
response to the new currency values. In the absence of such advantages, for-
eign firms would be unable to compete with U. S. firms in exploiting the lo-
cational advantage of lower production costs.
A more basic problem with attributing investment flows to exchange-rate
levels is that the post-Bretton-Woods regime of generalized floating has been
characterized by large swings in key rates, from apparent undervaluation to
apparent overvaluation and back again.9 Thus, the motive for increased U.S.
investments may lie less in the specific level of the exchange rate around the
time of the investment than in the high probability of future large movements.
Here internalization does play a key role-allowing increased costs in one
location to be offset by reduced costs elsewhere and permitting some flexibil-
ity in shifting marginal production between locations on different sides of a
major rate alignment.
9.1.5 The Role of Protection
No other incentive for foreign direct investment has received as much atten-
tion as import barriers. It seems almost self-evident that tariffs or quotas will
stimulate direct investments in the protected markets. Recent developments in
the U. S. auto and electronics industries offer visible support for the proposi-
tion. Yet statistical analyses of Canadian and U.S. data have failed to confirm
a systematic relationship between direct investment and protection.
The likely reason for the weak empirical findings is that protection by itself
confers only a locational advantage. Whether that locational advantage leads
to inward investment or simply alters conditions of domestic entry and exit
depends on other industry characteristics. In the absence of a firm-specific
competitive advantage optimally exploited through internalization, domestic
producers will be better able than subsidiaries of foreign companies to capture
the benefits of local production.
Important though they are in their own right, autos and electronics may be
exceptions to the general rule. In these industries, technological know-how
and managerial know-how are firm-specific advantages that allow foreign
producers (notably Japanese) to compete effectively with established domestic
firms.l0 By contrast, the highly protected U.S. apparel and footwear indus-
tries have seen almost no direct investments from abroad. For these low-
technology industries, firm-specific advantages are apparently too small to
offset the greater costs incurred by foreign investors.
Evidence at the country rather than the industry level also casts doubt on
the hypothesis that protection is a strong magnet for inward direct invest-
ments. Among the less-developed countries, open, export-oriented economies
have been more successful in attracting new investments than nations pursu-
ing import-substitution strategies. For U.S. outward investments, Canada, the
268 Jonathan S. Leonard and Rachel McCulloch
United Kingdom, and Germany, all with relatively liberal trade regimes, have
been the most important host countries.
Bhagwati (1985) suggests a more subtle link between protection and direct
investment using the concept of “quid pro quo” foreign investment-invest-
ment made to defuse protectionist pressure rather than to circumvent actual
current or anticipated protection. On this interpretation, Japanese investments
in the automobile industry were intended, at least in part, to avoid future in-
creases in protection (e.g., local-content requirements) rather than to circum-
vent existing import restrictions. Presumably, such investments would lessen
the perceived need for protection and also would shift the domestic political
balance toward a more liberal trade stance.
An alternative “strategic” interpretation is suggested by the oligopolistic
structure of the auto industry and the extensive publicity surrounding Japa-
nese entry into U.S. production. Japanese firms may wish to demonstrate to
their U.S. rivals that Washington cannot protect them from yielding part of
their customary shares in the U.S. market. The new Japanese entrants could
actually benefit from future increases in protection if their competitive advan-
tage translates into lower costs in U.S. production while trade barriers prevent
the Big Three from putting their own nameplates on captive imports from
Japan or elsewhere.
9.1.6 Foreign Investment and U.S. Labor
For host countries worldwide, the most important anticipated benefit from
foreign investment is the creation of new jobs. In this, the United States has
been no diEerent. Holding out the prospect of hundreds or even thousands of
new jobs, U.S. state and local officials have mounted formidable campaigns
to lure foreign plants, usually offering sizable financial incentives to supple-
ment the region’s other attractions.’’ Yet, as with the presumed job losses
associated with outward investments of U.S. multinationals, the overall ef-
fects of inward investments on domestic employment are far from clear.
First, while there are obviously “new jobs” created by new subsidiaries, to
some extent these new jobs will be offset by employment losses elsewhere. In
the most optimistic scenario, local production will simply substitute for goods
previously imported.’* At least for the industry, the effect on total employment
should be positive. But local production by foreign affiliates can also cut into
the market share of domestic competitors, so that the new jobs are matched
by layoffs elsewhere in the same industry. If the affiliates use more imported
inputs than their domestic counterparts, production and employment may be
reduced accordingly in the supplier industries.
A second concern is about the types of jobs created. Will foreign multi-
nationals use U.S. labor for routine assembly operations, keeping the “good
jobs” at home? Reich and Mankin (1986) interpret Japanese joint ventures in
the United States as “part of a continuing, implicit Japanese strategy to keep
the higher paying, higher value-added jobs in Japan and to gain the project
269 Foreign-Owned Businesses in the United States
engineering and production process skills that underlie competitive success .”
A related concern shared with other host countries is that foreign affiliates
allow little opportunity for local workers to rise into management ranks.
While the Japanese presence in U.S. manufacturing is still too small and
too new to offer much evidence on this issue, most analysts agree that U.S.
operations abroad have benefited U.S. managers and skilled workers at the
expense of less-skilled U.S. production workers, whose jobs have moved off-
shore.
9.1.7
The close link between FDI and the investing firm’s competitive advantage
suggests that the rise in inward foreign investments in the United States as
well as the slowing of U.S. direct investments abroad reflect the industrial
catch-up of other nations to the United States. Where the competitive advan-
tages were once controlled almost exclusively by U.S. companies, new rivals
have emerged in Europe, Japan, and even some of the developing countries.
As with the successful U.S. multinationals of earlier decades, these firms have
exploited their competitive advantages first through exports and later through
direct investment in the market countries.
Like other host countries over the years, the United States is reevaluating
the potential gains and losses from allowing free entry to foreign subsidiaries.
U.S. Competitiveness and Inward Direct Investment
9.2 Data Sources and Methods
Section 9.3 compares foreign-owned U.S. firms with their U.S.-owned
counterparts. Most of the data used are derived from foreign direct investment
series published by the Bureau of Economic Analysis of the U.S. Department
of Commerce. Extensive cross-sectional data for 1980 are from Foreign Direct
Investment in the United States, 1980, a survey of U.S. business enterprises
in which foreign ownership, either direct or indirect, was at least 10 percent.
A number of caveats apply to these data, particularly where comparisons
are made to domestic aggregates. First, FDI data are reported in consolidated
form for the U.S. affiliates. The activities of each establishment within multi-
establishment enterprises are not classified separately by their own industry
(as in the National Income and Product Accounts) but, rather, are classified
by the industry group accounting for the largest percentage of the enterprise’s
sales. This undercounts data for industries with many “owned” establishments
and overcounts for industries with many “owner” establishments. It overstates
cross-industry variance in sales.
Second, FDI data aggregate petroleum-related activities including extrac-
tion, refining, and retailing. These have been removed from their respective
industries and aggregated into a separate category that has been suppressed in
the industry detail presented here. In consequence, these foreign-owned activ-
ities are undercounted in their respective subindustries.
270 Jonathan S. Leonard and Rachel McCulloch
Third, enterprises that are entirely foreign owned but in which no single
foreign person owns at least 10 percent are not classified as foreign owned.
For this purpose, “person” is defined to include any individual, partnership,
associated group, or corporation, including members of a syndicate or joint
venture.
Fourth, compensation and employment data are collected only for U.S. af-
filiates whose assets, sales, or net income exceeded $1 million or whose land
ownership exceeded two hundred acres.
Fifth, data are annual averages for each enterprise’s fiscal year.
Sixth, employment is reported as annual average number of employees, not
as full-time equivalents. This will cause an understatement of compensation
or wages per employee as reported here, but it should not affect the compari-
son of foreign direct with domestic because analogous concepts are used for
domestic. Compensation per employee is also understated in some cases by
the use of part-year compensation and year-end employment in some newly
acquired enterprises and establishments.
Finally, all the data discussed here are aggregates of enterprise data. The
composition of the underlying sample changes over time, as the section on
new acquisitions and establishments shows. Any change in, say, compensa-
tion per employee may then be due to (1) pay raises within previously sampled
establishments, (2) deletion of low-wage establishments from the sample, (3)
addition of high-wage establishments, or (4) purchase of a high-wage estab-
lishment by an enterprise with greater sales in another industry.
9.3 Buying American
Foreign direct investment in the United States includes all firms in which
10 percent or more of the equity is foreign owned. The stock of FDI increased
more than fourfold (in nominal terms) from $34.6 billion in 1977 to $159.6
billion in 1984. In 1977, the value of FDI in the United States was equal to
23.6 percent of the value of U.S. direct investment in foreign countries built
up during earlier decades. By 1984, the reversal of net investment flows was
well along. FDI in the United States was 68.4 percent of U.S. FDI abroad.
Between 1977 and 1984, FDI in the United States more than doubled in pro-
portion to the total value of all stocks listed on the New York Stock Exchange,
the proportion rising from 4.3 percent to 10.1 percent. In part, this reflects
growing foreign investments in all forms of U.S. assets. However, table 9.1
also shows that foreign direct investments increased in value relative to
foreign-owned stocks and to U.S. investment abroad. Both developments in-
dicate that potential foreign investors see greater competitive, locational, and
organizational advantages to establishment of U.S. subsidiaries than in earlier
periods.
High interest rates affected many foreign investments in U.S. financial in-
271 Foreign-Owned Businesses in the United States
Table 9.1 International Assets, 1977-84
1977 1978 1979 1980 1981 1982 1983 1984
1. Private foreign investment in
2. Row 1 as % of private
private U.S. assetsa 157.9 189.8 242.1 308.7 380.1 477.0 559.2 630.5
U.S. fixed nonresidential
gross capital 5.3 5.6 6.2 6.9 7.7 9.0 10.1 10.9
3. FDI in the US.” 34.6 42.5 54.5 83.0 108.7 124.7 137.1 159.6
4. Row3as%of NYSEVal ue 4.3 5.2 5.7 6.7 9. 5 9.6 8.7 10.1
5. Row 3 as % of foreign-
owned U.S. stocks 86.9 101.0 112.8 128.5 168.3 162.4 140.9 166.4
6. Row 3 as % of U.S. direct
investment abroad 23.6 26.1 29.0 38.5 47.6 56.2 60.4 68.4
Sources: Survey of Current Business, various issues; Economic Report of the President, various
issues; Statistical Abstract of the United States.
a Billion current dollars.
Table 9.2 Flows of Foreign Investment in the United States, 1960-86 (billions
of U.S. dollars)
Total Total FDI as % Japanese % of FDI
Year Inflow FDI of Total FDI from Japan
~
1960 2.3 .3 13.7 NA NA
1970 6.4 1.5 23.0 .o 3.6
1972 21.5 .9 4.4 .o 2.0
1974 22.5 4.8 21 .o .2 4.4
1976 42.7 4.3 10.2 .6 13.5
1978 65.4 7.2 12.1 1 .a 12.5
1980 84.7 13.7 16.1 .7 5.3
1981 78.2 22.0 28.1 2.8 12.6
1982 109.7 10.4 9.5 I .7 16.8
1983 96.9 11.9 12.3 1.7 13.8
1984 108.2 25.4 23.4 4.4 17.2
1985 127.1 17.9 14.0 3. I 17.3
1986 213.3 25.6 12.0 4.7 18.5
Sources: Survey of Current Business, various issues (for 1972-86); Business Statistics, 1984 (for
1960 and 1970).
Note: Percentages calculated from unrounded flow data. Data for 1986 are preliminary.
struments. The composition effect created by the increase in foreign owner-
ship of U. S. financial assets overshadows a less noticed shift in foreign in-
vestment toward direct corporate ownership. Since 1977, foreigners have
increasingly been purchasing control of U.S. corporations. The value of in-
ward FDI, as a percentage of foreign ownership of U. S. stocks, rose from 87
percent in 1977 to 166 percent in 1984.
Table 9.2 shows the trends in the flow of new FDI. The total inflow of
foreign investments increased in the 1970s and 1980s. Although FDI as a
272 Jonathan S. Leonard and Rachel McCulloch
percentage of the total inflow of investment does not show a distinct trend
over this longer period, Japanese FDI has increased substantially. The flow
statistics offer less support for the hypothesis of changing competitive, loca-
tional, or organizational advantages to establishment of U.S. subsidiaries by
foreign-based multinational corporations. The absence of a trend in FDI as a
percentage of total investment inflow indicates that the growth in FDI may be
simply a manifestation of the growth in all forms of foreign investment in the
United States. The trend in Japanese FDI in the United States suggests that
economic advantage arguments may, however, apply to Japan and to the in-
dustries in which Japanese companies are highly visible.
9.3.1 Acquisitions and New Establishments
Newly acquired or established enterprises show one form of increased in-
vestment by foreigners in the United States (see table 9.3). The Bureau of
Economic Analysis (BEA) classifies as a new acquisition an existing U.S.
enterprise in which foreign ownership (directly or through U.S. affiliates)
passes 10 percent. However, this is only a small part of total investment, be-
cause additional equity investments in existing U.S. affiliates are not counted
once the 10 percent threshold has been passed, and because only enterprises
with assets exceeding $1 million or two hundred acres of U.S. land are in-
cluded. To illustrate, of the 2.1 million U.S. employees of foreign-owned
companies in 1980, 13 percent were in newly acquired enterprises, and .6
percent were in newly established enterprises. Compared to analogous rates
for total domestic industry, the acquisition rate is high and the start-up rate
low. Of the $522 million of foreign-owned 1980 assets, 8 percent were newly
acquired and 1.4 percent newly established. Roughly 80 percent of these in-
vestment funds came through existing U.S. affiliates. In 1980, 37 percent of
these investments were financed by U.S.-source funds. Only 2 percent of
Table 9.3 Outlays and Employment in U.S. Enterprises Newly Acquired or
Established by Foreign Direct Investors, 1979-85
Outlays Employment New Employment New
($ million current) Establishments Acquisitions
1979 15,317 15,467 314,548
1980 12,172 13,022 279,459
1981 23,219 14,072 428,745
1982 10,817 8,169 225,673
1983 8,091 5,556 102,557
1984 15,197 4,139 168,406
1985 19,547 7,772’ 235,667’
Sources: Outlays: Shea (1986, 47, table 1). Employment: same 1979, 1980, 1981, 1982, 1983,
1984, and 1985.
preliminary
273 Foreign-Owned Businesses in the United States
these investments (1982) were reported to receive specific state or local in-
vestment incentives or subsidies.
Measured either by employment or assets added in new acquisitions or es-
tablishments, inward FDI has been volatile and shows no clear trend. This is
misleading. Indeed, both assets and employment in foreign direct investments
have been growing steadily in the 1980s. The difference arises because most
of the growth has occurred in ongoing foreign-owned businesses. This is sim-
ilar to the growth process of domestic industry generally, which is also domi-
nated by the expansion of ongoing concerns.
9.3.2 Foreign Ownership by Industry
The nature of the industries in which foreigners invest does differ substan-
tially from domestic industry as a whole. Table 9.4 shows the industrial distri-
bution of employment in foreign-owned businesses. It bears greater resem-
blance to U.S. direct investment in other developed economies. Overall,
foreigners invest predominantly in U.S. manufacturing industries. While 22.1
percent of 1980 U.S. employment was in manufacturing, this sector ac-
counted for fully 54.3 percent of FDI. The service and retail trade sectors
show the fastest growth in FDI, but manufacturing still dominates. Employ-
ment in foreign-owned manufacturing doubled between 1977 and 1984.
Measured as a percentage of total industrial employment, foreign owner-
ship has advanced farthest in chemicals, where 39 percent of all employment
is in foreign-owned establishments; stone, clay, and glass (1 1 percent); pri-
mary metals (1 1 percent); food (9 percent); and electrical machinery (8 per-
cent).I3 The chemical industry stands out as a case in which foreign ownership
is approaching a majority of the industry. These are all manufacturing indus-
tries in which the foreign parent may have a competitive advantage due to the
importance of technology in determining business success. Of the sectors
Table 9.4 Industrial Distribution of Employment in Foreign-owned Business,
1977-84. (%)
1977 1978 1979 1980 1981 1982 1983 1984
Mining 1.31 1.12 1.03 1.23 1.65 1.67 1.46 1.18
Manufacturing 56.28 56.22 57.39 54.33 53.79 50.74 51.54 50.76
Wholesaletrade 12.55 12.03 11.18 10.67 10.51 11.44 10.61 10.79
Retail trade 11.65 11.96 13.46 14.95 14.23 16.26 16.51 16.72
Construction 1.07 1.61 1.60 2.11 2.40 2.12 1.98 1.55
Services 3.04 3.57 3.76 4.18 5.13 5.43 5.34 7.07
Residual 14.11 13.50 11.58 12.54 12.29 12.34 12.55 11.93
Total
(thousands) 1,219 1,430 1,753 2,034 2,417 2,448 2,526 2,715
Sources: U.S. Department of Commerce (1983) and Shea (1986).
Note: Foreign employment in industry i over total foreign employment (in %), 1977-84.
274 Jonathan S. Leonard and Rachel McCulloch
with highest foreign ownership, two (electronics and chemicals) are com-
monly considered to embody advanced and rapidly progressing technology.
In most other industries, the share of employment in foreign-owned busi-
nesses remains under 5 percent. Foreign employment as a share of total em-
ployment is notably low in communications and public utilities (. 4 percent),
services ( . 8 percent), agriculture (. 5 percent), and construction (.9 percent).
Regulation limits access to the first of these markets. The others are all non-
manufacturing industries with low domestic sales concentration.
Foreign ownership rates are highest within manufacturing. Foreign owner-
ship is increasing in almost every industry, including stagnant industries such
as primary metals. In recent years, it has increased fastest in such home-goods
industries as service, real estate, and retail trade.
9.3.3 Location Decisions
As new entrants to the U.S. employment market, foreign direct investors
have at times been characterized as locating in the low-wage South, the grow-
ing West, or the technologically advanced Northeast. From the perspective of
any of these stories, the surprising fact is just how closely the geographic
distribution of FDI employment parallels that of all domestic firms.
Table 9.5 shows, for each of the nine major Census geographic divisions,
the shares of foreign direct and of domestic employment for all sectors and for
manufacturing. The largest difference between FDI and domestic location oc-
curs in the Middle Atlantic states (New York, Pennsylvania, and New Jersey),
which account for 21 percent of FDI employment but just 17 percent of do-
Table 9.5 Geographic Distribution of Employment, 1980 (in thousands)
Geographic
Fraction of Total
Fraction of Total Manufacturing Manufacturing
Region Foreign Domestic Foreign Domestic Foreign Domestic Foreign Domestic
New England 122.9 5,474.5 .06 .06 72.2 1,524.6 .07 .07
Middle Atlantic 414.5 15,011.6 .21 .17 217.2 3,554.2 .20 . I7
South Atlantic 363.1 14,625.2 . I 8 .16 192.1 3,041.5 .I8 . I 5
East North
Central 368.1 16,826.8 .18 . I9 226.4 4,687.6 .21 .23
East South
Central 99.5 5,145.5 .05 .06 61.5 1,362.7 .06 .07
West North
Central 103.2 6,903.0 .05 .08 63.2 1,381.3 .06 .07
West South
Central 209.8 9,313.3 .10 .10 93.2 1,669.5 .08 .08
Mountain 64.0 4,488.1 .03 .05 29.1 563.2 .03 .03
Pacific 274.8 13,058.8 .14 .14 142.7 2,569.0 .13 .13
Total 2,019.9 90,846.8 1.00 1.00 1,097.7 20,353.6 1.00 1.00
Sources: U.S. Department of Commerce (1983, table F-7); U.S. Bureau of Labor Statistics.
275 Foreign-Owned Businesses in the United States
mestic. FDI employment is also more prevalent in the South Atlantic states
and less prevalent in the relatively depressed East North Central region. Aside
from these differences, the location decisions of foreign direct investors in the
United States look much like those of domestic employers. As we noted
above, much FDI involves the creation of wholesale, retail, and service estab-
lishments to support the international trade of the parent. Apparently, the geo-
graphic distribution of population in the United States is more important than
regional labor market differences in determining the location of these estab-
lishments.
9.3.4 Compensation Differences
U.S. production workers do not appear to suffer under foreign ownership.
In the manufacturing sector, the ratio of compensation per worker in foreign-
owned to the same measure for employees of U.S .-owned businesses in-
creased from .94 in 1977 to 1.08 in 1984 (table 9.6). Foreign ownership has
little effect on the mix of fringes to wages, so wages show a similar pattern.
While compensation per worker is 10 percent lower in the foreign-owned es-
tablishments of the food, primary metals, and instruments industries, in gen-
eral foreign gains in domestic industries have not been accompanied by rela-
tively low-wage labor.
Overall, workers in foreign-owned businesses enjoy a 20-30 percent ad-
vantage in compensation over employees of U.S.-owned businesses. The
compensation differential can be decomposed into a within-industry differen-
tial and a composition effect. Only about a third of the overall difference is
due to higher compensation in foreign direct employment within industry. For
the most part, the higher compensation found in the aggregate in foreign di-
rect employment is explained by the greater concentration of FDI in the high-
wage manufacturing sector.
For manufacturing, the BEA provides separate data on wages, hours, and
occupational structure. The hourly wages of production workers in foreign-
owned enterprises are 8 percent greater than the domestic average in 1980,
although the hours worked are 8 percent less. For production workers, in-
creased hourly pay is balanced by shorter hours. Foreign-owned firms in man-
ufacturing appear more top heavy, employing 64 percent production workers
in 1980 compared to the domestic average of 70 percent (see table 9.7).14
Since the overall 1980 compensation ratio is less than one, non-production
workers in foreign-owned firms appear to be paid less than their counterparts
in U. S . -owned firms.
As table 9.6 shows, foreigners appear to invest in high-wage industries. In
our discussion of the sources of advantage to FDI, we noted that locational
advantages may be conferred by protective trade policies, as in the automotive
industry, for example. However, successful direct investment requires that the
foreign investor also possess competitive advantage. Otherwise, only domes-
tic entry and pricing decisions are altered by the protection. When some trade
276 Jonathan S. Leonard and Rachel McCulloch
Table 9.6 Compensation Ratios (foreign compensation per worker over
domestic compensation per worker)
Industry 1977 1978 1979 1980 1981 1982 1983 1984
All industries
Mining
Manufacturing
All nonpetroleum industries
Durable
Transportation & equipment
Primary metal industries
Fabricated metal products
Machinery, except electrical
Electric & electronic
equipment
Nondurable
Textile products & apparel
Lumber & furniture
Paper & allied products
Printing & publishing
Rubber & plastics products
Stone, clay, & glass products
Food & kindred products
Wholesale trade
Retail trade
Construction
Services
Finance
Real estate
Insurance
Communication
Transportation
1.20 1.23 1.20 1.20 1.27
1.17 1.20 1.18 1.19 1.42
.94 1.02 .98 1.02 1.09
.94 .96 .94 .95 1.06
.81 1.07 .93 .97 .97
.83 .86 .80 .86 .86
.96 .98 .98 .90 .98
.96 .99 .98 .97 .99
.83 .86 .87 .90 .99
1.21 1.13 1.07 1.07 1.14
1.01 .93 .81 .89 .84
.99 1.00 .98 .96 1.22
1.02 1.03 1.07 1.07 1.10
.82 .83 .87 .95 .91
.90 .97 .97 1.08 1.11
.89 .91 .78 .83 .89
1.03 1.03 1.01 1.02 1.01
1.22 1.21 1.29 1.24 1.18
.80 1.10 1.16 1.06 1.16
1.10 1.17 1.11 1.04 .94
1.33 1.67 1.68 1.54 1.54
1.02 .96 1.40 1.38 1.34
1.00 1.02 1.01 .99 .94
1.31 .86 1.13 .74 .76
1.05 1.07 1.09 .90 .88
1.30
1.28
1.16
1.11
.98
.90
1.09
1.10
1 .OO
1.21
.83
1.24
1.11
.94
1.21
.90
1.11
1.21
1.23
.97
1.91
1.19
.93
.63
.93
1.29 1.26
1.28 1.24
1.16 1.14
1.08 1.08
.90 .96
.95 .99
.96 1.12
1.05 1.01
.96 .98
1.17 1.21
1.03 1.01
1.30 1.18
1.14 1.14
.92 1.06
1.17 1.12
.87 .91
1.19 1.13
1.13 1.05
1.26 1.14
1.03 1.05
1.87 1.71
1.12 1.12
.91 .91
.55 .57
1.14 1.07
Sources: U.S. Department of Commerce (1983); Survey of Current Business; National Income
and Product Accounts.
flows are restricted through commercial policy, factor flows are magnified. A
partial explanation of the tendency of foreigners to invest in high-wage U.S.
industries (and to pay the U.S. wage in those industries) is the export of capi-
tal to the United States as an alternative to export of the products of their
domestic high-wage industries.
9.3.5 Research and Development
Just as U.S. direct investment in other countries is dominated by industries
with substantial research and development (R&D), foreigners investing in
U.S. domestic industries use R&D to generate a competitive advantage.
While R&D scientists and engineers constitute only .5 percent of employees
of domestic-owned firms, they are 2.1 percent of the employees in foreign-
owned companies (see table 9.8). By this measure, R&D intensity is half
277 Foreign-Owned Businesses in the United States
Table 9.7 Proportion of Production Workers in Manufacturing, 1980
Ownership
Foreign Domestic
Manufacturing
Food & kindred products
Chemicals & allied products
Primary & fabricated metals
Primary metals
Fabricated metals
Machinery, except electrical
Electric & electronic
Textile products & apparel
Lumber & furniture
Paper & allied products
Printing & publishing
Rubber & plastics
Stone, clay, & glass
Transportation equipment
Instruments & related products
Machinery
.64
.68
.48
.72
.71
.74
.61
.57
.63
.81
.79
.74
.57
.75
.75
.66
.64
.70
.69
.57
.75
.77
.74
.64
.64
.64
.86
.82
.75
.56
.77
.71
.65
.60
Sources: U.S. Department of Commerce (1983); Statisrical Abstracr of the Vnired Srares.
Table 9.8
Research and Development Intensity, 1980
9% of Total Employees Who
are R&D Scientists and
Engineers
Foreign Domestic
All industries
Manufacturing
Food & kindred products
Chemicals & allied products
Primary metal industries
Fabricated metal products
Electric & electronic equipment
Textile products & apparel
Paper & allied products
Stone, clay, & glass products
Transportation & equipment
Instruments & related products
Machinery
Other manufacturing
2.13
S O
6.14
.72
2.40
3.62
3.78
.39
.43
.94
2.34
4.10
.46
.42
4.48
.78
.61
3.03
3.91
.09
1.09
.81
1.88
3.86
Sources: U.S. Department of Commerce (1983); National Science Foundation.
278 Jonathan S. Leonard and Rachel McCulloch
again as high in foreign-owned compared to domestic-owned manufacturing
businesses (3.1 vs. 2.0 percent) and more than ten times greater in nonmanu-
facturing (.9 vs. .07 percent). One need not be xenophobic to wonder about
the fate of U.S. comparative advantage in R&D-intensive industries when the
declining share of U.S. citizens in U.S. graduate science and engineering edu-
cation is coupled with the declining domestic-ownership share in U.S. R&D-
intensive industries.
9.3.6 Collective Bargaining
Among industrialized countries, the United States now has one of the low-
est union representation rates. Among the many explanations proffered, some
have pointed to differences in management attitude. Managers from some Eu-
ropean countries and Japan are often surprised at the unquestioned and vehe-
ment antiunion animus of their U.S. counterparts.
Higher unionization rates in home countries may be associated with greater
management tolerance of unionization and perhaps with greater skill in devel-
oping cooperative arrangements with unions. If foreign owners really take a
less antagonistic position toward unions, one might expect this to carry over
to their U.S. operations and reveal itself in higher unionization rates than in
U. S .-owned domestic operations.
On the other hand, once geographically removed from home country appro-
bation and leverage, the same cost considerations that drive U. S. companies
may dominate. To the extent that cost disadvantages of union firms can be
capitalized in the sales price of corporate assets, no difference in unionization
is expected on the basis of foreign ownership.
In 1980, 23 percent of U.S. employees were union members. Among
foreign-owned companies, 29 percent of employees were covered by collec-
Table 9.9 Union Density at Home and Abroad, 1980
% Covered by
Collective Bargaining
in Foreign Owned
% Unionized
All Industry Manufacturing Home Country
United States
All foreign owned
Canada
United Kingdom
Japan
Netherlands
Sweden
France
Germany
Switzerland
29.21
32.05
26.07
20.26
24.52
31.90
47.49
30.96
17.35
35.9
31.13
35.39
29.11
28.82
34.03
55.05
20.06
22.55
23.1
30.50
53.10
30.80
37.10
87.80
19.20
38.60
33.50
Sources: U.S. Department of Commerce (1983); Kokkelenberg and Sockell (1985); Troy and
Sheflin (1985).
279 Foreign-Owned Businesses in the United States
tive bargaining contracts (see table 9.9). Again, this difference is almost en-
tirely due to the greater concentration of FDI in the manufacturing sector.
Within manufacturing, foreigners leave any pro-union sentiments at home.
Whereas 36 percent of U.S. employees are union members, only 31 percent
of the employees in foreign-owned companies are covered by collective bar-
gaining contracts (a more inclusive measure).
Part of the difference in manufacturing union density may reflect union
avoidance on the part of foreigners. Part may reflect compositional differences
within manufacturing, and part may be due to a vintage effect. Newer firms
and industries are less unionized, and foreign ownership is presumably con-
centrated among these. In any case, owner attitudes inferred from home-
country unionization can be dismissed as an important factor. The growing
internationalization of the world economy has so far presented greater com-
petition attacking local rents and greater opportunities for union avoidance
than for the international application of union leverage to enforce union stan-
dards.
9.3.7 Home-Country Effects
The characteristics of FDI in the United States may differ systematically by
home country of the investor, although in general one may suspect that such
differences either are transient or represent industry-specific or firm-specific
effects. For the countries that are home to most of the ultimate beneficial own-
ers of FDI in the United States, table 9.10 compares a number of characteris-
tics.
Of these countries, investments owned by Australians and Dutch appear the
most successful. They show the highest return on assets and the highest pay.
The Dutch investments are also R&D intensive. Countries such as Germany
and Switzerland, with the greatest share of their investment in manufacturing,
show the worst rates of return. Japanese investments stand out only in their
avoidance of manufacturing, compared to the investments of other foreigners.
In general, the measures in table 9.10 appear to tell more about the common
characteristics of FDI in the United States than about differences systemati-
cally related to home-country factors.
9.4 Conclusions
The close link between foreign direct investment and the investing firm’s
competitive advantage suggests that both the rise in inward direct investment
in the United States and the slowing of U.S. direct investment abroad reflect
the industrial catch-up of other nations to the United States. Where the com-
petitive advantages that underlie successful foreign investment were once con-
trolled almost exclusively by U.S. companies, new rivals have emerged in
Europe and Japan, and even in some of the developing countries. As with the
U. S .-based multinationals that dominated global direct investment flows in
earlier decades, these foreign-based firms have exploited their competitive
Table 9.10 Characteristics of Foreign Direct Investment by Country of Origin, 1980
All Foreign
Countries
1. Net income/assets
2. Wagesiworkers*
3. Compensation/workers*
4. R&D employedlworkers
5. Manufacturing employed/workers
6. Wages/workers**
7. Compensation/workers**
,031
16.28
19.69
.021
.54
16.92
20.67
Canada
.022
16.92
20.68
.011
.52
17.19
21.90
United
Kingdom
,056
14.59
17.80
,017
.52
15.13
18.50
Japan
,026
16.26
18.57
,015
.31
16.35
19.25
Nether-
lands
,049
18.84
23.00
,040
.55
17.60
21.13
France
,011
18.58
22.76
.016
.58
20.33
25.64
Germany
Switzer-
land
Australia,
New Zealand,
South Africa
United
States
,006
15.40
18.73
.024
.64
15.76
19.38
.008
15.99
19.04
,038
.65
17.88
21.50
,097
18.17
21.35
,019
.29
17.97
21.12
.059
13.91
16.39
,005
.20
17.36
21.68
Sources: U.S. Department of Commerce (1983, tables B-8, E-2, F-2, F-13, F-14).
Nore: U.S. column gives comparable domestic averages.
*In thousands of dollars.
**In manufacturing.
281 Foreign-Owned Businesses in the United States
advantages first through exports and later through direct investments in market
countries.
As with U.S. companies investing abroad, foreign firms establishing sub-
sidiaries in the United States often rely on superior technology for the corn-
petitive advantage necessary to make their investments profitable. Although
these firms come to the United States to exploit an already-established corn-
petitive advantage, their U.S. operations employ a larger proportion of scien-
tists and engineers than U.S.-owned businesses in the same industry; foreign-
owned enterprises in the United States are on average more R&D intensive
than their domestic counterparts.
However, R&D intensity is the only large difference between foreign-
owned and US.-owned businesses that emerged from our statistical com-
parison. Indeed, it is striking how similar foreign-owned and U.S.-owned
businesses appear statistically. Although foreign-owned companies have a dif-
ferent industrial mix favoring manufacturing, retail and service establish-
ments are growing fastest. Wages and compensation of foreign-owned busi-
nesses are very similar to those of U.S.-owned businesses in the same
industries.
Notes
1. In U.S. statistics, the line is drawn at 10 percent of equity, although most U.S.
direct investments abroad and many foreign direct investments in the United States are
wholly owned subsidiaries of the parent. Other asset purchases, e.g., of private or
government bonds or smaller blocks of stocks, are termed portfolio investments.
2. The size of the ownership stake need not indicate the total size of the controlled
activity. A fall in measured direct investment could, in principle, be accompanied by
an increase in the extent of controlled activity. Alternative measures of foreign influ-
ence include sales, employment, and profits of the controlled enterprise.
3. Although appropriately classed as asset transactions, direct investments are dis-
tinctive in that no international transfer of financial capital, i.e., purchasing power,
need be entailed. In many cases, the contribution of the investing firm to a joint venture
consists primarily of proprietary technology or managerial expertise rather than finan-
cial capital. Even when financial capital is part of the investment “package,” the re-
quired funds may be borrowed locally in the host country. This was a common practice
in the 1960s for the U.S. firms establishing European subsidiaries.
4. In 1983, US. multinational corporations accounted for more than three-quarters
of U.S. exports and almost half of U.S. imports. However, these shares have been
declining from their peaks in the 1970s (the comparable percentages for 1977 were 84
and 58) , while the U.S. trade role of foreign multinational firms appears to have grown
over the same period. (See Barker 1986.)
5. This has been alleged in recent years about Japanese investments in the United
States but is also a longstanding complaint of less-developed host countries. These
nations invite foreign investments with the hope of reducing chronic balance-of-
payments difficulties. The usual experience is that induced imports of machinery and
components tend to offset any direct reduction in imports or rise in exports of the
282 Jonathan S. Leonard and Rachel McCulloch
product itself. This is, of course, consistent with the notion that balance-of-payments
difficulties are fundamentally macroeconomic problems requiring macroeconomic so-
lutions. On the macroeconomic roots of the U.S. trade deficit, see McCulloch and
Richardson (1986).
6. The idea that FDI requires a significant departure from conditions of perfect
competition was advanced by Hymer (1960) and expanded by Kindleberger (1969),
Caves, and many others. For a comprehensive survey of the literature, see Caves
(1982, chap. 2).
7. The three “necessary conditions” are elaborated by Dunning (198 1, and earlier
papers). Dunning uses this classification to explain the distribution of investment by
home and host country and industry. For a detailed analysis of direct investment as
international internalization, see Rugman (1980).
8. The U.S. investments by Japanese firms are a case in point. While this kind of
complementary relationship between exporting and direct investment was suggested
by Bergsten, Horst, and Moran (1978) for manufacturing investments, it may be even
more important in the case of service industries such as banking and insurance.
9. As long as rates are determined mainly by market forces, in one sense under- or
overvaluation cannot occur. These descriptions usually refer to a deviation of market-
determined rates from rates calculated using relative price levels (purchasing power
parity).
10. As discussed below, local policies to attract new investment may also play a
role.
11. These benefits include tax breaks, cheap loans, worker training, and free infra-
structure. According to a Kentucky legislative study, the state will spend $125 million,
or about $42 perjob, to attract Toyota’s new plant (Lore 1987).
12. In early 1987, the president of Ford Motor Co. called for further reductions in
auto imports from Japan to compensate for increased production by Japanese plants in
the United States.
13. Statistics are for 1984. Sources are U.S. Department of Commerce, Foreign
Direct Investment in the United States, and the Survey of Current Business.
14. This difference may be understated by the use of full-time equivalent counting
only for the domestic figure.
References
Barker, Betty L. 1986. U.S. Merchandise Trade Associated with U.S. Multinational
Companies. Survey of Current Business 66 (May):55-72.
Bergsten, C. Fred, Thomas Horst, and Theodore H. Moran. 1978. American Multi-
nationals and American Interests. Washington, D.C.: Brookings.
Bhagwati, Jagdish. 1985. Investing Abroad. Esmee Fairbairn Lecture, University of
Lancaster, 27 November.
Caves, Richard E. 1982. Multinational Enterprise and Economic Analysis. Cam-
bridge: Cambridge University Press.
Dewald, William, Harry Gilman, Harry Grubert, and Larry Wipf, eds. 1978. The
Impact of International Trade and Investment on Employment. Washington, D.C. :
U.S. Department of Labor.
Dunning, John H. 1981. Explaining the International Direct Investment Position of
Countries: Towards a Dynamic or Developmental Approach. Weltwirtschaftliches
Archiv 117:30-64.
283 Foreign-Owned Businesses in the United States
Frank, Robert H., and Richard T. Freeman. 1978. The Distributional Consequences of
Direct Foreign Investment. In The Impact of International Trade and Investment on
Employment, ed. William Dewald, Hany Gilman, Harry Grubert, and Larry Wipf.
Washington, D.C.: U.S. Department of Labor.
Graham, Edward M. 1978. Transatlantic Investment by Multinational Firms: A Rival-
istic Phenomenon? Journal of Post Keynesian Economics 1 (Fall):82-99.
Hymer, Stephen. 1960. The International Operations of Firms. Ph.D. diss., Massa-
chusetts Institute of Technology.
Kindleberger, Charles P. 1969. American Business Abroad: Six Lectures on Direct
Investment. New Haven, Conn.: Yale University Press.
Kogut, Bruce. 1983. Foreign Direct Investment as a Sequential Process. In The Multi-
national Corporation in the 1980’s ed. C. P. Kindleberger and D. B. Audretsch,
Cambridge, Mass.: MIT Press.
Kokkelenberg, E. C., and D. R., Sockell. 1985. Union Membership in the United
States, 1973-1981. Industrial and Labor Relations Review 38:497-543.
Lipsey, Robert E. 1987. Changing Patterns of International Investment in and by the
United States. In The United States in the World Economy, ed. Martin Feldstein,
475-545. Chicago: University of Chicago Press.
Lore, Dave. 1987. Japanese Auto Investments: Great Today but What about Tomor-
row? MidAmerican Outlook 1 O(Spring): 2-4.
McCulloch, Rachel, and J. David Richardson. 1986. U.S. Trade and the Dollar: Eval-
uating Current Policy Options. In Current US. Trade Policy: Analysis, Agenda and
Administration, ed. R. E. Baldwin and J. D. Richardson. Cambridge, Mass.: Na-
tional Bureau of Economic Research.
Musgrave, Peggy B . 1975. Direct Foreign Investment Abroad and the Multinationals:
Effects on the US. Economy. Washington, D.C.: Senate Foreign Relations Com-
mittee.
Reich, Robert B., and Eric D. Mankin. 1986. Joint Ventures with Japan Give Away
Our Future. Harvard Business Review 64(March/April):78-86.
Rugman, Alan M. 1980. Internationalization as a General Theory of Foreign Direct
Investment: A Re-appraisal of the Literature. Weltwirtschaftliches Archiv 116:365-
79.
Shea, Michael A. 1986. U.S. Business Enterprises Acquired or Established by Foreign
Direct Investors in 1985. Survey of Current Business 66(5):47-53.
Troy, L., and N. Sheflin. 1985. Union Sourcebook. West Orange, N.J.: IRDIS.
U.S. Department of Commerce. 1983. Foreign Direct Investment in the United States,
1980. Washington, D.C.: U.S. Government Printing Office.
Vernon, Raymond. 1971. Sovereignty at Bay. New York: Basic.
This Page Intentionally Left Blank
10 Immigration, International
Trade, and the Wages of
Native Workers
Peter Kuhn and Ian Wooton
The purpose of this paper is to develop and apply to U.S. data a theoretical
model with the following features. First, it should yield a set of predictions
regarding the effects of international factor movements, such as immigration,
on the rewards of all factors employed in the country, including labor disag-
gregated by skill level. Second, it should be consistent with the following
stylized facts: (i) the U.S. economy is “partially open” in the sense that it
produces both internationally traded and nontraded goods; and (ii) interna-
tional trade in goods has apparently not equalized factor prices between the
United States and the rest of the world.
These requirements play an important role in this paper because few of the
existing models that consider general-equilibrium effects of factor endow-
ments on factor prices satisfy them. For example, Hicks’s (1932, chap. 6)
classic analysis assumes that no goods are traded internationally. This work
predicted that the effect of factor quantities on factor prices was determined
by a set of within-industry elasticities of substitution as well as substitution
elasticities in consumption, and it stimulated several empirical attempts to
estimate these parameters (e.g., Fallon and Layard 1975). On the other hand,
the basic two-good, two-factor (2 X 2) trade model (Samuelson 1948) as-
sumes that all produced goods are traded. It predicts, unrealistically, that trade
alone should eliminate all factor price differentials between countries and thus
that factor endowments should have no effect on factor prices. Finally, among
the trade models that do allow for international factor movements to affect
factor prices (e.g., the 2 X 2 models of Kemp 1966; Markusen and Melvin
1979; Brecher and Choudhri 1982; and Rivera-Batiz 1982; the two-good,
three-factor (2 X 3) models of Batra and Casas 1976; Ruffin 1981; and Jones
Peter Kuhn is associate professor of economics, Department of Economics, McMaster Univer-
sity. Ian Wooton is associate professor of economics, Department of Economics, University of
Western Ontario.
285
286 Peter Kuhn and Ian Wooton
and Easton 1983; as well as the higher-dimensional treatments of Jones and
Scheinkman 1977; or Chang 1979), the only one that includes any nontraded
goods is Rivera-Batiz (1982).
The model in this paper can be thought of as an extension of Rivera-Batiz
(1982), which adds an extra traded good and an extra factor. We thus have
three factors, two traded goods, and one nontraded good. The additional
traded good allows us to have both an exporting and an import-competing
industry in the analysis and to compare how these two sectors are affected by
changes in factor endowments. The additional factor allows us to distinguish
between workers with different investments in human capital. Specifically, we
subdivide the labor force into skilled workers and unskilled workers. Immi-
grants of a particular type are considered to be perfect substitutes for native
workers of that same category. Our model may also be thought of as the addi-
tion of a nontraded sector to Ruffin’s (1981) 2 X 3 model.
The paper’s main theoretical results are twofold. First, we find that the di-
rections of the effects of factor endowments on factor prices, while not zero
as in the “standard” trade model, are still independent of the within-industry
technical substitution elasticities between inputs in production. This inde-
pendence property (which incidentally also holds in Rivera-Batiz’s lower-
dimensional model) dramatically illustrates the effects of allowing interna-
tional trade in even a subset of commodities on models of the functional
distribution of income. It arises because, contrary to the closed-economy
model, the fundamental determinants of factor price changes are not the abil-
ity to substitute factors in production; they are, instead, the tendency for fac-
tor prices to change in such a way as to maintain the international competitive-
ness of the country’s exporting and import-competing industries, as long as
those industries continue to operate.
Second, providing that a relatively weak “normality” condition holds, the
directions of all the factor quantity-factor price effects in our model can be
deduced directly from the relative intensities of factor use within the traded
sector of the economy only, as follows. First, an increase in the supply of any
factor lowers its own price. Second, with three factors, one will be “ex-
tremely” intensively used in exports, another in imports, and the third will be
the “middle” factor in the traded sector of the economy. Our model predicts
that an increase in the supply of either extreme factor lowers the price of the
other extreme factor and raises the price of the middle factor. Third, an in-
crease in the supply of the middle factor benefits owners of both extreme fac-
tors. Interestingly, the results given above are identical to those obtained by
Ruffin (1981) without a nontraded sector.
The paper’s main empirical result, based on factor intensities in 430 four-
digit U.S. manufacturing industries for the years 1960, 1970, 1980, and 1984,
is the following. For all definitions of traded versus nontraded goods consid-
ered, and for all years except 1960, skilled labor is extreme in exports and
unskilled labor is extreme in imports, with capital as the middle factor. Thus,
287 Immigration and Wages
our model predicts that, in the long run, the interests of both types of labor in
immigration issues should coincide and should conflict with those of capital.
Workers of both types should oppose all immigration but favor foreign invest-
ment in the United States, while owners of capital should favor immigration
of both types of workers.
Section 10.1 of the paper outlines the structure of the model. Section 10.2
solves the model for the effect of factor endowment changes on factor prices.
Section 10.3 characterizes the properties of that solution. Section 10.4 pre-
sents our empirical estimates of factor intensities for the United States and
their implications, while section 10.5 concludes.
10.1 The Model
Each of the three goods X , , X,, and X , is produced using the services of the
three factors of production, V, , V2, and V, , according to linearly homogeneous
production functions. We adopt the convention that good 3 is nontraded and
that, of the two traded goods, XI is imported and X , exported. Let qJ be the
quantity of factor i required to produce a unit of good j , where a,J depends on
the prices of the three factors w,, w,, and w,. Without loss of generality, we
number factors in such a way that
al l i u12 ‘2Iia22 u31iu32
in the initial equilibrium and assume the inequalities are strict. Thus, in Ruf-
fin’s (1981) terminology, when comparing factor intensities of the two traded
sectors, factor 1 is extreme in imports (XI), factor 2 in exports ( XJ , and factor
3 is the middle factor in the traded sector.
If the nominal prices of the three goods are p , , p,, and p, , then the zero-
profit conditions for production are
(1) a l l wl + %Iw2 + ‘3Iw.3 = PI?
( 2) a12wl + ‘2Zw2 + = p2 9
(3) a13wl + a23w2 + a33w3 = p3.
It is assumed that all three goods are produced in positive amounts, so the
market prices exactly reflect the costs per unit of output.
Full employment of the stocks of the three factors of production would
entail
Were all three goods prices exogenously determined, then equations ( l ) , ( 2) ,
and (3) would uniquely determine the factor-price vector, and factor prices
288 Peter Kuhn and Ian Wooton
would be independent of the factor endowments. However, while goods 1 and
2 are considered to be traded internationally at exogenously given world
prices, it is assumed that good 3 is nontraded, its price being endogenously
determined by domestic demand. In consequence, changes in factor en-
dowments, through immigration, may influence the returns to factors in the
economy.
Let domestic demand for good 3 be represented by a Hicksian compensated
demand function, that is,
Equilibrium in the market for the nontraded good occurs when domestic sup-
ply exactly meets domestic demand,
(7)
and “national’ utility is a function of the quantities of goods consumed in the
country by native and immigrant factors together:
(8) u = U[ C, , c,, q.
PI CI + P2 c2 = PIX, + PIX,,
x3 = g[ X, , x,; v,, v,, v31.
National utility is maximized subject both to the balanced trade constraint,’
( 9)
and to the constraints of technology and endowments,
(10)
Equations (I)-( 10) provide a complete description of the static general
equilibrium of the economy.
10.2 Factor Migration
Consider a change in the domestic supply of a factor of production as a
result of migration. This will directly affect production activity through the
change in the total factor supplies available for production. It will also affect
the (endogenous) price of the nontraded good, with consequent further
changes in output and induced changes in factor rewards.
Differentiating equations ( l ) , ( 2) , and ( 3 ) reveals the way in which the equi-
librium is disturbed by small changes in commodity prices:
e l l GI + e2]G2 + e31G3 = pI,
e12Gl + e22G2 + e32G3 = b2,
e13Gl + ez3G2 + e33G3 = p3,
where O,, =a,w,lp,, the distributive share of factor i in industry j , and a “hat”
(*) over a variable denotes a relative change (e.g., ~ = dw/w). Similarly, by
289 Immigration and Wages
differentiating equations (4), ( 5) , and ( 6) , the response to changes in factor
endowments can be determined:
(4’)
(5’)
(6’)
A,,rZ, + A$, + AIg3 = C, - {at GI + a: Gz + a: G3},
A,$, + AzzX2 + AZg3 = Qz - {u; GI + U: Gz + G3},
A31Xl + A3zrZz + A3g3 = QI - {a: GI + U: G, + U: G3},
where A, = a,Xj /Vi, the fraction of the total supply of factor i used in the jth
industry, and @ denotes the economy-wide substitution toward or away from
the use of factor i when factor k becomes more expensive, under the assump-
tion that each industry’s output is held constant. That is, a: =Ej A, Ei , where
Ei is the elasticity of demand for factor i with respect to w, in industry j ,
holding output and other factor prices constant.
Were commodity prices to remain unchanged after the factor movement (as
would occur if all goods were traded at exogenously given world prices),
then, from equations (1’), (2’), and (3’), factor earnings would also remain
constant, and hence the bracketed terms of equations (4’), ( 5’ ) , and (6’) would
all be zero, as there would be no substitution between factors in production.
Output change in response to changes in factor endowments would be influ-
enced only by the relative intensities with which factors are used in each of
the three industries. This behavior results from a higher dimensional analogue
of the familiar Rybczynski theorem (which was derived for a model with two
factors and two goods), and we shall call it a “pure Rybczynski effect.”
Good 3 is, however, not traded internationally, and the inflow of factors will
induce changes in both the demand for and the supply of that good. Differen-
tiating equation (8), and using equations (9) and (10) to determine the change
in utility resulting from factor immigration at constant commodity prices,
where p is the marginal utility of money income, and I is national income:
Rewriting equation (1 1) in terms of relative changes yields
( 1 1 ‘ )
0 = o x 8’ Qi,
where w = pl/U, and 8‘ is the share of factor i in national income,
290 Peter Kuhn and Ian Wooton
By the appropriate choice of utility scale, let o = 1 locally. Then equation
( 1 1 ’) becomes
(12)
0 = 8lV1 1- e2V2 + e3V3.
Differentiating the market-clearing condition, equation (7), to determine the
equilibrium responses to disturbances in the market for good 3,
(13)
where uI , v2, and v3 are the compensated price elasticities of demand for X, ,
and IJ. is the income elasticity of demand for X,. It will be convenient to re-
write (13), using (12), as
x, = v, p, + v d2 + v3p3 + KO,
(13’)
+ .. - phJ3 > 0.
p, = p,p, + pd2 + p3x3 + +{elV, + e2V2 + e3P3},
where PI ” -vlIv3, p, ’’ -v2/v, , p3 “ l/u3 < 0, and (if good 3 is normal)
Rewriting the output-response equations (4’)-(6’) in matrix form,
’I1 ’I2 ’13 u; 0; a;
[ ’21 ’22 ’231 f!] = [
’31 ‘32 ’33 - [ 2 2 $1 [
Solving this for X, ,
(14)
where
(15)
for k = 1, 2, 3; Sk = IAtl/lhl, for k = 1, 2, 3; \A1 is the determinant of the
lambda matrix
x3 = s,Q, + S2V2 + s3Q3 + ylGl + y2G2 + y3G3,
yk = -{s,a; + 6,a; + S,a:},
‘I1 ’I2 ‘13
’31 ’32 ’33
and
from the definitions of the “extreme” and “middle” factors. Note that the pure
Rybczynski effect (i.e., when factor prices are constant) of factor endowment
changes on output of good 3 is then
Substituting equations (3’) and (14) into (13), and solving, yields
291 Immigration and Wages
(17)
where
(18)
fori = 1, 2, 3; and
(19)
for i = 1, 2, 3. The comparative statics of factor rewards in this economy are
now completely determined by equations (1 '), (2'), and (17):
AIGl + A2G2 + A3G3 = PI@, + Pd2 + Zlpl + Z2p2 + Z3p3,
A, = (0,3 - P3YJ9
z, = P,@, - CLO'),
which is easily solved for GI , G2, and G3.
Since, in this paper, we focus on the effects of factor endowments on factor
prices, we henceforth set @ = @ = 0 and solve (20) for Gk in terms of pi, for
i , k = 1, 2, 3. The induced change in factor rewards when factor endowments
change is then given by
for i, k = 1, 2, 3,
The properties of equation (21) are analyzed in the following section.
10.3 Consequences for Factor Rewards
The consequences of (21) for the effects of factor endowment changes on
both nominal factor prices and the welfare of factor owners are analyzed in
turn below. We begin with properties of price changes that are independent of
a certain "normality" condition and then consider the additional restrictions
imposed by that condition. Finally, welfare effects are considered.
10.3.1 General Results
To develop an intuition for the effects of endowment changes on factor re-
wards, consider the effect of a change in the endowment of factor i on the
rewards paid to factor k in equation (21). The Zj term gives the effect of im-
292 Peter Kuhn and Ian Wooton
migration (pi > 0) or emigration (pi < 0) on the price of the nontraded good
X,. Recalling equation (19),
(19) z, = p3(s, - pet).
The first term in the parentheses is the supply effect of the change in output of
good X, resulting from the immigration of factor i, that is, the pure Rybczyn-
ski effect of equation (16). The second term is a demand effect, reflecting the
increased demand for good X, resulting from a higher level of national income
created by the increase in the economy’s endowment. We call the difference
between the terms the “modified Rybczynski effect.” Should this be positive,
then an excess supply of the nontraded good has been induced by the migra-
tion, triggering a fall in its price, p3 (because 6, < 0), and vice versa.
The other component of the effect of pi on w, in (21) is the ratio of the two
determinants. Changes in commodity prices in tbe traditional two-factor, two-
good trade model induce changes in factor prices according to the Stolper-
Samuelson Theorem. Were all three commodities in our model traded inter-
nationally, then the response of factor prices to a change in p , would be
a higher dimensional analogue to these familiar magnification effects. We
shall call it the “pure Stolper-Samuelson effect.” The magnitude of this re-
sponse can be measured by solving equations ( l ’ t ( 3’ ) (letting$, = ~3, = 0),
yielding
fork = 1, 2, 3, where 101is the determinant of the theta matrix,
‘11 ‘21 ‘31
[ I ‘I3 ‘23 ‘33
8 = ‘12 ‘22 ‘32 .
Good 3 is not traded, however, and faces a less than infinitely elastic demand.
As a result, changes in factor earnings have repercussions on the amount of X,
supplied, through within-sector substitution among factors, and this will in-
duce a further change in the commodity price. This effect is captured by the
P3y, terms that modify the denominator from 181, in the pure Stolper-
Samuelson effect of equation (22), to A in equation (21). We therefore call the
ratio of determinants in (22) a “modified Stolper-Samuelson effect” of p , on
w,. It can be shown, for any negative semidefinite economy-wide substitution
matrix, that
(23) sign(A) = sign((81),
and so the modified Stolper-Samuelson effect will be qualitatively identical to
the pure Stolper-Samuelson effect.2
The entire effect of a change in the endowment of factor i on the earnings
293 Immigration and Wages
of factor k is thus simply the product of the modified Rybczynski effect of V,
on p , and the modified Stolper-Samuelson effect of p , on w,. This interpreta-
tion and decomposition of (21) emphasizes a major feature of the model that
is independent of any assumptions regarding the structure of demand and
factor-intensity rankings and is true for changes in the rewards to all factors.
The only way that factor endowment changes can affect factor prices is
through inducing changes in price of the nontraded good, p, ; that is, factor
prices cannot change unless p , changes. Otherwise, were p, fixed, because
either the good was traded internationally or domestic demand was infinitely
elastic (p, = 0) , then factor prices would be uniquely determined by equa-
tions (1)-(3), independently of the endowment.,
10.3.2 The Normality Condition
Consider more closely the excess supply of X, induced by immigration of
factor i (6, - pel). The level of national income will always rise with factor
inflows, and, if this is compounded by a negative Rybczynski effect (6, < 0)
lowering the supply of X,, then there will undoubtedly be excess demand for
the nontraded good, inducing an increase in its price. However, suppose im-
migration of factor V, induces an expansion of X, production because 6, > 0.
The change in factor endowment has then raised both the demand for and the
supply of the nontradable. Thus, the potential exists for what we shall call a
“perverse demand” result that, with a sufficiently high national income elastic-
ity of demand for the nontraded good, an increase in the output of good 3
induced by a change in endowment will be accompanied by an increase in its
price.
In this subsection, we develop a pair of sufficient conditions, together called
the “normality condition,” that rule this out. One of these is a fairly weak
restriction on the structure of demand, while the other is a constraint on sup-
ply effects that, as we shall see, is clearly satisfied by our empirical evidence
on factor intensities. Together they ensure that
(24) sign(6, - kel) = sign@,);
thus, the modified Rybczynski effect, like the Stolper-Samuelson effect, is
qualitatively the same as the “pure” effect.
First, rewrite the expression for 8, in equation (16) in terms of distributive
shares:
where 0, = pj Xj / I, the share of good j in national expenditure. From our num-
bering of factors in terms of relative intensity of their use in the traded sector,
we know that
(26) sign(+,) = sign(+,) = -sign(+J.
294 Peter Kuhn and Ian Wooton
Thus, immigration of either of the two “extreme” factors will have opposite
effects to immigration of the “middle” factor on the production of the non-
traded good.
(27) sign(6,) = signlo(,
since l0,l > 0. A sufficient condition for 6, > 0 is then that
Furthermore, it is helpful to note that
‘ 32 ‘12 ‘22 ‘31 ‘I1 ‘21
‘ 33 ‘13’ ‘23 ‘33 ‘13’ ‘23
- <- - or - <- -.
In other words, an increase in the supply of the middle factor in the traded
sector ( V, ) raises the supply of the nontraded good if, relative to either one of
the traded goods, nontraded goods are extremely intensive in V, . Conversely,
if nontraded goods are extremely unintensive in V, relative to either X , or X, ,
the supply of X , will fall when V3 rises. Thus, through (27) and (26), the signs
of all the Rybczynski effects hinge on whether the nontraded sector uses the
middle “traded” factor intensively or unintensively.
Now assume that the two traded goods, XI and X, , are not, as a bundle,
inferior in demand, in the sense that the total expenditure on the two goods
( p , X , + p, X, ) does not decline when national income increases, at constant
prices. Differentiating the national income identity yields
i = el xl + 0, x, + 03x, .
i - 0, x, > 0.
XI and X , then are not inferior as a bundle if and only if
Noting that p = X3/f, this may be rewritten as
1 > k0,.
Multiplying by 0‘ and rewriting yields
Thus, the noninferiority of goods 1 and 2 ensures that the demand term mod-
ifying the Rybczynski effect is less than the wage bill of the immigrant factor
relative to the value of output of X, . We can therefore rule out the perverse
demand result if, when 6, > 0, 6, > 0Y0,. Using equation (25), this may be
rewritten to state:
sign 10.1
= sign
101
Suppose that factor 3 has a positive Rybczynski effect, that is, S3 > 0. As
are nega- noted earlier, this means that 101is positive. As both l0,l and
295 Immigration and Wages
tive, we are assured that l9,l exceeds 191,and, hence, the perverse demand
result cannot occur if goods 1 and 2 are noninferior.
If, instead, immigration of factor 3 directly reduces the production of the
nontraded good, that is, 6, < 0, then the potential for a perverse demand re-
sult arises with immigration of either factor 1 or factor 2. Now, as 191< 0 and
only I@,[ > 0, it is not possible to rule out l9,l/l9l<l f or j = 1 or 2 without
recourse to the data. Our normality condition therefore in general requires
both 11.8, < 1, which ensures a small enough effect of V, on the demand for
good 3, and 19Jl/191 > 1 for 191< 0 andj = 1 or 2, which ensures a large
enough supply effect in the cases where the nontraded sector is extremely
unintensive in the middle “traded” factor. The latter condition is clearly satis-
fied by the data on factor intensities that we consider below.
10.3.3 Effects of Factor Endowments on Factor Prices under the
Normality Condition
Under the normality condition, modified Rybczynski effects are qualita-
tively the same as the pure effects, as is also true for modified Stolper-
Samuelson effects. Furthermore, it is straightforward to show that the
Rybczynski effect of Vt on X , is always the same sign as the Stolper-
Samuelson effect of p , on wi. This can be seen by manipulating (25) and sub-
stituting into (23), yielding
(30)
19.1
sign(6,) = sign [TI.
Substituting (24) and (30) into (21), and noting that p, < 0, we have
(31) = sign( - s,s,~J,
for i, k = 1 , 2, 3. Irrespective of whether 191S 0, it is clear from (31) that
any factor is its own “enemy,” in that immigration of a factor will always
lower its own nominal wage, while emigration raises its earnings. Further, as
w, and w, have the same sign, the extreme factors, 1 and 2, will be mutual
enemies while being friends with the middle factor, 3. The qualitative effects
of factor quantities on factor prices thus depend only on which factors are
extreme in the traded sector of the economy, not on the relative factor intensi-
ties of the traded versus nontraded sector. The reason is that, whenever a fac-
tor supply shift raises the price of nontraded goods (by lowering supply),
higher nontraded goods prices lower the equilibrium price of that factor, and
conversely.
10. 3. 4 Welfare Effects
The effect of the above nominal changes in returns on the welfare of a factor
depends on the induced changes in p , ( p , and p , being fixed, by assumption).
Assessment of welfare effects is, however, made easy if we note that, by sub-
stituting (22) into (21),
296 Peter Kuhn and Ian Wooton
p3 = p, - 101(6, - kf?)+,.
A
Since the ratio of determinants is always positive, the general equilibrium
effects of V, on p3 are always the same sign as when factor prices are held
fixed. In more detail, immigration of factor Vi whose 6, > 0 will result in a
fall in p, , raising the real income of those factors whose nominal earnings
have increased. The other factors are faced with both reduced nominal earn-
ings and lower prices, with apparently ambiguous consequences for their wel-
fare. However, consider the relative change in the ratio of earnings of factor k
to the price of good 3,
(33)
[3 = w - p,
Substitute equations (21) and (32) into this expression and rearrange to find
(34)
t] = 8 3 [m 1% - 1).
Using (29), (31), and (32), equation (34) implies that sign [6, 1p3] =
sign 6,. Thus, under the normality condition, changes in nominal earnings
will outweigh changes in the price of the nontraded good, ensuring that
changes in welfare will have the same signs as the changes in factors’ nominal
returns.
Table 10.1 gives a complete listing of the predicted responses of variables
to changes in migration.
10.4 Empirical Estimates of Factor Intensities
Our estimates of U.S. factor intensities are based on an analysis of 430
four-digit manufacturing industries during the period 1960-84. These indus-
Table 10.1 Predicted Effects of Factor Endowment Changes under the
“Normality” Condition
Endowment Change
297 Immigration and Wages
tries constitute all the industries in the NBER Data Set on Trade, Immigration,
and Foreign Direct Investments (450 in all) with, data on both exports and
imports.6 This data set has the advantages of being at a finer level of aggrega-
tion than most existing empirical studies of factor intensities and trade and of
being available for a long series of years, as recently as 1984.’ It has the
disadvantage of being limited to manufacturing industries only. Thus, we
view our results here as (a) likely to be quite accurate for the manufacturing
sector relative to other studies but (b) relevant to the whole economy only to
the extent that the pattern we identify in manufacturing generalizes to other
industries.
Our basic procedure begins by allocating the total value added in each four-
digit industry into payments to one of the three factors in the model. Total
income of “unskilled” workers was defined as the industry’s total production-
worker payroll for the year. Income of “skilled” workers is the difference be-
tween this and the industry’s total payroll.8 Finally, subtracting total payroll
from total value added yielded the payments to “capital” in the industry. Thus,
in keeping with our model’s assumptions, we ignore indirect factor inputs and
assume no intermediate inputs in produ~t i on. ~
Next, the 430 four-digit industries were aggregated into three as follows.
First, nontraded goods were defined as those industries whose level of “open-
ness” to international trade, measured by the sum of exports plus imports
divided by total industry output, was under k%, where k took on alternative
values of 0, 5, 10, and 20. Then the remaining, traded goods were divided
into import-competing goods and exports, depending on whether the net im-
ports of the industry (imports less exports) were positive or negative, respec-
tively. This yielded 3 X 3 table of factor incomes, from which it was then
straightforward to calculate the matrix of factor intensities, 0.
Table 10.2 shows the factor intensities calculated in this manner for the year
1984 and with k = 10%. This criterion yielded a nontraded sector producing
$351 billion of value added, which is 36.6% of the total value added of $959
Table 10.2 Estimated Factor Intensities for U.S. Manufacturing, 1984
V,, Unskilled Labor V,, Skilled Labor V, , Capital
1. “Absolute” factor intensities (9J:
a. XI (import competing) ,2615 ,1504 ,5881
b. X, (exporting) ,2092 ,1858 ,605 1
c. X, (nontraded) ,2205 .I408 ,6387
a. 9,,/9,, (imports vs. exports) 1.2502 .SO96 ,9719
b. 8,,/9,, (imports vs. nontradeds) 1.1861 1.0681 ,9208
2. Relative factor intensities (9,/9,J:
c. 9,,/9,, (exports vs. nontradeds) .9487 1.3192 ,9473
Sources: NBER Data Set on Trade, Immigration, and Foreign Direct Capital Investments.
Note: Nontraded sector is defined as industries with (exports + imports)/output under 10%.
Determinant of 9 equals ,0019528.
298 Peter Kuhn and Ian Wooton
billion in the sample. Import-competing and exporting industries, respec-
tively, produced $361 billion (37.6%) and $247 billion (25.7%) of value
added. A list of industries producing more than $10 billion of value added, by
trade category, is given in table 10.3.
Row 2a of table 10.2 clearly indicates that, within the traded sector, un-
skilled labor is extremely intensive in imports and skilled labor is extreme in
exports, with capital as the middle factor. This accords well with the widely
held notion that the United States is well endowed with skilled labor and
poorly endowed with unskilled labor, relative to the rest of the world (e.g.,
Baldwin 1971).1° The list of industries in table 10.3 supports this notion, with
what are commonly considered “low-technology’’ industries such as paper and
steel appearing only in the import category and “high-technology’’ industries
like computers dominating in exports.
Finally, row 2b of table 10.2 indicates that the middle “traded” factor, V, ,
is used extremely intensively in the nontraded sector relative to imports.
Thus, in a well-defined sense, the nontraded sector is intensive in capital and
unintensive in both types of labor, relative to the traded sector. By previous
results, this means that 101 > 0 and that the “perverse demand” result cannot
occur as long as XI and X , are normal goods (either individually or as a
bundle). Thus, in combination with the results in table 10.2, our model pre-
dicts that immigration of either type of labor will, in the long run, hurt do-
Table 10.3 Four-Digit Industries with over $10 Billion of Value Added, by Wade
Category, 1984
1. Exports:
2869
3573
3721
2. Imports:
2621
291 1
3312
3662
3674
3679
371 1
3714
3861
Industrial organic chemicals, n.e.c
Electronic computing equipment
Aircraft
Paper mills
Petroleum refining
Blast furnaces
Radio and TV transmitting, signaling and detection equipment
Semiconductors and related devices
Electronic components, n.e.c.
Motor vehicles and passenger car bodies
Motor vehicle parts and accessories
Photographic equipment and supplies
3. Nontraded goods:
271 1 Newspapers, publishing
2752 Commercial printing, lithographic
2834 Pharmaceutical products
3079 Miscellaneous plastics products
3761 Guided missiles and space vehicles
Note: Nontraded goods deflned as (exports + imports)/output < 10%. N.e.c. means “not else-
where classified.”
299 Immigration and Wages
mestic owners of both types of labor. The reason is that immigration raises the
price of nontraded goods, which in turn benefits owners of capital. This higher
return to capital must be compensated for by a reduction in wages of both
types of workers if U.S. exports and import-competing industries are to re-
main internationally competitive.
How robust are these conclusions to changes in the definition of traded
versus nontraded sectors and to changes in factor intensities over time in the
U.S. economy? This question is explored in table 10.4, which summarizes
our estimates of factor intensities for various levels of k and for other years.
What is most striking about this table is that, for every year but 1960 and for
every level of k, the factor intensity rankings in the traded sector are the same
as in table 10.2. Also, (81 is the same sign, indicating that the relative factor
intensities of the traded versus nontraded sectors are unchanged and that nor-
mality of X , and X, in demand continues to be sufficient to rule out perverse
factor-demand results. Only in 1960 are the results mixed; we feel that this is
connected with changes in the U. S. economy over time that produced “Leon-
tief paradox”-type results in earlier years but no longer do so, as the United
States specializes more and more in knowledge-intensive industries. We con-
clude that our estimates of direct relative factor intensities (and the resulting
predictions about the effects of immigration) reliably summarize a broad
underlying pattern that has persisted in manufacturing since 1970 or even
somewhat earlier.
Table 10.4 Factor-Intensity Rankings for Various Definitions of the Nontraded
Sector, 1960, 1970, 1980, and 1984
Maximum Level of (Exports + Imports)/
Output in the Nontraded Sector
.o .05 .I0 .20
1960:
Extreme factors in imports and exports,
respectively (within traded sector)
Sign I 0 I
Extreme factors in imports and exports,
respectively (within traded sector)
Sign I 0 I
Extreme factors in imports and exports,
respectively (within traded sector)
Sign I e I
Extreme factors in imports and exports,
respectively (within traded sector)
Sign 10 I
1970:
1980:
1984:
u, s u, s
NA +
K , S
+
u, s
+
u, s
+
Note: K = capital, S = “skilled (nonproduction) labor, and U = “unskilled” (production) labor.
300 Peter Kuhn and Ian Wooton
How robust are our estimates to the inclusion of nonmanufacturing indus-
tries in the analysis? This question cannot be answered with the NBER data
but is explored using two-digit industry data from another source in the Ap-
pendix. Interestingly, this analysis indicates that the basic results do general-
ize. The basic reason for this is that services (which constitute the bulk of the
excluded, nonmanufacturing industries), contrary to widespread beliefs, are
not always nontraded and not always labor intensive. For example, real estate
(a very important nontraded service) is highly capital intensive, while educa-
tion (which is very intensive in skilled labor) is a substantial U. S. net export.
Still, owing to the considerable practical and conceptual problems involved in
estimating service trade (see the data sources listed in the Appendix), we view
this finding with considerable caution. Clearly, there is an outstanding need
for careful empirical analysis based on better-quality trade data for service
industries. The consistency of our results across definitions of sectors and
years is, however, suggestive and proves a useful illustration of the ease with
which our theoretical model can be implemented.
10.5 Conclusion
This paper has developed a simple, general equilibrium model of how fac-
tor endowments affect factor prices when a subset of the goods produced in
the economy is traded at internationally fixed prices. The result is a model that
makes unambiguous predictions that are independent of estimated elasticities
of substitution among factors in production.
Our empirical analysis, based on 430 four-digit manufacturing industries in
the years 1960, 1970, 1980, and 1984, indicated that, at least since 1970,
factor intensities in U.S. manufacturing follow a very consistent pattern:
skilled labor is used extremely intensively in exports, while unskilled labor is
extremely intensive in import-competing industries. Furthermore, the middle
factor in the traded sector, capital, is used intensively in the nontraded sector
relative to the traded sector. Thus, to the extent that the relative factor inten-
sities we find in manufacturing generalize to the whole economy, our model
predicts the following. Increased immigration of either skilled or unskilled
workers to the United States will, in the long run, hurt U.S. workers of both
types and benefit owners of capital. These effects should be associated with
an increase in nontraded goods prices, which, by reducing the international
competitiveness of the country’s traded goods, causes the reduction in wages.
They are also independent of the technical substitution elasticities in produc-
tion that so many analysts have attempted to estimate (for a recent summary,
see Hamermesh and Grant 1979).
Our model can, of course, be extended and improved on in various ways.
These include incorporating intermediate inputs (as is done in empirical work
by Baldwin 1971; and Stem and Maskus 1981), allowing for capital mobility
(as in Gerking and Mutti 1983), considering shorter run effects (as in Rivera-
301 Immigration and Wages
Batiz 1987), relaxing the extreme assumption that U.S. consumption and
production have no effect on the prices of traded goods, and considering non-
balanced trade. In addition, better data, which consider nonmanufacturing in-
dustries as well, may also be available in the relatively near future as more
trade statistics on services are collected. All these extensions could, of course,
change our specific predictions about the directions of factor price changes
here, which we view as suggestive but tentative. It is clear that they will not,
however, change what we view as the fundamental lesson of this paper. This
lesson is that the effects of immigration in a partially open economy may be
determined by a fundamentally different set of factors than in a closed econ-
omy, where technical substitutability between factors in production plays the
key role. In the partially open economy, factor prices are constrained to
change in a way that preserves the international competitiveness of its traded
goods, as long as those goods continue to be produced. This places tight re-
strictions on the kinds of factor-price changes that can occur. These restric-
tions deserve, we feel, greater prominence in the work of empirical and
policy-oriented researchers studying the effects of immigration to the United
States.
Appendix
Factor Intensity Estimates or the Entire US.
Economy from Two-Digit f ndustry Data, 1983
This appendix explores the generalizability of our results based on manufac-
turing only to the entire U.S. economy by constructing a factor-intensity
matrix for the entire U.S. economy from two-digit data for 1983. These data
were obtained from the following sources: total value added and compensa-
tion of employees from the Survey of Current Business (66 [July 19861: 1986,
table 6); merchandise trade data from the 1985 Statistical Abstract of the
United States (tables 1448, 1449); and rough estimates of service trade from
the Office of Technology Assessment (1986, table 2, using midpoints of inter-
vals). Thus, some services are classified as exports (e.g., education), others
as imports (e.g., insurance), and still others as nontraded (e.g., retailing).
Skilled workers were defined as those who completed high school; the per-
centage of the work force that was skilled was then taken from a 1980 Census
tabulation of occupation by industry (U.S. Department of Commerce series
PC-80-2-70, pp. 1-80) and their relative wage rates in 1983 from U.S. De-
partment of Commerce series Money Income of Husbands, Families and Per-
sons in the United States (P-10, no. 146, table 48). Payments to land as an
input are included by definition with capital, which functions as a kind of
“residual” factor here. Because of the small number of industries. we chose
302 Peter Kuhn and Ian Wooton
criteria of openness t o trade and a critical balance-of-trade level that yielded
nontraded, exporting, and importing sectors of roughly equal sizes (this
meant that some industries with low trade deficits were classed as exports).
This procedure yielded the following factor-intensity matrix, 8:
where the three factors are, respectively, unskilled labor, skilled labor, and
capital, as in table 10.2. Noting that, as in table 10. 2, the nontraded sector is
again relatively intensive in capital, we conclude that the factor-intensity
rankings we find in manufacturing may well generalize completely to the en-
tire U. S. economy when better data on service trade are available.
Notes
1. Since this is an undistorted economy, we know that it maximizes a weighted sum
of the utilities of its members subject to the balanced trade and resource constraints.
Under certain conditions (see, e.g., Gorman 1953), this is equivalent to maximizing a
community utility function of the type used here. We assume that such conditions are
satisfied here, which does not necessarily imply that all agents (including immigrants)
are equally endowed. This latter condition would of course make the analysis of im-
migration much less interesting.
2. To see this, note that the denominator can be written as
Through algebraic manipulation, the second term can be reduced to [p3/(0,10/)] u’Tu,
where T is a negative semidefinite matrix derived from the economy-wide substitution
matrix (see Jones and Scheinkman 1977), and u is a column vector. The product U‘TU
is nonpositive, while p, is negative, and 8, is positive. By simple manipulation it
follows that lel/A is always positive.
3. It is assumed that endowment changes do not move the economy out of the “cone
of diversification.”
4. The national income elasticity of demand equals the national utility elasticity of
demand since, locally, units of income and utility are equivalent. Also, it is easily
shown that, in the “perverse demand” case, the demand curve for factor V, in general
equilibrium is upward sloping. While this might be a fortunate situation for owners of
V, , it appears quite unlikely given the fairly weak sufficient conditions needed to rule it
out.
5. In the standard 2 X 2 model of international trade, the Rybczynski effect of the
immigration of one factor resulted in the output of one industry increasing while pro-
duction in the other industry diminished. Thus, the value of output of the expanding
303 Immigration and Wages
industry had to increase by a greater amount than national income. If both of the goods
were normal, then the increase in the value of production for the expanding industry
would then necessarily be greater than the increase in the value demanded of the good,
ensuring a nonperverse demand result. Our results show that, when there are three
goods and three factors, nonperversity is not guaranteed. To see this, suppose that
immigration of both factors 1 and 2 increases output of X, , through the Rybczynski
effect. Even with factor intensities unchanged (as product prices are constant), this
need not result in a fall in the output of both of the traded goods. Indeed, the increase
in the value of production of X , might be quite small relative to the increase in national
income when output of one of the traded goods also increases. Thus, the potential does
exist that the expansion in output of good 3 may be less than the increase in its demand,
even when goods 1 and 2 are jointly noninferior in consumption.
6. The factor income information in the NBER data set was taken from the Annual
Survey ofManufactures for various years. This could be matched with trade figures
from the Trading Monitoring System of the Bureau of Labor Statistics for 430 of the
450 four-digit industries. The data were collected and made available to us by John
Abowd and Richard Freeman.
7. See table 4.2 in DeardorfF (1984). The finest level of detail used in the studies
cited there is in Stem and Maskus (1981), who use 128 three-digit industries. Our
sample of 430 four-digit industries contrasts very favorably with all three studies.
8. An alternative, and perhaps preferable, definition of skilled vs. unskilled work-
ers might be based on total years of education. Unfortunately, information on years of
education by industry is not available in this data set. Our experiments using high
school completion rates with two-digit data, reported in the Appendix, lead us to ex-
pect that this would not change the results.
9. Ideally, our theoretical model should incorporate (both traded and nontraded)
intermediate inputs and specify how these should fit into the empirical analysis. Since
it appears that this extension of our model would significantly complicate our analysis,
it is not undertaken here. We feel that it is, however, an important area for further
research on this topic.
10. This notion has often been advanced as a possible explanation of the “Leontief
paradox” of capital-intensive U.S. imports (Leontief 1953). Interestingly, this paradox
does not arise in table 10.2 here (the combined labor shares in imports exceed those in
exports, and the capital share is greater in exports than imports). Three possible rea-
sons for this are the fact that we use four-digit, not two-digit, data, the fact that our
focus on manufacturing excludes natural resource industries, and the later time period.
Of these, the “time” explanation seems to be most convincing, for the following rea-
son: for the four different values of k considered in the experiments of table 10.3, the
capital share in exports exceeded the capital share in imports three out of four times in
1984, four times in 1980, twice in 1970, and zero times in 1960. This seems fairly
strong evidence of an increasing relative capital intensity of U.S. exports over time.
References
Baldwin, R. E. 1971. Determinants of the Commodity Structure of U.S. Trade. Amer-
ican Economic Review 61: 126-46.
Batra, R. N., and F. R. Casas. 1976. A Synthesis of the Heckscher-Ohlin and the
Neoclassical Models of International Trade. Journal of International Economics
6:21-38.
304 Peter Kuhn and Ian Wooton
Brecher, R. A., and E. U. Choudhri. 1982. The Factor Content of International Trade
without Factor-Price Equalization. Journal of International Economics 12:277-84.
Chang, W. W. 1979. Some Theorems of Trade and General Equilibrium with Many
Goods and Factors. Econometrica 47:707-26.
Deardorf€, A. 1984. Testing Trade Theories and Predicting Trade Flows. In Handbook
of International Economics, vol. 1, ed. R. W. Jones and P. B. Kenen. Amsterdam:
Fallon, P. R., and P. R. G. Layard. 1975. Capital-Skill Complementarity, Income
Distribution, and Output Accounting. Journal of Political Economy 83:279-302.
Gerking, S., and J. Mutti. 1983. Factor Rewards and the International Migration of
Unskilled Labor: A Model with Capital Mobility. Journal of International Econom-
ics 14:367-80.
North-Holland.
Gorman, W. 1953. Community Preference Fields. Econometrica 21:63-80.
Hamermesh, D. A., and J. Grant. 1979. Econometric Studies of Labor-Labor Substi-
tution and Their Implications for Policy. Journal of Human Resources 145 18-42.
Hicks, J. R. 1932. The Theory ofWages. New York: Macmillan.
Jones, R. W., and S. T. Easton. 1983. Factor Intensities and Factor Substitution in
General Equilibrium. Journal of International Economics 15:65-99.
Jones, R. W., and J. A. Scheinkman. 1977. The Relevance of the Two-Sector Produc-
tion Model in Trade Theory. Journal of Political Economy 85:909-36.
Kemp, M. C. 1966. Gain from International Trade and Investment. American Eco-
nomic Review 56:788-809.
Leontief, W. 1953. Domestic Production and Foreign Trade: The American Capital
Position Re-examined. Proceedings of the American Philosophical Society 97:
332-49.
Markusen, J., and J. Melvin. 1979. Tariffs, Capital Mobility, and Foreign Ownership.
Journal of International Economics 9:395-410.
Office of Technology Assessment. Congressional Board of the Ninety-ninth Congress.
1986. Trade in Services: Exports and Foreign Revenues, Summary. Washington,
D.C.: U.S. Government Printing Office.
Rivera-Batiz, F. 1982. Nontraded Goods and the Pure Theory of International Trade
with Equal Numbers of Goods and Factors. International Economic Review 23:
. 1987. Modeling the Short-Run Economic Effects of Immigration: Some Gen-
eral Equilibrium Simulations. In Modeling andSimulation, Vol. 17, ed. W. G. Vogt
and M. H. Mickle. Research Triangle Park, N.C.: Instrument Society of America.
Ruffin, R. J. 1981. Trade and Factor Movements with Three Factors and Two Goods.
Economic Letters 7: 177-82.
Samuelson, P. A. 1948. International Trade and the Equalization of Factor Prices.
Economic Journal 58: 163-84.
Stem, R. M., and K. E. Maskus. 1981. Determinants of the Structure of US . Foreign
Trade, 1958-1976. Journal of International Economics 11 ~207-24.
401-9.
11 Immigrants, Labor Market
Pressures, and the Composition
of the Aggregate Demand
Susan M. Collins
The purpose of this paper is to examine the effect of changes in the composi-
tion of aggregate demand on total labor requirements and on the requirements
for jobs typically held by immigrants, using the input-output (10) tables for
the United States. The paper asks two sets of questions. First, how are labor
requirements affected by a dramatic turnaround in the trade balance? Does it
matter whether the deficits are accompanied by an investment or a consump-
tion boom, and how are the resulting labor market pressures distributed across
industries? Second, how are immigrants distributed across domestic indus-
tries, are they differentially affected by shifts in the composition of demand,
and are the recent changes likely to have made immigrant workers more “vis-
ible,” providing one explanation for the increased attention they have received
in recent years?’
The basic approach is as follows. Changes in the composition of aggregate
demand will alter the distribution of labor requirements across sectors and
industries. While these shifts do not imply changes in labor demand or in
actual employment, they can be interpreted as indicating labor market pres-
sures in those sectors where demand has decreased. Because immigrants and
native workers are distributed quite differently across jobs, these pressures
will influence the two groups differently.*
Input-output analysis provide a useful framework to explore the linkages
between aggregate demand, labor market pressures, and immigrants because
it integrates both microeconomic and macroeconomic aspects. On the micro-
economic side, it considers the output and employment responses of particular
industries. On the macroeconomic side, it incorporates the key identity from
Susan M. Collins is associate professor of economics at Harvard University and a faculty re-
The author would like to thank 1. Abowd, H. Bowen, R. Freeman, and K. Lang for comments
search fellow of the National Bureau of Economic Research.
and A. Revenga for excellent research assistance.
306 Susan M. Collins
the National Income and Product Accounts, which highlights the linkages be-
tween net exports and the other components of aggregate demand. The coun-
terpart to an external imbalance (i.e., a deficit in U.S. goods and services vis
ii vis the rest of the world) must be an excess of investment over domestic
savings.
The macroeconomic focus is important because it points to a different set
of issues and conclusions than many of the industry studies. For example,
suppose that a researcher concluded that imports had been a “cause of injury”
in a particular industry and that restricting imports would be likely to raise
domestic output and employment. From the macroeconomic perspective, un-
less the policies to restrict imports were expected to reduce the savings invest-
ment imbalance, thereby reducing the total trade deficit, these policies merely
shift the trade deficit between sectors.
There are also some drawbacks to the I 0 analysis. By maintaining constant
input-output coefficients and fixing the commodity composition of each com-
ponent of aggregate demand, it rules out substitution on both the production
and the consumption sides. A related point is that it does not specify why
aggregate demand changes and how relative prices (including interest rates
and exchange rates) are affected. The answers to these questions will in turn
have implications for the composition of imports, consumption, and the other
components of demand. In order to incorporate these factors, it would be nec-
essary to imbed the I 0 framework into a macroeconomic model, which is
beyond the scope of the current paper.
The paper is composed of four remaining sections. Section 1 1.1 asks where
the immigrants are and examines the distribution of immigrant workers across
sectors. Section 11.2 turns to the key macroeconomic issues and discusses
changes in the composition of aggregate demand. Section 11.3 analyzes the
effect of shifts in the composition of aggregate demand on labor requirements
by sector and for immigrants and nonimmigrants. The section first spells out
the methodology and then discusses results. Concluding remarks are given in
the final section.
11.1 Where Are the Immigrants?
A number of authors have pointed out that immigrant workers tend to be
concentrated in different industries than native workers. In particular, immi-
grants tend to enter the labor market in low-wage, relatively unskilled posi-
tions, but the distribution of immigrants over industries and occupations be-
comes more similar to that of natives the longer they remain in the United
state^.^
Table 11.1 compares the 1982 employment distribution of foreign-born
(immigrants plus refugees) and native workers across sectors. These data are
derived from a special matched sample of respondents to supplementary CPS
surveys conducted in March and April 1983. Unfortunately, only about 75%
307 Immigrants, Labor Market Pressures, and the Aggregate Demand
Table 11.1 Employment Distribution by Sector
% of Total Group Employment
Foreign as % of Total
Total Foreign Native Sector Employment
Agriculture
Mining
Construction
Manufacturing
Transportation and
public utilities
Wholesale trade
Retail trade
Finance, insurance, real
estate
Private household
Other service
Public administration
Total (thousands)
3.4
.9
5.9
19.6
1.0
4.3
16.3
6.2
1.3
30.3
4.7
109,064
3.6 3.4
.I 1 .o
4.7 6.0
25.3 19.2
4.2 1.2
3.9 4.4
16.5 16.3
6.6 6.2
1.8 1.2
30.5 30.3
2.4 4.9
8,694 100,370
8.4
5.7
6.4
10.2
4.8
7.1
8.1
8.4
11.5
8.0
4.1
7.97
Source: Sehgal(1985).
of the April sample matched with the March sample, and no corrections were
made for missing value^.^
The first three columns of the table report the shares of total, foreign-born,
and native employment in each of eleven sectors. The final column gives the
percentage of foreign born in total sectoral employment for each sector. The
table shows that foreign born accounted for 7.97% of total employment but
that these workers were not evenly distributed across sectors. The largest dif-
ference between the two groups is in manufacturing, which accounted for
25% of foreign-born workers but only 19% of native workers. Furthermore,
manufacturing has the second highest concentration of immigrants, behind
private household services. Immigrants are relatively underrepresented in
public administration. However, it is not surprising that natives are twice as
likely to hold these jobs since many of them require citizenship. Immigrants
are also underrepresented in construction and in transport and public utilities.
Within manufacturing, immigrants are disproportionately located in nondur-
able goods: 9.5% of total employment in nondurables was immigrant com-
pared to less than 7% of total employment in durables.
Immigrants are disproportionately located in apparel, where they account
for over 19% of total employment. They also account for large employment
shares in textiles, footwear, leather, drugs, and cleaning and toilet prepara-
tions. They are relatively scarce in tobacco, petroleum refining, and chemical
product industries. The various durable goods industries each account for a
small share of total immigrant employment. Immigrants are relatively visible
in some sectors, such as miscellaneous manufacturing.
308 Susan M. Collins
In summary, immigrants account for less than 8% of total employment.
However, they are distributed across industries quite differently than native
workers are. Furthermore, they are extremely visible in some industries,
amounting to 12%-20% of the total work force. The remainder of the paper
explores the implications of recent changes in aggregate demand on the distri-
bution of job requirements across industries and asks whether reductions in
job requirements have been concentrated in industries where immigrants are
also concentrated.
11.2 Shifts in the Composition of Aggregate Demand
In fact, there have been large recent changes in the composition of aggre-
gate demand. The two identities from national income accounting given in (1)
and (2) are very useful for documenting the shifts in key macroeconomic var-
iables and for highlighting the linkages between the foreign sector and domes-
tic demand:
(1)
(2)
Y = C + 1 + G + (X - M),
( X - M) = S, + S, - 1 - R.
As usual, Y denotes GNP; C, I , and G denote private consumption, invest-
ment, and government spending, respectively; and X and M refer to exports
and imports of goods and services, S, and S, to private and government sav-
ings, and R to net other international transaction^.^ Equation (2) says that
foreign savings must equal the difference between domestic savings and in-
vestment.
Table 11.2 shows the U.S. experience during 1973-86. The top panel gives
the composition of aggregate demand as shares of GNP during each of four
subperiods, while the bottom panel shows the domestic savings and invest-
ment counterparts to net export performance. As shown, net exports declined
during 1977-79, improving somewhat during 1980-82 before the substantial
deterioration during 1983-86.
The two periods of poor trade performance differ in more than simply the
magnitude of the deficit. In 1977-79, investment rose by 2.2% of GNP rela-
tive to 1973-76, requiring additional domestic and/or foreign savings. Ap-
proximately one-third was met by foreign savings, as the trade deficit declined
by .8% of income. The remaining two-thirds was met by an increase in gov-
ernment savings. During 1980-82, investment fell, as did all three compo-
nents of savings. However, because of the larger decline in government sav-
ings, foreign savings did not return to its 1973-76 level-the trade balance
recovered only partially.
The 1983-86 period stands in stark contrast to 1977-79. Although the trade
balance deteriorated by 2.6% of GNP, investment rose by little more than .5
percent. Instead of increased domestic savings, government savings fell pre-
309 Immigrants, Labor Market Pressures, and the Aggregate Demand
Table 11.2 The U.S. Experience, 1973-86
Net
Year Consumption Investment Gov’t (Defense) Exports
The shifting composition
of aggregate demand (as
a percentage of GNP):
1973-76 62.6 15.8 20.2 ( 5. 5) 1.3
1977-79 62.7 17.9 18.9 (4.9) .5
1980-82 63.6 15.7 19.6 (5.6) 1 .o
1983-86 65.2 16.3 20.1 (6.4) - 1.6
Savings
Private Net
Private Gov’t Investment Exports Other
Decomposition of net
exports: savings -
investment (as a
percentage of GNP):
1973-76 18.1 - 1.5 15.8 1.3 .6
1977-79 17.9 - .2 17.9 .5 .8
1980-82 17.7 - 1.9 15.7 1 .o .9
1983-86 17.2 -3.3 16.3 - 1.6 .8
Source: Economic Report of the President, 1987
cipitously, by 1.5% of GNP, while private savings continued its trend decline.
From table 11.2, government spending rose by just .5 percent of GNP (al-
though this figure masks the large shift toward defense spending). Private con-
sumption, on the other hand, ranged from 64.8% to 65.6% of GNP during
1983-86. The jump is especially notable because private consumption has
been relatively stable at 63% of income since 1950 and has exceeded 64% in
only four years between 1950 and 1980. Thus, the 1983-86 trade deficit co-
incided with the large reduction in government revenues, which lowered gov-
ernment savings but raised private consumption. The next section of the paper
explores the labor market implications of these compositional shifts in de-
mand.
11.3 Final Demands and Labor Requirements
1 1 . 3. 1 The Framework
The I 0 tables provide a useful way to link changes in the composition of
demand to shifts in industrial output and labor requirements. As already dis-
cussed, the major shortcomings of the approach are that it does not consider
whether the demand shifts are associated with relative price changes and that
it rules out substitution-on both the demand and the supply side-by assum-
310 Susan M. Collins
ing constant coefficients. The results provide information about the labor that
would be required to produce the sectoral outputs consistent with a particular
final demand. These labor requirements may be very different from sectoral
employments-especially in the short run. Although the results of an I 0 anal-
ysis cannot be interpreted as indicting shifts in actual labor demands or em-
ployments, they do provide information about the likely labor market pres-
sures.
The analysis uses the eighty-five-industry-level disaggregation of the 1980
I 0 tables.'j Final demands and data used to compute the technical input-output
coefficients are valued in producer prices. The vector of labor requirements
per dollar output is matched to the 1977 10 tables.' The exercises discussed
below will consider different compositions of final demand, so that the index
k refers to the kth scenario, or aggregate demand composition.
The central relation is given in equation (3):
(3) L, = Y [ I ] . Q . dk9
where
L, = the total labor required in each industry, given the kth aggregate de-
y = the vector of labor requirements for a dollar of output industry
I = the identity matrix (85 x 85);
Q = the total requirements (direct and indirect) matrix of the output from
each industry required to prduce a dollar's worth of each commodity
(85 x 85) ;
d, = the vector of final demands for each commodity, given the kth aggregate
demand composition (85 X 1).
Equation (4) divides final demand into three parts:
(4)
where
mand composition (85 x 1);
(85 X 1);
d, = n * T,. D,
D =
TT, =
n =
total aggregate demand;
the vector of shares in total demand of each of the nine components-
private consumption, investment, inventory accumulation exports, im-
ports, and four types of government expenditures (9 X 1);
the matrix of demands for each of the eighty-five commodities per dol-
lar of each component of final demand (85 X 9).
To focus on the implications of recent shifts in demand composition, alter-
native final demand vectors were computed by varying the shares in aggregate
demand (T,) while holding everything else constant. In other words, total de-
mand and the commodity composition of each piece of final demand were
held constant as the shares of investment, consumption, imports, etc. were
varied.
311 Immigrants, Labor Market Pressures, and the Aggregate Demand
1 1.3.2 The Components of Final Demand
Variations in the composition of aggregate demand will influence labor re-
quirements even when total demand is held constant. This is because each
type of demand has a different commodity basket so that each concentrates its
spending on commodities with different labor requirements.
To help interpret the results in the next sections, table 11.3 compares the
total labor requirements to produce one million (1980) dollars worth of the
commodities in each demand component. (Thus, the entry for imports is pos-
itive .) Inventories are excluded because the commodity composition of inven-
tories varies substantially from year to year so that it is not particularly useful
to think of them as a fixed basket of commodities.
The first column of table 11.3 reports the total number of jobs required per
million dollars. The second column gives the average expenditure per job to
produce each commodity basket. The figures point out that there is substantial
variation. Government spending requires the most labor per dollar spent. A
million dollars of government expenditure requires 69 jobs when spent on
education and 55 jobs when spent on defense. (Equivalently, the figures imply
a total of $14,400 of educational expenditure per job and $18,200 of defense
expenditure per job .) Government expenditures are followed by fixed invest-
ment and private consumption, which require 42 and 41 jobs per million dol-
lars, respectively.
It may seem surprising that the external sector has the smallest labor re-
quirements and that labor requirements are slightly higher for exports than for
imports. Because of the relative capital abundance in the United States, im-
Table 11.3 Labor Requirements and Final Demand
All Workers Immigrants
Jobs per $million Jobs per $million
'Qpe of Demand Expenditure $/Job Expenditure % Total Jobs
Private consumption
Fixed investment
Exports
Imports ( - )
Federal government:
Defense
Other
State & local:
Education
Other
Addendum:'
Exports
Imports ( - )
40.7
42.0
26.6
23.0
24,600
23,800
37,600
43.500
3.5
3.2
2.1
2.1
8. 5
7.7
8.0
9.2
54.9
61.8
69.4
59.7
38.3
44.2
18,200
16,200
14,400
16,700
26,300
22,700
3.2
3.5
3.1
3.4
5.8
5.6
4.6
5.7
Source: Tabulated by author from I 0 tables, as described in text.
a Excludes petroleum, and noncomparable products.
312 Susan M. Collins
ports would be expected to be relatively labor using. However, traditional
trade theory has implications for the ratio of capital to labor embodied in
trade, not for the absolute amount of any single factor.
In fact, Leontief‘s paradox (that capital abundant countries have higher
capital-labor ratios embodied in their imports than in their exports) has been a
standard result in the empirical trade literature. For a review of this literature,
see Deardorff 1984. Early resolutions to the paradox have included disaggre-
gation of labor so as to include human capital as a separate factor and special
treatment of natural resource industries. More recently, Learner (1980) has
shown that it is not inconsistent with theory for capital abundant countries to
have higher capital-labor ratios in imports when they are also running trade
surpluses.
In the 1980 I 0 tables, further disaggregation of imports and exports gener-
ates the more intuitive result that imports are labor intensive relative to ex-
ports. Both export and import final demands from the I 0 tables include sub-
stantial expenditures on commodities that have little or no domestic labor
inputs. For example, 27.8% of exports are classified as commodities from the
“rest of the world,” including labor remittances. The relevant “industry” uses
no domestic labor. Nine percent of imports are from the rest of the world. In
addition, 14% are “noncomparable imports,” also with no domestic labor
usage, while 26% are on petroleum-related products that use relatively little
labor. The total labor requirement per million dollars spent on imports exclud-
ing these special categories is 44. This exceeds the labor requirements for
consumption and investment expenditures and the comparable figure for ex-
ports, which is only 38.
Total labor requirements per dollar of expenditure depend on the type of
final demand expenditure. It is also interesting to explore how type of expend-
iture is likely to influence the availability of jobs for immigrants versus non-
immigrants. To do this requires an additional assumption-that the share of
immigrants in total industry employment remains relatively constant. Then,
if immigrants account for 8% of agricultural employment, an increase of 100
agricultural jobs will generate approximately 8 jobs for immigrants and 92 for
nonimmigrants. Using the data on immigrants as a share of total industry em-
ployment from table 11.1 in this way, it is possible to split the total labor
requirements from column 1 of table 1 1.3 into immigrant and nonimmigrant.
The third and fourth columns of table 11.3 report the number of “immigrant
jobs” per million dollars of expenditure on each demand component together
with the jobs “held” by immigrants as a percentage of total labor require-
ments.
Because immigrants are not proportionally distributed across industries, the
share of total jobs that are likely to be filled by immigrants changes with the
component of demand. The last column of table 11.3 shows that the import
commodity basket is the one with the highest immigrant labor concentration.
Nine point two percent of the labor required to produce the (1980) import
313 Immigrants, Labor Market Pressures, and the Aggregate Demand
basket was likely to have been immigrant. This is not surprising given the
large percentages of immigrant workers in apparel, footwear, and other indus-
tries with strong import competition. Private consumption expenditures have
the second highest immigrant concentration, followed by exports and fixed
investment. Government expenditures, particularly on education, come at the
other end of the scale because the labor requirements to satisfy these demands
are concentrated in industries with relatively small shares of immigrants.
Thus, shifts in the composition of final demand will tend to put different
labor market pressures on immigrants and nonimmigrants. In particular, an
investment boom will generate a larger rise in total labor requirements than a
consumption boom, and fewer of those jobs are likely to go to immigrant
workers. Overall, a consumption boom will be relatively less beneficial for
native workers.
However, these aggregate figures mask differences in the intraindustry labor
requirements associated with the demand components and provide an incom-
plete picture of the likely labor market pressures associated with changes in
the composition of aggregate demand. The next step is to consider particular
aggregate demand vectors and to compare the implied labor requirements,
disaggregating both by immigranthathe and by industry.
11.3.3 Alternative Scenarios
The analysis below considers four final demand vectors corresponding to
four compositions of aggregate demand. These are shown in table 11.4. The
first column shows the base case-the actual composition in the 1980 I 0
tables.8 The second column shows a scenario like the 1983-86 period. The
Table 11.4 Composition of Aggregate Demand (percentages of GNP)
% A Alt. 1 From Alt. 2 1980 Alt. 3
1980 Base Trade Deficit Trade Deficit Investment Slump
Type of Demand % Total Consumption Boom Investment Boom Consumption Boom
Consumption
Investment:
Fixed
Inventory
Exports
Imports
Government,
federal:
Defense
Other Federal
State & local:
Education
Other
63.1
16.1
- .3
12.6
-11. 5
-
4.9
2.4
5.2
7.5
65.6
16.6
- .3
11.6
14.0
6.5
2.2
5.0
6.8
63.1
18.6
- . 3
11.6
- 14.0
-
5.4
2.6
5.4
7.6
65.6
14.1
- .3
12.6
11.5
4.9
2.2
5.1
7.3
Source: Tabulated by author from I 0 tables, as described in text
314 Susan M. Collins
trade balance shifts from surplus to deficit,while private consumption soars.
In addition, government spending shifts toward federal defense spending and
away from state and local expenditures. The third column shows an alternative
scenario with the same trade deficit as in column 2, but an investment instead
of a consumption boom. The final column maintains the consumption boom
from column 2 but assumes an investment slump instead of a trade deficit.
(Inventories are the same share of output in all scenarios.)
Because it is misleading to interpret the labor requirements as employment,
the results for the three alternative scenarios are presented as percentage
changes from the corresponding 1980 base. Table 1 1.5 provides an overview
of the effect of the compositional shifts in demand on labor requirements.
Even though total aggregate demand is held constant, total labor require-
ments rise in the first two scenarios and fall slightly in the third. The increases
for trade deficits combined with either a consumption boom (alternative 1) or
an investment boom (alternative 2) are not surprising given the relatively
small labor requirements per dollar of total imports. The labor requirements
rises somewhat more in scenario 2 because scenario 1 includes a shift toward
defense and away from other types of government spending and because in-
vestment requires relatively more labor than consumption per dollar expendi-
ture. This also explains the decline in the third scenario.
It is again interesting to decompose the total labor requirements into immi-
grant and nonimmigrant segments. This is done in the second and third rows
of table 1 1.5. The figures point out a more striking difference between the two
trade-deficit scenarios. In the actual 1983-86 combination described in alter-
native 1, there is little difference between the growth of “immigrant jobs” and
“nonimmigrant jobs.” However, alternative 2 , the historically typical combi-
nation of high investment and trade deterioration, implies a substantially
larger expansion of “nonimmigrant jobs” than “immigrant” jobs. Similarly, a
switch from investment to consumption with no trade change, as in alternative
3, implies an increase in jobs typically held by immigrants and a decline in
jobs typically held by natives.
To the extent that workers in declining sectors are laid off more quickly than
workers are hired in expanding sectors, any demand shift that alters the sec-
Table 11.5 Demand Shifts and Total Labor Requirements
% A Alt. 1 From Alt. 2 1980 Alt. 3
Consumption Boom Investment Boom Consumption Boom
Trade Deficit Trade Deficit Investment Slump
Immigrant 1.22
Native 1.27
Total 1.27
1.12
1.81
1.75
. I 6
- . 33
- .30
315 Immigrants, Labor Market Pressures, and the Aggregate Demand
toral distribution of labor requirements will tend to cause short-term unem-
ployment, even if aggregate labor requirements have increased. A key impli-
cation of table 11.5 is that native workers will have a relatively easier
adjustment to expanding imports when the trade deficit is associated with an
investment boom. The very different distribution of jobs when the trade deficit
coincides with a consumption boom places native workers more directly in
competition with immigrants for the jobs in expanding sectors.
The next step is to look at the distribution of the change in job requirements
implied by each scenario across industries. Table 11.6 examines the sectoral
decomposition of total labor requirements for all sectors and nondurable man-
ufacturing where immigrants are overrepresented. The first column gives the
share of each sector in the total labor requirements in the 1980 base. Columns
2-4 give the percentage change in sectoral labor requirements for each of the
three alternative scenarios.
The consumption boom-trade deficit alternative leads to a reduction of la-
bor requirements in manufacturing but to increases in service and government
sectors. (The decline in manufacturing, with its large concentration of immi-
grants, is offset by a spurt in household services and other particular indus-
tries.) In contrast, the investment boom-trade deficit scenario leads to in-
creased labor requirements in manufacturing but slower growth in the service
sectors.
An important implication from the top panel of table 1 1.6 is that we should
expect a trade deficit cum consumption boom to shift employment from man-
ufacturing to service sectors, but it is incorrect to conclude that such an em-
ployment shift is “caused” by an expanding trade deficit. A trade deficit of
equal magnitude (alternative 2) cum investment boom will tend to shift em-
ployment toward manufacturing and construction and away from services, fi-
nance, insurance, and real estate.
The analysis has identified two sets of labor market pressures that are likely
to arise from a consumption boodtrade imbalance but unlikely to be present
in the more standard investment boodtrade imbalance combination. The first
is that natives are relatively less concentrated in the sectors and industries with
large increases in labor requirements. The second is that labor requirement
decreases will be concentrated in the manufacturing sector and therefore more
likely to increase pressures for protectionism.
The bottom panel of table 11.6 presents further information about the
change in labor requirements in (durable and nondurable) manufacturing
under each alternative to make a third point. Within manufacturing, many of
the industries with large declines under the consumption boom-trade deficit
scenario also have high concentrations of immigrants. In particular, leather,
footwear, and plastics, three of the industries suffering the largest losses, have
immigrants accounting for 9.5%-16.5% of their work forces. Immigrants are
extremely “visible” in these industries. Similarly, of the three industries en-
316 Susan M. Collins
Table 11.6 Changes in Labor Requirements in All Sectors and in Nondurable
Manufacturing
1980 % A Alt. 1 From Alt. 2 1980 Alt. 3
Base Trade Deficit Trade Deficit Investment Slump
% Total Consumption Boom Investment Boom Consumption Boom
All sectors:
Agriculture
Mining
Construction
Manufacturing
Transportation,
communication
Wholesale & retail trade
Finance, insurance,
real estate
Services
Government
Total
Nondurable manufacturing:
Food & kindred products
Tobacco manufacturers
Fabric
Textile mill products
Apparel & other textile
Miscellaneous textile
Paper & allied products
Paperboard containers &
Pnnting & publishing
Chemical & allied
Plastic materials &
products
products
boxes
products
synthetics
DNgS
Paints & allied products
Petroleum & coal
Rubber & miscellaneous
Leather products
Footwear except
products
plastic products
rubber products
Total
1.95
.80
6.39
20.87
5.08
16.60
5.12
24.26
18.94
100.00
20.60
.84
6.64
1.38
15.86
1.89
6.24
2.53
15.09
6.36
2.68
3.78
.74
3.86
8.36
.28
2.86
100.00
- .44
- 7.04
- .78
- .53
1.32
2.35
2.46
2.35
1.28
1.27
1.59
1.32
-2.01
- 1.02
- .28
.33
-2.36
.08
1.18
-3.93
- 2.92
.23
- .27
- .50
- 1.75
-8.08
-5.70
.I0
-2.78
-6.56
8.60
.54
.55
1.54
.15
.26
4.46
I .75
- 1.80
- 1.94
- 4.40
.90
-4.77
- 1.62
-2.88
- .85
1.02
-4.11
-3.03
- 1.86
3.35
-2.44
- 1.25
- 12.67
- 11.01
-2.20
2.53
- .69
-6.72
- 1.85
.63
.79
2.61
2.42
-2.14
- .30
3.50
3.26
2.25
- 1.74
4.31
1.94
.92
.93
.70
- .03
- .08
2.64
-3.19
1.43
- .70
4.57
5.31
1.92
Source: Tabulated by author from I 0 tables, as described in text.
317 Immigrants, Labor Market Pressures, and the Aggregate Demand
joying the largest increases, two (tobacco and printing) have relatively small
concentrations of immigrants, while the third (food) is about average.
1 1.4 Concluding Remarks
This paper has made two major points. First, changes in the composition of
aggregate demand will shift the distribution of labor requirements across in-
dustries and sectors. It will also affect total labor requirements. Thus, it is
important to identify changes in the components of domestic absorption when
analyzing the likely labor market consequences of aggregate demand shifts.
By the same token, it is inappropriate to examine changes in imports or ex-
ports in isolation. An increase in imports must coincide with an offsetting shift
in some other component(s) of demand, and a consumption boom is not
equivalent to an investment boom or to a rise in government expenditures.
Instead, the total and the sectoral distribution of labor requirements will de-
pend critically on which has occurred.
The analysis showed that the 1980s consumption boom with trade deficit
has implied that decreased labor requirements were concentrated in the manu-
facturing and construction sectors. In contrast, an investment boom with trade
deficit would tend to raise labor requirements in both these sectors. Therefore,
it is incorrect to ask whether trade deficits tend to“deindustria1ize” the econ-
omy by shifting employment from manufacturing to services because the an-
swer also depends on the changes in other components of aggregate demand.
The second set of points concerns the role of immigrant workers versus
native workers. Immigrants are not evenly distributed across sectors but are
concentrated in some manufacturing industries and in private household ser-
vices. Thus, shifts in the distribution of labor market requirements should
have quite different short-run implications for immigrants than for native
workers.
The analysis showed that an investment boom cum trade deficit implies that
requirements for jobs typically held by natives will increase nearly twice as
quickly as requirements for jobs typically held by immigrants. Immigrants are
relatively better off in a consumption boom, in which case there is little differ-
ence between the job growth rates for the two groups. Furthermore, some of
the industries with the largest declines under the consumption boom are also
the ones with the heaviest concentrations of immigrants. Both these factors
may tend to make immigrants more “visible” in the labor market and help
explain the recent increase in concern over their presence.
Notes
1 . Another possible explanation, an increase in the number of legal and/or illegal
2. This approach is relevant for short-run analysis only. Over time, both immigrants
3. This issue is discussed in Sehgal(l985) and in Borjas (1987).
immigrants entering the country, is discussed in other papers in this volume.
and native workers will presumably move from contracting to growing sectors.
318 Susan M. Collins
4. For additional discussion of these data, see Sehgal(l985).
5. Net other international transactions are capital grants, net transfers, and interest
payments. This term also includes the statistical discrepancy.
6. For further description of these data, see “The Input-Output Structure of the U.S.
Economy, 1977” in Survey of Current Business (May 1984).
7. These data are from the Survey of Current Business, November 1985 and May
1986.
8. These data do not correspond exactly to the figures in table 1 1.1 because output
is measured in producer prices in the I 0 tables but in consumer prices in the National
Income and Product Accounts.
References
Abowd, J. M., and R. B. Freeman. 1990. Internationalization of the U.S. Labor Mar-
ket. NBER Working Paper no. 3321. Cambridge, Mass.: National Bureau of Eco-
nomic Research.
Aho, C. M. , and J. A. Orr. 1981. The Growth of Trade-Sensitive Employment.
Monthly Labor Review 104; no. 2(February):28-35.
Borjas, G. 1987. Self-Selection and the Earnings of Immigrants. American Economic
Review 77, no. 4(September):531-53.
Branson, W. H. , and J. P. Love. 1986. Dollar Appreciation and Manufacturing Em-
ployment and Output. NBER Working Paper no. 1972. Cambridge, Mass.: National
Bureau of Economic Research.
DeardorfS, A. V. 1984. Testing trade theories and predicting trade flows. In Handbook
of International Economics, vol. 1, eds. R. W. Jones and P. B. Kenen. Amsterdam:
North-Holland.
DeardorE, A. V., and R. M. Stem. 1986. The Michigan Trade Model of World Pro-
duction and Trade: Theory and Applications. Cambridge, Mass.: MIT Press.
Dickens, W. T., P. Shapiro, L. Tyson, and J. Zysman. 1985. The Employment Effects
of International Trade: A Review of the Literature. Department of Economics, Uni-
versity of California, Berkeley. Mimeo.
Grossman, G. M. 1986. Imports as a Cause of Injury: The Case of the U.S. Steel
Industry. Journal of International Economics 20, nos. 3/4(May):201-24.
. 1987. The Employment and Wage Effects of Import Competition in the
United States. Journal of International Economic Integration 2, no. l(Spring): 1-23.
Krueger, A. 0. 1979. The Impact of Foreign Trade on Employment in U.S. Industry.
In Current Issues in Commercial Policy and Diplomacy, ed. J. Black and B. Hindley.
London: Macmillan.
. 1980. Protectionist Pressures, Imports and Employment in the United States.
Scandinavian Journal of Economics 82(2): 133-46.
Learner, E. 1980. The Leontief paradox reconsidered. Journal of Political Economy
90, no. 3 (June): 495-503.
Orr, A. C., and J. A. Orr. 1983. Employment Adjustments in Import Sensitive Man-
ufacturing Industries, 1960-80. Department of Economics, Manhattan College.
Mimeo.
Schoepfle, C. K. 1982. Imports and Domestic Employment: Identifying Affected In-
dustries. Monthly Labor Review 105, no. 8 (August): 13-26.
Sehgal, E. 1985. Foreign Born in the U.S. Labor Market: The Results of a Special
Survey. Monthly Labor Review 108, no. 7 (July): 18-24.
U. S. Trade Related Employment, 1978-84. 1986. USITC Publication no. 1855, May.
Washington, D.C. : United States International Trade Commission.
111
Comparative Experiences:
Canada and Australia
This Page Intentionally Left Blank
12 An Analysis of the Earnings of
Canadian Immigrants
David E. Bloom and Morley Gunderson
The purpose of this paper is to analyze immigrant labor market progress in
Canada-a country that, as a matter of official policy, has screened most of its
immigrants on the basis of their expected “ability to assimilate.” In particular,
we compare earnings profiles for Canadian immigrants and natives and seek
to determine whether immigrant earnings profiles reflect any “vintage effects”
associated with year of immigration.
Over the past ten years, a number of studies of immigrant earnings have
focused on these same issues using data for U.S. immigrants. Among the best
known is that by Chiswick (1978), which fits a standard wage equation to
cross-sectional data on immigrants and natives in the 1970 Public Use Sample
of the U.S. Census. Chiswick’s results support the conclusion that, when they
first enter the labor market, immigrants earn approximately 25 percent less
than natives with comparable years of schooling and experience, marital
status, etc. However, Chiswick also finds that immigrants have steeper
experience-earnings profiles than “comparable” natives, with immigrant earn-
ings overtaking native earnings within roughly thirteen years of their entry
into the United States. A number of other studies have fit the same basic
model to similar data and have reached roughly identical conclusions (see,
e.g., Carliner 1980; Long 1980; and Borjas 1982).
David E. Bloom is professor of economics in the Department of Economics at Columbia Uni-
versity and a research associate of the National Bureau of Economic Research. Morley Gunderson
is professor of economics and director of the Centre for Industrial Relations at the University of
Toronto.
An earlier version of this paper was presented at the NBER Conference on Trade, Immigration,
and Labor in September 1987. The authors are indebted to McKinley Blackbum, Anne Hill,
Robert Komfeld, David Neumark, Andrew Newman, and Marcus Rebick for assistance in the
preparation of this paper and to Blackbum, Rebick, Richard Freeman, Mark Killingsworth, Jacob
Mincer, and Glenn Withers for helpful discussions and comments. This research was supported
by NIH grant HD18844-02 and by a grant from the Ford Foundation to the National Bureau of
Economic Research.
321
322 David E. Bloom and Morley Gunderson
The set of findings based on Chiswick’s approach to measuring immigrant
assimilation has been challenged by Borjas (1985), who argues that the steep-
ness of immigrant earnings profiles is inflated by cross-cohort declines in im-
migrant quality. Evidence supporting this argument is provided by using
pooled data from the 1970 and 1980 U.S. Population Censuses to measure
earnings growth in the intercensal period for individual entry cohorts of im-
migrants. On the basis of this analysis, Borjas concludes that “cross-section
studies of immigrant earnings provide useless and misleading insights into the
process of immigrant assimilation into the labor market” (p. 485).
Borjas’s conclusion deserves further examination. A priori theoretical rea-
soning is perhaps more consistent with Chiswick’s empirical conclusions than
with those of Borjas. Low entry wages for immigrants can plausibly be ex-
plained as a loss of (origin) country-specific human capital; rapid earnings
growth can be viewed as reflecting positive self-selection into immigration
(i.e., immigrants are above average in terms of their aggressiveness, ambi-
tiousness, willingness to work hard, etc.). In contrast, sizable cross-cohort
declines in immigrant quality are somewhat harder to accept given that it is
not overall quality that is hypothesized to have declined but rather that com-
ponent of overall quality that is unmeasured (i.e., the part of immigrant qual-
ity that is not measured by or correlated with variables such as schooling,
experience, marital status, country of origin, etc.). Borjas’s results, as he rec-
ognizes, may also reflect differential patterns of underenumeration in the
successive Censuses or nonrandom intercensal mortality and out-migration.
Indeed, out-migration, death, or undercounting of immigrants who are rela-
tively unsuccessful in the labor market are all alternatives to declining immi-
grant quality as an explanation of Chiswick’s cross-sectional results.
We will also use the example of Canada as an opportunity to gain some
insight into the importance of intercensal exiting from an immigrant popula-
tion. Although there is little information on either the covariates of immigrant
mortality or on differential Census undercounting of immigrants, there are
several established lines of inquiry on the subject of out-migration. According
to a group of imperfect information models, out-migration is an event that was
unplanned ex ante and that occurs primarily among migrants whose labor
market expectations are not satisfied (see Yezer and Thurston 1976; Allen
1979; Blejer and Goldberg 1980; and Lam 1986). These models suggest, at
the margin, that out-migrants will tend to be selected from the lower end of
the earnings distribution. On the other hand, intertemporal substitution mod-
els tend to view out-migration as a planned event among individuals who
make short-term moves in order to take maximal advantage of temporarily
favorable earnings opportunities (Stark and Bloom 1986; Fox 1987). These
models suggest that out-migration will be most prevalent among individuals
who are selectively active and successful in the labor market. Although the
results are far from definitive, empirical research by Jasso and Rosenzweig
(1987, 1989) and by Lam (1987) tends to support this view insofar as out-
323 An Analysis of the Earnings of Canadian Immigrants
migration of U.S. and Canadian immigrants is reported to be most prevalent
among those who are relatively successful.
Although they are extremely different in spirit, both the imperfect informa-
tion and the intertemporal substitution models of out-migration share an im-
portant empirical implication, namely, that the variance of residuals in a mi-
grant earnings equation will decline with duration of stay (i.e., under the
imperfect information models, exit occurs at the lower end of the distribution,
while, under the intertemporal substitution models, exit occurs at the upper
end). In contrast, job matching or asymmetric information models imply that
the residual variance in a wage equation will increase with duration of stay as
employers are increasingly able to observe the true productivity of migrants
(see Harris and Holmstrom 1982; Katz and Stark 1984). We attempt to infer
which set of forces tends to be stronger by examining patterns in the variance
and kurtosis of immigrant earnings by duration of stay. For example, we will
interpret an increase (decrease) in the variance of earnings with duration of
stay as evidence favoring the relative importance of the job-matching models
(imperfect information models).
Thus, we have four main goals in this paper. First, by fitting the models
proposed by both Chiswick and Borjas to data for Canada, we hope to assess
the extent to which it is generally true that cross-sectional studies of immi-
grant earnings are “useless and misleading.” Second, we hope that estimates
of these alternative models will lead to clear substantive conclusions regarding
the shape of immigrant earnings profiles and the importance of entry-cohort
effects on earnings. Third, by comparing corresponding results under different
Canadian immigration policies, we hope to shed some light on the signifi-
cance of a nation’s institutions in determining the economic benefits of immi-
grati