The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
August 2001 To Our Readers: I would like to take this opportunity to share some recent news at the Economic Policy Review. After many years of valuable service as editor of the Review, Paul Bennett has retired from the Federal Reserve Bank of New York. Paul is now Senior Vice President and Chief Economist at the New York Stock Exchange. In July, Erica Groshen, an assistant vice president in our domestic research area, succeeded Paul as editor. Erica, a Harvard-trained labor economist, has conducted research in the Federal Reserve System for thirteen years, headed our international and domestic research areas, and served on the editorial boards of the Review and other economic journals. Erica will be assisted by an active editorial board: Linda Goldberg, James Kahn, and Hamid Mehran. In addition, consistent with the main goal of the Review—to make the policy-oriented research of our economists available to a wide range of readers—we will be enhancing our electronic presentation of articles. The Economic Policy Review’s web site will begin to offer executive-level summaries of new refereed articles, a feature that will make the chief findings of these studies easier and faster to absorb. The summaries will also be interactive, enabling readers to link conveniently to key charts, related articles, and other resources. This feature, along with our current practice of posting articles prior to their print availability, is designed to take advantage of the many benefits of electronic publishing. I hope you will agree that these developments will broaden our ability to deliver timely and thoughtprovoking articles on important policy issues. To that end, we welcome your comments on the Review. Please do not hesitate to contact Erica or any member of her editorial board with your suggestions. We look forward to hearing from you. Christine M. Cumming Executive Vice President and Director of Research Jamie B. Stewart, Jr. Opening Remarks G ood morning. I am delighted to welcome you to our conference “Welfare Reform Four Years Later: Progress and Prospects.” In 1996, sweeping legislative changes in public assistance ushered in a period of remarkable change in how welfare is administered in New York, New Jersey, and the nation as a whole. The purpose of today’s conference is to explore the nature of the reforms introduced, the consequences of these changes, and the prospects for the future. It was the Personal Responsibility and Work Opportunity Reconciliation Act of 1996 that set welfare reform in motion these past several years. The name says it all: work and responsibility. The act was designed to encourage welfare recipients to find work and to take personal responsibility for their efforts. The motivation for the legislation was to reduce the number of people on public assistance while at the same time increasing the number of people with jobs. The goal, in short, was responsible economic self-sufficiency for the least advantaged Americans. Four years later, we may reasonably ask: Has welfare reform been successful? Have more people reached economic selfsufficiency as a result of this legislation? In broad terms, what we have found is that the number of people on public assistance has fallen dramatically in the past few years. Welfare caseloads today are one-half the peak they reached in 1994. There are a number of open questions, however, some of which we will explore today. One question concerns the issue of time limits. Under the 1996 legislation, each family on public assistance faces a maximum number of years it can receive welfare. For example, quite a few New Yorkers may run out of eligibility for welfare benefits in 2002. Will these families be able to find jobs? How much do macroeconomic conditions matter? At the level of the individual families, a related question we may well ask is what has happened to each of the families that left the welfare rolls? Some of these families may be economically self-sufficient, but others may have fared far less well. A second issue to consider is the impact of the legislation on high-risk women—women without much education or work experience. Does anything special need to be done to help these hard-to-employ women? Finally, we also want to think about our own area, the Second District, which encompasses New York, New Jersey, and Fairfield County. The 1996 legislation gave each state wide latitude to formulate its own welfare policy. Thus, an obvious question is how has the experience of the New York– New Jersey region differed from that of the rest of the country? Can we learn anything about these differences? The Federal Reserve Bank of New York is pleased to provide a forum to address these issues. In fact, one of our jobs here at the Bank is to facilitate the free exchange of ideas on public policy. We certainly think that monetary policy is important, Jamie B. Stewart, Jr., is first vice president of the Federal Reserve Bank of New York. FRBNY Economic Policy Review / September 2001 1 but we also recognize that social policy matters. Some of you may have been to our 1999 conference on income inequality. We are also organizing a conference on productivity to be held in November 2001. Our conferences are unique because they are nonpartisan and bring together academics, practitioners, and other experts. We hope that today’s conference will begin to provide answers to some of the questions I have raised. We also look forward to learning from each other. Our conference speakers represent diverse areas of expertise. We are pleased that leading representatives from the fields of economics and sociology are joining us today to present their latest research on welfare reform outcomes. The panel of discussants for the conference’s closing session includes 2 Opening Remarks distinguished representatives from business, public health, and community services. And we are especially fortunate to have the New York State Executive Deputy Commissioner of Labor, James Dillon, as our luncheon speaker. Finally this morning, I would like to emphasize how much we value the participation of all of you in the audience. Most of you here today also bring a wealth of experience in welfare reform that we want very much to incorporate in our conference proceedings. Thus, we have scheduled specific time for discussion at the end of the day. We also intend to leave time at the end of each of the individual sessions for open discussion. Today’s conference promises to be both informative and productive, and once again I am delighted to welcome you. Stephen V. Cameron, Robert A. Moffitt, and Carol Rapaport Summary of Observations and Recommendations I n 1996, Congress passed the Personal Responsibility and Work Opportunity Reconciliation Act in an effort to “end the dependence of needy parents on government benefits by promoting job preparedness, work, and marriage.” The welfare reform embodied by this legislation shifted the responsibility for policymaking to the states while imposing new federal mandates, such as time limits on receipt of welfare funds paid by the government and stricter work requirements and sanction policies. As part of this reform, the major cash assistance program for poor families became known as Temporary Assistance for Needy Families, reflecting the goal that such government aid should not be received on a longterm basis. The potential effects of welfare reform on low-income families have since become an issue of debate. Critics argue that although families have always left the welfare rolls voluntarily, the new legislation greatly increases the number who are being forced to depart. As evidence of these concerns about family well-being, the critics point to the steep national decline in the welfare caseload. Meanwhile, proponents of welfare reform contend that the caseload decline is an indicator of the success of the measures, rather than a cause for concern. To help put these issues in perspective, the Federal Reserve Bank of New York hosted the conference “Welfare Reform Four Years Later: Progress and Prospects.” The sessions, held in November 2000, focused on several key questions: What types of individuals have left the welfare rolls, and how have they fared since leaving? How much of the decline in the welfare caseload is attributable to a strong economy and how much is due to reform per se? How is welfare policy being implemented in New York and the nation? Finally, what new avenues are available to policymakers to encourage welfare recipients to find steady employment? More than 100 academic researchers, government officials, practitioners, and advocates for the poor participated in the day’s discussions. Stephen V. Cameron is an associate professor of economics and public affairs at Columbia University; Robert A. Moffitt is a professor of economics at Johns Hopkins University; Carol Rapaport is an economist at the Federal Reserve Bank of New York. The views summarized are those of the presenters and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. How Have Welfare Leavers Fared? A central concern of policymakers and welfare researchers is how people fare after departing the welfare rolls. Conventional wisdom holds that those women who initially left welfare after passage of the 1996 reform act were likely the most work-ready and should have done comparatively well. Conversely, those women remaining on the rolls and who might leave in the future probably have fewer job skills, less work experience, and a more dire prognosis for economic self-sufficiency. In the day’s first session, Pamela Loprest drew on the National Survey of America’s Families, a representative survey of U.S. households conducted in 1997 and 1999 that focuses on low-income families and the impact of welfare reform. Loprest compared individuals who left the welfare rolls in 1995-97 with those who left in 1997-99. Contrary to expectations, she found FRBNY Economic Policy Review / September 2001 3 that the 1997-99 leavers worked about the same, earned a bit more, and stayed with their employers significantly longer than the 1995-97 leavers did. After excluding Medicaid, food stamps, and money received by way of the earned income tax credit (EITC), Loprest found that median monthly family income, in inflation-adjusted 1999 dollars, was $1,306 among the later leavers and $1,204 among the earlier leavers after one year off welfare. However, many former recipients in both groups were found to have experienced economic difficulty: 40 percent of the earlier leavers and 50 percent of the later leavers reported problems paying mortgage, rent, or utility bills. Effects of the Economy versus Those of Welfare Reform The second session examined the relative influences of the business cycle and welfare reform. One could argue that a strong economy should be expected, by itself, to reduce the welfare caseload by removing the most job-ready families from the rolls and leaving behind those with fewer skills. Welfare reform should also be expected to reduce the caseload, but whether it pulls or pushes off the rolls the more skilled or less skilled individuals is an issue requiring further examination. Other forces—such as an increase in the generosity of the EITC and a rise in the minimum wage—could also be responsible for changes in the caseload and characteristics of those remaining on welfare. One study, by Rebecca Blank, separated the effects of the business cycle and welfare reform on the decline in the caseload, while another, by Robert Moffitt and David Stevens, examined how these forces affected the types of women who remained on the rolls. Blank found that changes in both macroeconomic factors and welfare policy were important contributors to the 50 percent decrease in the caseload. Positive macroeconomic forces, according to the author, explained between 25 and 50 percent of the caseload change in the early and mid-1990s. Moreover, the effects of welfare policy were fundamental in explaining the shrinking caseload in the postreform period. Blank noted as well that a 1 percent rise in the national unemployment rate historically has increased the caseload by 6 percent. Looking ahead, she expected that women who had relied on public assistance during periods of joblessness would now have to depend on unemployment insurance and other sources of aid. 4 Summary of Observations and Recommendations In their study, Moffitt and Stevens concluded that the skill levels of individuals on welfare are affected by the business cycle. Movements in the wage rates of recipients are countercyclical, they observed, because women with the greatest earnings potential tend to leave the welfare rolls, or not enter them, during upturns. Thus, the caseload has tended to become more job-disadvantaged during a strong economy. Nevertheless, the authors argued that welfare reform itself has had little impact above and beyond the effects of the declining unemployment rate, suggesting that the measures have pulled more job-ready and less job-ready individuals off the rolls in equal numbers. Moffitt and Stevens’ examination of trends in the types of individuals remaining on welfare in Maryland produced findings consistent with these national results. Their findings are also somewhat consistent with those of Loprest, who found that the types of individuals leaving the rolls had not changed much over time, although she did not separate the effects of the economy from those of welfare reform. Administering Welfare Policy in New York and the Nation Individual states and localities have a great deal of discretion in designing new public assistance programs. If variations in the individual programs do in fact lead to variations in outcomes, researchers can determine which programs are the most successful. LaDonna Pavetti and her coauthors began the third session by examining the organizations that act as intermediaries between the welfare system and employers. The increased emphasis on moving families into the workforce has led many welfare administrators to contract out this responsibility to for-profit and not-for-profit intermediaries. Yet critics have expressed concern that the intermediaries strive to find employment for only the most job-ready women because the agencies are often presented with cash incentives for the number of clients placed. As the basis of their study, Pavetti et al. drew on a unique data set of 120 intermediaries as well as conducted on-site interviews with welfare administrators, intermediaries, and employers. They found no evidence of intermediaries systematically targeting the most employable women and placing them in jobs. Moreover, successful welfare administration can differ greatly by site: some welfare offices used a single intermediary while others used many, and the contracts between offices and intermediaries varied in many of their details. The authors concluded by stressing the importance of clear information channels between intermediaries and welfare offices in effectively linking recipients with jobs. Next, Howard Chernick and Cordelia Reimers considered the consequences of welfare reform in New York City. They compared several thousand city households eligible for cash assistance, Medicaid, and food stamps in 1994 and 1995 with a post-reform group of welfare-eligible households in 1997 and 1998. Chernick and Reimers found a 33 percent drop in the cash assistance caseload between the two periods, as well as a modest decline in food stamp receipt, from 17 to 15 percent. In addition, Medicaid participation was found to be unchanged, as was the percentage of households using at least one of the three programs. Consequently, despite a strong economy and an administrative push to get people off public assistance, the authors concluded that there was no large drop in the number of New York City households receiving at least some benefit from social programs in the immediate aftermath of welfare reform. New Policies A primary goal of welfare reform is to move recipients into work and toward economic self-sufficiency. To fulfill that goal, policymakers have provided various incentives to promote paid employment, such as the imposition of sanctions and work requirements, the enforcement of lifetime eligibility limits, and the increased use of earnings disregards. An earnings disregard allows recipients to earn money without experiencing a complete reduction in benefits; Connecticut, for instance, disregards all earnings up to the poverty level. However, earnings disregards typically encourage more parttime work than full-time work. Another program, the EITC, has enabled individuals to work off the welfare rolls by supplementing their earnings. Yet the EITC is not restricted to full-time work, so it can also be used to subsidize part-time employment. In the day’s final session, Philip Robins and Charles Michalopoulos considered a program designed to encourage full-time work. Using data from three welfare-to-work demonstration projects, the authors predicted the effectiveness of a financial incentive program, similar to Canada’s SelfSufficiency Project, that would provide assistance only if an individual works at least thirty hours a week. Robins and Michalopoulos estimated that such a program would lead to a sizable increase in the number of welfare recipients working full-time, at only a modest cost to the government. Future Directions The conference offered a great deal of information on the effects of welfare reform. Women who have left the rolls since reform began, for example, have experienced fairly high employment rates, and the earnings obtained through this work essentially have replaced any lost welfare benefits. Moreover, although the incomes of these women generally have not been any higher than they were while on welfare, neither have they been any lower. In addition, the value of the earned income tax credit has become evident from the way in which the program’s supplements have boosted the total income of ex-recipients. An important caveat to these findings, however, is that there is still a subgroup of disadvantaged women who experience significant hardship after departing the welfare rolls. Accordingly, researchers and policymakers still face some important unresolved issues associated with welfare reform. For instance, how does one address the problems of those women who do not thrive off welfare? The effects of alternative sanction policies—which, as currently constituted, appear to affect mainly the most disadvantaged welfare recipients— would also benefit from additional review. Another key policy issue is how to increase the amount of full-time work through financial incentives while not withdrawing support for those who can only work part-time. Finally, consideration of the effects of future economic downturns must be high on the agenda. These and other issues will no doubt play a central role in the fiscal year 2002 congressional and public debates over the reauthorization of the Personal Responsibility and Work Opportunity Reconciliation Act. The views summarized are those of the presenters and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever. FRBNY Economic Policy Review / September 2001 5 Pamela Loprest How Are Families Who Left Welfare Doing over Time? A Comparison of Two Cohorts of Welfare Leavers Introduction O ne of the stated purposes of the Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA) of 1996, popularly known as welfare reform, was to “end the dependence of needy parents on government benefits by promoting job preparedness, work, and marriage.” To this end, this federal legislation, along with many other changes in state policies before and after passage, has increased incentives and requirements for families receiving benefits to move into work and eventually off welfare. The major cash assistance program for poor families is now named Temporary Assistance for Needy Families (TANF), reflecting the goal that receipt of cash assistance from the government should be a temporary situation for families. After passage of PRWORA, concerns began to grow about the effect of welfare policy changes on family well-being. These concerns were heightened by the large declines in welfare caseloads—more than 50 percent nationally from 1994 to 1999—and the claims by some that this meant that welfare Pamela Loprest is a senior research economist at the Urban Institute. reform was a success. Although there have always been families leaving the welfare rolls, these recent policy changes have done more to explicitly “create” leavers, mainly through stricter sanctions for failure to meet program requirements and the institution of time limits on benefits receipt. To address these concerns, a number of state and local welfare agencies as well as some independent researchers began conducting what have come to be known as leaver studies. These studies examine outcomes for families who left welfare over a certain period of time. Early results from these studies showed that a majority of leavers were working and that their wage rates were the same or higher than other similar groups in the labor market.1 Although results were not all positive (many leavers were not working and few had escaped poverty), it seemed that the goal of increasing work was being met. However, a cautionary note in interpreting these results, pointed out by many, was that future groups of leavers may not fare as well and that these early results may not be representative of future results. For example, if recipients who can most easily find work leave welfare more quickly, future This paper was funded by the Urban Institute’s Assessing the New Federalism Project, a multi-year project designed to analyze the devolution of responsibility for social programs from the federal government to the states. The project has received major funding from the Annie E. Casey Foundation, the W. K. Kellogg Foundation, the Robert Wood Johnson Foundation, the Henry J. Kaiser Foundation, the Ford Foundation, and the David and Lucile Packard Foundation. The author would like to thank Donald Alderson for his assistance with this research, as well as acknowledge the helpful comments of her conference discussant, Hilary Williamson Hoynes, and the comments of other conference participants. The views expressed are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. FRBNY Economic Policy Review / September 2001 9 cohorts could possibly have higher numbers of recipients with obstacles to work, such as inferior job skills and experience. Now, four years after passage of these welfare program changes, many additional efforts are under way to assess and evaluate whether the goals of reform have been met and how these policy changes have impacted families. Leaver studies have also progressed, in terms of the number and quality. The U.S. Department of Health and Human Services’ Office of the Assistant Secretary for Planning and Evaluation (ASPE) provided funding to fourteen states and local areas to conduct studies of families who left the welfare rolls, providing technical assistance to help bolster quality and enhance comparability. Results of these studies are now being released.2 This study is also a “leaver study”—describing the economic well-being of families who left welfare and using the National Survey of America’s Families (NSAF), conducted by the Urban Institute. It adds to the body of leaver studies by presenting a national picture, providing context for the individual state and local study results, and giving a sense of outcomes on average across the fifty state “experiments” in welfare policy. An initial study of welfare leavers using these data was carried out recently (Loprest 1999); that study presented results for families leaving welfare between 1995 and 1997, compared with other low-income families with children. This paper focuses on a comparison of outcomes for these early leavers with a more recent cohort of those leaving welfare between 1997 and 1999. It addresses two questions: Data and Definitions The data for this paper are drawn from the NSAF, a nationally representative survey of the civilian, noninstitutionalized population under sixty-five and their families. Two rounds of interviews using essentially the same instrument have been conducted. The first was between February and November 1997 and the second was between March and October 1999. These rounds provide two cross-sectional samples. The survey collected economic, health, and social characteristics for about 44,000 households, oversampling households with incomes under 200 percent of poverty and households in each of thirteen targeted states. The survey’s oversample of lowincome families generates a larger sample size of welfare leavers than most national surveys.3 My definition of leavers includes those who reported receiving welfare at some point in the two years prior to the interview and also reported that they stopped receiving benefits at some point in this same time period. Some of these leavers were also receiving TANF benefits at the time of the interview, meaning that they left the program and then returned. For much of the study, I focus on the subset that has not returned to TANF. The total unweighted sample of welfare leavers is 1,771 in the 1995-97 cohort and 1,206 in the 1997-99 cohort.4 All of the results reported in this paper are weighted. • Do the characteristics of leavers in the later period differ from the earlier period? • Are leavers in the later group doing better or worse economically than the earlier leavers? The paper is organized into the following sections. In the first section, I describe the data used and my definitions. The next section discusses the characteristics of leavers in the 199799 cohort and how they differ from the earlier 1995-97 cohort. The remainder of the paper examines the question of whether leavers in the later cohort are doing better or worse economically than the earlier cohort of leavers. I describe economic well-being by examining employment and job characteristics. I also examine whether the use of nonwelfare government benefits seems to have changed. Finally, I document leavers’ experiences of material hardship and whether this has changed compared with the earlier cohort of leavers. 10 How Are Families Who Left Welfare Doing over Time? Has the Composition of Welfare Leavers Changed over Time? The concern that newer cohorts of welfare leavers may fare progressively worse in the market as the time since passage of welfare reform increases stems in part from the idea that the most “job-ready” left welfare first. This, in turn, would mean that more of the remaining recipients have barriers to work. However, the implications of this hypothesis, if it is true, for the composition of cohorts of leavers is not clear. More recipients with barriers to work could mean fewer recipients leaving. This smaller group of leavers may look similar to the earlier group in its characteristics, if we believe that only those with a certain level of job readiness will leave. However, differences could be introduced because of the existence of time limits and work sanctions that can compel exit, regardless of barriers to work. Since time limits are being reached in some states during the period of the second cohort we study and since use of full family sanctions also increased over the 1995-99 period (U.S. General Accounting Office 2000), it is possible that the second cohort of leavers is composed of fewer job-ready former recipients on average. Caseloads continue to decline every year over the 1995-99 period, with some moderation toward the end of the period.5 The size of my leaver group also declines between the first and second cohort—from 2.1 million who left between 1995 and 1997 to 1.6 million who left between 1997 and 1999.6 Before examining whether characteristics associated with work differ across these cohorts, one important factor needs to be considered: the extent to which former recipients in both cohorts have returned to TANF. Returning to the TANF program is in itself an indicator of economic well-being and success (or lack of success) in transitioning from welfare to work. In the early cohort of leavers, by the time of the interview in 1997, 29.1 percent of former recipients were again receiving TANF benefits.7 For the second cohort of leavers, fewer returned to TANF, with 21.9 percent receiving benefits at the time of the interview in 1999. Fewer returns to TANF could signal that leavers in the second cohort are doing better than those in the first cohort. It could also be a reflection that as families grow nearer to “using up” their time-limited TANF benefits (or have already exhausted benefits), fewer are opting to (or are able to) return.8 Because TANF receipt affects the probability of outcomes such as work and receipt of other sources of income, the fact that fewer of the second cohort are receiving benefits could lead to differences in outcomes between the early and later groups of leavers. In order to focus on differences beyond returns to TANF, the rest of this paper compares subsets of the two leaver cohorts who were not receiving TANF benefits at the time of their respective interviews. The two groups of leavers studied here are made up of those leaving welfare over a fairly wide time frame. Although both cohorts are defined in the same way, a possible difference between them is the weighting of time since leaving welfare. However, I find that of former recipients who have not returned to welfare, the distribution of time since exiting is similar across cohorts, weighted, in both cases, more heavily toward those who left welfare in the past year (Chart 1). In both cohorts, about a quarter left welfare in the three months prior to the interview. Close to an additional third left welfare between three and twelve months prior to the interview. The rest exited TANF more than a year ago. For the most part, characteristics of leavers are similar across these two cohorts (Table 1). The ages, sex, and race of the two groups are not significantly different. More recent leavers have slightly fewer children and slightly younger children than the earlier cohort, although the distribution is not significantly different. They are somewhat more likely to have an unmarried partner, but the percentages who have never married are similar. Education levels across the two groups are also broadly similar, with a slightly higher percentage of the recent group having some years of college. The only characteristic that is significantly different is the indicator that an individual has a physical, mental, or other health condition that limits the kind or amount of work he or she can do. In the second cohort, a greater number of leavers, 22.1 percent, report having this health issue than the first cohort (15.8 percent). Given that the percentage of current recipients with health problems has not increased significantly from 1997 to 1999, this suggests a greater likelihood of exit for those with health problems.9 Chart 1 Former Welfare Recipients Who Have Not Returned to the Program, by Months since Having Left 1995-97 and 1997-99 Cohorts Percent 45 40 1995-97 1997-99 37 39 35 30 25 26 23 20 18 20 15 15 12 10 7 5 0 0-3 4-6 7-12 12+ Months since having left welfare 6 DK/RF Source: Author’s calculations, based on the National Survey of America’s Families. Notes: None of the differences between groups is significant at p<.10. DK/RF is don’t know/refuse to answer. FRBNY Economic Policy Review / September 2001 11 Are More Recent Welfare Leavers Better or Worse off Economically? Table 1 Characteristics of Former Welfare Recipients Who Have Not Returned to the Program 1995-97 and 1997-99 Cohorts (Percent) Former Recipients, 1995-97 Former Recipients, 1997-99 Sex Male Female 6.5 93.5 5.5 94.5 Age 18 to 25 26 to 35 36 to 50 51 to 65 30.5 44.0 23.5 1.9 28.6 40.0 29.1 2.5 Race Hispanic White Nonwhite, non-Hispanic 13.1 52.2 34.7 14.0 50.4 35.6 Number of children in family One Two Three More than three 31.5 35.1 19.7 13.6 33.5 32.4 19.4 14.8 Characteristic Age of youngest child in family Less than three years old Between three and six years old Six to twelve years old Thirteen years or older Marital status Married Unmarried partner Widowed/divorced/separated Never married Married spouse not interviewed Education Less than high school GED or high-school diploma Some college College degree Don’t know/refuse to answer/ not available Condition that limits worka Memo: Sample size Moving recipients into employment is a primary goal of the welfare legislation and an important factor in making the transition to self-sufficiency. In the more recent cohort of welfare leavers who have not returned to welfare, a slightly higher percentage are working than in the earlier cohort, 64.0 percent versus 61.3 percent (Table 2).10 This masks a larger, but still not significantly different, change in the employment rates of single-parent leavers, which increased from 65.6 percent to 71.0 percent across the cohorts. If we broaden the definition of work to include those former recipients who are not currently working but have recently worked (in the year of the interview—on average, the last six months), the percentage increases slightly. An additional 8.6 percent of the early group of leavers and 10.8 percent of the more recent leavers have worked recently (Table 2, bottom section). A recipient leaving welfare to work (or continuing work at higher earnings) is an oft-cited model of how to transition off Table 2 41.8 25.4 25.9 6.9 43.1 20.8 30.4 5.7 26.8 10.6 29.8 31.6 1.3 24.4 15.4 26.6 31.8 1.6 28.9 37.2 27.3 6.0 29.2 33.8 30.9 5.7 0.6 0.3 15.8 22.1 1,289 987 Employment of Former Welfare Recipients Who Have Not Returned to the Program 1995-97 and 1997-99 Cohorts Employment Measure Percentage employed All former recipients Single-parent former recipients Former recipients with spouse/partner Former recipients or spouse/partner in two-parent families All familiesa Percentage of former recipients not currently employed but recently employed (in year of interview) All former recipients Former Recipients, 1995-97 Former Recipients, 1997-99 61.3 65.6 54.0 64.0 71.0 53.7 89.4 74.5 90.2 78.6 8.6 10.8 Source: Author’s calculations, based on the National Survey of America’s Families. Note: None of the differences between groups is significant at p<.10. Source: Author’s calculations, based on the National Survey of America’s Families. a 12 The two groups are significantly different with p<.10. How Are Families Who Left Welfare Doing over Time? a Includes all former recipient families: employment of former recipient for single-parent families and employment of either former recipient or spouse/partner for two-parent families. welfare. However, even when a former recipient is not working, a family can be relying on the earnings of a spouse or partner. This is important, since a large percentage of former recipients (more than a third) are married or have an unmarried partner in both cohorts. In former recipient families with spouses or partners, the family employment rate (at least one of the two people working) is much higher, about 90 percent. This did not change between the two cohorts. Overall, this means that about 75 percent of former recipient families have at least one parent currently working; the figure is even higher for the second cohort (79 percent). The more recent cohort of leavers is working the same or to an even greater extent than the earlier cohort. Even with similar numbers of leavers working, it is possible that the jobs that the later cohort holds are of a lesser quality than those held by the earlier cohort. The first indicator of job quality is the hourly wage. Hourly wages for the 1997-99 cohort of leavers are similar to the hourly wages of the 1995-97 cohort of leavers across the wage distribution. Adjusting for inflation, median hourly wages for the later cohort are $7.15, compared with $7.08 for the earlier cohort (Table 3).11 Total earnings could be affected by a change in the hours that employed leavers work, but there is no significant difference in work effort among the employed across the two groups. In the newer cohort, 67.5 percent of employed recipients are working thirty-five hours or more, compared with 69.4 percent of recipients in the older cohort. The difference is not statistically significant. A slightly greater number of former recipients in the second cohort work multiple jobs, although again this is not statistically different. A similar percentage of former recipients in the two cohorts work in the private and government sectors. There is a small shift (again not statistically significant) within the private sector toward nonprofits, from 4.9 percent to 8.9 percent, but this is still a relatively small group of workers. Working mainly at night or on variable shifts can make finding child care difficult. There is no significant change in the percentage working mainly the day shift, from 71.8 percent to 73.2 percent. But these statistics mean that more than a quarter of employed former recipients are working more difficult night schedules. In two-parent families, some mothers may work night hours while a spouse or partner works day hours as a way of coordinating work and child-care needs. The survey asked whether spouses or partners worked different hours so they could take turns caring for their children. The percentage making these arrangements decreased from 62.4 percent in the first cohort to 53.4 in the second cohort, although this difference is not statistically significant.12 Time working for the current employer reflects a level of employment stability and can be related to higher wages. Contrary to the hypothesis that more recent leavers are less jobready, many more of the recent cohort of leavers have worked for more than two years at their current job, 18.4 percent versus Table 3 Job Characteristics of Employed Former Welfare Recipients Who Have Not Returned to the Program 1995-97 and 1997-99 Cohorts Job Characteristic Former Recipients, 1995-97 Former Recipients, 1997-99 Hourly wagesa 25th percentile Median 75th percentile $5.71 $7.08 $8.71 $6.05 $7.15 $9.00 Hours of work Less than 20 20 to 34 35 or more 6.1 24.5 69.4 8.7 23.8 67.5 8.0 10.1 Class of work Government Private company Nonprofit organization Self-employedb 11.4 76.9 4.9 6.8 11.0 73.3 8.9 6.8 Mostly work between 6 a.m. and 6 p.m. 71.8 73.2 Coordinated schedule with spouse for child carec 62.4 53.4 Time at current employerd Less than six months Six months to one yeare One to two yearse More than two years 31.2 42.8 16.2 9.7 32.8 33.4 15.4 18.4 Multiple jobs (two or more) Source: Author’s calculations, based on the National Survey of America’s Families. Notes: All figures are percentages, except where indicated. Numbers may not add up to 100 percent due to rounding or in some cases a small percentage of “don’t know” or “refuse” answers. a 1997 wages are reported in 1999 dollars using the CPI-X. Includes a small number without a regular employer who work only occasionally. c Asked only of two-parent families with both parents working and a child under thirteen. d Excludes the self-employed. e The two groups are significantly different with p<.10. b FRBNY Economic Policy Review / September 2001 13 9.7 percent. While the same percentage of leavers have worked at their job for less than six months in both groups, a smaller percentage of the recent leavers have been with their employer in the six-months-to-a-year range. These differences are statistically significant. This may be a reflection of the increasing number of women working while on welfare, some of whom may have continued on the same job after exiting welfare. Sources of Support after Leaving Welfare The most common measure of economic well-being, particularly for low-income families, is the percentage with incomes below the poverty level. I do not calculate a measure of total income or the percentage in poverty here because all sources of income are not available for the current time period, only for the past year. Since many leavers recently left welfare and therefore spent part of the previous year receiving benefits, last year’s income would not represent income after exiting. Instead, I examine in this section the total earnings of families and their receipt of other public benefits, in particular food stamps and Medicaid. Examination of earnings at least allows us to compare whether income from work is changing over time. Receipt of food stamps and Medicaid, although not traditionally counted as part of income, can add to family economic well-being, sometimes substantially.13 Putting together hourly wages and the usual amount of work of former recipients and their spouses/partners, I calculate the total monthly earnings of former recipient families with at least one employed adult. This is only a portion of many families’ total income, because they may have other sources of income and these amounts do not include the earned income tax credit for which most of these families are eligible. The median total family monthly earnings for the 1997-99 cohort is $1,360, only slightly higher than and not statistically different from the median of the earlier cohort of $1,204 (Chart 2).14 If work effort remained the same over the course of a year, this median would represent annual earnings of $16,320 for the recent cohort. However, most evidence from other research on low-income workers and other leaver studies shows that work effort is not stable over time. Thus, annual earnings are likely to be lower. Most welfare recipients receive food stamp benefits and many former recipients remain eligible. However, it has been 14 How Are Families Who Left Welfare Doing over Time? well documented that receipt of food stamp benefits drops off precipitously when families leave welfare (Zedlewski 1999; U.S. Department of Agriculture 1999). Food stamps can add substantially to family incomes. For example, in 1999, a single parent with two children and a full-time minimum-wage job would receive $260 per month in food stamps.15 For both cohorts of leavers discussed here, less than a third were receiving food stamps at the time they were interviewed, 31 percent in the early cohort and 29 percent in the later cohort (Chart 3). We might expect that those who have left welfare more recently may be more likely to receive food stamp benefits, and that as time since leaving increases former recipients are less reliant on benefits. This could happen if eligibility for food stamps declined over time because incomes are increasing. For both cohorts, the percentage of those who left in the past year receiving food stamps is higher than the percentage who left more than twelve months ago. For the recent group of leavers, 33 percent of those who left in the past year are receiving food stamps, compared with 25 percent of those who left more than a year ago. Medicaid is also a benefit that can greatly increase the wellbeing of families leaving welfare, since many low-wage jobs do not provide health insurance coverage. Again, most welfare Chart 2 Total Monthly Family Earnings of Employed Former Welfare Recipients Who Have Not Returned to the Program 1995-97 and 1997-99 Cohorts Dollars 2,500 1995-97 1997-99 2,000 1,500 $2,002 $2,080 $1,361 $1,204 1,000 $872 $960 500 0 Twenty-fifth percentile Median Seventy-fifth percentile Source: Author’s calculations, based on the National Survey of America’s Families. Notes: Earnings include those of the former recipient and spouse/ partner where at least one of them is working. All figures are in 1999 dollars. None of the differences between groups is significant at p<.10. Chart 3 Food Stamp Receipt by Former Welfare Recipients Who Have Not Returned to the Program, by Months since Having Left 1995-97 and 1997-99 Cohorts Percent 50 1995-97 40 1997-99 38 33 31 29 30 26 25 20 10 0 Total Less than twelve Twelve or more Months since having left welfare Source: Author’s calculations, based on the National Survey of America’s Families. recipients are covered by Medicaid and many continue to be eligible after leaving. Employed former recipients are eligible for transitional Medicaid benefits up to certain income and time limits. Expansions for children and the implementation of the Children’s Health Insurance Program (CHIP) in individual states have extended nonwelfare-related coverage to even higher income levels for children. However, only about a third of former recipient adults in both cohorts report having Medicaid coverage (Chart 4). This percentage is significantly higher for children, with 44 percent of the early cohort and 53 percent of the later cohort having coverage. The increase for children is likely related to the CHIP expansions and outreach efforts around these programs. Many former recipients remain uninsured. Forty-one percent of the adults in our early cohort and 37 percent of adults in our later cohort are uninsured. Given the increases in Medicaid, less children are uninsured in the later cohort, 17 percent, compared with 25 percent in the earlier group. Notes: The total includes all former recipients who have not returned to welfare. None of the differences between groups is significant at p <.10. Measures of Material Hardship Chart 4 Medicaid Coverage and No Insurance Coverage for Former Welfare Recipients Who Have Not Returned to the Program 1995-97 and 1997-99 Cohorts Percent 70 1995-97 60 50 40 1997-99 53 a 44 34 34 41 37 30 25 20 21 10 0 Adult with Medicaid Children with Medicaid Uninsured adults Uninsured children Source: Author’s calculations, based on the National Survey of America’s Families. Notes: Medicaid here includes state children’s health insurance programs. Children with Medicaid refers to the percentage of all children of former welfare recipients who have Medicaid coverage. a Differences between the groups are significant at p<.10. In addition to earnings and sources of income, another measure of economic well-being is whether and how often a family experiences certain material hardships, such as not having enough food or having problems paying the rent. Several questions of this type were asked in the NSAF in reference to the twelve months prior to the survey. Results for these indicators provide evidence, with a few exceptions, that both groups of former recipients are experiencing similar levels of hardship (Table 4). About a third of both groups of leavers say that they have had to cut the size of meals or skip meals because they did not have enough food in the past year. More than half of both groups have worried that food would run out before they received money to buy more. Among the more recent group of leavers, a significantly greater percentage had this worry often, compared with the earlier group of leavers. About half of both groups report that food did not last or that they did not have money for more food at some time in the past year, either often or sometimes. Problems paying rent or utility bills were also an issue for more than a third of both leaver groups. A significantly higher percentage of the more recent group of leavers, 46.1 percent, were unable to pay mortgage, rent, or utility bills in the past FRBNY Economic Policy Review / September 2001 15 year, compared with 38.7 percent for the earlier cohort. A smaller percentage in both groups had to move in with others because of this inability to pay bills, 7.1 percent in the early group and 9.2 percent in the later group. Table 4 Indicators of Economic Struggles over the Previous Year Former Welfare Recipients Who Have Not Returned to the Program (Percent) Former Recipients, 1995-97 Indicator Former Recipients, 1997-99 Had to cut size of meal or skip meals because there wasn’t enough food 33.4 32.7 Worried that food would run out before got money to buy more Often truea Sometimes true 17.9 39.0 25.0 35.1 Food didn’t last and didn’t have money for more Often true Sometimes true 11.8 37.6 14.6 39.9 A time in last year when not able to pay mortgage, rent, or utility billsa 38.7 46.1 7.1 9.2 Moved in with other people even for a little while because couldn’t afford to pay mortgage, rent, or utility billsb Source: Author’s calculations, based on the National Survey of America’s Families. Note: Approximately 1 percent of respondents in 1995-97 and 3 percent of respondents in 1997-99 did not answer the questions on food problems. a b 16 The two groups are significantly different with p<.10. Only asked of those who had an instance when they were not able to pay bills. How Are Families Who Left Welfare Doing over Time? Conclusions Despite concerns that later cohorts of leavers may fare increasingly worse in the labor market, I find relatively little evidence that there has been much change over the two groups of leavers studied here. The characteristics of the two groups are similar except for a larger percentage in the recent group reporting health conditions that may limit work. Despite this difference, employment and characteristics of jobs are also very similar across the two groups. About two-thirds of former recipients are working and three-quarters of families have an adult working (either the former recipient or the spouse/ partner). Wages are at about the same level for the more recent leavers and most are working full-time, as in the earlier group. One difference in work is the experience of the two groups, with a significantly higher number of more recent leavers having been on their job for more time. Receipt of nongovernment benefits is also similar across the two cohorts. About a third of each group are receiving food stamps and about a third of adults are covered by Medicaid. One difference is that a higher percentage of children are covered by Medicaid in the second cohort, potentially from expansions in state child health insurance programs for lowincome families. Finally, measures of material hardship show for the most part similar experiences of problems with food for early and late cohorts of leavers. Overall, there are few differences between these two groups of leavers. On face, these results seem to provide little evidence in support of the hypothesis that as the amount of time since the passage of the Personal Responsibility and Work Opportunity Reconciliation Act increases, subsequent groups of leavers are less “job-ready” and fare worse economically. However, the two groups of leavers are experiencing different labor markets in 1997 versus 1999. Average monthly unemployment rates for the whole labor force fell from 4.9 percent in 1997 to 4.2 percent in 1999. According to the National Survey of America’s Families data, employment at the time of the interview for unmarried women with children and less than a high-school education increased from 42.4 percent in 1997 to 47.9 percent in 1999. A similar increase (58.9 percent to 63.1 percent) was observed for unmarried women with children with less than or equal to a high-school education. Improvements in labor market outcomes over this time period mean that for a similarly job-ready group of former recipients we might expect to observe improvement in outcomes. The fact that we do not observe significant improvements in economic outcomes across leaver groups could indicate that the more recent cohort of leavers is less jobready. It could also indicate that the subset of former welfare recipients among all less educated single women with children did not experience improvement over this time period. We can only conclude from these results that the more recent cohort is not faring worse than the earlier cohort on an absolute level. Beyond this, it is also true that neither group is showing unequivocal success in transitioning off welfare. A relatively large percentage of leavers still have returned to welfare over this two-year time period, and about a quarter are in families without earnings at the time of the survey. Although this more recent group of leavers looks similar to earlier cohorts, the issues raised about the absolute well-being of earlier cohorts and whether some are “falling through the cracks” remain. Further analysis of subgroups of these data will help us to answer some additional questions about the distribution of outcomes for this group. FRBNY Economic Policy Review / September 2001 17 Endnotes 1. For reviews of some of these early studies, see Brauner and Loprest (1999) and U.S. General Accounting Office (1999). Loprest (1999) compares the wage rates of employed leavers between 1995 and 1997 with other employed low-income women with children who had not recently been on welfare and finds that the leavers’ wages were generally higher. 2. Many studies have links on the ASPE’s web page: <http://aspe.hhs.gov/hsp/leavers99/index.htm>. 3. For more information on the NSAF, see Brick et al. (1999). 4. The NSAF questions about current and former welfare receipt are asked of the adult in the family who is most knowledgeable about the children. The samples of leavers are therefore not exactly all adults who left welfare, but one adult per family who reports that he/she or the children received Aid to Families with Dependent Children (AFDC) or TANF at some point in the two years prior to the interview. Since most respondents are the children’s mothers and most AFDC recipients are women, this corresponds closely to a sample of mothers who left welfare. However, some single fathers and a small number of fathers in two-parent families (who are the adults most knowledgeable about the children and reported leaving welfare) are also included. 7. Because the survey does not ask for complete welfare histories, this may understate returns to welfare. Some families may be missed that left in the time period, returned, and left again, such that they are not receiving TANF at the time of the interview. These families are included in my “did not return to welfare” group. 8. Analysis of what factors are most important in predicting returns to TANF and whether they have changed over the two cohorts is being carried out as part of another study using these data. 9. This is supported by the increase in work among current recipients with multiple barriers to work (Zedlewski and Alderson 2001). 10. Working is defined as any positive weekly hours of work at the time of the survey interview. 11. Adjustments for inflation were made using the CPI-X. All wages are reported in 1999 dollars. 12. This question was asked only to two-parent families in which both parents were working and there was at least one child under age thirteen. The percentage of working former recipients meeting this criterion changed only slightly over the cohorts, from 22 percent to 24 percent. 5. Caseload numbers are reported at <http://www.acf.dhhs.gov/ news/welfare>. 6. As in most surveys, the NSAF undercounts TANF receipt compared with administrative data. The NSAF in both rounds finds about 70 percent of welfare receipt in the previous year, similar to the March Current Population Survey. This implies that my weighted count of welfare leavers reported here is also an undercount, although it is difficult to estimate the extent. 18 How Are Families Who Left Welfare Doing over Time? 13. The calculations needed to estimate total income and poverty and the results are presented in another study on this topic (Loprest 2001). 14. Monthly earnings are in 1999 dollars, adjusted using the CPI-X. 15. This assumes that the family has no income beyond earnings, a maximum child-care cost deduction for children older than two, and no excess shelter costs. References Brauner, Sarah, and Pamela Loprest. 1999. “Where Are They Now? What States’ Studies of People Who Left Welfare Tell Us.” Assessing the New Federalism, Brief, Series A, no. A-32. May. Washington, D.C.: Urban Institute. Brick, Pat, Genevieve Kenney, Robin Mc Cullough-Harlin, Shruti Rajan, Fritz Scheueren, and Kevin Wang. 1999. “Survey Methods and Data Reliability.” Assessing the New Federalism, Methodology Report no. 1. Washington, D.C.: Urban Institute. Loprest, Pamela. 1999. “Families Who Left Welfare: Who Are They and How Are They Doing?” Assessing the New Federalism, Discussion Paper no. 99-02. Washington, D.C.: Urban Institute. ———. 2001. “How Are Families that Left Welfare Doing? A Comparison of Early and Recent Welfare Leavers.” Assessing the New Federalism, Brief, Series B, no. B-36. April. Washington, D.C.: Urban Institute. U.S. General Accounting Office. 1999. “Welfare Reform: Information on Former Recipients’ Status.” GAO-HEHS-99-48. Washington, D.C. ———. 2000. “Welfare Reform: State Sanction Policies and Number of Families Affected.” GAO-HEHS-00-44. Washington, D.C. Zedlewski, Sheila. 1999. “Declines in Food Stamp and Welfare Participation: Is There a Connection?” Assessing the New Federalism, Discussion Paper no. 99-13. Washington, D.C.: Urban Institute. Zedlewski, Sheila, and Donald Alderson. 2001. “Before and after Reform: How Have Families on Welfare Changed?” Assessing the New Federalism, Brief, Series B, no. B-32. April. Washington, D.C.: Urban Institute. U.S. Department of Agriculture. Food and Nutrition Service. Office of Analysis, Nutrition, and Evaluation. 1999. “Who Is Leaving the Food Stamp Program? An Analysis of Caseload Changes for 1994 to 1997.” Washington, D.C. The views expressed in this article are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever. FRBNY Economic Policy Review / September 2001 19 Hilary Williamson Hoynes Commentary T he paper by Pamela Loprest provides a rich description of the characteristics and economic well-being of recipients who have left welfare. Loprest’s work differs from other recent “welfare leaver” studies in two important ways. First, while most leaver studies are based on data from a single state, her study uses the Urban Institute’s National Survey of America’s Families (NSAF), thereby offering a national picture. Second, by pooling several years of NSAF data, the study can provide a comparison of the circumstances of different leaver cohorts. There has been some concern that the most able and job-ready welfare recipients would leave welfare soon after the reforms were put in place, leaving the least job-ready on the caseloads. However, prior to this study, there were very few data to support this claim. Loprest’s paper fills a large gap in the literature and provides an important contribution to our understanding of the effects of welfare reform. In this commentary, I summarize the study’s results and discuss what I see as its limitations. Most of my comments apply not only to the Loprest paper, but to leaver studies more generally. However, given the unique nature of the NSAF data, there is an opportunity here to push beyond the usual descriptive analysis that characterizes leaver studies. Loprest compares the characteristics and economic wellbeing of two cohorts of welfare leavers: those leaving in the 1995-97 period and those departing in the 1997-99 period. The main results of her paper are summarized in Table 1. The composition of leavers is found to change very little between cohorts. The leavers are older, more likely to have a married partner, slightly more educated, and have fewer children. But none of these differences is statistically significant. The measures of economic well-being show somewhat mixed results. Changes in employment, earnings, wages, recidivism, and job tenure suggest increases in well-being between the cohorts. However, significant increases in the percentage of leavers citing a work limitation and material hardship suggest substantial decreases in economic well-being. The main goal of leaver studies is to assess the circumstances of welfare recipients after leaving welfare. In the Loprest paper, this analysis is extended to examine how the characteristics of leavers are changing over time. A strength of leaver studies— especially when using survey data like the NSAF data—is that one can design the questionnaire to include a broad list of measures (such as material hardship) that would not be found on the usual household surveys. I think there is a temptation, however, to interpret the results beyond what is valid. For example, since the characteristics and outcomes of the leavers have not changed substantially over time, can we conclude that job readiness has not declined? The answer, simply, is no, because the outcomes (employment, recidivism, earnings) of the two cohorts are taking place in different economic environments. One (probably very important) example of this is the changing labor market. To illustrate, Table 2 presents labor market characteristics for the United States in 1997 and 1999. This short period saw an improved economy and Hilary Williamson Hoynes is an associate professor of economics at the University of California, Davis. <hoynes@ssds.ucdavis.edu> The views expressed are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. FRBNY Economic Policy Review / September 2001 21 dramatic increases in the employment of less educated female heads of households. The changes found here for the second leaver cohort are substantially below the improvement of all less educated women. This could suggest a decline in job readiness. My general point is that the major limitation of leaver studies is that with a changing environment, you are limited in terms of the conclusions that can be drawn from the studies. In particular, one cannot make any causal links to try to explain the differences in outcomes. It is possible, I think, to extend the analysis to address this concern. The NSAF data are not limited to welfare recipients, but are nationally representative. So it is possible to explore this issue using “control groups,” such as all less educated female heads of households. Such an analysis would allow one to compare the change in economic well-being among leavers with a broader population of less educated women. This could be very useful and would provide a context for evaluating the observed changes of former recipients. Finally, a few additional suggestions. First, for comparison purposes, it would be useful to include tabulations for a leaver group in the pre-reform period. Is what we are seeing a leaver effect just like in the past? For example, are recidivism rates in the post-reform period higher or lower than they were in the pre-reform period? Are employment and earnings different? Second, the nature of the Temporary Assistance for Needy Families reforms is very different across states. For example, the earnings disregard has been liberalized in California, leading to large increases in employment while on welfare and somewhat smaller reductions in the caseload (that is, fewer leavers) compared with other states. In order to capture this aspect of reform, it would be nice to supplement Loprest’s work by selecting the sample of persons ever on welfare in the past two years (as opposed to her sample, which is ever left welfare in the past two years). In addition, there is significant value in using as inclusive a measure of family income as possible to examine changes in economic well-being. Here the analysis is limited to family earnings. The reason cited is that the only measure of total family income corresponds to income last year, which may include a period in which the family was on welfare. I think that this is still a useful measure and should be used. In sum, the Loprest paper provides very useful information describing the circumstances of former welfare recipients. It will surely be read and used by researchers and policymakers alike. Table 1 Table 2 Summary of the Loprest Study’s Main Results U.S. Labor Market Characteristics in 1997 and 1999 Variable Change between 1995-97 and 1997-99 Demographic characteristics Age Percentage nonwhite Number of children Percentage with unmarried partner Education Percentage with work limitation + – – ++ + + +* Economic well-being Recidivism Own employment rate Family employment rate Hourly wages Hours worked Job tenure Family earnings Food stamp receipt Material hardship –– + + + – + +* + –– + +* Unemployment Employment rate of less educated single ---/women (twelve years of education or --/-less) with children //////Worked at all last week //////Worked at all last year 1999 Change 4.9% 4.2% –0.7 0.58 0.71 0.64 0.77 +0.08 +0.06 Source: Author’s calculations, based on 1997 and 1999 Current Population Surveys. Notes: The +/– indicates whether the mean of the measure was higher or lower in the later cohort compared with the earlier cohort (++/– – is a large increase/decrease). An * indicates that the difference is statistically significant at the 10 percent level. 22 1997 Commentary The views expressed in this article are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever. Rebecca M. Blank Declining Caseloads/ Increased Work: What Can We Conclude about the Effects of Welfare Reform? I. Three Simultaneous Events I n 1996, Congress passed the Personal Responsibility and Work Opportunity Reconciliation Act, or PRWORA, which substantially restructured public assistance programs. PRWORA gave states almost entire discretion to design and operate cash assistance programs for families with children, reducing the role of the federal government in program operation and regulation. The federal government did continue to help states fund these programs through the newly created Temporary Assistance for Needy Families (TANF) block grant. In addition, the federal government required states to move an increasing share of their caseloads into work and also, for the first time in history, implemented time limits on how long most families could receive TANF-funded assistance. As a result of this legislation, states have made major changes to the structure of their family assistance programs. States have increased the incentives for public assistance recipients to move into work by reducing the rate at which benefits fall as earnings rise, by implementing more extensive job placement welfare-to-work programs, and by reinforcing the message of time limits that cash assistance will come to an end. States have also increased the penalties and sanctions for those who do not comply with work efforts, and have begun serious “diversion” programs aimed at diverting applicants from public assistance in the first place. Different states have Rebecca M. Blank is the Dean of the University of Michigan’s Gerald R. Ford School of Public Policy and the Henry Carter Adams Collegiate Professor of Public Policy. <blank@umich.edu> chosen different “packages” of these policies, so that one must understand the entire mix of policies in order to characterize the welfare programs in any state.1 For instance, states with low benefit-reduction rates—a more generous policy that allows clients to keep a higher share of benefits as they go to work— may offset this generosity with very strict sanction policies for those who do not participate in welfare-to-work programs. States with strong diversion programs may reinforce this “discouraging” effect on caseloads by also implementing short time limits. States with generous welfare benefit levels may run more intensive welfare-to-work efforts in an attempt to move people into work faster. These major policy changes in public assistance programs did not occur in a vacuum, but coincided with two other important changes in the economic environment in the mid1990s. First, the U.S. economy entered a period of strong and sustained growth. Unemployment rates fell to their lowest levels in thirty years, employment grew rapidly, and inflation remained relatively restrained. These economic changes disproportionately helped less skilled workers, cutting unemployment rates among high-school dropouts by more than half. By the late 1990s, unemployment rates among black and Hispanic workers were at all-time lows. As part of this boom, starting in the mid-1990s, wages among less skilled workers also began to rise for the first time in two decades. Average real weekly earnings among full-time The author thanks Lucie Schmidt for excellent research assistance. The views expressed are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. FRBNY Economic Policy Review / September 2001 25 male workers who did not have a high-school degree rose 5 percent between 1995 and 1999, while they rose 4 percent among full-time female workers.2 This combination of rising wages and rising job availability greatly strengthened the incentive to work. The last major change that occurred in the 1990s was the implementation of a series of policy changes focused on increasing the returns to work among less skilled and low-wage individuals. The minimum wage rose from $3.35 at the beginning of 1990 to $5.15 by 1997. Equally important, a series of expansions in the earned income tax credit (EITC) greatly increased the subsidies received by low-wage workers through the tax system. By 1999, a mother with two children working full-time in a minimum-wage job could receive over $3,500 in a refundable tax credit, a substantial addition to her income. A key design issue in the EITC is that one must work in order to receive any EITC benefits and—at least for low-wage labor market participants in low-income families—EITC benefits rise as work increases. As noted in Blank and Schmidt (2001), the combination of the EITC and the minimum-wage changes substantially increased after-tax wages among minimum-wage workers with children. By the late 1990s, a full-time minimumwage worker with two children had an income above the poverty line.3 Any of these three events—the enactment of major welfare reform, the economic expansion, and the expansion in work support programs—should have affected the behavior of less skilled workers and encouraged greater labor force participation. As it happens, all three of these occurred at about the same time, serendipitously producing a very large change in the rewards from, and incentives to, work, particularly among less skilled women. Regardless of cause, the behavioral changes over the 1990s among welfare recipients have been amazingly large. Public assistance caseloads have declined by half since their peak in 1994. (Even the strongest supporters of the 1996 legislation did not dare predict a change this large.) Chart 1 shows the magnitude of change in caseloads over the 1990s, with a sharp increase in the early 1990s followed by an even greater decline in the late 1990s. Workforce participation has increased at the same time. In March 1994, 23 percent of those receiving welfare in 1993 were observed at work. By March 1999, 40 percent of welfare recipients from 1998 were working.4 Labor market participation among single mothers with young children—the group historically most likely to rely on welfare—soared during this time period. Chart 2 plots the labor force participation rates among women by marital status and children from 1989 to 1999. Unmarried women with children under age eighteen have experienced more than a ten-point increase in labor force participation over the 1990s. At the same time, average incomes among less skilled single mothers have increased while poverty among single mothers has reached an historic low (Haskins 2001). Despite substantial declines in public assistance income, earnings have risen to offset the loss of welfare benefits, and income among less skilled mothers has not fallen. While there is evidence of economic stagnation among some of the more economically disadvantaged over the past several years, the majority of less skilled women appear to have higher incomes by the late 1990s.5 The record since the mid-1990s is quite incredible: there have been large and fast reductions in caseloads, increases in work, and overall declines in poverty. The speed and Chart 1 Chart 2 Total AFDC/TANF Caseloads Labor Force Participation Rates for Women by Marital Status and Number of Children Millions of AFDC/TANF households Women Ages Twenty to Sixty-Five 6 Percent 5 80 Single with no children 4 75 3 Single with children under eighteen Married with children under eighteen 2 70 1996 welfare reform 1 0 1976 78 82 84 86 88 90 92 94 96 98 Source: Department of Health and Human Services Administration for Children and Families. <http://www.acf.dhhs.gov/news/stats/3697.htm> Note: AFDC is Aid to Families with Dependent Children. 26 Married with no children 65 80 Declining Caseloads/Increased Work 60 1989 90 91 92 93 94 95 96 97 98 99 Source: Author’s tabulations of March Current Population Survey data. magnitude of these changes have driven researchers to try and understand their underlying causes, focusing on the factors discussed above. II. Evaluating the Effects of Welfare Reform Those who study welfare reform are particularly interested in disentangling the effect of the 1996 legislation from other changes. This is an extremely difficult analysis to undertake, for at least three reasons. First, as noted above, the timing of welfare reform coincides almost perfectly with the last round of minimum-wage and EITC expansions. The mid-1990s was also a period when the economic expansion became more vigorous and wages among less skilled workers started to rise. Whenever such major events occur at about the same time, it is difficult to identify their effects separately. Second, the economic and the policy changes were not just simultaneous, they were also endogenous and intercausal. For instance, a variety of states enacted precursor programs to federal welfare reform under a program that granted states waivers to experiment with stronger work enforcement among public assistance recipients. The states that enacted these programs had higher average unemployment rates than the states that chose not to enact them.6 This strongly suggests that the types of policies adopted following the 1996 legislation are also likely to be differentially chosen in states with different economic environments. Conversely, states that chose to adopt stronger measures that pushed welfare recipients into work faster after 1996 might have affected the wage and employment opportunities for less skilled workers in their labor markets.7 All of this suggests that it will be quite difficult to identify separately the economic versus policy effects. Third, there are likely to be a substantial number of indirect effects arising from economic growth that are hard to measure separately. Not only will the economic expansion increase job availability and earnings among current and past welfare recipients, but it will also increase earnings among their friends and relatives. Boyfriends and family may be more willing to share housing or to share income in good economic times, making it easier for women to leave welfare even if they themselves are not working more or earning more. In addition, the ready availability of jobs almost surely affected the speed and the nature of state design and implementation of welfare programs after 1996. Precisely because they did not have to focus on job availability, states were able to devote more time and attention to new program design and to focus on implementation of these programs. This suggests that the strong economy might have allowed states to move both further and faster as they redesigned their welfare programs. All of these factors make it difficult to identify separately the effects of the 1996 welfare reform. But even if there had been no economic boom and no other policy changes, providing an evaluation of the 1996 welfare reform legislation would be difficult. The federal legislation was implemented at close to the same time in all states. Between September 1996 and July 1997, all fifty states switched to running new TANF-funded welfare plans, with most states inaugurating their new programs within a few months of each other. This makes it difficult to rely on differences in implementation dates across states to identify differential program impacts. In addition, there are timing problems with current evaluations. We only have a few years of post-1996 data currently available. While states announced their new plans within a year of the 1996 legislation, in many of them implementation of these changes was much slower. Many state welfare programs were in flux for a year or two after the 1996 legislation was signed. Evaluating the effects of these programs using data from 1996, 1997, and even 1998 might be misleading, since many aspects of the programs were only partially implemented in these years. An additional concern is that some program changes have not yet fully taken effect. This is most true of time limits. Only a very small number of welfare recipients have currently hit their time limits, but over the next several years many more persons may face them. This may change the behavior of those who are still recipients and will increase the share of involuntary leavers among ex-welfare recipients.8 This suggests that our current evaluations could seem quite inadequate in only a few years. In short, evaluating the effect of welfare reform is inherently difficult by itself, and made even more difficult by the simultaneous occurrence of an economic boom and other policy changes. These caveats are important to keep in mind while reviewing the existing research on welfare reform. All of this research is subject to the problems discussed above. III. The Effects of Economy versus Policy One might validly ask, why try to disentangle these factors at all? If one’s interest is in changes in the well-being of ex- or current welfare recipients, then simply looking at outcomes might be adequate. Indeed, a good deal of the evaluation literature on welfare reform takes this approach. The growing volume of “leaver studies”—studies of the personal and FRBNY Economic Policy Review / September 2001 27 economic circumstances of families who were previously welfare recipients—makes no attempt to separate out causal factors, but simply looks at the work status and income of families at some future point after they have left welfare.9 There are at least two primary reasons for separating economic and policy effects. First, those who are operating and designing policy may validly want to evaluate the direct effects of their efforts. Understanding the comparative effects of different state approaches to the design of welfare programs might provide knowledge that will be useful in the future as states continue to redesign and evaluate these programs. Second, there are very different future implications if the current changes in behavior are primarily due to policy or to economy. If it is structural program changes that have been effective in reducing caseloads, increasing work, and raising incomes, then these effects may be expected to persist in the future. If it is the current economic boom that is the primary cause of these changes, then they may be quite changeable and temporary. For this reason, the question of what is driving caseload declines and work increases has deep political implications. Those who want to claim success for the 1996 legislation are more likely to favor policy-related explanations. Those who are critical of the legislation and concerned about its long-term impact are more likely to favor economy-related explanations. The existing research literature that tries to disentangle policy and economic effects generally suggests that both factors are important, although the relative magnitude of effects varies depending upon the time period and estimation strategy chosen. Table 1 presents the major empirical studies that attempt to separately assess the effects of policy and economy, utilizing data up to 1996. These studies are only indirectly relevant to the evaluation of the 1996 reforms—they focus on caseload changes in the earlier Aid to Families with Dependent Children (AFDC) program and do not go beyond the 1996 welfare reform in their analysis. The welfare reforms that they focus on are the waivers granted to states in the 1992-96 period, which allowed states to run welfare-to-work programs that were more strongly enforced and that covered a larger share of the welfare population. Table 2 summarizes the smaller group of studies that utilize data after 1996 and attempt to explicitly evaluate the 1996 legislation. Most of these studies focus solely on caseload changes, but a few investigate a broader range of outcomes. The approach in most of these studies is to use panel data on state outcomes—typically state caseload numbers from administrative data—and estimate the impact of economic variables—typically state unemployment rates—and policy variables while controlling for state- and year-fixed effects. A number of studies also control for state-specific time trends, or use more complex first difference or lagged dependent variable 28 Declining Caseloads/Increased Work models. The hope is that these extensive controls for fixed and trend effects will substitute for the large number of omitted variables in these regressions, such as differences in political and population characteristics. Among these omitted variables, I should note, is the effect of the labor market policies mentioned above. The federal minimum wage and the EITC changes are not explicitly controlled for in most of these regressions, but because these policies changed everywhere in the same year, they are assumed to be taken up in the year-fixed effects. This may not be fully adequate; for instance, changes in the minimum wage should have greater effects in low-wage states than in high-wage states. To the extent that the minimum-wage changes and the EITC changes are coterminous with welfare changes—and the welfare changes chosen in any state may partially reflect the presence of these policies—the estimated welfare effects may be biased upward due to these omitted variables. The studies in Table 1 identify the impact of welfare reform based on differential timing in the implementation of welfare waivers across states. Most of these studies use data similar to those of the Council of Economic Advisers (1997), or CEA, and reach similar conclusions. They find that economic factors explain somewhere between 25 to 50 percent of the observed change in caseloads. Welfare reform waivers typically explain a smaller share of the caseload change. Blank (2001) and Wallace and Blank (1999) are the only papers that differentiate between the periods of rising versus falling caseloads (that is, the period up to 1994 and the period from 1994 to 1996); other papers look at changes over a time period that spans both increases and declines in caseloads, typically 1993 to 1996.10 These first two papers find that waivers actually explain a negative share of caseload change between 1990 and 1994—that is, the number of caseloads was rising but the waivers should have caused it to fall. These papers suggest that waivers explain 13 to 31 percent of the decline in the 1994-96 period, when caseloads were falling. Ziliak et al. (2000) and Figlio and Ziliak (1999) find stronger economic effects and weaker waiver effects than the other papers listed in Table 1. In part, this must be due to their focus on the 1993-96 time period. The Wallace and Blank results suggest that the impact of waivers over this longer period must be less than their effects over the 1994-96 period, when caseloads declined. Figlio and Ziliak present a series of estimates suggesting that their results are closely connected to their use of first difference and lagged dependent variable models, with extended lags in many of the independent variables. In many of these models, the implementation of waivers has only a one-time effect—in the period when the waiver is adopted—and it is perhaps not surprising that the resulting coefficients are not large. Table 1 Major Research on Caseload Change Using Data prior to 1996 Welfare Reform Study Data Dependent Variable Included Variables Results on Key Variables Council of Economic Advisers (1997) Annual panel of state administrative data, 1976-96 Log (AFDC caseloads) ÷ total population Unemployment rates Waivers Benefit levels State effects Year effects State/year trends Share of caseload change due to economic factors: Levine and Whitmore (1998) Same as CEA Same as CEA Same as CEA, with more detailed data on waivers Economic effects of same size as CEA Waiver states have almost twice the caseload Ziliak, Figlio, Davis, and Connelly (2000) Monthly panel of state administrative data, 1987-96 Log (AFDC caseloads) ÷ female population (ages 15-44) Unemployment rates Employment/population ratios Waivers Benefit levels State effects Time trends (t, t 2, t 3) Estimate first difference and lagged dependent variable models No separate estimates of economic effects alone; Blank (2001) Annual panel of state administrative data, 1977-96 Same as CEA (Also separates this into AFDCchild only, AFDC-UP, and core remaining caseloads) Economic (including unemployment and wages) Program (including waivers and benefit levels) Demographic Political State effects Year effects Share of caseload change due to economic factors: Figlio and Ziliak (1999) Annual panel of state administrative data, 1976-96 Same as CEA Unemployment rates Waivers Benefit levels State effects Year effects State/year trends Dynamic models include first difference and lagged dependent variable models In static models: Share of caseload change due to economic effects: -10% to 36% in 1993-96 Share of caseload change due to waivers: 0% to 24% in 1993-96 reduction, but no difference in unemployment rate Wallace and Blank (1999) Annual panel of state administrative data, 1980-96 Same as CEA Moffitt (1999) Annual panel of state-level data based on March CPS, 1977-95 (Also aggregates data into education and age cells by state) ln (AFDC participation) ÷ total female population ages 16-54 a 24% to 31% in 1989-93 31% to 45% in 1993-96 Share of caseload change due to waivers: 13% to 31% in 1993-96 3% to 5% estimated change in AFDC caseloads due to one-point increase in unemployment rate 66% of change due to economic and seasonal factors in 1993-96 Share of caseload change due to waivers: -9% in 1993-96 2% estimated change in AFDC caseloads due to one-point increase in unemployment rate that lasts five months 29% in 1990-94 59% in 1994-96 Share of caseload change due to waivers:a -22% in 1990-94 28% in 1994-96 5% estimated change in AFDC caseloads due to one-point increase in unemployment rate In dynamic models: Share of caseload change due to economic effects: 18% to 76% in 1993-96 Share of caseload change due to waivers: -7% to 1% in 1993-96 6% to 9% long-run rise in caseloads due to one-point rise in unemployment rate Share of caseload change due to economic effects: Same as Blank 50% for 1990-94 47% for 1994-96 Share of caseload change due to waivers: -13% for 1990-94 22% for 1994-96 5% to 6% rise in caseloads due to one-point rise in unemployment rate Unemployment rate Waivers Benefit levels State effects Year effects State/year trends Reduction in participation due to waivers: -1.7 percentage points among women high-school dropouts, -0.8 to -1.0 percentage points among all women Among high-school dropouts, significant effects on weeks and hours of work; no significant effects on earnings or income WRSHUFHQWDJHSRLQWULVHLQparticipation due to one-point rise in unemployment rate Author’s calculations, not shown in paper. FRBNY Economic Policy Review / September 2001 29 I should note that more of the variation in these studies comes from explaining caseload increases rather than caseload declines, since caseloads rose sharply between 1990 and 1994. One of the striking aspects of these studies is that their combined estimates explain a very low share of the overall variation in caseloads during this time period once year, state, and seasonal effects are excluded. Blank (2001) investigates this at some length and focuses on the mix of programs that are included in the AFDC caseload numbers. She indicates that 57 percent of this caseload increase is due to increases in two- parent welfare-recipient cases (the AFDC-UP program) and in child-only cases (welfare cases in which there is no adult recipient—a category that rose rapidly in the 1990s). These two programs are responsive to quite different factors and are in turn very different from the “core” AFDC program, that is, benefits paid to single mothers and their children.11 Much of the large unexplained rise in caseloads in the 1990s is due to the growth in these two programs. As one might expect, waivers have stronger negative effects on core AFDC caseloads than they do on aggregate caseloads, which are the data used by most Table 2 Major Research on Caseload Change Including Data after 1996 Welfare Reform 30 Study Data Dependent Variable Included Variables Results on Key Variables Council of Economic Advisers (1999) Annual panel of state administrative data, 1976-98 Log (AFDC caseloads) ÷ total population Unemployment rate Waivers and TANF Benefit levels State effects Year effects State/year trends Share of caseload change due to economic factors: Wallace and Blank (1999) Monthly panel Same as CEA of state administrative data, 1980:1-1998:6 Unemployment rate Waivers and TANF State-specific month effects Models estimated in first differences and with lagged dependent variables Estimated caseload change due to economic factors: Grogger (2000) Annual panel of March CPS data, 1979-99 AFDC/TANF participation Unemployment rate Waivers and TANF Benefit levels Demographic State effects Year effects TANF and waivers have identical (negative) effects on Schoeni and Blank (2000) Annual panel of state-level data based on March CPS, 1977-99 (Aggregates data into education and age cells by state) Multiple variables Unemployment rate Waivers and TANF Benefit levels Demographic State effects Year effects State-specific time trends Waivers have a significant effect on AFDC Hill and O’Neill (forthcoming) Microdata from March CPS, 1983-2000 AFDC/TANF participation and employment last week Unemployment rate Waivers and TANF Benefit levels Demographic State effects Time trend State-specific time trends Economic factors have significant effects on both Declining Caseloads/Increased Work 26% to 36% in 1993-96 8% to 10% in 1996-98 Share of caseload change due to waivers: 12% to 15% in 1993-96 Share of caseload change due to TANF: 35% to 36% in 1996-98 20% to 36% in 1990-94 8% to 12% in 1994-98 Estimated caseload change due to waivers: -4% to -5% in 1990-94 26% to 31% in 1994-96 Estimated caseload change due to TANF: 28% to 35% in 1997:1-1998:6 participation, creating a 2.1 percentage point decline (exclusive of time limit effects) Time limits have significant negative effect on participation in families with younger children participation, labor market participation, earnings, income, and poverty rates, as well as marital status TANF has significant negative effects on welfare participation, larger than the effects of waivers TANF has relatively small but significant effects on earnings, poverty rates, and household structure Economic factors fully explain labor market changes in the TANF period welfare participation and employment in 1992-96 and 1996-99 Waivers have significant effects on employment, but not on welfare participation TANF has a significant effect on welfare participation and on employment Stronger effects on more educated single mothers other studies in Table 1. Indeed, the combination of economic, policy, demographic, and political variables in Blank’s study (2001) comes close to explaining fully the caseload changes in the core AFDC program. Table 2 summarizes the studies that include data from the post-1996 period and that try to estimate the effects of both waivers and the 1996 TANF block grant. Of these, the Council of Economic Advisers study (1999) is most comparable to the earlier work. This study essentially updates the 1997 CEA publication, including data from 1996-98. The results indicate that in this post-PRWORA period, the labor market has a smaller effect on caseloads (explaining 8 to 10 percent of the caseload decline), while welfare reform has a larger effect (explaining about 35 percent of the caseload decline). Wallace and Blank (1999), using monthly data and more dynamic specifications, estimate quite similar effects due to TANF over the 1997-98 period. Schoeni and Blank (2000) analyze the impact of the 1996 welfare reforms on a much wider range of variables beyond caseloads, including workforce participation, weeks and hours of work, earnings, income, and poverty rates. They calculate these data by age and education cells within each state and year, aggregating data from the Current Population Survey. They show that the welfare reform effects that they estimate—both for waivers and for the implementation of TANF—are strongest among the least skilled; they argue that this supports their claim that they are measuring the actual effects of policy changes. Schoeni and Blank include an extensive discussion of the problems of estimating the impact of the 1996 legislation in the existing data, and they try several different estimation procedures. Their results suggest that welfare reform in the post-1996 period had a larger negative effect on caseloads than did the earlier state waivers. In contrast, their labor force participation variables are positively affected by waivers, but appear to be largely unaffected by the 1996 reforms. Increases in work appear to be explained entirely by the strong economy after 1996. This is consistent with the idea that the 1996 legislation focused much more on getting people off of welfare—through sanctions, time limits, and diversion activities—while the waivers focused more on running strong welfare-to-work programs. Hill and O’Neill (forthcoming) investigate the determinants of welfare participation and employment using data on single mothers only from the Current Population Surveys. Unlike other research, their study does not aggregate observations by state, but uses the individual microdata, which make it difficult to compare its results with other research. Since the key variables of interest—unemployment rates and policy changes—are state-level variables, using individual-level data is likely to produce much smaller standard errors than in the state panel data analysis of other papers. They also do not include year-fixed effects (they only include time trends), which makes it more likely that changes in minimum wages and in the EITC might be contaminating their other state-level variables. Nonetheless, their results on welfare participation are similar to those of the other papers, indicating that TANF had a larger effect in reducing caseloads than did waivers. Hill and O’Neill also find strong effects of TANF on employment increases after 1996—a very different result than Schoeni and Blank produce. Schoeni and Blank go beyond caseloads and labor force participation to also look at earnings and income effects. They find sizable, positive, but poorly determined effects of the 1996 changes on family income among less skilled women, and significant negative effects of the legislation on poverty rates. Similar to tabulations of income data by Primus et al. (1999), Schoeni and Blank find some evidence that among all female high-school dropouts, those in the bottom part of the income distribution of this very disadvantaged group are not experiencing the same increases as those in the middle and top of the distribution. While Schoeni and Blank probably provide the most extensive evaluation of the 1996 welfare reform legislation to date, they are clear about the limitations of their study. It remains hard to identify an independent effect of the 1996 welfare reform act, given its rapid national implementation. The ongoing changes in these programs throughout the 199698 period also mean that any estimated effects may not reflect the impact of the more mature programs that were emerging by 1999. The authors choose not to decompose overall caseload or work behavior changes into the share due to policy effects versus economic effects, as the earlier literature did frequently, out of a concern that these two effects are quite simultaneously determined and the coefficients on the economic variables reflect far more than the direct effect of these variables on jobs or income. Summarizing the results in Tables 1 and 2, I would identify four major conclusions from these studies to date: 1. Most of the evidence suggests that both economy and policy have mattered; the exact nature of those effects varies across studies and time periods. More dynamic models appear to produce weaker policy effects, a fact that is likely to be related to the specifications used in those models. 2. The caseload increases in the early 1990s were due to a wide variety of factors, including growth in child-only payments and the mandatory implementation of AFDC among two-parent families in all states. As a result, explaining much of the aggregate caseload rise with simple econometric specifications is a difficult task. Waiver effects appear stronger when focusing only on caseloads among single mothers with children. FRBNY Economic Policy Review / September 2001 31 3. The economy seems to have mattered less in the post1996 period and welfare policy has mattered more in reducing caseloads than it did in the earlier period. This is entirely consistent with the fact that the state TANFfunded programs were typically more focused on sanctions, diversion, and time limits than were the waiver programs of the early 1990s. Both the economy and waivers appear to have raised employment in the early 1990s; studies that look at the effect of TANF on employment in the late 1990s show more mixed results. 4. While there is a serious need for more evidence on outcomes other than caseload declines, the existing evidence suggests that both waivers and the 1996 legislation might have had positive effects on income and negative effects on poverty. On average, less skilled female-headed families appear to be better off, and the 1996 legislation seems to be an important causal factor in this, even after controlling for economic effects. Among a group of the most disadvantaged, less skilled female family heads, there is some evidence that incomes have not risen and some evidence that poverty has become worse. At least one study suggests that the 1996 legislation has not had the same effects on this bottom group as on less skilled women as a whole (Schoeni and Blank 2000). IV. Future Research Issues For those who want to understand further the relationship between economy, policy, and observed behavioral changes in the 1990s, several future research projects recommend themselves. First, there may be ways to evaluate the impact of the 1996 policy changes by looking in a more disaggregate way at the policies that different states have implemented. A number of organizations are now regularly collecting and publishing information on state-specific program parameters, and the research community needs to experiment with various ways of describing these very different programs in a quantitative and comparable form. Some of the studies listed in Tables 1 and 2 have tried to estimate separately the effect of different types of program changes, such as time limits or work exemptions, as well as look at the overall effects of implementing welfare reform. Because data on any particular type of program are limited—implemented in only some states and over only limited time periods—many of these estimates are poorly determined. But as we acquire a growing amount of information on changes within states over time, it should be possible to do more to tease out the impact of specific types of policy changes. Second, the most obvious way to study the differential effects of policy and economy is to wait for the next economic 32 Declining Caseloads/Increased Work slowdown and see what changes. I am doubtful we will be able to say anything very conclusive about how much the economy has influenced the changes in the late 1990s until we collect some observations in a world with less robust economic growth and higher unemployment. Third, it is important to note the need for more studies that focus on overall measures of well-being. Too much of the existing work looks just at caseloads, a very limited measure that provides little information about how the less skilled population is faring. Declining caseloads are generally viewed as a good thing, but say nothing about work or income among those leaving public assistance. Increases in workforce participation are generally viewed as a good thing, but these data need to be balanced with information on overall disposable income as families face greater work-related and child-care expenses and lose welfare benefits. It will also be important to explore how unique our current set of results is to the current time period. This is not only a question of separating out the impact of the current economic boom, but also relates to some of the implementation questions raised above. As a growing share of the caseload hits time limits, some state programs may begin to operate differently. More and more states are proclaiming that they have “changed the culture” of their welfare offices, through retraining front-line workers. Most states are still working to better integrate their job placement and training efforts with their welfare efforts. As all of these changes occur, the long-run nature of TANF-funded welfare programs may be different from their operations in the immediate post-1996 period. Whether the long-term effects of welfare reform will be greater or smaller than the short-term effects is hard to predict. Some arguments suggest that the long-term effects may be larger: once recipients begin to hit time limits, there may be bigger effects; if recent state changes induce cross-state migration over time, there may be bigger effects; if states are successful in changing the culture of their welfare offices to make them more employment-oriented, that may result in bigger long-term effects. But the economic arguments suggest that the long-term effects of welfare reform will be smaller: a more typical economy will force states to expend more resources on finding jobs or creating public sector employment, which will take resources out of new programs; women will be less willing to be diverted or to avoid welfare in a higher unemployment economy, and states will again see increases in caseloads. In a slower growth economy, states will feel more economic budget pressure and will be less willing to focus as much time and money on welfare programs. At present, I think it is fundamentally unknowable what the long-term effects of the 1996 law are likely to be. We have seen enormous behavioral changes, including faster exit from and reduced entry onto public assistance; big increases in work; and reductions in take-up rates even among the eligible population. We have also seen enormous program changes as states have greatly modified their old AFDC programs. In the presence of this much program and behavioral change, it is simply hard to know what is permanent and what may change in the next round of program reform or during the next economic slowdown. V. Some Big Remaining Questions I close this paper with a few of the many unanswered questions about future program and behavioral issues that I think will be important in the years ahead. First, assuming that the rapid decline in caseloads is somewhat permanent, this means that the remaining caseloads are more disadvantaged than the AFDC caseloads of ten years ago. For instance, Allen and Kirby (2000) find that a growing share of the caseload is composed of women of color living in center cities, who are more likely to face a host of barriers to finding permanent employment. How will this change the politics of welfare? Will this make voters and politicians even less sympathetic to welfare recipients and lead to harsher measures designed to move these women off of public support? Or will this generate greater sympathy for welfare recipients, as voters realize that those who remain on welfare do face multiple barriers to work? Second, the changes in welfare program design have almost surely made less skilled women—and particularly single mothers—more vulnerable to the economy. As these women rely on earnings for an increasing share of their income, and as they face tighter restrictions on their access to public assistance, they will be more subject to the vagaries of the labor market. It may be that single mothers will become as responsive to the labor market as two-parent families have been in the past. For instance, Blank (2001) suggests that a one-point rise in unemployment raised AFDC caseloads by 6 percent among single mothers, but raised AFDC-UP (the program for married couples) caseloads by 9 to 17 percent. Women may cycle more frequently on and off of welfare, responding to changes in job availability in the private sector. A key question is whether these women will be able to access the unemployment insurance (UI) system during times of joblessness. Policy changes to UI could make it easier for part-time and shortterm workers to access benefits; this may be an alternative way to support work-eligible single mothers when they become jobless. Third, the current economy has allowed a large number of less skilled women (and men) to work more continuously than in previous decades. How will this help these women? Will their growth in labor market experience lead to significant wage growth?12 Will they be able to make contacts and create labor market networks for themselves that make it easier to find jobs in the future if they leave or lose employment? In short, will this extended labor market boom help provide a larger pool of workers who are willing and able to work? This is the positive version of the “hysteresis hypothesis” much discussed in Europe over the past decade, in which extended periods of high unemployment appear to result in more workers permanently disconnected from the labor market. The 1990s economic boom, providing a long-run decline in unemployment rates and time spent out of the labor market, may more permanently connect a group of disadvantaged workers to the labor market. Finally, there are a host of questions about the impact of these program and economic changes on family fertility and formation patterns. Some existing evidence suggests that recent welfare reforms can have an effect on marriage.13 Of course, a stronger economy might also have this effect, as less skilled men appear to be better marriage prospects. We need to move beyond a focus solely on income, labor market, and caseload changes among single mothers to observe how these program and economic changes are related to family formation, to fertility, and to the educational plans of younger, less skilled men and women. A closely connected question is how these changes impact the children in families in which the sole parent is now working and earning more and receiving less public assistance. Preliminary evidence suggests that there may be child-related impacts that vary by age (Duncan and ChaseLansdale 2001). In their evaluations of these programs, economists need to think broadly about program impacts and move beyond their usual set of income and earnings data. The answers to all of these questions will become clearer as time passes and we accumulate additional observations on programs, economic forces, and individual behavior. As such, the results from current evaluations of the impact of the 1996 welfare reforms must be considered highly preliminary. FRBNY Economic Policy Review / September 2001 33 Endnotes 1. For a discussion of the changes occurring inside states, see Nathan and Gais (1999). For information on the nature of the new programs adopted by states, see the data and related descriptive papers provided by the Urban Institute’s Assessing the New Federalism Project. 7. For a review of the evidence on this issue, see Bartik (2000). 2. Data were tabulated by the author from the outgoing rotation groups of the Current Population Survey. While significant, these increases were not enough to overcome the previous fifteen years of wage declines. Full-time male high-school dropouts experienced an 18 percent decline in real wages over the entire 1979-99 period, despite rising wages in the late 1990s. Among full-time female workers without high-school degrees, real wages fell by a smaller amount (5 percent) over the entire 1979-99 period. 9. A good review of studies of welfare leavers is provided by Brauner and Loprest (1999). 3. As part of the 1990s expansion in work support, one might also note the large increase in subsidies for child care over the 1990s. There was also a substantial expansion of Medicaid insurance in the late 1980s to cover all low-income children. 4. Data were tabulated by the author from the March Current Population Survey. 8. For a discussion of the current evidence on the effects of time limits, see Bloom and Pavetti (2001). 10. These papers appear to look at 1993-96 because the CEA report focuses on these years. However, the reason why the CEA report focuses on these years is purely political—the analysis starts in 1993 because this is the first year of the Clinton Administration. 11. All states were mandated to run an AFDC-UP program starting in 1990, and much of the increase in this program was due to new states beginning to serve the two-parent population. The rise in child-only cases is related to the growing use of sanctions (removing the adult from the payment unit), the rising number of immigrants (whose American-born children are eligible for assistance, but whose immigrant parents are not), changes in the structure and functioning of foster care programs, and a rising share of children living in households without a parent present. 5. For a review of the evidence on changes in income since welfare reform, see Haskins (2001). For a review of the evidence on changes in work and labor market behavior, see Blank and Schmidt (2001). 12. Gladden and Taber (2000) indicate that wages grow with experience even among very unskilled women. 6. See evidence in Schoeni and Blank (2000). 13. See Knox et al. (2000) or Schoeni and Blank (2000). 34 Declining Caseloads/Increased Work References Allen, Katherine, and Maria Kirby. 2000. “Unfinished Business: Why Cities Matter to Welfare Reform.” Center on Urban and Metropolitan Policy. July. Washington, D.C.: Brookings Institution. Gladden, Tricia, and Christopher Taber. 2000. “Wage Progression among Less-Skilled Workers.” In David Card and Rebecca M. Blank, eds., Finding Jobs: Work and Welfare Reform. New York: Russell Sage Foundation. Bartik, Timothy. 2000. “Displacement and Wage Effects of Welfare Reform.” In David Card and Rebecca M. Blank, eds., Finding Jobs: Work and Welfare Reform. New York: Russell Sage Foundation. Grogger, Jeff. 2000. “Time Limits and Welfare Use.” NBER Working Paper no. 7709. Blank, Rebecca M. 2001. “What Causes Public Assistance Caseloads to Grow?” Journal of Human Resources 36, no. 1 (winter): 85-118. Haskins, Ron. 2001. “The Second Most Important Issue: Effects of Welfare Reform on Family Income and Poverty.” In Rebecca M. Blank and Ron Haskins, eds., The New World of Welfare. Washington, D.C.: Brookings Institution. Blank, Rebecca M., and Lucie Schmidt. 2001. “Work and Wages.” In Rebecca M. Blank and Ron Haskins, eds., The New World of Welfare. Washington, D.C.: Brookings Institution. Hill, M. Anne, and June O’Neill. Forthcoming. “Gaining Ground? Measuring the Impact of Welfare Reform on Welfare and Work.” Manhattan Institute Civic Report. Bloom, Dan, and LaDonna Pavetti. 2001. “Sanctions and Time Limits: State Policies, Their Implementation, and Outcomes for Families.” In Rebecca M. Blank and Ron Haskins, eds., The New World of Welfare. Washington, D.C.: Brookings Institution. Knox, Virgina, Cynthia Miller, and Lisa A. Gennetian. 2000. “Reforming Welfare and Rewarding Work: Final Report on the Minnesota Family Investment Program.” New York: Manpower Demonstration Research Corporation. Brauner, Sarah, and Pamela Loprest. 1999. “Where Are They Now? What States’ Studies of People Who Left Welfare Tell Us.” Assessing the New Federalism, Brief, Series A, no. A-32. May. Washington, D.C.: Urban Institute. Levine, Philip B., and Diana M. Whitmore. 1998. “The Impact of Welfare Reform on the AFDC Caseload.” National Tax Association Proceedings–1997, 24-33. Washington, D.C.: National Tax Association. Council of Economic Advisers. 1997. “Technical Report: Explaining the Decline in Welfare Receipt, 1993-1996.” April. Washington, D.C.: Executive Office of the President. Moffitt, Robert A. 1999. “The Effects of Pre-PRWORA Waivers on Welfare Caseloads and Female Earnings, Income, and Labor Force Behavior.” In Sheldon H. Danziger, ed., Economic Conditions and Welfare Reform. Kalamazoo, Mich.: W. E. Upjohn Institute for Employment Research. ———. 1999. “Economic Expansion, Welfare Reform, and the Decline in Welfare Caseloads: An Update. Technical Report.” September. Washington, D.C.: Executive Office of the President. Duncan, Greg J., and P. Lindsay Chase-Lansdale. 2001. “Welfare Reform and Child Well-Being.” In Rebecca M. Blank and Ron Haskins, eds., The New World of Welfare. Washington, D.C.: Brookings Institution. Figlio, David N., and James P. Ziliak. 1999. “Welfare Reform, the Business Cycle, and the Decline in AFDC Caseloads.” In Sheldon H. Danziger, ed., Economic Conditions and Welfare Reform. Kalamazoo, Mich.: W. E. Upjohn Institute for Employment Research. Nathan, Richard P., and Thomas L. Gais. 1999. “Implementing the Personal Responsibility Act of 1996: A First Look.” Albany, N.Y.: Nelson A. Rockefeller Institute of Government. Primus, Wendell, Lynette Rawlings, Kathy Larin, and Kathryn Porter. 1999. “The Initial Impacts of Welfare Reform on the Incomes of Single-Mother Families.” Washington, D.C.: Center on Budget and Policy Priorities. Schoeni, Robert F., and Rebecca M. Blank. 2000. “What Has Welfare Reform Accomplished? Impacts on Welfare Participation, Employment, Income, Poverty, and Family Structure.” NBER Working Paper no. 7627. FRBNY Economic Policy Review / September 2001 35 References (Continued) Wallace, Geoffrey, and Rebecca M. Blank. 1999. “What Goes Up Must Come Down? Explaining Recent Changes in Public Assistance Caseloads.” In Sheldon H. Danziger, ed., Economic Conditions and Welfare Reform. Kalamazoo, Mich.: W. E. Upjohn Institute for Employment Research. Ziliak, James P., David N. Figlio, Elizabeth E. Davis, and Laura S. Connolly. 2000. “Accounting for the Decline in AFDC Caseloads: Welfare Reform or the Economy?” Journal of Human Resources 35, no. 3 (summer): 570-86. The views expressed in this article are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever. 36 Declining Caseloads/Increased Work Robert A. Moffitt and David W. Stevens Changing Caseloads: Macro Influences and Micro Composition he unprecedented decline in the caseload of the Aid to Families with Dependent Children (AFDC) program, retitled the Temporary Assistance for Needy Families (TANF) program in 1996, has been, by common agreement, remarkable. The caseload has declined by 50 percent since its peak in 1994 and is now at a level roughly similar to what it was in the late 1970s. It is also generally agreed that welfare reform has played a role in this decline, albeit simultaneously with the effects of the strong economy and of other policy measures. The welfare reform movement that was solidified in the 1996 Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA) actually began in the early 1990s, and contributed to the caseload decline prior to 1996. The economy played a stronger role in that period than did welfare reform. However, subsequent to 1996, the economy has played the lesser role, according to estimates from currently available studies (Mayer 2000; Moffitt 1999; Schoeni and Blank 2000; Council of Economic Advisers 1997, 1999). Also playing a role of rather uncertain magnitude have been expansions of the earned income tax credit and Medicaid eligibility; both of these reforms greatly increased the amount of resources made available to families off welfare. Research studies to date have also examined the effect of welfare reform on employment outcomes and other individual and family outcomes, as well as effects on the caseload. Two types of studies have been conducted. By far, the more numerous have been studies of welfare leavers: women who T have left the AFDC or TANF rolls after welfare reform began. These studies generally have shown leavers to have employment rates in the range of 50 to 70 percent, considerably higher than expected (Brauner and Loprest 1999; U.S. General Accounting Office 1999; Isaacs and Lyon 2000; Acs and Loprest 2001). Unfortunately, these studies do not estimate the effect of welfare reform per se because they do not control for the influence of the economy, which has improved considerably over the same period and could have contributed to these favorable outcomes.1 A second strand of research study examines the effect of welfare reform on employment and other outcomes of all single mothers, or sometimes all less educated women, regardless of their welfare participation status (Moffitt 1999; Schoeni and Blank 2000). These studies control for the state of the economy, and typically have found positive effects on employment and earnings.2 The issue addressed in this paper is how welfare reform has affected the types of women who have remained on the welfare rolls (sometimes called stayers, as opposed to leavers). This group has not been examined by either of the two types of studies just referred to. Yet those women remaining on the rolls are also of policy interest. By and large, it is expected that those women are the most disadvantaged recipients who have not yet been able to find jobs in the growing economy or who have some significant health or other problem that prevents them from being able to leave the rolls or to work. If this is the case, such a disadvantaged group, still in need of a safety net, Robert A. Moffitt is a professor of economics at Johns Hopkins University; David W. Stevens is the Executive Director of the University of Baltimore’s Jacob France Center. The authors would like to thank Robert Lerman, Susan Mayer, and June O’Neill for comments, and John Janak, Eva Sierminska, Sang Truong, and Zhang Zhao for research assistance. The views expressed are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. FRBNY Economic Policy Review / September 2001 37 deserves attention and calls for the development of policies to address its needs. However, as was the case in studies of leavers, ascertaining that more disadvantaged women remain on the rolls does not say whether that is a result of the economy or of welfare reform; a low unemployment rate would also tend to draw women with more labor market skills off the welfare rolls. Determining the net effect of welfare reform requires controlling for the business cycle, as some of the other studies cited above have done for other outcomes. Our analysis is composed of three parts. First, we provide a discussion of what the effects of welfare reform on the composition of the caseload—primarily measured by labor market skill level—should be, in principle. Perhaps surprisingly, we argue that different welfare reform policies have different effects on more skilled versus less skilled recipients, and that the net effect of the policies taken together is mixed and ambiguous. Second, we provide some new evidence from the nation as a whole using Current Population Survey (CPS) data, and from the state of Maryland using administrative caseload and earnings data. Third, we summarize what the few other studies of welfare stayers have shown. Our analysis indicates that, after controlling for the effects of the economy, there is little evidence in national CPS data that welfare reform has affected the composition of the caseload in its labor market skill distribution, indirectly implying therefore that leavers have been equally distributed across all skill types. The analysis of data from Maryland indicates, in addition, a disproportionate effect of welfare reform on long-term recipients on the welfare rolls, who are the most disadvantaged, although not necessarily resulting in their departure from welfare. Other studies comparing leavers with stayers find as a whole that the former are more job-ready than the latter, but this could be the result of the growing economy and is inconsistent with the CPS, which shows a decline in the skill level of the caseload prior to adjustment for the business cycle. On net, therefore, we find no strong evidence that welfare reform per se has been selective in who has left the rolls and who has stayed in terms of labor market skills. I. Expected Effects of Welfare Reform on Caseload Composition The common theory of the main determinants of why some women are on welfare and others are not is based on a standard economic framework, which views welfare participation as resulting from a trade-off between potential income off welfare and potential income on welfare. Holding constant the latter, usually measured by the level of the welfare benefit, women 38 Changing Caseloads with greater income off welfare are less likely to be on the rolls and those with less income off welfare are more likely to be on the rolls. Since labor market earnings are a major source of income off the rolls, this leads to the natural presumption that women with greater labor market skills should be off welfare and those with lesser skills should more likely be on welfare.3 The composition of the rolls over time can be expected to change, according to this framework, if either the benefit level or labor market opportunities off the rolls change. If benefits trend downward, for example, one should expect the caseload to become increasingly disadvantaged in terms of labor market skills, and the same should occur if labor market opportunities improve. The caseload should change in composition over the business cycle as well according to this framework. As the unemployment rate rises, one should expect women with more labor market skills to come onto the welfare rolls and hence the average skill level of welfare recipients should rise. Such women are ordinarily employed but lose their jobs during economic downturns. Likewise, as the unemployment rate falls, one should expect women with greater labor market skills to leave the rolls as they find jobs, leaving the caseload increasingly composed of more disadvantaged recipients. When the features of welfare reform in the 1990s are considered, a more detailed examination is required. The overall emphasis of 1990s welfare reform has unquestionably been to increase employment of welfare recipients, and to this extent one might expect the most employable women to leave the welfare rolls first and the least employable recipients to stay on the rolls and to leave later, if ever. However, there are countervailing pressures at work, as can be seen by a more careful consideration of the main elements of reform: work requirements, sanctions, more generous earnings disregards, and time limits. Work requirements should, at one level, make welfare less attractive in general and should lead some women to leave the welfare rolls. Naturally, the women who can leave most easily are those with greater labor market skills. An important question, however, is whether such requirements lead to work while on the rolls instead of work off the rolls. States that count earnings against the welfare grant, as most do, may make some women who earn sufficient amounts of money from employment ineligible for benefits and hence lead to their departure from welfare. However, those women who do not earn enough to render themselves ineligible will stay on the rolls and will combine welfare and work. The question regarding work requirements is how they will affect those women who have barriers to employment, such as health problems, low levels of education and work experience, or difficulties finding child care. To the extent that these more disadvantaged women are exempted from work requirements, they will be unaffected; but to the extent that they are not exempted (and the tendency in many states is to minimize the number of exempt categories), they will find work requirements more onerous to fulfill. This could lead to an inability to meet those requirements and lead to a departure from the rolls, possibly working in the opposite direction to the main effects of work requirements.4 Sanctions that are imposed for noncompliance with work requirements should, similarly, work toward the departure from the rolls of more disadvantaged rather than less disadvantaged women. Women who are more job-ready and have fewer barriers to work are most likely to be able to comply with work requirements and hence avoid sanctions, while women who have more barriers related to health, child-care problems, or difficulties at home or in their personal life are likely to have a more difficult time complying and hence are more likely to be sanctioned. Indeed, the evidence to date is that women who have left the rolls after being sanctioned have lower employment rates and higher poverty rates than other leavers and are, in general, a more disadvantaged group (Brauner and Loprest 1999; Moffitt and Roff 2000; U.S. General Accounting Office 2000). Thus, sanctions work against the usual presumption that the most advantaged are more likely to leave the rolls. More generous earnings disregards also work against this presumption, at least in relative terms.5 Such disregards have an employment-inducing effect by encouraging women to combine welfare and work and hence to hold jobs while still on welfare. They therefore tend to increase the welfare rolls by discouraging women from leaving welfare for work and encouraging women who might otherwise not have come onto welfare to do so, knowing that they can work while on the rolls. The women most capable of taking advantage of more generous earnings disregards are the more job-ready women who have sufficient education and work experience to find and retain employment. The women least able to take advantage of disregards are those with the poorest work skills and those with the most difficult problems in their personal and family life.6 Finally, the effects of time limits on caseload composition are complex and not easy to predict. In the short run, to the extent that the existence of time limits causes some women to leave the rolls before the time limit is reached, possibly in order to “bank” their benefits, it should be expected that more jobready recipients would be more easily able to find jobs and leave the rolls early. However, in the longer run, as time limits are reached, women who are more disadvantaged will remain on the rolls and will actually be observed to hit the limit and be terminated. At that point, the more disadvantaged women are more likely to leave welfare. States may grant extensions from the time limits to some of these types of recipients as well as use their 20 percent time-limit exemption for such women, thereby ameliorating their impact. But even these short-run and longrun effects depend on the extent to which state policy encourages women to work on the rolls before they hit the limit, and the extent to which such encouragement extends to disadvantaged as well as advantaged women (Moffitt and Pavetti 2000). The more women stay on the rolls to work prior to the limit, the more likely they will still be on the rolls when the time limit is reached. In summary, while the general tendency of welfare reform is to encourage more job-ready recipients and those with more education and work experience to leave the rolls—leaving behind the more disadvantaged women—there are tendencies in the opposite direction as well. Sanction policies and more generous earnings disregards, as well as elements of other policies, will tend to retain more job-ready women on the rolls and/or lead to the departure from the rolls of the more disadvantaged recipients. II. New Evidence Analysis of the Current Population Survey The ideal data set for a study of national trends in the composition of the AFDC and TANF caseloads would be a national data set with information on the characteristics of recipients over several years, including different periods of the business cycle. Many characteristics of families are of interest, including the education, work experience, health, and other characteristics of the single mother herself, as well as the number and ages of her children and their health status; also, information on others in the household and the type of income they can provide. Information on the mother’s history of welfare participation would be useful to determine whether she is a long-term recipient. Unfortunately, such a data set does not exist. Administrative data on recipient characteristics in all states have been collected in a series of changing formats since 1969, providing some information on recipients, but most data are drawn from the AFDC or TANF records themselves, leading to a variable list far shorter than the list noted above. National survey data sets generally are weak as well, often having very small sample sizes of recipients (as in the national longitudinal data sets) or a limited number of years available (the Survey of Program Dynamics). Probably the best national survey for this purpose is the Survey of Income and Program Participation, which has FRBNY Economic Policy Review / September 2001 39 been available since 1984, but it has been very slow to release data, and very little information subsequent to 1996 has yet been provided to the public. A next-best national data set is the March Current Population Survey, which is used here. The CPS has strong advantages for this type of study. It is available back to 1968 on an annual basis and through 1999, and it contains reasonably large sample sizes of single mothers. It is nationally representative and most questions have been consistently asked across the years. It contains information on nonwelfare recipients as well as welfare recipients, which is needed in order to disentangle trends in characteristics that have occurred for all single mothers from those that have been experienced by single mothers on welfare per se.7 However, the CPS has major disadvantages as well. The survey takes place in March of each year and obtains information on earnings, weeks of work, and welfare receipt during the prior calendar year, but respondents are not asked week-by-week questions, which would allow a determination of whether welfare receipt and work occurred at the same time. Most individual and family characteristics are measured as of the March interview, which does not coincide with the time at which welfare participation is measured. The characteristics of the single mother obtained are very sparse, and consist only of the usual crude socioeconomic markers—age, education, and so on. There is essentially no information on the indicators of serious disadvantage that are present in the worst-off portions of the welfare caseload—poor health of mother or children, substance abuse, a history of welfare dependency, very little work history, and so on. Also, the data are not longitudinal in nature, and hence a woman’s movements on and off welfare over time cannot be tracked. Nevertheless, the CPS is used here because it is the only nationally representative data set that has a long enough history to estimate business cycle effects.8 The main characteristics of the single mother that we use to indicate labor market skill are her level of education and the level of hourly wage rates of jobs she has held over the past calendar year.9 Hourly wage rates are the best single indicator of where in the hierarchy of skill in the labor market an individual is located. We also look at other characteristics in the CPS pertaining to family structure and marital status (family size, whether the single mother has ever been married), some other personal characteristics (age, race), and some labor market attachment variables (earnings, weeks of work, employment status). Note that these last three variables do not measure skill per se but rather outcomes that themselves are changed by the business cycle and, possibly, by welfare reform; they are not markers of whether the caseload is becoming more or less disadvantaged in terms of labor market skills. Charts 1 and 2 plot an education measure and the real hourly wage rate for AFDC recipients, respectively, together with the unemployment rate. The education measure is the percentage of recipients who have at least twelve years of education.10 It shows a strong upward trend over the past thirty years, indicating growth in the educational levels of welfare recipients. There is a slight countercyclical pattern in Chart 1, showing a positive correlation between the unemployment rate and the educational level of welfare recipients. The hourly wage rate measure in Chart 2 shows a steady decline from the 1970s to the 1980s, but with a slight recovery starting in the late 1980s and early 1990s. The relationship to the unemployment rate again appears to be roughly countercyclical, with the exception of the early 1980s. There appears to be a slight upward movement in wages after 1996. These charts are misleading, however, for they do not show trends in the single-mother population as a whole. Educational levels, for example, have been increasing for the entire Chart 1 Percentage of Welfare Recipients with Twelve or More Years of Education, 1968-99 Chart 2 Hourly Wage Rate of Welfare Recipients, 1975-98 Percent Percent 14 Twelve or more years of education 12 Scale Percent 0.7 14 0.6 12 10 0.5 10 8 0.4 8 0.3 6 6 Unemployment rate 0.2 4 2 0.1 2 0 1968 70 0 0 1975 4 40 Scale 75 80 85 90 95 99 Changing Caseloads Dollars 7 6 Hourly wage rate 5 Scale 4 3 Unemployment rate 2 Scale 1 0 80 85 90 95 99 population, both men and women, over the past three decades. Likewise, the hourly wage rates of women in general, and single mothers in particular, have been undergoing long-term trends that have affected all women, not just mothers on welfare. It would be incorrect to attribute long-term trends or any post1996 trend to welfare or any other factor if those trends were occurring for all single mothers. Charts 3 and 4 show the trends in the education measure and hourly wage measure, respectively, for welfare recipients relative to those same measures for nonwelfare recipients.11 Interestingly, the upward trend the in education of welfare recipients appears even here, reflecting a gain relative to nonwelfare recipients. As for the period following 1996, it appears that educational levels of the welfare recipient population are again rising, but it is not clear that they are rising any faster than would be expected from the long-term trend. The hourly wage rate shows a long-term, secular decline relative to nonwelfare recipients and without the gradual recovery that was visible in Chart 2. This decline in relative wages is probably the result of a deterioration in the demand for low-skilled labor that has affected other low-skilled workers in the U.S. economy over this same period. The wage rate appears to be countercyclical, as should be expected: as the unemployment rate rises, higher wage workers come onto the welfare rolls. The period in the early 1980s does not demonstrate this relationship, however, possibly because the 1981 OBRA reduced the number of higher wage welfare recipients at the same time that the unemployment rate was rising. After 1996, there appears to be a decline in the wage rates of welfare recipients, but again it is not clear whether it is any different from what would be expected from a trend. Table 1 reports the results of regressions in which these two welfare-nonwelfare ratios, as well as similar ratios for other variables, are regressed on a time trend, the unemployment rate, a dummy for OBRA 1981, and a dummy for 1996 and after. The trend coefficients in the first two rows confirm the graphical evidence that there have been significant long-term trends in both the education and hourly wage rates of welfare recipients relative to nonwelfare recipients. The unemployment coefficients are both positive, although statistically significant in only one case, indicating that higher unemployment rates draw onto the rolls more skilled women in terms of education and wage rates. This implies that both educational levels and wage rates in the post-1996 period should have been falling because of the business cycle alone. The coefficients on the 1981+ dummy for OBRA are both negative, indicating that more skilled recipients left the rolls because of that legislation. Finally, the coefficients in the last column show whether there has been a deviation from trend and cycle after 1996; the answer is that there has been no significant change. Although educational levels have been rising and hourly wage rates of recipients falling after 1996, these are not significantly different from what would be expected on the basis of trend and cycle. Therefore, the CPS provides no evidence that PRWORA has been strongly selective in ending welfare participation for either more or less disadvantaged women; the best conclusion is that both types of women have left the rolls in equal proportions. Table 1 also shows the results of similar regressions for other characteristics in the Current Population Survey. The caseload has been becoming younger, more white, and more composed of never-married mothers over the period, and these Chart 3 Ratio of the Percentage of Welfare Recipients with Twelve or More Years of Education to the Percentage of Nonwelfare Recipients, 1968-99 Chart 4 Ratio of Hourly Wage Rate of Welfare Recipients to Hourly Wage Rate of Nonwelfare Recipients, 1975-98 Percent Percent 12 Ratio of education percentages Scale 10 8 6 1.2 12 1.0 10 0.8 8 0.6 6 Scale Scale 2 0 1968 70 75 80 85 90 95 99 0.4 4 0.2 2 0 0 1975 1.0 0.8 0.6 Unemployment rate 4 1.2 Ratio of hourly wage rates Unemployment rate 0.4 Scale 0.2 0 80 85 90 FRBNY Economic Policy Review / September 2001 95 99 41 characteristics have been changing over the business cycle in expected ways: as the unemployment rate rises, women with smaller family sizes—who are younger and more likely to be white—come onto the rolls. However, there have been no post1996 changes in these recipient-nonrecipient ratios after netting out the effects of trend and cycle except for the proportion that has never been married, which has declined. Never-married recipients tend to be more disadvantaged than other recipients. The last four rows of the table show coefficients for regressions with labor market attachment as dependent variables—employment, weeks of work, hours of work, and Table 1 Regression Results for March Current Population Survey Welfare Recipient Characteristics Dependent Variable Twelve years of education or more Trend Unemployment OBRA81 PRWORA Rate Dummy Dummy .019* (.003) .023* (.008) -.065 (.044) -.053 (.044) Hourly wage rate -.006* (.002) .008 (.007) -.075* (.026) .032 (.023) Number in family .001 (.001) -.022* (.004) .021 (.023) .010 (.024) .009* (.001) .021* (.004) -.002 (.023) -.022 (.023) White .006* (.001) .005* (.003) .006 (.014) -.003 (.014) Never married .022* (.001) -.006 (.004) .065* (.022) -.066* (.022) .006* (.003) -.009 (.010) -.150* (.038) .128* (.034) .003 (.003) -.008 (.009) -.164* (.035) .122* (.031) .003 (.004) -.011 (.011) -.177* (.043) .150* (.038) -.002 (.002) -.006 (.008) -.159* (.030) .085* (.027) Less than twenty-five years old Employed over the year Annual weeks worked Hours worked per week Annual earnings Notes: Standard errors are in parentheses. The dependent variable is calculated as the ratio of the mean for welfare recipients to the mean for single mothers not on welfare in the year in question. Intercept not shown. *Statistically significant at the 10 percent level. 42 Changing Caseloads annual earnings. All four have risen significantly after 1996, even after accounting for trend and cycle. This suggests that welfare reform has, indeed, resulted in more work and earnings among welfare recipients than was the case prior to 1996.12 Evidence from Maryland Another source of data for examining trends in welfare recipient characteristics, albeit not national in scope, are administrative records from individual states and local areas. Many states have assembled records from welfare agency files of the characteristics of recipients over a fairly long period of time, and these typically have been matched to other administrative records, most commonly the earnings data from unemployment insurance (UI) files. Such data have the advantage of large sample sizes, relatively good administrative information on welfare receipt and simultaneous earnings, and a moderately long time period (1985 to 2000 in the case of Maryland).13 A disadvantage is that the data contain even less information on personal and family characteristics than the CPS does, and therefore cannot provide a comprehensive picture of well-being or an index of advantage and disadvantage. The major variable indexing skill comes from the match to UI files, where quarterly earnings are available.14 However, a better measure of disadvantage that can be constructed from this type of data comes from the availability of histories of welfare participation, for in this case we can classify recipients by their past level of welfare dependency. Long-term recipients are the most obvious subgroup on the rolls who are known from other research to be more disadvantaged in terms of labor market experience, education, health, and other problems; indeed, long-term recipiency is, in a sense, an overall measure that is a proxy for a large number of problems of disadvantage. We use a slightly more detailed classification based on the one initially proposed by Bane and Ellwood (1994), which divides the caseload into long-termers, short-termers, and cyclers. Longtermers are those with relatively long spells of welfare receipt and generally a relatively small number of individual spells; short-termers are those with short spells when on welfare and also a small number of spells; and cyclers are those with relatively short spell durations but a larger number of spells. Long-termers include the most disadvantaged women on welfare, while short-termers are presumed to be the least disadvantaged and cyclers are in between long-termers and short-termers in this dimension.15 We use this classification as our primary measure of disadvantage and examine whether the relative numbers of these types of recipients have trended over time, have varied with the business cycle, and have changed after PRWORA. The Maryland welfare and earnings data are available for all TANF recipients beginning in April 1985 and running through March 2000. These files are maintained by The Jacob France Center at the University of Baltimore through data-sharing agreements with Maryland’s Department of Human Resources (DHR) and Department of Labor, Licensing, and Regulation (DLLR). Data are available for all Maryland welfare recipients, but the diversity of Maryland’s economy led us to limit the analysis reported here to Baltimore City welfare recipients alone. We use the longitudinal dimension of the data to classify women by their welfare dependency status; we use a five-year window to do so, using welfare participation within that window to classify women into the three dependency groups. To examine trends over time, we select different birth cohorts of women, each cohort consisting of all women in that cohort who were on welfare at least once during the five-year period. In the results reported here, we select women who were nineteen in the initial year. Thus, for example, our earliest cohort consists of women whose nineteenth birthday fell between April 1, 1985, and March 31, 1986, whom we follow from 1985:2 to 1990:1. Our second cohort consists of women whose nineteenth birthday fell between April 1, 1986, and March 31, 1987, whom we follow from 1986:2 to 1991:1, and so on. The final cohort was age nineteen between the same dates in 1995 and 1996, and is followed from 1995:2 to 2000:1. For each cohort, we extract all monthly welfare events represented in the DHR records and quarterly DLLR wage records over the relevant five years and select all women with at least one welfare record. We have eleven cohorts; comparing these cohorts over time tells us whether the caseload is changing in terms of composition, controlling, clearly, for age—each cohort is at the same point in its life cycle.16 We define a woman as a cycler if she had three or more spells during the five years, a long-termer if she had only one or two spells and an average spell duration of twenty-one months or more, and a short-termer if she had only one or two spells and an average spell duration of twenty months or less.17 Some components of welfare reform in Maryland began with a federal waiver in October 1995, so we have many months of observations after the official beginning of the reforms. However, state and local observers encourage the use of October 1996 as an appropriate date to expect welfare leaving and employment profiles to show a reform effect, for that is the approximate date of post-PRWORA TANF implementation. The Maryland TANF program has two-year work requirements stipulating a minimum of twenty hours per week (in accordance with federal law), full family sanctions, a 35 percent earnings disregard, and a five-year time limit. The cyclical pattern of the unemployment rate in Maryland over the 1985-2000 period is roughly similar to that in the United States as a whole, although lower in level. It fell from 4.6 percent in 1985 to a trough of 3.7 percent in 1989, then rose to a peak of 6.7 percent in 1992, and has since fallen steadily to 3.5 percent in 1999. Table 2 shows the trend in cohort size as well as the relative fractions of women in short-term, cycler, and long-term welfare dependency categories over time. The cohort size column indicates that the number of young women ever receiving welfare in Baltimore City in the five-year period rose for the first few cohorts, most of whose observation periods fell in the period of rising unemployment from 1989 to 1992. It peaked for the 1989-94 cohort and then fell markedly, reaching its level for the first cohort by the 1991-96 period. Subsequently, it has declined only slightly thereafter through the last cohort (in fact, it rose for the last two cohorts). The lack of decline in cohort size in the last, post-PRWORA periods reflects the fact that entry rates in Baltimore did not decline very strongly, at least through 1997 or 1998 (Mueser et al. 2000, Figure 2). The caseload did decline, however, because exit rates rose. Table 2 Percentage Distribution of Maryland AFDC-TANF Caseload by Welfare Dependency Status and Cohort, Ages Nineteen to Twenty-Three Welfare Dependency Status (Percentage Distribution)b Cohorta 1985:2-1990:1 1986:2-1991:1 1987:2-1992:1 1988:2-1993:1 1989:2-1994:1 1990:2-1995:1 1991:2-1996:1 1992:2-1997:1 1993:2-1998:1 1994:2-1999:1 1995:2-2000:1 Cohort Size LongTermer ShortTermer Cycler 1,865 2,234 2,354 2,307 2,388 2,090 1,874 1,604 1,518 1,751 1,754 35 33 40 41 47 45 46 45 43 19 15 65 64 56 52 44 47 44 47 48 62 67 0 3 4 7 9 8 10 8 9 19 18 a Five-year observation period for women who were age nineteen in the year beginning with the first quarter indicated. b Long-termers have one or two spells in the five-year observation period and an average spell of twenty-one months or more; short-termers have one or two spells in the five-year observation period and an average spell of twenty months or less; cyclers have three or more spells in the five-year observation period. FRBNY Economic Policy Review / September 2001 43 The other columns in Table 2 separate the cohorts into short-termer, cycler, and long-termer components. Most of the young Baltimore City welfare recipients—between 44 and 67 percent across all cohorts–are designated as short-termers. There are very few cyclers in general, although the number has been gradually rising over time. Long-termers are in between in terms of size. The trends in composition up through the 1993-98 cohort are partly explainable in terms of the business cycle. In the late 1980s and early 1990s, as the local unemployment rate and welfare caseload rose, the percentage of long-term recipients drifted upward to a high of 47 percent for the 199095 cohort and fell modestly over the next four cohorts as the unemployment rate declined, though not falling perhaps as much as would be expected. Mirroring this trend, the percentage of the cohort composed of short-termers fell initially and then rose slightly through the 1993-98 cohort. Interestingly, the percentage composed of cyclers rose during the rise in the unemployment rate as well, but then roughly stabilized. The last two cohorts show a marked change in composition, with a sharp drop in the percentage of long-termers and a sharp rise in the percentage of short-termers and cyclers. The unemployment rate was continuing to decline over this period, but at a steady rate that could not explain the suddenness of the caseload composition change, which is therefore almost surely the result of welfare reform. The abruptness of the change also suggests that welfare reform in Baltimore affected primarily those young recipients who had recently entered the rolls, for the last two cohorts are observed for almost their entire fiveyear period after welfare reform. The earlier cohorts began their observation period prior to reform. These findings go against the conventional wisdom for how welfare reform should affect the composition of the caseload, for the usual presumption is that the percentage of the caseload composed of long-termers should markedly rise after reform, as short-termers and cyclers leave the rolls for the labor market. The opposite has occurred in Maryland, where long-termers have declined as a fraction of the ever-on five-year caseload. It is quite likely that women who would have been long-termers in the absence of reform are now short-termers and cyclers, and that welfare reform has caused a reduction in the number of long spells while on welfare. Note that this does not imply that those who would have been long-termers have left the rolls; indeed, the cohort size rose slightly over the last two cohorts.18 However, it does imply that it is among the longtermers where welfare policy has had its greatest impact in Baltimore. Tables 3-5 offer additional details on how the characteristics of the young Baltimore caseload has changed over time. Table 3 shows trends in the total percentage of time on welfare over the 44 Changing Caseloads five-year period—sometimes called the “total time on”—which is one of the best overall measures of welfare dependency. The first column shows a marked rise in welfare dependency, from 31 percent of the five years on welfare to a high of 51 percent for the 1991-96 cohort. The percentage of time on subsequently declined at about the same time as the unemployment rate, and then dropped more precipitously as the 1990s ended, returning to the beginning level of approximately one-third of the five years spent in welfare dependency. Again, this abrupt decline is almost surely the result of welfare reform. The other columns show that this welfare reform change was the result of two, complementary changes: a drop in the total time on among those who remained as long-termers and a slight drop among those who were short-termers and cyclers. Even if these welfare dependency levels within groups had not changed, the shift from long-termers to short-termers and cyclers apparent in Table 2 would have generated a reduction in overall total time on. The reductions in total time on within each group, particularly among long-termers, reinforce this.19 Table 4 shows trends in the mean quarterly earnings of the women during the quarters in which they were not receiving welfare benefits. Real earnings rose steadily through the mid- Table 3 Percentage of Five-Year Period on AFDC-TANF, by Cohort and Welfare Dependency Welfare Dependency Statusb Cohorta All LongTermer ShortTermer Cycler 1985:2-1990:1 1986:2-1991:1 1987:2-1992:1 1988:2-1993:1 1989:2-1994:1 1990:2-1995:1 1991:2-1996:1 1992:2-1997:1 1993:2-1998:1 1994:2-1999:1 1995:2-2000:1 31 34 39 43 49 49 51 49 48 37 32 57 56 59 64 71 73 74 73 72 67 54 18 23 25 25 24 24 25 25 25 20 20 22 39 44 51 59 59 58 59 60 61 57 Note: Percentage of time on welfare is defined as the fraction of the sixty months in the five-year observation period in which the woman received an AFDC or TANF payment. a Five-year observation period for women who were age nineteen in the year beginning with the first quarter indicated. b Long-termers have one or two spells in the five-year observation period and an average spell of twenty-one months or more; short-termers have one or two spells in the five-year observation period and an average spell of twenty months or less; cyclers have three or more spells in the five-year observation period. Table 4 Mean Quarterly Earnings While off Welfare, by Cohort and Welfare Dependency, in Real 1999 Dollars Welfare Dependency Statusb Cohorta 1985:2-1990:1 1986:2-1991:1 1987:2-1992:1 1988:2-1993:1 1989:2-1994:1 1990:2-1995:1 1991:2-1996:1 1992:2-1997:1 1993:2-1998:1 1994:2-1999:1 1995:2-2000:1 All LongTermer ShortTermer Cycler 1,313 1,498 1,572 1,502 1,626 1,773 1,855 1,752 1,965 1,845 1,889 1,201 1,232 1,232 1,131 1,302 1,546 1,737 1,661 1,961 2,398 1,716 1,361 1,618 1,794 1,763 1,879 1,896 1,957 1,785 1,983 1,724 1,981 1,762 1,468 1,286 1,344 1,588 1,862 1,655 1,888 1,873 1,794 1,625 a Five-year observation period for women who were age nineteen in the year beginning with the first quarter indicated. b Long-termers have one or two spells in the five-year observation period and an average spell of twenty-one months or more; short-termers have one or two spells in the five-year observation period and an average spell of twenty months or less; cyclers have three or more spells in the five-year observation period. Table 5 Earnings Relative to Human Capital Potential, by Cohort Cohort 1985:2-1990:1 1986:2-1991:1 1987:2-1992:1 1988:2-1993:1 1989:2-1994:1 1990:2-1995:1 1991:2-1996:1 1992:2-1997:1 1993:2-1998:1 1994:2-1999:1 1995:2-2000:1 Five-Year Earnings as a Percentage of Human Capital Potential Human Capital Potential as a Percentage of Full-Time, Full-Year Minimum Wage 44 46 48 48 45 43 42 42 44 46 52 51 58 61 58 63 69 72 68 76 72 73 Notes: Human capital potential is defined for each individual as her mean quarterly earnings, taken over those quarters in which she had earnings, multiplied by twenty quarters. The full-time, full-year minimum wage is the earnings amount for 2,000 hours per year for five years at the minimum wage, $5.15 per hour. 1990s, both overall and for the individual dependency groups. If earnings are taken as a measure of wage rates, which they proxy only slightly, this is consistent with the more skilled women being on the rolls than was the case for cohorts in which the unemployment rate was lower. However, earnings have more or less leveled off over the past few cohorts, even though the unemployment rate has declined (there is considerable fluctuation, but no consistent upward or downward trend). These results are fairly surprising, for the marked decline in the unemployment rate should have led to a decline in wage rates as the caseload became less skilled (again, quarterly earnings is at best a proxy for wage rates); however, real earnings did, at least, level off, and no longer continue to rise.20 These figures do not capture labor market skill levels in the same way the hourly wage does because they do not control for the employment rate and for hours of work. The former can at least partly be adjusted for by calculating what earnings over the entire five-year period would be if each individual had worked in all twenty quarters and had earned in each quarter the off-welfare amounts shown in Table 4; we term this their “human capital potential.”21 The first column of Table 5 shows how the five-year earnings of each cohort have changed relative to this human capital potential, showing that they have risen gradually and then increased sharply recently. This calculation implies a more definitive increase in employment and work effort than was implied in Table 4. The second column shows the ratio of this human capital potential to a measure of fullyear, full-time work at the minimum wage over the full five years. This measure has also increased over the eleven cohorts, but with some unevenness. There has been some increase in the average earnings capacity of the caseload, but the effects in the last three cohorts, which are the main post-reform periods, are not as strong relative to previous periods as might be expected. III. Other Evidence There is a scattering of additional evidence from other sources relevant to the issues discussed thus far, even though none is definitive and none controls for the business cycle. One source is administrative data from AFDC and TANF records on the characteristics of recipients over time. Such data have been collected sporadically since 1969, but not always on a comparable basis. For present purposes, the main variable of interest is educational attainment, which has been collected on and off over the years. The figures in Table 6 show how it has changed over time. There was a dramatic improvement in the educational level of AFDC adults from 1969 to 1994 in terms of FRBNY Economic Policy Review / September 2001 45 the percentage of recipients who have at least twelve years of education, and some improvement from 1986 to 1995 in the fraction with some college education. Unfortunately, for 1996 and after, only the percentage with twelve or more years of education has been published. There appears to be some negative selection on education in 1995 and after, for while the fraction with twelve or more years grew from 1986 to 1994, it then dropped in 1995 and afterward. This is consistent with the CPS, which also showed negative selection in this period, but there it was ascribed to the business cycle.22 A second source of additional evidence comes from some of the studies of welfare leavers conducted in the past several years that have compared leavers with stayers (most leaver studies do not conduct such comparisons). All of these studies focus on post-1996 data, so no control for the business cycle can be made. For example, Loprest and Zedlewski (1999) find that stayers have lower levels of education and more obstacles and barriers to work and, among those with obstacles, stayers are less likely to work. However, stayers and leavers did not differ on some other dimensions (such as health). Cancian et al. (2000) estimate probit equations for the probability of leaving TANF in Wisconsin and find that probability to be positively related to education, age, age of the youngest child, the number of other adults in the household, and work experience, and to be negatively correlated with the number of children and years on welfare. These all accord with selection on job readiness. A further investigation of the likelihood of leaving welfare in Wisconsin, using dependency categories similar to those used here for the Maryland data, finds that short-termers and cyclers are more likely to leave welfare than long-termers (Ver Ploeg 2001). A study comparing leavers with stayers in Illinois likewise finds stayers to be worse off in terms of education, experience, and marital history (Institute for Public Affairs and School of Social Work 2000). A similar study in the state of Washington finds leavers to be better educated, younger, in better health, and to have fewer children than stayers (Fogarty and Kraley 2000).23 In an examination of welfare leavers in Michigan, Danziger (2000) finds them to have higher levels of education, better adult and child health, more work experience and job skills, and fewer transportation problems than stayers. While this evidence is a bit mixed, the general tendency is nevertheless consistent with negative selection on skill and with the implication that better-off recipients have more likely left the rolls. Again, this is consistent with the CPS, although with that data set it was attributable to the favorable state of the economy. A leaver examination issued by the Three-City Study is also indirectly relevant. Moffitt and Roff (2000) found that leavers in three cities (Boston, Chicago, and San Antonio) contained a wide diversity of different types of women—ranging from more educated women who were in better health and had relatively high employment and earnings to less educated women who were often in poor health and had much lower employment and earnings. The implication of these findings is Table 6 Educational Percentage Distribution of AFDC and TANF Adults, 1969-99 Fiscal Year Fiscal Year Years of Education May 1969 May 1975 March 1979 1986 1988 1990 1992 1994 1995 1997 1998 1999 8 or less 9-11 12 13-15 16+ 1-6 7-9 10-11 12+ 37.5 39.2 20.4 2.6 0.3 — — — 23.3 21.8 41.3 30.9 5.1 0.9 — — — 31.9 18.3 40.0 36.2 5.2 0.8 — — — 42.2 11.9 35.5 42.9 8.4 1.2 — — — 52.5 13.2 35.3 42.0 9.4 1.4 — — — 52.8 12.2 34.6 40.5 11.9 0.8 — — — 53.2 9.2 35.2 41.9 12.7 0.9 — — — 55.5 7.4 32.5 44.6 14.3 0.9 — — — 59.8 10.0 28.9 45.4 14.6 1.1 — — — 51.1 — — — — — 5.2 14.1 28.0 52.6 — — — — — 5.0 13.2 29.9 51.9 — — — — — 5.2 12.3 31.5 51.1 Sources: For 1969-95, U.S. Congress (1998, Table 19); for 1997-99, U.S. Department of Health and Human Services (1997, Table 16; 1998, Table 17; 1999, Table 17). Notes: Figures shown represent the originals inflated by the fraction nonmissing. Figures for 12+ for 1969-95 are derived by summing the figures for 12, 13-15, and 16+. 46 Changing Caseloads that leavers are composed not only of the better-off recipients in the caseload, but many of the more disadvantaged recipients as well. This is the flip side of the coin indicating that stayers are likewise composed of more advantaged as well as disadvantaged recipients. Both stayers and leavers are composed of a diverse, heterogeneous set of women who have a wide range of labor market skills and other characteristics. This again belies the conventional view of leavers as composed solely of better-off (former) recipients and stayers as composed solely of worse-off recipients.24 Another set of relevant studies are the few leaver studies that have compared multiple cohorts of post-1996 leavers. The conventional wisdom suggests that successive waves of leavers should be progressively worse off, presuming that the better-off and more job-ready recipients left first. The available studies provide much less support for this supposition than would be expected. While a study in Illinois found early leavers to be slightly better off than later leavers in terms of work experience, education, and marriage history (Institute for Public Affairs and School of Social Work, University of Illinois 2000), a succession of leaver cohorts examined in South Carolina found no difference in employment rates, hardship, or other measures (South Carolina Department of Social Services 2000). A study of Wisconsin leaver cohorts in 1995 and 1997 found lower earnings in the later cohort but no significant differences in employment or income (Cancian et al. 2000), and a review of three states with multiple cohort leavers— Arizona, Washington, and Wisconsin—found very mixed evidence of any trend in employment (Isaacs and Lyon 2000). The finding of little evidence of selectivity is not consistent with many of the previous findings, but since most of these multiple cohort studies have examined only post-1996 leavers—when the unemployment rate decline has slowed and welfare reform effects have been more important—they may be more consistent with the CPS findings reported above, which indicate no statistically significant selectivity after 1996. IV. Conclusions The discussion in this paper concerns the effect of welfare reform on the composition of the caseload, and on whether reform has led to more or less disadvantaged recipients leaving the rolls. A consideration of the theoretical effects of welfare reform on the composition of the caseload suggests that while most policies should lead to a departure from the rolls of those who are more job-ready, who have more labor market skills, and who are in general less disadvantaged, several welfare reforms—most notably, sanctions and more generous earnings disregards—work in the opposite direction. Our analysis of national Current Population Survey data indicates that the skill level of the welfare caseload has tended to decline, but this has been primarily the result of the improvement in the economy; welfare reform per se, after one nets out the effects of the economy, has had little effect on the composition of the caseload in its labor market skill distribution. An analysis of data from Maryland indicates that welfare reform has had its major impact on long-term recipients, who are the most disadvantaged. A survey of other studies comparing leavers with stayers, multiple cohorts of leavers, and diversity among leavers and stayers, offers several findings. One finding is that while stayers may have been worse off in general than leavers—although our analysis implies that this is the result of the business cycle and not welfare reform— more recent cohorts of leavers are not much different than earlier cohorts of leavers. In addition, studies examining the diversity of leavers have found many worse-off former recipients who have left welfare, suggesting that leavers have not been composed solely of better-off former recipients. The policy implication of these findings is that policy should recognize that there are at least two types of low-income single mothers: those who have more job skills and are better off and those who have much lower skill levels and a much greater set of problems. More important, both types of women are found both on and off the welfare rolls. Therefore, any additional assistance to either or both groups—say, greater labor market supports to the more job-ready women and more basic assistance to the more disadvantaged women—should be directed not just at women still on TANF, but also at women off TANF. Policies must be designed to assist women in need of assistance who are in these multiple situations. FRBNY Economic Policy Review / September 2001 47 Endnotes 1. For an exception, see Mueser et al. (2000), who find that welfare reform had no effect on employment rates of leavers in five urban areas through 1997, after controlling for the economy. 2. Because these studies examine a more comprehensive group, they can capture the effects of welfare reform on discouraged entry onto welfare as well as increased exit. 3. It is somewhat less obvious what the influence of other sources of nonwelfare income should be. For example, whether women who have more income available off welfare from unearned sources—say, help from other family members—have more labor market skills or less skills is not as clear-cut. In addition, it is also less clear how labor market skill is correlated with the likelihood of moving on and off welfare as the result of changes in marital status. 4. Similar effects should occur in terms of welfare entry. Women with more job market skills are the least likely to come on the rolls for the most part, but they are also more likely to be able to fulfill the work and job-search requirements—often imposed by formal diversion programs—than are women in more disadvantaged situations. See Moffitt (1996) for a general discussion of entry effects in welfare programs. 5. As of October 1997, one state (Illinois) disregarded 67 percent of earnings and a number of states (such as California) disregarded 50 percent of earnings, usually beyond a threshold. Other states had smaller disregards and a few remained with the AFDC disregard of zero. See Gallagher et al. (1998). 6. The 1967 and 1981 federal changes in earnings disregards in the AFDC program had these effects. The 1967 change increased earnings disregards, which led to an increase in the employment rate and earnings of welfare recipients and hence an increase in the skill level of those on welfare. The 1981 Omnibus Budget Reconciliation Act’s (OBRA) elimination of earnings disregards removed many workers from the welfare rolls and led to a reduction in the employment rate of welfare recipients. In addition, since 1996, states that have had more generous earnings disregards have had higher employment rates of recipients on TANF (U.S. Congress 2000, Chart 7-5). 7. Another advantage of the CPS is that trends in recipient characteristics will capture the effects of economic and policy changes working through entry rates as well as exit rates. 8. Another disadvantage of the CPS is that it appears to be increasingly undercounting the number of AFDC and TANF recipients compared 48 Changing Caseloads with counts in administrative data. This is a serious but currently unresolved problem. It will not affect the results given here if the undercount is not related to the measures of disadvantage we use (education and hourly wage rate). 9. Unfortunately, hours of work per week in the past calendar year have been collected only since 1976. Therefore, hourly wage rates are measurable only from 1976 to 1998, unlike the other variables, which go back to 1968. 10. The sample is composed of all single mothers age sixteen to sixtyfour who reported public assistance income in the prior calendar year. Education is measured at the time of the March interview and the hourly wage is the average wage rate over the prior calendar year, in real 1997 dollars. 11. For the wage rate measure, only those with hourly wage rates of less than $30 per hour are included, for those constitute a better comparison group than all single mothers. 12. As emphasized previously, the inability to know from these data whether the work periods were in the same weeks as the welfare participation periods over the year leaves somewhat ambiguous the issue of whether this increased work occurred while on or off the rolls. This illustrates one of the weaknesses of the CPS for this type of question. However, evidence from many other sources (such as U.S. Department of Health and Human Services [2000]) clearly indicates that there has been a large increase in employment and earnings among TANF adults subsequent to PRWORA. 13. A minor timing problem arises because UI earnings are available quarterly but welfare data are available monthly, so it is not possible to know precisely in some cases whether work and welfare periods overlap within a quarter. However, this is a minor problem relative to the major timing issues in the CPS. 14. Another disadvantage is that there is no information on hours of work over the quarter, so hourly wage rates—the preferable measure—cannot be calculated. 15. See Moffitt (2001) for an analysis of the background characteristics of these three types of recipients. Somewhat surprisingly, the analysis indicates that cyclers are, in some dimensions, worse off than longtermers. How these groups are defined affects the answer to this question. Endnotes (Continued) 16. Other age groups could, of course, be examined. We reserve that for future work. 17. A “spell” in our definition is a consecutive run of months of welfare receipt that is not interrupted by two or more months of consecutive nonreceipt (one-month gaps are allowed). Left-censored and right-censored spells are included as spells. The twenty-month criterion for separating long-termers from short-termers is used because twenty months is the mean “spell” length among those two groups combined. 18. Another way to say this is to suppose that the impact of welfare reform in Baltimore had taken place by shortening the spells of shorttermers and cyclers only, who, although having come onto welfare in the first place, left earlier than they would have otherwise; this is the usual hypothesis. In that case, the relative proportions of the three groups in Table 2 would not have changed at all. 19. It is interesting to note that the total time-on figures for cyclers have risen over time to equal those of long-termers. However, this is a long-standing trend and not a result of welfare reform. 20. Real earnings levels tend to be highest for short-term recipients over most of the period, with cyclers between short-term and longterm recipients, who have the lowest levels. Note that this is not a statistical artifact of their assignment to long-term status because only nonoverlapping quarters with some earnings are used to calculate the subpopulation average amount. However, these differences have gradually declined and have led, in particular, to a closer match between long-term recipients and cyclers. 21. On average, the young Baltimore caseload worked seven to nine quarters over the five-year period. There was a slight increase from eight to nine quarters for the last three cohorts. 22. These data are of sometimes dubious quality, for often upwards of 40 percent of the sample is missing education information in some of the years. 23. A study of Medicaid leavers and stayers finds as well that leavers are in better health than stayers (Garrett and Holahan 2000). 24. The Danziger study (2000, Table 3) also broke out leavers and stayers each into those who are working and not working. Interestingly, Danziger found that working stayers and working leavers were much more similar in characteristics (such as education and work experience) than stayers and leavers as a whole, and that nonworking stayers and nonworking leavers were also more similar than stayers and leavers as a whole. This has the same implication: there is a mix of better-off and worse-off types of women in both stayer and leaver groups. FRBNY Economic Policy Review / September 2001 49 References Acs, G., and P. Loprest. 2001. “Studies of Welfare Leavers: Methods, Findings, and Contributions to the Policy Process.” In R. Moffitt and M. Ver Ploeg, eds., Data Collection and Research Issues for Studies of Welfare Populations. Panel on Data and Methods for Measuring the Effects of Changes in Social Welfare Programs, Committee on National Statistics. Washington, D.C.: National Academy Press. Bane, M. J., and D. Ellwood. 1994. Welfare Realities: From Rhetoric to Reform. Cambridge: Harvard University Press. Brauner, S., and P. Loprest. 1999. “Where Are They Now? What States’ Studies of People Who Left Welfare Tell Us.” Assessing the New Federalism, Brief, Series A, no. A-32. May. Washington, D.C.: Urban Institute. Cancian, M., R. Haveman, D. Meyer, and B. Wolfe. 2000. “Before and after TANF: The Economic Well-Being of Women Leaving Welfare.” Special Report no. 77. Madison, Wis.: Institute for Research on Poverty, University of Wisconsin. Council of Economic Advisers. 1997. “Explaining the Decline in Welfare Receipt, 1993-1996.” Washington, D.C.: Executive Office of the President. ———. 1999. “The Effects of Welfare Policy and the Economic Expansion on Welfare Caseloads: An Update.” Washington, D.C.: Executive Office of the President. Danziger, S. 2000. “Approaching the Limit: Early Lessons from Welfare Reform.” Paper presented at the Joint Center for Poverty Research Conference, Rural Dimensions of Welfare Reform. Fogarty, D., and S. Kraley. 2000. “A Study of Washington State TANF Leavers and TANF Recipients: Findings from Administrative Data and the Telephone Survey: Summary Report.” State of Washington. Gallagher, L. J., et al. 1998. “One Year after Federal Welfare Reform: A Description of State Temporary Assistance for Needy Families (TANF) Decisions as of October 1997.” Washington, D.C.: Urban Institute. Garrett, B., and J. Holahan. 2000. “Do Welfare Caseload Declines Make the Medicaid Risk Pool Sicker?” Report no. 00-06. Washington, D.C.: Urban Institute. 50 Changing Caseloads Institute for Public Affairs and School of Social Work. University of Illinois. 2000. “Illinois Study of Former TANF Clients: Final Report, Executive Summary.” Issacs, J., and M. Lyon. 2000. “A Cross-State Examination of Families Leaving Welfare: Findings from the ASPE-Funded Leavers Studies.” Paper presented at the National Association for Welfare Research and Statistics Conference. Scottsdale, Ariz. Loprest, P., and S. Zedlewski. 1999. “Current and Former Welfare Recipients: How Do They Differ?” Discussion Paper no. 99-17. Washington, D.C.: Urban Institute. Mayer, S. E. 2000. “Why Welfare Caseloads Fluctuate: A Review of Research on AFDC, SSI, and the Food Stamps Program.” New Zealand Treasury. Moffitt, R. 1996. “The Effect of Employment and Training Programs on Entry and Exit from the Welfare Caseload.” Journal of Policy Analysis and Management 15: 32-50. ———. 1999. “The Effect of Pre-PRWORA Waivers on AFDC Caseloads and Female Earnings, Income, and Labor Force Participation.” In S. Danziger, ed., Economic Conditions and Welfare Reform. Kalamazoo, Mich.: W. E. Upjohn Institute for Employment Research. ———. 2001. “Experience-Based Measures of Heterogeneity in the Welfare Caseload.” In R. Moffitt and M. Ver Ploeg, eds., Data Collection and Research Issues for Studies of Welfare Populations. Panel on Data and Methods for Measuring the Effects of Changes in Social Welfare Programs, Committee on National Statistics. Washington, D.C.: National Academy Press. Moffitt, R., and L. Pavetti. 2000. “Time Limits.” In Finding Jobs: Work and Welfare Reform. New York: Russell Sage Foundation. Moffitt, R., and J. Roff. 2000. “The Diversity of Welfare Leavers.” Policy Brief and Working Paper from the Three-City Study. Johns Hopkins University. <http://www.jhu.edu/~welfare> Mueser, P., J. Hotchkiss, C. King, P. Rokicki, and D. Stevens. 2000. “The Welfare Caseload, Economic Growth, and Welfare-to-Work Policies: An Analysis of Five Urban Areas.” Unpublished paper, July. References (Continued) Schoeni, R., and R. Blank. 2000. “What Has Welfare Reform Accomplished? Impacts on Welfare Participation, Employment, Income, Poverty, and Family Structure.” NBER Working Paper no. 7627. ———. 1999. “Characteristics and Financial Circumstances of TANF Recipients: Fiscal Year 1999.” Washington, D.C.: Administration for Children and Families. <http://www.acf.dhhs.gov/programs/ opre/> South Carolina Department of Social Services. 2000. “Survey of Former Family Independence Program Clients: Cases Closed during April through September 1998.” ———. 2000. “Temporary Assistance for Needy Families (TANF) Program: Third Annual Report to Congress.” Washington, D.C.: Administration for Children and Families. U.S. Congress. Committee on Ways and Means. 1998. “Background Material and Data on Programs within the Jurisdiction of the Committee on Ways and Means.” Washington, D.C.: U.S. Government Printing Office. <http://aspe.hhs.gov/gb98/ intro.htm> U.S. General Accounting Office. 1999. “Welfare Reform: Information on Former Recipients’ Status.” Washington, D.C.: U.S. Government Printing Office. ———. 2000. “Background Material and Data on Programs within the Jurisdiction of the Committee on Ways and Means.” Washington, D.C.: U.S. Government Printing Office. U.S. Department of Health and Human Services. 1997. “Characteristics and Financial Circumstances of TANF Recipients: July-September 1997.” Washington, D.C.: Administration for Children and Families. <http://www.acf.dhhs.gov/programs/opre/> ———. 2000. “Welfare Reform: State Sanction Policies and Number of Families Affected.” Washington, D.C.: U.S. Government Printing Office. Ver Ploeg, M. 2001. “Pre-Exit Benefit Receipt and Employment Histories and Post-Exit Outcomes of Welfare Leavers.” In R. Moffitt and M. Ver Ploeg, eds., Data Collection and Research Issues for Studies of Welfare Populations. Panel on Data and Methods for Measuring the Effects of Changes in Social Welfare Programs, Committee on National Statistics. Washington, D.C.: National Academy Press. ———. 1998. “Characteristics and Financial Circumstances of TANF Recipients: Fiscal Year 1998.” Washington, D.C.: Administration for Children and Families. <http://www.acf.dhhs.gov/programs/ opre/> The views expressed in this article are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever. FRBNY Economic Policy Review / September 2001 51 Susan E. Mayer Commentary I n 1994, welfare rolls began to fall precipitously. In her study, Rebecca Blank reviews the research that tries to separate the effect of the economy on this decline from the effect of policy changes. The paper by Robert Moffitt and David Stevens attempts to determine whether the Temporary Assistance for Needy Families (TANF) program has changed the labor market characteristics of welfare recipients. An important motivation for both papers is to help predict what will happen when the economy takes a downturn. If changes in caseloads are mostly due to the booming economy, caseloads will probably increase rapidly in a recession. If, instead, the decline in the rolls is due to program changes, caseloads might increase more slowly in a recession. Both Blank and Moffitt and Stevens rely on the standard economic model of caseloads. In this model, caseloads depend on program parameters that affect eligibility and the economic attractiveness of participation. The macroeconomy affects caseloads by changing the attractiveness of work, which is an alternative to program participation. But not everyone who is eligible for welfare takes it. Only about two-thirds of families eligible for Aid to Families with Dependent Children (AFDC) actually participated in the program. Furthermore, the standard economic model is unable to explain fully early increases in AFDC caseloads or the current decline. This fact, combined with relatively low take-up rates for welfare, suggests that something other than the standard economic variables may be important for explaining changes in welfare rolls. In a well-known paper, Moffitt (1983) invoked welfare stigma to help explain low take-up rates. Stigma is only one aspect of the norms and values that affect caseload changes. The standard economic model treats norms and values as constants, not as variables. Over the long run, changes in social norms and values affect caseloads by affecting demographic characteristics such as marriage and fertility, which affect eligibility. In the shorter run, changes in norms and values affect the success of program changes. In turn, program parameters are often meant to change norms and values. Program parameters have a smaller effect when they are contrary to strongly held norms and values. Since 1967, AFDC recipients have been required to seek work. But in 1967, there was a lot of social ambiguity about whether mothers of young children should work. In 1970, less than a third of married mothers of pre-school-age children worked at all and many of these women worked part-time. In the absence of a strong work norm among mothers of young children, there were few social supports, including child care, for working mothers. In addition, work rules for AFDC were vague, reflecting the ambivalence of legislators and the public about mothers of young children working. Together, these factors made it easy for caseworkers to make “excuses” for clients who did not show up for job interviews or otherwise seek employment, and few welfare recipients were sanctioned for such behavior. As more and more middle-class mothers of young children went to work, social approval for working mothers increased (as did Susan E. Mayer is an associate professor at the University of Chicago’s Irving B. Harris School of Public Policy Studies and the Director of the Northwestern University/University of Chicago Joint Center for Poverty Research. The views expressed are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. FRBNY Economic Policy Review / September 2001 53 availability of child care). This shift clearly encouraged legislators to pass increasingly aggressive work rules for welfare recipients and encouraged caseworkers to feel more confident in sanctioning mothers who did not cooperate with these rules. Norms and values can also affect many aspects of program implementation. For example, high levels of social hostility toward welfare recipients by society as a whole could lead caseworkers to treat potential recipients in ways that discourage their participation in welfare programs. This social hostility stigmatizes welfare recipients, reducing participation among those who are eligible. This means that if the take-up rate among eligibles does not change, changes in norms and values regarding welfare cannot be a big factor in caseload changes.1 Blank (2001) finds that almost all of the change in caseloads between 1984 and 1995 was attributable to changes in eligibility, and little was due to changes in take-up rates. However, take-up rates among those eligible for the food stamp program and Medicaid seem to have fallen since TANF was implemented, so take-up rates for welfare benefits may also have declined. Both program rules and the economy can affect norms and values. Conservatives believed not only that AFDC provided disincentives for single mothers to work, marry, and control their fertility, but also that it fostered a “culture of poverty.” By this, they meant that a set of social responses to incentives provided by AFDC had been internalized into norms and values that perpetuated poverty even when incentives changed. The low value placed on work and marriage reduced the extent to which they responded to changes in the economy and in welfare rules. Not working, it was argued, was due to attitudes toward work, not to the unavailability of jobs. TANF was supposed to change behavior as well as these norms and values. Changes in norms and values that result from a change in program parameters in turn affect future responses to economic change and changes in program parameters. Imagine a state in which strong work rules are implemented during a strong economy. More single mothers become employed, so fewer are eligible for welfare. The welfare caseload declines. With the increase in job opportunities, the stigma associated with welfare receipt increases. Caseworkers become less sympathetic to mothers who do not work and treat welfare applicants more harshly. This causes some eligible mothers to 54 Commentary reject welfare. Caseloads decline further. Because fewer mothers receive welfare, the availability of information on how to receive it decreases. When unemployment increases, higher stigma and less information persist for some period, delaying an increase in caseloads. According to this scenario, norms and values can also affect the composition of the caseload. As welfare becomes more disfavored, advantaged women get jobs while the least advantaged remain eligible but are less likely to participate. Thus, welfare rules can affect take-up rates differently for women with different skill levels. There is no easy way to measure norms and values directly, so it is not surprising to find that empirical evidence on how norms and values influence welfare caseloads and vice versa is at best suggestive. Blank (2001) and Wallace and Blank (1999) find that Democratic governors and state representatives are associated with higher caseloads in a state. Political parties with more liberal attitudes toward welfare may create a political climate in which families feel like they can ask for help and in which state civil servants see their job as helping recipients rather than discouraging them from taking welfare.2Moffitt (1983) argues that changes in stigma explained changes in takeup rates in the late 1960s. Social networks that include welfare recipients increase the likelihood that a person will receive welfare. Gottschalk (1992) finds that among women eligible for welfare, those who grew up in families that received welfare were more likely to receive it themselves than those who grew up in families that were eligible for but did not receive welfare. Bertrand et al. (1999) find that among non-English speakers, exposure to others who speak your language increases welfare use more for individuals from language groups with high welfare use than for individuals from language groups with low welfare use. These studies imply that welfare use results from either shared norms and values or shared information. These effects may not be trivial. Bertrand et al.’s estimates suggest that a policy change that would increase welfare caseloads by 1 percent in a group in the absence of networks can be expected to actually result in an observed increase of between 15 and 25 percent in that group. Although norms and values may not be the most important determinants of caseload changes, they remain an understudied and potentially important source of such changes. Endnotes 1. If take-up has declined, it does not prove that norms and values have changed, since several factors affect take-up rates. 2. Of course, more liberal regimes may also implement more liberal AFDC policies, but these studies try to control for this possibility. References Bertrand, Marianne, Erzo F. P. Luttmer, and Sendhil Mullainathan. 1999. “Network Effects and Welfare Cultures.” Joint Center for Poverty Research Working Paper no. 62, January. Blank, Rebecca M. 2001. “What Causes Public Assistance Caseloads to Grow?” Journal of Human Resources 36, no. 1 (winter): 85-118. Moffitt, Robert. 1983. “An Economic Model of Welfare Stigma.” American Economic Review 73, no. 5: 1023-35. Wallace, Geoffrey, and Rebecca M. Blank. 1999. “What Goes Up Must Come Down? Explaining Recent Changes in Public Assistance Caseloads.” In Sheldon H. Danziger, ed., Economic Conditions and Welfare Reform. Kalamazoo, Mich.: W. E. Upjohn Institute for Employment Research. Gottschalk, Peter. 1992. “The Intergenerational Transmission of Welfare Participation: Facts and Possible Causes.” Journal of Policy Analysis and Management 11, no. 2: 254-72. The views expressed in this article are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever. FRBNY Economic Policy Review / September 2001 55 June O’Neill Commentary F ew policy analysts fully anticipated the extraordinary plunge in welfare caseloads that has occurred over the past few years. In fact, one would have to go back to the surge in caseloads from the mid-1960s to the early 1970s to find a comparably dramatic change in the number of welfare recipients, albeit a change in the opposite direction. Changes of these magnitudes seldom occur in ongoing social programs, and it is noteworthy that the two episodes coincided with significant changes in welfare policy. The surge that started in the 1960s took place in a period of liberalization of the welfare program; the decline in the 1990s occurred in a period of program deliberalization. Given the large magnitude of the caseload changes and their coincidence with policy change, it seems likely that policy played a significant role in the caseload changes. However, other factors also could have contributed. The economic boom of the late 1990s is an obvious candidate for explaining or helping to explain the recent caseload decline (although the economy, which was booming in the late 1960s, seems unlikely to have played a leading role in the welfare surge in the 1965-72 period). Rebecca Blank summarizes the results of the growing number of research studies measuring the separate effects of welfare reform and the economy on the caseload decline of the 1990s as well as on the less studied rise in the work participation of single mothers—the predominant demographic group receiving welfare benefits. The various studies differ considerably in the data and methodology used and in the period of time covered. Blank’s summary provides a useful table succinctly describing those differences along with the major findings of each study. Not surprisingly, the findings differ. But Blank identifies a few results that might qualify as conclusions. She notes the generally consistent evidence that both the economy and welfare policy contributed to the caseload decline of the 1990s. Another finding of a number of studies is that the implementation of welfare reform through state waivers in the period before the Temporary Assistance for Needy Families (TANF) program—1992 to 1996—had a weaker effect on caseload decline than did the implementation of TANF, from late 1996 through 1997. However, the decline in unemployment generally is found to have played a more important role in reducing caseloads in the pre-TANF period than in the post-TANF period, when its contribution was smaller than that of the TANF policy reforms. In a new analysis of the determinants of change in both welfare and work participation, Hill and O’Neill (forthcoming) incorporate one more year of post-TANF observation than has been included in most studies and use microdata rather than the more usual aggregated or caseload data. Nonetheless, their results concerning the effects of policy and economic factors on June O’Neill is the Morton Wollman Professor of Economics at Baruch College and the Director of Baruch’s Center for the Study of Business and Government. The views expressed are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. Blank FRBNY Economic Policy Review / September 2001 57 welfare participation are consistent with the mainstream conclusions described above. Their results differ, however, from those of Schoeni and Blank (2000) with respect to the relative effects of policy and the economy on work participation. Hill and O’Neill find that policy contributed much more to the increase in the work participation of single mothers during the TANF period than did the decline in unemployment, although unemployment tended to be more important than policy in the waiver period. Schoeni and Blank, however, find that only economic factors affect work participation in the TANF years. One reason why Hill and O’Neill’s results might differ is that they restrict the analysis to nonmarried mothers, while Schoeni and Blank include all women in their population sample. The work participation of married women and unmarried women without children is unlikely to be affected by changes in the welfare program, but it certainly could be influenced by the economy. Statistical analysis of the effects of the 1990s welfare reform on various outcomes is bound to be problematic. The changes occur over time and coincide with a major economic expansion, making it difficult to isolate the effect of reform. Moreover, it is difficult to measure the relevant explanatory variables as precisely as one would wish. It is particularly difficult to ascertain the actual content of the welfare reforms initiated in the different states and, even more so, the manner in which they were implemented. However, lack of precision in measuring policy variation would tend to bias results away from finding a strong effect of welfare reform. One important factor that is often neglected in studies of determinants of welfare participation is the potential wage rate welfare recipients could earn if they worked. Some studies have used federal and state minimum-wage levels as a measure of that wage. But this is a questionable practice on several grounds. For one thing, most single mothers who work earn more than the minimum wage. A finding that an increase in the minimum wage is associated with caseload reduction is likely to be the result of a positive association of increases in the minimum wage and increases in the wage level generally. It would be misleading to infer from this finding that increasing the minimum wage would increase earning opportunities for welfare recipients. A minimum-wage increase that boosted the wage above the productivity level of welfare recipients would reduce their employment prospects, not improve them. Moffitt and Stevens Only a few papers that have analyzed the relationship between welfare reform in the 1990s and changes in welfare 58 Commentary participation have examined how the results differ among population subgroups differentiated by skill, race, and other characteristics. Hill and O’Neill (forthcoming) show that the observed percentage point decline in welfare participation among single mothers in the 1990s was greatest for those who might be regarded as having the greatest disadvantages— high-school dropouts, black and Hispanic women, young mothers with young children. Yet one frequently hears the comment that as the caseload declined, those left behind were increasingly disadvantaged. Therefore, the paper by Robert Moffitt and David Stevens, which focuses on the issue of compositional change in the caseload, is particularly welcome. Moffitt and Stevens first provide a conceptual discussion of the expected effects of welfare reform on the composition of the caseload. In general, it is true that welfare recipients are more likely to be those with weaker skills and less education than others because their potential earnings and income off welfare would be lower. Most studies of welfare duration (or of entry onto and exit from welfare) have shown this to be the case. However, that relationship pertains to time periods when Aid to Families with Dependent Children was the nation’s welfare program. The passage of the Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA) and some of the policy changes introduced earlier through state waivers dramatically changed the relative attractiveness of being on welfare. Time limits ultimately restrict the choice, work requirements and tough sanctions alter life on welfare, and the enhanced-earnings disregards adopted in some states make it possible to earn more income without losing benefits. Moffitt and Stevens examine how these and other policy changes would influence women in different circumstances, and conclude that some changes would disproportionately discourage or encourage the less advantaged—while others would similarly affect the advantaged—with no clear net impact. I agree with their conclusion, although I would place somewhat different emphasis on the expected effects of particular policy changes on the different categories of women. The time-limit and work requirements (which in twenty states allow no exemption for mothers of children over the age of six months) represent the most dramatic change for those who would have accumulated more than five years of welfare allotment under the old system and those who are less predisposed to work. These typically are women with low skills, and I would expect the changed policy to reduce their entry onto welfare, as well as to increase their exit rates, disproportionately. Potential recipients would have an incentive to postpone entry to save up the five years of allotment for a rainy day. Some may be shocked into rethinking their life situation and follow a different path: stay in school longer, acquire more work skills, postpone a first birth. However, while the change in policy appears most radical for the disadvantaged, those with more education may respond more quickly because they are better informed and more capable of adjustment. Moffitt and Stevens present two types of empirical analysis to investigate compositional changes in the welfare population. The first, based on data from the Current Population Survey (CPS), investigates how a series of skill-related characteristics has changed among the welfare population compared with the total population of single mothers. They then use regression analysis to identify the effect of PRWORA on these characteristic-intensity measures after controlling for business cycle effects and other factors. The analysis indicates that PRWORA has not been associated with an increased concentration of welfare recipients with disadvantaged traits. The second empirical investigation conducted by Moffitt and Stevens utilizes data on eleven successive cohorts of female welfare recipients from Baltimore who have been followed over five-year periods, starting in 1985. All of the women in each cohort were age nineteen at the start of the five-year period and participated in welfare in at least one of the five years. Six of the cohorts completed the five survey years prior to the enactment of PRWORA, while the rest were increasingly exposed to the policy reform. The purpose of constructing the cohort samples is to observe changes in the characteristics of the caseload before and after the implementation of reform. Unfortunately, the Baltimore data do not contain any independent information on important personal characteristics, such as education, that typically is used to measure skill or disadvantage. It is hard to get around this deficiency, and the attempt to use years on welfare during the five-year observation window does not really work. Changes in the percentage of time spent on welfare by each cohort over the five-year window largely reflect the effects of policy and the economy. Thus, the percentage of time on welfare rises during the early to mid-1990s and then declines at the end of the period. Time on welfare might be a better proxy for level of disadvantage if it referred to duration prior to the period in which exit rates are measured. However, that would require panels of older women as well as a method for adjusting for the effects of the economy. The first portion of the Moffitt-Stevens paper, which analyzes CPS data, provides evidence that the national welfare caseload has not become more disadvantaged as it shrank. That should help dispel the concerns of those who fear that welfare reform has not touched people who lack education and other work-related skills. FRBNY Economic Policy Review / September 2001 59 References Hill, M. Anne, and June E. O’Neill. Forthcoming. “Gaining Ground? Measuring the Impact of Welfare Reform on Welfare and Work.” Manhattan Institute Civic Report. Schoeni, Robert F., and Rebecca M. Blank. 2000. “What Has Welfare Reform Accomplished? Impacts on Welfare Participation, Employment, Income, Poverty, and Family Structure.” NBER Working Paper no. 7627. The views expressed in this article are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever. 60 Commentary LaDonna Pavetti, Michelle K. Derr, Jacquelyn Anderson, Carole Trippe, and Sidnee Paschal Changing the Culture of the Welfare Office: The Role of Intermediaries in Linking TANF Recipients with Jobs Introduction T he Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA), enacted in August 1996, brought sweeping changes to the country’s welfare system. Through the elimination of the sixty-one-year-old Aid to Families with Dependent Children (AFDC) program and the creation of the Temporary Assistance for Needy Families (TANF) block grant, the new law shifted the emphasis of the welfare system from providing ongoing cash assistance to placing welfare recipients in jobs. Local welfare offices have relied on a number of different strategies to shift to a more work-oriented assistance system. Some have expanded the role of former income maintenance (eligibility) workers to include more tasks related to helping welfare recipients find employment, or they have hired additional staff to perform these functions. Others have created closer alliances with, or transferred primary responsibility for employment-related activities to, the local workforce development system. Nearly all have increased their use of “intermediaries”: private or public organizations that act as brokers between the welfare system and employers. Because states have been given so much flexibility to decide how to structure their TANF programs, there has been considerable emphasis on understanding the variation in states’ policy choices. However, there has been far less emphasis on trying to understand how states and local welfare offices LaDonna Pavetti is a senior researcher, Michelle K. Derr a researcher, Jacquelyn Anderson a research analyst, Carole Trippe a senior researcher, and Sidnee Paschal a research assistant at Mathematica Policy Research, Inc. <http://www.mathematica-mpr.com> have reorganized to provide the services TANF recipients need in order to make the transition from welfare to work. Clearly, a state’s policy choices are important; these choices set the parameters that define what is expected of clients, what consequences they will face if they do not meet those expectations, and what benefits will be provided to them and under what circumstances. But a state or local TANF office’s employment service delivery system is an equally important piece of the new social contract we have imposed upon families who turn to the government for support. It is through this service delivery system that TANF clients receive or do not receive the assistance they need to do what is expected of them, which in most states is finding work as quickly as possible. In an effort to increase our understanding of the design and structure of this new service delivery system, this study examines the characteristics of intermediary organizations and explores the role they play in linking welfare recipients with jobs. The study was undertaken because there was widespread belief—but limited systematic evidence—that many welfare offices transferred to intermediaries significant responsibility for helping welfare recipients find jobs. Time limits on the receipt of benefits and full family sanctions for noncompliance with work requirements have increased the importance of providing TANF recipients with the assistance they need to find employment quickly and to maintain it over the long term. To the extent that intermediaries are able to link clients who would be unable to find employment on their own with employers The research for this paper was supported by the U.S. Department of Health and Human Services, Office of the Assistant Secretary for Planning and Evaluation. The views expressed are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. FRBNY Economic Policy Review / September 2001 63 who are willing to hire them, they will contribute to the success of welfare reform. However, if the intermediaries are unable to help clients make these linkages, they could contribute to the failure of welfare reform and increase the risk of clients being adversely affected by new policies such as time limits. Consequently, even though some states historically may have used intermediary organizations to help welfare recipients find employment, the increased emphasis on work increases the importance of understanding their role and what influence, if any, these organizations may have on the success or failure of welfare reform. Review of the Literature To reduce the cost and size of government, state and local governments are increasingly contracting with private agencies to provide public services. In recent years, the privatization of such social services as transportation, mental health care, corrections, health, and education has increased among state agencies (Council of State Governments 1993; Kettner and Martin 1995; U.S. General Accounting Office 1996). City governments have also increased the number and types of public services contracted to private organizations (International City/County Management Association 1994). Given this trend, it is not surprising that welfare offices would turn to “intermediaries” to help welfare recipients find employment. Privatization efforts have had mixed success in reducing government costs and streamlining services (National Academy of Public Administration 1999; Snell 2000). In a study comparing privatized child support services with public agencies in four states, in some cases researchers found that the cost-effectiveness of agencies varied widely with savings achieved, while in other cases costs increased (U.S. General Accounting Office 1996). Kettner and Martin (1993) also found mixed success in the use of performance contracting within human services organizations. In addition, a study conducted by the U.S. General Accounting Office (1997) revealed a number of key factors that contribute to the success of privatizing services, including support from political leadership, attention to the implementation structure, legislative and resource changes that support privatization, reliable cost data, strategies for workforce transition, and effective monitoring and oversight. Overall, the research examining the privatization of employment services is limited. Most of the current literature on the privatization of services in public welfare is geared more broadly to include all human services organizations and does not focus specifically on privatization of employment- or 64 Changing the Culture of the Welfare Office welfare-related services. The research presented here aims to expand our knowledge of privatization as it relates to welfare reform by: • describing the characteristics of intermediaries; • discussing the key decisions local welfare offices have made regarding the use of intermediaries; • providing in-depth information on the types of services intermediaries provide, the process they use to link welfare recipients with employers, and the challenges they face; • identifying lessons that can benefit policymakers and other or newly emerging intermediaries and assessing the implications of the findings for future research on welfare employment efforts. Study Design The devolution of responsibility from the federal government to the states for developing and implementing assistance policies for needy families has spawned a broad range of approaches to transforming the welfare system into a workbased assistance system. To capture the way intermediaries function in these diverse policy environments, we gathered information for this exploratory study through in-depth visits to twenty sites: one urban and one rural in ten different states. Defining an Intermediary Given the broad range of organizations that might be classified as intermediaries in any one community, we sought to develop a definition of an intermediary that would allow valid comparisons across communities. After considering several definitions, we established two criteria that an organization had to meet to be classified as an intermediary for the purposes of this study: • it must provide services that help link welfare recipients with jobs; • it must have a formal relationship with the welfare office or other administrative entity that has responsibility for moving welfare recipients into the labor market.1 Although narrow in some respects, this definition makes it possible to gather and compare information on the universe of intermediaries within select communities in a relatively short time frame and with modest financial resources (see box). We include intermediaries funded with TANF and welfareto-work (WtW) dollars in this study. TANF employment What Is an Intermediary? Intermediary An organization that has responsibility for linking TANF recipients with jobs through a formal relationship with the state or local entity responsible for the administration of TANF or welfare-to-work employment programs. Primary Intermediary An intermediary that operates a job search and placement assistance program targeted to most TANF recipients who are required to find employment. Secondary Intermediary An intermediary that operates a work experience, education, training, supported work, job retention, advancement, or other specialized employment program for a limited pool of TANF recipients. Site Selection programs generally are targeted to the entire TANF caseload while WtW programs are targeted more narrowly to hard-toemploy TANF recipients. TANF employment programs usually are administered by the welfare department, although a state or local community can choose to transfer this responsibility to another organization, such as the Department of Labor or a local workforce development board. The WtW program is administered through the Department of Labor at the federal level and through the workforce development system at the state and local levels. In four of the study sites, both programs were administered by the workforce development system; in the remaining sites, TANF employment programs were administered by the welfare department and WtW programs by the workforce development system. Sites were selected to provide broad regional representation; a mix of large, medium, and small TANF caseloads; different approaches to moving welfare recipients into employment; and a diversity of administrative and service delivery structures. Using these criteria, we selected ten states to include in the study: Arizona, Arkansas, California, Connecticut, Florida, Minnesota, Nebraska, Ohio, Texas, and Virginia (Exhibit 1). Once we selected the ten states, we then chose two communities—one urban and one rural—in which to conduct our analysis. In selecting the urban sites, we chose one of the three largest urban areas in each state. In choosing the rural sites, we limited our pool of potential sites to localities with a Exhibit 1 Study Sites Olmsted St. Paul Scotts Bluff Napa Columbiana Cleveland Hartford New London Omaha Richmond Yavapai Wise San Diego Little Rock Phoenix Jefferson Jacksonville Uvalde Urban study site San Antonio Suwannee Rural study site FRBNY Economic Policy Review / September 2001 65 TANF caseload of between 500 and 1,000 families at the time of site selection. Whenever possible, we selected a rural site that was close to the urban site. We purposefully selected some rural sites because they had exceptionally high unemployment rates or because they had implemented innovative approaches to using intermediaries. Except for a few of the rural communities, sites were not chosen based on their use of intermediaries. the characteristics of intermediaries, key decisions regarding the use of intermediaries, the implementation of the intermediary function, implementation challenges and lessons learned, and ways researchers and policymakers might expand our understanding of intermediary organizations and the role they play in linking welfare recipients with jobs. The Characteristics of Intermediaries Data Collection Information for this study was gathered through site visits conducted between April and August 1999. During these visits, two-person research teams met with staff from the welfare office, an agency from the workforce development system, selected intermediaries, and employers who actively hire welfare recipients through intermediaries. We obtained information on intermediaries with whom we did not interview through meetings with staff from the welfare office and workforce development system, written material collected during our site visits, and an information request sent to individual intermediaries. Data Analysis From the site visits, general information about intermediaries was collected and entered into a database containing information on all of the intermediaries in each site. This information falls into four key areas: 1) program responsibility, which identifies how program responsibilities are allocated among agencies within the services delivery system, 2) payment information, which includes information on how and how much intermediaries are paid for their services, 3) services, which lists the types of services intermediaries provide, and 4) characteristics, which provides basic data on each intermediary, such as type of agency, funding sources, and the types of clients who are served. In all, the database included information on 120 intermediary organizations. A broad range of organizations act as intermediaries for welfare recipients. These organizations include nonprofits, for-profit companies, educational institutions, and government or quasigovernment agencies. The organizations that act as intermediaries bring a broad range of expertise to the task of linking welfare recipients with jobs. The overwhelming majority of the intermediaries in the study sites are wellestablished nonprofit organizations. These organizations account for 67 percent of the intermediaries overall and 74 percent of the intermediaries in the urban sites (see chart). The intermediaries in the rural areas are more equally split among the various types of organizations. While a few sites rely only on nonprofit organizations, most use a mix of nonprofit, for-profit, and public organizations, as well as educational institutions, to link welfare recipients with jobs. The majority of the nonprofit organizations are of two types: 1) local entities or local affiliates of national organizations (such as the Urban League, Salvation Army, Goodwill) that have a long history of providing employmentrelated services to disadvantaged populations and 2) organizations with expertise in addressing the supportive service, and sometimes the employment, needs of special populations such as ex-offenders, persons with disabilities, or Types of Organizations Acting as Intermediaries Number of organizations 100 Nonprofit For-profit Educational 80 60 40 Study Findings 20 Information gathered from site visits was synthesized into several key findings. In this section, we summarize and discuss 66 Changing the Culture of the Welfare Office 0 Urban Rural Total Public persons who speak limited English. Only a few nonprofit organizations are new to the communities in which they provide services or have no experience providing employment services to or working with welfare recipients. Represented among the for-profit intermediaries are organizations that have been providing employment services to welfare recipients for many years and organizations that are new to the employment service arena. Most of the for-profit intermediaries are large organizations with a national presence, although a few are smaller local organizations. The educational institutions that act as intermediaries include community colleges, adult education programs, and local school districts. The public or quasi-public agencies that act as intermediaries include city governments, local employment and training agencies, and public housing authorities. For-profit companies account for a relatively small share of all intermediaries in the study sites. However, because most for-profits serve large numbers of TANF clients, they expect to serve almost half of all TANF recipients who are referred to intermediaries for services. On average, the intermediaries included in this study expect to serve 370 TANF clients per year, but the range of clients served is wide, with the smallest intermediary expecting to serve only twenty-five recipients and the largest expecting to serve 4,000. On average, for-profit organizations expect to serve 985 clients, compared with 240 for nonprofits. Forty percent of the for-profit intermediaries in the study sites expect to serve more than 500 clients, compared with only 10 percent of the nonprofit organizations. For-profit organizations are projected to serve 45 percent of the total TANF clients to be served by intermediaries, even though they account for only 15 percent of the intermediaries. Key Decisions Regarding the Use of Intermediaries Within a work-based assistance system, a broad range of tasks must be performed to provide families with cash assistance and help them make the transition to self-sufficiency. The primary employment-related services provided to most TANF recipients are case management and job-search and placement assistance. Secondary employment-related services, provided on a more limited basis, include work experience, education, training, supported work, job retention, and advancement programs. In deciding how to use intermediaries to provide these services, local welfare offices or their designees face three key decisions: 1) how much responsibility to transfer to intermediaries, 2) whether to transfer responsibility to a single or multiple intermediaries, and 3) how and how much to reimburse intermediaries for the services they provide. Using these three key decisions as our framework, we examine the choices the local sites made regarding how to use intermediaries to help welfare recipients make the transition to employment. How Much Responsibility to Transfer to Intermediaries Localities transfer to intermediaries various levels of responsibility for providing employment-related services. While some localities transfer responsibility for job-search assistance and case management, others transfer responsibility only for job search and some do not transfer any responsibility. Of the twenty study sites, eighteen transfer some responsibility for providing employment-related services to intermediaries. Due to their smaller size, it is less common for rural offices to transfer responsibility for employment-related services to intermediaries; the two sites that do not transfer any responsibility to intermediaries are both rural sites that provide all employment-related services in-house or rely on existing resources in the community. (Sites were not selected for this study based on their use of intermediaries. Thus, prior to conducting the study, we did not know whether the sites had transferred any responsibility to intermediaries.) The majority of the study sites—seven urban and five rural—transferred responsibility for case management and job-search assistance to intermediaries. When case management responsibilities are transferred, intermediaries are responsible not only for linking TANF recipients with jobs but also for assessing client needs, working with clients to develop self-sufficiency plans, and linking clients with the resources they need to achieve the goals outlined in their plans. Given the emphasis on shifting the focus of the welfare office from determining eligibility to helping TANF recipients make the transition to unsubsidized employment, it is notable that so many of the sites transferred primary responsibility for providing case management services to intermediaries. When case management responsibility is transferred to intermediaries, welfare office staff often are responsible only for eligibility determination, just as they were under the AFDC program. Only four of the sites have expanded the role of former eligibility staff to include case management responsibilities. The other sites that have not transferred all responsibility for case management to intermediaries have separate case management staff, usually working in a specialized unit, who provide case management and/or job-search assistance to all or a portion of the TANF caseload. When these units exist, they often function and are treated the same as other intermediaries. FRBNY Economic Policy Review / September 2001 67 Whether to Transfer Responsibility to One or Multiple Intermediaries Most of the urban sites, but only a few of the rural sites, transferred responsibility for providing job-search assistance and/or case management to multiple intermediaries. Seven of the urban sites and three of the rural sites transferred responsibility for providing job-search assistance and/or case management assistance to multiple intermediaries. Especially in the urban sites, the number of intermediaries determines how many clients each intermediary will serve. Some sites have a small number of intermediaries that each serve a large number of clients while others have a larger number of intermediaries that each serve a smaller number of clients. In the urban sites, clients are allocated to multiple intermediaries based on geography or a discretionary process, with each intermediary providing the same services to a portion of the TANF caseload. In the rural areas, multiple intermediaries’ functions are more specialized, providing employment services to specific subgroups of the TANF caseload or a narrowly defined set of employment services to all TANF clients. The decision to transfer responsibility to one intermediary or multiple intermediaries has important implications for the range of organizations that act as intermediaries. Generally, when multiple intermediaries are selected to provide services, the range of organizations that act as intermediaries is broader—small-community-based organizations can compete and provide services alongside large for-profits. Especially in urban areas, when intermediaries are required to serve large numbers of clients, many smaller nonprofits do not have the expertise or capacity to compete with large for-profit organizations. The study sites initially focused their employment-related efforts on increasing their capacity to provide job-search assistance for applicants and recipients who are required to find employment. Now that these services are in place, sites have begun to expand the employment-related services to include options other than job search. These options include short-term training, subsidized employment, specialized services to promote job retention and advancement, and specialized services for the hard-to-employ. Few sites provide all of these services. Instead, individual sites have focused their efforts on a few of these options. Often these services are provided through the Department of Labor’s Welfare-to-Work program and operate outside of the primary TANF employment service system. So far, these programs have served a relatively small number of recipients. While some of the intermediaries that provide these more specialized services also provide job-search assistance, most do not. 68 Changing the Culture of the Welfare Office How and How Much to Reimburse Intermediaries for the Services They Provide In addition to making critical decisions about how much responsibility to transfer to intermediaries and how to structure the delivery of services at the local level, local welfare offices or their designees must also decide how and how much to reimburse intermediaries for the services they provide. The most common payment structures are cost-reimbursement, where organizations are paid for the costs they incur, or payfor-performance, where organizations are paid based on their accomplishments. Prior to welfare reform, most welfare offices paid intermediaries on a cost-reimbursement basis. Because there has been a general shift toward developing more performance-based arrangements with contractors, and the stakes are much higher under TANF than under previous welfare employment efforts, there has been more interest among welfare offices in considering which option would work best. The experiences of the study sites suggest that while a few localities have shifted to performance-based payment arrangements, most still reimburse intermediaries on a costreimbursement basis. Some localities combine the two methods of payment, reimbursing the intermediary for part of its costs through a cost-reimbursement mechanism and the remainder through a performance incentive structure. The local sites that rely on cost-reimbursement payment mechanisms often include performance criteria in their costreimbursement contracts and evaluate the success of their intermediaries against these criteria. Critics of pay-for-performance reimbursement mechanisms argue that this payment structure encourages program operators to “cream,” that is, to provide services to job seekers who are the most likely to succeed rather than to those most in need of assistance. Critics of cost-reimbursement payment systems argue that program operators get paid even if the services they provide do not produce results, wasting taxpayers’ money and reducing incentives to meet high performance standards. It is too soon to know whether the way in which intermediaries are reimbursed for their services or the amount they are paid influence program outcomes. Welfare offices or other relevant administrative entities that reimburse intermediaries on a cost basis believe that they can demand high levels of performance from intermediaries as long as clear program goals are established and performance is monitored on an ongoing basis. Those that reimburse intermediaries based on performance believe that pay-for-performance systems play a critical role in emphasizing the importance of placing recipients in jobs, not just engaging them in employment preparation activities. All agree that administering a pay-forperformance reimbursement system is much more complicated than administering a cost-reimbursement system. Regardless of the way in which intermediaries are reimbursed for their services, there is wide variation in the amount that intermediaries are paid for these services. This variation exists between the sites and between intermediaries within some of the sites. In the eight study sites where we were able to obtain comparable reimbursement data, intermediaries were paid as little as $355 and as much as $6,250 per recipient served (see table). Some, but not all, of this variation reflects differences in the services provided. On average, intermediaries that provide only job-search and placement assistance are reimbursed $1,320 per person, while those that provide specialized employment services are reimbursed an average of $2,970 per person. Comparisons across four of the urban sites that used multiple intermediaries to provide primary TANF employment services suggest that there is considerable variation within and between the sites in how much intermediaries are reimbursed, even when they provide similar services. The average per-person reimbursement across the four sites ranges from $1,045 to $2,360. The sites with the highest and lowest average reimbursement provide comprehensive services—job-search and placement assistance and case management—to TANF clients, suggesting that differences in the range of responsibility transferred to the intermediaries do not fully account for the variation in the amount they are reimbursed for their services. In three of the four sites, the minimum and maximum payment amounts vary dramatically, even though the intermediaries have responsibility for providing the same services. In one site, the highest-paid intermediary is paid almost four times the lowestpaid intermediary. In sites where payments are comparable across intermediaries, program administrators negotiate a similar price with intermediaries regardless of how much they indicate it will cost to provide services. In sites where there is considerable variation, program administrators accept the price set by intermediaries in their response to the agency’s request for bids to provide services. The Implementation of the Intermediary Function In a work-based assistance system, intermediaries are an important link in a complex process that starts at the welfare office and ends when a recipient is placed in a job. Features that distinguish one intermediary from another include the structure of their job-search programs, their ability to link clients with ancillary services, and the extent to which services are provided after a client finds employment. Another important feature that distinguishes one intermediary from another is the approach to job development, especially the ability to establish ongoing working relationships with employers. Key to this success is the development of strong links to, and ongoing communication with, the welfare office or other administrative entity that controls the flow of TANF clients to them. In this section, we examine the process through which intermediaries link welfare recipients with jobs. We start by examining how welfare recipients are linked with intermediaries. We then discuss the services intermediaries provide to prepare recipients for employment and how they identify job openings to place them in employment. Linking TANF Recipients with Intermediaries The path that a welfare recipient takes to get to an intermediary ranges from a simple referral from the welfare office to a complex chain of referrals from one intermediary to another. The process of linking welfare recipients with intermediaries is complex and highly dependent on the service delivery structure in which intermediaries operate. As a result, there is considerable variation in the way in which welfare recipients are linked with intermediaries and the ease with which this process occurs. The success that sites have in linking welfare recipients with intermediaries is determined in part by how streamlined the referral process is and how well the different agencies communicate. Regardless of how much responsibility is transferred to intermediaries, the referral process starts at the welfare office, usually when an eligibility worker determines whether a TANF applicant or recipient is required to look for work (Exhibit 2). The actual transfer of clients to an intermediary ranges from an automatic electronic transfer to a more complicated decisionmaking process that takes into account client needs and the unique characteristics of intermediaries. In most sites, staff from the welfare office refer TANF clients directly to intermediaries. In a few sites, clients are first referred to the workforce development system and then to intermediaries. In sites where responsibility for case management is transferred to intermediaries, staff from the welfare office make the initial referral to an intermediary, but all subsequent referrals to other intermediaries are made by an intermediary. FRBNY Economic Policy Review / September 2001 69 To enforce mandatory participation requirements and achieve high work participation rates, the referral process is often tightly defined and monitored, making it difficult for intermediaries outside the primary TANF employment system to receive referrals. In all of the local sites, participation in employment-related activities is mandatory. Most of the sites have developed their referral and client monitoring systems expecting that clients will participate in programs offered by intermediaries directly under their purview. In developing these systems, the organizations responsible for managing TANF employment programs aim to achieve two different goals: 1) to ensure that clients who are mandated to find work have access to job-search and placement assistance and 2) to ensure that the intermediaries to which they have transferred responsibility for providing these services have the opportunity to provide them. In the sites where multiple intermediaries provide job-search and placement assistance, intermediaries generally did not feel that they were competing with each other for clients. However, the situation is quite different for intermediaries providing services other than job search. In sites where TANF and welfare-to-work employment programs are operated by different entities, WtW intermediaries often have difficulties (over and above those related to eligibility criteria) receiving referrals for TANF clients. In some sites, WtW providers are dependent upon other intermediaries to refer clients to them; in others, they are dependent upon welfare office staff to consider them along with primary TANF employment intermediaries as potential service providers for their clients. Especially in sites where there is excess service capacity, welfare administrators who encourage referrals to WtW providers run the risk of having even greater excess capacity among their own providers. When the primary TANF employment and the WtW programs are managed by the same administrative entity, it is easier for WtW and TANF providers to receive equal consideration. As WtW intermediaries become more established and their programs more distinguishable from those provided by TANF intermediaries, some of the issues WtW intermediaries currently face may be alleviated. Reimbursement per Person for Employment Services Provided by Primary and Secondary Intermediaries Type of Site Method of Reimbursement Four urban sites with multiple intermediaries Site 1 (comprehensive servicesa) Site 2 (job search and placement) Site 3 (comprehensive services) Site 4 (job search and placement) Minimum (Dollars) Maximum (Dollars) Cost Performance Cost Performance 1,900 1,100 935 580 3,055 3,995 1,135 2,520 2,360 2,130 1,045 2,090 NA 580 635 355 1,000 6,250 4,640 4,775 5,000 1,785 2,390 2,660 1,680 NA 400 930 355 1,010 5,000 3,055 6,250 5,000 1,320 1,825 2,605 2,970 745 4,745 3,685 Type of organizationb Nonprofit For-profit Educational Public Type of services provided (TANF)b Job search and placement Comprehensive servicesa Training Specialized (hard-to-employ) Welfare-to-work Note: TANF is the Temporary Assistance for Needy Families program. a Comprehensive services include case management and job-search and placement assistance. Based on data from eight sites: San Diego, Calif.; Napa County, Calif.; Hartford, Conn.; St. Paul, Minn.; Olmstead, Minn.; Cleveland, Ohio; Columbiana County, Ohio; and Richmond, Va. b 70 Average (Dollars) Changing the Culture of the Welfare Office Exhibit 2 Linking Welfare Recipients to Intermediaries: Referral Models in the Study Sites Welfare Office Case Management Modela Other service provider Welfare office Intermediary Intermediary Case Management Modelb Welfare office Intermediary Intermediary Other service provider Workforce Development System Case Management Modelc Workforce development system Welfare office Intermediary Workforce Development System Progression Modeld Welfare office Workforce development system Intermediary 1 (assessment) Intermediary 2 (case mgmt., job search) Intermediary 3 (placement) aSites: Cleveland, Columbiana County, Jefferson County, Phoenix, Richmond, Scotts Bluff, Yavapai County. bSites: Jacksonville, Little Rock, Napa County, Olmstead County, Omaha, San Antonio, San Diego, St. Paul, Suwannee County, Uvalde County. cSite: Hartford. dSite: New London. The Services Intermediaries Provide Intermediaries that provide job-search and placement assistance to welfare recipients differ little in the specific services they provide. These intermediaries do, however, differ in their approach to providing these services and the context in which the services are provided. In a work-first environment, the primary effort intermediaries are engaged in is preparing TANF clients to enter the labor market as quickly as possible. Thus, most intermediaries that provide job-search assistance and/or case management provide a fairly standard set of services, including assessment, orientation, job-search skills development, and post-placement assistance. The dimensions on which these programs differ are often quite subtle and include factors such as 1) the extent to which they assess client strengths, needs, and employment interests, 2) the amount of guidance provided to TANF recipients to help them find employment, and 3) the amount of emphasis placed on the development of jobreadiness skills and/or addressing job-retention or advancement issues. Intermediaries also are distinguished by their ability to link TANF clients with ancillary services. Intermediaries that provide comprehensive services to disadvantaged families often are able to access a broader range of services for their TANF clients than intermediaries that provide only job-search assistance. FRBNY Economic Policy Review / September 2001 71 Linking TANF Recipients with Employers In the current economic climate, it is relatively easy for most intermediaries to link job-ready TANF recipients with employment opportunities. Still, intermediaries rely on a variety of strategies to help TANF clients find employment. An intermediary’s success in linking welfare recipients with employment is crucial to the short- and long-term success of the organization. Finding employment for job-ready welfare recipients in the current economic environment is an easy task for most intermediaries; employers are looking for qualified employees and are eager to work with intermediaries who can supply them with job-ready applicants. Intermediaries use a broad array of strategies to link welfare recipients with jobs. For the most established intermediaries, job development often involves filling job orders for employers. In other instances, intermediaries build relationships with employers by inviting them to participate in job fairs and mock-interviewing sessions with job seekers, or by creating internships and work experience programs that allow employers to “test out” clients. Job developers in all but the most established intermediaries also rely on “cold calls” to employers with whom they have not developed a relationship. Implementation Challenges and Lessons Learned This research was designed to be exploratory in nature. Thus, it represents a first step in trying to understand which organizations are acting as intermediaries and the role they play in linking welfare recipients with jobs. In this section, we discuss the challenges that the intermediaries in the study sites encountered and present broad lessons that can be gleaned from their experiences. The Importance of Clearly Defined Roles and Responsibilities In many of the sites, numerous organizations are involved in providing assistance to TANF clients. Consequently, clearly defined roles and responsibilities and procedures for transferring information between organizations are critical to the successful operation of a work-based assistance system. 72 Changing the Culture of the Welfare Office Intermediaries are operating in a complex policy and administrative environment. Regardless of how TANF is administered and how much responsibility is transferred to intermediaries, the process of linking welfare recipients with jobs is a shared responsibility. Welfare office staff remain responsible for referring clients to intermediaries, imposing sanctions on clients who do not participate in work-related activities, and authorizing work supports such as food stamps and Medicaid when clients are no longer eligible for cash assistance. When the welfare office and the workforce development system are both involved in the administration of TANF or providing employment-related services to TANF recipients, clearly defined roles and responsibilities and clear procedures for transferring information between agencies are even more critical. Unfortunately, many state or local automated data collection systems were not designed with intermediaries in mind. As a result, the development of clear roles and responsibilities often requires establishing detailed—and sometimes cumbersome—procedures for transferring information between agencies. As a result, it is an ongoing challenge to develop and maintain a system of communication that provides all involved parties with the information they need and that is not overly burdensome on front-line staff. Risks in Providing Services to TANF Clients Intermediaries are operating in a new and changing environment, where the flow of clients is rarely steady and predictable. Some intermediaries are serving more clients than they anticipated, while others are serving less. All intermediaries struggle with high no-show rates among the TANF clients referred to them. When intermediaries enter into a formal agreement with the welfare office or its designee, they do so with the expectation that they will serve a specified number of clients. However, in a rapidly changing environment, it has been difficult to predict accurately how many TANF recipients will need to be served by intermediaries. In some of the urban sites, intermediaries are serving more clients than they anticipated. In the sites with the largest caseload declines, intermediaries are serving far less TANF clients than they anticipated. Even when intermediaries receive sufficient referrals, they have to account for extremely high levels of nonparticipation. Intermediaries report that they generally can expect only about half of the clients referred to them to participate in the program. High no-show rates reduce the number of clients an intermediary can serve and create a huge paperwork burden, since clients who do not show up for services usually are referred back to the welfare office for sanctioning. In an effort to reduce the number of clients who do not participate in their programs, a few intermediaries have put outreach activities into place. Outreach activities include calling the client the day before he or she is scheduled to begin participation and sending follow-up reminder cards. Other outreach activities are more intensive, and may include visits to clients at home. experience, there is no evidence to suggest that one particular strategy for transferring responsibilities to intermediaries will produce better results than another. Instead, what appears to matter is creating an infrastructure that builds on the strengths of the local community. It is also important to note that the decisions one makes regarding how much responsibility to transfer to intermediaries can affect the types of organizations that are qualified to function as an intermediary. In particular, when responsibilities are broadly defined and the number of clients to be served is large, nonprofit organizations may be less likely to act as an intermediary than large for-profit organizations with a national infrastructure. Serving the Hard-to-Employ As TANF caseloads decline, intermediaries are concerned that there is a mismatch between the limited services they are being asked to provide and the needs of the clients they are being asked to serve. As TANF caseloads decline, many intermediaries feel that they are working with more clients who have multiple barriers to employment. Most intermediaries believe they could do a better job serving these families if they had more time to work with clients and could provide a broader range of services. Over time, it is possible that jobsearch programs will be redefined to address the more diverse needs of the families remaining in the TANF caseload. There may also be an increasing demand for longer term supported work programs. Given the more specialized knowledge required to address the needs of some families with chronic barriers to employment, it is possible that a new set of intermediaries will be called upon to provide these services. Alternatively, existing intermediaries may begin to collaborate with organizations that have more expertise in providing these more specialized services. Building on the Strengths of the Local Community There are a variety of ways to transfer employment-related responsibilities to intermediaries. Given that localities have different resources, needs, and priorities, a service delivery structure that works in one locality may not necessarily work in another. The local sites examined for this study transferred responsibility to intermediaries in a number of different ways. The decisions they made reflected differences in their in-house resources, administrative structure, prior experience with intermediaries, and perceptions of the relative effectiveness of government and the private sector. Based on their early Next Steps: Expanding Our Knowledge Base This study has provided one of the first examinations of the role intermediaries are playing to help welfare recipients find employment. Clearly, intermediaries are an important part of a complex array of actors that are attempting to help welfare recipients find and maintain stable employment. Therefore, their ability to link welfare recipients with jobs may substantially influence the overall success of localities’ efforts to reform the welfare system. Especially over the long term, it would broaden our understanding of welfare reform if we explored the role of intermediaries in further detail. There are no accepted standards on how to measure performance in work-first programs, making it difficult to compare performance across programs. Commonly used measures of performance include program enrollment, program completion, job placement, and job retention. However, even around these measures there is considerable variation in what constitutes “success.” In some programs, clients may have to work in a job for a minimum period of time, such as thirty days, before they can be counted as a successful placement, while in other programs, clients may have to work for only one day. Gaining a better understanding of how programs define success and considering the relative merits of various measures is a critical first step in being able to identify the characteristics of successful programs. Currently, there is no evidence on whether intermediaries with certain characteristics perform better than others. Investing in research to examine this issue could potentially help local welfare offices develop more effective TANF employment service delivery systems. In the current environment, many intermediaries are being asked to provide FRBNY Economic Policy Review / September 2001 73 the same set of services to welfare recipients. However, intermediaries differ in a number of dimensions that may influence their performance. Key characteristics that may influence performance include 1) the number of clients served, 2) a previous history of providing employment-related services, 3) an expertise serving hard-to-employ populations, 4) the payment mechanism, 5) the payment amount, 6) the type of organization, 7) links to the business community, and 8) the administrative structure in which the intermediary is operating. Work-first programs—consisting primarily of job-search and placement assistance—are at the heart of most current efforts to increase employment among welfare recipients. As these programs become more established, it would be useful to know whether one work-first approach is more effective than another. Job-search assistance is the core service provided by most primary intermediaries. While these programs are similar in many ways, often there are subtle differences. Some of the dimensions on which these programs vary include 1) the length of the program, 2) the amount of structure, 3) the level of employer involvement, 4) the extent to which life skills issues are addressed, and 5) the length and extent of follow-up. Currently, there is no information available to indicate whether different approaches to providing job-search assistance have any influence on program outcomes. Additional information on what makes a “good” job-search program may help to improve the overall quality of these programs. 74 Changing the Culture of the Welfare Office Conclusion In many communities, intermediaries provide the primary link between welfare recipients and the paid labor market. Although a service delivery system that effectively links the welfare office, the workforce development system, and intermediaries is in place in some communities, in others an integrated service delivery system is still being created. Given the changing nature of the Temporary Assistance for Needy Families caseload and shifting priorities, the system for providing employment-related services to TANF clients is likely to be in transition for some time. Over the next several years, states and localities will be implementing the Workforce Investment Act, which may encourage some local communities to again rethink how they transfer responsibility to intermediaries. Examining how these transitions take place and how they affect the role intermediaries play in linking welfare recipients with jobs will help to broaden our knowledge of what it takes to create a stable work-based assistance system. Understanding the implementation of welfare reform is an extremely complex undertaking. Clearly, such implementation cannot be understood fully without taking into account the role intermediaries play in linking welfare recipients with jobs. Because many implementation decisions are being made at the local level, the focal point for many implementation studies is the local welfare office. This study suggests that in some communities, the scope of inquiry may need to expand beyond the welfare office. This is especially true for the analysis of implementation issues that involve significant worker-client interaction, such as assessment practices, the implementation of sanction policies, and efforts to link clients with ongoing work supports such as food stamps and Medicaid. Although we often think of these tasks as within the purview of welfare office staff, it is clear that intermediaries have an important role to play in making sure that clients are aware of what is expected of them and the benefits to which they are entitled. Endnotes 1. In an effort to maintain a focus on intermediaries that link welfare recipients with jobs, we explicitly excluded two potentially large groups of organizations that often are thought of as intermediaries. These are 1) organizations that provide only support services (such as child care, transportation, or legal assistance) and 2) organizations that offer only education or training services without a job-placement component (such as adult basic education and general equivalency diploma programs and some community college education or training programs). FRBNY Economic Policy Review / September 2001 75 References Council of State Governments. 1993. State Trends and Forecasts: Privatization, vol. II: no. 2. November. Lexington, Ky. Snell, L. 2000. Child-Welfare Reform and the Role of Privatization. Los Angeles: The Reason Foundation. International City/County Management Association. 1994. International City/County Management Association Municipal Year Book 1994: Alternative Service Delivery in Local Government, 1982-1992. Washington, D.C. U.S. General Accounting Office. 1996. “Child Support Enforcement Privatization: Challenges in Ensuring Accountability for Program Results.” GAO/T-HEHS-98-22. Washington, D.C.: General Accounting Office. Kettner, P. M., and L. L. Martin. 1993. “Performance, Accountability, and Purchase of Service Contracting.” Administration in Social Work 17, no. 1: 61-79. ———. 1997. “Privatization: Lessons Learned by State and Local Governments.” GAO/GGD-97-48. Washington, D.C.: General Accounting Office. ———. 1995. “Performance Contracting in the Human Services: An Initial Assessment.” Administration in Social Work 19, no. 2: 47-61. Yates, J. 1997. “Case Studies on Nonprofits’ Involvement in Contracting for Welfare Services.” Welfare Information Network, October 27, 2000. <http://www.welfareinfo.org/case.htm> National Academy of Public Administration. 1999. Privatization 1999, the Reason Foundation Public Policy Institute’s 13th Annual Report on Privatization. Washington, D.C.: Alliance for Redesigning Government. The views expressed in this article are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever. 76 Changing the Culture of the Welfare Office Kathryn Edin and Rebecca Joyce Kissane Commentary L aDonna Pavetti and her coauthors have described an interesting study in several carefully and logically chosen sites. The goal of the study is “exploratory” and descriptive—to see what role “intermediaries” are playing as the story of welfare reform unfolds. Such studies can be thought of as a special type of process-implementation study, a research genre that has enjoyed a renaissance in the welfare reform era. Process-implementation studies are valuable for specifying what the “treatment” is and thus help to unpack the “blackbox” process between policy and product. We seek to place this study in a broader context. As it stands, the study focuses on what is occurring with the welfare department as well as on the type of services that have been devolved to intermediary organizations. Howard Becker’s classic essay “Whose Side Are We On?” (1967) admonished scholars to look at a given problem from all points of view. The paper takes the point of view of the local welfare department, but the intermediaries themselves and the clients that must rely on such intermediaries for services also deserve some interest. Thus, the story is only one-third told. Enlarging our scope to encompass the agency and client points of view takes us somewhat out of the welfare reform context and into the larger world of low-income families and the various public and private institutions that attempt to serve them. To illustrate why such points of view might be of interest to those who make or study policy for the poor, we draw on Kathryn Edin is an associate professor at Northwestern University’s Department of Sociology and a faculty fellow at Northwestern’s Institute for Policy Research; Rebecca Joyce Kissane is a Ph.D. candidate at the University of Pennsylvania’s Department of Sociology. three sources: an ethnographic study of two large multiservice nonprofits (NPs) conducted in the early 1990s (Edin and Lein 1998); a longitudinal study (with two rounds of interviews conducted between 1998 and 2000 under the auspices of Manpower Demonstration Research Corporation) of thirty nonprofits in a single, large metropolitan area that serve welfare clients;1 and a small exploratory study of the prospective clients of these same nonprofits (Kissane 2001). We deem these studies of nonprofits and their actual and prospective clients relevant because most intermediary agencies identified by Pavetti and her colleagues were nonprofits. Some of the NPs we refer to in these studies were the types of intermediary agencies that Pavetti et al. studied: they had a contract with the welfare department or another entity that controlled welfare dollars, such as welfare-to-work. Other NPs received welfare dollars more indirectly through reimbursement for services rendered (such as child-care or support services). Finally, some of our NPs received no welfare or other government monies, but some had considered doing so. What’s Happening with the Agencies Welfare reformers hoped that nonprofit social services agencies would work together to weave a private safety net to partly The views expressed are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. FRBNY Economic Policy Review / September 2001 77 neither helped the children nor expressed their own values. In the end, they gave up the contract in order to regain control over their program management. Our second example of how NPs may alter their function is more directly relevant to the welfare reform context. The example is drawn from the longitudinal study of thirty nonprofit social service agencies in Philadelphia, all of which serve welfare clients. One complaint sometimes voiced by directors is that “all the dollars” were going to programs that took a “work-first” approach.3 Agencies feared that they would not secure, or would risk losing, funding if they did not offer “rapid immersion” or incorporate job readiness into their current programs. Some of these NPs had long offered job-placement programs for welfare clients, but these programs generally were focused more on employability (that is, education, training, mental health, soft skills, supportive services) than on rapid immersion into the labor force. Some lost their funding in the mid-1990s, prior to our first round of interviews. At least two directors suspected that this was the result of a work-first bias on the part of their funders. One director of a jobplacement program described the trend in funding as follows, “Agencies are in a pickle because the funders are all ‘jobs, jobs, jobs’ . . . that’s all they want from the social services. But we know that more needs to be done than just that. You need to juggle the stuff that we already do [like basic education] with the jobs stuff.” Another director reflected, “[I tell the staff,] ‘let’s move these people.’ But are we really doing a service [just shoving them out]? And we don’t want to—and we really can’t because we are a small agency and because we are a community [agency] and because our staff is representative of the community, there is a lot more empathy to what the real issues are. I find the staff truly, truly struggling with these issues.” Pavetti et al. find some of the same forces at work (see also De Vita [1999]; Smith [1999]). replace the public safety net they were dismantling. Although this expectation caused considerable scholarly debate (for example, see De Vita [1999]; Diaz et al. [1996]; Smith [1999]), it caused surprisingly little public debate. To the extent that policymakers and the general public are relying on these agencies to provide a sort of private safety net as the public safety net shrinks, the health and financial well-being of such agencies should be of concern. The various NPs that we and our colleagues have studied have expressed many concerns over the eroding public safety net and their actual or prospective role in substituting their services for the ones the government used to provide. As part of the Urban Change study, data from NPs in four cities show that NPs have largely not taken on the mantle imposed on them by welfare reform’s architects and have few plans for meeting the demands that time limits or other aspects of welfare reform might impose (Fink and Widom 2001). Our Philadelphia NPs, which also participated in the above study, have reacted similarly. Some have taken on government contracts or have received welfare dollars for services rendered to welfare clients. Yet even if they receive no government money, all of our NPs serve welfare clients, and as we have talked to the directors and observed the daily operations of some agencies, we have identified several ways in which welfare reform might be affecting them. 1. 78 Change of Function. Scholars have long argued that nonprofits that take on government contracts might profoundly change in ways that might not serve the interests of their clients (Lipsky and Smith 1989-90; Lipsky and Smith 1995; Smith 1999). Our example of this phenomenon is drawn from an NP that received a government contract for the first time (see Edin and Lein [1998]). Although the contract was not with the welfare department, the example below illustrates the pressures that government contracts of any kind may place on a small grassroots nonprofit. At the beginning of Laura Lein’s study of several nonprofits in the early 1990s, “All Service” accepted federal support to provide meals for eligible neighborhood children.2 In order to comply with the program guidelines of the funder, the agency had to guarantee that only children under seventeen would receive the meals. While the children were fed in a fenced-in, outdoor pavilion, other community members occasionally gathered outside the facility to watch the meal in progress. In this Mexican-American community, children are expected to follow social norms of mutual exchange and offer food to other people who wished it. Thus, agency volunteers soon found themselves intervening to prevent children from sharing their food with other people they knew. The staff quickly realized that they were spending valuable time and energy on a regulatory activity that Commentary 2. The Paper Trail. Some directors in the longitudinal study of nonprofits complained about what they viewed as the ruination of the organizational culture that occurs when NPs begin to rely on government dollars. One director of a grassroots agency pointed to several other NPs that “used to be smaller, social service/advocacy organizations in the neighborhood,” but had become “bureaucratic nightmares themselves” and are now “arms of the department of public welfare.” The agencies that had taken on such contracts talked at length about the added accountability and paperwork burdens they now faced. More bureaucracy means higher overhead costs, and unless a nonprofit has funding that allows for crosssubsidization of programs or specifically for overhead (both of which are increasingly rare), nonprofits are left financially vulnerable, a topic we turn to next. 3. Show Me the Money. Another complaint voiced by agency personnel in the longitudinal study of nonprofits— particularly those that have actually had contracts with the welfare department or received welfare-to-work dollars—relates to the financial risk involved in serving as an intermediary agency. Pavetti et al. report that “the most common payment structures” in their study “are cost reimbursement, where organizations are paid for the costs they incur, or pay-for-performance, where organizations are paid based on their accomplishments.” Many of the reimbursement contracts also had performance incentive structures built in and/or performance criteria that had to be recorded for evaluation purposes. Both cost-reimbursement and performance-based contracts require NPs, which are often on a shoestring budget (Urban Institute 1983), to take huge financial risks. Agencies with either type of contract faced a similar problem: getting clients. Pavetti et al. also find that the number of clients referred to intermediaries is often unpredictable. Theoretically, clients are supposed to come from the welfare department, via caseworker referral. One director we interviewed whose agency had a performancebased contract waited eight months before seeing a single client for the program. Because of bad past experiences with other such contracts, she had decided not to hire any program staff until the clients had actually arrived. Though her prudence probably saved the agency from financial disaster, she believes that clients may have suffered from her wariness because when they did arrive, she had no trained staff to meet them. Reasons for the lack of caseworker referrals remain mysterious, both to the NPs and to us, and we were fascinated to see that the Pavetti et al. sample reports similar problems. The problem could lie with the caseworkers, with the intermediaries, or with potential clients, for reasons we speak directly to below. One can well imagine that overworked caseworkers might have little time to keep track of which agencies are offering services when and which clients are eligible. However, Pavetti et al. find that even when the welfare department did make referrals, the clients would only show up and enroll in the program about half the time. Furthermore, when nonprofits do manage to get clients, clients do not always “perform” to the standards of the contract. One director we interviewed pointed out the obvious financial difficulty this imposed on the agency—an agency cannot base staff pay solely on client performance. Even for reimbursement-based programs, like those that Pavetti et al.’s respondents most often described, the process of securing a contract and establishing a program nearly always required large up-front investments that NPs feared may not be recouped for a long period of time. One director of a very large agency told us that they had applied for a performance-based contract funded by welfare-to-work. The agency was awarded the contract, but after talking with other NPs that had received such contracts, it turned down the money. The director said, “I was so glad that we turned it down, because it sounded like it was a nightmare. . . . I talked to a number of other agencies and they all had such nightmare stories, I decided this is crazy. [We would have to pay staff] and we wouldn’t get the money until people were in jobs for three months, six months, a year.” She went on to complain that, “the department of welfare was supposed to be feeding us the clients and they weren’t feeding the clients to these other agencies, so I decided that it just wasn’t worth it for us to do it. Even though the department of welfare swore that this was changing—that they were really going to feed the clients—[I didn’t believe them]. Even afterwards, I was talking to people and they were still having a hard time getting the clients.” She concluded, “I didn’t feel I could jeopardize the organization by taking that contract. [The start-up capital comes from us] and then [we’d have to] hope that we’d get reimbursed.” A director of another large nonprofit said that a performance-based contract requires a large volume of clients to remain solvent, and if the flow of referrals is low or unpredictable and/or other problems occur, there is a lot of “anger and bad feeling between the vendor and the contracting agent.” Even agencies providing support services reported that the slow and often Byzantine process of getting reimbursed could cause great fiscal strain. One executive director of an agency that provides child care (among other services) said, “We cannot continue to hemorrhage from not receiving the payments” from the welfare department for the children in its child-care center. “The welfare department always finds a way to screw things up. We are still in the black and we can’t go through another year like this . . . it could close this agency and ones like it.” The Client Point of View Pavetti et al. readily acknowledge that the intermediaries expect that only half of the individuals referred to them will actually come into their programs. This is especially notable given the often severe consequences attached to not attending the programs (such as sanctioning). While a client-based study is beyond the scope of the current paper, other studies hint at what might be going on here. We and our colleagues have been involved in two such studies (Edin and Lein 1998; Kissane 2001), and we refer to them here to speculate about what might be going on in the minds and lives of clients who must increasingly turn to intermediaries to get the services they might have formerly gotten directly from the welfare department. Edin and Lein (1998) study two community FRBNY Economic Policy Review / September 2001 79 organizations (one of which is the aforementioned All Service) and interview clients as well as agency personnel. In taking into account both points of view, they find that policies that were rational for the agency were often irrational from the client point of view. In both agencies, fiscal constraints, the demands of funders, and agency personnel’s need to feel that they were “doing a good job” conspired to create a set of policies that Edin and Lein dubbed “targeting,” “rationing,” and “investing.” In their early years, both agencies had provided a diverse array of services in their communities. Over time, however, one agency (“Community Cooperative”) began to feel increased pressure to target one or two programs rather than provide a broad range of services.4 This decision provided it with a clear mission, which both pleased its funders and satisfied its own need to feel good about the work it was doing. The agency also increasingly rationed the amount of services any one client could receive, allowing it to serve a larger number of community members and thereby increasing its perceived “effectiveness” in the eyes of funders. While such targeting and rationing strategies were rational procedures for the agency to adopt, such actions looked different from the client point of view. First, targeting meant that clients who had needs in multiple domains (for example, help paying for both food and prescription drugs) had to utilize a number of agencies, each of which targeted a specific domain. Second, rationing meant that clients often had to approach several agencies to meet their needs even in a single domain (such as food). The third practice—investing—consisted of targeting discretionary resources to those clients viewed as more likely to “succeed” and less likely to return for additional services. Investing, while rational from the agency point of view (agencies could claim a higher “success rate”), meant that the most able clients could command a lion’s share of the agency’s discretionary resources, while the least able (and most needy) could not access these resources. While the targeting, rationing, and investing strategies of NPs may dissuade some potential clients from using social services, other factors may also play a role in nonparticipation. Kissane (2001) finds that although none of the Philadelphia agencies limited their services to particular racial or ethnic groups, they did tend to serve one ethnic or racial group more frequently. Members of other racial or ethnic groups living in a certain section of the city associated high stigma in utilizing the services of an agency that was not identified with their racial or ethnic group or that was identified with other types of “undesirables” (such as the homeless). Respondents told Kissane that they preferred using services with clientele that were “more normal” and “more like” themselves. Often, stigma was high enough to keep these women from claiming any available services. This was particularly true if the agency 80 Commentary was located in a neighborhood that was perceived to be “unsafe” or the “territory” of another racial or ethnic group. Respondents who talked about such agencies said that it would be “too dangerous” to make use of the services offered, and that they would likely be mugged or molested. Ironically, these same respondents did not express similar fears about going to the welfare office, also located in a “dangerous” area, nor did concerns over the “types” of people using welfare dissuade them from getting welfare themselves. Kissane’s respondents also reported that they felt more stigma when utilizing certain nongovernmental services than those provided directly through the welfare department. Such reluctance often meant that these potential clients had unmet needs. It is possible that similar forces (stigma, racial or ethnic identification, location) may affect the willingness of welfare recipients to utilize programs offered by intermediaries, and they may partly explain the problems in client flow that Pavetti et al. have identified. Both the Edin and Lein and the Kissane findings suggest that the use of intermediaries might prove problematic from the client point of view, even if the policies such intermediaries implement are rational and well-meaning. Conclusion What lessons do we draw from considering, however speculatively, the points of view of the nonprofit agencies themselves and of the potential and actual clients of such agencies? First, contracting with the welfare department may significantly alter the function of nonprofits, influence their fiscal well-being, and create huge administrative demands for which they may not be compensated. Agencies that experience such demands might well get out of the business of welfare reform altogether, and other agencies that might have been considering such a role may choose not to participate based on the experiences of those who have. Alternatively, some agencies that manage to make welfare contracts work for themselves might find that they have significantly altered their function, sometimes to the detriment of clients. Second, clients who are sent to such agencies to receive the assistance they formerly received at the welfare department might, for reasons outlined above, have to utilize several services (thus increasing clients’ transaction costs significantly) to meet all of their needs. Third, if stigma, racial or ethnic identification, or fears about the agency’s location prove to be more salient in the case of intermediates than for the welfare office itself, increased use of intermediaries might mean that fewer eligible clients will receive assistance. Endnotes 1. The longitudinal nonprofit data used in this paper were collected as part of Manpower Demonstration Research Corporation’s Project on Devolution and Urban Change. We would like to thank Gordon Berlin, Barbara Goldman, our many funders, and our collaborators for their support of this work. We also thank Laura Lein for her comments. 3. Interestingly, the dollars they were referring to were not only public, but private as well. Apparently, even private funders have caught the work-first fever (perhaps in response to what they feel is an unassailable mandate by the public sector). 4. Community Cooperative is a pseudonym. 2. All Service is a pseudonym. FRBNY Economic Policy Review / September 2001 81 References Becker, Howard S. 1967. “Whose Side Are We On?” Social Problems 14, no. 3: 239-47. Kissane, Rebecca Joyce. 2001. “What’s Need Got to Do with It: Why Poor Women Don’t Use Nonprofits.” Unpublished paper. De Vita, Carol J. 1999. “Nonprofits and Devolution: What Do We Know?” In Elizabeth T. Boris and C. Eugene Steuerle, eds., Nonprofits and Government: Collaboration and Conflict. Washington, D.C.: Urban Institute. Lipsky, Michael, and Steven Rathgeb Smith. 1989-90. “Nonprofit Organizations, Government, and the Welfare State.” Political Science Quarterly 104, no. 4: 625-48. Diaz, William A., Dwight F. Burlingame, Warren F. Ilchman, David A. Kaufmann, and Shawn D. Kimmel. 1996. “Welfare Reform and the Capacity of Private Philanthropy.” In Dwight F. Burlingame et al., eds., Capacity for Change? The Nonprofit World in the Age of Devolution. Indianapolis, Ind.: Indiana University Center on Philanthropy. Edin, Kathryn, and Laura Lein. 1998. “The Private Safety Net: The Role of Charitable Organizations in the Lives of the Poor.” Housing Policy Debate 9, no. 3: 541-74. ———. 1995. Nonprofits for Hire: The Welfare State in the Age of Contracting. Cambridge: Harvard University Press. Smith, Steven Rathgeb. 1999. “Government Financing of Nonprofit Activity.” In Elizabeth T. Boris and C. Eugene Steuerle, eds., Nonprofits and Government: Collaboration and Conflict. Washington, D.C.: Urban Institute. Urban Institute. 1983. “Serving Community Needs: The Nonprofit Sector in an Era of Government Retrenchment.” Progress Report no. 3. Fink, Barbara, and Rebecca Widom. 2001. “Community-Based Organizations and Welfare Reform.” New York: Manpower Demonstration Research Corporation. The views expressed in this article are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever. 82 Commentary Howard Chernick and Cordelia Reimers Welfare Reform and New York City’s Low-Income Population I. Introduction T he goal of this paper is to evaluate the effects of welfare reform on the economic well-being of low-income families in New York City. To do so, it is important to examine changes in both the social safety net and the income and earnings of vulnerable households and families. For families with low earnings capacity, programs providing cash and/or in-kind assistance may be the source of all or most of the economic resources available, or they may provide vital supplements to earnings. To investigate the extent to which the safety net is still in place in New York City, we use the New York City sample of the Current Population Survey (CPS) to compare program receipt before and after the passage of the Personal Responsibility and Work Opportunity Reconciliation Act of 1996 (PRWORA). We use the income and earnings data from the CPS to compare economic status. Cities around the country have benefited from the strong economic growth in the 1990s. The most recent data show that for the nation as a whole, between 1998 and 1999, the number of central-city residents in poverty fell by 1.8 million and household income of central-city residents, although still substantially lower than in the rest of the country, grew faster than elsewhere (U.S. Census Bureau 2000). Job growth has also been strong in New York City in this period, actually surpassing the national rate in the most recent years. From 1997 to 1999, Howard Chernick and Cordelia Reimers are professors of economics at Hunter College and at the Graduate Center of the City University of New York. Please direct correspondence to <Howard.Chernick@hunter.cuny.edu> or <Cordelia.Reimers@hunter.cuny.edu>. New York City job growth exceeded 2 percent each year, outperforming any equal span of time during the past three decades. The expanding New York economy has increased demand and possibly wages for low-skilled workers. Increases in the earned income tax credit (EITC) and the minimum wage have also made work more attractive to low-skilled individuals in recent years, and New York State supplements the national EITC with its own refundable credit.1 It is difficult to disentangle the effects of welfare reform from the influence of these other factors on welfare receipt and incomes of the vulnerable groups in a single city. Moreover, without longitudinal data, it is not possible to trace the flows between work and benefits programs in detail. We can only observe net changes in program receipt, employment, and income. Our goal in this paper is therefore more modest: to compare public transfer program participation and economic status among New York City households before and after the 1996 welfare reform. We also investigate the extent to which the economic good news has translated into higher earnings and household income for families with low levels of education or single mothers. For those in the groups that have lost public assistance, we ask to what extent earnings have replaced the lost income. Are such families doing better, or about the same? Are more families able to combine public benefits programs with earnings, and how much has their household income changed?2 The authors would like to thank Swati Desai and Kathryn Edin as well as their discussant, Gary Burtless, for excellent comments. They would also like to thank Makada Henry for research assistance. The views expressed are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. FRBNY Economic Policy Review / September 2001 83 Although our analysis compares outcomes before and after PRWORA, it should be made clear that because the formal state plan for welfare reform did not take effect until 1999, we are not really evaluating welfare reform in New York City. Instead, our results primarily reflect the net effect of changes in city administrative policies—characterized as push factors— and the pull of economic growth on the receipt of public assistance. The plan of the paper is as follows. Section II discusses the changes in welfare law and administrative policy in New York and their potential effect on public assistance recipients. Section III describes the data source. Section IV addresses the issue of the packaging of programs and the extent to which the social safety net has been preserved. Section V considers differences among ethnic groups in changes in public assistance receipt. Section VI describes the changes in income and earnings among New Yorkers at risk of needing public assistance. The final section summarizes our findings and highlights the most striking results. II. Legal and Administrative Changes to Programs The major cash programs in the social safety net are Aid to Families with Dependent Children (AFDC) and Temporary Assistance for Needy Families (TANF)—known in New York as Family Assistance—and Supplemental Security Income (SSI). General Assistance, previously known as Home Relief and now called Safety Net Assistance, has also been very important, particularly in New York City. As we use the terms in this paper, “public assistance” or “welfare” includes both AFDC/TANF and Home Relief/Safety Net Assistance, but not SSI. In New York City, a nontrivial number of households get both public assistance and SSI. The major in-kind programs are food stamps and Medicaid. Since the public assistance rolls hit a peak in 1995, New York City has been engaged in a vigorous program to reduce the number of public assistance recipients. According to monthly caseload data from the New York City Human Resources Administration (HRA), the number of public assistance recipients—including both Family Assistance and Safety Net Assistance—dropped by 50 percent, from 1,160,593 in March 1995 to 576,723 in May 2000. New York City has one of the largest mandatory workfare programs in the country, with 32,771 cases engaged in the Work Experience Program (WEP) in June 2000. PRWORA severed the automatic eligibility link between public assistance, food stamps, and Medicaid. The entitlement 84 Welfare Reform and New York City’s Low-Income Population to welfare under the TANF program was ended, with a lifetime limit of five years of welfare receipt, and states were given considerable discretion in designing programs that substituted work for cash assistance. In general, the intention of the law was not to reduce eligibility for, or participation in, food stamps and Medicaid. In fact, there has been a concerted effort to expand Medicaid participation. The exception to this statement is that the eligibility of immigrants (noncitizens) for the various programs was restricted. Immigration is very important in New York City, and our results may be driven by differences between citizens and noncitizens. Therefore, we briefly describe the changes in the law regarding immigrant eligibility for public benefits programs. Historically, naturalized citizens and refugees have been eligible for the same benefits as native-born citizens, but legal permanent residents have been subject to “deeming” and “public-charge” restrictions, and temporary and undocumented immigrants have been ineligible for benefits. Under PRWORA, undocumented immigrants and those on temporary visas remain ineligible for benefits (other than Medicaid emergency services). Except for refugees and asylees, legal immigrants arriving after August 22, 1996, are barred from all federal means-tested benefits (other than Medicaid emergency services) for at least five years, and effectively until they naturalize. For legal immigrants who were in the United States before August 22, 1996, the sponsor-income deeming period was extended for up to ten years for most types of benefits. PRWORA also barred noncitizen immigrants who were in the United States before August 22, 1996, from food stamps unless they had worked in the United States for ten years. Some states, including New York, have at least partially replaced the federal food stamp program with their own food subsidy programs. However, state replacement in New York is limited to those under eighteen, over sixty-five, and/or disabled. Subsequently, the federal government restored eligibility for this same population. Another federal law enacted in 1996, the Illegal Immigration Reform and Immigrant Responsibility Act, tightened the requirements for sponsors to support immigrants. It requires sponsors to sign a legally enforceable affidavit to support the immigrant, if necessary, and authorized government-funded agencies to sue for reimbursement of means-tested benefits. Given the changes in the law, and the increased administrative hurdles that the city has raised to getting public assistance, our expectation was that New York City would show a reduction in the number of families getting the full package of programs—public assistance, food stamps, and Medicaid. Nationally, the intent of the law was to reduce the receipt of public assistance, with less reduction in food stamps and perhaps an expansion in Medicaid coverage. However, food stamps might be expected to decline more in New York than nationally because many new immigrants arrived in New York after 1996 and most of them are ineligible for food stamps until they become citizens. The receipt of public assistance depends both on eligibility rules and on the way in which the intake process is administered. The city has tried to rename its welfare offices “job centers,” with a change in goals from determining eligibility in a relatively straightforward way to actively discouraging applicants by “diverting” them into employment. Advocates for the poor have argued that in fact the way diversion works is that applicants are frequently misinformed about their eligibility and are improperly sent away from the welfare office with only minimal help finding jobs (Sengupta 2000). As evidence that diversion has been important, we note a sharp rise in the number of applicants who were rejected for public assistance, from 26 percent to 56 percent, and a 77 percent increase between 1993 and 1998 in the number of fairhearing complaints by applicants who were denied access to public assistance (City of New York, various years). In the vast majority of these hearings, the city’s actions have been overturned and applicants have been declared eligible for public assistance.3 In response to complaints by advocates, the City of New York has been investigated by the U.S. Department of Agriculture for illegally denying potentially eligible persons the opportunity to apply for food stamps, and a federal judge has ordered the city government to cease the conversion of welfare offices into job centers (Welfare Law Center 2000). These administrative and legal developments suggest that the food stamp rolls might be dropping in tandem with (or at an even greater rate than) the public assistance rolls. By contrast, New York City has made active efforts to enroll eligible persons in Medicaid, particularly low-income women during pregnancy and when they enter the hospital to give birth. III. Accuracy of the Current Population Survey Our data source is the March Current Population Survey. To conform to most other studies, our unit of observation is the household. Because the questions about receipt of most program benefits are asked about the household rather than the person, a household is treated as participating in a particular program if anyone in the household receives benefits from that program. The New York City sample of the March CPS consists of 2,123 households in 1995, 1,579 in 1996, 1,586 in 1998, and 1,568 in 1999. To increase our sample sizes before and after welfare reform, we pooled 1995 and 1996 (“before”) and 1998 and 1999 (“after”). This gives us 3,702 households in 1995-96 and 3,154 households in 1998-99. Because the March CPS asks about income and program participation in the previous year, we refer to the “before” period as 1994-95 and the “after” period as 1997-98. Due to the sample rotation pattern in the CPS, there is approximately a 50 percent overlap in our sample for two adjacent years; consequently, the standard errors of our estimates are biased downward. Because we are dealing with the low-income population, we ignore the topcoding of income data in the CPS. We use the March CPS household weights throughout, with Passel’s corrected weights and race codes for 1995 (Passel 1996). It is well known that the CPS underreports welfare receipt compared with administrative records. Throughout the late 1980s and early 1990s, estimates of AFDC receipt from the March supplement to the CPS were about four-fifths the number of AFDC cases found in program records nationwide (Bavier 2000). After 1994, CPS underreporting became more severe, so that by 1998 the CPS estimates were only about twothirds the actual number of AFDC/TANF cases. In New York City, the CPS indicates that in 1994-95, on average 325,863 households per year received public assistance in at least one month. By contrast, New York City’s welfare agency, the HRA, reports an average of 472,177 public assistance households for December 1994 and December 1995. The 1997-98 average for the CPS is 252,718. The HRA numbers for December 1997 and December 1998 average 314,946.4 The ratio of CPS households to administrative households goes from 69 percent in the earlier period to 80 percent in the later period. The HRA reports a 33.3 percent decline in the caseload between December 1994-95 and December 1997-98, while the CPS indicates a 22 percent decline in households getting public assistance. Thus, while underreporting of public assistance receipt in the CPS was somewhat greater in New York City than nationally before welfare reform, in the later period there was less underreporting in New York than nationally. We have no explanation for the decrease in underreporting in the CPS in the later period. If caseloads were declining more rapidly in the later period than in the earlier period, then one might expect that the end-of-year administrative measure would be smaller relative to the “ever-on” measure in the CPS. However, the rates of caseload decline were very similar between 1994 and 1995 (14.1 percent) and 1997 and 1998 (15.2 percent). One possibility is that changes in the CPS sampling frame caused the changes. However, experts at the Bureau of Labor Statistics say that the changes in the CPS FRBNY Economic Policy Review / September 2001 85 sample in New York during the period were normal ones that were unlikely to cause a sharp change in reported rates of benefits receipt.5,6 When we look at the number of persons living in households with at least one public assistance recipient, the CPS shows 1,105,000 in 1994-95 and 884,000 in 1997-98. These numbers are very close to the administrative counts of recipients, which were 1,115,000 in February 1994 and 792,000 in February 1998.7 This close correspondence does not mean that the CPS correctly counts all those getting public assistance. Person-weighting counts every person in the household as getting public assistance. This leads to an overcount of the number of persons, since in some households not all members receive public assistance—for example, child-only cases or cases where the adult gets SSI. Nonetheless, we take it as reassuring that the CPS count of the total number of persons benefiting from public assistance is close to the total number of actual recipients in New York City. Because the program definition of a food stamp household is much closer to the census definition of a household than is the case for public assistance, we expected food stamp receipt by households to be reported more accurately than public assistance. The CPS reports between 76 and 80 percent of the number of food stamp households reported by the HRA. Hence, while there is less underreporting of food stamps than public assistance in the “before” period, the degree of underreporting is similar in the “after” period.8 IV. Packaging of Programs To examine multiple program receipt, we look at both the overall population and that part of the population at risk of receiving public assistance (AFDC/TANF or General Assistance). “At-risk” households are defined as those that, by virtue of education or family structure, are likely to have low earnings capacity. We include all households whose head is under age sixty-five and has less than a high-school education, plus all female-headed households with children under age eighteen.9 Chart 1 shows the rate of receipt among all households for each of the programs separately. Between 1994-95 and 199798, there was a drop in public assistance receipt from 11.3 to 8.4 percent of households. Food stamp receipt also went down, from 17 to 15 percent. Medicaid receipt remained constant at 25.2 percent. By contrast, SSI receipt increased over the period, from 8.6 to 9.3 percent. Among the population at risk of needing public assistance, rates of program receipt are of course much higher (at least two times higher for public assistance, food stamps, and Medicaid). However, the pattern of changes in receipt across programs is very similar to that seen for the overall population.10 The “any benefits” bars in Chart 2 represent those households that participate in at least one of the four programs. They show that the proportions receiving some benefit stayed about the same over the period. Thus, even with the strong Chart 2 Receipt of Public Benefits Packages By New York City Households, 1994-95 and 1997-98 Chart 1 Percentage receiving benefits Public Benefits Receipt in New York City 60 1994-95 and 1997-98 1994-95 1997-98 50 Percentage receiving benefits 30 1994-95 1997-98 25 30 20 20 15 10 10 0 5 0 PA Food stamps SSI Medicaid All households Note: PA = AFDC/TANF or Home Relief/Safety Net Assistance; SSI = Supplemental Security Income. 86 40 Welfare Reform and New York City’s Low-Income Population , , , , s id its id its its fit ef ef FS ef FS ca PA ca PA i ne i n n n e d d e e e & & o o e b e b b b y M n y M n 4 PA 4 PA 3- cl. 3- cl. An An n n i i All households Households “at risk” Note: PA = AFDC/TANF or Home Relief/Safety Net Assistance; SSI = Supplemental Security Income; FS = food stamps. economy and the administrative push to get people off public assistance, we do not find a large drop in the number of households receiving at least some benefit from the social safety net in the immediate aftermath of welfare reform. The fact that public assistance receipt declined by more than food stamp or Medicaid receipt, while the proportion participating in at least one program stayed the same, suggests that some of those who lost public assistance retained other program benefits. To examine this issue directly, we look next at changes in multiple program receipt and the degree of “packaging” of the various public assistance programs. Chart 2 shows multiple program receipt for all households and for those “at risk.” Table 1 shows benefits packaging in more detail and the benefit combinations received by different ethnic groups. Households are grouped according to whether they did or did not get public assistance. The first pair of bars in each half of Chart 2 shows a substantial drop in the proportion getting the full package of public assistance and at least two of the other three programs: Medicaid, food stamps, and SSI. Among all households, the drop is from 10.4 to 7.4 percent, while among households at risk the drop is from 32.9 to 23.9 percent. This drop closely parallels the decline in public assistance discussed above. The second pair of bars shows that the proportion of those getting a package including Medicaid, but not public assistance, goes up by an approximately equal amount. On its face, this pattern would seem to suggest that most people losing public assistance retained their Medicaid benefits. People losing public assistance can either exit the welfare system entirely or retain other program benefits. Longitudinal data, which track people on public assistance after they leave the rolls, would be required for a precise determination of the proportions in each group. However, our cross-sectional data suggest that both patterns occurred. For those getting public assistance, the most common pattern is also to get food stamps and Medicaid. Of the 8.6 percentage point drop in the Table 1 Receipt of Benefits “Packages” by Households in New York City All Households Percentage receiving All four programs PA+FS+MC PA+MC+SSI SSI+FS+MC PA+FS PA+MC SSI+MC FS+MC PA only FS only MC only None Total Three to four programs, including PA Medicaid without PA Any program Sample size “At-Risk” Householdsa All Hispanic Households All Black Non-Hispanic Households All White and Asian Non-Hispanic Households 1994-95 1997-98 1994-95 1997-98 1994-95 1997-98 1994-95 1997-98 1994-95 1997-98 1.5 8.6 0.3 3.8 0.2 0.7 2.9 2.0 0.1 0.9 5.4 73.6 1.0 6.1 0.3 4.2 0.2 0.8 3.8 2.5 0.1 1.0 6.5 73.6 4.0 27.7 1.2 4.7 0.1 1.4 3.1 3.0 0.1 1.4 7.1 46.2 2.8 20.3 0.8 5.0 0.9 2.1 4.8 5.4 0.1 1.6 10.4 45.8 2.9 20.1 0.6 6.7 0.2 1.6 5.1 3.4 0.1 1.5 6.0 51.9 3.0 11.1 0.3 6.8 0.1 1.1 6.5 4.1 0.1 1.9 10.5 54.6 2.2 12.6 0.8 4.3 0.3 0.8 2.9 3.4 0.0 1.1 7.4 64.2 0.7 11.7 0.8 3.7 0.7 2.1 4.1 4.5 0.2 0.9 8.6 62.1 0.5 1.7 0.0 2.3 0.1 0.2 2.0 0.8 0.1 0.6 4.2 87.6 0.2 1.2 0.1 3.3 0.1 0.1 2.5 0.9 0.0 0.7 3.8 87.3 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 10.4 14.2 26.4 7.4 17.0 26.4 32.9 17.9 53.8 23.9 25.6 54.2 23.5 21.2 48.1 14.5 27.9 45.4 15.6 18.0 35.8 13.2 20.8 37.9 2.3 9.3 12.4 1.5 10.4 12.7 3,702 3,154 1,095 925 1,255 1,117 727 603 1,720 1,434 Note: PA = AFDC/TANF or Home Relief/Safety Net Assistance; FS = food stamps; MC = Medicaid; SSI = Supplemental Security Income. a Head is a nonelderly high-school dropout or a female with children under eighteen. FRBNY Economic Policy Review / September 2001 87 proportion of at-risk households who were getting the full package of public assistance, food stamps, and Medicaid (and maybe SSI as well), about a third (2.7 percentage points) lost only public assistance. Moreover, the proportion of Medicaidonly households increases by 3.3 percentage points. If all of the increase in Medicaid-only receipt comes from households that have lost both public assistance and food stamps, then one could conclude that of those who have lost public assistance, about 70 percent [(2.7 + 3.3)/8.6] have retained their Medicaid coverage. This would imply that at least 30 percent of those who got the full package before welfare reform and then lost public assistance have exited the public welfare system entirely. If Medicaid-only was expanding for reasons other than a shift from a package of programs to just Medicaid, then the proportion exiting the system would be correspondingly larger. Finally, if exits from the public welfare system of this magnitude have occurred, why has the overall percentage of the population getting some benefit not gone down? The answer lies in the increase in SSI receipt. The proportion getting SSI without public assistance increased by 2 percentage points, and 100 percent of SSI recipients also get Medicaid. In other work, we have found that this increase in SSI is due almost entirely to an increase in program receipt among elderly noncitizens. V. Ethnic Patterns of Decline in Public Assistance Flows off of public assistance are influenced by economic conditions, the characteristics of individual households, and changes in administrative rules and procedures. For example, the growth in low-skill, low-wage jobs in the New York economy could reduce the probability of being on public assistance more for those with less education. More stringent administrative procedures could impose a higher hurdle for those who are not fluent in English. To investigate the question of which groups are more likely to have left public assistance, we first focus on ethnicity. We divide the population into three groups—black non-Hispanics, Hispanics, and all others (including non-Hispanic whites, Asians, and Native Americans)11—and look at changes in the rate of receipt of public assistance. Next, we subdivide the Hispanic population by citizenship status and Puerto Rican or other origin. We then present a multivariate analysis of changes in public assistance receipt, which allows us to control for a number of demographic characteristics. Chart 3 shows the change in the proportion of households receiving public assistance (AFDC/TANF and Home Relief/ Safety Net Assistance) between 1994-95 and 1997-98. What 88 Welfare Reform and New York City’s Low-Income Population stands out is the large drop in the rate of receipt among Hispanics (9.8 percentage points) compared with blacks (less than 1 percentage point). In 1994-95, the rate of public assistance receipt is 50 percent higher among Hispanic households than among blacks, yet just three years later the rates are the same. The difference between the rates of decline for Hispanics and blacks is easily significant at the 1 percent level. The percentage point decline among whites and Asians is also small. However, because the white and Asian population is large, the decline still represents a substantial number of persons. Since initial rates of receipt differ sharply among the three groups, in Chart 3 we also show the percentage drop in public assistance receipt. The rate of decline is 38 percent among Hispanics, 36 percent among non-Hispanic whites and Asians, but only 3 percent among blacks. We next ask whether the drop among Hispanics affects only certain groups of Hispanics, or is similar for all Hispanics. In Chart 4, we divide Hispanics into Puerto Ricans (whether born in the mainland United States or in Puerto Rico), other Hispanic citizens, and other Hispanic noncitizens. The chart shows that the decline is substantial among all groups of Hispanics, but the biggest drop (42 percent) occurs among Puerto Ricans. What explains the relatively large drop in rates of public assistance receipt among Hispanics compared with blacks? The greater decline could result from greater improvement in labor market opportunities, or from changes in the characteristics of households that put them at lower risk of receiving welfare, such as a greater decline in the proportion of female-headed families. Faster decline could also be due to increased Chart 3 Receipt of Public Assistance By New York City Households, 1994-95 and 1997-98 Percentage receiving public assistance 60 1997-98 1994-95 50 -31% 40 30 -3% -38% -3% 20 -24% 10 -36% 0 Hispanic Black Other nonnonHispanic Hispanic All households Hispanic Black Other nonnonHispanic Hispanic Households “at risk” administrative barriers making it relatively more difficult for Hispanics to navigate the welfare bureaucracy. To determine whether the greater decline in receipt rates among Hispanics remains statistically significant when we control for other factors that affect the probability of welfare receipt, we estimate a set of linear probability models of public assistance receipt. These models include ethnicity and the change from 1994-95 to 1997-98 for each ethnic group, plus various combinations of demographic controls. In some models, the effect of the controls is allowed to vary over time. The demographic controls are dummy variables for female headship, presence of children under age eighteen, whether the household head is under age sixty-five, whether he or she lacks a high-school diploma, and whether he or she is a citizen.12 The change from 1994-95 to 1997-98 for whites and Asians, and the changes for blacks and Hispanics relative to white and Asian non-Hispanics, are summarized in Table 2. The t-statistic offers a statistical test of whether the drop in receipt is significantly greater among blacks or Hispanics than these others. Model 1 corresponds to the division of households into whites and Asians, blacks, and Hispanics (Chart 3). The results indicate that the greater decline in receipt among Hispanics remains statistically significant under all specifications. Without any controls, the decline is 8.8 percentage points greater for Hispanics than for whites and Asians (column 1). Including the full set of controls and allowing their effects to vary over time reduces this difference to 6.2 percentage points (column 9). Allowing the effect of family structure to vary over time (columns 6, 8, and 9) has the greatest impact on the probability of welfare receipt, because female-headed house- Chart 4 Receipt of Public Assistance By New York City Hispanics, 1994-95 and 1997-98 Percentage receiving public assistance 50 1994-95 1997-98 40 30 20 10 0 Puerto Ricans Other Hispanic citizens All households Other Hispanic noncitizens holds with children experienced an above-average decline in welfare receipt since 1995, and Hispanics are more likely than whites and Asians in New York City to be single mothers. By contrast, the regression shows no significant change in the rate of welfare receipt among blacks.13 Among whites and Asians, the decline is at or close to statistical significance until the effect of age and education is allowed to vary over time. When simple controls for the household head’s age and education are included, the decline for whites and Asians becomes significant at the 5 percent level. However, when we allow the effect of age and education to vary over time, the change for whites and Asians is always insignificant. This last result indicates that the effect of the household head’s age and education on the change in the probability of household welfare receipt completely explains the change in the rate of receipt by whites and Asians. Model 2, like Chart 4, divides the Hispanic group into Puerto Ricans, other Hispanic citizens, and Hispanic noncitizens. As expected, the results for whites and Asians and blacks are unchanged from Model 1. However, among Hispanics, only Puerto Ricans continue to show significantly greater drops in rates of welfare receipt when we allow the effect of being a single mother to vary over time. The differential rate of decline for Puerto Ricans is reduced from 11.6 to 8.3 percentage points by the full set of controls in column 8. Among other Hispanics, the estimated declines are only about half as large as for Puerto Ricans, but the decline is measured more precisely for noncitizens than citizens. In fact, for Hispanic citizens, the decline between 1994-95 and 1997-98 is not significantly greater than for whites and Asians. For noncitizen Hispanics, the decline is significantly greater at the 6.5 percent level, even when we control for single motherhood, age, and education. However, when we control for the differential effect of single motherhood in the later year (columns 6 and 8 of Table 2), the decline for Hispanic noncitizens also becomes insignificant. This insignificance indicates that if a household is at risk of welfare receipt in 199798 because it is headed by a female, then there is no additional likelihood that non-Puerto Rican Hispanics lost public assistance. Thus, once we introduce controls for the characteristics that put families at risk of receiving public assistance, the greater decline for Hispanics seems to have occurred mainly among Puerto Ricans. Given the greater rate of decline in public assistance for Hispanics, it is also of interest to see whether the change in the packaging of benefits differs for this group. Chart 5 (Table 1) shows for Hispanics only the grouping of programs according to public assistance receipt, Medicaid receipt, and any benefit. The pattern is similar to that seen for all groups in Chart 2, but the changes are greater. There is a bigger drop in the FRBNY Economic Policy Review / September 2001 89 Administrative records indicate a decline of 157,000 cases overall during the sample period, while the CPS shows a drop of 73,000 households of all ethnic groups. Because the black population of New York City increased slightly, the administrative records imply a decline in the rate of public assistance receipt among blacks. The question then becomes, why does this drop not show up in the CPS? Although it is possible that the patterns of underreporting of welfare receipt by different ethnic groups have changed since welfare reform, proportion with three or four benefits, including public assistance and food stamps, and a bigger increase in the proportion getting Medicaid, but no public assistance. The only substantive difference between Hispanics and the overall population is that there is a slight increase (2.7 percentage points) in the proportion of Hispanic households getting no benefits. We were surprised by the fact that the CPS shows virtually no drop in the rate of public assistance receipt among blacks. Table 2 Linear Probability Models of Public Assistance Receipt, by Ethnicity and Period Difference in Differences Relative to White and Asian Non-Hispanics, with Various Controls (1) (2) (3) (4) (5) (6) (7) (8) (9) -0.009 1.76 -0.009 1.71 -0.012 2.25 -0.012 2.16 -0.012 2.17 0.015 1.72 -0.001 0.13 0.008 0.62 0.007 0.56 0.004 0.18 0.005 0.26 0.004 0.18 0.005 0.26 0.006 0.29 0.023 1.12 0.006 0.26 0.020 1.05 0.022 1.12 -0.088 4.98 -0.084 5.06 -0.079 4.54 -0.076 4.62 -0.076 4.66 -0.065 3.90 -0.077 4.10 -0.061 3.43 -0.062 3.49 -0.009 1.76 -0.009 1.71 -0.012 2.24 -0.012 2.15 — — 0.014 1.67 -0.002 0.17 0.008 0.57 — — 0.004 0.18 0.005 0.26 0.004 0.185 0.005 0.26 — — 0.022 1.11 0.005 0.26 0.020 1.04 — — Puerto Ricans t-statistic -0.116 4.35 -0.105 4.19 -0.104 4.05 -0.095 3.92 — — -0.088 3.54 -0.103 3.89 -0.083 3.32 — — Other Hispanic citizens t-statistic -0.040 1.18 -0.053 1.71 -0.037 1.13 -0.048 1.54 — — -0.036 1.16 -0.036 1.08 -0.033 1.06 — — Hispanic noncitizens t-statistic -0.060 2.00 -0.054 1.98 -0.055 1.86 -0.050 1.85 — — -0.032 1.16 -0.053 1.68 -0.033 1.12 — — No No No No Yes No No No No Yes No No Yes Yes No No Yes Yes Yes No Yes No No Yes No Yes No Yes Yes Yes No Yes Yes Yes Yes Yes Model 1 (all Hispanics) Change from 1994-95 to 1997-98 White and Asian non-Hispanics t-statistic Change from 1994-95 to 1997-98, relative to white and Asian non-Hispanics Black non-Hispanics t-statistic Hispanics t-statistic Model 2 (Hispanics by citizenship) Change from 1994-95 to 1997-98 White and Asian non-Hispanics t-statistic Change from 1994-95 to 1997-98, relative to white and Asian non-Hispanics Black non-Hispanics t-statistic Controls Female head, children under eighteen Dropout, nonelderly Noncitizen Interactions of controls and year Note: Number of observations = 6,856. 90 Welfare Reform and New York City’s Low-Income Population Chart 5 Receipt of Public Benefits Packages By New York City Hispanics, 1994-95 and 1997-98 Percentage receiving benefits 50 1994-95 1997-98 40 30 20 10 0 3-4 benefits, including public assistance Medicaid, no public assistance Any benefits All Hispanic households this seems unlikely. Further investigation of this puzzling result is clearly warranted. We are analyzing other data sets to see whether the same result is found. VI. Income and Earnings of Low-Income New Yorkers The previous section showed that many, but not all, of those who were on welfare apparently continue to participate in other benefits programs, particularly Medicaid. We also found a particularly sharp drop in public assistance among Hispanics. We now turn to the broader question of how New Yorkers with low household earnings capacity are faring after welfare reform. For households with low education levels or headed by a female, how has the mix of income sources shifted between public assistance and earnings, and how have the levels of income and earnings changed? Given the differential decline in public assistance for blacks, Hispanics, and whites, are the changes in household income different for these groups? A second question focuses on public assistance recipients. An improving job climate in New York City, and increased sanctions for not working, might be expected to increase the proportion of public assistance recipients who are combining cash assistance and earnings. Are those who were still on public assistance in 1997-98 more likely to combine cash assistance and earnings than in 1994-95, and has total household income increased for this group? We note at the outset that the March CPS asks whether anyone in a household got public assistance or earnings in any month during the previous year, but it does not tell us whether the two were received at the same time. Those reporting both public assistance and earnings may have received them at different times during the year. Along with a number of other states, New York has raised the earnings disregard and lowered the benefit reduction rate for TANF recipients with earnings (New York State Office of Temporary and Disability Assistance 2000; Giannarelli and Wiseman 2000). Eventually, these changes should lead to an increase in the proportion of public assistance cases that also receive earnings. However, the changes in the disregard and the benefit reduction rate did not take effect until November 1999. Hence, they should have no impact on the changes in the likelihood of combining cash assistance and earnings between 1994-95 and 1997-98. Chart 6 shows the mixing of income sources for at-risk households for blacks and Hispanics separately. Whites and others are excluded because the sample size is small and because the patterns are very close to those for Hispanics. Overall, the increase in the proportion of the at-risk population that gets both public assistance and earnings is small, going from 9 to 11.5 percent. As shown in the chart, there was a substantially bigger drop among Hispanics than blacks in the proportion getting only public assistance: 13 percentage points versus 8.1 percentage points. What stands out is the difference in where those leaving the “just public assistance” category go. Among blacks, almost all apparently wind up getting both public assistance and earnings. The increase in the percentage getting both public assistance and earnings is almost 90 percent Chart 6 Welfare and Earnings Receipt By “At-Risk” Households, 1994-95 and 1997-98 Percentage receiving 70 1994-95 60 1997-98 50 40 30 20 10 0 Welfare, no Both earnings Earnings, no welfare Hispanics Welfare, no Both earnings Earnings, no welfare Black non-Hispanics FRBNY Economic Policy Review / September 2001 91 of the decrease in public assistance alone. By contrast, for Hispanics, the proportion getting income from both earnings and public assistance does not change, while the increase in the proportion with earnings-only is equal to 85 percent of the drop in those getting only public assistance. The above results show that in the first years after welfare reform, Hispanics were more likely than blacks to leave public assistance entirely, while blacks were more likely to combine public assistance and earnings. The differential pattern of shifts between Hispanics and blacks among public-assistance-only, earnings-only, and both suggests that for many Hispanics, earnings have increased enough to end eligibility for public assistance. However, for blacks, the earnings increase seems to have been more modest and therefore a higher proportion retain eligibility for public assistance. Chart 6 shows that the proportion of the “at-risk” population getting both public assistance and earnings is unchanged for Hispanics, but increases substantially among blacks. Chart 7 would seem to contradict this story. It shows that among those getting public assistance, the proportion of recipients who also get earnings increased almost as much among Hispanics (12 percentage points) as among blacks (15.6 percentage points). The explanation for this apparent inconsistency is that among Hispanics, two things appear to have been going on at the same time. Of those getting only public assistance in the earlier period, a substantial number also got earnings in the second period. However, of those Hispanics getting both sources of income in the earlier period, many lost their public assistance benefits and wound up having only earnings. By contrast, among blacks, only the first “movement” occurred. Households moved from publicassistance-only to public assistance and earnings, but very few households lost their public assistance benefits entirely. We would also like to know whether household income increased among those combining public assistance and earnings. As reported in our CPS samples, nominal household income actually went down for those combining public assistance and earnings (from $18,193 to $16,524). Unfortunately, the sample size for this group is quite small (about 100 households in each period), so our estimates are not very precise. The small sample size makes it impossible to determine the reasons for the drop (or lack of increase) in household income among those combining public assistance income and earnings. One possibility stems from the fact that a household may have received income from both sources during a year, but not at the same time. Of those reporting both public assistance and earnings, some may have gotten public assistance toward the beginning of the year and earnings toward the end. Before welfare reform, the group leaving public assistance for work would have consisted mainly of households “pulled” off welfare by attractive employment. After welfare reform, however, more families may have been “pushed” off welfare into low-wage jobs. On balance, this may have led to household income being lower than it was before reform for those receiving both public assistance and earnings in the same year. Lastly, we ask, how has economic well-being changed between 1995 and 1998 for New York City households with low earnings capacity? We examine changes in income and earnings both for those with positive earnings and for the entire at-risk group, again dividing the sample into Hispanics, blacks, and whites and Asians. The results are summarized in Charts 8 and 9. Chart 8 shows the change in the proportion of Chart 8 Households with Earnings “At-Risk” Households, 1994-95 and 1997-98 Chart 7 Earnings by Households on Welfare In New York City, 1994-95 and 1997-98 Percentage with earnings 50 1994-95 Mean earnings Percentage with earnings 70 1994-95 1997-98 60 30 50 25 40 20 30 15 20 10 10 5 1997-98 40 30 20 0 0 10 Hispanic 0 Black non-Hispanic All households on public assistance 92 Black non-Hispanic Percentage with earnings Hispanic 35 Welfare Reform and New York City’s Low-Income Population Hispanic Black non-Hispanic Mean earnings if greater than 0 Note: Mean earnings are in thousands of 1999 dollars. households with earnings and the average amount of earnings (in 1999 dollars) for those with some earnings. The proportion with earnings went up by 11 percentage points among Hispanics, as opposed to 4.9 percentage points among blacks. Among whites and Asians, there was no change. The increase in the proportion with earnings is statistically significant both among all “at-risk” households and among Hispanics, but insignificant among the other groups. The second half of Chart 8 shows the change in average household earnings for those with positive earnings. (All figures are adjusted to 1999 dollars, using the New York City values of the consumer price index.) Among blacks, average real annual earnings decreased by $3,277, while among Hispanics, average earnings went up by $2,171. Among whites, there was a decline of $1,268 (not shown). None of these changes is statistically significant, however. Although it is not statistically significant, the difference in the change between blacks and Hispanics is consistent with the greater decline in public assistance receipt among Hispanics than among blacks, as discussed above. The first half of Chart 9 shows earnings among all Hispanic and black households who are at risk of receiving public assistance. Average real earnings increased by $4,161 (30 percent) for Hispanics, but fell by $798 for blacks. Only among Hispanics was the increase in earnings statistically significant. Among all households, the change in average real household earnings, although positive, was not significantly different from zero. Household income, shown in the second half of Chart 9, shows a pattern of change almost identical to household earnings, rising a statistically significant 22 percent among Hispanics and falling among blacks. Chart 9 Household Earnings and Income “At-Risk” Households, 1994-95 and 1997-98 Mean income Mean earnings 25 25 1994-95 1997-98 20 20 15 15 10 10 5 5 0 0 Hispanic Black non-Hispanic Mean earnings Hispanic Black non-Hispanic Mean income Note: Mean earnings and mean income are in thousands of 1999 dollars. VII. Summary and Conclusions The 1996 welfare reform law marked a major change in national policy toward public assistance. Over the time period covered by our research, the City of New York has also been engaged in a vigorous effort to reduce the welfare rolls. To evaluate the initial effects of the new law and the change in city policies, we use the Current Population Survey to compare receipt of public benefits programs, income, and earnings among households with low earning capacity in New York City in 1994-95 and 1997-98. The CPS shows a 22 percent drop in the number of households getting public assistance. This estimate is well under the 33 percent decline in the caseload reported by the Human Resources Administration. However, food stamp and Medicaid receipt appears to be more accurately reported. The undercount suggests that some caution is warranted in interpreting our findings. Between 1994-95 and 1997-98, the CPS shows a drop in the proportion of New York City households getting public assistance, from 11.3 to 8.4 percent. Food stamp receipt went down by 2 percentage points, from 17 to 15 percent, while the rate of Medicaid receipt remained constant. The proportion getting at least one benefit (Medicaid, public assistance, SSI, or food stamps) stayed about the same over the period. Of those who had been getting public assistance, food stamps, and Medicaid and then lost their public assistance, we estimate that at least 30 percent have exited the public welfare system entirely. At most, 70 percent have retained some other program benefit. Surprisingly, the reduction in rates of public assistance receipt among blacks is negligible. The decline in public assistance receipt is significantly greater among Hispanic households than among other ethnic groups. When we divide the Hispanic population into various groups, the greatest rate of decline is among Puerto Ricans. When we control for other factors that might affect the rate of public assistance receipt, the significantly greater rate of decline holds up statistically only for Puerto Ricans. We also look at changes in income and earnings of public assistance recipients and households at risk of needing public assistance. Overall, we find only a small increase in the proportion of the at-risk population that is combining earned income and public assistance. However, among those who remained on the public assistance rolls in 1997-98, the increase was more substantial, with the proportion also receiving earnings going up from 27 to 43 percent. This increase probably results from both an economic pull—an improving job climate—and an administrative push—more emphasis on work requirements and greater sanctions for not working. Blacks were more likely than Hispanics to combine both sources of income in the later period. However, based on a very FRBNY Economic Policy Review / September 2001 93 limited sample, we find no evidence of significantly increased income among those who did combine the two sources of income. The proportion of “at-risk” households with earnings rose from 62 to 69.2 percent, but went up more for Hispanics (by 11 percentage points) than for blacks (4.9 percentage points). Among those with earnings, the average level of household earnings went down for blacks and up for Hispanics, but these conditional earnings changes are not significant for either group. Among the entire “at-risk” group, including those with zero earnings, there was a statistically significant increase in average real household earnings (30 percent) and income (22 percent) for Hispanics, but not for the other ethnic groups. We conclude by highlighting what we consider to be the most striking results from this research. First, although there was a sharp drop in the rate of receipt of public assistance, the same proportion of the city’s households (26.4 percent) received at least some benefits under the social safety net in 1997-98 as in 1994-95. This result reflects the strong fiscal incentives to maintain Medicaid enrollment and the increase in the number of SSI recipients. Second, despite the strong economy in New York, real earnings and income for at-risk households show no significant gain over the period studied. Almost 20 percent of those who relied on public assistance alone in 1994-95 had substituted earnings for public assistance by 1997-98. Another 10 percent combined earnings and public assistance. Nonetheless, for those with low education levels or headed by a single mother, total household earnings and income remained basically unchanged. The lack of an increase in earnings could result in part from the depressing effect on wages for low-skill jobs caused by the entry of many former welfare recipients into the labor market. It should be noted, however, that our measure of income does not take into account the earned income tax credit, which was increased substantially not only in 1993 but also in 1996. Differences between Hispanics and blacks may be characterized as “gap closing,” in that rates of receipt of public assistance and earnings levels of Hispanics converge on those of blacks. The next step in our research is to use the March 2000 CPS to determine whether public assistance rates continue to decline more rapidly for Hispanics than for blacks, and earnings and income continue to increase more rapidly, or whether the rates have trended together as the economic expansion continues in New York City. Possible explanations for the observed gap closing involve data accuracy, language barriers, and economic factors. First, there is a question of data accuracy. Although the decline in public assistance receipt among Hispanics is consistent with the overall caseload decline in the 94 Welfare Reform and New York City’s Low-Income Population administrative data, the especially sharp decline among Puerto Ricans and the negligible change among blacks are surprising. We find this result for blacks hard to believe. More and better data are required to determine whether the rate of public assistance receipt actually did not drop among blacks, or whether our result reflects anomalies in the CPS data. The greater decline in rates of public assistance receipt among Hispanics between 1994-95 and 1997-98 would seem to be consistent with the hypothesis that language is an important barrier to understanding the new rules and policies implemented by New York City. However, this hypothesis is contradicted by the fact that the greatest drop in rate of receipt was among Puerto Ricans, who might be expected to face fewer language barriers than other Hispanics. An alternative explanation for the sharp decline among Puerto Ricans is that it reflects a complicated interaction between greater administrative barriers to receipt, differences in family structure and resources, and the “pull” effects of a stronger economy. Suppose that Puerto Ricans were more likely to cohabit, or live in extended families, and therefore were better able to draw on extended family economic resources than blacks. If those resources were increasing relatively rapidly because of the stronger economy, then the additional administrative hurdles, even if relatively uniform for all groups, could make Puerto Ricans more likely to leave the welfare rolls. On the earnings side, only Hispanics show consistent and statistically significant increases in employment, income, and earnings. Hispanics “at risk” for needing assistance started out the period with household earnings only 75 percent of the earnings of blacks. By 1997-98, their household earnings had risen to 105 percent of the earnings of blacks. Why did lowskilled Hispanics do better in the labor market than other groups, particularly blacks? Kathryn Edin has suggested to us that one consequence of welfare reform may be a switch from informal to formal earnings, and greater reporting of those earnings on sample surveys such as the CPS. If Hispanics were more likely to rely on informal and unreported earnings than blacks, then the increase in earnings among Hispanics could represent a difference in reporting, rather than a real change in relative economic circumstances. Another possibility is that the characteristics that help to determine income, such as education level, changed more for Hispanics than for blacks. Although the data do show an increase in education level among Hispanics, and a drop in rates of single motherhood relative to blacks, the differences are not great enough to explain the difference in outcomes. Moreover, the fact that our at-risk group is based on single motherhood and low education means that those experiencing sharp increases in education or changes in headship would be selected out of the at-risk group. A third explanation for the increased employment of Hispanics is that the demand for Hispanics in the labor market has increased relative to blacks. This change could reflect employer discrimination, or the fact that the ability to speak Spanish is increasingly valued by employers in New York. Employer preferences for Hispanics over blacks have been reported in interview surveys conducted in a number of cities (Moss and Tilly 2000). The fact that both employment and earnings went up for Hispanics is consistent with both of these stories. Finally, the fact that Hispanics left the public assistance rolls at such high rates may have been related to their increase in earnings. Exit from public assistance reflects both push and pull factors. If the push factor of administrative hassling had a greater effect on Hispanics than on blacks, it may have forced Hispanics to increase their employment and earnings more than other groups. To conclude, it is axiomatic that researchers always call for more research. In this case, however, we feel particularly justified in doing so. In an era in which welfare policies are changing rapidly, patterns in receipt of public benefits, income, and earnings are highly important in understanding the wellbeing of New York City’s low-income residents. There are some genuine puzzles presented by the data, and we hope that future research, by ourselves and others, will be able to explain the results more conclusively. FRBNY Economic Policy Review / September 2001 95 Endnotes Tables containing the information presented in the charts are available from the authors. 1. The New York State credit was expanded after 1997, so it now equals 22 percent of the federal EITC. 2. For a national analysis along these lines, see Primus et al. (1999). 3. From fiscal year 1994 through fiscal year 1997, the percentage of fair-hearing rulings in the client’s favor ranged from 85 percent to 91 percent. In fiscal year 1998, the measure was changed, making it impossible to compare with the earlier period. The last statement is based on a communication with Glenn Pasanen, Associate Director of the City Project, on December 13, 2000. 4. The HRA counts were prepared for us by the Office of Policy and Program Analysis of the Human Resources Administration. 8. A priori, we would expect the CPS to show a bias toward overreporting because the CPS measure is a measure of “ever received” the program during a year, while the administrative records are point-in-time measures. Because of turnover, the former number is larger than the latter in welfare programs. 9. A more targeted group at risk for AFDC/TANF would require both low education and female headship. It would include only female household heads with children whose mother lacks a high-school diploma. However, sample sizes are substantially reduced for this restricted group and are too small for fruitful analysis. Moreover, this would exclude the population at risk for General Assistance. 10. Tables containing the rates of receipt by at-risk households are available from the authors. 5. Communication with Anne Polivka, Bureau of Labor Statistics, November 19, 2000. 11. Throughout this paper, for the sake of brevity, we use “whites” to refer to non-Hispanic whites and “blacks” to refer to non-Hispanic blacks. The group “whites and Asians” also includes Pacific Islanders, American Indians, Aleuts, and Eskimos. 6. One possible explanation for this increase in the reporting of public assistance receipt in the later Current Population Surveys was suggested to us by Kathryn Edin. Changes in the official names of many state welfare programs after PRWORA might be confusing to respondents, and could be expected to lower reporting rates for public assistance in the CPS. In the case of New York, the name change from AFDC and Home Relief to Family Assistance and Safety Net Assistance may have had the effect of increasing the reporting of these programs in the CPS because the names conform more closely to the wording of the census question on receipt of public assistance. 12. The most inclusive specification of Model 1, shown as column 9 in Table 2, is3URE 3$UHFHLSW FRQVWDQWβ1(Yr9798)β2 %ODFN β3 %ODFN <U β4 +LVSDQLF β5 +LVSDQLF <U β6 VLQJOHPRP β7 VLQJOHPRP <U β8 GURSRXW/7 β9 GURSRXW/7 <U β10 QRQFLWL]HQ β11 (noncitizen*Yr9798) + error. Model 2 breaks up each “Hispanic” term into three separate terms: Puerto Rican, other Hispanic citizen, and Hispanic noncitizen. The specifications in column 1 include only the terms identifying ethnicity. The specifications in columns 2-8 also include various subsets of the variables labeled “controls” in Table 2. 7. The fact that the person count is much closer to the administrative count of persons receiving public assistance, while the CPS household count is between 69 and 80 percent of the number of cases, indicates that CPS households reporting welfare receipt typically are larger than caseload units. This reflects the frequency with which public assistance units live with other relatives. The upward bias from counting persons in a household who are not part of the case unit offsets the underreporting bias. 13. This result is obtained by adding !0.009 (row 1) and 0.004 (row 3), resulting in an insignificant !0.005. 96 Welfare Reform and New York City’s Low-Income Population References Bavier, Richard. 2000. “Accounting for Increases in Failure to Report AFDC/TANF Receipt.” Working draft, U.S. Office of Management and Budget. March 1. City of New York. Various years. “Mayor’s Management Report.” City of New York. Department of Human Resources. 1999. “HRA Facts.” Unpublished paper, February. Giannarelli, Linda, and Michael Wiseman. 2000. “The Working Poor and the Benefit Door.” September. Moss, Philip, and Chris Tilly. 2000. Stories Employers Tell: Race, Skill, and Hiring in America. New York: Russell Sage Foundation. Passel, Jeffrey. 1996. “Problems with March 1994 and 1995 CPS Weighting.” Memorandum, Urban Institute. November 12. Primus, Wendell et al. 1999. “The Initial Impacts of Welfare Reform on the Incomes of Single-Mother Families.” Center on Budget and Policy Priorities. August 22. Sengupta, Somini. 2000. “At One Center, a Study in Welfare Cuts.” New York Times, June 27. U.S. Census Bureau. 2000. “Money Income in the U.S., 1999.” Current Population Reports, Series P60-209. September. Welfare Law Center. 2000. November. <http://www.welfarelaw.org> New York State Office of Temporary and Disability Assistance. 2000. “New York State Plan and Executive Certification: Administration of the Block Grant for Temporary Assistance to Needy Families.” <http://www.dfa.state.ny.us/tanf/> The views expressed in this article are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever. FRBNY Economic Policy Review / September 2001 97 Gary Burtless Commentary H oward Chernick and Cordelia Reimers have written a very useful paper that sheds light on issues that many citizens and policymakers care deeply about: How have the employment and earnings of low-income Americans been affected by welfare reform? How has reform affected the mix of welfare benefits and earned income received by low-income families? Chernick and Reimers go about answering these questions in a straightforward and illuminating way. My comments will focus on the effect of reform on the wellbeing of New York City’s low-income population. I want to suggest a couple of extensions of the authors’ analysis that might shed even more light on this crucial aspect of reform. When President Clinton and Congress were considering reform back in 1993, many policymakers and researchers wanted to know the possible impact of time limits and work requirements on the welfare-dependent population. Soon after the Administration took up the issue of reform, in the spring of 1994 the Urban Institute organized a conference on the topic of work requirements.1 When the conference volume was ultimately published, but well before Congress had acted on reform, the Urban Institute held a press conference to publicize the volume’s main lessons. Because of wide public interest in reform, the press conference was very well attended, and one question repeatedly came up: If Congress enacts a law that imposes strict time limits, strong work requirements, and tough sanctions on recipients who fail to comply with new welfare rules, how will the reform affect the well-being of the low-income population? Many people obviously were concerned that children in single-parent families might be harmed as a result of time limits and tough work requirements. Using a variety of indirect measures of well-being, Chernick and Reimers try to answer this crucial question. It is obviously impossible to answer the question in isolation. Many other things have changed since 1994 besides the public assistance law and welfare administration. The economy is in much better shape in 2000 than it was in 1994. The earned income tax credit (EITC) is also more generous, and a number of states have established or expanded EITC programs of their own. In addition, the Children’s Health Insurance Program (CHIP) now provides subsidized health insurance protection to many low-income working families who would have been ineligible for such coverage in 1994. All of these changes in the environment have affected family earnings, net incomes, and well-being. Nonetheless, it is still useful to try to answer the question posed in 1994: What is the situation of the population at risk of receiving welfare today compared with the situation it faced in 1993 or 1994? Many people, including President Clinton and other architects of reform, believe that reform involved changes in addition to those directly connected to the welfare system. Reform also involved liberalization of the EITC, implementation of CHIP, more generous provision of child- Gary Burtless is a senior fellow at the Brookings Institution. The views expressed are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. FRBNY Economic Policy Review / September 2001 99 care subsidies, and expansion of earnings disregards. President Clinton also believed, correctly, that he and the Federal Reserve deserve some of the credit for the healthy job market. What do Chernick and Reimers find? • Employment has risen in the New York City population at risk of receiving welfare. • Unconditional average earnings have increased 7 percent in the at-risk population (an increase that is not statistically significant). • Unconditional household income has increased 9 percent in the at-risk population (an increase that is almost statistically significant). The authors do not tell us whether or how much these gains are due to welfare reform as narrowly defined, to welfare reform more broadly defined (to include the expansion in the EITC, for example), to ordinary economic progress, and to extraordinary labor market tightness. Chernick and Reimers have not attempted to determine how much of the employment and income gains in the at-risk population can be traced to tougher work requirements, strict time limits on benefits, generous disregards, a liberalized EITC, or six years of strong economic growth and two or three years of tight labor markets. This is understandable, because it is famously difficult to disentangle the separate effects of each factor. Nonetheless, the authors have given us a helpful overview of the changes in employment and income that have followed in the wake of New York City’s welfare reform. It would be useful if the analysis could be extended to consider two other questions. First, is it possible to give readers an indication of the changes in broader measures of well-being? The present paper shows changes in the employment rate, unconditional earnings, and unconditional household income of the at-risk population. The employment rate, by itself, provides an ambiguous indicator of well-being. Some critics of recent U.S. economic performance suggest that jobless people in western Europe enjoy a higher standard of living than lowwage job holders in the United States. If the United States has increased the ranks of job holders by withholding transfer benefits from people who are jobless, the increase in job holding in the at-risk population might be consistent with a decline rather than an improvement in the living standards of the at-risk population. Perhaps, as some European critics suggest, the increased employment rate of low-productivity American workers is not a reliable indicator that they are better off. The improvement in unconditional earnings in the at-risk population also gives an ambiguous signal that the population at risk of receiving welfare is better off today than it was in the past. If the gain in earnings has been offset by an equal or even greater loss in public transfers to the at-risk population, the 100 Commentary well-being of the poor may have declined even as average earned income increased. The average income received by the at-risk population offers a less ambiguous indicator of well-being. If the loss of government transfers had offset the gain in earned income, average household income would have declined. But the authors’ tabulations show that average income climbed 9 percent while labor earnings rose just 7 percent. Most students of American poverty, however, recognize that cash household income is a deficient measure of family well-being. • Do the authors’ tabulations include state and federal EITC payments? If the earned income tax credit is excluded, then the tabulations understate the gains that some families have made as a result of moving from the public assistance rolls into employment. • Do the tabulations subtract from household income a reasonable estimate of the cost of caring for children when the custodial parents are at work? If they do not, they overstate the improvement in family spendable income that occurs when potential breadwinners move off welfare and into jobs. • Do the authors’ calculations include plausible imputations of federal, state, and local tax withholdings? Because wage earnings are taxed while government transfers are untaxed or very lightly taxed, ignoring tax payments can bias the assessment of a breadwinner’s relative position when he or she moves off the public assistance rolls and into employment. • Do the tabulations include consumption enjoyed by the family that is not paid for with spendable household income? For example, do they include rent subsidies received by residents of public housing? Food purchases made possible with free school lunches or food stamps? Consumption of medical care that is financed by Medicaid, CHIP, or a group health plan subsidized by an employer? • Do the tabulations adjust household incomes to reflect differences in the number of people who must divide the incomes? Most people agree that families containing more members must receive more income to enjoy a standard of living comparable to that of a family with the same income but fewer members. One crude adjustment to reflect such differences is to calculate each family’s income-to-needs ratio, that is, the ratio of its spendable income to its poverty threshold. The authors’ calculations show that unconditional household income rose 9 percent in the at-risk population. If family size also rose, well-being did not increase by 9 percent; if family size fell, well-being probably improved by more than 9 percent. No adjustment for family size differences seems to have been made in the first version of the paper. Although it is difficult to make the calculations required to derive a meaningful measure of family or personal well-being, it is not impossible. Most of the required data are available in the March Current Population Survey, which provides information on estimated tax liabilities and EITC payments, noncash income sources, health insurance coverage, and family size. Using methods proposed by the Census Bureau, we can also make defensible estimates of work-related expenses.2 A second question worth considering involves the distribution of gains and losses in the at-risk population. The authors show us how gains and losses differ by racial and ethnic group. Their emphasis on race is a sad commentary on the huge significance of race in U.S. policy evaluation. Americans care more passionately about this difference than they do about a distinction that may be much more meaningful, namely, the difference between workers and nonworkers. How has reform affected the comparative well-being of workers versus nonworkers in the at-risk population? It should be clear that people in families containing working breadwinners are better off as a result of the changes in the economic and policy environment over the past few years. The EITC and CHIP have improved the potential living standards of families containing children and a low-wage breadwinner. However, low-income families without a working breadwinner may be significantly worse off. It is now more difficult to obtain cash public assistance than it was in the past. Once people become entitled to cash benefits, it is now more difficult to remain steadily entitled to benefits. The comparison between workers and nonworkers is complicated by the fact that changes in the economic and policy environment have increased the percentage of at-risk potential breadwinners who actually work. Many low-wage people now hold jobs who would not have been at work if the environment of the early 1990s had remained unchanged. Thus, it would be interesting to assess the shifting fortunes of three groups of at-risk people: those who are members of families where an adult would have worked in either the old or the new regime; those who are members of families where no adult would have worked in either the old or the new regime; and those who are members of families where there would have been no adult worker in the old environment, but where an adult has been induced to find employment in the new one. My guess is that families in the first group have seen an improvement in their well-being while families in the second group are now worse off. I do not know whether families in the third group are better off or worse off now than they were under the old regime. It would be worthwhile to find out. A major extension of Chernick and Reimers’ excellent paper is needed before we will know. FRBNY Economic Policy Review / September 2001 101 Endnotes 1. The papers presented at the conference were later published in Smith Nightingale and Haveman (1995). 2. See Short et al. (1999). References Short, K., T. Garner, D. Johnson, and P. Doyle. 1999. “Experimental Poverty Measures, 1990-1997.” U.S. Bureau of the Census Report no. P60-205. Washington, D.C.: U.S. Government Printing Office. Smith Nightingale, Demetra, and Robert H. Haveman, eds. 1995. The Work Alternative: Welfare Reform and the Realities of the Job Market. Washington, D.C.: Urban Institute. The views expressed in this article are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever. 102 Commentary Philip K. Robins and Charles Michalopoulos Using Financial Incentives to Encourage Welfare Recipients to Become Economically Self-Sufficient I. Introduction On August 22, 1996, President Clinton signed the Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA), which radically altered the structure of the welfare system in the United States. Among other things, the act replaced the Aid to Families with Dependent Children (AFDC) program, a federal entitlement, with the Temporary Assistance for Needy Families (TANF) program, a system of block grants to states. One of the primary goals of TANF is to move welfare recipients into work and economic self-sufficiency. Although states were given much flexibility in how to achieve this goal, the federal government imposed some guidelines in the form of requirements that welfare recipients be participating in a workrelated activity (“work participation requirements”) and time limits on length of welfare receipt. The focus of this paper is on alternative financial incentive schemes that are being used or could be used to help states meet the work participation requirements specified by the federal legislation. In particular, the paper considers whether an earnings supplement conditioned on full-time work would encourage more people to work than the enhanced earnings disregards currently being used or tested by many states. Philip K. Robins is a professor of economics at the University of Miami; Charles Michalopoulos is a senior research associate at the Manpower Demonstration Research Corporation. The remainder of the paper is organized as follows. Section II provides a background of the PRWORA legislation and describes methods that states have been using to encourage employment and economic selfsufficiency among the welfare population. The discussion focuses on various financial incentive schemes adopted by the states. Section III describes a financial incentive scheme currently not being used in the United States (but being used on an experimental basis in Canada) that conditions benefits on full-time employment. Section IV discusses how such a scheme might be implemented in the United States. Section V presents estimated effects of such a scheme based on results from a microsimulation model. Finally, Section VI summarizes the results and offers some concluding observations. II. Background The federal PRWORA legislation stipulated that 25 percent of the caseload in a particular state had to be participating in work activities by fiscal year 1997.1 The minimum work The research for this paper was supported by a grant to the Manpower Demonstration Research Corporation from the Annie E. Casey Foundation. An earlier version of the paper was presented at a seminar at Manpower Demonstration Research Corporation. The authors acknowledge the helpful comments of Gordon Berlin, Barbara Goldman, Judy Gueron, Christopher Jencks, Thomas MaCurdy, and the seminar participants. The views expressed are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. FRBNY Economic Policy Review / September 2001 105 participation requirement has been and will be increasing by 5 percent each year until fiscal year 2002, when it will reach 50 percent. States failing to meet the work participation requirements might not receive the full value of the federal TANF block grant. Since 1997, continued economic prosperity and substantial declines in welfare caseloads have left states with substantial TANF surpluses, and no state thus far has failed to meet the work participation requirements (U.S. Department of Health and Human Services 2000, pp. 41-3).2 The federal legislation defines an “allowable work activity” as unsubsidized employment, subsidized private sector or public sector employment, on-the-job training, job-search assistance for up to six weeks, community service programs, vocational education training for up to one year, and education for persons who have not yet completed high school. The legislation emphasizes work activities and places caps on the number of people who can be placed in educational activities. Reducing the caseload can also count toward the participation requirement. States have considerable latitude in penalizing household heads who fail to comply with the work activity requirements. Benefits can be reduced or terminated, at state discretion. States can exempt certain people from the requirements, such as single parents of young children, but they must meet federal requirements for the percentage of their caseload participating in work activities. The work requirement provisions of PRWORA make it crucial for states to find effective ways of moving welfare recipients into work. Many studies have shown that a significant portion of the caseload spends more than sixty months receiving benefits (the maximum time limit specified under PRWORA, although many states have opted for shorter time limits). Bane and Ellwood (1994), for example, estimate that the median length of total welfare receipt (not necessarily a continuous spell) is about fortyeight months. Pavetti (1995) estimates that 76 percent of the welfare caseload at any point in time (which is dominated by long-term recipients) will eventually receive welfare for at least sixty months. She finds that among those who received welfare for sixty months or more, 63 percent lacked a high school diploma (or GED) at the time they started collecting welfare, 39 percent had no work experience, 53 percent were under twenty-five years of age, 58 percent had never been married, and 52 percent had a child under the age of one year. Clearly, in the absence of effective actions by the states, many individuals are likely to be in financial despair when the time limit is reached. The wide latitude given to states in implementing the 1996 legislation has led to many innovative welfare-to-work 106 Using Financial Incentives programs throughout the country. To stimulate work by household heads, states have designed programs that provide both services and financial incentives (see, for example, U.S. General Accounting Office [1998]). Until now, most of the emphasis has been on services, particularly those, such as job-search assistance, aimed at preparing welfare recipients for immediate employment. Less attention has been paid to financial incentives, although most states have modified their benefit formulas to provide financial incentives to work. Prior to 1996 (and since 1982), a working welfare recipient lost one dollar of cash assistance for each dollar of earnings (after four months of earnings). That is, benefits were reduced on a dollar-for-dollar basis with earnings. Such a high “tax rate” provided a powerful disincentive to work. Beginning in the early 1990s, some states were granted waivers to the AFDC program rules, and several of these states introduced enhanced disregards that excluded a certain amount of earnings when calculating welfare benefits. Since PRWORA, establishment of enhanced disregards accelerated. According to Gallagher et al. (1998), between January 1992 and October 1997, forty-one states had adopted some form of enhanced disregard. Eleven of these states had established their enhanced disregard prior to August 1996. Since 1997, an additional six states have adopted enhanced disregards (U.S. Department of Health and Human Services 2000, pp. 201-3).3 Table 1 shows the earnings disregards being used by states under TANF as of January 2000. The disregards have two components: a flat component and a variable component. The flat component is a fixed dollar amount of exempt earnings. The variable component is a percentage of earnings above the flat disregard (either fixed or varying with the level of earnings, time spent on welfare, or caseload status). Prior to TANF (from 1981 to 1996), the AFDC program had a flat disregard of $120 for the first twelve months of earnings and $90 thereafter. The variable disregard was one-third of earnings above $120 for the first four months of earnings and zero thereafter, thus creating a “tax” (or “benefit reduction”) rate of 100 percent on earnings.4 After TANF, many states adopted very liberal (or enhanced) disregards. For example, Connecticut currently disregards all earnings up to the poverty level until families encounter the state welfare program’s time limit, so that the effective benefit reduction rate is zero for all families with income below the poverty level. Other states, such as Nevada, disregard all earnings initially, but then phase in decreasing disregards over time. A substantial number of the states disregard between 20 and 50 percent of earnings and have a flat disregard of between $100 and $200 per month. Table 1 Earnings-Disregard Policies for TANF Recipients January 2000 State Alabama Alaska Arizona Arkansasc California Colorado Connecticut Delaware District of Columbia Florida Georgia Hawaiif Idaho Illinois Indiana Iowag Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Flat Disregarda Variable Disregardb 0 100% for first three months, 20% thereafter $150 33%, 25%, 20%, 15%, 10% for years one to five, zero thereafter $90 30% 0 68% $225 50% Same as pre-TANFd Same as pre-TANFe 0 100% (up to poverty level) Same as pre-TANFd Same as pre-TANFe $100 50% $200 50% Same as pre-TANFd Same as pre-TANFe $250 48.8% 0 40% 0 67% Same as pre-TANFd Same as pre-TANFe 0 60% $90 40% Zero for first two months, same as pre-TANF after two monthsd 100% for first two months, same as pre-TANF thereaftere $1,020 for first six months, $120 thereafter — $108 50% 0 35% $120 50% $200 20% 0 38% 100% for first six months if employed full-time within one Zero for first six months if employed full-time within one month after first benefit or start of formal job-search activity month after first benefit or start of formal job-search activity, $90 thereafter, $90 otherwise $90 67% for first twelve months, zero thereafter $200 for first twenty-four months, $100 thereafter 25% for first twenty-four months, zero thereafter 0 20% Zero for first twelve months, 100% for first three months, 50% for next nine months, $90 thereafter if monthly earnings less than $450 20% thereafter if monthly earnings exceed $450 0 50% 0 100% for first month, 50% thereafter $150 50% $90 46% 0 100% for first three months, 27.5% thereafter Zero if monthly earnings less than $333.33, 27% thereafter $182 for first eight months, $145 for next two months, $108 for next two months if monthly earnings less than $333.33, zero if earnings exceed $333.33 Source: Adapted by authors from U.S. Department of Health and Human Services (2000, pp. 201-3). Note: TANF is the Temporary Assistance for Needy Families program. a The flat disregard is the initial amount of earnings that is disregarded when calculating benefits. The variable disregard is the percentage of earnings above the flat disregard that is disregarded when calculating benefits. c Disregard stipulated as 20 percent and 60 percent of remainder. d Pre-TANF flat disregard is $120 for first twelve months, $90 thereafter. e Pre-TANF variable disregard is one-third for first four months of earnings, zero thereafter. f Disregard stipulated as 20 percent, then $200, then 36 percent of remainder. g Disregard stipulated as 20 percent and 50 percent of remainder. h Disregards are the same as pre-TANF for families not subject to time limits. If earnings exceed poverty level, families are not eligible for benefits. i Formally, the variable disregard operates as “fill-the-gap budgeting,” rather than as an earned income disregard. j Wisconsin has no benefit formula. Benefits are zero for families with earnings. b FRBNY Economic Policy Review / September 2001 107 Table 1 Earnings-Disregard Policies for TANF Recipients (continued) January 2000 State Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginiah Washington West Virginia Wisconsinj Wyoming Flat Disregarda Variable Disregardb $250 $120 0 0 $170 Zero for first four months, $100 thereafter $90 $150 $120 $100 $150 0 0 0% 50% 50% 50% 50% 50% 50% for first four months, zero thereafter 20% 0 90% for first four months, zero thereafter 50% 25% Zero for income below poverty level, 100% otherwisei 50% Varies, averages 40% — — $200 per spouse 0 During the period in which states were incorporating financial incentives into their welfare benefit formulas, work effort among welfare recipients increased dramatically. From 1993 to 1997, employment among single mothers on welfare rose by 14 percentage points. According to Blank, Card, and Robins (2000), welfare mothers accounted for close to one-half of the rise in work by all single mothers over this period. As the authors explain, these rises are especially notable in view of the rapid decline in welfare use over the same period, which might have been expected to shift the pool of remaining welfare participants toward a more disadvantaged and less work-ready population. Even with this potential selection effect, however, work effort among welfare recipients rose. Of course, the work effort of welfare recipients was probably affected by other changes that occurred during this period. One important change was a substantial expansion of the earned income tax credit (EITC). Blank, Card, and Robins show that while some of the rise in employment among welfare recipients is undoubtedly due to the expansion of the EITC, some of it is also probably due to the adoption of enhanced welfare disregards. A randomized experiment in Minnesota also shows that enhanced disregards encourage work (Miller et al. 2000). Despite substantial increases in work effort among welfare recipients in recent years, most recipients remain out of work or are working too few hours to be economically 108 Using Financial Incentives self-sufficient. In a study of welfare leavers in Michigan, Danziger (2000) finds that one reason why poverty has not declined as fast as welfare caseloads is that few former recipients are working full-time, full-year. 5 Given the existence of time limits, it is crucial that recipients become employed full-time before exiting welfare.6 III. Encouraging Full-Time Work by Welfare Recipients: The SSP Program Few of the enhanced disregards being used by states are structured to encourage full-time work. The same may also be said for the EITC. The reason is that the financial rewards from working can be achieved at low levels of work effort as well as at high levels. For example, in 1999, the EITC for a family with two children increased with earnings at the rate of about 40 percent up to earnings of $9,500 per year and was constant between earnings of $9,500 and $12,500. For incomes above $12,500, the subsidy was phased out at the rate of about 21 percent, or until earnings reached $30,850. Thus, a person receiving a wage of $6 per hour would have received the maximum EITC subsidy of $3,816 (more than $300 per month) by working full-time (for example, forty hours per week). However, a substantial subsidy could also have been received for part-time work, and the full subsidy can be received at parttime work if the person is earning much more than $6 per hour. Similarly, welfare recipients in most states can benefit from enhanced disregards at less than full-time work as well as at full-time work. The fact that full-time work is relatively infrequent among welfare recipients suggests a possible need for restructuring financial incentives to encourage more full-time work. A social experiment being conducted in the Canadian provinces of British Columbia and New Brunswick is testing a financial incentive program that rewards welfare recipients only if they work full-time.7 The program is called the Self-Sufficiency Project, or SSP. Under SSP, which began in late 1992, longterm, single-parent welfare recipients (those receiving benefits for at least a year)8 who take a full-time job within one year are eligible to receive an earnings supplement for up to three years.9 The SSP supplement is quite generous: in certain cases, it can double a person’s earnings. For example, in New Brunswick, someone earning $10,000 per year (say, working forty hours per week for fifty weeks at $5 per hour) would receive supplementary payments totaling $10,000 per year.10 As long as the recipient continues to work full-time, the supplement can be received for up to three years.11 The SSP supplement bears some resemblance to the negative income tax (NIT), which was proposed as an alternative to welfare more than thirty years ago. There are three main differences between SSP and the NIT, however. First, SSP only pays benefits if the recipient works full-time. Second, it is targeted to welfare recipients, whereas the NIT was envisioned as a universal program. Third, SSP is available only for a limited period (three years), whereas the NIT did not have a time limit. Because of these differences, SSP strongly encourages work, whereas the NIT was found to discourage work. SSP has been remarkably successful during its early years of implementation. In the fifth quarter after the program began, full-time employment of the program group was more than double the full-time employment of the control group, 29 versus 14 percent (see Card and Robins [1998] and Michalopoulos et al. [2000]). SSP achieved this effect primarily by moving people from nonemployment to full-time employment, but a significant number of people also switched from part-time to full-time employment. Although SSP increased government transfer payments by about $55 per month (net of taxes), each $1 the government spent on additional transfer payments brought more than $2 of increased earnings and led to more than $3 of additional income for program group members. By way of contrast, the NIT generated less than $1 of additional earnings for each $1 of additional government transfer payments (Keeley et al. 1978). SSP also reduced poverty (the fraction of the program group having family incomes below the low-income threshold) by 12 percentage points and increased spending on food, clothing, and shelter.12 The early success of SSP in Canada raises the intriguing question of whether such a program would generate similar effects in the United States. Although the welfare systems in Canada and the United States are similar, differences make it difficult to draw comparisons. This has been especially true since PRWORA was enacted. As we indicated earlier, the U.S. system now imposes a time limit on welfare receipt, and there is a strong emphasis on placing recipients in work activities before the time limit is reached. The Canadian welfare system currently does not have time limits, although there is an emphasis on promoting economic self-sufficiency through work, as evidenced by the Canadian government’s willingness to test SSP on a pilot basis. In designing SSP, researchers at Manpower Demonstration Research Corporation used a microsimulation model to predict the impacts of alternative program models. As described in Greenberg et al. (1995) and Michalopoulos (1999), the model performed extremely well in predicting the eventual effects of the SSP program tested. Given its proven accurate predictive ability, we use the model in this paper to estimate the effects of an SSPtype financial incentive program in the United States.13 IV. Implementing SSP in the United States Because welfare reform efforts are already under way in the United States and because the EITC has been expanded significantly since 1994, the effects of an SSP financial incentive superimposed on the old AFDC system are not of particular policy relevance. Instead, it is of interest to examine what would have happened if states had coupled the nonfinancial components of their welfare-to-work programs with an SSPtype earnings supplement instead of with enhanced earnings disregards.14 To answer this question, we use data from three welfare-towork programs currently operating in the United States that are similar to TANF programs and are being evaluated using an experimental design. They are the Portland (Oregon) Job Opportunities and Basic Skills Training (PJOBS) Program being evaluated as part of the National Evaluation for Welfareto-Work Strategies, the Florida Family Transition Program (FTP), and the Minnesota Family Investment Program (MFIP). The features of these three welfare-to-work programs are described in Table 2. FRBNY Economic Policy Review / September 2001 109 Table 2 Features of the Welfare-to-Work Programs in Three States Program/ Study Period Minnesota (MFIP)/ 1994 to 1995 Florida (FTP)/ 1994 to mid-1998 Financial Incentives Other Features Employment Characteristics of Enrolled Families 1. Benefits increased by 20% for workers and reduced by 62% with earnings 1. Mandatory employment-focused activities 1. 52% of long-term recipient full MFIP program group in urban counties employed in quarter seven (N=676) 2. Benefits may not exceed benefits for nonworkers 2. Direct child-care payments to providers 2. 42% of long-term recipient MFIP financial incentives only program group in urban counties employed in quarter seven (N=681) 3. Food stamps cash-out 3. 38% of long-term recipient control group in urban counties employed in quarter seven (N=687) 1. Twenty-four-month time limit on benefits 1. 53% of program group employed in last quarter of year two (N=1,405) 2. Intensive case management 2. 45% of control group employed in last quarter of year two (N=1,410) 1. Disregard of $200 plus one-half of remaining earnings 3. Enhanced employment and training services 4. Parental responsibility mandates Oregon (PJOBS)/ 1993 to mid-1996 1. Disregard of $30 plus one-third of remaining earnings for first four months, no disregard after four months (pre-TANF rules) 1. Mandatory employment-focused activities that were strictly enforced 1. 46% of program group employed in last quarter of year two (N=3,529) 2. Integrated case management 2. 35% of control group employed in last quarter of year two (N=2,018) 3. Employment-focused Sources: Miller et al. (2000); Bloom et al. (2000); Scrivener et al. (1998). Note: MFIP is the Minnesota Family Investment Program; FTP is the Family Transition Program; PJOBS is the Portland (Oregon) Job Opportunities and Basic Skills Training Program; TANF is Temporary Assistance for Needy Families. Each program in Table 2 is similar to the TANF programs currently operating at the state level. In the table, a distinction is made between the financial incentive features of the programs and their other features. As the table indicates, over the study periods, two of the three programs (MFIP and FTP) had enhanced disregards. Furthermore, as indicated in Table 1, Oregon subsequently introduced a 50 percent variable disregard into its TANF program. All three programs use intensive case management and mandatory employmentfocused services, as outlined in the TANF legislation. Florida has instituted a shorter intermediate time limit (twenty-four or 110 Using Financial Incentives thirty-six months, depending on how job-ready a recipient is) than is required by the federal legislation. Minnesota has cashed out the food stamp program, and has turned food stamps, general assistance, and AFDC into one welfare program. Having one welfare program makes MFIP more similar to the Canadian Income Assistance program. Follow-up survey data are available on program and control group members for each of these three programs. For PJOBS and FTP, the data are available for two years; for MFIP, they are available for three years. The microsimulation analysis uses follow-up data from the second follow-up year for all three studies. By this time, members of the program group in each of the studies would have had some chance to respond to the TANF-like provisions they faced. Furthermore, the second follow-up year falls between 1995 and 1997 for all three studies; therefore, members of the program group probably had also responded to the expanded EITC. Using the survey data from each program and the microsimulation model, it is possible to estimate what additional effects, if any, an SSP-type program would have. In the case of Oregon and Minnesota, it is also possible to estimate how the SSP financial incentive would compare with the enhanced disregards adopted by those states in response to TANF. The post-TANF enhanced disregard currently in existence in Florida is identical to the financial incentive used in the FTP study. If adopted in the United States, an SSP program could be operated as a separate program, as it is in Canada. Berlin (2000) suggests that if a separate program was not created, and SSP operated as part of the existing welfare system, it might make sense for the TANF time-limit clock to stop ticking for people working full-time. Thus, the time spent working full-time would not count against the sixty-month (or less) TANF time limit. Of course, an SSP program could also have a time limit (as it does in Canada), which would limit its cost.15 With a time limit, the spirit of the TANF legislation would be maintained, but in a separate context that uses financial incentives to encourage full-time work. Two SSP financial incentive schemes are examined in this paper. All are patterned after the programs being tested in British Columbia and New Brunswick. To be included in the simulation, a welfare recipient must have been receiving AFDC at the end of the two-year follow-up period as well as in eleven of the twelve prior months. Once eligible, the welfare recipient qualifies for the earnings supplement if a full-time job of thirty or more hours per week paying at least the minimum wage is taken. In addition, the recipient cannot simultaneously receive welfare and the earnings supplement. The SSP financial incentive operates by paying people who meet the full-time work requirement a supplement equal to onehalf the difference between a “target” earnings level and actual earnings (see endnote 10 for the exact formula used). For the purposes of this paper, we examine the effects of programs using two target earnings levels: $20,000 and $30,000.16 For the target earnings level of $20,000, if the person works forty hours per week for fifty weeks per year and earns $7 per hour (about the average wage in our samples), the annual SSP subsidy would be $3,000, or roughly one-fifth of annual earnings of $14,000. For the target earnings level of $30,000, the SSP subsidy would be $8,000, or roughly three-fifths of annual earnings. Although these subsidy amounts seem substantial, it should be kept in mind that the recipient is required to give up AFDC (TANF) benefits in order to receive the subsidy. In Oregon, the average annual AFDC benefit was close to $6,000 per year, which is substantially more than the SSP subsidy under the program with the lowest target earnings level ($20,000) and about $2,000 less than the SSP subsidy under the program with the highest target earnings level ($30,000). Furthermore, people who work might also pay federal and state income taxes, which further reduce the government costs of the program. However, people who work qualify for the EITC and some people who would have left AFDC (TANF) without being offered the SSP financial incentive will receive subsidies. The chart shows, for the three sites, the net weekly income by weekly work effort for a single mother of two earning $8 per hour under SSP programs with a $20,000 and $30,000 target earnings level. For reference, net weekly income is also shown under the traditional AFDC earnings disregard ($120 per month disregarded, taxed at the rate of 100 percent thereafter) and the post-TANF earnings disregards adopted by states (a 50 percent variable disregard in Oregon, the same $200 flat and 50 percent variable disregards as those used in the experimental FTP program in Florida, and a 38 percent variable disregard in Minnesota that differed from the disregard used in the experimental MFIP program).17 The three panels of the chart illustrate how the traditional AFDC earnings disregards provide little incentive for welfare recipients to work. As shown in the top panel, for example, net weekly income in Oregon is relatively constant, between four and eighteen hours of work per week for a mother of two who earns $8 per hour. This is the range over which earnings have exceeded the AFDC flat disregard ($120 per month) and are deducted dollar-for-dollar from the AFDC benefit. The post-TANF enhanced earnings disregards adopted by the states improve the financial incentives to work parttime, but leave the financial incentives to work full-time pretty much unchanged. In contrast, SSP does not provide an incentive to work less than thirty hours per week but substantially increases the incentive to work thirty or more hours. This is especially true of the SSP program simulated with a $30,000 target level. For each of the SSP programs simulated, we report effects (or changes) that would occur as a result of the SSP program on annual labor force outcomes (full-time and part-time employment, hours of work, and earnings), welfare outcomes (receipt of AFDC, food stamps, and SSP), and various components of net income (AFDC, food stamps, EITC, SSP, and income taxes).18 For comparison, estimated effects on these outcomes of the TANF earnings disregards are also reported. It is important to note that these effects are for the chosen sample. Namely, these are effects for people FRBNY Economic Policy Review / September 2001 111 who were receiving AFDC for almost the entire second year of the follow-up period—that is, sample members who would be eligible for the SSP supplement at the end of the second year of follow-up according to how the program operates in Canada. In general, the effects of changes in Budget Constraints under Alternative Financial Incentive Schemes: Single Mother with Two Children and Wage Rate of $8 per Hour Net weekly income (dollars) 500 Oregon SSP/$30K 400 300 TANF SSP/$20K AFDC 200 100 0 500 Florida SSP/$30K 400 300 TANF SSP/$20K AFDC 200 100 0 500 Minnesota SSP/$30K 400 TANF 300 AFDC SSP/$20K 200 100 0 0 5 10 15 20 25 30 Hours per week 35 40 45 50 Source: Robins, Michalopoulos, and Pan (2000). Note: SSP is the Self-Sufficiency Project (a Canadian program); TANF is Temporary Assistance for Needy Families; AFDC is Aid to Families with Dependent Children. 112 Using Financial Incentives TANF earnings disregards for the full welfare population will differ from their effects for the simulation samples because the full welfare population includes people who received welfare for shorter periods of time. The characteristics of the three welfare-to-work samples are presented in Table 3. The first three columns show the average characteristics of all people who were randomly assigned to the program group in each site. The latter three columns show the characteristics of the samples used in the simulations. The simulation samples have fewer people than the program groups because they were limited to people who were receiving AFDC at the end of the second year of follow-up and had been receiving AFDC for eleven of the twelve months prior to the end of the second year. This sample has real-world relevance because an SSP-type financial incentive offered in the United States would most likely be offered only to those people still receiving TANF benefits at the time the SSP program would be introduced. Almost all families in the program groups are headed by never-married women. In Oregon, about two-fifths were never married at the beginning of the follow-up period; in Florida, three fifths were never married; in Minnesota, about two-thirds were never married. The average mother was about thirty years old and more than two-thirds had children less than six years of age. In Oregon, about onequarter were black; in Florida, more than half were black; and in Minnesota, about two-fifths were black. Roughly one-third of the sample lacked a high-school diploma or a GED and roughly one-third of the sample received AFDC as a child. At the start of the follow-up period, just over 10 percent of the samples worked, with only a small fraction working full-time. In the year prior to the start of the welfare-to-work program, between one-fourth and two-fifths of the mothers were employed. Thus, many of these mothers had some sort of work experience. In Oregon and Florida, about half of the program group received AFDC or food stamps the full year, while in Minnesota more than three-quarters received AFDC or food stamps the full year. The Minnesota sample is somewhat more disadvantaged than the Oregon and Florida samples because it includes only long-term recipients from urban counties. These are sample members who had been on AFDC for at least twenty-four of the thirty-six months prior to random assignment, and it is the group that was most affected by the Minnesota program (Miller et al. 2000). Each of the experimental welfare-to-work programs significantly increased employment (Bloom et al. 2000; Miller et al. 2000; Scrivener et al. 1998).19 Table 4 shows selected effects of each of these programs, measured as differences Table 3 Characteristics of Program Group Members in Three Welfare-to-Work Programs Program Group Members in Simulation Samples All Program Group Members Characteristic Characteristic at baseline Percentage female Percentage married, living with spouse Percentage never married Age Percentage with child under age six Percentage black Percentage white Highest grade completed Percentage with high-school diploma or GED Percentage receiving AFDC as a child More than five years Less than five years Percentage employed Full-time (more than thirty hours per week) Part-time (less than thirty hours per week) Characteristic in year prior to baseline Percentage employed Months receiving AFDC Months receiving food stamps Months receiving either AFDC or food stamps Percentage receiving AFDC in every month Percentage receiving food stamps in every month Percentage receiving either AFDC or food stamps in every month Characteristic during follow-up period Percentage employed Full-time (more than thirty hours per week) Part-time (less than thirty hours per week) Total hours worked Earnings Average hourly wage Sample size Oregon (1) Florida (2) Minnesota (3) Oregon (4) Florida (5) Minnesota (6) 94.9 2.7 42.7 31.6 65.1 27.4 62.7 11.2 67.2 24.7 13.9 10.9 12.9 1.7 11.2 99.0 0.7 57.9 28.9 70.8 56.2 42.4 11.1 57.9 19.8 13.9 6.0 11.7 4.8 6.9 99.7 0.3 69.9 29.0 72.4 39.2 47.8 11.5 65.8 37.9 25.1 12.8 11.1 3.5 7.6 98.7 1.3 53.2 32.6 70.1 38.2 55.3 11.0 55.8 31.3 14.9 16.4 6.5 1.3 5.2 100.0 0.0 69.2 29.3 74.6 73.8 26.2 11.0 49.2 31.7 21.7 10.0 10.8 6.2 4.6 100.0 0.0 68.9 29.8 70.4 44.6 42.7 11.5 65.3 39.9 27.8 12.1 9.2 2.8 6.4 39.9 8.0 8.5 9.1 45.1 53.5 42.1 8.1 9.2 9.4 48.8 60.9 30.1 10.6 10.4 10.6 75.5 72.0 23.4 9.6 10.0 10.4 66.2 75.3 27.7 10.3 11.1 11.2 71.2 78.8 27.8 10.7 10.5 10.7 79.0 75.8 58.2 65.2 75.5 80.5 86.4 79.0 65.3 17.5 47.8 652 $4,882 $7.41 75.6 21.4 54.2 818 $4,740 $6.49 66.1 11.0 55.1 607 $4,714 $7.57 26.0 0.0 26.0 51 $295 $6.54 48.5 3.0 45.5 252 $1,135 $4.45 61.6 2.7 58.9 442 $3,015 $6.98 297 299 372 77 66 219 Source: Manpower Demonstration Research Corporation calculations using baseline information forms and two-year client survey data from the Portland (Oregon) Job Opportunities and Basic Skills Training Program evaluation, two-year client survey data from the Florida Family Transition Program evaluation, thirty-six-month client survey data from the Minnesota Family Investment Program evaluation, and unemployment insurance earnings records and public assistance benefit records data from Oregon, Florida, and Minnesota. Note: The simulation samples include program group members who were not living with a spouse or partner at the time of the follow-up interview and who received Aid to Families with Dependent Children (AFDC) in the twenty-fourth follow-up month and in at least eleven of the twelve months prior to that. FRBNY Economic Policy Review / September 2001 113 Table 4 Effects on Employment and Earnings in the Oregon, Florida, and Minnesota Studies Outcome Percentage ever employed Quarter 4 Quarter 5 Quarter 6 Oregon (1) Florida (2) 7.1*** 9.0*** 11.1*** 4.5*** 4.3** 5.9*** Minnesota (3) 11.7*** 15.0*** 17.4*** Earnings Quarter 4 Quarter 5 Quarter 6 $191*** $201*** $267*** $90** $104** $150*** $150** $235*** $264*** Sample size a 5,547 2,815 1,363 Sources: The Oregon data are from Scrivener et al. (1998); the Florida data are from Bloom et al. (2000); the Minnesota data are from Miller et al. (2000). Notes: The Minnesota data are for long-term recipients in urban counties. The Minnesota study defined long-time receipt as two years or more in the prior three years. Random assignment dates were: Oregon, February 1993-December 1994; Florida, May 1994-October 1996; Minnesota, April-December 1994. Following the Oregon and Minnesota studies, Quarter 1 is defined as the calendar quarter in which random assignment falls. In the Florida study, Quarter 1 was defined as the quarter after the calendar quarter in which random assignment fell. Quarter 4 in the table therefore corresponds to Quarter 3 in Bloom et al. (2000). Effects are measured as differences between outcomes for program and control groups. A two-tailed t-test was used to assess the statistical signifiance of differences between outcomes for program and control group members. Rounding may cause slight discrepancies in sums and differences. a Sample sizes differ from those in Table 2. Sample sizes include program and control group members who responded or did not respond to the follow-up survey. *** Statistically significant at the 1 percent level. *** Statistically significant at the 5 percent level. between outcomes for program and control group families. Because the welfare-to-work programs had already successfully increased employment, adding the SSP financial incentive may generate smaller increases in employment than if the SSP program was superimposed on the old AFDC system. A similar argument can be made for the EITC. In Canada, SSP was introduced in an environment without an EITC-type program. Hence, some of SSP’s effects in Canada may have already occurred for the types of welfare recipients who responded to the EITC in the United States. When interpreting the effects of the SSP program in the United States, therefore, one should keep in mind that many of the people who would have 114 Using Financial Incentives responded to SSP by finding full-time employment may already have responded to either the EITC or the welfare-towork program. Columns 4-6 of Table 3 indicate that the simulation samples (those who were on AFDC at the end of the second year of follow-up and were on AFDC for at least eleven of the twelve prior months) are somewhat more disadvantaged than the full program groups. They are less likely to have a high-school diploma (although not by very much in Minnesota), they are less likely to have worked in the year prior to random assignment, they are much less likely to have worked during the follow-up period, and their average wages are lower. V. Simulation Results Estimated Effects of the TANF Earnings Disregards and SSP Table 5 reports the simulated outcomes under the AFDC disregard in the three samples.20 These outcomes include effects of all features of the experimental welfare-to-work programs except for the enhanced disregards. About one-fifth of the Oregon sample is employed and about one-half of the Florida and Minnesota samples are employed. Most of the employment is part-time, with few sample members employed full-time. This, of course, partly reflects the fact that for many welfare recipients fulltime employment would make them ineligible for benefits. Table 6 reports estimated effects of the TANF earnings disregards and the two SSP programs for each of the three samples. The first column for each sample (columns 1, 4, and 7) shows the estimated effect of the TANF earnings disregard used in the state. Each effect is measured relative to the preTANF AFDC flat earnings disregard of $120. The next two columns for each sample (columns 2-3, 5-6, and 8-9) show the estimated effects of the two SSP-type financial incentive programs. In Oregon, column 1 illustrates strikingly how variable earnings disregards currently being used by many states may be quite successful in moving welfare recipients to work, yet at the same time might not succeed in moving recipients to economic self-sufficiency. The TANF earnings disregard currently used in Oregon is estimated to have increased employment considerably among long-term recipients (recall that the simulation selects program group members in the study who were still on AFDC in the last month of the second year of follow-up and in at least eleven of twelve prior months). The 50 percent earnings disregard is estimated to have increased employment by just over 23 percentage points. All of this employment, however, is part-time. Comparing columns 2 and 3 with column 1 provides an estimate of what would have happened if Oregon had instituted an SSP-type financial incentive instead of a 50 percent earnings disregard. Unlike the 50 percent earnings disregard, the SSP-type financial incentive programs increase full-time employment. Furthermore, for both SSP-type programs, they do so without any net increase in cash transfers from the government. An SSP program with target earnings of $20,000 would increase the full-time employment rate by 5.2 percentage points, annual earnings by $406, and net annual Table 5 Annual Outcomes under the AFDC Disregard for the Simulation Samples Oregon (1) Florida (2) Minnesota (3) 22.1 0.0 22.1 51 299 45.5 3.0 42.4 227 1,027 51.6 6.9 44.8 379 2,725 Welfare outcomes (percent) AFDC receipt Food stamp receipt SSP receipt 100.0 96.1 0.0 95.5 97.0 0.0 99.1 NA 0.0 Income outcomes (dollars) AFDC Food stamps EITC SSP Taxes Net government assistancea Net incomeb 5,879 3,797 97 0 23 9,749 10,048 3,715 4,447 388 0 79 8,471 9,498 8,753 NA 858 0 250 9,361 12,085 Labor force outcomes Employment (percent) Full-time employment (percent) Part-time employment (percent) Hours of work Earnings (dollars) Source: Manpower Demonstration Research Corporation simulation model, using data from the Oregon, Florida, and Minnesota welfare-towork programs. Note: AFDC is Aid to Families with Dependent Children; SSP is the SelfSufficiency Project (a Canadian program); EITC is the earned income tax credit; TANF is Temporary Assistance for Needy Families. a Net government assistance is TANF + food stamps + EITC + SSP - taxes. b Net income is net government assistance + earnings. income by $261 (column 2). All of this occurs while net cash transfer payments from the government to recipients decrease by $146. An SSP program with target earnings of $30,000 would increase the full-time employment rate by 10.4 percentage points, annual earnings by $884, and net annual income by $946, without significantly increasing the amount of cash transfers to recipients by a statistically significant amount (column 3). Despite their sizable effects on full-time employment, the SSP-type programs do not generate nearly as much employment (either full- or part-time) in Oregon as the TANF earnings disregard does. The SSP program with target earnings of $30,000 would increase employment by 6.5 percentage points and the SSP program with target earnings of $20,000 would increase employment by only 1.3 percentage points. Nonetheless, the SSP-type incentives are estimated to reduce AFDC and food stamp payments substantially. Such reductions are possible because many of those who are estimated to go from part-time to full-time work when offered an SSP supplement are people working very few hours and receiving practically their full welfare grant amount. These results imply that an SSP program as generous as the ones tested in Canada would modestly increase employment and income among this group of persons in the Oregon welfare-to-work program at no additional cost to the government. Full-time employment would have increased by a somewhat greater amount. For the more generous SSP-type program, net family income would increase by almost three times the estimated increase generated by the TANF earnings disregard. It is important to emphasize that these effects are those in addition to the effects already generated by the expanded EITC. Without the EITC, the effects of SSP and the TANF disregard might have been substantially larger. Thus, in the present welfare environment, if the policy objective is to increase full-time employment and income without any additional cost to the government, our results suggest that a moderately generous SSP program could be somewhat more effective than the enhanced disregard actually adopted by Oregon under its TANF program. Simulation results for Florida’s FTP program are presented in columns 4-6 of Table 5. Unlike Oregon’s experimental welfare-to-work program, the FTP included an enhanced disregard ($200 flat and 50 percent variable) as part of its program. The enhanced disregard in Florida’s TANF program is identical to the enhanced disregard tested in the FTP program. The estimated effects of this enhanced disregard for the simulation sample are shown in column 4 of Table 5.21 Columns 5 and 6 show estimated effects of the two SSP-type programs. FRBNY Economic Policy Review / September 2001 115 Table 6 Simulated Annual Effects of SSP Financial Incentives on Participants in Three Welfare-to-Work Programs Incentives Introduced after Two Years for Families Still Receiving AFDC a Oregon Labor force outcomes Employment (percent) Full-time employment (percent) Part-time employment (percent) Hours of work Earnings (dollars) Florida Effect of SSP Effect of SSP $30,000 Target (3) Effect of TANF Disregardc (4) $20,000 Target (5) 1.3 5.2** -3.9* 66** 406** 6.5** 10.4*** -3.9* 144*** 884*** 0.0 0.0 0.0 26* 112 Effect of TANF Disregardb (1) $20,000 Target (2) 23.4*** 0.0 23.4*** 47*** 331*** Minnesota $30,000 Target (6) $20,000 Target (8) $30,000 Target (9) 0.0 15.2*** -15.2*** 126*** 595*** 1.5 24.2*** -22.7*** 240*** 1,159*** 12.3*** -0.9 13.2*** 87*** 565*** 0.0 13.2*** -13.2*** 45*** 295*** 0.0 23.7*** -23.7*** 144*** 929*** 0.5 NA 0.0 -2.7** NA 17.8*** -11.0*** NA 28.8*** 52 e 371*** 0 14 409*** 974*** -493*** e -25 806*** 96*** 193*** 488*** -1,486*** e -227*** 2,747*** 443*** 591*** 1,520*** Welfare outcomes (percent) TANF receipte Food stamp receipt SSP receipt 0.0 0.0 0.0 -5.2** -5.2** 5.2** -10.4*** -9.1*** 10.4*** 4.6* 3.0 0.0 -13.6*** -12.1*** 16.7*** -22.7*** -24.2*** 25.8*** Income outcomes (dollars) TANF Food stamps EITC SSP Taxes Net government assistancef Net incomeg -31 -64*** 109*** 0 25*** -11 320*** -260** -170** 69* 277** 62* -146** 261** -536*** -312*** 79** 1,077*** 245*** 62 946*** 168*** 33 57** 0 9 249** 361*** -384*** -549*** 242*** 1,046*** 70*** 286** 881*** -736*** -1,051*** 195*** 2,912*** 283*** 1,037*** 2,196*** Source: Manpower Demonstration Research Corporation simulation model, using data from Oregon, Florida, and Minnesota welfare-to-work programs. Note: SSP is the Self-Sufficiency Project (a Canadian program); AFDC is Aid to Families with Dependent Children; TANF is Temporary Assistance for Needy Families; EITC is the earned income tax credit. a Effects are changes relative to outcomes under the welfare-to-work program with a flat earnings disregard of $120 per month and a child support disregard of $50 per month (the pre-TANF disregard). b Oregon’s TANF program has a 50 percent variable earnings disregard. c Florida’s TANF program has a $200 flat disregard and a 50 percent variable disregard. d Minnesota’s TANF program has a 38 percent variable disregard. e Minnesota’s TANF benefit includes the cash value of food stamps. f Net government assistance is TANF + food stamps + EITC + SSP - taxes. g Net income is net government assistance + earnings. *** Statistically significant at the 1 percent level. *** Statistically significant at the 5 percent level. *** Statistically significant at the 10 percent level. 116 Using Financial Incentives Effect of SSP Effect of TANF Disregardd (7) Column 4 shows that the simulation predicts no changes in employment from the FTP (TANF) earnings disregard. Hours of work are estimated to increase slightly, but the incentive does not induce anyone to start working, nor does it induce anyone to switch from part-time to full-time work. Furthermore, the FTP financial incentive is estimated to increase the welfare caseload. The FTP financial incentive does increase annual net income, but much of this increase (more than two-thirds) comes from an increase in government spending. Both of the SSP programs would have substantially increased full-time employment, according to the simulation model. Under the less generous SSP program (target earnings of $20,000), full-time employment would have risen by just over 15 percentage points. Under the more generous SSP program (target earnings of $30,000), full-time employment would have risen by just over 24 percentage points. For the less generous SSP program, all of the increase in full-time employment is estimated to be the result of people switching from part-time to full-time work, with no net increase in overall employment. For the more generous SSP program, nearly all of the increase in full-time employment is estimated to be the result of people switching from part-time to full-time work, with only a 1.5 percentage point increase in overall employment. The SSP programs would have increased annual hours of work by between 126 and 240 hours and annual earnings by between $595 and $1,159. The less generous SSP program would have increased net family income by $881, at a net cost to the government of $286. The more generous SSP program would have increased net family income by more than $2,000 (which represents more than a 20 percent increase), at a net cost to the government of just over $1,000. The less generous SSP program costs about the same as the enhanced disregard of the FTP program, although the sources of costs are different for the two strategies (the SSP induces greater decreases in welfare payments and more taxes, but adds EITC costs plus SSP supplementary payment costs). However, SSP more than doubles the effect on family income, primarily because it induces a substantial amount of full-time employment. Overall, then, the main difference between SSP and the enhanced disregard in the FTP is the greater full-time employment associated with the SSP program. Thus, our simulation model predicts that many of the people still in the FTP and working part-time because of the EITC and the nonfinancial components of the FTP would have been induced to work full-time under an SSP program that conditioned benefits on full-time employment. Simulation results for Minnesota’s MFIP program are given in columns 7-9 of Table 6. Column 7 presents the estimated effects of Minnesota’s earnings disregard under TANF. As indicated earlier, Minnesota implemented an enhanced disregard that was slightly different from the one used in the experimental MFIP programs. The earnings disregard currently used in the Minnesota TANF program is a 38 percent variable disregard with no flat disregard. According to column 7, the simulation model predicts that the TANF earnings disregard increased part-time employment for our sample, but had virtually no effect on full-time employment.22 With the TANF financial incentive, overall employment is estimated to be 12.3 percentage points higher, annual hours of work eighty-seven hours higher, and annual earnings $565 higher than if there had been no enhanced earnings disregard. The simulation model predicts that the TANF enhanced earnings disregard has virtually no effect on welfare receipt for our sample. Net government costs are about $400 higher and net family income is close to $1,000 higher because of the enhanced earnings disregard. Similar to the Florida sample, the Minnesota simulations predict that the SSP programs would not increase employment, but would cause a substantial number of people to switch from part-time to full-time work (columns 8 and 9). Under the less generous SSP program ($20,000 target), full-time employment would increase by just over 13 percentage points. Under the more generous SSP program ($30,000 target), full-time employment would increase by almost 24 percentage points. For the less generous SSP program, net government assistance would rise by $193 and net family income would increase by more than twice that amount. For the more generous SSP program, net government cost would increase by $591 per person, while annual net family income would again rise by more than twice that amount. Effects of the EITC It is important to emphasize that all effects reported in this paper are in addition to changes already resulting from the EITC. To the extent that the EITC caused large employment gains, the effects of enhanced earnings disregards or of an SSPtype financial incentive may be reduced because those who are most apt to respond to financial incentives will already be working in response to the EITC. Table 7 shows the effects on labor force outcomes, welfare outcomes, and family income of taking away the EITC from the study samples. The table shows that much of the employment observed in the study samples was indeed generated by the EITC. If there had been no EITC, the employment rate would have been 12 percentage points lower in Oregon, 11 percentage FRBNY Economic Policy Review / September 2001 117 Table 7 Simulated Annual Effects of Taking Away the EITC Oregon Labor force outcomes Employment (percent) Full-time employment (percent) Part-time employment (percent) Hours of work Earnings (dollars) Welfare outcomes (percent) AFDC receipt Food stamp receipt Income outcomes (dollars) AFDC Food stamps EITC Taxes Net government assistanceb Net incomec Florida Outcomes under AFDC Disregarda (1) Effect of Taking Away EITC (2) Outcomes under AFDC Disregarda (3) 22.1 0.0 22.1 51 299 -11.7*** 0.0 -11.7*** -19*** -97*** 45.5 3.0 42.4 227 1,027 100.0 96.1 0.0 0.0 5,879 3,797 97 23 9,749 10,048 29* 15*** -97*** -7*** -46** -143*** Minnesota Effect of Taking Away EITC (4) Outcomes under AFDC Disregarda (5) Effect of Taking Away EITC (6) -10.6*** -1.5 -9.1** -73*** -349*** 51.6 6.9 44.8 379 2,725 -7.3*** 0.9 -8.2*** -63*** -346*** 95.5 97.0 3.0 1.5 87.2 99.1 1.8** -0.9 3,715 4,447 388 79 8,471 9,498 165*** 79 -388*** -27*** -118* -467*** 5,290 3,463 858 250 9,361 12,085 440*** -44*** -858*** -2 -460*** -806*** Source: Manpower Demonstration Research Corporation simulation model, using data from the Portland (Oregon) Job Opportunities and Basic Skills Training (PJOBS) Program, Florida’s Family Transition Program (FTP), and the Minnesota Family Investment Program (MFIP). Note: EITC is the earned income tax credit; AFDC is Aid to Families with Dependent Children; SSP is the Self-Sufficiency Project (a Canadian program). a Outcomes are for an AFDC program with a flat earnings disregard of $120 per month and a child support disregard of $50 per month. Outcomes for Florida and Minnesota samples are simulated outcomes. See text for details of the simulation. b Net government assistance is AFDC + food stamps + EITC + SSP - taxes. c Net income is net government assistance + earnings. *** Statistically significant at the 1 percent level. *** Statistically significant at the 5 percent level. *** Statistically significant at the 10 percent level. points lower in Florida, and 7 percentage points lower in Minnesota. In Oregon, the reduction represents over half of the observed employment in the sample. According to the simulations, very few persons in the study samples were combining welfare and full-time work before the EITC was taken away. Thus, taking away the EITC results in mostly a reduction in part-time employment. It is important to note that these estimated effects of the EITC are conditional on receiving welfare benefits. It is quite possible that the EITC has induced many persons to work full-time and leave welfare, but our simulations do not capture this effect because our samples only consist of persons on welfare. 118 Using Financial Incentives VI. Summary and Conclusions Since the enactment of the Personal Responsibility and Work Opportunity Reconciliation Act in 1996, most states have included financial incentives in their overall welfare-to-work programs. These financial incentives have taken the form of enhanced disregards that allow recipients to keep more of their welfare benefits as earnings increase. Prior to PRWORA, under the Aid to Families with Dependent Children program, benefits after the first year of earnings were reduced dollar-for-dollar with earnings. This meant that recipients faced a 100 percent “tax rate” on their earnings, which can impose a significant disincentive to work. Given the existence of time limits under PRWORA and the importance of moving welfare recipients into the workforce, most states have reduced this 100 percent tax rate through enhanced disregards. The intent of the enhanced disregards is to provide a work incentive and to ease the transition from welfare to work. One of the problems with enhanced disregards is that they often provide only an incentive to work part-time. Faced with a sudden loss in welfare benefits when the time limit is reached, many recipients may find that part-time earnings are not enough to allow them to be economically self-sufficient. Although these families will still be eligible for the earned income tax credit, the tax credit also provides only limited incentives to work full-time. This paper has presented results from a simulation model to examine whether an alternative form of financial incentive could increase full-time employment among long-term welfare recipients. The alternative financial incentive scheme considered is based on the Self-Sufficiency Project, an experimental program being tested in two provinces in Canada. SSP provides a direct incentive to work full-time because it conditions benefits on full-time work. The program is being evaluated using a random assignment design, and the results thus far indicate that it is generating sizable increases in full-time employment, at little additional transfer cost to the Canadian government. Because the welfare-to-work programs under Temporary Assistance for Needy Families in the United States are different from the welfare-to-work programs in Canada, and because Canada does not have an EITC, it is not clear what additional effects an SSP-type program might have in the United States. This paper has used microsimulation analysis to predict what the effects might be if an SSP-type program was adopted in the United States. Our results indicate that an SSP-type program in the United States—in place of the enhanced disregards currently being used—could have significantly greater effects on full-time employment for long-term welfare recipients at modest additional cost to the government. Perhaps the most attractive feature of SSP is its ability to achieve gains in family income that are as much as three times the increase in government cost. Such a high “efficiency ratio” is rarely seen in financial incentive programs. Thus, an SSP-type program in the United States may be an attractive way to ease the transition from welfare to work under a system of time-limited welfare benefits. FRBNY Economic Policy Review / September 2001 119 Endnotes 1. These are the rates for all families. The legislation stipulates much higher rates for two-parent families. In fiscal year 1999, for example, the two-parent family participation requirement was 90 percent. 2. Several states have failed to meet the two-parent participation rate requirement, however. 3. Enhanced earnings disregards are not a new policy. From 1967 until 1981, the federal AFDC program provided modest financial incentives for welfare recipients to work, in the form of a 33 percent earnings disregard. Some of the earnings disregards introduced since the early 1990s are similar to the pre-1982 disregard. 4. From 1967 to 1981, the AFDC program had a flat disregard of $30 and a variable disregard of one-third throughout the duration of a welfare spell. 5. Danziger’s research is summarized in Joint Center for Poverty Research (2000). 6. We emphasize the importance of full-time work purely from the standpoint of economic self-sufficiency. We realize that full-time work among single parents may have drawbacks (such as adversely affecting child development) and that a case may be made against encouraging full-time work for single parents. 7. The welfare system in Canada is called Income Assistance. Canada has no food stamp program, so cash benefits in Canada generally are higher than they are in the United States. 8. The restriction to long-term recipients is intended to minimize “entry effects” (people applying for welfare in order to receive the supplement) and “windfall effects” (benefits accruing to recipients who would have left welfare and worked full-time anyway in the absence of the earnings supplement). As indicated in Card, Robins, and Lin (1998), this provision substantially limited entry and windfall effects. 9. Full-time work under SSP is defined as thirty or more hours per week. 10. Formally, the SSP subsidy is given by .5(E* - E), where E* is “target” earnings ($30,000 in New Brunswick and $37,000 in British Columbia, both in Canadian dollars, when the SSP began) and E is actual earnings. The subsidy is available only to people working thirty hours per week or more, and has been adjusted upward slightly for inflation since 1992. 120 Using Financial Incentives 11. Because the benefits are targeted to long-term welfare recipients, there is some horizontal inequity because similar workers not on welfare have lower income. However, horizontal inequities exist for any program in which some recipients mix welfare and work. 12. A companion experiment, conducted on a group of new applicants for welfare in British Columbia, did not lead to any net increase in government cash transfer payments and had similar effects on employment, income, and poverty (see Michalopoulos, Robins, and Card [1999]). 13. For a detailed discussion of the microsimulation model, see Robins, Michalopoulos, and Pan (2000). The model incorporates the notion of welfare stigma and utilizes the economic framework developed by Moffitt (1983). 14. Perhaps an equally interesting question concerns the effects of an SSP-type earnings supplement program in addition to enhanced earnings disregards. The effects of such a policy are presented in Robins, Michalopoulos, and Pan (2000). 15. TANF costs would be reduced in the short run as persons shifted from TANF to the SSP. Of course, these reduced welfare costs would be offset by SSP costs. For the SSP program in Canada, the cost of the SSP slightly exceeded the reduced welfare cost for long-term recipients, but was about the same as the reduced welfare cost for new applicants (net of the additional income taxes resulting from the additional full-time work). Part of the additional SSP cost was “windfall,” resulting from SSP benefits being paid to persons who would have left welfare and worked full-time anyway in the absence of the SSP-type financial incentive. 16. Recall that in the actual SSP programs being tested in Canada, the target earnings levels (in Canadian dollars) are $37,000 in British Columbia and $30,000 in New Brunswick. At an exchange rate of .75 U.S. dollars per Canadian dollar, these target earnings levels in U.S. dollars are $27,750 and $22,500, respectively. 17. Minnesota’s TANF variable earnings disregard was 36 percent until October 1999, when it increased to 38 percent. The simulations presented in this paper were performed using the 38 percent disregard. 18. Although the effects are presented as changes in annual outcomes, the simulation model is based on weekly decisions concerning these outcomes. Hence, it is implicitly assumed that all predicted weekly changes occur for each week during the year. Although this is a Endnotes (Continued) reasonable assumption given the nature of the SSP earnings supplement offer, a more comprehensive simulation model would incorporate decisions on how many weeks to work as well as how many hours to work per week. The simulations are based on an underlying economic model that assumes welfare recipients choose how much to work and whether to receive welfare in order to maximize their economic well-being. Receiving welfare is assumed to be stigmatizing. The welfare recipient is assumed to weigh the benefits of the additional income from SSP with the reduced time in activities outside of work (such as child-rearing and leisure-time activities). The parameters of the underlying economic model are taken from Moffitt (1983) and are updated to the present time. For full details on the mechanics of the simulation model and how all outcomes are calculated, see Robins, Michalopoulos, and Pan (2000). 19. Changes in employment over time for members of the program group do not necessarily represent effects of the welfare-to-work program because other changes are also occurring that affect employment. For one to measure effects validly, behavior of a randomly selected control group must also be tracked and compared with the behavior of the program group. Such an “experimental” approach to measuring program effects is being used in each of the welfare-to-work programs examined in this paper. 20. Because the Oregon experimental program used the AFDC disregard, these are actual mean outcomes for the Oregon sample. 21. These results do not represent experimental effects of the financial incentive component of Florida’s FTP program. Such a program was never tested experimentally. Instead, the numbers represent the simulated effects of the FTP’s enhanced earnings disregard for the simulation sample of long-term welfare recipients. Furthermore, the numbers were derived from taking away the financial incentive from the simulation sample. To the extent that the sample of long-term recipients would have been different if there had not been a financial incentive as part of the FTP program, the effects of offering the financial incentive will differ from the effects of taking away the financial incentive from people who were long-term recipients when the financial incentive was offered. 22. In results not shown here but reported in Robins, Michalopoulos, and Pan (2000), we found that the MFIP experimental earnings disregard reduced full-time employment and increased part-time employment. FRBNY Economic Policy Review / September 2001 121 References Bane, Mary Jo, and David Ellwood. 1994. Welfare Realities: From Rhetoric to Reform. Cambridge: Harvard University Press. Berlin, Gordon. 2000. “Encouraging Work, Reducing Poverty: The Impact of Work Incentive Programs.” March. New York: Manpower Demonstration Research Corporation. Blank, Rebecca M., David Card, and Philip K. Robins. 2000. “Financial Incentives for Increasing Work and Income among Low-Income Families.” In Rebecca M. Blank and David Card, eds., Finding Work: Jobs and Welfare Reform, 373-419. New York: Russell Sage Foundation. Bloom, Dan, James J. Kemple, Pamela Morris, Susan Scrivener, Nandita Verma, and Richard Hendra. 2000. “The Family Transition Program: Final Report on Florida’s Initial Time-Limited Welfare Program.” December. New York: Manpower Demonstration Research Corporation. Card, David, and Philip K. Robins. 1998. “Do Financial Incentives Encourage Welfare Recipients to Work? Evidence from a Randomized Evaluation of the Self-Sufficiency Project.” In Solomon W. Polachek, ed., Research in Labor Economics, vol. 17, 1-56. Stamford, Conn.: JAI Press. Card, David, Philip K. Robins, and Winston Lin. 1998. “Would Financial Incentives for Leaving Welfare Lead Some People to Stay on Welfare Longer? An Experimental Evaluation of ‘Entry Effects’ in the Self-Sufficiency Project.” NBER Working Paper no. 6449, March. Citro, Constance F., and Eric A Hanushek, eds. 1991. Improving Information for Social Policy Decisions: The Uses of Microsimulation Modeling, vol. i. Washington, D.C.: National Academy Press. Danziger, Sheldon. 2000. “Approaching the Limit: Early Lessons from Welfare Reform.” Paper presented at the Joint Center for Poverty Research Conference “Rural Dimensions of Welfare Reform.” May. Gallagher, L. Jerome, Megan Gallagher, Kevin Perese, Susan Schreiber, and Keith Watson. 1998. “One Year after Federal Welfare Reform: A Description of State Temporary Assistance for Needy Families (TANF) Decisions as of October 1997.” Assessing the New Federalism, Occasional Paper no. 6, June. Washington, D.C.: Urban Institute. 122 Using Financial Incentives Greenberg, David H., David Long, Daniel Meyer, Charles Michalopoulos, and Philip K. Robins. 1995. “Using Microsimulation to Help Design Pilot Demonstrations: An Illustration from the Canadian Self-Sufficiency Project.” Evaluation Review 19, no. 6 (December): 687-706. Joint Center for Poverty Research. 2000. “How They’re Faring: Work and Earnings under Welfare Reform.” Poverty Research News 4, no. 5 (September-October): 5-7. Keeley, Michael C., Philip K. Robins, Robert G. Spiegelman, and Richard W. West. 1978. “The Labor Supply Effects and Costs of Alternative Negative Income Tax Programs.” Journal of Human Resources 13, no. 1 (winter): 3-36. Lin, Winston, Philip K. Robins, David Card, Kristen Harknett, and Susanna Lui-Gurr. 1998. “When Financial Incentives Encourage Work: Complete Eighteen-Month Findings from the SelfSufficiency Project.” September. Ottawa: Social Research and Demonstration Corporation. Michalopoulos, Charles. 1999. “Comparison of Simulated Impacts of SSP and MFIP with Results from Evaluation.” Unpublished paper, Manpower Demonstration Research Corporation, May 14. Michalopoulos, Charles, Philip K. Robins, and David Card. 1999. “When Financial Incentives Pay for Themselves: Early Findings from the Self-Sufficiency Project’s Applicant Study.” May. Ottawa: Social Research and Demonstration Corporation. Miller, Cynthia, Virginia Knox, Lisa A. Gennetian, Martey Dodoo, Jo Anna Hunter, and Cindy Redcross. 2000. “Reforming Welfare and Rewarding Work: Final Report on the Minnesota Family Investment Program.” September. New York: Manpower Demonstration Research Corporation. Moffitt, Robert. 1983. “An Economic Model of Welfare Stigma.” American Economic Review 73, no. 5 (December): 1023-35. Pavetti, LaDonna A. 1995. “Who Is Affected by Time Limits?” In Isabel Sawhill, ed., Welfare Reform: An Analysis of the Issues. Washington, D.C.: Urban Institute. Robins, Philip K., Charles Michalopoulos, and Elsie Pan. 2000. “Financial Incentives and Welfare Reform.” August. New York: Manpower Demonstration Research Corporation. References (Continued) Scrivener, Susan, Gayle Hamilton, Mary Farrell, Stephen Freedman, Daniel Friedlander, Marisa Mitchell, Jodi Nudelman, and Christine Schwartz. 1998. “Implementation, Participation Patterns, Costs, and Two-Year Impacts of the Portland, Oregon, Welfare-to-Work Program.” May. New York: Manpower Demonstration Research Corporation. U.S. General Accounting Office. 1998. “Welfare Reform: States Are Restructuring Programs to Reduce Welfare Dependence.” GAO/ HEHS-98-109, June. Washington, D.C.: U.S. Government Printing Office. U.S. Department of Health and Human Services. 2000. “Temporary Assistance for Needy Families (TANF) Program.” Third Annual Report to Congress, August. Washington, D.C.: U.S. Government Printing Office. The views expressed in this article are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever. FRBNY Economic Policy Review / September 2001 123 Christopher Jencks Commentary P hilip Robins and Charles Michalopoulos make a strong case for two propositions. First, if we offer earnings supplements to single mothers for full-time but not part-time work, more mothers will work full-time. Second, because such an earnings supplement induces some mothers to shift from part-time to full-time work, it can raise single mothers’ overall income without costing the taxpayer more than the existing system. These propositions are qualitatively consistent with the results of Canada’s Self-Sufficiency Project. The exact magnitude of such a program’s effects in the United States inevitably involves some guesswork, but I want to focus not on whether the authors have given us the right numbers but on whether their goal of getting more single mothers to work fulltime rather than part-time is a sensible objective for government policy. The principal goal of the federal welfare legislation enacted in 1996 was to move single mothers into the labor force. This goal had overwhelming public support. Indeed, it has been an avowed goal of federal welfare policy since 1967. But while most Americans clearly want single mothers to work, it is not clear how much work the public thinks single mothers should do. In order to become completely self-sufficient, a single mother with two children who needs paid child care would have to earn at least $20,000 a year. For a mother earning, say, $6.50 an hour, that would mean working a bit over 3,000 hours a year, or sixty hours a week. If asked, most Americans would probably think that sixty hours of work per week is too much, especially for a single parent with young children. But if sixty hours is “too much,” how much is “enough”? Should we aim for forty hours a week, thirty hours a week, or what? Most Americans’ ideas about how much mothers should work depend partly on how old the mother’s youngest child is. When Aid to Families with Dependent Children was established in 1935, the law assumed that mothers would stay home until their children reached the age of eighteen. This norm was outmoded by 1970. Most married mothers now work outside their home once their children are in school, so the argument that unmarried mothers with school-age children should be exempt from work requirements commands little political support. But most Americans still think that mothers of newborns should stay home for a while. But it is unclear as to for how long. The norms that state legislatures set for unskilled single mothers will inevitably reflect the behavior of other members of society. If most unskilled married mothers work full-time, legislators will expect unskilled single mothers to do the same. Likewise, if most skilled single mothers work full-time, legislators are quite likely to think that unskilled mothers should do the same. Chart 1 shows the rates of full- and part-time employment for both single and married mothers with children under age three. Even among mothers whose youngest child is less than a year old, at least half work. But only a third work full-time. Christopher Jencks is the Malcolm Wiener Professor of Social Policy at Harvard University’s John F. Kennedy School of Government. The views expressed are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. FRBNY Economic Policy Review / September 2001 125 Chart 1 Percentage of Married and Unmarried Mothers with Children under Three Who Worked in 1998-99 By Child’s Age Percent 80 70 Child under 1 Child 1-2 Child 2-3 60 50 40 30 20 10 0 Married: Full-time Unmarried: Full-time Married: Unmarried: Full- or part-time Full- or part-time Source: U.S. Department of Labor, Bureau of Labor Statistics, “Employment Characteristics of Families in 1999.” News release, June 15, 2000. Among unmarried mothers, the percentage who work fulltime rises sharply as children get older, presumably because it is easier to find child care for a one- or two-year-old than for a newborn. Chart 2 shows variation by educational level in the rate of full-time employment among mothers with preschool children. Among those who did not finish high school, only 19 percent work full-time when their children are under age three. In part, this reflects the economics of work. Highschool dropouts earn very little, so if they have to pay for child care, they barely cover their costs. For mothers with graduate degrees, the economic incentives look quite different. But these differences in rates of employment have a nonmonetary dimension as well. Taken as a group, the better educated mothers are likely to be much better at managing their time, juggling the competing demands of work and motherhood with fewer adverse effects on both themselves and their children. Thus, even if their wages were the same, we would expect to find better educated women working more. Consistent with this view, the effect of a mother’s education on her chances of working full-time falls as her children get older and require less intensive care. In a society where only half of the best educated mothers manage to combine caring for a young child with holding a full-time job, it seems unrealistic to expect that most unskilled mothers will have the energy, competence, or financial resources to do this. Thus, as time limits begin to bite, I think we will need to think more carefully about how much work we should expect of unskilled single mothers. We should, I think, expect them to do some work. But if we want them to work full-time, I think we will need to establish a much denser network of child-care centers and subsidies, so that single mothers are not forced to devote huge amounts of energy to organizing, reorganizing, and financing child care. At the moment, there is no political support for creating such a system in America. Under the Temporary Assistance for Needy Families program, federal time limits are based entirely on one’s lifetime use of public assistance. Some states apparently take the child’s age into account, at least informally, when enforcing time limits, but we are still a long way from a consensus on what rules the new system should follow. I fear, however, that if states really expect all single mothers with young children to hold down a full-time job, they will be building a system in which many unskilled mothers fall through the cracks. If I am right, building a support system for unskilled mothers who work part-time while their children are young should be an even higher priority than improving the support system for single mothers who work full-time, as Robins and Michalopoulos propose. Chart 2 Percentage of Mothers Employed Full-Time By Education, Marital Status, and Child’s Age Percent 80 Years of school 70 60 50 Less 12 13-15 16 More than than 12 16 40 30 20 10 0 Child under Child 3-5 years: Child under Child 3-5 years: 3 years: All mothers 3 years: Married mothers All mothers Married mothers Source: Current Population Survey (March 2000). The views expressed in this article are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever. 126 Commentary Thomas MaCurdy Commentary I n their paper, Philip Robins and Charles Michalopoulos project the impacts of an earnings-supplement program modeled after Canada’s Self-Sufficiency Project (SSP).1 The distinguishing characteristic of this program involves a benefit structure designed to encourage welfare recipients to work fulltime. This commentary addresses the following questions: • What are the key features of the proposed program? • How does the authors’ study evaluate the program? • How reliable is this evaluation? Description of the Proposed Earnings-Supplement Program The critical element of SSP is that it pays benefits based on the number of hours per week an individual works. The benefits schedule offers nothing until a person reaches thirty hours per week, then it pays a large amount exactly at thirty hours. This amount is inversely related to a person’s wage rate. As a person increases hours of work beyond thirty hours, benefits are reduced much in the same way as they are in other welfare programs. Although an hours-limitation feature is an uncommon feature of U.S. welfare programs, many programs in Europe have such elements. In Great Britain, for example, the Family Credit program gives a bonus to families when they Thomas MaCurdy is a professor of economics at Stanford University and a senior fellow at the Hoover Institution. reach sixteen hours per week, and another bonus for thirty hours. To gauge the size of SSP payments for the most generous version of the program considered by Robins and Michalopoulos, Chart 1 compares the benefits paid by the federal earned income tax credit (EITC) with those paid by the SSP program with a target annual earnings level of $30,000. When an individual works more than thirty hours per week, he is eligible to receive ½∗($30,000 - actual earnings) in addition to his actual earnings. The chart depicts the supplement for three cases, differentiated by the wage of the worker: a worker who earns $6 per hour, a second who earns $9, and a third who receives $12. The $6-per-hour worker begins to receive the supplement when his monthly earnings are $780, the amount he would earn if he worked exactly thirty hours per week. The initial supplement of $860 for this hypothetical worker more than doubles his earnings. For each additional dollar he earns, his supplement is reduced by fifty cents. The chart shows that individuals receiving hourly wages of $9 and $12 become eligible for the SSP benefit at higher levels of monthly earnings, $1,170 and $1,560, respectively. It also highlights the different incentives attached to the EITC, which does not base benefits on hours worked, as compared with the SSP. Under the EITC, each additional hour of work is rewarded with a positive transfer up until monthly earnings reach $780. In contrast, when total weekly hours are less than thirty, an additional hour of work does not result in any supplement according to the SSP program. The views expressed are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. FRBNY Economic Policy Review / September 2001 127 Chart 1 Monthly Benefits Paid by Federal EITC and SSP Programs for Different Earnings and Hourly Wages Dollars Earned income tax credit/transfer 1000 E* = $30,000 and wage = $6 800 E* = $30,000 and wage = $9 600 E* = $30,000 and wage = $12 400 Federal EITC 200 0 783.3 0 1175.0 1566.7 2666.7 Monthly earnings To illustrate further the implications of the SSP program on work incentives, Chart 2 shows changes in monthly disposable income as a family’s earnings increase in conjunction with benefits paid from the federal EITC, the Temporary Assistance for Needy Families (TANF) program, and the SSP program. The chart shows how disposable income changes for an additional $100 in monthly earnings under two scenarios: one in which a family receives TANF and EITC benefits (termed a TANF family) and a second in which the family also qualifies for the SSP program (designated an SSP family). In addition to an adjustment for benefit changes, all computations account for the payment of Social Security and income taxes. The changes displayed in the chart are based on the TANF benefit schedule for California, which pays benefits that are about 15 percent higher than the most generous level paid by the states considered by Robins and Michalopoulos. According to the chart, disposable income rises approximately $110 for the first $100 earned by either a TANF or an SSP family. The three large spikes in the chart depict the large increase in disposable income occurring when the SSP program first goes into effect at thirty work hours per week for eligible recipients. The locations of these spikes reflect the different monthly earning thresholds at which workers with different hourly wages initially receive the SSP benefit. To highlight the impact of the SSP program on work incentives, Chart 3 duplicates Chart 2, except that 1) the vertical scale is reduced to magnify changes in disposable income ranging from -$25 to $125 and 2) the chart isolates changes only for a family that makes $9 per hour. The chart shows that changes in disposable income across monthly earnings are identical for TANF and SSP families when monthly earnings range from $0 to $1,100, the point at which the SSP benefit is first given. When earnings increase from $1,100 to $1,200, the TANF family sees a decrease in disposable income of $2.80 while the SSP family witnesses a one-time increase of approximately $390. However, the increase in disposable income for SSP families for the next $100 in additional earnings—from $1,200 to $1,300—is only $6. When earnings increase from $1,500 to $1,600, the Chart 2 Chart 3 Change in Monthly Disposable Income from Additional $100 in Earnings for TANF and SSP Families for Different Hourly Wages Change in Monthly Disposable Income from Additional $100 in Earnings for TANF and SSP Families Dollars Dollars Change in disposable income 500 Change in disposable income 125 E* = $30,000 and wage = $9 400 E* = $30,000 and wage = $12 E* = $30,000 and wage = $6 300 200 E* = $30,000 and wage = $9 75 EITC and TANF 50 EITC and and TANF TANF EITC 100 0 25 0 -100 -25 0 500 1000 1500 2000 2500 Monthly earnings 128 100 0 500 1000 1500 Monthly earnings Commentary 2000 2500 relationship between the two programs is reversed; this increase in earnings causes about a $30 increase in disposable income for TANF families and only a $6 increase for SSP families. Changes in disposable income occurring to families under TANF remain higher than those under SSP until monthly earnings reach $2,600, when these programs are no longer relevant. Thus, the work incentive created by the SSP program varies according to the level of weekly hours worked. For incremental changes at low hours, TANF and SSP families face the same work incentives, since the SSP program pays nothing. The SSP program offers a large incentive for a family to increase hours up to thirty. After thirty hours, work incentives are generally worse for the SSP family. So, if the objective is to induce workers to raise hours from thirty to forty, a typical definition of full-time work, SSP works against achieving this goal. Approach for Predicting Impacts of the Earnings-Supplement Program To predict the impact of introducing an SSP program, Robins and Michalopoulos posit a specific utility optimization framework and presume that families in the target population select hours of work according to this model. In essence, they implement a “simulation” approach to solve this optimization problem and to forecast responses. Specifications Selected by the RobinsMichalopoulos Study The functional forms for preferences assumed by Robins and Michalopoulos in conducting their analysis come from Moffitt (1983), a well-known empirical study analyzing the effects of Aid to Families with Dependent Children (AFDC) benefits on welfare participation and hours of work. The maintained labor supply equation takes the form: 1) H = α + β W( ± γ t ) + δ ( N + γ G ) for a person who receives welfare, and H = α + β W + δ N 2) for nonwelfare participants. An individual participates in welfare when 3) P∗ > β ± δH P∗ = γδ Wt + ln ----------------- ± ϕ . β ± δ H The variable H represents weekly hours of work, W is the gross hourly wage rate, N is nonwage income excluding welfare, t reflects the benefit reduction rate associated with the welfare program, and G is welfare benefits at zero hours of work. A person does not work when H < and P∗ > or when H < and P∗ ≤ . The coefficient α is an intercept dependent on a linear function of age, race, education, unemployment, and family composition; β measures the substitution effect; δ determines the income response; and the parameters γ and ϕ allow for the presence of welfare stigma. Robins and Michalopoulos assume coefficient values and distributions broadly consistent with Moffitt’s econometric model. For coefficients, they select values estimated by Moffitt. Also, as in Moffitt, the authors interpret the intercept of the labor supply function, α , and welfare sigma, ϕ , as being randomly distributed across families; α and ϕ equal normally distributed error terms added to fitted values estimated by Moffitt. Robins and Michalopoulos estimate a predicted value for wages using a simple linear regression equation, and they add a normally distributed error to this value to assign wages for nonworkers. Overview of the Robins-Michalopoulos Simulation Procedure To achieve their underlying goal of imputing distributions for wages, ϕ , and α , Robins and Michalopoulos simulate hours worked and welfare participation for each low-income family in their sample, given the AFDC benefit schedules actually faced by the family. To construct these distributions, Robins and Michalopoulos first draw one random variable, an error determining the value of ϕ , for each member of the sample who works. Using data on an averaged hourly wage, they then evaluate hours from equations 1 and 2 depending on whether the family participates in welfare. This hours calculation implicitly determines a value for the coefficient α (the intercept of the labor supply equation). Using the value drawn for ϕ and the constructed values of wages and hours, equation 3 determines whether the family receives welfare. If the resulting outcomes agree with the observed hours of work and welfare status of the sample member—and are consistent with highest utility—these imputed values of wage, ϕ , and α are assigned to this family. If, however, the outcomes disagree with the observed data, then the procedure is repeated with new random draws for ϕ until agreement is achieved. The process then moves on to assignments for the next sample member. For sample members who do not work, Robins and Michalopoulos modify the above procedure by drawing two additional random components: an error for wages and a FRBNY Economic Policy Review / September 2001 129 disturbance determining the value of α . Computing wages as the sum of a fitted value and the drawn error, they calculate hours and welfare participation from equations 1-3 using this value of the wage and the drawn realizations of ϕ and α . If the simulation reveals outcomes inconsistent with observed behavior, the process is repeated with new random draws. Once agreement with observed data is found, the constructed values of wage, ϕ , and α are assigned to the sample member. At the end of this procedure, each family has been assigned values for the random variables of the model, which are then fixed for conducting counterfactual exercises. To forecast impacts of the SSP program, Robins and Michalopoulos alter benefit schedules to reflect the addition of the SSP benefits and then calculate changes in hours of work implied by their behavioral model. They then use these new hours to calculate changes in disposable income and program costs. Concerns about the Reliability of Predicted Impacts Carrying out a counterfactual analysis of the type performed by Robins and Michalopoulos always involves making compromises subject to criticism. This discussion briefly notes three categories of potential shortcomings: 1) problematic features of the underlying economic/empirical model, 2) incompatibilities between the simulation model and its estimated variant, and 3) modifications needed to conduct the simulation exercise. random survey week cannot be used to infer a family’s annual, quarterly, or monthly hours. Furthermore, one cannot assess the degree to which benefit programs encourage more hours worked per week at the expense of fewer weeks worked. Robins and Michalopoulos focus on hours per week because this is the target variable of the SSP program, even though they consider payments from this program in an annual context analogous to the way in which the EITC program operates. The reasoning underlying the authors’ linkage between hours per week and hours per year is questionable. Predictions rely critically on the applicability of the labor supply function maintained in the simulation exercise, and the static linear specification assumed here is difficult to justify. The most fundamental shortcoming is that the labor supply function must apply globally for all ranges of wages and income observed in the data. This is a challenge not attained by most empirical specifications with only one source of randomness in tastes. Moreover, this same specification determines labor force participation, meaning that its parameters govern whether or not a person works. Unfortunately, such specifications have been found to be grossly incompatible with the data whenever tested. The static character of the assumed specification also gives rise to some concern, for it presumes that individuals ignore impacts of current work experience on future choices and opportunities, thus ruling out trade-offs between hours across periods. Moreover, such static specifications ignore responses motivated to avoid sanctions and time limits, which have become critical elements of states’ welfare systems. Selected Calibration of the Simulation Model Conceptual Features of the Economic Model A major shortcoming concerns the presumption by Robins and Michalopoulos that all adjustments in annual hours of work come in the form of changes in weekly hours instead of shifts in the number of weeks worked per year. Even a casual examination of data on annual hours worked reveals that exactly the opposite is true; most adjustments occur in changes in the number of weeks worked per year. Moreover, the estimated empirical model used by the authors to calibrate their model merely measures the impact of changes in wages and income on a person’s weekly hours, but it offers minimal capacity to assess effects on number of weeks worked over any extended period. Unless participation is determined independently across weeks for the given family, or is perfectly correlated, knowledge of the probabilities of working in a 130 Commentary In addition to questions about the applicability of the behavioral underpinnings of the Robins-Michalopoulos simulation model, the authors’ selection of parameter values and distributional assumptions raises concerns about the accuracy of hours projections. To be accurate, the model must be calibrated using values associated with the circumstances relevant for the simulation. There are two shortcomings in this regard. First, the treatment of missing wages in this analysis creates problems with predictive accuracy. Robins and Michalopoulos simply impute wages using fitted values from conventional regression estimation, ignoring potential sample selection that will alter predictions for particular disadvantaged groups. In contrast, Moffitt (1983) accounts for sample selection in his estimation of missing wages. This adjustment leads to systematic and significant differences between the expected value and other moments of the wage distribution used in the simulated and the estimated versions of the model. Second, whereas Robins and Michalopoulos use their model to predict the behavioral responses of a highly dependent population of welfare recipients, their choice of coefficient values and evaluation points for parameters specifying distributions comes from Moffitt, who estimates his model on a nationally representative sample of single-female households. Only 10 percent of the Robins-Michalopoulos sample worked at the baseline, and 80 percent received AFDC and food stamps for at least eleven months in the previous year. These numbers far exceed those for the representative population of femaleheaded families; Moffitt reports that only 35 percent of his sample received welfare benefits. The resulting parameters presented in Moffitt are unlikely to be applicable to the RobinsMichalopoulos population. Consequently, even if all the functional forms of distributions correctly describe outcomes for welfare populations, the values at which the authors evaluate parameters yield distributions that do not fit their data in the baseline simulation. Conceptual Problem with the Simulation Approach When random variables enter specifications nonlinearly, simulation methods dictate that many draws must be assigned to each sample observation to calculate distributions. For example, if a researcher wishes to infer the distribution of the quantity gi ( x ) , and x follows a density f ( x ) , then constructing a histogram of the values gi ( xk ) , k , …, N , with each xk representing an independent draw from f ( x ) , computes this distribution. A single draw and the resulting value gi ( xk ) do not estimate this distribution. Instead, one requires a sufficient number of simulated values to obtain consistent estimates of the statistic of interest. As interpreted above, Robins and Michalopoulos conduct their simulation with only a single draw assigned to each sample member. This might be appropriate to compute statistics if the gi ’s are linear in their random components, or if many welfare families are observationally equivalent possessing identical gi ’s—identical abstracting from the value of xk . However, neither of these conditions holds in the authors’ analysis. Nonlinear budget constraints and nonconvexities alone rule out linearity of the gi analogues. The existence of families residing in different states and with differing economic endowments implies that gi ’s vary across observations. Thus, a proper analysis should include many assigned simulated draws for each sample member. Conclusion The paper by Robins and Michalopoulos is an enlightening contribution to a topic that is central to the debate on welfare reform. The above commentary suggests that researchers should consider three modifications in future applications. First, to evaluate features of earnings-supplement programs aimed at influencing weekly hours, the underlying empirical/ economic model should not only incorporate hours per week, but also weeks worked per year or some other relevant period allowing for trade-offs between weekly hours and weeks worked. Moreover, to describe the behavior of any population with a substantial segment that does not work, the model must allow for factors to impact interior solutions for hours of work different from labor force participation decisions. Second, one needs to calibrate the model to fit the sample used in simulation. This requires adjusting coefficient values and/or distributions to account for how a simulation sample differs from the data used in estimating parameters of the model. Estimating parameters of the model using the simulation sample offers one method for accomplishing this task, but less onerous options are available. Lastly, the simulation implementation must assign enough random draws for each sample observation to compute distributions and statistics reliably. FRBNY Economic Policy Review / September 2001 131 Endnotes 1. Portions of this commentary pertain to an unpublished technical appendix to the Robins-Michalopoulos paper. References Moffitt, Robert. 1983. “An Economic Model of Welfare Stigma.” American Economic Review 73, no. 5 (December). The views expressed in this article are those of the author and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever. 132 Commentary