Finding Jobs: Work and Welfare Reform 0871541165, 9780871541161

Citation preview

Finding Jobs

Finding Jobs Work and Welfare Reform

David E. Card and Rebecca M. Blank, Editors

Russell Sage Foundation. New York

The Russell Sage Foundation The Russell Sage Foundation, one of the oldest of America's general purpose foundations, was established in 1907 by Mrs. Margaret Olivia Sage for "the improvement of social and living conditions in the United States." The Foundation seeks to fulfill this mandate by fostering the development and dissemination of knowledge about the country's political, social, and economic problems. While the Foundation endeavors to assure the accuracy and objectivity of each book it publishes, the conclusions and interpretations in Russell Sage Foundation publications are those of the authors and not of the Foundation, its Trustees, or its staff. Publication by Russell Sage, therefore, does not imply Foundation endorsement. BOARD OF TRUSTEES Ira Katznelson, Chair Alan S. Blinder Christine K. Cassel Thomas D. Cook Robert E. Denham Phoebe C. Ellsworth

Jennifer L. Hochschild Timothy A. Hultquist Ellen Condliffe Lagemann Cora B. Marrett Neil J. Smelser

Eugene Smolensky Marta Tienda Eric Wanner

Library of Congress Cataloging-in-Publication Data Finding jobs: work and welfare reform / David Card and Rebecca Blank, editors p. cm. Includes bibliographical references and index. ISBN 0-87154-116-5 (cloth) ISBN 0-87154-159-9 (paperback) 1. Welfare recipients-Employment-United States. 2. Unskilled labor-United States. I. Card, David E. (David Edward), 1956- II. Blank, Rebecca M. HV95.F48 2000 362.5' 8' 0973-dc21 00-020794 Copyright © 2000 by Russell Sage Foundation. First papercover edition 2002. All rights reserved. Printed in the United States of America. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior written permission of the publisher. Reproduction by the United States Government in whole or in part is permitted for any purpose. The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences-Permanence of Paper for Printed Library Materials. ANSI 239.48-1992. Text design by Suzanne Nichols. RUSSELL SAGE FOUNDATION 112 East 64th Street, New York, New York 10021 10 9 8 7 6 5 4 3 2 1

Contents

CONTRIBUTORS Introduction

vii

THE LABOR MARKET AND WELFARE REFORM Rebecca M. Blank and David E. Card

PART I

THE DEMAND FOR LOW-WAGE WORKERS

Chapter 1

THE EMPLOYMENT, EARNINGS, AND INCOME OF LESS SKILLED WORKERS OVER THE BUSINESS CYCLE

21

23

Hilary W. Hoynes

Chapter 2

DISPLACEMENT AND WAGE EFFECTS OF WELFARE REFORM Timothy J. Bartik

PART II Chapter 3

72

WAGES AND JOB CHARACTERISTICS IN THE LESS SKILLED LABOR MARKET

123

JOB CHANGE AND JOB STABILITY AMONG LESS SKILLED YOUNG WORKERS 125

Harry J. Holzer and Robert J. Lalonde

Chapter 4

WAGE PROGRESSION AMONG LESS SKILLED WORKERS 160

Tricia Gladden and Christopher Taber

Chapter 5

GENDER DIFFERENCES IN THE LOW-WAGE LABOR MARKET 193

Jane Waldfogel and Susan E. Mayer

Chapter 6

HEALTH INSURANCE AND LESS SKILLED WORKERS Janet Currie and Aaron Yelowitz

Chapter 7

PART III Chapter 8

233

EMPLOYEE-BASED VERSUS EMPLOYER-BASED SUBSIDIES TO LOW-WAGE WORKERS: A PUBLIC FINANCE PERSPECTIVE Stacy Dickert-Conlin and Douglas Holtz-Eakin

262

PUBLIC POLITICS TO INCREASE EMPLOYMENT AND EARNINGS OF LESS SKILLED WORKERS

297

PUBLIC SERVICE EMPLOYMENT AND MANDATORY WORK: A POLICY WHOSE TIME HAS COME AND GONE AND COME AGAIN? 299

David T. Ellwood and Elisabeth D. Welty

I

v

Contents

FINANCIAL INCENTIVES FOR INCREASING WORK AND INCOME AMONG LOW-INCOME FAMILIES Rebecca M. Blank, David E. Card, and Philip K. Robins

"373

CHILD CARE AND MOTHERS' EMPLOYMENT DECISIONS Patricia M. Anderson and Phillip B. Levine

420

PART IV

THE IMPACT OF WElFARE REFORM

463

Chapter 11

USE OF MEANS-TESTED TRANSFER PROGRAMS BY IMMIGRANTS, THEIR CHILDREN, AND THEIR CHILDREN'S CHILDREN Kristin F. Butcher and Luojia Hu

465

TIME LIMITS Robert A. Moffitt and LaDonna A. Pavetti

507

INDEX

5"37

Chapter 9

Chapter 10

Chapter 12

vi

Con tribu tors REBECCA M. BLANK is dean of the Gerald R. Ford School of Public Policy and Henry Carter Adams Collegiate Professor of Public Policy at the University of Michigan. She is also research associate of the National Bureau of Economic Research. DAVID E. CARD is Class of 1950 Professor of Economics and head of the Center for Labor Economics at the University of California, Berkeley. He is also research associate of the National Bureau of Economic Research.

PATRICIA M. ANDERSON is associate professor of economics at Dartmouth College. She is also faculty research fellow of the National Bureau of Economic Research. TIMOTHY J. BARTIK is senior economist at the W.E. Upjohn Institute for Employment Research. KRISTIN F. BUTCHER is on leave from the Department of Economics at Boston College and is program officer at the MacArthur Foundation. JANET CURRIE is professor of economics at the University of California, Los Angeles. She is also research affiliate of the National Bureau of Economic Research. STACY DICKERT-CONLIN is assistant professor of economics and senior research associate in the Center for Policy Research at Syracuse University. DAVID T. ELLWOOD is Lucius N. Littauer Professor of Political Science at the John F. Kennedy School of Government, Harvard University. He is also research asso-

ciate at the National Bureau of Economic Research. TRICIA GLADDEN is a graduate student in the Department of Economics at Northwestern University. DOUGLAS HOLTZ-EAKIN is professor of economics and associate director of the Center for Policy Research at Syracuse University. He is also a faculty research fellow of the National Bureau of Economic Research. HARRY J. HOLZER is professor of economics at Michigan State University.

vii

Contributors

HILARY W. HOYNES is assistant professor of economics at the University of California, Berkeley. She is also faculty research fellow of the National Bureau of Economic Research and affiliate of the Institute for Research on Poverty, and the Northwestern/University of Chicago Joint Center for Poverty Research. LUOJIA Hu will be assistant professor in the Department of Economics at Northwestern University in the fall of 2000. ROBERT J. LALONDE is professor of public policy at the Irving B. Harris Graduate School of Public Policy Studies at the University of Chicago. He is also faculty research fellow of the National Bureau of Economic Research. PHILLIP B. LEVINE is associate professor of economics at Wellesley College. He is also research associate of the National Bureau of Economic Research. SUSAN E. MAYER is associate professor at the Irving B. Harris Graduate School of Public Policy Studies at the University of Chicago. She is also director of the Northwestern/University of Chicago Joint Center for Poverty Research. ROBERT A. MOFFITT is professor of economics at Johns Hopkins University and is affiliated with the Institute for Research on Poverty, the Joint Center for Poverty Research, and the National Bureau of Economic Research. LADoNNA A. PAVETTI is senior researcher at Mathematica Policy Research, Inc. PHILIP K. ROBINS is professor of economics at the University of Miami and a research affiliate of the Institute for Research on Poverty at the University of Wisconsin, Madison. CHRISTOPHER TABER is assistant professor in the Economics Department and research associate of the Institute for Policy Research at Northwestern University. JANE WALDFOGEL is associate professor of social work and public affairs at the Columbia University School of Social Work and research associate at the Centre for Analysis of Social Exclusion at the London School of Economics. ELISABETH D. WELTY is research assistant in the Malcolm Wiener Center at the John F. Kennedy School of Government, Harvard University. AARON YELOWITZ is assistant professor of economics at the University of California, Los Angeles. He is also faculty research fellow of the National Bureau of Economic Research.

viii

Introduction The Labor Market and Welfare Reform Rebecca M. Blank and David E. Card

T

he coordinated push to move an increasing number of welfare recipients off assistance and into full-time work has raised a number of key questions about the nature of the labor market for less skilled workers: Will employment opportunities for former welfare recipients be vulnerable to future recessions? How quickly will workers' wages grow as they gain labor market experience? What is the effect of eligibility time limits on those who remain on welfare despite financial incentives and administrative prodding to leave? The twelve chapters in this book address these and many other important questions about the labor market prospects facing less skilled workers in the aftermath of recent welfare reform legislation.

IMPLEMENTING THE 1996 WELFARE REFORM ACT Concern about job prospects for less skilled workers is hardly new. Interest in the issue has grown substantially, however, in the wake of the Personal Responsibility and Work Opportunity Reconciliation Act of 1996,the latest effort at welfare reform. This legislation created powerful incentives for the states to move welfare recipients into jobs. It mandated specific employment targets for the welfare caseload and offered states new freedom in designing their welfare benefit schedules and implementing welfare-to-work programs. The long-established Aid to Families with Dependent Children (AFDC) program, which had provided cash assistance to needy parents with children for more than forty years, was abolished and replaced with a federal block grant known as Temporary Assistance for Needy Families (TANF). Whereas AFDC was a welfare program operating under a complex set of federally mandated rules, TANF is a funding stream that states can use in a variety of ways. Nevertheless, a number of key restrictions affect state eligibility for TANF funds. 1 Many of these restrictions relate to the employment behavior of adult beneficiaries. In order to continue receiving full TANF funding, states must meet or exceed a rising target for the proportion of welfare recipients who work at least

I

1

Finding Jobs

thirty hours a week. Many states are attempting to achieve these targets by expanding welfare-to-work programs and by encouraging new welfare applicants to look for work before entering the program (so-called diversion activities). A second key provision of welfare reform legislation is the establishment of a fiveyear lifetime eligibility limit for the receipt of federally funded benefits by adults. Time limits on federal funding create an additional incentive for states to move recipients off welfare and into work as quickly as possible. The new federal policies have led to a virtual revolution in the design of assistance programs for low-income families. The fifty states have introduced a host of new TANF-funded programs that vary by how time limits are enforced, how benefit payments are related to work effort, how cases are managed, and how sanctions for noncompliance are enforced. 2 At the same time, public assistance caseloads have plummeted, and the employment rates of welfare recipients and single mothers as a whole have risen. State governors routinely cite these trends as evidence that their welfare reform efforts are succeeding. The apparent success of welfare reform efforts has been aided by a remarkable economic climate. Labor market conditions have improved steadily since the end of the 1991 recession. By the end of the decade, unemployment was at a thirty-year low, the proportion of the population with jobs was continuing to rise, and the real (inflation-adjusted) wages of less skilled workers were showing steady growth for the first time in many years. In this extraordinarily favorable macroeconomic environment, most states were able to focus on redesigning and implementing new programs, with little concern for job availability. The strong economy has been particularly helpful to less skilled workers. Figure I.1 shows unemployment rates by education level over the past six years. The unemployment rate of high school dropouts fell from 11.1 percent in January 1993 to 6.7 percent in May 1999. 3 Figures 1.2 and 1.3 show trends in average hourly wages for men and women at the median (or middle) of the wage distribution and at the tenth and twentieth percentiles! Real wages of most groupsespecially less skilled men-have declined since the late 1970s. Only in the past four years has there been evidence of a sustained turnaround. For those at the tenth percentile of the male wage distribution (typically, young men with less than a high school education), real hourly wages rose by 6.9 percent between 1994 and 1998. Over the same time period, real wages of low-paid women rose by 5.3 percent. Reinforcing the strong labor market, other policy changes in the 1990s also increased the incentives for low-wage workers to enter the labor market and made it easier for states to strengthen their welfare-to-work efforts. The Earned Income Tax Credit (EITC), which provides a refundable credit to individuals with low earnings who live in low-income families, was increased in 1990 and expanded greatly in 1993. The cumulative effect was a tripling of the real (inflation-adjusted) value of the maximum refundable credit available to low-wage workers with two or more children between 1990 and 1998. (The maximum credit for a lowwage worker with one child rose by 90 percent.) The minimum wage was also increased in 1990, 1991, 1996, and 1997, with a cumulative 9 percent increase in

2

I

FIGURE 1.1

/

Unemployment, by Education Level

12.0

10.0

8.0

4.0

2.0

1993

1994

1995

1996

1997

1998

1999

Source: Data from u.s. Department of Labor, Bureau of Labor Statistics, monthly unemployment releases (2000).

FIGURE 1.2

14

/

Hourly Wages of Men Aged Sixteen and Over

5th decile (median)

12

[/J

.... ::l 10 "0 0

I'-.. 0\ 0\

"'"

2d decile 8

6 1st decile 4 1979

1982

1985

1988

1991

1994

1997

Source: Tabulated by the Council of Economic Advisers from the Current Population Survey.

FIGURE I.3

/

Hourly Wages of Women Aged Sixteen and Over

14

12

5th decile (median)

6 1st decile

4

1979

1982

1985

1988

1991

1994

1997

Source: Tabulated by the Council of Economic Advisers from the Current Population Survey. FIGURE 1.4

/

Welfare Caseloads, from 1970 to 1998

5.5

5.0 4.5 'Vi' >::

;.§

~ ]co [Jl

4.0

1996 Welfare Reform Act

3.5

Q)

""

3.0

2.5 2.0

1970

1974

1978

1982

1986

1990

1994

Source: Data from u.s. Department of Health and Human Services, Administration for Children and Families (2000).

1998

Introduction FIGURE 1.5

/

Labor Force Participation of Women

90.0

Single without children 80.0

Married without children •.................•.......... / .. 00000000000 00.

0.. 00 O... , 00 00 0••.. ··"/=0,-::-:.:-. -0 _ .........

/

4/

;~

70.0

~ (l) P-

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

_-",'

","

","

;/'

~

;,

;;/

cd

~

oJ_._. -':::'-=

60.0

Single with children

",/'

50.0

40.0

30.0 1969

1974

1979

1984

1989

1994

1998

Source: Tabulated by the author from the Current Population Survey data.

real terms from 1990 to 1998. Both of these policy changes presumably increased the incomes of low-wage workers (and their families), although they may have also had some offsetting effect on hours or employment. At this stage there are few indications of significant disemployment effects, perhaps because of the strong economic climate. s In the short term, welfare reform, combined with other policy changes and a strong and growing economy, has produced substantial declines in public assistance caseloads. Figure 1.4 plots the number of families receiving AFDC or TANF benefits from 1970 through 1998. After hitting a peak in 1994, caseloads have fallen dramatically, with particularly fast declines after 1996. Current research suggests that the strong economy accounts for about a fifth of the recent caseload decline; much of the remainder is presumably related to policy changes. 6 As caseloads decline, there has been a parallel increase in labor force participation among women with children. Figure 1.5 plots the various rates of labor force participation of married and single women with and without children. Among single women with children, labor force participation has risen 10 percentage points, from 68 percent in 1989 to 78 percent in 1998. The increase is even larger among women who have children and have never been married (who are more likely to receive welfare) than among divorced, separated, or widowed mothers. By comparison, childless women have shown little change in

f

5

Finding Jobs

their labor force participation over the decade. Analysts have credited the relative rise in participation of single women with children to the EITe expansions, as well as to welfare reform. 7 All of these changes are quite recent. It is still far too early to evaluate the long-term impact of welfare reform, particularly on the ability of younger and poorly educated parents to find and keep jobs that will adequately support their families over the course of any future recession.

KEY QUESTIONS ABOUT JOBS, WORK, AND POLICY DESIGN In the spring of 1997, we formulated twelve broad questions we thought were critical to understanding the long-term prospects for employment-focused welfare reform. We then recruited twelve researchers to refine and investigate these questions; their findings constitute the twelve chapters of this book. Part I relates the demand side of the labor market to outcomes for low-wage workers. The strong macroeconomy and the concomitant strength in demand for workers has obviously been very important to the progress of welfare reform since 1996.

Do Less Skilled Workers Respond Differently to Macroeconomic Cycles? In chapter 1, Hilary Haynes examines the effect of business cycle fluctuations on the employment, earnings, and incomes of individuals in different gender, race, and education groups. A key innovation in her approach is her study of labor market outcomes at the local (city) leveL This provides much more data than an analysis at the national level and allows her to analyze potential changes in the cyclical responsiveness of different groups over time. Another key feature of Hoynes's chapter is a focus on a full set of labor market outcomes-employment, hours, earnings, and hourly wages. The chapter provides a far more comprehensive picture of the cyclical sensitivity of different skill groups than has been available in the past. Hoynes's findings confirm that less educated workers, nonwhites and (in some cases) women, are more heavily impacted by business cycle changes than are highly educated white men, whom she treats as a reference group. For instance, when the employment rate of white men with more than a high school education rises by 1 percentage point, the employment rate of less educated white men increases by 1.37 points, the employment rate of less educated nonwhite men increases by 2.92 points, and the rate for less educated nonwhite women increases by 1.52 points. Earnings of these groups are also more cyclically sensitive. The employment and earnings of less skilled white women are somewhat less responsive to the cycle than those of more highly skilled white men. Total family income, which includes government transfers, is also more 6 I

Introduction

cyclical among less skilled and nonwhite groups-but less so than earnings, confirming the role of transfers in smoothing earnings fluctuations. These results help to explain the relative employment and earnings gains of less skilled workers over the 1990s, during a long period of economic expansion. However, they also raise many questions about the sustainability of these gains during any future economic downturn, since less skilled workers suffer greater employment and income losses in a recession. The impact of welfare reform on the future cyclical responsiveness of less skilled workers is hard to predict. For instance, less skilled black women's employment has been more cyclical than that of less skilled men. Historically, black women have also made greater use of public assistance benefits (largely because they are more likely to be single mothers). This may have made it easier for them to move in and out of the labor market as economic conditions change. It is interesting to speculate whether welfare reform efforts and time limits will decrease the cyclicality in labor force participation among less skilled black women because it limits their access to support through the public assistance system or whether, on the contrary, it will increase their cyclicality as they find they are more affected by movements in labor market demand.

Will the Movement of Former Welfare Recipients into the Labor Market Reduce Opportunities for Other Workers? Welfare reform is expected to add between 1 million and 2 million people to the labor force between the mid-1990s and the middle of the following decade. In chapter 2, Timothy Bartik investigates the potential impact of this labor force influx on wages and unemployment rates. Because of the modest size of the inflow relative to the overall U.s. labor market, welfare reform is unlikely to have an appreciable aggregate effect. It may have a larger effect on the labor market for less skilled women, however, because newly working welfare recipients constitute a much larger share of that group's labor supply. Bartik pulls together a range of possible estimates of the effects of this labor supply increase on the outcomes of less skilled groups. His estimates suggest there could be a significant effect on both employment and wages of currently employed less skilled women as a result of the increase in labor supply among welfare recipients. These effects may be avoided only if the demand elasticities for less educated women are quite large and the labor market clears quickly. Given existing evidence on the behavior of low-wage labor markets, this seems unlikely. Bartik's chapter raises a number of key theoretical and policy questions. An uncomfortably large range of "reasonable" estimates for key parameters can be used to predict the impact of welfare reform on wage and employment outcomes of less skilled workers. It is unclear how best to model the labor market for less skilled women. Do increases in the labor supply among this population induce demand changes that might permanently increase their job opportunities (as appears to happen with new immigrants)? Is the labor market for less skilled workers close to a zero-sum game between new entrants and similar women I

7

Finding Jobs

who were already working? At the present time there is no clear way to choose one model over another; analysts are thus left with a disturbingly wide range of potential impacts. Future research, based on what actually happens in the aftermath of welfare reform, may help us choose better among these models. From a policy perspective, Bartik's findings provide information on the labor market costs of welfare reform. An influx of less skilled workers looking for jobs may have a negative impact on the earnings and employment opportunities of other less skilled workers that is particularly large in certain local labor markets. For instance, inner-city areas may experience bigger negative effects of welfare reform than will medium-size towns and suburbs.

THE LABOR MARKET FACED BY LESS SKILLED WORKERS Those who want to help move welfare recipients into economic self-sufficiency must be concerned about the nature of the jobs available to less skilled workers. For instance, welfare-to-work transitions will be harder if these jobs are markedly less stable, provide fewer rewards to experience, or provide far less compensation because they limit nonwage benefits. Part II explores these questions.

Why Are Unemployment Rates of Less Skilled Workers So High? It is sometimes argued that low-skilled workers have less difficulty actually finding a job than they do in keeping a job. In chapter 3, Harry Holzer and Robert LaLonde investigate the dynamic behavior underlying the relatively low employment rates-and correspondingly high unemployment rates-experienced by younger and less educated workers. They find that unskilled workers differ from more highly skilled workers both in the rate at which they find jobs when they are out of work and in the rate at which they lose jobs and enter unemployment. Moreover, less skilled workers who lose a job are more likely to exit the labor market altogether rather than move to a new job immediately or begin an active job search. These findings suggest that differences across skill groups in both job-finding and job-retention rates are important, although the gap in jobfinding rates accounts for a larger share of the overall employment gap. Holzer and LaLonde find some evidence that both workers with little education and those with higher education gradually age out" of job instability. Finally, they note that longer duration of employment in previous jobs appears to coincide with increased stability in the current job. From a policy perspective, these results indicate that attention should be directed both to increasing the rate of job acquisition of welfare recipients and other less skilled workers and to lowering the likelihood of job loss. It may also be important to provide assistance in locating a new job if the current job ends, as a way of helping people avoid having to exit the labor market and begin a period of welfare dependence. II

8

I

Introduction

How Much Wage Growth Can Less Skilled Workers Expect over Time? Economists typically assume that wages grow with job experience. According to this reasoning, working today will yield higher wages tomorrow and increase the prospects for self-sufficiency among former welfare recipients. Some analysts have argued that the payoff to work experience can be quite substantial, while others have argued that less skilled workers rarely (if ever) experience substantial wage growth. In chapter 4, Tricia Gladden and Christopher Taber investigate the extent of wage growth among less skilled workers. An innovative feature of their analysis is the use of detailed data on actual job experience rather than imputed or "potential" experience based on age. Remarkably, they find that different skill groups experience about the same percentage growth in wages per year of actual work experience. Moreover, less skilled women experience about the same wage growth per year of work as less skilled men. Differences in accumulated wage growth over time among these groups reflects differences in their actual accumulated work experience, not differences in the returns to work experience. The evidence in this chapter is the best available research on this topic, and it suggests that less skilled workers who remain employed can indeed expect steady, albeit relatively modest, wage growth. Gladden and Taber also investigate the role of job mobility in wage growth. They find that high school dropouts who change jobs voluntarily once a year experience higher wage growth than those who stay in their old job. More frequent job changes lead to lower wage growth, however, and involuntary job changes lead to wage declines. These patterns suggest that many low-skilled workers can achieve higher wage growth through selective job changes. Indeed, if low-skill jobs are concentrated in small firms or in establishments with only limited job ladders, job changing may be the best way to increase wages over time. These results reinforce those presented in the previous chapter and emphasize the importance of long-term attachments to the labor market. They also imply that job changing per se is not a problem for low-wage workers and may in fact be a key to sustained wage progression. Joining these to the findings in chapter 3, one can conclude that effective job-training programs are those that help less skilled workers enter the labor market, maintain steady employment, and find new jobs in the event of a job dissolution and also prepare them to actively seek better employment opportunities when and where they are available.

How Are Labor Market Outcomes Different for Less Skilled Women and Less Skilled Men? The recent policy focus on welfare reform has led to an new interest in the lowwage labor market for women and how it may differ from the labor market for men. In chapter 5, Susan Mayer and Jane Waldfogel investigate differences in wages among men and women by skill group and the evolution of these differ-

I

9

Finding Jobs

entials over time. The past two decades were particularly difficult for younger, less educated men, whose real wages declined by 10 to 20 percent. Less skilled women have not fared as badly, and consequently the gender gap in wages among the less educated has narrowed steadily. The employment gap has also closed somewhat, partly because of declines in the labor force participation of less skilled men and partly because of rises in the participation of less skilled women. Using a standard model of human capital earnings, Mayer and Waldfogel show that the narrowing of the gender wage gap has occurred largely through a convergence in the relative rewards received by women for higher education, longer work experience, and other characteristics. In addition, less skilled men have been hurt by their disproportionate representation in certain sectors (for example, heavy manufacturing) that have experienced difficulties in recent decades. Mayer and Waldfogel underscore the fact that although the gap in wages between less skilled men and women has narrowed, the real earnings of younger, less educated women have not risen appreciably over the past two decades. Rather, the gender gap has narrowed because of the fall in real earnings for less skilled men. In this light, policies such as the EITC may have played a critical role in determining the relative well-being of many female-headed families and of families headed by couples with low levels of education. Without increases in hours worked or in the real value of supplementary benefits available to less skilled workers, their income position would not have improved over the past two decades. A further interesting result in this chapter is Mayer and Waldfogel's finding that the historically negative impact of children on the wages of more skilled women has declined over time, but that children have a greater negative effect on the wages of less skilled women now than in the past. The latter may be indicative of the relative difficul ties faced by lower-wage women in obtaining reliable child care.

Can less Skilled Workers Obtain Health Insurance Through Their Jobs? Just as higher wages make employment more attractive, so does the availability of health insurance. Because most welfare recipients are automatically eligible for Medicaid (public health insurance), going to work can lead to the loss of health benefits, at least for the adults in a family (children in low-income families are covered by Medicaid even if the family receives no public assistance). In chapter 6, Janet Currie and Aaron Yelowitz explore the issue of health insurance availability for less skilled workers. They find that the extent of private employer-sponsored health insurance declined from the mid-1980s through the early 1990s but has been stable in the late 1990s. These changes in the availability of private health insurance seem to be largely explained by a combination of three factors: a general deterioration in the labor market opportunities for less skilled workers, rising health insurance costs, and" crowd-out," the process by 10

/

Introduction

which availability of public health insurance leads employers and employees to drop private health insurance. Currie and Yelowitz emphasize the role of nonwage issues in encouraging employment among less skilled workers. State welfare-to-work programs typically allow families to retain their Medicaid eligibility for some period of time. A small but growing number of states are experimenting with more extensive ways of providing health insurance to low-wage workers. Subsidized health insurance-like child care subsidies-can be viewed as a form of wage subsidy, which increases the relative attractiveness of employment for workers in lowwage jobs.

POLICIES TO INCREASE EMPLOYMENT AND EARNINGS OF LESS SKILLED WORKERS A major focus of policy design at both the federal and state level has been to make work more economically rewarding for less skilled workers. Wage subsidy schemes (including child care subsidies) as well as other financial incentive programs are one way to do this. In the absence of private sector demand, the provision of public sector employment is another way to encourage employment among the less skilled. Part III reviews the current evidence on the effectiveness of these programs.

Can Subsidies-Either to Workers or to Employers-Improve Outcomes for Less Skilled Workers? One way to improve labor market opportunities for less skilled workers is to offer subsidies to employers who hire certain workers or to workers who find suitable employment. In chapter 7, Stacy Dickert-Conlin and Douglas HoltzEakin use an analytical framework to compare these subsidy schemes and discuss the pros and cons of the two approaches. A key concern is the question of targeting: will subsidies to low-wage workers raise the incomes of low-income families? Dickert-Conlin and Holtz-Eakin present a variety of data showing that many low-wage workers are in middle-income families. This means that an employerbased subsidy scheme has to rely on other eligibility criteria-over and above the wage rate or earnings offered on a job-to effectively target individuals in low-income families. Such eligibility criteria may stigmatize potential employees. In contrast, worker-based subsidies through the tax system (such as the EITC) can target low-income families directly. The certification problems with targeted employer-based subsidies are one reason take-up rates in these programs are far lower than participation rates in the employee-based EITe. In addition to the issue of targeting efficiency, Dickert-Conlin and Holtz-Eakin's analytical framework suggests other differences between an employee-based subsidy scheme and an employer-based scheme, which they also discuss. I

11

Finding Jobs

There is evidence that both employer-based and employee-based subsidies can raise the labor force participation of targeted groups. The evidence on the positive effects of the EITC on the labor supply of single mothers over the past decade is particularly compelling. It is also worth noting that the EITC should raise the overall income levels and change the income distribution between lower- and higher-wage workers, although this is an effect that is hard to analyze with available data (the chapter cites evidence on the extent to which the EITC has lowered poverty rates). If the Bartik chapter suggests that there may be negative effects on wages or unemployment from increasing work effort among less skilled women, this chapter discusses policy options that might plausibly offset some of those negative effects.

What Can We Learn from Past Efforts to Stimulate Demand Through Public Sector Employment Programs? An alternative way to bolster the labor market opportunities of less skilled workers is through the creation of public sector jobs. David Ellwood and Elisabeth Welty analyze past and present public sector employment (PSE) programs in chapter 8, providing a comprehensive discussion of the design issues and potential effects of PSE programs. Although the record suggests that some PSE programs have been wasteful and inefficient, leading mainly to the "displacement" of preexisting public sector jobs, the authors argue that other PSE programs have been able to increase employment and produce socially valuable output with modest displacement effects. In particular, they argue that displacement effects are minimized if the program is targeted at workers with characteristics different from those of regular public sector workers (for example, less skilled), if the program offers only short-term employment, and if PSE employees perform jobs that are substantially different from regular public sector jobs. Only a handful of states are currently trying to implement PSE programs as part of their welfare reform efforts. Given the evidence in chapter 1 showing the cyclical sensitivity of labor market opportunities for less skilled workers, however, more states may become interested in PSE programs during the next economic downturn as a way to maintain work effort among welfare recipients. This chapter provides useful guidance to states on the design of effective PSE programs, indicating the trade-offs they will face in program design and providing evidence on how past programs have been more or less effective.

Can Financial Incentives Stimulate Greater Work Among Less Skilled Workers? Over the years, a variety of financial incentives have been used to promote work among less skilled workers. Chapter 9, which we co-wrote with Philip Robins, reviews the evidence on the efficacy of a new set of financial incentive programs, some of which have been designed as part of larger welfare-to-work efforts. 12

I

Introduction

These programs range from the EITe (discussed in a slightly different context in chapter 7), to enhanced disregards of earnings for welfare recipients, to alternative assistance programs that operate outside the realm of regular welfare, in the mold of the negative income tax proposals of the 1970s. There is clear evidence that financial incentive programs-which essentially raise the percentage of any additional earnings that a welfare recipient or former recipient can keep for herself or himself-can increase work effort, raise incomes, and lower poverty. A large number of states have adopted various financial incentive schemes as part of their TANF-funded public assistance programs, and so over time there will be additional information on the effects of these programs. A key issue in the design of these programs is the extent to which so-called "windfall" beneficiaries-individuals who would have been working even in the absence of the incentives-can participate and claim benefits. Making benefits available to windfall beneficiaries increases the antipoverty effects of the program but may substantially raise its costs, as well. Several of the new programs limit costs by targeting long-term welfare recipients (rather than the general population of low-wage workers) or by restricting eligibility to full-time workers. Because these financial incentive programs are typically more expensive to run than existing welfare programs, states are forced to make a choice among policy priorities. If a state places a priority on poverty alleviation, a broad-based financial incentive program available only to full-time workers can have a significant income effect without leading to reductions in work effort. If a state is more concerned with budget expenditures than poverty alleviation, then it may find a narrowly targeted financial incentive program more appropriate. In any case, these programs provide options for states that clearly reinforce the effects of more traditional welfare-to-work programs. In fact, there is evidence that combining financial incentive programs with job search requirements and job support services can lead to greater employment and income gains than either program would produce alone.

How Do Child Care Costs Affect Employment? Any discussion of welfare reform and the goal of higher employment among single mothers must inevitably confront the issue of child care. Welfare reforms over the past few years have led to large increases in the child care subsidies available to low-income women in most states. In chapter 10, Patricia Anderson and Phillip Levine investigate how child care subsidies influence employment outcomes of single mothers. Low-skilled women are less likely to use paid child care and pay less for the care that they do use. Yet, child care expenses are still a higher fraction of earnings for less skilled women than for women with higherlevel job skills. Anderson and Levine estimate that the elasticity of labor force participation with respect to the price of child care is between - 0.06 and - 0.36. This narrows the range of estimates substantially compared with earlier studies and suggests that child care subsidies can have a significant but modest effect on I

13

Finding Jobs

labor force participation among women, although the effects are larger for less skilled women and for unmarried mothers with young children. Anderson and Levine's findings underscore the importance of nonwage issues for low-skilled workers. If less educated mothers have to pay full market rates for child care, their net wage may be close to zero. Subsidized child care is potentially as important as other types of wage subsidies in providing an incentive for increased work among former welfare recipients.

THE IMPACT OF WELFARE REFORM The specific institutional details that were part of the welfare reform legislation can have an impact on who is affected and how. In part IV, the final two chapters in this book explore the impact of two controversial aspects of the 1996 legislation, namely, cuts in public assistance eligibility for immigrants and the implementation of time limits for TANF-funded services.

How Do Immigrants and Their Children Use Work and Welfare? One group that has been greatly affected by welfare reform is immigrants. In addition to converting AFDC into the TANF block grant, the 1996 Personal Responsibility and Work Opportunity Reconciliation Act made sweeping changes to rules governing the availability of food stamps and Supplemental Security Income (SSI) to immigrants. In chapter 11, Kristin Butcher and Luojia Hu compare the participation rates of immigrants and natives in means-tested welfare programs and explore the sources of the differences. Butcher and Hu find that immigrants are more likely than the native born to use transfer programs, particularly in-kind transfers (food stamps or Medicaid). Once observable characteristics such as education and location are taken into account, however, immigrants are less likely than the native born to participate in these programs. Butcher and Hu then extend their analysis to study utilization of public assistance by second-generation immigrants (that is, the U.S.born sons and daughters of immigrants) compared with natives. They find that second-generation families are less likely to utilize public assistance than the native born with similar characteristics. They also find little evidence that by itself the use of welfare by immigrants makes their children any more likely to use welfare. The results in this chapter provide little indication that immigrants abused the public assistance system before the 1996 legislative changes; indeed, immigrants used public assistance programs less than natives with similar individ ual characteristics. Hence, major cuts in immigrant access to these programs can be justified only by budgetary pressure, not by abnormal or objectionable behavior on the part of immigrant recipients. There is also little evidence that the second generation will utilize public assistance more extensively than people whose families have been in the United States for three or more generations, although this conclusion may change as the characteristics of immigrants and their countries of origin change over time.

14

I

Introduction

What Is the Projected Impact of TANF Benefit TIme Limits on Less Skilled Workers? One of the most controversial and widely discussed aspects of the 1996 welfare reform legislation was the imposition of a five-year time limit on the receipt of program benefits funded by the TANF block grant. In chapter 12, Robert Moffitt and LaDonna Pavetti use historical welfare participation data, as well as recent program data from the states, to assess the likely effects of these time limits. Microeconomic data from the era of the AFDC program suggest that about 40 percent of welfare recipients will hit the five-year time limit over an eight-year period. Younger, less educated, and never-married women are most likely to experience long spells of welfare use. Of course, if TANF recipients respond to the threat of time limits by taking work requirements more seriously, many fewer than 40 percent will be affected by time limits. Some states have adopted maximum benefit periods that are even shorter than the sixty-month limit prescribed in the federal legislation; but in only a few states have individuals actually begun to hit these limits. Moreover, some of these early" states have found ways to exempt families that are running into time limits. Most importantly, many states have begun to utilize their new powers under the TANF legislation to deny continued benefit eligibility to people who fail to meet job search requirements or other work criteria. These" sanctioning" policies potentially eliminate many long-term welfare recipients who would otherwise hit the five-year time limit. Thus, it may be that the threat of sanctions will have a greater effect on welfare durations and caseloads than time limits per se. The introduction of time limits raises the prospect of a large influx of former welfare recipients into the labor market, if and when substantial numbers of women hit the limits. Currently, however, it is unclear exactly how many people will be subject to time limits. Beyond the obvious monitoring and enforcement problems, states have some flexibility to avoid imposing limits on women they believe are satisfying the program rules. In the extreme case, they can cover their benefit costs out of non-TANF funds. Women who are not cooperating with employment programs are likely to be sanctioned well before their time limit. The potential interactions between time limits and labor market participation are still unclear, however, and a fuller analysis of the impact of time limits is an important future research agenda. U

WHAT DOES THIS RESEARCH TELL US, AND WHAT DO WE STILL NEED TO KNOW? The research findings in this volume contain both good and bad news about the long-run prospects for welfare reform. On the positive side, there is clear evidence that certain policies-including wage subsidies (chapter 7), public sector employment (chapter 8), and financial incentive programs (chapter 9)-can increase employment and hours and simultaneously raise the incomes of less

I

15

Finding Jobs

skilled workers. Variants of these programs are already operating federally and in many states, and we expect to see widening use of financial incentives, and perhaps wage subsidies and public sector employment programs, in the next few years. The findings on job stability (chapter 3) and wage progression (chapter 4) indicate that low-wage workers who can find and keep jobs will experience modest wage growth over time. Workers who are able to voluntarily move between jobs will likely experience even greater wage gains. All of this suggests that it is possible to increase the earnings and work effort of younger and less educated parents who have traditionally relied on public assistance programs. Nevertheless, wage subsidies and financial incentives (such as the EITC and enhanced earnings disregards) still involve transfer payments to low-skilled workers. Unlike traditional welfare programs, however, they are linked to work effort and may be more politically palatable. On the other hand, there are also a number of cautionary lessons in these chapters. The greater cyclicality of the low-wage labor market (chapter 1) suggests there may be real problems maintaining employment gains during an economic downturn. Welfare reform may also have some displacement effect on other less skilled workers (chapter 2). So far these effects may have been masked by the strong economic climate of the late 1990s. A further cautionary note is that the long-run success of welfare-to-work efforts may depend on more than just the effectiveness of job search programs and initial job placements. The availability of child care (chapter 10) is important to employment gains; declines in the number of jobs offering health insurance to less skilled workers (chapter 6) may also cause problems. Wages of less skilled women have not risen much over the past two decades, and wages of less skilled men have actually declined (chapter 5). In the presence of these trends, the increased availability of wage supplements (particularly the EITC and expansions in child care subsidy dollars) has been one of the few bright spots for individuals and families at the bottom of the skill distribution curve. Finally, it is clear that simply getting workers into a job may not assure stable employment; less skilled workers experience greater problems in job instability and more difficulty in reentering employment after a job loss (chapter 3). Many questions remain. First, this volume focuses entirely on the labor market side of employment programs and does not investigate how these programs are affecting family formation, family behavior, or child-rearing patterns. (In part, we ignore these issues because other major projects are focusing on them.) Second, it is clearly too early to do any real long-term evaluation of the 1996 welfare reform legislation. Some parts of the legislation (such as time limits) are still being implemented. Other aspects of welfare reform are particularly hard to assess in a period of extraordinarily strong economic growth and low unemployment. The real long-term effectiveness of welfare-to-work efforts will be revealed only in the next economic downturn, when we can observe what happens to employment, income, and welfare participation among those who have left the welfare rolls over the past several years. If we have simply produced greater

16

I

Introduction

cycling in and out of the labor market, rather than long-term labor market gains, recent reforms will be seen as less effective. Third, we are just beginning to see preliminary results from studies that track those who leave public assistance in the midst of the currently falling caseloads. s (A harder research problem is to identify and track those who would have sought public assistance benefits in the past but do not even apply for public assistance in the current environment.) Such studies will investigate the extent to which adults find and keep jobs over time and how their income changes over time. These studies are crucial to any evaluation of the net effects of the 1996 welfare reform legislation. Fourth, many specific policy interventions need to be more fully evaluated. The interaction between employment programs and financial incentive programs needs to be better understood. Actual demonstration projects that try to experimentally estimate the impact of subsidies for child care and health care on employment behavior would provide much more reliable evidence on the value of these subsidies to low-wage workers. Greater experimentation with and evaluation of public sector employment programs would provide a better sense of "best practices" in this area. Similarly, we need more experimentation with the mechanics of effective programs that assist workers in maintaining jobs or in searching for new and better jobs in the low-wage labor market. Better information on whether certain occupations or industries offer greater promise of wage growth or wage-enhancing job changes would help assist those who provide job counseling on welfare-to-work programs. The enormous range of experiments currently under way in different states' assistance programs is both a boon and a bane for researchers. On the one hand, we have the promise of many future "natural experiments" as states adopt a wide variety of alternative programs. These "experiments" may be particularly useful for answering labor market questions, because states will differ in both their economic growth rates and in the nature and type of the TANF-funded programs they implement. On the other hand, the growing diversity of public assistance programs across states means that much analysis will have to occur at the state level. This will require detailed information on state program parameters, as well as data sets that are sufficiently large to measure state-level impacts. Researchers will face the dual challenge of finding new data sources and trying to compare the outcomes of a vast array of different programs. So far, the evidence suggests that welfare reform is proceeding as well as or better than most analysts had expected. In terms of declining caseloads and increasing work effort among single mothers, welfare reform has been an astonishing success. The evidence is still preliminary, however. Although it is clear that we have run some effective programs that have increased employment among the least skilled, it remains uncertain exactly which gains will be maintained over the business cycle. Only over time will we be able to do the detailed studies of the income changes, job changes, family changes, and wage changes experienced by welfare recipients after they enter the labor market and leave public assistance. The research in this book suggests that we are on the right

/

17

Finding Jobs

track with many of our policy efforts. However, maintaining good policies and adapting them to a changing economic environment will be a challenge in the years ahead.

NOTES 1. For a more detailed description of welfare reform and the process leading up to it, see Blank (1997). 2. For a description of the nature of these changes, see Gais and Nathan (1998). 3. Data on unemployment rate by education level are only available starting in 1993. 4. The hourly wage data in figures 1.2 and 1.3 are based on data reported in chapter 3 of the Council of Economic Advisers (1999), using Outgoing Rotation Groups from each month of the Current Population Survey. Each month's data is deflated by the Consumer Price Index (CPI-UX1) and then averaged across twelve months to produce a annual wage number. Included are all persons sixteen and over who report wage and salary income during the survey week. 5. For instance, see Card and Krueger (1995) for evidence on the 1990 and 1991 increases. See Bernstein and Schmitt (1998) for evidence on the 1996 and 1997 increases. 6. See Wallace and Blank (1999). 7. See Eissa and Liebman (1996) and Meyer and Rosenbaum (1999). 8. For instance, see the recent review of studies by the U.s. General Accounting Office (1999).

REFERENCES Bernstein, Jared, and John Schmitt. 1998. Making Work Pay: The Impact of the 1996-1997 Minimum Wage Increase. Washington, D.C.: Economic Policy Institute. Blank, Rebecca M. 1997. It Takes A Nation: A New Agenda for Welfare Reform. Princeton, N.J.: Princeton University Press. Card, David, and Alan B. Krueger. 1995. Myth and Measurement: The New Economics of the Minimum Wage. Princeton, N.J.: Princeton University Press. Council of Economic Advisers. 1999. Economic Report of the President. Washington: u.s. Government Printing Office. Eissa, Nada, and Jeffrey B. Liebman. 1996. "Labor Supply Response to the Earned Income Tax Credit." Quarterly Journal of Economics 3(2): 605-37. Gais, Thomas L., and Richard P. Nathan. 1998. Overview Report: Implementation of the Personal Responsibility Act of 1996. Albany, N.Y.: Nelson A. Rockefeller Institute of Government. Meyer, Bruce, and Daniel T. Rosenbaum. 1999. "Welfare, the Earned Income Tax Credit, and the Employment of Single Mothers." Working paper 7363. Cambridge, Mass.: National Bureau of Economic Research. U.s. Department of Health and Human Services, Administration for Children and Families. 2000. Data downloaded from the world wide web located at: http://www.acf. dhhs.gov / news/ stats/3697.htm.

18

I

Introduction u.s. Department of Labor, Bureau of Labor Statistics. 2000. Data downloaded from the world wide web located at: http://www.bls.gov/webapps/legacy/cpsatab3.htm. u.S. General Accounting Office. 1999. Welfare Reform: Information on Former Recipients' Status. Washington: U.s. General Accounting Office. Wallace, Geoffrey, and Rebecca M. Blank. 1999. "What Goes Up Must Come Down? Explaining the Recent Changes in Public Assistance Caseloads." In Economic Conditions and Welfare Reform, edited by Sheldon Danziger. Kalamazoo, Mich.: Upjohn Institute.

I

19

Chapter 1 The Employment, Earnings, and Income of Less Skilled Workers over the Business Cycle Hilary W. Hoynes

O

ne of the most substantial risks facing workers is the potential for job loss, either permanent or temporary. The possibility of a loss in earnings and employment is likely to be of greater concern to less skilled workers because of difficulties in replacing lost income with savings and the earnings of secondary earners. Many government transfer programs have been established to reduce the variability of family income over the business cycle. Because of recent changes in welfare programs, however, there is some uncertainty as to the role that the safety net can and will play in subsequent recessions. Recent evidence suggests that state and federal policy changes are leading to increases in employment among recipients of benefits from Aid to Families with Dependent Children (AFDC) (see chapter 9 in this volume). With increases in attachment to the labor market comes the potential for increases in family income and earnings. However, with increasing labor market attachment also comes the risk of recession and loss of family income. The potential for cyclical fluctuation in earnings is very different from the relatively constant transfer that a family could expect from AFOC. This chapter examines the impact of changes in local economic conditions on the employment, earnings, and income of individuals in different skill groups. The skill groups are defined by sex, race, and education level. The emphasis here is on the relative impact of cycles across these demographic groups. Our findings consistently show that the labor market outcomes of less skilled workers exhibit more variability over business cycles than those of higher-skilled groups. Nonwhites and those with lower education levels are more impacted by changes in local economic conditions. Furthermore, high-skilled women exhibit significantly less sensitivity to business cycles than low-skilled women, especially low-skilled nonwhite women. These patterns hold for both recessions and recoveries. These groups are more likely to have reductions in employment and earnings during a downturn and are also more likely to have gains in recoveries. An examination of individuals in isolation, however, gives an incomplete picture of the effect of cycles on well-being. The results of our study also show that government transfers decrease the differences between groups, resulting in more

I

23

Finding Jobs

skill-group-neutral effects of business cycles on family income than individual earnings. Many previous studies have examined the effects of business cycles and local labor markets on individuals and families. That prior work generally shows that less skilled workers are more impacted by cycles. My research contributes to the literature in three important ways. First, this chapter is focused on a comparison of the responses across demographic groups, defined by sex, education, and race. Second, the chapter examines a broad range of individual and family outcomes, including employment, hours, earnings, and income. Most previous studies focus on one or two individual measures. The combination of detailed outcome measures and comparisons across demographic groups allows me to present a very complete picture of the way families are impacted by cycles-a picture that has not been generated by the existing literature. Third, the study examines whether these effects vary across business cycles. This is all done in a simple empirical framework that makes use of variation in the timing and severity of recessions across metropolitan statistical areas (MSAs). This variation is much richer than that used in time-series studies. The focus in the chapter on relative responses across demographic groups and different individual and family outcomes can be used to inform the policy process. The results shed light on particular problems that may exist for less skilled workers, for women versus men, or for whites versus nonwhites. These differences may result from differences in labor market attachment, industry, or geographic location. Furthermore, the examination of different outcome measures may shed light on the success or failure of particular government transfer programs designed to insure against fluctuations in income. For example, policies such as unemployment insurance and job training are geared toward individuals, and their effects will show up in individual outcomes. Other policies, such as Aid to Families with Dependent Children and the Earned Income Tax Credit, are geared toward families, and their effects will show up in family income.

PREVIOUS LITERATURE This study has connections to many different areas of research, including the literature on wage, earnings, and income inequality; trends in employment and earnings for women; determinants of labor market outcomes and differences between groups; and worker displacement. It is neither feasible nor desirable to present here a comprehensive review of the literature. Instead, this review focuses explicitly on those studies that examine the effect of local labor market conditions on employment and income outcomes. In particular, I focus on three features of these studies: the variables used to control for the characteristics of the area labor market, the outcome measures, and the degree to which differences across groups is explored.

24

I

Employment, Earnings, and Income of Less Skilled Workers

The applications that are most relevant to this analysis include those that examine the effect of business cycles and local labor markets on employment outcomes (Bartik 1991, 1993a, 1993b, and 1996; Blanchard and Katz 1992; and Holzer 1991), real wages (Bils 1985; Blank 1990; Keane, Moffitt, and Runkle 1988; and Solon, Barsky, and Parker 1994), racial differences in labor market outcomes (Bound and Holzer 1993 and 1995), labor market outcomes of disadvantaged youth (Acs and Wissoker 1991; Bound and Freeman 1992; Cain and Finnie 1990; Freeman 1982, 1991a, and 1991b), and family income, poverty, and income inequality (Bartik 1994; Blank 1989; Blank 1993; Blank and Blinder 1986; Blank and Card 1993; and Cutler and Katz 1991). These studies almost universally find an important role for local labor market conditions. The studies of disadvantaged youth relate labor market outcomes to local (typically, MSA) unemployment rates. That literature has consistently found that higher local unemployment rates lead to reductions in employment and earnings (Acs and Wissoker 1991; Bound and Freeman 1992; Cain and Finnie 1990; Freeman 1982, 1991a, 1991b), with larger effects for blacks, younger workers, and less educated workers (Acs and Wissoker 1991; Freeman 1991a). The studies of family income and poverty have typically used either national (Blank 1989, 1993; Blank and Blinder 1986; Cutler and Katz 1991) or regional (Blank and Card 1993) variation in unemployment rates or gross national product (GNP). The studies have found a consistent negative relation between unemployment rates and inequality and poverty. Of particular interest is Rebecca Blank (1989), who disaggregates household income into many components and examines the relative cyclicality of the components. She finds earnings and capital income to be procyclical and some transfer income to be countercyclical. Overall, she finds greater variation over the cycle for those who are young, male, and nonwhite. The literature that is most relevant for this study is the literature that uses variation across MSAs in labor market conditions to examine labor market outcomes across different demographic groups (Bartik 1991, 1993a, 1993b, 1994, and 1996; Bound and Holzer 1993 and 1995). The studies by Timothy Bartik use growth in employment, changes in the manufacturing share of employment, and changes in the average wage premium implied by the area's industry mix. John Bound and Harry Holzer (1993 and 1995) use skill-group-specific measures of employment growth, using as weights the skill group's participation in each industry at the beginning of the period. The results differ somewhat across the studies, but they generally show that changes in labor demand lead to larger changes for blacks, younger persons, and those with lower education levels. The patterns seem to hold for men and women. Distinct from the above literature on labor market outcomes are studies that use panel data to examine the cyclicality of real wages. The literature uses primarily aggregate measures of business cycles (national unemployment rates of GNP growth) and asks to what degree aggregate wage fluctuations over the cycle are the result of changes in the composition of the workforce. The results

I

25

Finding Jobs

vary somewhat across the studies but generally find that the composition effect alone leads to countercyclical wage patterns. Accounting for this composition effect, wages are found to be procyclical, with greater fluctuations for those who are male, young, and working in private sector. Overall, these studies raise several possible explanations for the differences across groups in sensitivity to business cycles. An often-cited explanation is variation across demographic groups in mobility rates: the larger the long-run supply elasticity for the demographic group, the lower the expected effect of a demand shift on wages and employment. Those with lower rates of population mobility will have larger effects. A second explanation is that different demographic groups tend to be employed in different sectors and occupations that may be associated with greater or lesser risks of layoff.

DATA The study uses the Outgoing Rotation Group (ORG) (covering the period from 1979 to 1992) and the March Annual Demographic File (ADF) data (covering the period from 1975 to 1997) from the Current Population Survey (CPS). The advantage of the ORG data, which pools monthly survey observations, is that its samples are about three times the size of the ADF sample-which is particularly important when presenting results by skill group within MSAs-but the labor market outcomes included in the survey are limited. I use indicators for employment in the previous week, full-time employment in the previous week, and earnings in the previous week, where "full-time employment" denotes a workweek of at least thirty-five hours. The data covers the period from 1979 to 1993, with about 325,000 observations each year. 1 Ultimately, broader measures of individual and family well-being are important. The ADF provides comprehensive data on employment, earnings, and income over the past year. The analysis of the ADF data uses hourly earnings, annual hours, annual earnings, family earnings, family transfer income, and total family income. When evidence from the two data sources is combined, the results tell a comprehensive story about the impact of business cycles on workers and families. The data also cover a relatively long time period, allowing for examination of the recessions over three decades. The ADF (or March CPS) includes labor market and income information for the previous year, at the individual and family level. Many different measures of individual and family outcome are considered, including whether household members were working, whether the work was full time all year, number of weeks worked, average hourly earnings, annual hours, annual earnings, family earnings (head and spouse), family transfer income, and total family income. All measures are annual and correspond to the calendar year previous to the survey. "Full time" is defined as at least thirty-five hours a week in the previous year,

26

I

Employment, Earnings, and Income of less Skilled Workers

and full year" is defined as fifty or more weeks in the previous year. The ADF data is available beginning with the 1964 survey year. Because of major changes in the survey beginning in 1976, this study uses the surveys from 1976 to 1998, covering the years from 1975 to 1997. 2 The sample size is approximately 150,000 persons a year. The earnings data are "top-coded" in both surveys.3 In the aRC data, weekly earnings are top-coded at $999 through 1988 and $1,923 from 1989 on. In the ADF data, annual earnings are top-coded at $50,000 through 1981, $75,000 from 1982 to 1984, $100,000 from 1985 to 1988, and about $200,000 from 1989 on. Following Lawrence Katz and Kevin Murphy (1992) and more recently Francine Blau (1998), the earnings of top-coded individuals are adjusted to be 1.45 times the top-coded value. Beginning in 1996, instead of giving each top-coded observation the value of the top-code, the CPS assigns the mean among the sample of top-codes (by demographic group). The earnings figures can be as high as $600,000 in this period. T make no adjustment for top-coding in these years. There is no apparent top-coding of family earnings or family income. Real earnings and income are constructed using the CPT-U-X1 deflator. For most of the analysis, the micro economic data is collapsed into cells defined by MSA, year, and skill group. Skill groups are defined by education, race (white, nonwhite), and sex. The nonwhite group includes both blacks and white Hispanics. The ORC data identifies forty-four MSAs, whereas the ADF data (beginning a few years earlier) identifies thirty-five MSAs. In order to better approximate labor market areas, the MSAs are combined in their consolidated MSA (CMSA) units where applicable. Examples of CMSAs include New York, Los Angeles, and Chicago. The final sample includes thirty-five MSAs or CMSAs in the ORC data and twenty-seven MSAs or CMSAs in the ADF. For the remainder of the chapter, these geographic units will be referred to as MSAs. In 1990, the MSA sample accounts for about 60 percent of the total metropolitan population (or 50 percent of the total population). The sample accounts for virtually all of the metropolitan population in 1975. The lack of complete coverage in the later part of the period comes from the need to create metropolitan areas that are consistent geographic units over the entire time period. Thus, metropolitan areas that are added in the middle of the period, for example, are not included in the sample: The median MSA in the ORC data contains about two hundred observations a year, compared with about seventy-five observations a year in the ADF. Once the cells are further refined to skill groups, some cells get quite small. When possible, data are combined into two-year periods to reduce the problem of small skill group, MSA, and year cells. All analyses in this chapter are weighted. s The same sample selection criteria are applied to both the aRC and ADF data. The sample includes persons between twenty-two and sixty-two years of age. The self-employed, those working without pay, and those with positive earned income but zero hours of work are excluded. Following Katz and MurII

I

27

Finding Jobs

phy (1992), we exclude individuals with real weekly earnings of less than $67 in 1982 dollars (that is, one half of the value of the minimum wage assuming a forty-hour workweek). The final sample has about 220,000 observations a year in the ORC sample and 70,000 observations a year in the ADF sample.

TRENDS IN LABOR MARKET OUTCOMES AMONG SKILL GROUPS This period is characterized by important secular trends in employment and earnings as well as containing three recessionary periods. The experiences vary dramatically across demographic groups.

Definitions Skill groups are defined by education, race, and sex. "Low-skilled workers" are typically defined using education level, and the term most often denotes persons with less than a high school education. This analysis uses data covering a period of three decades and is concerned with making comparisons across groups, over time, and across cycles. It is important for the analysis that the skill groups are defined to be relatively comparable over time. However, education levels have been rising over time for all demographic groups. In the presence of rising education levels, even if the distribution of earnings and income are unchanged over time, one would expect that the relative position of persons with low levels of education (for example, high school dropouts) would decline over time. That is, over time, this group would become more and more disadvantaged. To illustrate this point further, figures 1.1 to 1.4 present trends in the percentage of persons with various education levels in the ADF sample, by race and sex. These figures show that the percentage of persons with less than a high school education has fallen dramatically in this period. For example, between 1975 and 1997, the percentage of white men with less than a high school education declined from 25 to less than 10 percent (figure 1.1). The percentage of nonwhite men with less than a high school education declined from 50 to 30 percent (figure 1.2). During the same period of time, among white men, the percentage with only a high school diploma has not changed significantly, and the percentage of those with greater than a high school diploma has increased. Among nonwhites, both groups-those with only a high school diploma and those with education beyond high school-are increasing. The trends are even more dramatic for women (figures 1.3 and 1.4).6 In the presence of these increases in education levels over this period, the main analyses in this chapter compare those with a high school education or less

28

I

FIGURE 1.1

/

Distribution of Population by Education, White Men, 1975 to 1997

--!:r- High school only -0- Less than high school --0- More than high school

60

40

20

o 1975

1985

1995

Source: Annual Demographic Files.

FIGURE 1.2

/

Distribution of Population by Education, Nonwhite Men, 1975 to 1997

--!:r- High school only -0- Less than high school --0- More than high school

60

o 1975 Source: Annual Demographic Files.

1985

1995

FIGURE 1.3

/

Distribution of Population by Education, White Women, 1975 to 1997

----l:r- High school only -0- Less than high school --D- More than high school

60

40

20

o 1975

1985

1995

Source: Annual Demographic Files.

FIGURE 1.4

/

Distribution of Population by Education, Nonwhite Women, 1975 to 1997

----l:r- High school only

-0- Less than high school

60

-{]- More than high school

o 1975 Sou rce: Annual Demographic Files.

1985

1995

Employment, Earnings, and Income of Less Skilled Workers

with those with more than a high school education. Those with a high school education or less are defined to be less skilled workers. This group will be less disadvantaged than high school dropouts, however, and where possible I examine outcomes across all four education groups (less than twelve years, twelve years, thirteen to fifteen years, more than sixteen years)?

Trends in Employment, Earnings, and Income, Using Annual Demographic Files Figures 1.5 to 1.8 present trends in the ratios of employment to population (EPOP) for men between 1975 and 1997 by race and education. There are two definitions for the employment to population ratios. "EPOP: Any Work" represents the employment-to-population ratio for those who worked at all the previous year. "EPOP: FTYR" represents the employment-to-population ratio for those who worked full time (at least thirty-five hours a week) and for the full year (at least fifty weeks) during the previous year. Several observations can be noted from these simple figures. As expected, EPOP ratios are higher for those with higher levels of education. Among less educated men, nonwhites tend to have higher EPOP ratios than whites. A striking trend in the figure is the declining employment-to-population ratios among men with less than a high school education. By the mid-1990s, fully 30 percent of men are not working at all over the year. This is undoubtedly in part a result of the changing composition of the lowest education group over this time period. The graphs also provide insight into the impact of cycles on different groups. The figures suggest that employment rates of those with lower education levels and nonwhites exhibit more cyclical variation. There seems to be more cyclical fluctuation in the full-time employment rates (figures 1.7 and 1.8) than in the "any work" employment rates (figures 1.5 and 1.6). For nonwhites with less than a high school education, the EPOP: Any Work graph (figure 1.6) also varies significantly over the cycle. This is striking given that the measure is any work in the entire calendar year. The high rate of nonwork in the trough of the recession in the early 1980s is consistent with the persistently high unemployment rates for this group. Note that the variability in the measures for nonwhites reflects small sample sizes, especially for higher education groups. Figures 1.9 to 1.12 present trends in annual hours worked and earnings for the same groups of men. It is important to note that these earnings and hours figures are averages for all individuals in the group under study, which includes both workers and nonworkers. Therefore, the change in earnings is comprehensive and reflects changes in hours, weeks, and hourly wages as well as changes in the composition of the work force. In general, the pattern for annual hours worked is similar to the trends for the EPOP ratios. These figures show that, to a greater extent than in other measures, both annual hours and real annual earnings show cyclical variation for coUegeeducated white men. Whereas the average hours and earnings of less educated

I

31

Finding Jobs

individuals also show cyclical variation, the relative variability of hours and earnings from low-level to high-level education groups is less dramatic than the variability in the employment figures. In graphs not presented here, the variation in annual hours worked comes more from variation in weeks worked each year than from hours worked each week. That may be attributable to measurement error in hours worked each week, or it may reflect the nature of employment reductions that firms elect to implement. 8 Figures 1.13 to 1.20 present similar figures for women. These figures show that employment and earnings for women are increasing secularly over this period for all groups, but at a substantially slower rate for nonwhite women and those with low education levels. These trends are so strong that it is difficult to make any inferences about the variation over the business cycle. Although not shown here, family earnings and family income also show cyclical variation. Family income and, to a lesser extent, family earnings show less variability across demographic groups compared with the fluctuations in individual earnings. Our empirical model explores the reasons for this difference.

EMPIRICAL MODEL The goal of this analysis is to estimate how individuals in different demographic groups are affected by changes in macroeconomic conditions. One approach to estimating this effect is to take the time-series trends presented above and regress the outcomes on a measure of the business cycle, such as the unemployment rate. This approach is not taken here for two reasons. First, aggregate measures of business cycles do not necessarily capture the relevant cycle if there is area variation in the timing or severity of the cycle. Second, the unemployment rate (or some other aggregate measure of employment) can be mechanically related to the dependent variable (for example, the EPOP ratio for less skilled persons); this reflection or endogeneity problem makes the interpretation of such estimates difficult. One approach used in the literature is to use an instrumental variables method to account for endogeneity in the unemployment rate (Bound and Holzer 1993 and 1995). As an alternative, this analysis treats the shock to a local area as unobserved and compares the response to the shock among different groups. This avoids the reflection problem and has the added advantage of differencing out an MSA effect. All of the comparisons across groups are made within MSAs, which takes advantage of the wide regional variation in the timing and severity of recessions. For this and all remaining analyses in this chapter, I start by collapsing the data into cells defined by MSA (m), time (t), and skill group (j). Let Yjmt be the mean of a given labor market outcome for group j in area m in year t. Suppose one could observe some exogenous measure of the business cycle in the MSA in time t, represented by Ymt. Putting the variables Y in (Text continues on p. 41.)

32

I

Employment, Earnings, and Income of Less Skilled Workers FIGURE 1.5

/

Annual White Male Employment Outcomes, EPOP: Any Work, by Education, 1975 to 1997

1.0

C1J

co

0.9

.lS

cC1J

u

....

~ 0.8

0.7 1975

D Recession ---fr-

FIGURE 1.6

/

High school. diploma

1985 -0- Less -0-

1995

than high school diploma

Some college

-+- College degree

Annual Nonwhite Male Employment Outcomes, EPOP: Any Work, by Education, 1975 to 1997

1.0

C1J

co

0.9

.lS c

~

~ 0.8

0.7 1975

D Recession ---fr-

High school diploma

1985 -0- Less -0-

1995

than high school diploma

Some college

-+- College degree

Source for figures 1.5 and 1.6: Annual Demographic Files. Note: EPOP: Any Work is the employment-to-population ratio for those who worked at all during the previous year. EPOP: FTYR is the employment-to-population ratio for those who worked full time (at least thirty-five hours a week) and a full year (at least fifty weeks) during the previous year.

33

Finding Jobs FIGURE 1.7

/

Annual White Male Employment Outcomes, EPOP: FTYR, by Education, 1975 to 1997 1.0 0.9

(I.)

00

0.8

nl

"E 0.7 (I.)

u

....

(I.)

0....

0.6 0.5 0.4 1975

D Recession -fr-

FIGURE 1.8

/

High school diploma

1985 --0- Less -0-

1995

than high school diploma

Some college

-+- College degree

Annual Nonwhite Male Employment Outcomes, EPOP: FTYR, by Education, 1975 to 1997 1.0 0.9

(I.)

00

0.8

nl

"E 0.7 (I.)

u

....

(I.)

0....

0.6 0.5 0.4 1985

1975

D Recession -fr-

High school diploma

--0- Less -0-

1995

than high school diploma

Some college

-+- College degree

Source for figures 1.7 and 1.8: Annual Demographic Files. Note: EPOP: Any Work is the employment-to-population ratio for those who worked at all during the previous year. EPOP: FTYR is the employment-to-population ratio for those who worked full time (at least thirty-five hours a week) and a full year (at least fifty weeks) during the previous year.

34

I

Employment, Earnings, and Income of Less Skilled Workers FIGURE 1.9 /

Annual White Male Employment Hours, by Education, 1975 to 1997

2,100

~ 1,800

o

:r: ]

1,500

1,200

'--'--'-....I..-L.....J..--'-...L......JL........l..--'--'---'----'---'--'---'----'---'--'---'---'--'---'

1975

o ---fr-

1985 Recession

High school diploma

1995

-D- Less than high school diploma -0-

Some college

-+- College degree

Source: Annual Demographic Files.

FIGURE 1.10 /

Annual Nonwhite Male Employment Hours, by Education, 1975 to 1997

2,100

~ 1,800

o

:r: C;;

j

1,500

1,200

L....L--'-..1........l--L.....L.-l........L.--'-..1........l--L.....L.-L......L--'-..1........l--L.....L.-'--'--l

1975

o ---fr-

1985 Recession

High school diploma

1995

-D- Less than high school diploma -0-

Some college

-+- College degree

Source: Annual Demographic Files.

/

35

Finding Jobs FIGURE 1.11

/

Annual White Male Earnings (1997 Dollars), by Education, 1975 to 1997

60,000

Vi" 1'0.. 0\ 0\

50,000

C !/l

b()

.~

40,000

....

:: (J) u '(J)"' 13.0

- - Female heads of household - - Other females

i.s

5 12.0 2

ro

11.0 10.0 ~ (J) 9.0 S :>, 0 8.0 ~ E 7.0 (J) c 6.0 ~

;:J

5.0 4.0 1979

1981

1983

1985

1987

1989

1991

1993

1995

1997

Year Source: Data are weighted U.S. means from CPS-Outgoing Rotation Group. Note: For group definitions, see figure 2.1.

would have worsened both groups' labor market outcomes but not necessarily the differential between the two groups. The wage curve model (table 2.6) suggests that when unemployment is low, labor supply shocks have larger negative effects on wages, as wages are more sensitive to a given-sized change in unemployment when the initial unemployment rate is low. Because wages adjust more to supply shocks when unemployment is initially low, labor demand also adjusts more to supply shocks, and supply shocks have fewer effect on unemployment. Therefore, the booming U.s. economy may have reduced the effects of welfare reform on hiking the relative unemployment rate of female household heads but would be expected to increase the negative effects of welfare reform on the relative wage rate of female household heads.

Reduced Form Estimates of the Effects of Recent Reductions in Welfare Rolls on Wages and Unemployment Rates Table 2.7 presents reduced form estimates of how reductions in welfare rolls are associated with changes in wages and unemployment for the same five groups analyzed in the wage curve model. The adult population is divided by gender and by education into four groups. A fifth group is separated from the female group without a college degree: female heads of household, age sixteen to fortyfour, with other relatives in the household.'2

106 /

TABLE 2.7

j

Reduced Form Estimates of the Effects of Reductions in Welfare Rolls Due to Welfare Reform on the Wage and Unemployment Rates of Various Groups (1)

(2)

(3)

(4)

Effects on In (Wages) of Reduction in In(Wel£are to Population) Ratio by -0.645 (Maximum Effect in Year 2005 According to Prediction) (Absolute Values of t-Statistics in Parentheses) Female heads Other women with less than college degree Female college graduates Men with less than college degree Male college graduates Average wages

0.065 (8.36) 0.063 (13.69) 0.056 (9.59) 0.088 (16.10) 0.045 (7.92) 0.070 (16.05)

-0.365 (3.43) -0.321 (3.62) -0.278 (3.40) -0.376 (3.52) -0.198 (3.08) -0.324 (3.59)

-0.235 (2.13) 0.036 (0.88) 0.053 (1.04) -0.074 (1.11) 0.036 (0.74) -0.017 (0.37)

-0.195 (1.68) 0.059 (1.09) 0.061 (0.93) -0.001 (0.01) 0.047 (0.69) 0.024 (0.45)

Effect on Unemployment Rate, Defined as [In(Labor Force) - In(Employment)], of Reduction in In(Welfare Receipt) Rate by - 0.645 (t-Statistics in Parentheses)

-0.052 -0.077 -0.094 -0.063 (8.63) (1.94) (1.76) (1.11) -0.021 Other women with less than college -0.037 0.052 0.057 degree (10.67) (1.67) (2.81) (1.92) -0.007 Female college graduates 0.024 0.004 0.007 (4.59) (2.01) (0.28) (0.49) -0.022 Men with less than college degree -0.077 0.034 0.058 (1.27) (9.16) (3.89) (1.52) Male college graduates -0.005 0.024 0.019 0.024 (2.79) (4.38) (1.56) (1.80) Average unemployment rate -0.019 -0.D48 0.030 0.045 (11.06) (3.72) (1.47) (1.58) Note: Column headings are as follows: (1) OLS estimates; (2) instrumental estimates with In (state welfare benefits) as instrument; (3) instrumental variable estimates with state welfare waiver as instrument; (4) instrumental variable estimates with welfare waiver as instrument and with controls for lagged wages and unemployment rates. Five groups are all between age sixteen and sixty-four. Female heads are female household heads without a college degree, with other relatives present, and who are also between the ages of sixteen and fourty-four. Data used in estimation are pooled annual data on fifty states (plus the District of Columbia) from 1979 to 1997. All models include complete set of year and state dummies. All models are estimated using 1979 total state population as weights. The model in the first column is estimated using weighted least squares. The other models use weighted two-stage least squares. The second model uses In(state welfare benefits) as an instrument. The third and fourth models use dummy variable for whether state has welfare waiver as instrument; in 1997, all states are assumed to have waivers. The fourth model includes two annual lags in wage and unemployment rate of group and in overall wages and unemployment rates. All estimates are effects of lowering In(welfarej population) by - 0.645, which is this study's estimate of the effect of welfare reform on welfare rolls in 2005. The fourth model presents dynamic simulation estimates of effects of this shock after five years. The" t-statistics" for this simulation are ratios of parameter values to standard deviation of estimates from 1,000 Monte Carlo repetitions of simulation. Female heads

Finding Jobs

Estimates use pooled annual time series and cross-section data on all fifty states (plus the District of Columbia) from 1979 to 1997. All estimates include year dummies and state dummies, to control for unobserved time-period or state influences. The last estimates control for past trends in labor markets by including lags in wages and unemployment. In the table, the estimated effects based on past changes in welfare rolls are extrapolated into the future. The table simulates the effects of reducing welfare rolls by an amount equal to the maximum predicted reduction in welfare rolls (see table 2.2). The estimates in column 1 make no attempt to correct for biases that occur because the In(welfare receipt rate) will be endogenous. We expect lower unemployment or higher wages to reduce welfare rolls. Thus, it is not surprising that column 2 shows that lower welfare rolls are associated with higher wages and lower unemployment for all groups. This association probably reflects causation from the economy to welfare rolls rather than the reverse. The other estimates in the table attempt to control for the endogeneity of welfare rolls. The estimation strategy is to use state policy variables affecting welfare rolls as instruments for state welfare rolls. State welfare policy variables will be good instruments if state policy is not directly or indirectly caused by state labor market outcomes. 43 Column 2 reports estimates that use state welfare benefits as an instrument. These estimates reverse the sign of the effects of welfare reform on wages but not on unemployment: welfare reform is now estimated to have negative effects on wages, but welfare reform is still estimated to reduce unemployment. These estimates are unconvincing and suggest that welfare benefits is a poor instrument. It is difficult to believe that welfare reform reduces unemployment for all groups. There is evidence from some studies that state welfare benefits may be set higher when state unemployment is high, possibly in response to political pressure from typical voters (Baicker 1998). If this is so, then using state welfare benefits as an instrument will lead to biased results, which is what we seem to get. Columns 3 and 4 report estimated effects of welfare reform using as an instrument another policy variable: whether the state has been granted a waiver that year to experiment with its welfare program. These estimates are closer to expectations. The negative effects of welfare reform on wages are concentrated on female heads of household. Welfare reform's effects on unemployment for all groups are statistically insignificant, although the estimates are imprecise, and the point estimate suggests that welfare reform lowers unemployment for female heads. These reduced form estimates of the effects of welfare reform are unconvincing. The estimates suggest that a state policy variable such as welfare benefits is probably endogenous, which raises doubts about the whole strategy of using state policy variables as instruments. Convincing reduced form estimates of the effects of welfare reform will require instruments that shift the labor supply of welfare recipients independently of state policy or state labor market outcomes. Such instruments are hard to find. 44

108

I

Displacement and Wage Effects of Welfare Reform

CONCLUSION Based on the theories and studies reviewed in this chapter, there is enough evidence for two conclusions about the labor market effects of welfare reform: 1. Welfare reform is unlikely to have large effects on the overall national labor market. The labor supply shock from welfare reform is not large compared with the national labor force, and it is difficult to create a plausible model in which a small labor supply shock would result in large overall wage or displacement effects. The overall labor market is flexible enough to respond to a small supply shock without huge adjustment problems for average wages or unemployment rates. 2. It is likely that the labor supply shock from welfare reform will have substantial effects on labor market outcomes for the particular demographic groups on which the shock is most concentrated, such as female heads of household and female high school dropouts. The supply shock of welfare reform is a relatively large proportion of the labor supply of these demographic groups. The best evidence from current research suggests that a large relative labor supply shock will have significant labor market effects. This conclusion can only be avoided if one is willing to assume a labor market that quickly clears and has unusually large labor demand elasticities for less educated women. Yet the wage curve literature suggests that in the short run, supply shocks will affect relative unemployment. The minimum-wage literature suggests that in the long run, it will take quite a bit of a reduction in relative wages for relative labor demand to adjust to a supply shock. Although these predicted effects of welfare reform are supported by theory and empirical evidence, it should be noted that as of yet, little effect of welfare reform on unemployment or wages is observable, although welfare reform has already had readily observable effects on the labor supply of some groups. However, the evidence should soon become much stronger. Over the next few years, as welfare reform is aggressively pursued, there are likely to arise adverse trends in the real earnings of less educated women that are large enough to be observable. This chapter has made testable predictions. If these predictions prove false, the appropriate models and estimates for understanding how labor demand and wages for different demographic groups respond to labor supply shocks will need to be reexamined. What are the policy implications of these findings? The most obvious is that we might consider how we can reduce the effects of welfare reform on the unemployment and wage rates of less educated women. A short-run policy option for reducing the effects of welfare reform on increasing the unemployment rate of less educated women would be to create additional job openings for less educated women. This could be done through public service or community service jobs, as discussed in chapter 8 of this volume. Alternatively, additional job openings for less educated women might be created through wage subsidies to

I

109

Finding Jobs

private employers, as discussed by Katz (1998). Although welfare reform is estimated to increase labor supply by 1.4 million people, we would not need to create 1.4 million job slots to offset the unemployment effects of welfare reform. As is apparent in figure 2.4 and the accompanying text discussion, the economy on its own, without policy intervention, creates additional jobs to offset much of this labor supply shock and allocates some additional jobs to less educated women. In the medium run, we might need to create two or three additional jobs for female household heads for every ten labor force participants added to the labor market by welfare reform!5 For a labor supply increase of 1.4 million, this implies the need to create between three hundred thousand and four hundred thousand jobs for female household heads if we want to fully offset the effects of welfare reform on increasing the unemployment rate for this group. Over the long run, more of the negative effects of welfare reform on less educated women are likely to take the form of lower wages. These lower wages might be offset to some degree by further expansions of the Earned Income Tax Credit or the minimum wage. In addition, the perennially popular American policy of more education might also help offset these wage reductions. In addition to helping those educated, a program that significantly raised the education levels of the population would be expected to have aggregate effects, reducing the wages and earnings of those who are already educated and raising wages and earnings of those left behind in the less educated group. Some estimates of the magnitudes of such aggregate effects of education are provided in Bartik (1999). One important overall message of this chapter is the need for policy makers to think about aggregate effects and spillover effects when considering labor market policies. The aggregate and spillover effects of policies such as welfare reform can be significant. Policy makers should consider such aggregate effects when designing any large-scale policies addressing the low-wage labor market: welfare reform, job-training, education, public service jobs, or subsidized jobs. Researchers need to develop the appropriate models and empirical estimates to help provide reliable estimates of the aggregate effects of labor market policies.

I appreciate extensive comments on the manuscript by Larry Katz, my discussant at the November 1998 conference. I also appreciate extensive written comments by two anonymous referees and by Rebecca Blank and David Card on the November 1998 draft of this chapter. David Card also provided useful comments on the January 1999 draft. I also benefited from comments on previous versions of a portion of this manuscript from Rebecca Blank and Phillip Levine and on the entire manuscript by Kevin Hollenbeck. I appreciate helpful comments during the preconference from all the attendees but in particular David Card. I also received good comments from attendees at seminars at the University of Michigan and Western Michigan University. I appreciate help with data from Daniel McMurrer and Phillip Levine. I appreciate research assistance with this manuscript from Ken Kline and Wei-Jang Huang, library support from Linda Richer, help

110

I

Displacement and Wage Effects of Welfare Reform with graphics from Ken Kline, Leslie Lance, and Claire Black, and secretarial support from Claire Black. The research for this chapter was conducted in part with financial assistance from the Joint Center for Research on Poverty, the Russell Sage Foundation, the Rockefeller Foundation, and the W. E. Upjohn Institute for Employment Research. The findings and opinions of this chapter are those of the author and do not necessarily represent the views of any of the sponsors of the research or of any of the persons mentioned above.

APPENDIX The wage curve model used in this chapter is summarized in table 2A.1. On the demand side, the model includes an equation for overall labor demand, equations for the share of demand for five different types of labor, and a personal income equation that allows expansions in state employment to feed back into expanded local demand. On the supply side, the model includes a wage curve expressing overall wages as a function of overall unemployment, five relative wage curve equations determining relative wages for each type of labor, and labor force participation and population migration equations for each type of labor. All equations in the model include fixed state and time-period effects to allow for unobserved influences of state characteristics and the national economy. The model is identified by lags. It is assumed that labor demand, labor supply, wages, and migration respond to their various determinants only after a lag. Although this assumption of lagged response seems plausible, it is imposed by assumption rather than estimated. The lagged structure of the model does impose the implicit assumption that there will in the short run be some immediate displacement in response to a labor supply shock: it takes time for unemployment to affect wages and, in turn, cause labor demand to increase in response to the supply shock. Adjustment is not allowed to take place instantaneously in the model. 46 As shown at the bottom of table 2A.1, where the model is estimating wellestablished parameters, the estimates are consistent with this previous research. The overall labor demand elasticity is about the expected size, the labor supply responses to wages and unemployment rates are plausible, the sensitivity of wages to unemployment is consistent with research by Blanchflower and Oswald (1994), lower unemployment has greater effects on wages at low unemployment rates as was found by Blanchflower and Oswald/7 and the responses of migration to wages and unemployment seem plausible. The model also estimates various parameters for which no real consensus is available from previous research. Relative labor demand is not very sensitive to relative wages. 48 However, the model also allows for relative demand for each labor type to be directly affected by the availability of different types of labor. The rationale for this direct supply effect is that employer hiring of different types of labor may depend on who applies for the job. 49 This direct effect of relative supply on relative demand is highly statistically significant in the model.

I

11l

Finding Jobs TABLE 2A.l

/

Summary of a Wage Curve Model of State Labor Markets

Type of Equation

Dependent Variable

Overall labor demand (1 equation) Employment share demand (5 equations)

In(state employment) In(share of employment in group)

Overall wage curve (1 equation) Relative wage curves (5 equations)

In(wage)

Labor force participation rate (5 equations)

In (labor force participation rate of group)

Migration (5 equations) Income (1 equation)

In(population of group) In (state personal income)

In(wage of group/ overall wage)

Independent Variables'

In(wage) In(personal income) In(wage of group/overall wage) Current value of In(labor force share of group) (endogenous, lagged labor force share used as instrument) In(unemployment rate) Some function of relative unemployment of group, with functional form chosen for each group after preliminary testing In(wage of group) Unemployment rate of group In(AFDC benefits) for female heads only Same as for labor force participation rate In(wage) In(employment) In(population) Includes current as well as lagged values of these variables

Note: All estimates based on pooled annual time-series cross-section data for all states, 1979 to 1997. All estimates are weighted by 1979 state population. All estimates use weighted least squares except employment share demand, which is weighted 2SLS. Summary of empirical results from preliminarv runs of the model: Overall labor demand elasticity with respect to wages is - 0.7 after five years, holding state income constant. In the long-term, labor demand is significantly more elastic when state personal income is allowed to vary. Relative labor demand is not particularly sensitive to relative wages (elasticity is less than 0.1 in absolute value for all groups, except female heads is - 0.13). Current labor force share has important effects on current employment share: between 0.3 and 0.5 for all groups except other females with less than college degree (0.8) Overall wages significantly affected by unemployment, with 1 percent lower unemployment increasing wage by 2.4 percent after five years, at average unemployment rate over sample (6.8 percent). Wage curve is nonlinear. Starting at unemployment of 5.0 percent, 1 percent lower unemployment increases wages by 3.3 percent after five years; starting at 9.9 percent unemployment, 1 percent lower unemployment increases wages by 1.6 percent. (Five percent and 9.9 percent are lowest and highest national unemployment rates from 1979 to 1997). Relative wages are not sensitive to relative unemployment, with 1 percent change in relative unemployment affecting relative wages by less than 0.3 percent after five years. Labor force participa tion rates are not particularly sensitive to wage rates, with all elasticities after five years less than 0.1. Labor force participation rates are somewhat sensitive to unemployment rates, with 1 percent lower unemployment increasing overall labor force participation rate by about 0.5 percent. Low-education groups' labor force participation rates are more sensitive than high-education groups. Migration is not particularly sensitive to wage rates, except for high-education males and fe-

112

Displacement and Wage Effects of Welfare Reform TABLE 2A.l

/

Continued

males, with elasticities of 0.4 and 0.2. Migration is somewhat sensitive to unemployment rates, with 1 percent lower unemployment increasing population by 0.6 percent. 'In addition to variables listed, all equations include year and state dummies and two lags of dependent variable. All independent variables are included with two lags, and unless otherwise noted, there are no current values.

The model also estimates that relative wage curves are not very sensitive to relative unemployment rates. I know of no previous findings on relative wage curves and relative unemployment. 5o More details on this model are available in Bartik (1999).

NOTES 1. The source for the Burtless quotation is Louis Uchitelle, "Welfare Recipients Taking Jobs Often Held by the Working Poor," New York Times, April 1, 1997, p. A-I. 2. The source for the Summers quotation is Louis Uchitelle, "Welfare Recipients Taking Jobs Often Held by the Working Poor," New York Times, April 1, 1997, p. A-I. 3. Although both President Ronald Reagan and President George Bush granted states waivers for various welfare reforms, President Clinton began approving more radical waivers and approved waivers that covered a larger percentage of the national caseload. 4. For example, see chapter 8 in this volume, which discusses the labor market effects of public service employment coupled with mandatory work requirements for welfare recipients. 5. Previously, federal welfare assistance to states was provided on a matching basis to states, which meant that the federal government paid a substantial share of any extra welfare costs incurred by a state. 6. These figures are derived from the March 1997 Current Population Survey. The percentages reporting receipt of Aid to Families with Dependent Children (AFDC) in the CPS are increased by one-third based on Rebecca Blank's (1997) estimates that AFDC caseloads in the CPS are only 75 percent of the true AFDC caseload. 7. A more extensive review of these studies is provided in my working paper, "The Labor Supply Effects of Welfare Reform" (Bartik 1998). 8. Kathryn Edin and Laura Lein's (1997) book suggests that official figures understate the work of welfare recipients, because many welfare mothers engage in unreported or underground work. However, their research also shows that unreported and underground work is usually performed only a few hours a week. Low-income working mothers also engage in unreported and underground work and do so almost as much as welfare mothers. Hence, underground and unreported work does not much alter the labor supply effects of welfare reform. See Bartik (1998) for more discussion. 9. Most surveys reviewed by Jack Tweedie and Dana Reichert (1998) ask about employ-

ment at the time of the survey rather than employment at any time after leaving welfare.

113

Finding Jobs

10. Sandra Danziger and Sherrie Kossoudji's (1995) study did ask about labor force participation rates of former general assistance recipients in Michigan. As of two years after the abolition of general assistance, 39 percent of nondisabled younger (forty years old or less) former recipients reported being employed, and 43 percent reported being unemployed, for an unemployment rate of more than 50 percent. 11. The group examined is female heads of household, age sixteen to forty-four, with relatives in the household and less than a college degree, who received AFDC last year. Unemployment rates, using sample weights, for each year from 1992 to 1997 are 41 percent, 36 percent, 40 percent, 39 percent, 35 percent, and 33 percent, respectively. Each year's estimate uses about four hundred observations. 12. The 63 percent labor force participation rate is derived by dividing the employment rate of 40 percent by one minus the unemployment proportion of .37. One could use the same unemployment rate to calculate labor force participation for welfare recipients. If the employment rate of single-parent cases is .13, and their unemployment rate is .37, then their labor force participation rate would be .21. The change in labor force participation rates for single-parent cases would be .42, close to the .47 assumed in this chapter. However, unemployment rates for those still on welfare would be lower than for those excluded from welfare, as welfare reduces the incentive for job search. Labor force participation rates of current welfare recipients would be lower than .21, and the change in labor force participation rates would be greater than .42. 13. Under the welfare bill, the percentage reduction in a state's caseload since 1995 reduces the work requirement percentage by the same amount, unless the federal government can show that the caseload reduction resulted from the state's changing welfare eligibility rules. Because the federal government bears the burden of proof, it is likely that most reductions in caseloads owing to state administrative procedures and other policies are likely to help states meet work requirements. 14. In addition, the 1996 bill provides for bonus grants for states that are" succeeding" at welfare reform, which provides a financial incentive for states to increase work by welfare recipients. 15. This is implicitly assumed because welfare reform is assumed to account for the unexplained change in the log of the welfare receipt rate. 16. It should be noted that this estimated labor supply increase represents the increase as of an average point in time, not labor force participation any time during the year. Therefore, the estimated percentage increase in labor supply owing to welfare reform already implicitly adjusts for the fact that many former welfare recipients will be in the labor force for only part of any given year. In addition, one might wonder whether the real labor supply increase owing to welfare reform might be considerably lower if many former welfare recipients work only a few hours. The available evidence suggests, however, that many former welfare recipients work close to full time. In most of the recent follow-up studies, more than half of welfare leavers who are employed work more than thirty hours a week. In the four studies that presented such data, mean weekly hours of work of employed welfare leavers ranged from thirty-four to thirty-six hours a week (Bra under and Loprest 1999). In a recent Urban Institute survey of welfare leavers and similar low-income families, more than twothirds of employed welfare leavers work more than thirty-five hours a week, and

114

Displacement and Wage Effects of Welfare Reform their work hours seem quite similar to those of other employed low-income mothers (Loprest 1999). 17. Using the "best forecast," prediction, I assume that labor supply effects on single parents are effects on women, and effects on two-parent households are effects on men. The estimated labor supply increase from 1993 to 2005 for women is 1.268 million. I compare this forecast with employment numbers for women with less than a college education because the March 1997 CPS shows that 98.5 percent of women receiving AFDC have less than a college education. In addition, wage trends of persons with "some college" are more similar to those of persons with only a high school diploma than they are to persons with a college degree. This suggests that dividing the population into those with and without a college degree makes more sense than dividing the population into those with just a high school diploma from those with "some college" or a college degree. Whether high school dropouts should be analyzed separately is a more difficult question. Some later analyses do look at the welfare reform supply shock's impact on female high school dropouts; see tables 2.4 and 2.5. 18. These are standard results in economics, and they can be found in many places in the literature-for example, in Freeman (1977). These equations can be derived by specifying the equilibrium condition qW) + G = LiW), where Ls(W) is labor supply as a function of wages, G is the supply shock, and Ld(W) is labor demand as a function of wages. To derive the formulas, totally differentiate this equilibrium condition with respect to G and evaluate at G = O. 19. These effects of labor supply and demand slopes could also be shown qualitatively by redrawing figure 2.2 several times with different slopes. 20. Thus, the relevant demand elasticity is a general equilibrium elasticity in response to the supply shock. The output and other general equilibrium effects for a shock of 1 percent of the labor force will be small. However, the direct effect of the supply shock on overall wages, holding general equilibrium effects constant, is also small, so the general equilibrium effects still make a difference. If we write a factor price function with the log of the overall wage a function of the log of overall labor and the log of GDp, and suppress capital for short-run analysis, the factor price equation is W = f(L, Y). The elasticity of the wage with respect to labor, holding output constant, is k the partial derivative with respect to L. The elasticity allowing Y to change will be A + jy YL. These elasticities will be quite different even if the change in L is small. 21. The demand elasticity reported in the text is consistent with Hamermesh's conclusion that the "output-constant" elasticity of demand for overall labor is (- 0.3). An outputconstant elasticity is irrelevant because output will increase because of a labor supply shock. In the short run, the relevant elasticity is one divided by the "factor price elasticity for overall labor," where the factor price elasticity shows how the wage that employers are willing to pay varies with the quantity of labor, holding capital constant. Hamermesh (1993) shows that the output-constant elasticity of demand for overall labor is - (1 - s)SUB and the factor price elasticity is - (1 - 5)(1/ SUB), where 5 is the factor share of labor and SUB is the elasticity of substitution between labor and capital. Research suggests that SUB is close to 1 and s is close to 0.7. Hence, the relevant short-run demand elasticity is 1/ ( - 0.3) = - 3.33. 22. One could argue that the long-run elasticity of demand for overall labor is infinite. Lower wages will increase profits, leading to investment. This investment will in-

115

Finding Jobs crease labor demand, leading to wage increases and profit declines. If long-run capital supply is perfectly elastic at a fixed profit rate, this process will continue until profit rates are restored to their original level. If production is constant in returns to scale in capital and labor, the original profit rate cannot be restored until increased labor demand has restored wages to their original level. With constant returns to scale, the profit rate and wage rate can both be expressed as functions of the ratio of capital to labor alone, so restoring the original profit rate implies restoring the original wage rate. 23. Feedback effects from increased output are not that important in determining the slope of the demand function for a labor input with a small factor share. Suppose we write a factor price equation expressing the log of the wage of labor of type i as a function of the log of labor of type i and the log of GDP, or W; = f(L;, Y). (We suppress other inputs that should also enter this function.) Then the elasticity of the wage of i with respect to the quantity of labor of type i is fLi + fYY/i. This is similar to the equation for overall labor in a previous footnote. However, in this case we would expect Y/; to be considerable smaller than was YL. That is, the effect on GDP of a percentage change in some labor input that is only a small portion of overall labor will be much smaller than the effect on GDP of the same percentage change in overall labor. 24. Although these findings can be reconciled with some specifications of the production function, for many commonly used production functions these findings are contradictory. 25. When employment is the dependent variable, unobserved demand shocks will move us along a supply curve, resulting in a positive correlation between the residual and wages, biasing the negative coefficient on wages toward zero. When the wage is the dependent variable, unobserved demand shocks will also lead to a positive correlation between the equation residual and employment, biasing the negative coefficient on employment toward zero. This will bias the implied slope of the labor demand curve, which is one over this coefficient on employment, away from zero. 26. One could argue that the literature on the labor market impacts of immigration, which shows modest impacts, suggests that labor demand elasticities for less skilled workers must be large in absolute value. I argue later in this section that immigration is not a pure labor supply shock, and so its impacts cannot be easily used to suggest likely labor demand elasticities. 27. Blanchflower and Oswald estimate that the effect of In (unemployment rate) on In(wage) is 0.1. At an unemployment rate of 6 percent, a 1 percent decrease in unemployment would increase wages about 1.6 percent. Cross-section studies suggest that a 1 percent decrease in unemployment increases labor force participation by perhaps 0.5 percent (Bowen and Finegan 1969). Combining these relationships, a 1.5 percent increase in employment is associated with about a 1.6 percent increase in wages. 28. Blanchflower and Oswald (150-53) do not find huge differences in the sensitivity of different groups' wages to overall unemployment; if some group's unemployment had huge effects on wages, greater differences would be expected. Also, see estimates below that suggest little sensitivity of relative wages of different groups to relative unemployment. 29. David Card's review of The Wage Curve in the Journal of Economic Literature argues that although "most readers ... will accept [the] conclusion that wages are negatively

116

Displacement and Wage Effects of Welfare Reform correlated with the local unemployment rate ... , of deeper interest to many readers will be the interpretation of this correlation.... Even if the question of why local unemployment rates affect pay remains unsettled, as I believe it does, the existence of a wage curve relation is an important addition to our knowledge about the modern labor market" (Card 1995, 785, 798; italics in original). Joe Stone's review of The Wage Curve in Industrial and Labor Relations Review (Stone 1997) reaches a similar conclusion. 30. Agglomeration economies are productivity advantages associated with the size of a local economy (urbanization economies) or the size of a particular industry in a local economy (localization economies). For example, larger cities may be better able to support diverse networks of local suppliers or may encourage innovative ideas. 31. The November 1998 conference version of this chapter also analyzed wage effects of welfare reform using two other multisector models of immigration: see Borjas, Freeman, and Katz (1997) and Card (1997). These two models, however, do not distinguish between the effects of shocks to female labor supply versus male labor supply, which limits the relevance of these two models to welfare reform. For interested readers, the November 1998 version of this chapter is available at the Upjohn Institute's web site, www.upjohninst.org. 32. One could raise some doubts about whether Juhn, Murphy, and Topel's elasticities do hold in the long run. Some of their elasticity numbers are implicitly derived from wage-curve-style short-run responses, for example, by analyzing pooled annual timeseries cross-section data on wages and employment of different groups in different regions and years. 33. This effect can be approximately calculated by taking the ratio of the labor supply elasticity of 0.4 times the change in the female high school dropout wage in table 2.5 to the percentage shock reported for female high school dropouts and then multiplying by minus one. 34. These empirical results are based on annual national wage data, from 1973 to 1997, derived from the Current Population Survey, Outgoing Rotation Group. These data were downloaded from the web site of the Economic Policy Institute (www.epi.org). The regression I estimated, following Katz and Murphy'S approach, was to regress the In (average wages for the some college group) on In (average wages of the high school graduate group) and the In(average wages of the college graduate group) with no intercept. The resulting coefficients and t-statistics are 0.771 (t = 12.79) on the high school graduate wage and 0.236 (t = 4.57) on the college graduate wage. If an intercept is added, the college graduate wage variable becomes inSignificant, with a t-statistic of 0.23, whereas the high school graduate wage variable has a t-statistic of 10.38. 35. Although this is speculative, I doubt if results would much change if I used a breakdown of "high school or less" and "more than high school." I doubt if more than 20 percent of my "less educated female heads" group has "some college" education, and an even lower percentage of welfare recipients are in the "some college" category. Hence, if I used the "high school or less" category, the welfare reform labor supply shock, as a percentage of the female head "high school or less" group, would probably go up by less than 20 percent compared with the shock to the female head group that I use. On the other hand, if the female head group is defined more narrowly, it is likely that more of its labor market outcomes will depend on aggregate variables and

117

Finding Jobs less on group-specific variables. Therefore, the estimated effects for this narrower group might not go up as much as the size of the welfare reform supply shock. 36. An alternative way of modeling national effects would be to include national variables in the model (with year effects specified as random) and then to simulate what happens if national variables are equal to the same variables for the typical state. However, national variables generally have insignificant effects or the "wrong sign" when entered into these state estimating equations. This suggests that unobserved national effects may be biasing these random-effect estimates. Estimates with a national dummy for each year may be less biased; it is these estimates that are described in the appendix and reported in the text, tables, and figures. 37. This finding seems to hold in a variety of specifications. Relative unemployment has statistically significant effects on relative wages. The magnitude of these effects, however, is quite small compared with the effects of overall unemployment on overall wages. 38. As discussed in the appendix, in this model there are some modest direct effects of labor force composition on the composition of who is hired. These direct effects are insufficient for labor demand for a type of labor to fully accommodate a supply shock targeted at that type of labor. 39. Displacement is here defined as the loss of jobs in a given group or overall, compared with the gain in jobs for all welfare recipients who enter the labor market because of welfare reform, regardless of what group the welfare recipient is in. For more detail on the displacement estimates, see the November 1998 conference version of this chapter. 40. From 1995 to 1997, the In(wage) for female heads declined from 2.130 to 2.116, while the In (wage) for other women without a college degree increased from 2.141 to 2.152. From 1995 to 1997, unemployment for female heads stayed at 11 percent, whereas unemployment for other women without a college degree decreased from 6.0 percent to 5.4 percent. 41. Based on the CPS, the 1997 population of female heads of household was 6.859 million. From 1995 to 1997, labor force participation for this group increased from 69.5 percent to 75.7 percent, while the labor force participation for other females without a college degree increased by only 0.4 percent. The extra labor supply of female heads resulting from an increase in their labor force participation of more than 0.4 percent is [(75.7 - 69.9)/100] mult 6.859 million = .398 million. These extra labor force participants are 7.7 percent-that is, .398/(.757 * 6.859)-of the total female head labor force in 1997. lf unemployment had decreased from 11 percent to 10.4 percent among the "original" labor force of female heads of household, overall employment among female heads of household would stay at 11 percent if unemployment among the new participants was 18.3 percent. Data from the 1992 to 1997 March Current Population Surveys suggest that unemployment among female heads who received welfare the previous year averaged 37 percent, whereas unemployment among female heads who did not receive welfare the previous year averaged 7 percent. Suppose unemployment of the "original" female head group was 10.4 percent and unemployment of new participants was 18.3 percent. Then employment of the new participants is .324 million out of total female head employment of 4.619 million, or 7.0 percent of total employment. For compositional effects to explain a relative wage decline cif 2.5 percent in the In(wage), the In(wage) of the new participants must be 35.7 percent lower than those of the original labor force. Data from the March CPS from 1992 to 1997

118

Displacement and Wage Effects of Welfare Reform show that female heads on AFDC last year had an average In (wage) of 1.786 versus an average In(wage) of 2.190 for female heads not on AFDC last year. 42. This is the closest one can come to consistently defining" single mothers" for all the years from 1979 to 1997 in the CPS aRC data. 43. In addition, to be a good instrument, state policy variables must be good predictors of state welfare rolls, which they are. The F-tests on the instruments in the first-stage estimation prediction of state welfare rolls always have large values: 21.69 for the estimates in column 2, 12.07 for the estimates in column 3, and values greater than 30 in all cases in column 4. 44. One might consider national policy changes with differential effects across states, for example, the changes implemented in 1981 by President Ronald Reagan in income disregards, or changes in the 1950s and 1960s in federal matching rates for state welfare spending. 45. This calculation is based on the estimate that the displacement rate for female household heads per job created by welfare reform is about 0.3 in the short run and 0.2 in the long run. This finding is mentioned earlier, and further estimates and discussion are in the November 1998 version of this chapter. 46. The model does allow an immediate response of the employment share of a group to changes in the labor market share of the group. The assumption is that employer hiring responds to who happens to be applying for jobs. 47. The estimation of the wage curve tested entering unemployment into the equation in various ways: the log of the unemployment rate, one over the unemployment rate, the unemployment rate, and the unemployment rate and unemployment rate squared together, with the final model chosen using the Akaike Information Criterion. The preferred model for overall wages uses the log of the unemployment rate; the preferred model for relative wages uses the unemployment rate itself, except for female heads and other females with less than a college degree, which use one over the unemployment rate. 48. Note that the specified model, with log(employment share) as a function of log(relative wages), could be derived from a production function with a labor aggregate and a constant elasticity of substitution among all the labor types. However, the estimation does not impose the complete restrictions that would be implied by a CES function. Each employment share is allowed to have its own response to relative wages. The model simulation then adjusts the shares to force them to add up to one. A preliminary version of the model allowed employment for each type of labor to respond to all five wages; this model showed even fewer effects of relative wages on relative employment. 49. The labor force share was added to the model because preliminary estimates showed relatively little response of the employment shares to shifts in relative supply; the intent was to allow another avenue of response. However, as shown in the main text of this chapter, even with this extra avenue of response, employment shares do not completely adjust to supply shocks in any medium-term time period. 50. A preliminary version of the model allowed each group's wage to be affected by unemployment rates for all five groups. This model showed even less sensitivity of relative wages for a group to relative unemployment.

119

Finding Jobs

REFERENCES Baicker, Katherine. 1998. "Fiscal Federalism and Social Insurance." Ph. D. diss., Harvard University. Bartik, Timothy J. 1996. "The Distributional Effects of Local Labor Demand and Industrial Mix: Estimates Using Individual Panel Data." Journal of Urban Economics 40 (September): 150-78. --.1998. "The Labor Supply Effects of Welfare Reform." Staff working paper 98-53. Downloaded on January 5, 2000, from the world wide web at: http://www.upjohninst.org. Kalamazoo, Mich.: Upjohn Institute. - - . 1999. "Aggregate Effects in Local Labor Markets of Supply and Demand Shocks." Staff working paper 99-57. Downloaded on January 5, 2000, from the world wide web at: http://www.upjohninst.org. Kalamazoo, Mich.: Upjohn Institute. Berger, Mark. 1983. "Changes in Labor Force Composition and Male Earnings: A Production Approach." Journal of Human Resources 18(2, Spring): 77-96. Bernstein, Jared. 1997. "Welfare Reform and the Low-Wage Labor Market: Employment, Wages, and Wage Policies." Technical paper 226. Washington, D.C: Economic Policy Institute. Bishop, John H. 1998. "Is Welfare Reform Succeeding?" Working paper. Ithaca: Cornell University, Industrial and Labor Relations School. Blanchflower, David G., and Andrew J. Oswald. 1994. The Wage Curve. Cambridge, Mass.: MIT Press. Blank, Rebecca M. 1997. "What Causes Public Assistance Caseloads to Grow?" Working paper 6343. Cambridge, Mass.: National Bureau of Economic Research. Borjas, George J., Richard B. Freeman, and Lawrence F. Katz. 1997. "How Much Do Immigration and Trade Affect Labor Market Outcomes?" Brookings Papers on Economic Activity 1. Washington, D.C: Brookings. Bowen, William G., and T. Aldrich Finegan. The Economics of Labor Force Participation. Princton, N.J.: Princeton University Press. Braunder, Sarah, and Pamela Loprest. 1999. "Where Are They Now? What States' Studies of People Who Left Welfare Tell Us." New Federalism Series working paper A-32. Downloaded on January 5, 2000, from the world wide web at: http://www.newfederalism.urban.org. Washington, D.C: Urban Institute. Card, David. 1995. "The Wage Curve: A Review." Journal of Economic Literature 33 (June): 785-99. - - - . 1997. "Immigration Inflows, Native Outflows, and the Local Labor Market Impacts of Higher Immigration." Working paper 5927. Cambridge, Mass.: National Bureau of Economic Research. Chernick, Howard, and Andrew Reschovsky. 1996. "State Responses to Block Grants: Will the Social Safety Net Survive?" Technical paper. Washington, D.C: Economic Policy Institute. Daly, Mary. 1997. "Labor Market Effects of Welfare Reform." Economic letter 97-24. Downloaded on January 5, 2000, from the world wide web at: http://www.bsf.org/ econrsrch/wklyltr/eI97-24.html. San Francisco: Federal Reserve Bank of San Francisco. Danziger, Sandra K., and Sherrie A. Kossoudji. 1995. "When Welfare Ends: Subsistence

120

Displacement and Wage Effects of Welfare Reform Strategies of Former GA Recipients." Final report of the General Assistance Project. School of Social Work, University of Michigan, Ann Arbor. Danziger, Sheldon, Robert Haveman, and Robert Plotnick. 1981. "How Income Transfer Programs Affect Work, Savings, and the Income Distribution: A Critical Review." Journal of Economic Literature 19(5eptember): 975-1028. Duncan, Greg L Kathleen Mullan Harris, and Johanne Boisjoly. 1998. "Time Limits and Welfare Reform: How Many Families Will Be Affected?" Working paper. Downloaded on January 5, 2000, from the world wide web at: http:j jwww.nwu.edujIPRjpublicationsjnuprjnuprvo3n1j duncan.html. Evanston, Il.: Northwestern Policy Research. Edin, Kathryn, and Laura Lein. 1997. Making Ends Meet: How Single Mothers Survive Welfare and Low-Wage Work. New York: Russell Sage Foundation. Freeman, Richard B. 1977. "Manpower Requirements and Substitution Analysis of Labor Skills: A Synthesis." In Research in Labor Economics: An Annual Compilation of Research, vol. 1, edited by Ronald G. Ehrenberg. Greenwich, Conn.: JAI Press. Grant, James. 1979. Labor Substitution in U.S. Manufacturing. Ph.D. diss., Michigan State University. Greenwood, Michael J., and Gary L. Hunt. 1984. "Migration and Interregional Employment Redistribution in the United States." American Ecollomic Review 74(5): 957-69. Hamermesh, Daniel S. 1993. Labor Demand. Princeton: Princeton University Press. Holzer, Harry J. 1996. "Employer Demand, AFDC Recipients, and Labor Market Policy." Discussion Paper 1115-96. Madison, Wisc.: Institute for Research on Poverty. Johnson, George E. 1998. "The Impact of Immigration on Income Distribution Among Minorities." In Help or Hindrance: The Economic Implications of Immigration for African Americans, edited by Daniel Hamermesh and Frank D. Bean. New York: Russell Sage Foundation. Juhn, Chinhui, and Daae Il Kim. 1999. "The Effects of Rising Female Labor Supply on Male Wages." Journal of Labor Economics 17(1): 23-48. Juhn, Chinhui, Kevin Murphy, and Robert Topel. 1991. "Why Has the Natural Unemployment Rate Increased over Time?" Brookings Papers on Economic Activity 2: 75-142. Katz, Lawrence F. 1998. "Wage Subsidies for the Disadvantaged." In Generating Jobs: How to Increase Demand for Less-Skilled Workers, edited by Richard B. Freeman and Peter Gottschalk. New York: Russell Sage Foundation. Katz, Lawrence F., and Kevin M. Murphy. 1992. "Changes in Relative Wages, 1963-1987: Supply and Demand Factors." Quarterly Journal of Economics 107(1): 35-78. Levine, Phillip, and Diane Whitmore. 1997. "Technical Report: Explaining the Decline in Welfare Receipt, 1993-1996." Downloaded on January 5,2000, from world wide web at: http:j jwww.whitehouse.gov jWHjEOP jCEAjWelfarejTechnical-Report.html. Washington, D.C: President's Council of Economic Advisers. Loprest, Pamela. 1999. "Families Who Left Welfare: Who Are They, and How Are They Doing?" New Federalism Series discussion paper 99-02. Downloaded on January 5, 2000, from the world wide web at: http:j jwww.newfederalism.urban.org. Washington, D.C: Urban Institute. McMurrer, Daniel P, Isabel V. Sawhill, and Robert 1. Lerman. 1997a. Spreadsheets Used to Estimate Effects of Welfare Work Participation Requirements. Personal communication from Daniel P. McMurrer. - - - . 1997b. "Welfare Reform and Opportunity in the Low-Wage Labor Market." Opportunity in America Series working paper 5. Washington, D.C: Urban Institute. Meyer, Bruce D., and Dan T. Rosenbaum. 1998. "Welfare, the Earned Income Tax Credit, and the Labor Supply of Single Mothers." Working paper November 32. Downloaded

I

121

Finding Jobs on January 5, 2000, from the world wide web at: http:j jwww.jcpr.orgjwpjwpprofile.cfm?ID = 32.0. Chicago: Joint Center for Poverty Research. Michigan Family Independence Agency (MFIA). 1998. "April 1996 AFDC Case Closures Due to JOBS Sanctions." Report. Downloaded on January 5, 2000, from the world wide web at: http:j jwww.mfia.state.mLusjsanctionjpart-a.htm. Lansing, Mich.: Michigan Family Independence Agency. Mishel, Lawrence, and John Schmitt. 1995. "Cutting Wages by Cutting Welfare: The Impact of Reform on the Low-Wage Labor Market." Briefing paper 58. Washington, D.C: Economic Policy Institute. Muth, Richard F. 1971. "Migration: Chicken or Egg?" Southern Economic Journal 37Ganuary): 295-306. Smith, James P., and Barry Edmonston, eds. 1997. The New Americans: Economic, Demographic, and Fiscal Effects of Immigration. Washington, D.C: National Academy Press. Solow, Robert M. "Guess Who Pays for Workfare?" 1998a. New York Review of Books, November 5. Downloaded on January 5, 2000, from the world wide web at: http:j j www.nybooks.comjnyrevjwwwarchdisplay.cgi?19981105027F. New York: New York Review of Books. - - - . 1998b. Work and Welfare. Princeton, N.J.: Princeton University Press. Stone, Joe A. 1997. Review of The Wage Curve, by David G. Blanchflower and Andrew J. Oswald. Industrial and Labor Relations Review 50(3): 526-27. Tweedie, Jack, and Dana Reichert. 1998. "Tracking Recipients After They Leave Welfare: Summaries of State Follow-Up Studies." Report. Downloaded on January 5, 2000, from the world wide web at: http:j jwww.ncsl.orgjstatefedjwelfarejfollowup.htm. Denver: National Conference of State Legislatures. U.s. Bureau of Economic Analysis. 1998. Regional Economic Information System. CD-ROM with 1969 to 1996 data .. Washington, D.C: U.s. Department of Commerce. U.S. Department of Health and Human Services. 1998. "Welfare Caseloads: Families and Recipients, 1936-1998." Downloaded on January 5, 2000, from the world wide web at: http:j jwww.acf.dhhs.govjnewsj3697.htm. Washington, D.C: Administration for Children and Families. - - - . 2000. "Temporary Assistance for Needy Families (TANF) Percent of Total U.S. Population, 1960-1999." Downloaded on January 5, 2000, from the world wide web at: http:j jwww.acf.dhhd.govjnewsjstatsj6097rf.htm. Washington, D.C: Administration for Children and Families. u.s. Department of Labor, Bureau of Labor Statistics. 1998. "Employment Characteristics of Families in 1997." Downloaded on January 5, 2000, from the world wide web at: http:j j www.stats.bls.govjnewsrels.htm.

122

I

Chapter 3 Job Change and Job Stability Among Less Skilled Young Workers Harry J. Holzer and Robert J. Lalonde

T

o what extent does job or employment instability contribute to the problems of less skilled workers in the labor market?! For which skill group is job instability most severe? What factors are associated with such instability, both among and within demographic groups? Labor economists and policy makers have long been interested in these questions, and a significant body of research on these topics has emerged over the years. Changes in the labor market for less skilled workers over the past few decades, however, raise new concerns about these questions. For one thing, inequality between skill groups has grown rapidly over the past few decades, while real wages of the less skilled (especially among men) have apparently declined. 2 Furthermore, employment rates among less skilled men have declined in association with these losses (Juhn 1992; Murphy and Topel 1997), and employment rates of less educated women have improved less rapidly than those of more educated women (Blau 1998). The extent to which job and employment instability contribute to the widening gaps in employment rates between the more and less educated and the degree to which enhanced job stability might help to improve both the employment prospects and the real wages of these workers clearly need to be determined. Interest in these questions also has been stimulated by the recent enactment of welfare reform legislation designed to increase the participation of less educated women in the labor market. Some observers (McMurrer, Sawhill, and Lerman 1997; Holzer 2000) have expressed concern that job turnover, perhaps associated with child care and transportation as well as work performance problems, will limit the earnings of welfare recipients as they enter the labor market and, with it, their potential for wage growth over time. To what extent is job turnover a particular problem for less educated females, and what factors are associated with these problems? Are they more severe for some parts of this population, such as minorities, than others? Are there policy approaches, such as the provision of early work experience or child care, that can help to remedy these problems once they have appeared?

I

125

Finding Jobs

PREVIOUS EMPIRICAL LITERATURE ON JOB CHANGE AND STABILITY Over the years, a number of empirical findings have emerged regarding job change and job stability among various groups of workers, especially by race, gender, age or experience, and educational attainment. For instance, it is well known that job separations decline with labor market experience and job tenure (see, for example, Leighton and Mincer 1982; Farber 1998), though this appears to be somewhat less true among low-income young black males (Ball en and Freeman 1986). Average separation rates from jobs vary systematically among demographic groups. For instance, employment and job instability appear to be higher among women, minorities, and those less educated than their counterparts (Parsons 1986; Farber 1998). However, these differences by race and gender disappear, or even are reversed, when controls are included for observable differences in personal or job characteristics (see, for example, Light and Ureta 1992). Voluntary separation rates appear particularly lower among blacks when these controls are included (for example, Blau and Kahn 1981), though layoffs and discharges are still higher for them (Jackson and Montgomery 1986; Ferguson and Filer 1986). By contrast, differences in turnover by level of educational attainment persist within each gender or race group, especially among females (Light and Ureta 1992). The differences among these groups become even clearer when we realize that job-to-job changes that have potentially positive effects on the earnings of young workers (see, for example, Topel and Ward 1992) are relatively infrequent among young less educated women, whereas job-to-nonemployment changes occur much more frequent among this group (Royalty 1998). To what extent are the employment problems of less skilled and particularly minority workers accounted for by their higher employment stability (or transitions out of employment), as opposed to longer nonemployment spells (or lower transitions into employment)? The work of Kim Clark and Lawrence Summers (1982) and John Ballen and Richard Freeman (1986) indicate that the latter have accounted for most of the differences in employment rates across racial groups, suggesting that employment instability has not been not as great a concern for this group as the difficulty of reentering employment, once it had left the workforce. 3 All of this work was performed on data from two or more decades ago, however, and little of it focused on differences in employment rates between the more and less skilled per se. Most recent work by Chinhui Juhn, Kevin Murphy, and Robert Topel (1991) and Murphy and Topel (1997) also suggests that recent declines in employment rates among less educated (or low-wage) men largely reflect increasingly lengthy durations of nonemployment and nonparticipation in the labor force, though the potential role of employment instability is not considered in this work. Thus, the extent to which employment instability contributes to the lower overall employment rates of less skilled workers of any race or gender has not been analyzed explicitly in any of these studies. 126

I

Job Change and Job Stability Among Less Skilled Young Workers

What do we know about the labor market consequences of these lower employment rates over time among less skilled workers? Although it is well known that some degree of job change has positive effects on the earnings of young workers (Topel and Ward 1992), it has also become increasingly apparent that early nonemployment among some groups of young workers leads to significant losses in earnings over time, as their general labor market experience and tenure are reduced. This appears to be true for both blacks and whites (Bratsberg and Terrell 1998), for women as well as men (Light and Ureta 1995), and among the less educated (see chapter 4, this volume). The extent to which early nonemployment generates later nonemployment (as opposed to lower wages), especially once unobserved heterogeneity among individuals is accounted for, was questioned in some well-known papers many years ago (for example, Ellwood 1982; Meyer and Wise 1982). More recent research evidence (for example, Neumark 1997; Rich 1994) calls these findings into question. 4 Furthermore, the growing tendency of low wages to be associated with lower employment rates further suggests an indirect mechanism through which early employment losses might persist over time, one that might matter more now than it did a few decades ago. A few other findings in this literature are noteworthy as well. Henry Farber (1994) finds that tenure of the most recent job has a stronger effect on job changes than does earlier employment experience. Ballen and Freeman (1986) suggest that, among black youth, these previous employment experiences do not necessarily improve employment prospects over time; but John Ham and Robert LaLonde (1996) find that a year of employment in the National Supported Work Demonstration program raised subsequent employment durations among very low-income adult women. Thus, the extent to which some early employment experience, even if it is in the public sector, provides returns in terms of subsequent private sector job instability remains unclear. Furthermore, the effects of past or current job characteristics on employment stability, controlling for personal characteristics, remain unclear as well (see, for example, Brown 1982; Ferguson and Filer 1982). It might be the case that access to "good" jobs, rather than any employment at all, is a more reliable predictor of employment stability. Marital status and the presence of children also continue to be important determinants in many studies of employment stability and wages among women. s Finally, it is quite noteworthy that, despite the growing interest recently in earnings returns to cognitive skills, independently of educational attainment, we know of no analysis to date of their relation to employment stability.6 Thus, we find a need to update important parts of the previous literature on employment stability among less skilled workers, particularly in light of the major changes that have occurred in the labor market for these workers. More attention needs to be placed on less skilled workers more generally, who can be identified on the basis of academic achievement through test scores as well as educational attainment. In addition, we need to consider how other determinants of employment stability, such as job characteristics, previous employment experiences, and family status, affect this group compared with other workers. I

127

Finding Jobs

DATA AND SUMMARY STATISTICS We use data from the 1979 National Longitudinal Survey of Youth (NLSY79) to analyze job and employment stability. This data set consists of a sample of more than twelve thousand individuals who were aged fourteen through twenty-one in 1979. To be included in our sample, respondents had to be interviewed in 1994, maintain complete job histories during the years they were scheduled to be interviewed, and have been employed at least once between 1978 and 1993. Accordingly, our sample contains job histories of respondents through the 1994 wave of the survey, and the statistics we present are weighted according to the sample weights given in that year. We limit our analysis to jobs that are held only when the respondent was not enrolled in school full time. The NLSY79 public use file contains a work history file that provides information on each weekly activity over the preceding year. It also contains information on up to five jobs a year, providing the week the job started (and perhaps ended), whether the job continues from the previous year or into the next, and usual weekly hours worked on each. It also provides information on occupation, industry, and wage level on the job, as well as why the job might have ended. We use all of this information in our computations below. We follow Ann Royalty (1998) and focus on job-to-job transitions as well as job-to-nonemployment transitions. We distinguish between voluntary separations in many cases. However, following Farber (1998), we analyze transitions at the weekly rather than monthly or annual level. This allows for the well-documented nonlinearity in hazards during the early history of a job. During this period, the probability of leaving a job in any given week increases before declining sharply after the fourth month of employment. For employers who use probationary periods (either formal or informal) of a few weeks or months before committing to employing workers for longer periods, such nonlinearities should be expected. By analyzing job transitions at the weekly level, we can examine the determinants of instability during the early phase of the job. Our primary focus in this chapter is on the stability of regular jobs during the first eighteen months of the job. We define these jobs as those in which respondents worked at least thirty hours a week. However, in circumstances in which the respondents worked in a part-time job that did not overlap with a regular job, we also include those spells in our analysis. In some cases these part-time jobs became full-time jobs. We have discarded jobs that were already in progress (that is, left-censored spells) before January 1978. When several jobs were held at the same time, we have chosen the job with the longest duration as the regular job, as long as the respondent usually worked full time for some period during that job spell. Thus a job that started as a part-time job, in which the respondent reported usually working less than thirty hours, and became a full-time job in a subsequent interview would be considered a regular job from the date that job began as a part-time job until the employee's relationship with the employer was severed. Accordingly, we mea-

128

I

Job Change and Job Stability Among Less Skilled Young Workers

sure duration of this job from the time it began as a part-time job. However, because some jobs began as a part-time job and then became a full-time job we created a variable indicating whether the regular job was once a part-time job. We also kept track of employment in other jobs during this regular job spell with a variable indicating whether during the current week the respondent held two or more jobs.

THE WORK EXPERIENCE OF YOUNG ADULTS We begin our analysis of the data by examining some results on the employment experiences of workers who are in their late twenties or thirties from recent panels of the NLSY79 and how these compare with results for this same cohort of workers when they were in their early to middle twenties several years earlier. Such a comparison can at least suggest the extent to which the stability of new jobs bears on the employment experience that youths and young adults accumulate during their careers and the extent to which their early experiences manifest a pattern that continues to hold as these individuals approach middle age. Table 3.1 presents the percentage of total time between 1991 and 1995 that workers were employed. By the end of this period, these workers were primeage adults, being thirty-one through thirty-eight as of 1995 (or twenty-seven through thirty-eight over the entire period).7 We disaggregate the figures in the table by individuals' educational attainment and, separately, by gender or race. The results indicate that the percentage of time spent employed varies considerably by education, race, and gender. Within each race and gender group, the least educated work the least frequently. Among white males, college graduates work roughly 95 percent of the time, whereas high school dropouts do so only 75 percent of the time during the five years studied. These data thus reflect the trend toward lower labor force participation among less educated men that has been frequently documented in recent years (by Juhn, Murphy, and Topel [1991], TABLE 3.1

/

Time Spent Employed, by Level of Education (Percentage)

Less than high school diploma General education diploma High school diploma Some college College graduate

Men

Women

Whites

Blacks

75.2 76.5 88.3 88.0 95.4

48.8 61.5 69.8 76.1 81.1

68.8 74.4 81.0 82.6 88.2

48.2 57.3 71.1 77.6 90.9

Source: Number of jobs, labor market experience, and earnings growth: Results from a longitudinal survey (U.S. Department of Labor 1998, table 2). Note: Figures represent percentage of weeks spent in employment for individuals age thirtyone to thirty-eight, from 1991 to 1995.

I

129

Finding Jobs

among others). However, among the other demographic groups the differences between the participation rates of the most and least educated are even larger than for white males. Indeed, both female and black high school dropouts tend to work less than half of the time. Moreover, consistent with recent evidence regarding the equivalence of a high school diploma and the General Educational Development (GED) diploma, individuals with only a GED participate substarttially less in the workforce than those with a high school diploma (Cameron and Heckman 1993). Among males, participation rates of high school dropouts and those with a GED are nearly the same, though women with a GED work considerably more than high school dropouts. In table 3.2, we examine these same individuals earlier in their life cycle, at ages twenty-three through twenty-seven. We present these data by gender and education, and on percentage of time spent in the military, unemployed, and out of the labor force as well as time spent employed. Overall, we find qualitatively similar relations between educational attainment and employment experience as presented in the earlier table. Specifically, male high school dropouts between the ages of twenty-three and twenty-seven report working about 70 percent of the time during this five-year period-just a bit less time than they spent working during their early thirties. High school dropouts were employed for nearly as many weeks as high school graduates, but the latter group spent a much larger share of their nonemployed time in the military, whereas the former spend it almost exclusively being unemployed or out of the labor force. Thus, the changing skill composition of the military over the past few decades may account for much of the diverging employment experiences of male high school dropouts and graduates during this time period. The corresponding figures for young adult females are substantially more varied among the educational groupings than are those of males. Female college graduates accumulate approximately the same amount of work experience during their early to middle twenties as male college graduates. Time spent either unemployed or out of the labor force also is similar. However, as we turn to increasingly less educated groups, we find that the time spent employed during this period drops sharply and is approximately offset by a rise in the percentage of time spent out of the labor force. Indeed, female high school dropouts were employed only about one-third of the time during their early to middle twenties. Among females, a natural explanation for these differences in participation rates are differences in childbearing patterns, because less educated women's first births generally occur earlier than those of more educated women (Geranimus and Korenman 1992). However, by comparing tables 3.1 and 3.2, we see that the timing of births cannot fully explain the low participation rates of less educated females early in their careers. The less educated work markedly less than their better educated counterparts, even as they age into the part of the life cycle in which births among more educated women are relatively more likely. The substantial persistence of their low employment rates even into nonchildbearing years might reflect other costs associated with their early childbearing (such as child care costs, lost work experience, and welfare dependence) or other factors that relate primarily to their low skills. s 130

/

Job Change and Job Stability Among Less Skilled Young Workers TABLE 3.2

I

Labor Force Participation of Young Adults (Percentage) Labor Market Status'

Demographic Group Males High school dropout High school graduate Some postsecondary (no schooling after age twentytwo) College graduate (no schooling after age twentytwo) Attends college after age twentytwo Females High school dropout High school graduate Some postsecondary (no schooling after age twentytwo) College graduate (no schooling after age twentytwo) Attends college after age twentytwo

Employed

Military

OLF

Unemployed

75.8 80.6 83.4

0.0 5.1 8.1

11.9 6.8 4.1

12.2 7.4 4.4

88.1

3.5

5.2

3.2

74.0

4.9

16.3

4.8

36.8 65.4 76.8

0.0 0.4 0.8

55.9 29.2 18.5

7.3 5.1 3.8

86.2

0.8

9.9

3.1

77.4

1.1

17.7

3.7

Source: Data from National Longitudinal Survey of Youth 1979. Note: Sample size is 9,295 observations. Observations are weighted using sample weights. Individuals in the sample must have birth years between 1957 and 1964 and a complete job history covering the five-year time period between ages twenty-three and twenty-seven. "OLF" is defined as out of the labor force and not serving in the armed forces. The" College graduate" category includes only those who do not acquire any additional postsecondary schooling after age twenty-two. "Attends college after age twenty-two" refers to persons with at least some postsecondary schooling by age twenty-two who also report attending school between ages twenty-three and twenty-seven. "Figures are percentage of weeks spent in various categories during the five-year period from the age of twenty-three to the age of twenty-seven.

THE TRANSITION RATES INTO AND OUT OF EMPLOYMENT BY YOUTHS AND YOUNG ADULTS Less skilled young adults exhibit less attachment to the employed workforce than other workers, and this pattern is maintained as they mature. This lack of attachment appears to be a barrier to future employment and to the acquisition of productivity- and wage-enhancing on-the-job training. By definition, the prob131

Finding Jobs

ability of being employed in any period for an individual or group is determined by their transition rate from nonemployment to employment (Pne) and the transition rate from employment to nonemployment (p",,).Y The latter reflects the frequency or incidence of spells of nonemployment, which in turn reflects employment instability; the former reflects the probability of getting a job or the inverse of the average durations of spells of nonemployment. To illustrate the importance of these two transition rates in explaining the difference in employment rates among educational groups, we follow Ballen and Freeman (1986) and compute the respective fraction of employment and nonemployment spells that end during any given week. As shown by table 3.3, our

TABLE 3.3

/

Transition Rates Into and Out of Jobs, by Race, Gender, and Education (Percentage of Jobs Ending in a Given Week)

Category

Pen

P ne

Race and gender White males Black males White females Black females

1.0 1.3 1.2 1.3

5.0 2.7 2.6 1.6

1.8 0.9 0.6

1.7 3.4 4.0 5.0

2.0 1.8

1.4 2.6

1.3 1.1

2.4 3.6

0.9 0.9

3.0 4.3

0.5 0.6

5.1 4.9

Education level High school dropouts High school graduates Some college College graduates Education by race High school dropouts Black White High school graduates Black White Some college Black White College graduates Black White

1.1

Source: Authors' calculations from the NLSY79. Employment histories are computed using the weekly employment summaries in the NLSY Work History file. Note: Pcn and Pne represent the weekly probabilities of transition from employment to nonemployment and from nonemployment to employment, respectively.

132

Job Change and Job Stability Among Less Skilled Young Workers

computations indicate that a larger portion of the difference in employment rates by race and gender results from differences in the probability of getting a job, whereas differences in transitions out of employment are relatively minor. to This finding is consistent with Ballen and Freeman's results using the early waves of NLSY79 and the National Bureau of Economic Research's Inner-City Youth Survey. Among educational groups, the results are a bit more mixed. Although the low transition rates out of nonemployment clearly explain the larger share of the gap between the employment rates of high school dropouts and high school graduates, table 3.3 also reveals that those with low levels of education are markedly more likely to leave employment for nonemployment. The figures in the table indicate that in any given week, a high school dropout is 63 percent more likely (that is, [1.8 - 1.1]/[1.1 X 100 percent]) to move from employment to nonemployment than is a high school graduate. We can illustrate the importance of these differences among educational groups by considering the steadystate employment rates implied by these transition rates. The figures in the table imply an employment rate of 50 percent for high school dropouts compared with 77 percent for high school graduates. Of the resulting 27 percentage point gap, approximately one-third is a function of high school dropouts' higher transition rates out of employment and two-thirds a function of their lower transition rate out of nonemployment. Thus, a closer examination of the causes of and distinction between job and employment instability among less skilled and more skilled workers appears to be in order.

EMPIRICAL TRANSITION RATES FROM JOBS In tables 3.4 and 3.5, we present means on transitions out of jobs rather than from employment more generally. From each job two transitions are possible: from job to nonemployment and from job to job. Previous research suggests that the latter of these transitions is most likely associated with earnings growth and is more likely among the skilled workers (Topel and Ward 1992; Royalty 1998). By contrast, the former of these transitions is less likely to lead to earnings growth and is more likely among the less skilled. These transitions also correspond more closely to transitions from employment to nonemployment that were considered in the previous table. We present the mean transition rates out of jobs as well as the mean rates into nonemployment and other jobs. We also distinguish between voluntary and involuntary transitions-that is, quits versus layoffs, discharges, and the like. The results appear separately by gender and by educational attainment and quartile of the math test score distribution. As shown in tables 3.4 and 3.5, during the first eighteen months of a job, the weekly probability of leaving that job is about 2 percent. This transition rate is a bit higher than that observed by Royalty (1998), most likely because of differences between the way we constructed our samples and in the way we defined

I

133

Finding Jobs TABLE 3.4

/

Type of Job Ending Any job transition Involuntary transition Voluntary transition Job-to-job transition Involuntary transition Voluntary transition Job to nonemployment transition Involuntary transition Voluntary transition

Job Transition Rates, by Gender and Level of Education (Percentage of Jobs Ending in a Given Week) High School Dropouts

High School Graduates

Some College

College Grads

Males

Females

Males

Females

Males

Females

Males

Females

2.4

2.8

2.1

2.0

2.1

1.8

2.0

1.9

1.5

1.5

1.3

1.1

1.3

1.0

1.3

1.1

0.8

1.2

0.7

0.8

0.8

0.8

0.7

0.7

0.6

0.5

0.6

0.5

0.6

0.5

0.4

0.4

0.3

0.2

0.3

0.2

0.3

0.2

0.2

0.2

0.3

0.3

0.3

0.2

0.3

0.2

0.2

0.2

1.7

2.3

1.4

1.5

1.5

1.4

1.6

1.5

1.2

1.3

1.0

0.9

1.0

0.8

1.1

0.9

0.5

1.0

0.4

0.6

0.5

0.5

0.5

0.5

Source: Authors' calculations from the NLSY79. Sample limited to persons who were inter-

viewed in 1994, had complete job histories during the years that they were scheduled to be interviewed, and were employed at least once between 1978 and 1993. Job histories are constructed from the NLSY work history file. Transition rates are for the first eighteen months of a job.

transitions. 1I However, our transition rates are fairly comparable to those reported by Farber (1998). Overall, transition rates out of jobs do not vary greatly by gender. Women who are high school dropouts, however, are approximately 16 percent more likely to leave a job in a given week than their similarly educated male counterparts (table 3.4). As shown by the last row of the table, this gap in transition rates results because women are twice as likely to voluntarily leave a job for nonemployment. This finding is consistent with the notion that many less educated women leave jobs for reasons related to childbearing and child rearing. The transition rates presented in table 3.4 do not vary substantially among educational categories of high school graduate or more. However, when we limit our analysis to individuals who are twenty-two years or older, we find that college graduates have lower transition rates from jobs than do these with a high school diploma or some college education. More striking in the table is the dif-

134

I

Job Change and Job Stability Among Less Skilled Young Workers TABLE 3.5

/

Type of Job Ending

Any job transition Involuntary transition Voluntary transition Job-to-job transition Involuntary transition Voluntary transition Job to nonemployment transition Involuntary transition Voluntary transition

Job Transition Rates, by Gender and Math Scores, Armed Forces Qualifying Test (Percentage of Jobs Ending in a Given Week) 1st Quartile

2d Quartile

3d Quartile

4th Quartile

Males

Females

Males

Females

Males

Females

Males

Females

2.4

2.2

2.3

2.2

2.1

2.0

2.0

1.7

1.5

1.2

1.5

1.2

1.3

1.1

1.2

1.1

0.9

1.0

O.S

0.9

O.S

O.S

0.7

0.6

0.6

0.4

0.4

0.7

0.5

0.6

0.5

0.5

0.3

0.2

0.2

0.3

0.2

0.3

0.2

0.3

0.3

0.2

0.2

0.3

0.3

0.3

0.3

0.2

1.S

1.S

1.6

1.7

1.5

1.5

1.4

1.4

1.2

1.0

1.1

1.0

1

0.9

1.0

0.9

0.5

O.S

0.4

0.7

0.5

0.6

0.4

0.4

Source: Authors' calculations from the NLSY79. Sample limited to persons who were interviewed in 1994, had complete job histories during the years that they were scheduled to be interviewed, and were employed at least once between 1978 and 1993. Job histories are constructed from the NLSY work history file. Transition rates are for the first eighteen months of a job.

ference between the job transition rates of high school dropouts and all other education groups. The figures indicate that during the first eighteen months of a job a female high school dropout is approximately 50 percent more likely to leave that job in any given week than are other women. Male dropouts are about 20 percent more likely to leave a job than are other men. The greater job instability for less educated women is again a result of higher voluntary transition rates into nonemployment. By contrast, for less educated males job instability results from higher involuntary transitions into nonemployment. Turning to table 3.5, we see that similar patterns appear to hold for our measures of cognitive achievement: the mathematics portion of the Armed Forces Qualifying Test (AFQT). We also obtained similar results when using the verbal portions of the AFQT. In our work here we focus on the math score as a explanatory variable because it appears to be more strongly related to earnings in much of the literature (for example, Murnane, Levy, and Willett 1995), and it may be less subject to criticisms that the test is inherently racially or ethnically biased. To see whether these transition rates vary with tenure on the job, we present

I

135

Finding Jobs

weekly Kaplan-Meier hazard rates in figures 3.1 through 3.6 for roughly the first eighteen months in the job. We present these hazard rates separately by gender and education (that is, high school dropouts versus all other educational categories). We also present them separately for transitions into nonemployment and into other jobs. In our sample, the pattern of transition rates from new jobs corresponds to the familiar pattern depicted elsewhere in the literature (for example, Farber 1998, figure 5). For every demographic group, the hazard rates rise during the first few weeks on the job and reach a peak after approximately twelve to sixteen weeks. After this point, the hazard declines-at first sharply and then more slowly through the first eighteen months of the job. This pattern is the same for all educational groups, but the hazard rates are generally higher for high school dropouts than for other workers (see figures 3.1 and 3.2). The pattern of transition rates for women is similar to those for men. However, during the first sixteen weeks of a job the transition rates of the least educated women are especially high, and they remain much higher than those of more educated groups for much of the first year on the job. Even after eighteen months, the hazard rates of women high school dropouts remain above those of their more educated counterparts. This pattern suggests that at least a portion of the low employment rates experienced by less educated women can be accounted for by very high transition rates out of jobs, especially during the early parts of their job spells. As shown in figures 3.3 through 3.6, we find that during the first eighteen months on the job there are differences between the patterns of transitions from jobs into nonemployment and from one job to the next. Among males, the patFIGURE 3.1

/

Overall Job Transitions, Males, by Education

5.0 4.5

- - High school dropouts - - High school graduates

4.0 3.5 (i)

be ru

3.0

~ (jJ

2.5

'"

2.0

u

(I)

p..,

1.5 1.0 0.5 0 1

Week Source: Data from National Longitudinal Survey of Youth, 1979. Note: "High school graduates" refers to all workers with at least a high school diploma.

136

I

Job Change and Job Stability Among Less Skilled Young Workers FIGURE 3.2

/

Overall Job Transitions, Females, by Education

5.0 4.5

- - High school dropouts - - High school graduates

4.0 3.5 ~ 3.0 ....!:::Ol 2.5 OJ ~ 2.0 Il.. 1.5 1.0 0.5 0

1

5

9

13 17 21 25 29 33 37 41

45 49 53 57 61

65 69

73 77

Week

Source: Data from National Longitudinal Survey of Youth 1979. Note: "High school graduates" refers to all workers with at least a high school diploma.

tern of job-to-nonemployment transitions (that is, the initial rise in hazard rates and subsequent decline) depicted in figure 3.3 is similar to but much more striking than the pattern depicted for job-to-job transitions in figure 3.4. Moreover, the difference between the job-to-nonemployment transition rates of high school FIGURE 3.3

/

Transitions from Job to Nonemployment, Males, by Education

4.5 4.0

- - High school dropouts - - High school graduates

3.5 ~

3.0

.... 2.5 Ol

8iu '"OJ Il..

2.0 1.5 1.0 0.5 0

1 Week

Source: Data from National Longitudinal Survey of Youth 1979. Note: "High school graduates" refers to all workers with at least a high school diploma.

I

137

Finding Jobs FIGURE 3.4

/

Transitions from Job to Job, Males, by Education

1.4 1.2

- - High school dropouts - - High school graduates

1.0 HS School

All < HS HS > HS

All < HS HS > HS

Organized

Family

All < HS HS > HS

Relative

All < HS HS > HS

Parent

nnnn

All < HS HS > HS

In-home

Type of Care

Source: Authors' calculations from 1990 to 1993 SIPP panels. Note: Each bar represents the percentage of children using each type of care (paid or unpaid), and the shaded region represents the percentage using paid care of each type. Numbers above bars represent the average weekly dollar amount paid for child care. HS refers to high school.

tives' substituting for fathers as caregivers for the children of unmarried mothers. Most notable in terms of costs is that the amount paid tends to increase with skill, but given the income disparities across groups, a much larger fraction of income is spent by the low-skilled mothers who pay for care. Because of the much lower incomes of families headed by a single mother, the fraction of family income spent on child care is much larger for unmarried mothers. Finally, for all groups the probability of paying for care generally rises with the skill level of the mother, with the notable exception being the case of care by a relative, for which it drops. What do these results tell us regarding the potential usage of day care services by the additional million or so mothers with children under the age of six who may enter the labor market as a result of welfare reform?12 One place to look for indications is the decisions made by welfare recipients in the past. Within our sample from the SIPP are 553 welfare recipients with children under the age of six who are using child care. Although these sample sizes are too small to reliably investigate the types of care used, we can calculate that 36.7 percent of these AFDC mothers with children under the age of six pay for care, and those who pay spend an average of fifty-seven dollars a week for that care. 432

Child Care and Mothers' Employment Decisions

Historically, however, most welfare recipients have not worked (perhaps because of child care constraints), and those who did may not be representative of those who will be forced to enter the labor market following welfare reform. An alternative reference group is all single mothers who have children under the age of six and who have dropped ou t of high school. For these women, as can be seen in table 10.1, a slightly higher fraction use paid care, at 41.8 percent. Weekly costs are fairly similar for this group and the AFDC group, at fifty-nine dollars. We can use these statistics to conduct back-of-the-envelope calculations regarding the cost of providing fully subsidized day care for these one million women, assuming that there is an unlimited supply of the type of care currently used.13 If only 41.8 percent of these women required paid care to enter the labor market, as their reference group does, we would expect the total cost to equal $1.28 billion per year (1 million times fifty-nine dollars a week times fifty-two weeks a year times 41.8 percent using paid care). Realistically, however, we can expect that more of the women who would otherwise turn to welfare will require paid care than this reference group, because if free care had been available through, say, a relative, they would most likely have utilized those services already. Alternatively, then, one could provide an upper bound on the cost by assuming that all will use paid care. In that case, the additional cost to provide fully subsidized day care for each of the one million women would be $3.08 billion a year. These estimates compare with an increase in federal funding brought about by welfare reform of $4 billion over six years, or about $667 million a year. Thus, the currently planned increase in federal spending could provide a subsidy of between 22 and 52 percent for each of the women likely to enter the labor market as a result of welfare reform.

EXISTING EVIDENCE ON CHILD CARE COSTS AND EMPLOYMENT The existing evidence on the impact of child care costs on the employment decisions of women comes mainly from econometric studies using nonexperimental data, although a few do take a "natural experiment" approach to the issue. Government-sponsored demonstration projects are another potential source of information. Many of the pilot welfare-to-work programs included a child care component, and the evaluation of the outcomes of these projects may shed light on the role child care expense plays in the employment decisions of poor women.

Evidence Obtained from Demonstration Projects Government-sponsored demonstration projects often offer considerable advantages in examining the impact of alternative policies on individual behavior. Well-designed projects randomly divide a sample of individuals eligible for a particular program into treatment and control groups that are comparable on any (observable or unobservable) dimension. Those in the treatment group reI

433

Finding Jobs

ceive the benefits and pay the costs of the relevant policy, and those in the control group face an environment no different from what they would ordinarily experience. A simple comparison of the aggregate behavior of members of the treatment and control groups identifies the impact of the policy. Policies that reduce or eliminate the cost of child care have been included in a number of recent demonstration projects targeted at improving the labor market performance of different subgroups of less skilled workers. These projects are summarized in table 10.4, which categorizes them according to both the policy initiatives that inspired the demonstration and the target population. Many of the demonstrations began as evaluations of early welfare reform policies, like the JOBS (Job Opportunities and Basic Skills) program initiated under the 1988 Family Support Act, and federal waivers granted to states to introduce alternative welfare policies under the AFDC system. They all include a large array of financial incentives and participation requirements, in addition to enhanced services, including child care. Because so many different policies are incorporated into the treatment, it is difficult to quantify the specific role that the child care component has played if large employment gains are observed. '4 On the other hand, if little or no increase in employment is observed, one may reasonably conclude that the provision of additional child care services probably has done little to improve the labor market success of these women. 'S Results from the different demonstrations are somewhat mixed, depending upon the target population and the services and incentives provided. Those demonstrations conducted in conjunction with the JOBS program and with state waiver policies generally did increase employment levels of the welfare population. Although some of the employment gains were impressive in percentage terms, the absolute gain in employment rates was typically quite small. For instance, in the third year of the California Greater Avenues for Independence (GAIN) demonstration, 40 percent of the treatment group worked at some point during the year, compared with 34 percent of the control group-about an 18 percent increase in employment rates. With a moderate employment effect such as this, along with the large array of other components of the treatments, it seems unlikely that child care played a very large role in increasing employment. '6 Moreover, in the evaluations targeted at teen mothers that included child care assistance, little success in improving employment rates was observed. Although the confluence of services, mandates, and incentives in these demonstrations suggests caution in interpreting their results, based on this evidence it seems reasonable to conclude that subsidized child care may have a modest effect, at best, in increasing employment levels of very low-skilled, single mothers with small children.

Econometric Evidence A substantial body of research has used data generated in a nonexperimental setting in combination with econometric techniques to explore the relationship

434

I

/

Minnesota's Family Investment Program

National Evaluation of Welfare-to-Work Strategies (the JOBS Evaluation)

Location and Time of Random Assignment

Job search assistance; short-term education and training

Mandatory education; job search

Other Components of Treatment

Single-parent new applicants and longterm (two or more years) welfare recipients

Seven counties in Minnesota, 1994

Financial incentives increasing the returns to work for all participants; mandatory employment and

Evaluations of State Programs Implemented Under Federal Waivers

Seven sites across the nation, early results available from three sites (Atlanta, Ga.; Grand Rapids, Mich.; and Riverside, Calif.)

Six counties in California

Evaluations of the JOBS Program

Heads of unmarried families with children mostly age six and older and all heads of two-parent families Heads of unmarried families with children age three and older and all heads of twoparent families

Target Population

Demonstration Projects with Child Care Components

California's Greater Avenues for Independence (GAIN)

Demonstration

TABLE 10.4

(Table continues on p. 436.)

Employment and earnings increased for long-term welfare recipients who faced both financial incen-

Treatment focusing on rapid labor force entry led to increased employment and reduced welfare use. Treatment focused on skills development had little impact in the short term

Increases in employment and earnings and reductions in welfare payments

Results

Mothers age sixteen to twenty-two who gave birth as teenagers, mostly high school dropouts

Teenage mothers who began collecting AFDC for the first time

New Chance Demonstration

Teenage Parent Demonstration

One county in Florida, 1994

Illinois and New Jersey, 1987 to 1991

sixteen sites in ten states, 1989 to 1992

Enhanced case management; self-sufficiency plans; transportation assistance

Employability development classes; vocational training; work internships, job placement

Evaluations Targeted at Teen Mothers

Vast majority of welfare caseload

Florida's Family Transition Program

training activities for some long-term recipients Large array of work incentives, including sanctions for nonparticipation, time limits on benefit receipt, and increased income disregards

Did not improve labor market or public assistance outcomes for members of the treatment group in any discernible way Some success in improving the labor market outcomes of teen mothers on welfare in the short run, but these effects dissipated after the program's services and mandates ended

tives and mandatory employment-related activities Increased employment and earnings but no reduction in welfare receipt

Child Care and Mothers' Employment Decisions

between child care costs and female labor supply. The existing literature has employed many different methodologies, data sets, and sample restrictions, so it is perhaps not surprising that the range of estimates of the elasticity of employment to child care costs is reasonably large. In this review we present a summary (which is not intended to be exhaustive) of the econometric evidence that has been gathered to date, which focuses narrowly on this one important issue in the child care market. One useful way to categorize these studies is to consider the source of the variability in child care costs and the methodological approach taken to study the impact of that variability on employment decisions. Table 10.5 summarizes selected previous research in this manner. The most common approach uses the variation in child care costs across individuals. These costs vary because of things like the type of care utilized and regional differences in day care costs. Some studies also rely on differences in the tax treatment of child care expenditures that vary nonlinearly with family income. Most of these papers employ a similar methodology. A probit model is estimated on the discrete employment decision, and the key covariates are the wage rate and child care costs. Because both of these variables are only observed for those women who have chosen to work, predicted values obtained from regression models for employed women and those who use paid child care are used instead. Corrections are made for the potential sample selection bias that may result from estimating these prediction equations on only these subsets of women. Other methodological approaches are also utilized that take advantage of individual variation in child care costs. Charles Michalopoulos, Philip Robins, and Irwin Garfinkel (1992) and David Ribar (1995), for instance, formulate a complete structural model based on utility-maximizing behavior and specific functional form assumptions and estimate the parameters of the model. Susan Averett, Elizabeth Peters, and Donald Waldman (1997) apply the approach introduced by Jerry Hausman (1981) for incorporating the nonlinearities of the budget constraint faced by individuals because of the specific features of the tax system. In their case, they exploit the variation created by the Child Care Tax Credit. They estimate a maximum likelihood model that incorporates the probability that an individual's choices rest on any particular segment of that budget constraint. Some studies use only geographic variation in the costs of child care. Methodologically these studies are quite similar to those described earlier that estimate probit models, including sample-selection-corrected predictions of child care costs and wages; the obvious difference is that the child care cost varies only by location of residence and not across individuals. The final methodological approach examines the impact of events that exogenously separate women into different child care cost categories, called natural experiments, and then compare employment patterns across the different groups. For instance, Mark Berger and Dan Black (1992) consider differences in employment between two groups of low-income women in Kentucky, one of

I

437

Finding Jobs

TABLE 10.5

I

Reference

Summary of Selected Previous Research Estimating Elasticity of Employment to Child Care Costs

Methodology

Data

Population

Estimated Employment Elasticity

Individual Variation in Child Care Costs Averett, Peters, and Waldman (1997)

ML estimates of kinked budget constraint model (Hausman 1981)

1986 wave of the National Longitudinal Survey of Youth

Connelly (1992)

Probit model with sample selection corrections

Wave 5 of 1984 SIPP panel (winter 1984 to 1985)

Government Accounting Office (1994)

Probit model with sample selection corrections

1990 National Child Care Survey

Kimmel (1995)

Probit model with sample selection corrections

Kimmel (1998)

I'robit model with sample selection corrections

Wave 6 of the 1987 SIPI' panel and wave 3 of the 1988 SIPI' panel (July 1988 to December 1988) Wave 6 of the 1987 SIPP panel (July 1988 to December 1988)

438

I

Married women between the ages of twenty-one and twentynine with children under the age of six Married women between the ages of twenty-one and fifty-five with children under the age of thirteen All mothers with at least one child under the age of thirteen Single mothers in poverty

Single and married mothers with children under the age of thirteen

-0.78

-0.20

- 0.50 (poor) -0.34 (near poor) -0.19 (nonpoor) -0.35

- 0.22 (single) -0.92 (married)

Child Care and Mothers' Employment Decisions

TABLE 10.5

/

Reference

Continued

Methodology

Data

Population Single and married mothers with children less than age eighteen Married women with children under the age of fifteen Married women with children under the age of fifteen

Michalopoulos, Robins, and Garfinkel (1992)

Estimation of structural model

Wave 5 of 1984 SIPP panel (winter 1984 to 85)

Ribar (1992)

Probit model with sample selection corrections

Wave 5 of 1984 SIPP panel (winter 1984 to 1985)

Ribar (1995)

Estimation of structural model

Wave 5 of 1984 SIPP panel (winter 1984 to 1985)

Estimated Employment Elasticity

0.00 for both married and single mothers

-0.74

-0.07 to -0.09

Geographic Variation in Child Care Costs Blau and Robins (1988)

Probit model with sample selection corrections

Employment Opportunity Pilot Projects 1980 Survey

Han and Waldfogel (1998)

Probit model with sample selection corrections

1991 to 1994 March CPS

Married women under the age of 45 with one child under the age of fourteen Single and married mothers with children under the age of six

-0.38

Unmarried child care subsidy recipients and those on waiting list in Kentucky

not available

- 0.31 (single) -0.21 (married)

Natural Experiments Berger and Black (1992)

Probit model with sample selection corrections

Telephone survey

(Table continues on p. 440.)

439

Finding Jobs TABLE 10.5

Reference

Gelbach (1997)

I Continued Methodology

IV model using child's quarter of birth as instrument

Data

1980 census (5 percent sample)

Population

Single mothers age fifty and under with one fiveyear-old child

Estimated Employment Elasticity

-0.13 to -0.36

which comprises women enrolled in a program that provides subsidized day care, and the other, women who are on a waiting list for the program. Because both groups have at least attempted to enroll in the program, one could argue that they must have similar observable and unobservable characteristics. A second natural experiment, studied by Jonah Gelbach (1997), examines employment of single mothers whose five-year-old children differ by their quarter of birth. Because public schools effectively provide fully subsidized day care, mothers of children born just before a school's enrollment cutoff date should be more likely to work in the year their children turn five compared with mothers whose children were born just after that date. All of these research designs contain potential weaknesses. The sample-selection correction techniques employed by most studies require for proper identification variables that are related to an individual's decision to work but unrelated to potential wages or child care costs. Similarly, for the final probit to be identified there must be variables related to potential child care costs that are not independently related to the employment decision. Different studies take different approaches to this issue, and many of the decisions seem ad hoc, at bese7 Those papers that estimate structural models encounter additional difficulties in that a number of functional form assumptions need to be made that mayor may not accurately reflect individual behavior. Perhaps the well-conceived natural experiments provide the strongest methodology, but their results, based on the experience of low-income women in Kentucky and single mothers of five-yearolds, cannot necessarily be generalized to the entire population of mothers with children. Despite these potential weaknesses, these studies do virtually uniformly find a negative relationship between child care costs and maternal employment, regardless of the econometric technique. Nevertheless, as can be seen in table 10.5, the range of estimated employment elasticities with respect to a change in the cost of child care across studies is rather large, ranging from just over zero to almost one. There does seem to be some clustering of estimates around an elasticity of about - 0.3 to - 0.4 or SO.18 However, the papers taking a structural approach are uniformly on the low end of the estimates, yielding estimates between 0 and - 0.1. Aside from the methodological differences discussed above, some studies use data on all mothers, others use single mothers, and still others 440

I

Child Care and Mothers' Employment Decisions

concentrate on married mothers. Some focus on mothers in low-income families. Some restrict their analysis to women with preschool-age children (under the age of six), and others include women with children up to the age of fifteen. Under these circumstances, pinning down the specific factors that generate the discrepancies across studies is virtually impossible. Additionally, none of the existing studies focuses directly on differences across skill levels, the key issue for this study. With these limitations of previous research in mind, we perform our own econometric analysis using the pooled SIPP data described earlier, with an eye toward both evaluating the overall stability of the estimates and investigating the effect of skill level.

ECONOMETRIC ANALYSIS USING DATA FROM THE SURVEY OF INCOME AND PROGRAM PARTICIPATION, 1990 TO 1993 For our econometric analysis, we employ the most common methodological approach used in this literature, a probit model relating maternal employment to wages and child care costs, controlling for sample selection problems. To specify the estimating equations, one must first consider what the underlying structural equations look like. There are four such underlying equations: a wage equation, a market price of care equation, a conditional use of paid care equation, and a labor force participation equation. As discussed in more detail below, reducedform versions of these latter two are necessary for estimating selection-corrected versions of the former two. Thus, it is the labor force participation equation that provides us with estimates of the elasticity with respect to child care costs, as measured by the market price. Note that in measuring child care expenditures in this way, we directly estimate the policy-relevant concept of the overall response to a subsidized price of child care. The basic approach starts by specifying the wage equation as a function of human capital variables and labor market characteristics. Similarly, the market price of care is specified as a function of demographic characteristics thought to influence the type of care chosen and child care market characteristics. The probability of using paid care is expected to be a function of similar demographics reflecting preferences, variables representing availability of unpaid options, and the market price of care. Finally, labor force participation is modeled as a function of variables capturing nonmarket opportunities, demographics that may reflect preferences for labor versus leisure, the market wage, and the market price of care. Although the exact specification varies by study, all follow this basic approach. We obtain our specification by beginning with the reasonably parsimonious specification used by the General Accounting Office (1994), altering it somewhat to reflect the available data and our judgment about appropriate variables. Starting, then, with our specification of the wage equation, the model underlying our estimates is I

441

Finding Jobs

In (hourly wage) = f(race, education, age, age squared, marital status, number of children, urban status, region, disability status, state unemployment rate, year). (10.1) In contrast with the procedures of the General Accounting Office (GAO) (1994), we first switch from experience and experience squared to age and age squared. Because the GAO measured experience simply as age minus education minus six, this is not a large change. To better capture likely work experience, we add a variable for the number of children, to account for the possibility that time has been spent out of the market. We add an indicator for a health problem of the mother. This can proxy for both lost time owing to health problems and restrictions on the types of jobs that can be done. We also add the unemployment rate in the state in that year, to capture the tightness of the labor market. 19 Finally, we add year dummies to account for any other changes in the wage structure over the time period. Our specification of the market price of care equation, which incorporates some slightly more significant changes to the GAO model, is as follows: In(market price of child care per hour worked) = f(number of children under six, number of children age six to twelve, presence of older children, presence of an adult other than self or spouse, presence of an unemployed adult, other family income, age, education, marital status, urban status, region, year). (10.2) In this case, we first switch the dependent variable from price of care per week to the log of the price of care per hour worked. We also switch from an indicator for the presence of additional income beyond one's own and one's spouse's earnings to the amount of that other income. We then add age, as it may correlate with a preference for different types of care. Additionally, older mothers may be less likely to have parents that are viable child care providers. We also add an indicator for whether an unemployed adult is in the household (this is in addition to the indicator for any other adult). By contrast, the GAO report uses this variable only in the probability of paid care equation. However, the labor force status of the adult may impact both the availability of such an adult to provide child care and how much (if any) is paid for that care. Recall that for close to 28 percent of the children in relative care, that care is paid for. Year dummies are added to account for any other changes in the child care market over the time period. A participation equation is important both for the final estimates and for creating selection correction terms for the wage and cost of care estimates. Our equation is P(participation in labor force) = f[ln(hourly wage), In(hourly market price of child care), other family income, number of children under six, number of children age six to twelve, age, marital status, education, disability (10.3) status, urban status, state maximum AFDC benefits],

442

I

Child Care and Mothers' Employment Decisions

where, again, we have made some significant alterations to the GAO model. First, we add other family income (as above) and maximum AFDC benefits for a family of three in the state in that year to capture nonlabor market income opportunities. We also add the indicator for mother's health problem, because such problems can interfere with market work. Finally, we add age and education, to proxy for cohort and peer group differences in the propensity to work. Although we do not estimate a structural use of paid care equation, it is useful to specify one before turning to the reduced-form version necessary for the selection model. Our specification is P(pay for child care using child care) = f[ln(hourly wage), In(hourly market price of child care), presence of older children, presence of an adult other than self or spouse, presence of an unemployed adult, race, age, education, marital status]. Essentially, we assume here that presence of potential caregivers will be important and that age, education, and race may be correlated with preferences for formal versus informal care. Based on these specifications, we obtain the following reduced-form equations: P(participation in labor force) = f(race, education, age, age squared, marital status, number of children, urban status, region, disability status, state unemployment rate, year, number of children under six, number of children age six to twelve, presence of older children, presence of an adult other than self or spouse, presence of an unemployed adult, urban status, other family income, state maximum AFDC benefits), (10.4) P(pay for child care) f(race, education, age, age squared, marital status, number of children, urban status, region, disability status, state unemployment rate, year, number of children under six, number of children age six to twelve, presence of older children, presence of an adult other than self or spouse, presence of an unemployed adult, other family income). (10.5) Equation 10.4 is then used to construct a selection correction term that is included in wage equation 10.1. A bivariate probit is used to jointly estimate equations 10.4 and 10.5 to obtain two separate selection correction terms that are included in the market price of care equation 10.2. The predicted wage and predicted market price of care are then used in equation 10.3 to estimate the effect of child care costs on labor force participation. Thus, the selection correction model used to predict the wage is identified by excluding the number of children under the age of six, the number of children age six to twelve, the presence of older children, the presence of an adult other than self or spouse, the presence of an unemployed adult, other family income,

I

443

Finding Jobs

state maximum AFDC benefits, and the state unemployment rate from the wage equation. Similarly, the selection correction model for the market price of child care is identified by excluding the state unemployment rate, the state maximum AFDC benefits, the total number of children, disability status, and age squared from the cost of care equation. At the same time, the effect of the wage is identified in the final probit by virtue of the exclusion of race, region, year, total number of children, the state unemployment rate, and age squared from the structural probit. Similarly, the cost of care is identified by excluding the presence of older children, the presence of an adult other than self or spouse, the presence of an unemployed adult, region, and year from the structural probit. 2o Not everyone may agree with the choices made, but the methodology does appear to result in a relatively stable model, as detailed below, much more so than the GAO specification upon which it is based. The stability of our model compared with the GAO specification is investigated in table 10.6. For this stability check, we first estimate the baseline specifications just described on the full sample of all mothers with children under the age of thirteen. These specifications are shown as column 1 in table 10.6. We then make a series of small changes to the specifications. In column 2, the bivariate selection model used in the prediction of the price of child care is changed to a simple univariate selection model based only on the use of paid care. Maintaining the univariate selection, the model in column 3 measures the price of child care as the level of the price per hour worked, whereas in column 4 it is measured as the level of the price per week. These changes represent the main types of alternatives represented by the past literature. In the top panel of the table, we see that across all models, our specification leads to elasticities with respect to the market price of child care of between - 0.3 and - 0.5, which is a relatively tight range. By contrast, the bottom panel presents estimates of this elasticity that range from about 0 to - 0.7. As Jean Kimmel (1998) also finds, the switch to the univariate selection does not seem to affect the estimates very much. The small changes in the way the cost of child care is measured are more important, though, especially in the bottom panel. This divergence of the estimates seems to underscore the importance of carefully thinking through the specification and associated identifying assumptions. Overall, then, these results are very supportive of Kimmel's findings on the sources of variation in the past estimates. The table also presents the estimated elasticities with respect to the wage, which are generally more stable. This increased stability most likely reflects both that the changes were focused on the price of care variable and that selection-corrected wage equations are quite standard in the literature, leading to more agreement on proper specification. Given that it is probably easier to obtain stable estimates of the wage elasticity, it is worth considering the relationship between the wage and cost of child care elasticities. Clearly, an increase in the cost of child care per hour worked can be viewed as analogous to a wage cut. However, this does not actually imply that the negative value of the wage elasticity should be comparable to the elasticity of employment to price of child care but rather that the pure behavioral

444 I

Child Care and Mothers' Employment Decisions TABLE 10.6

/

Test of the Stability of Alternative Estimates of the Effect of the Market Price of Child Care and Wages on Labor Force Participation, for All Women with Children Under the Age of Thirteen

Alternative estimates based on main specification Market price of child care

In(hourly wage)

Alternative estimates based on GAO-type specification Market price of child care

In(hourly wage)

(1)

(2)

(3)

(4)

-0.217 (0.033) [ -0.358] 0.347 (0.026) [0.572]

-0.233 (0.035) [ -0.384] 0.367 (0.027) [0.604]

-0.130 (0.018) [ -0.257] 0.400 (0.029) [0.660]

-0.005 (0.001) [ -0.528] 0.385 (0.028) [0.635]

-0.413 (0.046) [ -0.681] 0.573 (0.028) [0.944]

-0.362 (0.044) [ -0.597] 0.556 (0.028) [0.917]

-0.017 (0.015) [ -0.018] 0.374 (0.026) [0.616]

0.000 (0.001) [0.012] 0.343 (0.029) [0.565]

Note: Models are estimated using a probit model, but marginal effects are shown, along with standard errors in parentheses. Elasticities are shown in brackets. All models in the top panel also include other family income, number of children under the age of six, number of children age six to twelve, age, education, marital status, disability status, urban status, and state maximum AFDC benefits. All models in the bottom panel also include number of children under the age of six, number of children age 6 to 12, race, marital status, and urban status. Predictions of the cost of child care and of the wage also differ across panels; see the text for details. Column 1 is the base model, with the cost of child care measured as the log of the price per hour worked. Column 2 changes the bivariate selection model used to predict the price of child care to a univariate selection model. In addition to this, column 3 measures cost of child care as the level of the price per hour worked, and column 4 measures the cost as the level of the weekly price.

response to a dollar-for-dollar change in the wage and in the actual amount paid for care will be equal. That is, that

where Plfp is the probability of labor force participation, w is the wage, and c is the cost of care actually used. Recall, though, that we have measured the policyrelevant response of a change in the market price, not this behavioral response. Thus, we measure ap,fp where m is the market price of care, and, if Pp is the probability of usinlpaid care, then c = m * Pp or m = clPp. Assuming Pp to be constant implies that

I

445

Finding Jobs

and thus that

Finally, because a given dollar increase in the cost of care is not the same percentage change as that same dollar change in the wage, one must take into account the ratio of the cost to the wage. Thus, with the proper additional information, the behavioral wage elasticity can be used to predict the policy-relevant elasticity with respect to the market price of care as follows:

(10.6) making it possible to compare the response implied by our estimated child care elasticities with the response implied by our estimated wage elasticities. Having chosen a relatively stable model to work with, we now turn to an investigation of the effect of child care cost on labor force participation, with a focus on differences across skill groups. We also examine differences across skill levels within several subgroups, defined by marital status, the age of the child, and family poverty status, the categories previously examined in the literature. In every case, though, the prediction of the wage and market care price are estimated using the full sample of women with children under the age of thirteen. 21 Only the final probit is estimated separately by subgroup, by skill,22 Tables 10.7, 10.8, and 10.9 present the results from these final probit model estimations. In all cases, the table presents the derivative obtained from the probit estimates, along with the elasticity of participation, with respect to both the market price of child care per hour worked and the hourly wage. Also shown is the price of care elasticity implied by the estimated wage elasticity, as calculated in equation (10.6) above, using the average predicted probability of paying for child care, market price of care, and wage. We refer to this measure as the alternate care elasticity. Table 10.7 presents results for all women with children under the age of thirteen, by education and marital status. Looking first at the top panel, we see that there is a significantly negative estimated effect of child care cost on the labor force participation of these women. As expected, there is also a significantly positive effect of the wage. Overall, we estimate an elasticity of participation with respect to the market price of care to be - 0.358, similar to the central tendency of the past results. Looking across skill groups, the least skilled are slightly more responsive to the child care price, with an elasticity of - 0.394, than the most skilled, with an elasticity of - 0.293. The high school graduates are somewhere in between, at - 0.328. A similar pattern can be seen in terms of the

446

I

Child Care and Mothers' Employment Decisions TABLE 10.7 /

Estimates of the Effect of the Market Price of Child Care and Wages on Labor Force Participation, by Education and Marital Status for All Women with Children Under the Age of Thirteen All

All women In(hourly wage) In (market price of care) Elasticity Wage Care Alternate care n Participation rate Married women In(hourly wage) In(market price of care) Elasticity Wage Care Alternate care n Participation rate Unmarried women In(hourly wage) In(market price of care) Elasticity Wage Care Alternate care n Participation rate

Less than High School

High School

More than High School

0.347 (0.026) -0.217 (0.033)

0.328 (0.045) -0.142 (0.076)

0.276 (0.044) -0.202 (0.052)

0.300 (0.046) -0.205 (0.047)

0.572 -0.358 -0.055 20,587 0.607

0.912 -0.394 -0.103 3,684 0.360

0.447 -0.328 -0.044 8,152 0.617

0.427 -0.293 -0.040 8,751 0.702

0.336 (0.034) -0.190 (0.039)

0.402 (0.069) -0.069 (0.107)

0.306 (0.057) -0.190 (0.062)

0.274 (0.056) -0.204 (0.053)

0.535 -0.303 -0.052 14,895 0.629

0.983 -0.168 -0.104 2,055 0.409

0.487 -0.301 -0.047 5,736 0.629

0.396 -0.296 -0.039 7,104 0.691

0.320 (0.045) -0.260 (0.065)

0.235 (0.060) -0.205 (0.107)

0.199 (0.074) -0.239 (0.095)

0.319 (0.083) -0.173 (0.099)

0.582 -0.473 -0.055 5,692 0.550

0.790 -0.688 -0.097 1,629 0.298

0.339 -0.408 -0.034 2,416 0.587

0.428 -0.232 -0.033 1,647 0.745

Note: Models are estimated using a probit model, but marginal effects are shown, along with standard errors in parentheses. The market price of care is measured per hour worked. The alternate care elasticity is calculated based on the wage elasticity, predicted probability of paying for care, wage, and market price of care. See text for details. All models also include other family income, number of children under the age of six, number of children age six to twelve, age, education, marital status, disability status, urban status, and state maximum AFDC benefits.

I

447

Finding Jobs

wage elasticity, for which responsiveness declines with skill. Here the decline is much more steep, going from 0.912 to 0.447 to 0.427. These estimated wage elasticities imply alternate care elasticities that decline from - 0.103 to - 0.044 to - 0.040 across the skill groups and an elasticity of - 0.055 overall. Thus, although the pattern of decline by skill group is similar for the alternate care elasticities, the levels are quite different. In this case, the elasticities are down at the bottom of the range of estimates in the past literature. There are several possible explanations for this finding. One is simply that, in fact, individuals do not respond to changes in the cost of child care as merely a change in the net wage. 23 It seems unlikely, though, that this explains the entire disparity. Another possibility is that for those not currently using care, unobserved attributes (such as a lack of near-by relatives) may make access to unpaid care much less likely than would be predicted based on observed attributes. In this case, our predicted probability of using paid care will be underestimated, and hence our alternate care elasticities will be underestimated. However, even if we assume that the probability of using paid care is one, the effect implied by the wage elasticity will still be well below the directly estimated effect. For example, in the case of all women with children under the age of thirteen, this upper bound on the alternate care elasticity is - 0.159. 24 An alternative explanation involves the source of identification for the model. Differences in child care costs across regions are one of the main sources of variation leading to identification of our model. It may be the case, however, that regional differences in child care prices reflect more general differences in consumer prices and, thus, in other work-related expenses. For a given wage, then, a dollar increase in the cost of child care may represent much more than a dollar decrease in the net wage, because it correlates with many other increased work-related expenses. If this is the case, then our calculated elasticities will clearly overstate the effect of a change only in the price of care." Most of the studies reviewed in table 10.5 using this methodology also rely on geographic variation in one form or another. It is perhaps not surprising, then, that the smallest effects are found in studies estimating structural models. Note that it is typical in such models to begin with the assumption that women respond to the net return to hours worked. Overall, then, the two elasticities are probably best viewed as providing a range of estimates. This range is toward the lower end of that provided by the past literature. In any case, it seems clear that the responsiveness of labor force participation to price incentives decreases with skill level for women with children under the age of thirteen. Within marital status groups, only for unmarried women does the directly estimated price of care elasticity decline with skill level, falling from - 0.688 to - 0.408 to - 0.232. For the married women, the smallest elasticity, - 0.168, is estimated for the least skilled, and the other two groups have elasticities of about - 0.3. However, the alternate care elasticities derived from wage elasticities imply that responsiveness declines with skill for both married and unmarried women. In both cases, this decline is relatively sharp, although the

448

I

Child Care and Mothers' Employment Decisions

move from - 0.104 to - 0.047 to - 0.039 for married women is slightly smoother than that from - 0.097 to - 0.034 to - 0.033 for the unmarried women. Comparing across marital status groups, we see that unmarried women, with an elasticity of - 0.473, appear more sensitive to the price of care than do married women, who have an estimated elasticity of - 0.303. This pattern is also maintained for the lower-skill groups, although for the most skilled the married women are slightly more responsive. Overall, there is little difference by marital status in the alternate care elasticity implied by the wage elasticity, at - 0.052 and - 0.055 for the married and unmarried women, respectively. However, judging by skill level, although the estimates remain close, it is the married women for whom the elasticities are slightly larger. The opposite is generally true for the directly calculated elasticities, where only for the most skilled women is it the case that the care elasticity is not larger for the married women. Table 10.8 repeats the estimates of table 10.7 but limits the sample to women with at least one child under the age of six. Overall, the results are quite similar, although the smaller sample leads to somewhat larger standard errors. The directly estimated elasticity with respect to the market price of care is - 0.511 for all women with children under the age of six, slightly higher than before. Similarly, the alternate care elasticity of - 0.089 is also slightly larger than was found for all women with children under the age of thirteen. Both overall and by marital status, the care elasticity is generally declining with skill, whether calculated directly or from the wage elasticity. As is the case in table 10.7, the pattern by marital status is less clear. In general, the main findings of table 10.7 are robust to this change in sample. Because the sample change does not seem very important, table 10.9 returns to the full sample of all women with children under the age of thirteen but focuses on differences by both skill and family income level. We separate the sample into two groups, the nonpoor, defined as women in families with incomes at least 185 percent above the poverty line, and the poor and near-poor, defined as women in families whose incomes are less than 185 percent above the poverty line. 26 Starting with the nonpoor, the directly estimated care elasticity is - 0.186 overall, somewhat smaller than was found for all women in table 10.7. The alternate care elasticity of - 0.034 is similarly smaller. Although this alternate care elasticity is largest for the least skilled group, the directly calculated care elasticity reveals the opposite pattern. However, given the small number of nonpoor women with less than a high school education, the care elasticity for this group is imprecisely estimated, making it difficult to draw clear conclusions. Turning to the bottom panel, we observe that the estimated care elasticity for the poor and near-poor, at - 0.375, is larger than for the nonpoor. The same is not true for the alternate care elasticity, which at - 0.032 is similar to that for the nonpoor. There is little difference in either elasticity across the two higher-skilled groups, although both the care and wage elasticities are fairly imprecisely estimated for these groups. For the least skilled group, the care elasticities, - 0.649 when directly calculated or - 0.051 when calculated from the wage elasticity, are much larger than for the higher-skilled groups. Thus, the general finding of the

I

449

Finding Jobs

TABLE 10.8

/

Estimates of the Effect of the Market Price of Child Care and Wages on Labor Force Participation, by Education and Marital Status for All Women with Children Under the Age of Six All

All women In( hourly wage) In(market price of care) Elasticity Wage Care Alternate care

n Participation rate Married women In(hourly wage) In (market price of care) Elasticity Wage Care Alternate care n Participation rate Unmarried women In(houriy wage) In(market price of care) Elasticity Wage Care Alternate care n Participation rate

Less than High School

High School 0.240 (0.065) -0.269 (0.070)

More than High School

0.316 (0.038) -0.277 (0.044)

0.250 (0.056) -0.275 (0.090)

0.219 (0.074) -0.193 (0.064)

0.583 -0.511 -0.089 12,458 0.543

0.811 -0.891 -0.145 2,432 0.308

0.436 -0.488 -0.071 4,738 0.551

0.341 -0.300 -0.050 5,288 0.642

0.319 (0.051) -0.266 (0.051)

0.361 (0.089) -0.342 (0.131)

0.307 (0.088) -0.286 (0.085)

0.178 (0.090) -0.189 (0.071)

0.556 -0.463 -0.085 9,045 0.575

0.995 -0.943 -0.177 1,275 0.362

0.540 -0.503 -0.088 3,297 0.568

0.278 -0.296 -0.041 4,473 0.640

0.262 (0.060) -0.267 (0.083)

0.183 (0.075) -0.181 (0.122)

0.143 (0.099) -0.256 (0.125)

0.277 (0.137) -0.140 (0.156)

0.572 -0.585 -0.089 3,413 0.457

0.736 -0.725 -0.132 1,157 0.249

0.279 -0.499 -0.045 1,441 0.513

0.423 -0.214 0.057 815 0.655

Note: Models are estimated using a probit model, but marginal effects are shown, along with standard errors in parentheses. The market price of care is measured per hour worked. The alternate care elasticity is calculated based on the wage elasticity, predicted probability of paying for care, wage, and market price of care. See text for details. All models also include other family income, number of children under the age of six, number of children age six to twelve, age, education, marital status, disability status, urban status, and state maximum AFDC benefits.

450

I

Child Care and Mothers' Employment Decisions TABLE 10.9

/

Estimates of the Effect of the Market Price of Child Care and Wages on Labor Force Participation by Education and Poverty Status for All Women with Children Under the Age of Thirteen All

Nonpoor women In(hourly wage) In (market price of care) Elasticity Wage Care Alternate care n

Participation rate Poor and near-poor women In (hourly wage) In (market price of care) Elasticity Wage Care Alternate care n Participation rate

Less than High School

High School

More than High School

0.278 (0.033) -0.137 (0.037)

0.479 (0.094) 0.067 (0.148)

0.175 (0.056) -0.118 (0.061)

0.221 (0.050) -0.146 (0.049)

0.377 -0.186 -0.034 12,416 0.738

0.818 0.115 -0.075 972 0.586

0.237 -0.160 -0.021 4,604 0.738

0.291 -0.192 -0.027 6,840 0.759

0.117 (0.038) -0.153 (0.053)

0.118 (0.047) -0.181 (0.083)

0.031 (0.065) -0.094 (0.082)

0.026 (0.105) -0.112 (0.113)

0.287 -0.375 -0.032 8,171 0.407

0.422 -0.649 0.051 2,712 0.279

0.068 -0.205 -0.007 3,548 0.459

0.053 -0.227 -0.006 1,911 0.495

Note: Models are estimated using a probit model, but marginal effects are shown, along with standard errors in parentheses. The market price of care is measured per hour worked. The alternate care elasticity is calculated based on the wage elasticity, predicted probability of paying for care, wage, and market price of care. See text for details. All models also include other family income, number of children under the age of six, number of children age six to twelve, age, education, marital status, disability status, urban status, and state maximum AFDC benefits. Poor and near-poor women are defined as women with family income under 185 percent of the poverty line.

least skilled being most responsive to the market price of child care appears to hold even within this lower-income group. To put all the previous results into context, table 10.10 presents a simulation for selected groups of women of a fifty-cent drop in the per hour market price of child care. Note that this is equivalent to an annual subsidy of about $1,000 for a full-time, full-year worker using paid child care. Obviously, larger subsidies would have larger effects, but recall that the average low-skilled woman using paid care spends about $2,962 annually on child care. 27 Thus, this does imply a

I

451

Finding Jobs TABLE 10.10

/

Simulations of the Effect of a 50¢ Decrease in the Per Hour Market Price of Child Care for Selected Groups of Women, by Education (Percentage)

All

Less than High School

High School

More than High School

All women with children under thirteen years of age Initial predicted labor force participation After a 50¢ drop in per hour market cost After a 50¢ X (prob pay) increase in wage

0.608 0.687 0.618

0.360 0.433 0.372

0.617 0.698 0.626

0.702 0.766 0.710

Unmarried women with children under thirteen years of age Initial predicted labor force participation After a 50¢ drop in per hour market cost After a 50¢ X (prob pay) increase in wage

0.552 0.657 0.562

0.298 0.415 0.308

0.587 0.690 0.594

0.746 0.811 0.754

Poor and near-poor women with children under thirteen years of age Initial predicted labor force participation After a 50¢ drop in per hour market cost After a 50¢ X (prob pay) increase in wage

0.408 0.476 0.412

0.279 0.380 0.284

0.460 0.501 0.461

0.495 0.537 0.496

All women with children under the age of six Initial predicted labor force participation After a 50¢ drop in per hour market cost After a 50¢ X (prob pay) increase in wage

0.543 0.628 0.555

0.309 0.424 0.321

0.552 0.641 0.562

0.643 0.695 0.650

Unmarried women with children under the age of six Initial predicted labor force participation After a 50¢ drop in per hour market cost After a 50¢ X (prob pay) increase in wage

0.459 0.555 0.470

0.249 0.334 0.260

0.514 0.608 0.520

0.656 0.701 0.666

Note: Poor and near-poor women are defined as women with family income under 185 percent of the poverty line. Simulations are based on the models estimated in tables 10.7, 10.8, and 10.9. See text for details of the simulation procedure.

subsidy of at least one-third of the cost of child care for the least skilled women. 28 The table also presents the results of simulating an equivalent change in the wage. This change is calculated by multiplying the fifty-cent drop in the market price by the predicted probability of using paid care. Reflecting the smaller alternate care elasticities calculated from the wage elasticities, the simulated increase in participation from a fifty-cent drop in the price of care is larger than that from an equivalent increase in the wage. Looking first at all women with children under the age of thirteen, our estimated model implies that a fifty-cent drop in 452

I

Child Care and Mothers' Employment Decisions

the price of care increases the participation rate from 0.608 to 0.687, a jump of almost 8 full percentage points. By contrast, the equivalent increase in the wage results in just a 1 percentage point increase, to 0.618. 29 Looking across skill levels, the absolute increases in participation are fairly similar across the groups. Thus, the sharp decline in the elasticities with skill are mainly the result of the lower initial participation rates for the lower-skilled mothers. For the simulations based on the change in the price of care, this absolute increase is often relatively large for the least skilled group. Take, for example, the case of unmarried women with low skills (less than a high school education) and with children under six years of age, a likely welfare population. Table 10.10 indicates that this subsidy could possibly increase their participation by more than a third, from 0.249 to 0.334. This increase is as large as or larger than that typically found in "successful" welfare-to-work programs (U.S. Department of Labor 1995) and suggests that child care subsidies may be a good way to increase rates of labor market entry among welfare recipients. What do these results tell us about the wisdom of expanding child care subsidies as a means of assisting less skilled women enter the labor market in the wake of welfare reform? If higher subsidies have the potential to increase labor force participation as much as the best previously attempted welfare-to-work programs, then perhaps more money should flow in that direction. Several important caveats are required before such a strong conclusion can be drawn based on our results. 3D First, and perhaps most important, our simulations suggest that even with subsidized child care about two-thirds of these women remain out of the labor force-a percentage too large to meet the work participation requirements and time-limit provisions imposed by welfare reform. Child care constraints are apparently only one small part of the difficulty faced by these women in the labor market. In addition, one may prefer to base estimates of the predicted response to a child care subsidy on the broader measure of changes in the returns to work, the wage elasticity. This approach provides considerably less optimistic predictions, indicating that the labor force participation rate of these women would increase by only 1 percentage point to 0.260 from 0.249. Moreover, past attempts to increase labor force participation of welfare recipients through more traditional welfare-to-work strategies have never been combined with the size of the stick that welfare reform threatens. Gauging the impact of such policies based on earlier demonstrations may not be so wise. At best, one might reasonably conclude that child care subsidies should be used as part of a bundle of services offered to welfare recipients to get them into the labor market.

SUMMARY AND CONCLUSION As the employment levels of women with children have rapidly risen over the past several decades, the importance of child care as a labor market issue has emerged as well. In this study we have focused on this intersection of child care and mothers in the labor market, with an eye toward examining differences I

453

Finding Jobs

across skill groups. Most government expenditure programs aimed at providing child care assistance, most notably the Child Care and Development Fund block grant program, are geared at low-income families, many of whom will be made up of low-skilled workers. The main government tax program, the Child Care Tax Credit, is used most extensively by families from the middle to upper range of the income distribution, though, where the least skilled workers are less likely to be found. The least skilled workers who use child care are less likely to pay for this care, and they tend to pay less for it when they do pay. Nonetheless, as a percentage of income, this group pays more for child care, even when the youngest child is of school age. However, children of the least skilled mothers are about twice as likely as children of the most skilled to be cared for by a relative. Again, this is true across age groups, even for school-age children, who are less likely to be mainly at school during the mother's work hours. The youngest children are, in the main, less likely to be in family-based or organized day care, whereas preschoolers are less likely to be in nursery school or organized day care. It is the case, though, that when using relative care, the least skilled are more likely to pay for it. The past literature on the effect of child care costs on the employment of mothers has been mixed, a result we attribute mainly to differences in the choices made in identifying the price of care effect in the final participation probit. These past estimates of the elasticity of participation with respect to the price of child care range from about 0 to almost -1 and provide no information on differences across skill levels. Similarly, most demonstration projects with a child care component are aimed at low-skilled mothers, and the results are difficult to interpret with respect to child care costs, although they appear to imply small effects. Our econometric results narrow the likely range of elasticities for all women with children under the age of thirteen to somewhere between - 0.055 to - 0.358. Additionally, we consistently find that this elasticity is larger for less skilled workers and declines with skill. One must bear in mind, however, that there are inherent weaknesses to all of the econometric methods. A well-designed and well-executed child care demonstration project, or a broader project carefully designed to allow the effect of the child care component to be separately identified, would likely add significantly to our knowledge of this issue. Until such a study has been done, however, we are left with our results indicating that the effect of child care costs on the labor supply of women as a whole is modest. For some groups, though, most notably less skilled unmarried women with young children, who are likely to be welfare recipients, child care subsidies could possibly lead to large relative gains. Nevertheless, the labor force participation of this group of women would still remain far below that of other groups, indicating that significant additional obstacles to employment remain, and far below the levels prescribed by the recently enacted welfare reform legislation. Although we believe that reducing the cost of child care for these women will help meet the goals of welfare reform, it will be insufficient on its own, and other types of policies will be required as well.

454

I

Child Care and Mothers' Employment Decisions

Prepared for the Joint Center for Research on Poverty Conference on Labor Markets and Less Skilled Workers, November 5 and 6, 1998. We are grateful for the comments provided by Becky Blank, David Card, and Sandra Hofferth, along with participants of both the November conference and the March preconference. We also thank Doug Staiger for helpful discussions.

APPENDIX TABLE 10A.1

/

Coefficients of Reduced-Form Participation Equation and SelectionCorrected Wage Equation

Reduced-Form Participation Probit

Coefficient

Standard error

Number of children under six Number of children six to twelve Presence of children thirteen to eighteen Presence of another adult Presence of an unemployed adult Other family income Age Age squared Education Urban area South West Northeast Nonwhite State unemployment rate Total number of children Disability State AFDC benefits (family of three) Married 1990 1991 Intercept

-0.252 -0.123 0.069 0.195 -0.155 0.000 0.146 -0.002 0.062 -0.023 -0.002 -0.035 -0.111 0.051 -0.D48 -0.089 -0.685 0.000 -0.037 -0.079 -0.045 -2.298

(0.032) (0.031) (0.044) (0.031) (0.039) (0.000) (0.010) (0.000) (0.004) (0.022) (0.029) (0.033) (0.030) (0.025) (0.008) (0.028) (0.040) (0.000) (0.023) (0.025) (0.025) (0.179)

Selection-corrected wage equation coefficients Education Nonwhite Age Age squared

0.094 -0.025 0.131 -0.002

(0.002) (0.014) (0.008) (0.000) (Table continues on p. 456.)

/

455

Finding Jobs

TABLE 10A.1

/ Continued

Total number of children Urban area South West Northeast State unemployment rate Disability Married 1990 1991 Lambda Intercept

Source: Authors' calculations.

456

I

-0.103 0.128 -0.026 0.097 0.079 -0.008 -0.326 0.022 -0.041 -0.008 0.435 -1.760

(0.007) (0.012) (0.013) (0.016) (0.016) (0.005) (0.034) (0.012) (0.014) (0.014) (0.042) (0.152)

/

Source: Authors' calculations.

(0.042) (0.040) (0.056) (0.039) (0.050) (0.000) (0.013) (0.000) (0.005) (0.028) (0.036) (0.042) (0.039) (0.031) (0.010) (0.037) (0.053) (0.029) (0.032) (0.032) (0.000) (0.229) 0.505

-0.061 -0.118 0.005 0.109 -0.046 0.000 0.158 -0.002 0.014 0.053 0.044 -0.020 0.060 0.019 -0.024 -0.078 - 0.534 0.082 -0.013 -0.085 0.000 -1.931

Participation Standard Coefficient Error

(0.047) (0.064) (0.040) (0.053) (0.000) (0.014) (0.000) (0.005) (0.028) (0.030) (0.036) (0.037) (0.031) (0.010) (0.044) (0.060) (0.029) (0.031) (0.032) (0.245)

-0.311 -0.060 -0.303 0.000 0.107 -0.002 0.052 0.065 0.053 0.045 -0.136 -0.100 -0.036 -0.124 -0.224 -0.161 0.040 -0.030 -2.275 (0.010)

(0.048)

-0.114

0.446

Use of Paid Care Standard Coefficient Error Number of children under six Number of children six to twelve Other family income Age Education Urban area South West Northeast Presence of children thirteen to eighteen Married Presence of another adult Presence of unemployed adult 1990 1991 Lambda from participation Lambda from use of paid care Intercept

Price of Care Equation

(0.025) (0.000) (0.003) (0.007) (0.025) (0.026) (0.031) (0.042) (0.070) (0.038) (0.041) (0.068) (0.025) (0.028) (0.218) (0.206) (0.191)

0.031 0.136 -0.006 -0.061 -0.168 -0.123 0.232 -0.326 -0.693

(0.058)

Standard Error

0.002 0.000 0.004 0.034 0.257 -0.044 0.126 0.191

0.320

Coefficient

Coefficients of Reduced-Form Bivariate Participation and Use of Paid Care Probit and Selection-Corrected Price of Care Equation

Number of children under six Number of children six to twelve Presence of children thirteen to eighteen Presence of another adult Presence of unemployed adult Other family income Age Age 2 Education Urban area South West Northeast Nonwhite State unemployment rate Total number of children Disability Married 1990 1991 State AFDC benefits Intercept Rho

Bivariate Probit

TABLE 10A.2

Finding Jobs

NOTES 1. See Sandra Hofferth (1991) and Council of Economic Advisers (1997) for a broad discussion of important issues in the analysis of child care markets. For the most part, research has addressed quality separately from the impact of child care costs on the quantity of child care purchased, although David Blau and Alison Hagy (1998) is an important exception. 2. This discussion is based mainly on information available in The Green Book (U.s. House of Representatives 1996, 1998). 3. The source of these data is U.S. Department of Treasury (1997). William Gentry and Hagy (1996) present similar information for 1989. 4. The CCTC figure is from 1996, but expenditures have been relatively constant over time. 5. As is done for Census Bureau publications, weights in the 1992 and 1993 panels are scaled so that the pooled data properly reflect the fall 1993 population. 6. Hofferth (1996) provides a summary of the overall utilization and cost of child care from 1965 to 1993. 7. We split the sample into three groups: those with less than a high school education, high school graduates, and those with more than a high school diploma. The overall numbers of women in these last two groups are about equal and provide reasonable sample sizes for the econometric analysis that follows. To maintain consistency, we do not separate out college graduates for the descriptive analysis, even though differences in labor force participation lead to quite different sample sizes here. 8. One may think that the share of the mother's earnings, rather than family income, that goes to child care is the appropriate measure because it better reflects the extent to which child care costs reduce the returns to work. However, family well-being is better reflected by the income share measure. For this reason, and for the purposes of consistency with previous work, we present costs as a percentage of income. Patterns across groups of costs as a share of earnings are qualitatively similar to those reported here. 9. This is calculated as the percentage paying for care multiplied by the percentage of income paid. 10. Because the majority of children age six to twelve are in school, we present the breakdown of parents' marital status only for children under the age of six. 11. Because parent care is defined by the survey to be unpaid, and because few children use in-home care or other arrangements, these categories are not reported. 12. See chapter 2 in this volume for a review of estimates of the number of women projected to enter the labor market in response to welfare reform. Between August 1996, when welfare reform legislation was enacted, and December 1998, the number of families on welfare declined from 4.4 million to 2.8 million (Department of Health and Human Services, http://www.acf.dhhs.gov/news/stats/aug-sep.htm. July 27, 1999). 13. These estimates are likely to be understated because they assume unlimited availability of the types of care currently used by unmarried, low-skilled mothers at the current market price. As more mothers demand these services, their cost may rise.

458

Child Care and Mothers' Employment Decisions 14. To our knowledge, the only large-scale randomized trial in which the treatment consisted solely of changes in child care costs was conducted in conjunction with the Gary (Indiana) Income Maintenance Experiment in the 1970s. The results of this study were largely inconclusive because of small sample sizes and very low take-up rates of the subsidy among treatment group members (Behrens 1978). Regardless of the results, it is difficult to assess the relevance of an experiment regarding female labor force participation from so long ago, when so much smaller a share of the female population worked. The only other large-scale randomized trial that has ever been planned was the Expanded Child Care Options (ECCO) Demonstration in New Jersey. It was on the drawing board for several years (see Gueron and Pauly 1991) but unfortunately was never conducted. 15. An important caveat before drawing such a conclusion is that it is difficult to ascertain the value of the child care component of these treatments. Although members of the control group were not eligible for the special child care services offered to members of the treatment group, they were still eligible for any ongoing programs through which they would normally be eligible to receive child care assistance. Therefore, if no treatment effect is observed, one may alternatively conclude that the benefits provided to mothers in the treatment group were not significantly greater than those offered to members of the control group. 16. Of course, if the observed increase in participation could be attributed solely to the child care treatment, the effect of subsidized care would be considered relatively large. 17. The importance of these decisions is underscored by Kimmel (1998), who concludes that most of the difference between the highest and lowest elasticity estimates using this methodology can be attributed to the specific explanatory variables included in the model. In performing this sensitivity analysis to determine the factors that underlie the difference in estimated elasticities reported by Rachel Connelly (1992) and David Ribar (1992), she dismisses the importance of the type of probit (bivariate or univariate) used for the selection correction and the metric used for the child care cost measure. 18. Gordon Cleveland, Morley Gunderson, and Douglas Hyatt (1996) find a similar elasticity of - 0.39 for married mothers of preschoolers in Canada. 19. This requires dropping the small states that are not separately identified in the SIPP. 20. A typical identification strategy is to rely on state child care regulations. We have estimated models comparable to those reported here that include state fixed effects (rather than regional effects), which would subsume any differences across states in the types of child care regulations that have been used to identify other models. The results from these model specifications are very similar to those reported here. 21. The results from this prediction stage are shown in the appendix tables. 22. Preliminary estimates indicated that the stability of the model suffers with smaller sample sizes when the predictions are done separately for each subgroup. For samples with about five thousand observations or more, however, the results are similar to those presented here. 23. We feel fairly confident that our wage elasticity estimates are reasonable. Ribar (1992) reports a wage elasticity of between 0.58 and 0.68 for the married women in his child

459

Finding Jobs care study. Studies of the Earned Income Tax Credit (EITC) find elasticities of between 0.35 and 0.70 for single parents (Dickert, Houser, and Scholz 1995) and of about 0.29 for married women with less than a high school education (Eissa and Haynes 1998). Also, there is evidence that participation is the main margin of adjustment for labor supply, implying that the labor force participation elasticity may be close to the overall labor supply elasticity; see Mroz (1987), for example. 24. Given the predicted probabilities of using paid care, the upper bound estimates are generally two to three times the size of the alternative care elasticities presented here. 25. Prima facie evidence for this proposition is found by reestimating our model, including regional controls in the probit model for all women with children under the age of thirteen. The weak identification leads to a fairly imprecise coefficient on the price of child care of - 0.063, with a standard error of 0.046. This much lower point estimate implies a care elasticity of - 0.104. The wage coefficient is essentially unaffected, implying an alternate care elasticity of - 0.058. 26. Separating out the poor from the near-poor leads to relatively small sample sizes by skill group and to very imprecise estimates. Even within these broader groups, sample sizes of less educated nonpoor and more educated poor and near poor are quite small, and regression coefficients for these groups are estimated with a great deal of imprecision. This most likely explains the opposite sign on the point estimate of the care elasticity from what one would expect for nonpoor women with less than a high school diploma. 27. This figure is obtained from table 10.1 as the per week cost for those paying $56.96 a week multiplied by 52 weeks a year. 28. Simulating a subsidy that is much larger than this would imply using our model to estimate far out of sample and would likely be unwise. 29. Assuming that the probability of using paid care is 1 (that is, simulating a fifty-cent increase in the wage) implies a change in participation between the two values presented but a figure typically still well below that for a fifty-cent decrease in the market price of care. For example, for all women with children under the age of thirteen, a 3 percentage point increase is implied. 30. One issue that often arises in the analysis of government subsidy programs is the extent to which funding goes to those who would have undertaken the intended behavior anyway. Although this concern emerges here as well, it is perhaps not quite as important as it is in other contexts. For instance, although it is true that a mother who would have worked anyway may receive a child care subsidy, she may use these additional funds to purchase higher-quality child care that is more expensive. Therefore, estimating the magnitude of the inefficiency is more complicated in the specific case of a child care subsidy.

REFERENCES Averett, Susan L., H. Elizabeth Peters, and Donald M. Waldman. 1997. "Tax Credits, Labor Supply, and Child Care." Review of Economics and Statistics 79(February): 125-35. Behrens, Jean. 1978. The Demand for the Gary Subsidized Child Care Program. Princeton, N.J.: Mathematica Policy Research.

460

Child Care and Mothers' Employment Decisions Berger, Mark C, and Dan A. Black. 1992. "Child Care Subsidies, Quality of Care, and the Labor Supply of Low-Income, Single Mothers." Review of Economics and Statistics 74(November): 635-42. Blau, David M., and Alison P. Hagy. 1998. "The Demand for Quality in Child Care." Journal of Political Economy 106(1): 104-47. Blau, David M., and Phillip K. Robins. 1988. "Child Care Costs and Family Labor Supply." Review of Economics and Statistics 70(August): 374-81. Cleveland, Gordon, Morley Gunderson, and Douglas Hyatt. 1996. "Child Care Costs and the Employment of Women: Canadian Evidence." Canadian Journal of Economics 29(February): 132-51. Connelly, Rachel. 1992. "The Effect of Child Care Costs on Married Women's Labor Force Participation." Review of Economics and Statistics 74(February): 83-90. Council of Economic Advisers. 1997. "The Economics of Child Care." Unpublished paper. Washington, D.C: Council of Economic Advisers. Dickert, Stacy, Scott Houser, and John Karl Scholz. 1995. "The Earned Income Tax Credit and Transfer Programs: A Study of Labor Market and Program Participation." Tax Policy and the Economy 9: 1-50. Eissa, Nada, and Hilary Williamson Hoynes. 1998. "The Earned Income Tax Credit and the Labor Supply of Married Couples." Working paper 4. University of California, Berkeley: Center for Labor Economics. Gelbach, Jonah. 1997. "How Large an Effect Do Child Care Costs Have on Single Mothers' Labor Supply? Evidence Using Access to Free Public Schooling." Unpublished paper. Cambridge, Mass.: MIT. Gentry, William M., and Alison P. Hagy. 1996. "The Distributional Effects of the Tax Treatment of Child Care Expenses." In Empirical Foundations of Household Taxation, edited by Martin Feldstein and James M. Poterba. Chicago: University of Chicago Press. Gueron, Judith M., and Edward Pauly. 1991. From Welfare to Work. New York: Russell Sage Foundation. Han, Wen-Jui, and Jane Waldfogel. 1998. Child Care and Women's Employment. Unpublished manuscript. Columbia University School of Social Work. Hausman, Jerry A. 1981. "Labor Supply." In How Taxes Effect Economic Behavior, edited by Henry J. Aaron and Joseph A. Pechman. Washington, D.C: Brookings. Hofferth, Sandra L. 1991. "Comments on 'The Importance of Child Care Costs to Women's Decision Making.''' In The Economics of Child Care, edited by David M. Blau. New York: Russell Sage Foundation. - - - . 1996. "Child Care in the United States Today." The Future of Children 6(SummerFall): 41-61. Kimmel, Jean. 1995. "The Effectiveness of Child Care Subsidies in Encouraging the Welfare-to-Work Transition of Low-Income Single Mothers." American Economic Review 85(May): 271-75. - - - . 1998. "Child Care Costs as a Barrier to Employment for Single and Married Mothers." Review of Economics and Statistics 80(2): 287-99. Long, Sharon K, Gretchen G. Kirby, Robin Kurka, and Shelley Waters. 1998. Child Care Assistance Under Welfare Reform: Early Responses by the States. Washington, D.C: Urban Institute. Michalopoulos, Charles, Philip K. Robins, and Irwin Garfinkel. 1992. "A Structural Model of Labor Supply and Child Care Demand." Journal of Human Resources 27: 166-203. Mroz, Thomas A. 1987. "The Sensitivity of an Empirical Model of Married Women's Hours of Work to Economic and Statistical Assumptions." Econometrica 550uly): 765-99.

I

461

Finding Jobs

Ribar, David. 1992. "Child Care and the Labor Supply of Married Women: Reduced Form Evidence." Journal of Human Resources 27(Winter): 124-65. - - . 1995. "A Structural Model of Child Care and the Labor Supply of Married Women." Journal of Labor Economics 13: 558-97. Stoney, Louise, and Mark H. Greenberg. 1996. "The Financing of Child Care: Current and Emerging Trends." The Future of Children 6(Summer-Fall): 83-102. U.s. Department of Labor. Office of the Chief Economist. 1995. What's Working (and What's Not); A Summary of Research on the Economic Impacts of Employment and Training Programs. Washington: US. Government Printing Office. U.s. Department of the Treasury. Internal Revenue Service, Statistics of Income Division. 1997. Individual Tax Returns, 1994. Washington: US. Government Printing Office. US General Accounting Office. 1994. "Child Care Subsidies Increase Likelihood That Low-Income Mothers Will Work" Unpublished paper. Washington: US. General Accounting Office. US. House of Representatives. The Green Book. 1996, 1998. Washington: US. Government Printing Office.

462

Chapter 11 Use of Means-Tested Transfer Programs by Immigrants, Their Children, and Their Children's Children Kristin F. Butcher and Luojia Hu

T

he past two decades have brought more immigrants to the United States than any comparable period since the" great" migration at the turn of the last century. Researchers and policy makers have always been concerned about the impact of immigration on a host of socioeconomic outcomes. However, since the changes in immigration law in 1965, the predominant sending regions shifted from European countries to countries in Latin America and Asia. These "new" immigrants tend to have lower levels of education, on average, than the native born in the United States. This shift, combined with declines in real wages for low-skilled workers in the United States in general, has increased concerns about immigrants' effect on the public coffers through their use of public transfer programs, participation in bilingual and other special programs in public schools, and criminal activity. In addition, policy makers express concern that generous welfare programs may act as a "magnet," perhaps attracting lowskilled immigrants to the United States. 1 Public concern over immigrants' use of welfare culminated in the Personal Responsibility and W~rk Opportunity Reconciliation Act (PRWORA) of 1996, popularly known as the Welfare Reform Act. Welfare reform radically changed the welfare system in the United States. Its impact on low-skilled U.S. citizens is the subject of intense debate and study. Although the changes brought by this law affect all welfare recipients, noncitizens were expressly singled out. The Welfare Reform Act affects the panoply of government-sponsored support more profoundly for noncitizens than for any other group. The changes in the federal law have resulted in much greater variation across states in immigrants' access to social programs. Although it is still too early to assess the impact of these changes on the immigrant population, we can examine the extent to which the immigrant population relied on the welfare system at the time the law was passed. Support for welfare reform as it pertains to immigrants came out of the perception that the United States currently attracts immigrants who are particularly prone to reliance on welfare, and that by changing these rules we could both save public funds and attract immigrants more likely to be successful in the U.S. labor market. One concern is that if immigrants rely

/

465

Finding Jobs

on welfare, they may raise their children to rely on welfare. Current immigration may have long-term implications for the welfare state. Indeed, the most important impact of immigration on the United States in many dimensions is likely to be long term. Currently, immigrants constitute about 10 percent of the population; those with at least one foreign-born parent also make up about 10 percent of the population, and those with at least one foreign-born grandparent, 40 percent of the population. The larger the immigrant flow, the larger the legacy it will leave. With that in mind, it is worth investigating what intergenerational welfare use looked like for immigrants under the prereform system. Was it the case, before welfare reform, that the first generation's welfare use escalated across generations, or does the second generation assimilate to the level of welfare use of the native born? To investigate these questions, throughout our data analysis we present information separately for immigrants (the foreign born, or the first generation), children of immigrants (the second generation), and those with native-born parents (the native born, or the third-and higher-generation).2 In addition, we examine the intergenerational correlation in welfare receipt, using data on immigrants from the 1970 census and data on the second generation in the Current Population Survey (CPS) from 1994 to 1996. To our knowledge, this is the first attempt to measure the intergenerational correlation in welfare use for immigrants in the United States. In addition to providing information on changes in welfare rules pertaining to immigrants, the contribution of this research to the literature on immigrants and welfare is threefold. First, we use the most recent data available before the large overhaul in the welfare rules, and those data have rich information on a wide range of transfer programs. Therefore, this work should help paint a clear picture of immigrants' reliance on welfare before welfare reform. Second, the data allow us to examine several groups of interest, in particular, immigrants, the second generation, and the third (and higher) generation. Finally, combining the CPS data with 1970 Census Bureau data allows us to calculate intergenerational correlation in welfare use between immigrants and their children. Thus, this chapter should shed some light on the possible effects, both short term and long term, of welfare reform. Our results show that, on average, immigrants are more likely to use transfer programs than are the native born. The differences are larger for in-kind transfer programs like food stamps and Medicaid coverage than for cash transfer programs. However, we also find that immigrants are less likely to participate in these programs than are the native born with similar observable characteristics. The second generation is always less likely than the third generation, both on average and controlling for characteristics, to use transfer programs. Finally, we find that the first generation's receipt of welfare has a positive and significant effect on the second generation's receipt. However, the coefficients are generally less than one-indicating that differences between the immigrant generation and the native born tend to die out over time. In addition, in most cases there is no significant effect of the first generation's participation in cash transfer programs on the second generation when the education level of the second genera-

466

I

Use of Means-Tested Transfer Programs

tion is held constant. Although this is far from a perfect test of welfare dependency, our results suggest that the transmission of welfare use between immigrants and their children works through skills; there is little evidence that welfare use by the parents' generation has an added effect over and above the fact that if one's parents are poor, one is also likely to be poor.

CHANGES IN ELIGIBILITY AFTER WELFARE REFORM Before August 1996, when the Personal Responsibility and Work Opportunity Reconciliation Act was signed into law/ the same welfare eligibility standards were applied to u.s. citizens and legal immigrant noncitizens: The new standards are very different for citizens and noncitizens and are extremely complex. The description here of the changes in eligibility standards draws heavily on the work of Wendy Zimmermann and Karen Tumlin (1999), which provides a thorough description of the changes in federal laws and a detailed analysis of states' responses to the federal changes. 5 The Welfare Reform Act has added to the complexity of determining immigrants' eligibility by creating several different categories of immigrants. Besides distinguishing between citizens and noncitizens, noncitizens are now further categorized according to their date of arrival. "Current" or "preenactment" immigrants are those who arrived before the signing of the welfare reform bill. "New" or "postenactment" immigrants are those who arrived after the bill was signed into law." In addition, immigrants are classified as "qualified" or "unqualified." Qualified immigrants include lawful permanent residents, refugees and asylees, persons paroled into the United States for at least one year, and battered spouses or children whose visa status is pending. Unqualified immigrants include undocumented immigrants, asylum applicants, and those with temporary status, such as students and tourists. A brief description of eligibility is difficult, because the rules are complicated and changing. Eligibility for various federal and state programs differs depending on the category into which an immigrant falls. In addition, there have already been several modifications to the Welfare Reform Act's rules governing immigrants' eligibility. Initially, the changes in welfare rules denied both Supplemental Security Income (551) and food stamp benefits to most legal immigrants who were in the United States as of August 22, 1996, but there was great public outcry at the perceived injustice of denying SSI benefits, and these benefits were restored by the Balanced Budget Agreement of 1997; this reversal does not, however, extend to new legal immigrants, who are permanently disqualified from receiving 55!. The government continues to revisit these provisions of the PRWORA. On June 23, 1998, the president signed the Agriculture Research, Extension, and Education Reform Act (public law 105-185), which restores food stamps to certain aliens who were denied benefits under the PRWORA. The restoration of food stamp benefits was patterned after the restoration of SSI benefits; thus, new immigrants are still denied food stamp benefits.

I

467

Finding Jobs

In addition to changing the name of the main cash program from Aid to Families with Dependent Children (AFDC) to Temporary Assistance for Needy Families (TANF), welfare reform changed its nature. Responsibility for the program now mainly lies with the state. Even qualified noncitizens who were in the United States before the passage of the welfare reform bill are eligible for TANF only if each state decides to extend benefits to this group. Noncitizens who have worked in the United States for forty quarters or who are military personnel (or their families) are eligible for TANF. In addition, refugees and asylees are eligible for TANF for their first five years in the United States, and at the state's discretion after that. The Medicaid rules pertaining to immigrants who arrived before August 22, 1996, are similar to those for TANF. The treatment of refugees and asylees is slightly different: they are eligible for their first seven years in the United States and then at the state's discretion. Following the federal restorations, most qualified immigrants are now eligible for food stamps and SS!. However, the treatment of refugees is the same as under Medicaid: they are eligible for their first seven years, after which the states have the option of offering benefits. For immigrants arriving after the signing of the welfare reform bill, the situation is quite different. Even lawful permanent residents are barred from receiving SSI or food stamps, and they are barred from Medicaid and TANF receipt for their first five years in the United States. After five years these two programs are available at the state's discretion. Welfare reform has made the social safety net for immigrants vastly more complex, and it may have opened up several holes, although the extent to which immigrants are affected by these holes depends on the state. Table 11.1, which is adapted from Zimmermann and Tumlin, presents an overview of noncitizen eligibility for the four largest budget items among the myriad programs that make up the welfare state. As the table demonstrates, in most cases immigrants are eligible for state and local public benefits if states decide to cover them. Zimmermann and Tumlin provide invaluable information: the authors surveyed states to determine their responses to changes in the welfare rules pertaining to immigrants. As might be expected, states' responses range from simply mimicking the eligibility rules set up by the federal government to purposefully trying to fill the gaps in coverage for immigrants. For example, Washington state pioneered the idea of purchasing food stamps from the federal government to offer to newly ineligible immigrants (Zimmermann and Tumlin 1999). In many cases, the states that have given the most consideration to providing benefits to newly ineligible immigrants are those with sizable immigrant populations or the most generous existing system of benefits (and, sometimes, both). For example, California-the state with the largest immigrant population-has opted to provide food programs, TANF, and Medicaid to new immigrants during the fiveyear bar (Zimmermann and Tumlin 1999, table 5). New York state, on the other hand, has opted to allow participation in state food programs only for targeted groups of immigrants (for example, children). Although some states are trying to fill the gaps in coverage for some groups

468

I

/

Qualified immigrants

Food Stamps

Medicaid

Eligible

Eligible

Eligible

Eligible for first seven years

Eligible

Eligible

Eligible for first seven years

Eligible for first five years; afterward, at state's option

Eligible

Eligible

State Option

Ineligible

Barred for first five years; afterward, at state's option

Barred for first five years; afterward, at state's option

New Immigrants (Arriving After August 22, 1996)

Eligible for first seven years; afterward, at state's option

Eligible

State Option

Eligible2

Eligible'

Ineligible

Temporary Assistance to Needy Families

Current Immigrants (Arriving On or Before August 22, 1996)

Supplemental Security Income

Noncitizen Benefit Eligibility

Qualified immigrants Exempted groups With forty quarters of work in United States Military personnel and their families Refugees and asylees

TABLE 11.1

(Table continues on p. 470.)

At state's option

Eligible for first five years; afterward, at state's option

Eligible

Eligible

State Option

State and Local Benefits

Eligible for first five years; afterward, at state's option Ineligible'

Eligible for first five years; afterward, at state's option Ineligible

Eligible for first seven years; afterward, at state's option

Ineligible "

Eligible for emergency services only'

Ineligible'

Eligible for first seven years

5States may provide state and local public benefits to unqualified immigrants only if they pass a law after August 22, 1996.

4American Indians born in Canada and certain other tribal members born outside the United States are eligible. Hmong and Lao tribe members (and their spouses and children) are eligible for food stamps.

'Immigrants formerly considered Permanently Residing Under the Color of Law (PRUCOL) who were receiving S5I on August 22, 1996, are eligible for SSI and for Medicaid in states where Medicaid eligibility is linked to SS!.

'Qualified immigrants who were lawfully residing in the United States on August 22, 1996, and are: under eighteen years; disabled or blind; or older than sixty-five years are eligible. All other qualified immigrants are ineligible unless exempted.

'Qualified immigrants receiving 551 on August 22, 1996, are eligible. Qualified immigrants lawfully residing in the United States on August 22, 1996, who are or become disabled are also eligible. All other qualified immigrants are ineligible unless exempted.

Eligible

Eligible for first seven years

Eligible

Barred for first five years; afterward, eligible Eligible

Barred for first five years; afterward, at state's option Eligible

Barred for first five years; afterward, eligible Eligible

Barred for first five years; afterward, eligible Eligible

Source: Zimmermann and Tumlin 1999, figure 1, p. 15.

Unqualified immigrants

Exempted groups With forty quarters of work in United States Military personnel and their families Refugees and asylees

Use of Means-Tested Transfer Programs

of newly ineligible immigrants, the state rules add yet another layer of complexity to eligibility. States may offer various programs, each with different eligibility restrictions. Some of the more common restrictions include targeting only certain groups (children, for example), imposing sponsor" deeming," requiring naturalization procedures, and imposing durational residency requirements.' The lasting impression from Zimmermann and Tumlin's exhaustive investigation is that although welfare reform created an enormous amount of interstate variation in the social safety net for citizens, that variation is even greater for immigrants. Some states are working to fill these gaps for some groups of immigrants, but there still appear to be groups that are uncovered for certain needs. In particular, Zimmermann and Tumlin emphasize, no state program fills the needs for new immigrants that are met by the federal Medicaid program. The effect of these gaps on the affected population remains to be seen, but the Zimmermann and Tumlin paper points to outcomes-for example, health outcomes-that will be particularly important to track in the future. As mentioned earlier, one of the primary motivations for welfare reform as it applies to immigrants was the need to restrict costs. The complexity of the new laws makes any estimate of cost savings to the public very difficult. However, we can use our data (described below) to get a very rough estimate of cost savings at the federal level. Table 11.1 indicates that new immigrants, who are not refugees, arriving after August 22, 1996, will be ineligible for TANF, food stamps, Medicaid, and SSI. Using the most recent data available before welfare reform, we calculated the fraction of recipients of AFDC, SSI, food stamps, and Medicaid who are noncitizens, who arrived within the past five years. 8 If welfare reform had been implemented five years earlier, these individuals would have been ineligible, so transfer receipt among this group gives some indication of the short-term savings (assuming welfare reform does not alter the incentive to migrate). Immigrants in this category constitute the following fractions of transfer recipients: 3.7 percent for AFDC, 2.0 percent for SSI, 3.7 percent for food stamps, and 3.6 percent for Medicaid. Using 1994 federal expenditures (Department of Commerce 1997) for each of these programs yields approximate savings of $522 million for AFDC, $470 million for SSI, $947 million for food stamps, and $2.96 billion for Medicaid. Although these numbers are substantial, it is important to remember that they represent a tiny fraction of government expenditures. Furthermore, they are an overestimate because they do not take into account the fact that refugees in this group continue to receive benefits. More important, these numbers do not represent the value of savings to the public because, as described above, much of the responsibility for the social safety net now falls on states. Thus, much of the savings at the federal level may simply emerge as costs shifted to the public at the state level. States' costs will be affected by the fraction of the needy who are immigrants and by states' decisions to provide programs compensating for the lack of federal aid. The extent to which the immigrant population will be affected by these changes depends on how extensively they have relied on various programs and how the states in which they reside have reacted to the federal changes. An

I

471

Finding Jobs

examination of the ways in which immigrants used welfare benefits in the United States before passage of the Welfare Reform Act might allow some insight into the effect welfare reform will have on future immigrants.

IMMIGRANT PARTICIPATION IN WELFARE PROGRAMS We rely primarily on data from the March Current Population Survey (CPS) from 1994 to 1996. These data have several advantages and disadvantages. The survey covers participation in a wide range of welfare programs including AFDC, SSI, food stamps, Medicaid, subsidized school lunches, public housing, reduced rent, and energy assistance. 9 We focus on the first four of these programs because they are much larger items in the government budget and they are the items that have been most affected by welfare reform. lO In addition to covering a wide range of programs, these data are the most recent available before the advent of welfare reform. Therefore, they should help inform policy makers as to where the impacts of welfare reform will be most keenly felt. One drawback to the CPS data for our purpose is that there appears to be an undercount of welfare recipients compared with administrative records. Although there has always been something of a census undercount, it has increased in recent years. From 1987 to 1993, the ratio of AFDC recipients reported in the CPS to the number found in administrative data hovered around 80 percent. From 1994 onward, this percentage has fallen, and from 1994 to 1996-the period covered by our data-the CPS seems to have found only about 70 percent of the cases (Bavier 1999). At present, not much is known about this decline in coverage in the CPS. For our purposes, the decline in coverage affects our ability to use CPS data to make statements about the total number of immigrants or the native born who receive welfare. However, assuming the undercount is not concentrated among the immigrant population, we may still use the CPS data to draw inferences about the relative use of welfare programs by immigrants and the native born. We currently have no evidence as to whether the undercount is concentrated among some groups; however, our results comparing immigrant and native-born participation in welfare programs are very similar to results found by George Borjas and Lynette Hilton (1996), who rely on the Survey of Income and Program Participation (SIPP), which suffers less from an undercount problem. With these caveats in mind, we can use these data to examine immigrants' participation in welfare programs.

IMMIGRANTS AS A PERCENTAGE OF STATE CASELOADS Table 11.2 presents demographic information for immigrants and several groups of the native born: those with one parent born abroad, those with two parents born abroad, and those with native born parents. ll All data come from the March CPS from 1994 to 1996, and only individuals age eighteen and older are in-

472

I

Use of Means-Tested Transfer Programs TABLE 11.2

I

Individual Characteristics by Nativity Groups Age (Standard Deviations)

Foreign Born

Both Parents Foreign Born

One Parent Foreign Born

U.s. Born and U.S.-Born Parents

Other

n Percentage of sample

37,592 10.88

15,122 4.48

14,762 4.43

47,194 78.70

5,737 1.51

Percentage of high school dropouts

20.82

5.94

3.73

67.31

2.20

9.51

1.84

3.21

82.89

2.54

0.512 (0.500) 42.53 (16.94) 0.366 (0.482) 0.243 (0.429) 0.184 (0.387) 0.207 (0.405) 0.074 (0.262) 0.443 (0.497) 0.207 (0.405) 0.034 (0.180) 0.624 (0.484) 0.655 (0.475)

0.539 (0.499) 56.97 (21.67) 0.254 (0.435) 0.323 (0.468) 0.240 (0.427) 0.183 (0.387) 0.020 (0.141) 0.192 (0.394) 0.054 (0.226) 0.013 (0.114) 0.527 (0.499) 0.452 (0.498)

0.514 (0.500) 49.32 (19.06) 0.161 (0.368) 0.320 (0.466) 0.275 (0.447) 0.244 (0.429) 0.026 (0.160) 0.133 (0.339) 0.023 (0.149) 0.014 (0.118) 0.583 (0.493) 0.645 (0.479)

0.522 (0.500) 43.45 (16.91) 0.164 (0.370) 0.349 (0.477) 0.277 (0.447) 0.211 (0.408) 0.132 (0.338) 0.030 (0.170) 0.003 (0.055) 0.009 (0.094) 0.579 (0.494) 0.729 (0.445)

0.524 (0.500) 41.11 (16.11) 0.279 (0.449) 0.274 (0.446) 0.256 (0.437) 0.190 (0.393) 0.069 (0.254) 0.468 (0.499) 0.065 (0.246) 0.041 (0.198) 0.518 (0.500) 0.684 (0.465)

Percentage of single female heads with children in household Personal characteristics Female Age High school dropout High school degree Some college College graduate African American Hispanic Asian Other race Married Worked last year

2

(Table continues on p. 474.)

I

473

Finding Jobs TABLE 11.2

/

Continued

U.s. Born and U.S.-Born Parents

Both Parents Foreign Born

One Parent Foreign Born

2.370 (0.818)

2.399 (0.804)

2.318 (0.773)

2.248 (0.746)

2.604 (1.539) 0.099 (0.299) 0.261 (0.439)

2.768 (1.401) 0.120 (0.325) 0.230 (0.421)

2.969 (1.456) 0.138 (0.345) 0.221 (0.415)

3.208 (1.592) 0.182 (0.386) 0.275 (0.446)

0.085 (0.177)

0.134 (0.213)

0.182 (0.234)

0.220 (0.248)

0.142 (0.256)

0.167 (0.273)

0.196 (0.290)

0.231 (0.304)

0.197 (0.282)

0.316 (0.333)

0.382 (0.336)

0.371 (0.316)

0.131 (0.276)

0.145 (0.297)

0.103 (0.256)

0.081 (0.222)

0.355 (0.411) 0.240 (0.331) 0.195 (0.300)

0.253 (0.373) 0.330 (0.380) 0.229 (0.331)

0.163 (0.308) 0.324 (0.371) 0.273 (0.348)

0.166 (0.311) 0.348 (0.381) 0.275 (0.353)

0.272 (0.383) 0.278 (0.355) 0.256 (0.344)

Fraction receiving various transfer programs 55I 0.032 (0.175) 0.034 AFDC (and other cash) (0.182) 0.112 Medicaid (0.316)

0.022 (0.146) 0.010 (0.100) 0.064 (0.245)

0.017 (0.131) 0.019 (0.136) 0.059 (0.236)

0.023 (0.151) 0.027 (0.163) 0.075 (0.264)

0.051 (0.220) 0.074 (0.261) 0.157 (0.364)

Foreign Born Log average hourly earnings

2.187 (0.773)

Household characteristics Number of 3.772 (1.987) people 0.217 More than one family (0.412) Female0.183 headed fam(0.387) i1y Fraction age 0.226 less than (0.238) eighteen Fraction age 0.240 eighteen to (0.293) thirty Fraction age 0.339 thirty-one to (0.291 ) fifty-five Fraction age 0.083 fifty-six to (0.212) sixty-five Education distribution for adults High school dropout High school degree Some college

474

I

Other

Use of Means-Tested Transfer Programs TABLE 11.2

/ Continued

Foreign Born Food stamps' Reduced rent' Public housing! Subsidized schoollunch1 Energy assistance l

0.115 (0.319) 0.017 (0.129) 0.029 (0.167) 0.184 (0.387) 0.023 (0.150)

Both Parents Foreign Born

One Parent Foreign Born

0.047 (0.212) 0.009 (0.097) 0.024 (0.152) 0.047 (0.213) 0.022 (0.146)

0.056 (0.229) 0.010 (0.099) 0.019 (0.137) 0.053 (0.223) 0.023 (0.150)

U.s. Born and U.s.-Born Parents 0.079 (0.270) 0.012 (0.108) 0.024 (0.152) 0.079 (0.270) 0.030 (0.171)

Other 0.164 (0.370) 0.032 (0.175) 0.063 (0.244) 0.152 (0.359) 0.054 (0.227)

Source: March Current Population Survey, 1994 to 1996. Note: These data are for individuals over eighteen years of age. The sample is weighted by

the March Supplement Weight. Standard deviations are in parentheses. The variable is equal to one if anyone in the individual's household participates in the program. ISomeone in the individual's household receives the transfer. 'Twenty percent random sample of the third generation.

cluded. 12 Immigrants are disproportionately represented among the low-skilled population, making up about 10.9 percent of the population but about 20.8 percent of those who did not finish high school. On the other hand, although immigrants are more likely to live in larger families and are more likely to have children present in the household, they are less likely to be in female-headed families and are more likely to be married than are the native born. They make up only 9.5 percent of single female heads of household with children under the age of eighteen in the household.!3 In addition, as has been documented elsewhere, immigrants are much more likely to be Hispanic or Asian than are the native born, they are less likely to have worked in the previous year than those with U.s.-born parents, and they have lower log average hourly earnings than any of the native-born groups." Based on demographic characteristics (other than marital status), one would expect immigrants to be more represented among welfare recipients than they are among the population in general. In addition, immigrants tend to be concentrated in only a few states in the United States, and their characteristics vary considerably by state. Therefore, immigrant concentrations among the needy populations are likely to vary greatly across states. Figures 11.1 and 11.2 map the concentration of immigrants in AFDC and food stamp caseloads across states. The numbers are the percentage of all individuals reporting AFDC and food stamp receipt who are immigrants, weighted by the

I

475

Finding Jobs FIGURE 11.1

/

Percentage of AFDC Recipients Who Are Immigrants

D Medium (1 percent to 10 percent) •

High (greater than 10 percent)

Source: March CPS, 1994 to 1996. Note: All numbers are weighted by months of receipt.

number of months of receipt. ' s We divide the states into three broad categories according to immigrant concentration. In the high concentration states, immigrants make up more than 10 percent of the welfare caseload, in the low concentration states, less than 1 percent. As is clear from the figures, there is great variation in the portion of caseloads composed of immigrants across states. When we repeat this exercise to show which states have caseloads heavily composed of noncitizen immigrants, we obtain similar results. Assuming that new immigrants' choice of state of residence follows earlier immigrants' patterns, the states with the largest current immigrant caseloads will be most affected by welfare reform.'6 If states choose to mirror the federal eligibility rules, then presumably this will result in a substantial decline in the "new immigrant" caseload and attendant cost savings. On the other hand, it is also the case that a substantial portion of their needy populations will go unserved. In reality, the states with heavy immigrant caseloads-California and Massachusetts, in particular-have tried to offset changes in the federal safety net with state programs (Zimmermann and Tumlin 1999). For these states, welfare reform is likely to result in a substantial increase in expenditures on welfare programs for immigrants.

476

I

Use of Means-Tested Transfer Programs FIGURE 11.2

/

Percentage of Food Stamps Recipients Who Are Immigrants

D Medium (1 percent to 10 percent) •

High (greater than 10 percent)

Source: March CPS, 1994 to 1996. Note: All numbers are weighted by months of receipt.

COMPARISONS BETWEEN IMMIGRANTS AND THE NATIVE BORN As discussed earlier, part of the motivation for welfare reform as it pertains to immigrants was to change both the incentives for immigration and the characteristics of the immigrant flow toward those who are less likely to use welfare. Although it is too soon to begin to evaluate whether those goals have been met, we can look back and examine welfare use among immigrants before welfare reform.

Previous Literature Because of data limitations, much of the previous work on immigrants' use of welfare programs in the United States has focused on cash transfer programs.

I

477

Finding Jobs

Francine Blau (1984) uses the 1976 Survey of Income and Education to analyze family receipt of welfare and social insurance.!7 Families were defined as natives or immigrants based on the status of the head of the family. Blau finds immigrants were more likely to use these programs than the native born but much less likely than the native born who had similar characteristics. This pattern of findings has become quite common in the literature. George Borjas and Stephen Trejo (1991) and Borjas (1995) have demonstrated, using several years of Census Bureau data, that immigrants were more likely to receive cash transfers than were the native born in the raw data. When controls for education of the head of household were entered, however, and especially when controls for race and ethnicity were entered, immigrants were significantly less likely to receive cash transfers than the native born. Although immigrants were more likely than natives to use cash public assistance programs on average, the differences tended to be fairly small. There are many other, larger budget, transfer programs in the social safety net in the United States. If the findings of immigrants' use of these programs differs from the findings of their use of cash programs, then the numbers based on cash transfers may understate or overstate immigrants' effect on the overall cost of the welfare system. Works by Borjas and Hilton (1996) and Janet Currie (1997) represent two efforts to fill this gap in our knowledge. Currie reports that about 18 percent of immigrant children were covered by Medicaid, compared with 14 percent of the native born. Currie uses variation across states in the Medicaid rules to instrument for individual eligibility. Similar to the work cited above, in which immigrants were found to be less likely to use welfare than natives with similar characteristics, the take-up rates conditional on eligibility were lower among immigrants, with about 50 percent of the eligible being covered among the immigrants and about 66 percent among the native born. However, she found that the utilization of routine care increased by the same amount among immigrants and natives as eligibility increased, but the utilization of expensive hospital care rose only for natives. In a comprehensive work that uses the Survey of Income and Program Participation (SIPP) to investigate immigrants' use of the whole range of assistance programs, Borjas and Hilton (1996) analyze both cash programs like AFDC, SSI, and General Assistance and noncash programs like Medicaid, food stamps, the Supplemental Food Program for Women, Infants, and Children (WIC), Low Income Energy Assistance, Housing Assistance, and School Lunch and Breakfast Programs. They divide the data into immigrant and native-born households based on the nativity of the household reference person. When they limit their analysis to receipt of cash programs in 1990 and 1991, they find that 12.3 percent and 18.2 percent of the native-born and immigrant households, respectively, participated in these programs. However, when they expand the list of programs to include all cash and noncash benefits, these percentages rise to 16.3 percent and 26.1 percent for native-born and immigrant households, respectively. Thus, the differential in immigrants' use of all programs-cash and noncash alike-was larger than the differential in their use of cash programs alone.

478

/

Use of Means-Tested Transfer Programs

Much work on other aspects of immigration relies on Census Bureau data, partly because of the large samples of individual-level data. Although immigration has increased over the past two decades, immigrants in 1990 still represented only about 10 percent of the population. Thus, very large data sets are often necessary to get large enough subsamples of immigrants and to break those immigrant groups into subgroups of interest. The Borjas and Hilton paper (1996) is the first attempt to fully document immigrants' participation in the welfare state; the data set it uses, the SIPP, is specifically designed to thoroughly investigate households' participation in transfer programs. However, their sample includes only 2,449 immigrant households. Our chapter adds to the literature by giving a second, independent measure of immigrants' use of many of the same welfare programs and by using the most recent data available from the CPS (combining data from 1994 through 1996), which allow larger sample sizes than the SIPP. In addition, the CPS data allow us to examine several groups of interest, in particular, the foreign born (immigrants) and the second generation (the native born whose parents were born abroad). This latter category is particularly interesting because the lasting effect of immigration on the U.S. economy and society depends on the behavior of the children and grandchildren of immigrants to an even larger extent than on the behavior of immigrants themselves. Additionally, combining the CPS data with 1970 Census Bureau data allows us to calculate the intergenerational correlation in welfare use between immigrants and the children of immigrants.

Welfare Use in the March Current Population Surveys, from 1994 to 1996 The bottom panel of table 11.2 shows the fraction of individuals18 receiving various benefits from welfare programs as reported in the CPS from 1994 to 1996. We can see a similar pattern across the columns for all of these programs. Except for energy assistance, immigrants are more likely to receive these transfers than are the second generation or those with two native-born parents. The differences in average receipt between immigrants and the third generation are small for cash programs like AFDC and SSI (0.007 and 0.009) but are larger for in-kind transfers like food stamps and Medicaid (0.036 and 0.037). The receipt of these transfer programs is, in general, lower among second generation immigrants than among any other group. Although the results across groups are intriguing, they give little insight into what might be driving these differences. For example, is the low participation in school lunch programs among the second generation attributable to the fact that they were exceptionally prosperous, or is it merely that they are older and less likely to have children present in the household? Is the low use of energy assistance among immigrants attributable to the fact that they have less access to information about these programs, or is it simply that they tend to live in warmer climates, where winter heating is much less a concern? Ideally, one

I

479

Finding Jobs

would like to be able to decompose differences in welfare participation into different components. These differences are often characterized as differences in eligibility versus differences in "taste" or reactions to "stigma" or information about eligibility. The reasons for differences in welfare use may have different policy implications, and so it may be important to separately identify them. However, these underlying reasons for welfare use are notoriously difficult to disentangle, and before expectations are raised too high, we want to stress that we will not be able to do so here. Instead, in the following analysis, we compare participation in various welfare programs of immigrants, the second generation, and those with native-born parents, controlling for individual characteristics. These results can only answer the following question: Compared with the third generation with similar observable characteristics, are immigrants more or less likely to use welfare? As stressed by Orley Ashenfelter (1983), one cannot answer questions about welfare stigma simply by regressing participation in welfare programs on individual characteristics. There may be unobservable characteristics associated with both eligibility and immigrant status that drive differences in welfare participation. 19 However, the comparisons are interesting for several reasons. First, some of the differences between these groups may be mechanical: the fact that the second generation has far lower use of welfare programs may be driven by the fact that they are older and are less likely to have children; individually, they may be just as likely to use welfare as those with similar age profiles. In addition, comparisons between those with similar characteristics is at least instructive. If immigrants are found to be less likely to use welfare than those with similar characteristics, we know that either they are less welfare prone or they are less likely to be eligible, for some unobservable reason. Other data sources may allow future research to fO,cus on identifying those unobservables. 20 Finally, as policy makers change welfare and immigration criteria to try to affect immigrants' participation in welfare, it is worth knowing, at least, how differences in observable characteristics correlate with differences in welfare use for immigrants and the native born. That said, table 11.3 presents linear probability models for receipt of 551, AFDC, Medicaid, and food stamps.21 The numbers reported are the coefficients on the indicator variables representing different groups of interest. The omitted category is the third (and higher) generation (that is, the native born with US.born parents). Column 1 simply presents average differences between the firstand second-generation immigrants and the comparison group. Because there are substantial age differences across these groups (in particular, the second generation is much older) and because participation in these programs may follow a distinct age profile, we include age and age squared in column 2. In addition, different household characteristics may affect the relative receipt across different groups. For example, immigrants are more likely to live in larger households, and their household age structure is tilted toward the young. Note that individuals are defined as receiving food stamps if anyone in the household receives food stamps, so there will be a mechanical relationship between household size

480

I

Use of Means-Tested Transfer Programs TABLE 11.3

/

Linear Probability Models for Various Welfare Benefits (1)

(2)

(3)

(4)

(5)

(6)

0.008 (0.001) -0.002 (0.002) -0.006 (0.001) 0.028 (0.003) 0.0008

0.010 (0.001) -0.011 (0.002) -0.010 (0.002) 0.027 (0.003) 0.0192

0.011 (0.001) -0.010 (0.002) 0.009 (0.002) 0.028 (0.003) 0.0216

-0.000 (0.002) -0.010 (0.002) -0.007 (0.001) 0.020 (0.003) 0.0458

-0.007 (0.002) -0.012 (0.002) -0.007 (0.002) 0.015 (0.003) 0.0477

-0.003 (0.002) -0.011 (0.002) -0.007 (0.001) 0.022 (0.003) 0.0276

0.007 (0.001) -0.017 (0.001) -0.008 (0.002) 0.047 (0.004) 0.0020

0.006 (0.001) -0.003 (0.001) -0.001 (0.001) 0.036 (0.003) 0.0894

0.001 (0.002) -0.007 (0.001) -0.003 (0.002) 0.034 (0.004) 0.0926

-0.005 (0.002) -0.008 (0.001) -0.002 (0.001) 0.029 (0.003) 0.1036

-0.010 (0.002) -0.007 (0.001) -0.001 (0.001) 0.027 (0.004) 0.1088

-0.001 (0.001) -0.006 (0.001) -0.001 (0.001) 0.040 (0.004) 0.0246

0.037 (0.002) -0.011 (0.003) -0.016 (0.003) 0.082 (0.005) 0.0035

0.033 (0.002) -0.015 (0.003) -0.014 (0.003) 0.069 (0.005) 0.0729

0.030 (0.003) -0.015 (0.003) -0.014 (0.003) 0.070 (0.005) 0.0790

0.005 (0.003) -0.017 (0.003) -0.010 (0.003) 0.052 (0.005) 0.1147

-0.009 (0.003) -0.019 (0.003) -0.009 (0.003) 0.042 (0.005) 0.1216

0.010 (0.003) -0.017 (0.003) -0.010 (0.003) 0.065 (0.005) 0.0515

0.036 (0.002) -0.032 (0.002) -0.024 (0.003) 0.085 (0.006) 0.0042

0.022 (0.002) -0.017 (0.002) -0.014 (0.002) 0.065 (0.005) 0.1031

0.030 (0.003) -0.009 (0.002) -0.007 (0.002) 0.071 (0.005) 0.1093

0.005 (0.003) -0.011 (0.002) -0.003 (0.002) 0.053 (0.005) 0.1490

-0.007 (0.003) -0.011 (0.003) -0.001 (0.002) 0.042 (0.005) 0.1591

0.006 (0.003) -0.018 (0.002) -0.009 (0.002) 0.064 (0.005) 0.0654

n = 120,407

Receipt of SSI Foreign born Both parents foreign born One parent foreign born Other nativity R2

Receipt of AFDC Foreign born Both parents foreign born One parent foreign born Other nativity R2

Receipt of Medicaid Foreign born Both parents foreign born One parent foreign born Other nativity R2

Receipt of food stamps! Foreign born Both parents foreign born One parent foreign born Other nativity R2

(Table continues on p. 482.)

I

481

Finding Jobs TABLE 11.3

/

Continued

Age, age' Household characteristics State and urban-rural Education Sex, marital status Race and ethnicity

(1)

(2)

(3)

(4)

(5)

(6)

No

Yes

Yes

Yes

Yes

Yes

No No No No No

Yes No No No No

Yes Yes No No No

Yes Yes Yes Yes No

Yes Yes Yes Yes Yes

No No Yes No No

Source: Data are from the March Current Population Surveys, 1994 to 1996. Note: Individuals are age eighteen years or older. In all columns, the omitted category is the third generation (those native born with native-born parents). Column 2 includes controls for household size and age distribution, and female-headed family, and the individual's age and age squared. Column 3 adds controls for the state of residence and whether the individual lives in a rural area. Column 4 adds controls for the individual's sex, marital status, and education. Column 5 adds controls for the individual's race and ethnicitv. Column 6 controls for age, age squared, and education. March supplement weights used. Standard errors in parentheses. 'The receipt of food stamps has a value of one if anyone in the individual's household participates in the program.

and receipt. Because many of these programs target children, having children will make one more likely to be eligible for the programs. Thus, we also add controls for household size and structure in this column. Column 3 adds controls for state of residence and urban residence. Employment opportunities vary by state and urban-rural residence and may affect welfare receipt. In addition, eligibility rules vary by state for some programs, and because immigrants are concentrated in a few states, place of residence may be an important determinant of welfare receipt.22 Column 4 adds controls for personal characteristics, induding education, sex, and marital status. Column 5 adds race and ethnicity. If there is discrimination in the labor market against minorities, then immigrants, who are much more likely to be Hispanic and Asian than the native born, may have more difficulty translating their skills into income than white non-Hispanics in the U.S. labor market. Finally, column 6 controls only for education and age. Some policy makers have suggested using education as a screening device for immigration in the future. Results in this column indicate the effect this may have on welfare participation. The results in the first column of the upper panel show that immigrants are about 0.8 percent more likely to receive SSI than is the third generation. Adding age and household characteristics increases their relative SSI receipt, but once education is added, the foreign born are no more likely to receive SSL The foreign born are 0.7 percent less likely to participate in SSI than the third generation when all controls are entered. The AFDC results are similar, except that each additional control decreases the relative receipt rate of the foreign born. Once age, household characteristics,

482

I

Use of Means-Tested Transfer Programs

and geographic controls are added, the foreign born are no more likely to receive AFDC than the third generation. When race and ethnicity are held constant, immigrants are significantly less likely to receive AFDC. The results change somewhat for the in-kind transfer programs in the lower portion of table 11.3. The initial differences in immigrant usage of Medicaid and food stamps are larger than for the cash transfer programs, and the coefficients do not decline to insignificance quite as quickly with the addition of control variables. However, when race and ethnicity are held constant, as depicted in column 5, immigrants are significantly less likely to use these programs than the third generation. In column 6, however, in which only age and education are held constant, immigrants are shown to be about 1 percent more likely to be covered by Medicaid and about 0.6 percent more likely to receive food stamps. The case of Medicaid is particularly interesting. Unlike the other programs analyzed here, Medicaid has a private alternative-namely, medical insurance purchased through an employer. 23 In our data, immigrants are less likely to report being employed than similar natives. Immigrants' lower employment rates may affect their ability to gain access to private health care. In addition, the value reported here is for coverage by Medicaid, not actual usage of medical services, so it is difficult to say how these figures might translate into dollars spent. On the other hand, Medicaid is not a traditional insurance program in that people can enroll once they need services. If immigrant adults find out that they are covered by the program when their children have become ill, this may account for some of the differences (although clearly controlling for the presence of children in the household, as we do in column 2, does not knock out the coefficient). Another possibility is that immigrants who come from countries with poor public health services may be sicker than the native born. Upon seeking services for their illnesses they would learn of their eligibility for Medicaid coverage. 24 The results here show that in general, immigrants are more likely to use transfer programs, on average, than is the third generation. 25 However, immigrants are almost always significantly less likely to use programs than the third generation with similar characteristics. 26 In general, simply controlling for age and education drives the difference between immigrants and the comparison group to insignificance. As mentioned above, the fact that the immigrant coefficient tends to be negative and significant when immigrants are compared with the third generation with similar characteristics may have several interpretations. It may mean that immigrants have a greater distaste for welfare programs, even though they are eligible for them. It may simply mean that immigrants have less information about programs than comparable native-born individuals. It may mean that immigrants have unobservable characteristics that lead them to have higher incomes and thus lower eligibility. In order to shed some light on these competing explanations, we ran regressions of log average hourly earnings on the same right-hand-side variables described above. Immigrants have lower hourly earnings, on average and controlling for all the characteristics described here. Although there may be other unobservable characteristics that affect their eligi-

I

483

Finding Jobs

bility (savings, or their sponsor's income, for example), these unobservable characteristics do not appear to be positively correlated with hourly earnings. Future research may be able to exploit relative usage across the various programs and variation in state eligibility criteria to shed some light on this issue. 27 Suppose for a moment that eligibility criteria were identical across the different programs. Then differences in receipt rates for immigrants could be attributed to differences in "tastes" for the program (or stigma, or information). Although it is not true that the eligibility is the same across all the programs, the fact that the immigrant coefficients tend to be largest (on average and controlling for characteristics) for Medicaid, school lunch, and food stamps programs suggests that these are either programs immigrants particularly value or that they are programs immigrants have information about (or both).28 The extent to which immigrants' use of welfare programs matters for the long-term solvency of public transfer programs in the United States may ultimately depend on whether these immigrants' children become dependent on these programs. The news on this front is optimistic: the second generation (those with two foreign-born parents) tend to have the lowest transfer program participation rates in the data from 1994 to 1996. 29 Of course, this does not mean that the children of immigrants who arrive here in the 1990s will necessarily follow this same pattern, because the characteristics of the parents of the second generation in the sample from 1994 to 1996 are quite different from those of immigrants arriving in the 1990s. Appendix table llA.4 shows linear probability models for transfer receipt in the 1970 census: immigrants in those data are not significantly more likely to receive public assistance than the third generation. This is the usual caveat we should bear in mind, as stressed by Borjas (1985), whenever we try to use evidence from cross-sectional analyses to extrapolate to longitudinal effects.

INTERGENERATIONAL CORRELATION IN WELFARE USE David Card, John DiNardo, and Gena Estes, in their study on the intergenerational correlations in education and wages between immigrants and their children (2000), find that the second generation has both higher wages and more education, using three cross-sections: the 1940 U.S. census, the 1970 U.s. census, and the CPS from 1994 to 1996. They also find high rates of intergenerational correlation in these outcomes. For earnings, they find correlations between the earnings of immigrant fathers and their children on the order of 0.4 to 0.6. Given this evidence, we would expect to find fairly high rates of intergenerational correlation in welfare use, given that a large part of what determines whether one uses welfare programs is one's earnings. There is a large literature on intergenerational welfare transmission in the United States. 30 Most studies have found consistent evidence of strong correlations between parental welfare receipt and daughter's welfare receipe1 Some studies, one by John Antel (1992), for example, go further to investigate the

484

I

Use of Means-Tested Transfer Programs question of whether a parent's use of welfare has an independent effect on his or her children's use of welfare once the children are grown, above and beyond the individual and family characteristics predisposing poverty. One possibility is that children whose parents use welfare learn how to navigate that system, whereas children whose parents work in the labor market learn how to get and keep jobs. After controlling for observed and unobserved heterogeneity, Antel finds that a mother's welfare participation increases the likelihood of her daughter's later welfare dependency. We are less ambitious and carry out a quite different exercise here. First, because of data limitations, we are not actually able to link parents and children with respect to their usage of welfare. So our main goal is to provide some descriptive evidence of the intergenerational correlations of welfare use for immigrants rather than to separate true state dependence from heterogeneity. We apply a grouping estimation strategy, similar to that used by Card, DiNardo, and Estes (2000). To fix ideas, we consider a simple descriptive model at the aggregate level of country-of-origin groups instead of the microeconomic level of individuals:

ph

where Tj2 and Tjl are the mean receipt rates of transfers by the origin group of the second generation from 1994 to 1996 and immigrants in 1970, respectively. We are interested in estimating the coefficient b, which tells us the percentage point increase in the mean receipt rate for the second generation that is associated with a 1 percentage point increase in the immigrant generation's mean receipt rate. A coefficient equal to one implies that the first generation's welfare use will be duplicated in the second generation, indicating that there is no assimilation of the second generation to the level of use of the native born. A coefficient of zero indicates that the first generation's welfare use has no impact on the second generation's use; this can be interpreted as complete assimilation. More specifically, we use data on the foreign born from the ages of eighteen to sixty-six from the 1970 census to identify the immigrant generation and data from the March CPS for 1994 to 1996 for those who were native born, age eighteen to sixty-six, with two foreign-born parents to identify the second generation. We divide each generation into the same thirty-five origin groupS.32 We then apply the following estimation strategy: first, we estimate the age-adjusted mean rates of receipt of cash assistance for the immigrants in 1970 by origin group.33 Second, we follow the same procedure to estimate the adjusted mean participation rates for different programs by the second generation from 1994 to 1996. Third, we regress the second generation's mean rates of participation (for different programs) from 1994 to 1996 on the mean rates of receipt of cash assistance for the immigrants in 1970. The coefficient from this regression gives us an estimate of the intergenerational transmission of welfare use. 34 It is unfortunate that the 1970 census does not contain information on use of other types of transfer programs, because it is possible that there is a closer

I

485

Finding Jobs

relationship between parents' use of a particular type of program and their children's use of that type of program. Borjas and Hilton (1996) find evidence that the types of benefits received by earlier immigrants from a particular country influence the types of benefits received by subsequent immigrants from that country. This type of connection might exist between generations, as well. Because the 1970 data contain information on cash transfers only, we focus our attention here on the receipt of SSI and AFDC among the second generation. The coefficients for these regressions are reported in the first column of table 11.4. The coefficient for the parent generation's receipt of public assistance on second generation receipt of AFDC is 0.533. This implies that 1 percent more recipients in the parents' generation is correlated with about 0.5 percent higher receipt in the second generation. Although all of these coefficients are positive and significant, any that are less than one indicate that the differences in the welfare use between the immigrants and the underlying population mean will die out after a number of generations. For example, assuming the intergeneraTABLE 11.4

/

Effect of Parent's Generation's Outcome on Second Generation's Outcome Estimated Coefficient Without Additional Controls (1)

Estimated Coefficient, Controlling for Second Generation's Education (2)

Receipt of SSI

0.470 (0.082) [0.498]

0.160 (0.195) [0.552]

Receipt of AFDC

0.533 (0.119) [0.376]

0.384 (0.290) [0.414]

Single female headed household with children

1.043 (0.236) [0.371]

0.506 (0.261) [0.552]

Second Generation Outcome

Source: March Current Population Survey, 1994 to 1996. Note: The dependent variable is the cell mean for the outcome for the second generation (age eighteen to sixty-six), adjusted for age and age squared. The right-hand-side variable is the cell mean for the outcome for immigrants in the 1970 census (age eighteen to sixty-six), adjusted for age and age squared. In the first two rows, the first generation outcome is receipt of cash transfers. In the third row, the outcome is single-female-headed household with children. Column 2 adds controls for the fraction of the cell that were high school dropouts or high school graduates or attended some college. The cells are defined as country of origin for the immigrants, or country of origin for the mother for those who report that both parents were born abroad. There are thirty-five country-of-origin groups included here. The regressions are weighted by the cell size for the second generation. Standard errors are in parentheses, R2 in brackets.

486

I

Use of Means-Tested Transfer Programs

tional correlation between the immigrants who arrive in the 1990s and their descendants will be the same as that between the immigrants in 1970 and their offspring in the 1990s, we can calculate how quickly the difference in AFDC receipt between current immigrants and the native born will die out. The raw difference in the AFDC receipt between the immigrants and the third generation from 1994 to 1996 (0.7 percent) will become insignificant between the third and fourth generations." The coefficient for SSI is somewhat lower than that for AFDC, as one might expect since SSI is the more exclusive program. Figures 11.3 and 11.4 show the relationship between receipt of SSI and AFDC by the second generation and receipt of cash public assistance by the parent's generation. The data points are plotted and labeled by the origin group. The fitted regression line is displayed.'" In the figures, we can examine where each country-of-origin group lies in relation to the regression line. The United States is included for comparison, although these numbers are not used to determine the regression line. In both figures, four countries exhibit very high welfare receipt for the parent's generation: Cuba, the Dominican Republic, EI Salvador, and Mexico. If not for these four origin groups, the relationship between welfare receipt in the first and second generations would be much less clear.37 However, because immigrants from these four countries represent a sizable portion of the foreign born from 1994 to 1996 (Cuba, 3.7 percent; the Dominican Republic, 2.1 percent; El FIGURE 11.3 / '>D

(J\

(J\ ,..., ...,0 ~

(J\ (J\

,...,

r..J CI

IJ;.,

-< ......

...,0p.

';:J u

(J)

~ ~

.g ...roc:; ~

(J)

CJ "0 ~

0 u

(J)

CIl

.09 .085 .08 .075 .07 .065 .06 .055 .05 .045 .04 .035 .03 .025 .02 .015 .01 .005 0

Intergenerational Correlation in Receipt of Cash Transfers

Dam. Rep.

u.s.

Jamaica

ElSal

Japan

Phil

R lid Haiti Czech India Portugal Gree~~nce Korea Mid. l'lJliItColblTlBiajS. AmGuatem.kJicar.

.01

.015

.02

.025

.03

.035

.04

.045

.05

Cuba

.055

.06

.065

.07

.075

Immigrant's Receipt of Cash Public Assistance, 1970

Source: Authors' tabulations.

487

Finding Jobs FIGURE 11.4

'Cl 0\

0\ ,.... ....0 -.j'

0\ 0\

,....

gf ""'0

....p..

'iiJ u

OJ

~

~

0

.~

....OJ ~

OJ

(j 't ~

0

u OJ CIl

.03 .025 .02 .015 .01 .005 0 -.005 -.01 -.015 -.02 -.025 -.03 -.035 -.04

/

Intergenerational Correlation in Receipt of Cash Transfers

Dom. Rep.

c./s.1A-Rt USSR Mid. East

fffi'a1

J~iln

G eee Phil HunQiiYada Ko a

India ~sia Col~8b/ Af

Gua~malt,

eua or

&.~d Eur Ilnce

Czech

.01

.015

.02

.025

.03

.035

.04

.045

.05

.055

.06

.065

.07

.075

Immigrant's Receipt of Cash Public Assistance, 1970

Source: Authors' tabulations.

Salvador, 2.85 percent; and Mexico, 27.0 percent), the intergenerational relationship in welfare use is likely to continue to be important in the future. Cuba is a particularly interesting case. Cuban immigrants were refugees and would have been eligible for many of the standard transfer programs immediately upon arrival. In addition to the standard programs, however, there are many special transfer programs specifically for Cuban refugees. 38 In 1969 the federal government spent $70.7 million on these special programs for Cuban refugees (Congressional Information Service 1970).39 These programs included funds for welfare assistance (cash transfers), resettlement (transport to areas within the United States), education, health services, and transport of refugees from Cuba. The 1970 census data indicate that Cubans took advantage of the availability of these programs: Cuban immigrants in 1970 have the third highest rate of welfare receipt of any group-lower than the Dominican Republic and El Salvador but higher than Mexico. This does not appear to have translated into welfare dependency among their children. Cuba is almost always below the regression line, which indicates that the second generation's welfare use died out sooner than one would have expected given the parent generation's welfare use. Mexico is important because it has always represented a large proportion of immigrants to the United States (7.55 percent in 1970 and 27.0 percent from 1994 to 1996). The immigrant generation shows high levels of welfare receipt, but the

488

I

Use of Means-Tested Transfer Programs

second generation's receipt is lower than the u.s. average. In both graphs, Mexico is very close to the regression line, because this observation is the leverage point for these regressions. The regressions are weighted by the second generation cell size, and Mexico has by far the largest number of second generation individuals. Because most AFDC recipients were single mothers and their children, it is also interesting to examine the intergenerational transmission of the rate of single motherhood. Owing to data limitations, instead of using the ideal definition of "single mother," we investigate this issue by looking at individuals who report that they are single female heads in households with children under the age of eighteen. The left- and right-hand-side variables are the fraction of each origin group in each year reporting that they fall into this category. (For brevity, we call this fraction the "rate of single headship.") We have adjusted for age as described above. Results are reported in the bottom row of table 11.4. In this exercise, the coefficient is not significantly different from one, suggesting that the rate of single headship among immigrants in 1970 will be duplicated among the second generation in the 1990s. In 1970, the rate of single-female headship was 0.014 among immigrants and 0.012 among (all) native born. In 1990, the rate of single-female headship was 0.047 among immigrants and 0.055 among (all) native born. This suggests that if the assimilation process is stable over time, then the offspring of current immigrants in the 1990s should continue to have lower rates of single-female headship than the rest of the native born. It is important to recall that the results in the first column of table 11.4 do not adjust for anything that would capture the immigrant or second generation's income level. Thus, much of the correlation between the first and second generation's welfare use must come through correlations in earnings. These results, then, are consistent with Card, DiNardo, and Estes (2000), who find a high correlation in earnings across generations. They also find that in the cohort from 1970 to 1995, the intergenerational transmission from the immigrant parents to their native-born children works only through education. The second column in table 11.4 reports the regression coefficients when we control for education of the second generation (the cell means for fraction high school dropouts, fraction high school graduates, and fraction with some college). When we add these controls, we find no statistically significant relationship between welfare use in the immigrant generation and receipt of either of the transfer programs by the second generation. 4o When we control for the second generation's education level in the single-female headship regressions, the coefficient falls to 0.506 and is only marginally statistically significant. There are some caveats to bear in mind about our results reported above. First, as pointed out by Card, DiNardo, and Estes (2000), there is slippage from this grouping estimation, because there is no guarantee that the immigrants identified in the 1970 census are the parents of the second generation identified in the cps from 1994 to 1996. Further restrictions of age on both sides may improve the match. 41 Second, additional problems arise if there is significant selective out-migration of immigrants. If immigrants who do poorly in the United

I

489

Finding Jobs

States do not stay, then their children will not be in the sample, and we will tend to overestimate the economic progress across generations. Information on out-migration is not readily available because the United States does not keep statistics on emigration. However, Wei-Yin Hu (1998) uses panel data from the Health and Retirement Study to track immigrants over time. He finds that out-migration may impart significant bias to synthetic cohort estimates of assimilation. Interestingly, he finds that selective out-migration is more prominent among non-Hispanic whites than among other groups of immigrants. Although his paper mainly focuses on the biases that may result in estimating intra generational assimilation, his cautions are germane here as well. In particular, out-migration is likely to affect the results for the second generation, because these will be the children of immigrants who did relatively well. In addition, we do not know whether the results for immigrant participation in the 1970 census include immigrants who would eventually leave the United States because they disliked their outcomes. To probe the importance of these problems for our analysis, we performed two additional investigations. First, we ran our regressions again, restricting the first generation sample to women with children under the age of eighteen present in the household and the second generation sample to women. Families with children-many of whom are likely to be native born-are arguably less likely to return home. When we use these data, we obtain similar results. The coefficients are 0.397 and 0.547 for 551 and AFDC, respectively. Secondly, Hu (1998) finds less selective out-migration for Hispanics, so we restrict the first generation in 1970 to those who report they are Hispanic. We then match the second generation from 1994 to 1996 to the first generation, using country of origin. When we use the remaining country-of-origin groups, the coefficients for both programs are about half their original magnitude (0.202, but insignificant, for SST and 0.277 for AFDC). These results suggest that differences in welfare use between immigrants and the native born tend to die out over time, even under the old welfare rules. However, one should use caution in extrapolating the results from the cohort from 1970 to 1995 to later cohorts. The coefficients estimated here will depend on the economic environment. 42 As that changes, the amount of assimilation is likely to change, as well, and there is no doubt that the environment surrounding welfare receipt is changing.

SUMMARY AND CONCLUSION The Personal Responsibility and Work Opportunity Reconciliation Act of 1996 brought dramatic changes to the welfare system in the United States. Although welfare reform in general devolved power from federal to state governments, leading to a great deal of interstate variation in the generosity of the social safety net, this variation is even greater for immigrants. States have taken very differ-

490

I

Use of Means-Tested Transfer Programs

ent approaches to providing services for immigrants. Assuming immigration patterns persist, those states with heavy immigrant caseloads will either have to face substantial portions of their needy populations remaining unserved, or they will have to bear the costs of programs to cover their needs. In general, states with the heaviest immigrant caseloads and those with the most generous existing welfare programs have led the way in designing programs to fill the holes in the social safety net (see Zimmermann and Tumlin 1999). Part of the underlying motivation for welfare reform in general was to change incentives for welfare recipients. For example, TANF emphasizes work, with an eye toward eliminating individuals' incentives to reduce their work effort in order to qualify for-or opt into-welfare programs. Policy makers worried about the additional layer of incentives of welfare programs as they pertained to immigrants. Was the United States attracting poor immigrants because of the generosity of its welfare programs? If so, that fact might have both immediate and long-term consequences for the United States-would immigrants have children who were also likely to be welfare dependent? Given this motivation for welfare reform, it is worth examining immigrants' use of welfare programs before reform. We find that immigrants are more likely to use welfare programs than the native born, on average. This is particularly the case for in-kind programs like food stamps and Medicaid. Because newimmigrant eligibility for these two programs was essentially eliminated, outcomes related to food consumption and medical coverage will be particularly important to monitor once welfare reform takes full effect. We also find that immigrants are less likely to use welfare programs than natives with similar characteristics. There are several possible explanations for these results, which are impossible to separate here. Unobservable characteristics may make immigrants ineligible for these programs, which could be good news about the immigrants the United States has attracted, depending on the nature of these unobservables-if, for example, immigrants have substantial savings or other resources that make them ineligible. On the other hand, immigrants may be less likely than similar natives to use welfare because they perceive more welfare stigma or they simply may have lacked information about the available programs. In either of these cases, it is hard to imagine that immigrants were attracted to the United States by the welfare state. A more important issue is whether the children of immigrants are welfare dependent. First generation immigrants-who are likely to have trouble with English-may never be particularly successful in the U.s. labor market. To a large extent, the most important impact of immigration is long term. We present evidence on the intergenerationallink in welfare use between immigrants in the 1970s and the children of immigrants from 1994 to 1996. Although we find that there is a positive and significant correlation between the welfare receipt of the first and second generations, the coefficient is less than one, suggesting that welfare use among immigrant generations converges to that of the native born across generations. In addition, we find a great deal of variation in welfare use

I

491

Finding Jobs

across different country-of-origin groups and across generations. Some groups, like Cubans, with very high welfare receipt in the first generation had very low receipt in the second generation. Finally, we find little evidence of an intergenerational link in welfare use, other than the fact that if parents are poor, their children are also likely to be poor. Given that, the absence of a social safety net for immigrants after welfare reform is well under way does not augur well for the socioeconomic outcomes of their children. Future research will have to monitor the economic outcomes for immigrants and their children to see if welfare reform achieves its goals or simply increases the depth of poverty in this segment of the low-skilled population.

The authors would like to thank Orley Ashenfelter, Becky Blank, David Card, Anne Case, John DiNardo, Hank Farber, Phil Levine, Anne Piehl, Robert Schoeni, Karen Tumlin, seminar participants at Princeton University, and participants at the Labor Markets and Less Skilled Workers preconference at the Joint Center for Poverty Research for helpful comments. We thank Karen Tumlin and Wendy Zimmermann for allowing us to excerpt their work on changes in immigrant eligibility after welfare reform. Kristin Butcher would like to thank the Industrial Relations Section at Princeton University for generous support. Luojia Hu gratefully acknowledges financial support from the Woodrow Wilson Society of Fellows. The views expressed here are those of the authors and not necessarily those of our funders or employers. All errors are our own.

(Text continues on p. 502.)

492

I

/

Some college

High school degree

High school dropout

Age

Individual characteristics Female

Percentage of single female heads with children in household

0.547 (0.498) 52.79 (19.31) 0.525 (0.499) 0.223 (0.416) 0.097 (0.296)

94,606 7.24 8.90 8.62

Foreign Born

0.528 (0.499) 53.83 (14.02) 0.508 (0.500) 0.304 (0.460) 0.091 (0.287)

114,433 8.75 . 10.41 5.06

Both Parents Foreign Born

Individual Characteristics by Nativity Groups

n Percentage of sample

TABLE 11A.l

APPENDIX

0.525 (0.499) 46.67 (17.15) 0.375 (0.484) 0.345 (0.476) 0.145 (0.352)

77,765 5.95 5.22 4.14

One Parent Foreign Born

0.520 (0.500) 44.24 (18.88) 0.529 (0.499) 0.289 (0.453) 0.101 (0.302)

67,664 5.18 6.41 10.36

Other

(Table continues on p. 494.)

0.526 (0.499) 42.22 (17.34) 0.405 (0.491) 0.342 (0.475) 0.141 (0.348)

190,543 72.88 69.07 71.83

U.s.-Born Parents

u.s. Born and

/

Continued

Female headed

More than one family

Household characteristics Number of people

Log average hourly earnings

Worked last year

Married

Other race

Asian

Hispanic

African American

College graduate

TABLE 11A.1

0.155 (0.362) 0.025 (0.157) 0.137 (0.344) 0.052 (0.223) 0.008 (0.089) 0.670 (0.470) 0.538 (0.499) 1.150 (0.762)

3.098 (1.702) 0.019 (0.135) 0.073 (0.261)

0.098 (0.297) 0.005 (0.071) 0.053 (0.225) 0.018 (0.133) 0.001 (0.034) 0.727 (0.445) 0.628 (0.483) 1.289 (0.707)

Foreign Born

3.115 (1.792) 0.034 (0.181) 0.071 (0.256)

Both Parents Foreign Born

3.337 (1.827) 0.019 (0.137) 0.069 (0.253)

0.133 (0.339) 0.007 (0.084) 0.060 (0.238) 0.010 (0.097) 0.003 (0.053) 0.710 (0.454) 0.669 (0.471) 1.244 (0.749)

One Parent Foreign Born

3.477 (1.918) 0.024 (0.154) 0.076 (0.264)

0.112 (0.315) 0.116 (0.320) 0.019 (0.137) 0.002 (0.041) 0.005 (0.067) 0.702 (0.457) 0.686 (0.464) 1.088 (0.756)

U.s.-Born Parents

u.s. Born and

3.131 (2.009) 0.024 (0.154) 0.084 (0.278)

0.081 (0.273) 0.167 (0.373) 0.119 (0.324) 0.007 (0.085) 0.009 (0.094) 0.558 (0.497) 0.646 (0.478) 0.910 (0.950)

Other

0.503 (0.442) 0.300 (0.389) 0.095 (0.245) 0.102 (0.262) 0.024 (0.152)

0.037 (0.188)

0.071 (0.167) 0.063 (0.189) 0.427 (0.412) 0.224 (0.365) 0.215 (0.372)

0.500 (0.430) 0.241 (0.353) 0.106 (0.250) 0.152 (0.304)

0.166 (0.244) 0.152 (0.269) 0.258 (0.302) 0.153 (0.295) 0.271 (0.395)

0.023 (0.148)

0.388 (0.439) 0.339 (0.415) 0.140 (0.305) 0.133 (0.307)

0.099 (0.195) 0.173 (0.326) 0.398 (0.410) 0.159 (0.326) 0.171 (0.349)

0.528 (0.436) 0.286 (0.386) 0.102 (0.260) 0.084 (0.240) 0.056 (0.229)

0.035 (0.184)

0.140 (0.231) 0.249 (0.360) 0.326 (0.375) 0.127 (0.292) 0.158 (0.337)

0.410 (0.465) 0.339 (0.440) 0.139 (0.320) 0.112 (0.298)

0.106 (0.207) 0.276 (0.400) 0.378 (0.420) 0.123 (0.306) 0.118 (0.307)

Source: 1970 Census, 15 percent Public Use Microdata Samples, age eighteen years and older. Note: The "U.s. born" are those who were born in the United States, and whose parents' were born in the United States. This is a 20 percent random sample of this group. Standard deviations are in parentheses.

Received cash public assistance

College graduate

Some college

High school degree

Education distribution for adults High school dropout

Fraction age older than sixty-five

Fraction age thirty-one to thirty-five Fraction age fifty-six to sixty-five

Household age distribution Fraction age younger than eighteen Fraction age eighteen to thirty

/

0.009 (0.094) 5,298 0.008 (0.088) 5,239 0.021 (0.144) 5,298 0.016 (0.126) 3,045

0.037 (0.188) 12,571 0.037 (0.189) 11,575 0.062 (0.242) 12,571 0.070 (0.255) 5,846

(1) Foreign Born

(2) Both Parents Foreign Born

2,905

5,172 0.033 (0.180)

5,190 0.032 (0.175)

0.021 (0.144) 5,172 0.017 (0.128)

(3) One Parent Foreign Born

Average AFDC and Food Stamps Receipt, Using Alternative Definitions

AFDC receipt Individual receipt, Individual defines category n Individual receipt, Family reference person defines category n Individual receives if anyone in family receives, individual defines category n Family receives if anyone in family receives, family reference person defines category n

Definition

TABLE 11A.2 (4)

44,092

83,509 0.050 (0.217)

77,540 0.042 (0.202)

0.030 (0.170) 83,509 0.027 (0.162)

U.S. Born and U.s.-Born Parents

1,166

2,056 0.119 (0.324)

2,043 0.100 (0.300)

0.071 (0.257) 2,056 0.069 (0.253)

(5) Other

/

Continued

0.047 (0.212) 5,298 0.043 (0.204) 3,045

0.123 (0.328) 12,571 0.124 (0.330) 5,846

(1) Foreign Born

2,905

5,172 0.057 (0.231)

0.055 (0.228)

One Parent Foreign Born

44,092

83,509 0.094 (0.292)

0.084 (0.278)

(4)

U.s. Born and U.s.-Born Parents

(3)

(2)

Both Parents Foreign Born

(5)

1,166

2,056 0.186 (0.390)

0.165 (0.371)

Other

Source: March Current Population Survey 1995. Note: Sample of adults, age eighteen years or older. AFDC receipt is asked for individuals and for the family. Individual receipt is equal to one if the person reports AFDC income. Family receipt of AFDC is equal to one if the family reports receipt of AFDC income. Food stamp receipt is asked in the form, "Does anyone in this household receive food stamps?" Individual receipt of food stamps is equal to one if anyone in the person's household receives food stamps. Family receipt is equal to one if anyone in the family'S household receives food stamps. "Category" refers to the nativity categories. If the individual defines the category, that means the individual's information on nativity was used. If the family reference person defines the category, that means the reference person's information on nativity was used.

Food stamp receipt Individual receives if anyone in household receives, individual defines category n Family receives if anyone in house hold receives, family reference person defines category n

Definition

TABLE llA.2

/

R2

Other nativity

One parent foreign born

Both parents foreign born

Receipt of energry assistance Foreign born

R2

Other nativity

One parent foreign born

Both parents foreign born

-0.007 (0.001) -0.008 (0.002) -0.007 (0.002) 0.024 (0.003) 0.0007

0.104 (0.003) -0.032 (0.002) -0.027 (0.002) 0.073 (0.005) 0.0154

(1)

-0.007 (0.001) -0.011 (0.002) -0.007 (0.002) 0.020 (0.003) 0.0252

0.068 (0.003) 0.006 (0.002) -0.006 (0.002) 0.047 (0.005) 0.2153

(2)

0.000 (0.001) -0.008 (0.002) -0.005 (0.002) 0.023 (0.003) 0.0343

0.074 (0.003) 0.017 (0.002) 0.002 (0.002) 0.054 (0.005) 0.2229

(3)

Linear Probability Models for Receipt of Various Welfare Benefits

Receipt of school lunch program Foreign born

n = 120,407

TABLE 11A.3

-0.009 (0.002) -0.008 (0.002) -0.004 (0.002) 0.016 (0.003) 0.0472

0.056 (0.003) 0.016 (0.002) 0.006 (0.002) 0.040 (0.005) 0.2509

(4)

-0.010 (0.002) -0.008 (0.002) -0.003 (0.002) 0.016 (0.003) 0.0488

0.037 (0.003) 0.010 (0.002) 0.004 (0.002) 0.015 (0.005) 0.2622

(5)

-0.017 (0.001) -0.011 (0.002) -0.005 (0.002) 0.019 (0.003) 0.0170

0.078 (0.003) 0.007 (0.002) -0.004 (0.002) 0.052 (0.005) 0.0815

(6)

No No No No No No

0.005 (0.001) -0.002 (0.001) -0.002 (0.001) 0.020 (0.002) 0.0007

0.005 (0.001) -0.000 (0.002) -0.004 (0.001) 0.040 (0.003) 0.0011

Yes Yes No No No No

0.007 (0.001) -0.002 (0.001) -0.001 (0.001) 0.018 (0.002) 0.0167

0.010 (0.001) -0.007 (0.002) -0.006 (0.002) 0.037 (0.003) 0.0286

Yes Yes Yes No No No

0.004 (0.001) -0.005 (0.001) -0.003 (0.001) 0.015 (0.002) 0.0188

0.010 (0.001) -0.007 (0.002) -0.006 (0.002) 0.035 (0.003) 0.0323

Yes Yes Yes Yes Yes No

0.002 (0.001) -0.005 (0.001) -0.003 (0.001) 0.013 (0.002) 0.0227

0.003 (0.002) -0.008 (0.002) -0.005 (0.002) 0.030 (0.003) 0.0419

Yes Yes Yes Yes Yes Yes

-0.000 (0.001) -0.004 (0.001) -0.002 (0.001) 0.012 (0.003) 0.0251

-0.005 (0.002) -0.008 (0.002) -0.004 (0.002) 0.024 (0.003) 0.0526

Yes No No Yes No No

0.002 (0.001) -0.003 (0.001) -0.001 (0.001) 0.D18 (0.002) 0.0047

-0.002 (0.001) -0.008 (0.002) -0.005 (0.002) 0.036 (0.003) 0.0136

Note: See note to table 11.3. The left-hand side variables are equal to one if anyone in the individual's household participates in the specified program. Standard errors are in parentheses.

Source: March Current Population Survey, 1994 to 1996.

Age, age2 Household characteristics State and urban-rural Education Sex, marital status Race and ethnicity

R2

Other nativity

One parent foreign born

Both parents foreign born

Receipt of rent subsidy Foreign born

R2

Other nativity

One parent foreign born

Both parents foreign born

Receipt of public housing Foreign born

Finding Jobs TABLE 11A.4

/

n = 545,011 Foreign born Both parents foreign born One parent foreign born Other nativity Household characteristics Age, age 2 State and urban or rural Education Sex, marital status Race and ethnicity R2

Linear Probability Models for Receipt of Cash Public Assistance, 1970

(1 )

(2)

(3)

(4)

(5)

(6)

0.001 (0.001) -0.012 (0.001) -0.013 (0.001) 0.020 (0.001 )

-0.018 (0.001) -0.021 (0.001) -0.016 (0.001) 0.014 (0.001)

-0.019 (0.001) -0.019 (0.001) -0.015 (0.001) 0.014 (0.001)

-0.022 (0.001) -0.021 (0.001) -0.014 (0.001) 0.007 (0.001)

-0.021 (0.001) -0.016 (0.001) -0.011 (0.001) 0.001 (0.001)

-0.016 (0.001) -0.021 (0.001) -0.014 (0.001) 0.013 (0.001)

No

Yes

Yes

Yes

Yes

No

No

Yes

Yes

Yes

Yes

Yes

No No

No No

Yes No

Yes Yes

Yes Yes

No Yes

No

No

No

Yes

Yes

No

No 0.0013

No 0.0471

No 0.0519

No 0.0642

Yes 0.0739

No 0.0299

Source: Data are from 1970 u.s. Census, 15 percent Public Use Microdata Sample. Note: Individuals are age eighteen years or older. Column 2 includes controls for household size and age distribution; indicators for a multifamily household and female-headed family, and the individual's age and age squared. Column 3 adds controls for the state of residence and whether the individual lives in a rural area. Column 4 adds controls for the individual's sex, marital status, and education. Column 5 adds controls for the individual's race and ethnicity. Column 6 only includes controls for individual's age, age squared, and education. Standard errors are in parentheses.

500

I

Use of Means-Tested Transfer Programs TABLE 11A.5

/

Country-of-Origin Groups and Sample Sizes for the Second Generation

Country of Origin Asia (NEC)' Canada Central and South America (NEC)' China Colombia Cuba Czechoslovakia Dominican Republic Ecuador EI Salvador Europe (NEC)! France Germany Greece Guatemala Haiti Hungary India Ireland Italy Jamaica Japan Korea Mexico Nicaragua Caribbean (NEC)! and Africa Peru Philippines Poland Portugal USSR United Kingdom Yugoslavia Middle East Other

Second Generation Sample Size

69 474 107

134 69 242 38 115

62 65 252 38 455

121 32 67 132 60 330 919 44

97 30 2,103 32 57 30 226

344 105 257 243 47 60 680

Source: March Current Population Survey, 1994 to 1996. Note: The groupings are consistent across the 1970 census and the Current Population Surveys from 1994 to 1996. The "second generation" are those people who say both parents were born abroad; the categories are based on mother's place of birth. The sample sizes are for the second generation in the 1994-96 CPS data. 'Not elsewhere classified (NEC); includes all countries in the region whose sample sizes are too small to be separately identified in CPS data. I

501

Finding Jobs

NOTES 1. For research addressing these concerns, see Borjas (1998), Borjas and Hilton (1996), Betts and Lofstrom (1998), and Butcher and Piehl (1998a, 1998b, forthcoming). 2. The "third generation" includes all native born with native-born parents. Their immigrant ancestors may be further back than their grandparents. 3. In addition to the Welfare Reform Act, the Illegal Immigration Reform and Immigrant Responsibility Act of 1996 addressed some similar issues. 4. Illegal immigrants were already excluded from almost all public aid (emergency medical treatment is an exception). 5. In addition, see Espenshade, Baraka, and Huber (1997) for description and analysis of the laws affecting immigrants' eligibility for means-tested aid programs. 6. "Preenactment" and "postenactment" are accurate, but cumbersome, designations. We use the terms "current" and "new" immigrants throughout this chapter for the sake of brevity. 7. "Deeming" means that the immigrant's sponsor's income is assumed to be available to the immigrant in its entirety-the immigrant is eligible if his or her income plus the sponsor's income is below the income cutoff for a given program. States have an incentive to encourage naturalization, because immigrants who become citizens become eligible for programs for which the federal government shares costs. Some states have programs to help with naturalization procedures. On May 17, 1999, the Supreme Court held state residence requirements for citizens to be unconstitutional. (Rita L. Saenz, Director, California Department of Social Services et aI., Petitioners, v. Brenda Roe and Anna Doe, etc.). 8. It is not possible in the CPS to determine which individuals arrived as refugees. However, the INS Statistical Yearbook reports that between 1994 and 1996, 13 to 15 percent of arrivals were refugees (table 2, "Immigrants Admitted, by Major Category of Admissions, FY-94-96,"