Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
7 Effects of Welfare Reform N ew federal public polices, which began to be enacted in the 1980s, have had a particularly significant effect on work and family trends among low-income families in the United States. The 1996 Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA) replaced Aid to Families with Dependent Children (AFDC), the major program of cash assistance to families with children, with Tem- porary Assistance for Needy Families (TANF), a work-oriented time-lim- ited program designed to encourage single mothers to become self-sufficient (see Box 7-1). This legislation ended what had become known as "wel- fare." Prior to the passage of welfare reform, efforts had been made to "make work pay" by increasing the federal minimum wage and expanding the earned income tax credit (EITC), which subsidizes the earnings of low- income families. Together, welfare reform and expansion of the EITC built on and furthered ongoing policy developments with origins in the federal 1988 Family Support Act, as well as 1980s and 1990s state welfare reform "waiver" programs that sought to increase assistance to low-income work- ing families and to require, prepare for, and support labor market partici- pation among welfare recipients and would-be recipients (e.g., Ellwood, 1988). The 1996 act went beyond earlier reform efforts by turning responsibil- ity for program design to the states, ending the entitlement to cash assis- tance to poor single mothers, and time-limiting eligibility for federal cash assistance. Proponents of welfare reform sought to increase employment among low-income single mothers with preschool-age children (who made 199
200 WORKING FAMILIES AND GROWING KIDS BOX 7-1 Key Provisions of Welfare Reform in Temporary Assistance for Needy Families (TANF) 1. Replaces welfare entitlement with block grants to states. 2. Establishes a 60-month limit on federal assistance for families with adults, with shorter time limits permitted at state discretion. However, up to 20 percent of a state's TANF caseload may be exempt from the time limit for hardship. 3. Requires states to ensure that recipients engage in work and work-related ac- tivities. For example, half of nonexempt single mothers must be engaged in work or work-related activities by 2002. Federal law requires 20 hours of work activity per week for single mothers with children under age 6 and 30 hours for those with only older children. However, states can offset the work require- ment with a "caseload reduction credit" that reduces the percentage of the caseload required to be engaged in work activities by 1 percentage point for each percentage point reduction in the caseload since fiscal year 1994 for rea- sons other than changes in eligibility rules. States may exempt mothers of children under 1 year of age from work requirements. 4. Consolidates streams of federal assistance to states for child care and increas- es funding. 5. Makes legal immigrants ineligible for TANF for their first five years in the coun- try, subject to limited exceptions. 6. Gives significant control over program design to the states. up approximately two-thirds of AFDC cases). And, indeed, the years fol- lowing welfare reform have seen major increases in labor force participa- tion among single mothers with children. This chapter reviews the evidence on the effects of TANF on working families, first by looking at its effects on adult employment, earnings, pov- erty, fertility, and marriage and then by looking at its effects on the well- being of parents which, in turn, affects the development of children and adolescents in these families. EFFECTS OF TEMPORARY ASSISTANCE FOR NEEDY FAMILIES Employment, Earnings, Income, and Welfare Use Reviewers of evidence on the effects of the 1996 welfare reform ob- serve, first, that dramatic changes occurred in the 1990s in the employment
EFFECTS OF WELFARE REFORM 201 and welfare participation of adults at highest risk of welfare use and, sec- ond, that the 1996 reforms were not subject to experimental evaluation (Blank, 2002; Moffitt, in press).1 As a result, the precise causal contribu- tion of welfare reform to the behavioral changes observed in the 1990s is uncertain and likely to remain so. Estimating the causal impact of policy changes such as welfare reform must begin with a clear statement of both the treatment and the counterfactual to treatment for those affected (see National Research Coun- cil and Institute of Medicine, 2000:Chapter 4 and Appendix B). Although impossible, the ideal study would allow an analyst to observe the same individuals exposed to and not exposed to welfare reform. Experimental evaluation approximates this ideal by randomly assigning individuals either to a treatment group subject to welfare reform policies or to a control group subject to some alternative policy, such as traditional AFDC. Be- cause assignment to a policy regime is done at random, if the experiment is carried out correctly and successfully, characteristics of individuals--even characteristics difficult for analysts to measure--should be unrelated to assignment to treatment or control. In principle, the average effect of the policy on the treated group is simply the difference between the average outcomes (such as earnings or employment rate) of the treatment and con- trol groups. In the absence of random assignment, analysts use statistical proce- dures to simulate experimental evaluation. Analysts attempt to measure and control characteristics of individuals related both to the outcome of interest and to the probability of treatment. Failure to adjust for differences in such characteristics between treatment and comparison populations can lead to biased estimates of program effects (selectivity bias). Despite the widespread understanding among social scientists and policy analysts of the importance of specifying a treatment and counterfactual to the treatment, in research on the effect of welfare reform, an explicit statement of the treatment and the counterfactual may be prob- lematic (Blank, 2002). Specifying the treatment is difficult because welfare reform legislation has spawned a complex variety of welfare policy changes by states and substate entities (see the section below on state reactions to welfare reform). Indeed, a central purpose of welfare reform was to in- crease state control over program design and the ability of states to tailor programs to state-specific needs and environments. Specifying the appropriate counterfactual to welfare reform has also proven difficult because it requires the analyst to project what would have 1 This section draws on these two major recent reviews of TANF and the estimates of its impact, borrowing some organizational features from Blank (2002).
202 WORKING FAMILIES AND GROWING KIDS happened to those affected by welfare reform in the absence of the policy changes. For example, one possibility would be to compare behavior under the 1996 reforms to that under the traditional AFDC program. However, a before-after study of welfare reform will not always result in a compari- son to traditional AFDC because about half the states had been granted major federal waivers under AFDC prior to the passage of the 1996 act. In short, both the treatment and the counterfactual vary across states. Fur- thermore, as noted, welfare reform was part of a broader social policy reform package, one intention of which was to end cash assistance to able- bodied adults, at the same time increasing the take-home pay of low-in- come working families through expansion of the EITC and increasing the minimum wage. This raises the question of whether it is appropriate to evaluate the effects of TANF provisions of welfare reform in isolation from (i.e., controlling for) the increased generosity of the EITC and other nonwelfare provisions intended to support low-income working families. Because the 1996 welfare reforms were not subject to random assign- ment evaluation, it is difficult to know what the labor market outcomes and related behaviors of the population affected by welfare reform would have been in the absence of welfare reform. In the absence of experimental evaluation, analysts have attempted to draw causal inferences about the effects of the 1996 welfare reforms from three types of studies: nonexperimental (observational) statistical analysis; extrapolation of re- sults from random assignment evaluations of earlier state waiver reform efforts that share features of the 1996 reforms; and "leaver studies" (studies of recipients who leave welfare after the implementation of welfare reform). Evaluations of the effects of welfare reform generally begin by reviewing aggregate trends in welfare caseloads and employment rates of single moth- ers over the 1990s that might plausibly be attributed to welfare reform. State Reactions to Welfare Reform Although research on the effect of welfare reform has properly esti- mated effects on recipient populations, federal welfare reform is mediated by the choices that states make in how they implement their welfare poli- cies. As noted, the welfare reform "treatment" received by those on welfare or at high risk of welfare use depends importantly on the state of residence. Blank (2002) concludes that variability across states in program character- istics was generally increased by welfare reform. Some generalizations about policy choices are possible. Maximum benefit for families with no other income continued to decline in real value in the 1990s, as they had in the 1970s and 1980s. The maximum benefit for a family of three in the median state fell about 21 percent (adjusted for inflation) between 1995 and 2000 (Blank, 2000:Table 2). Many states opted to provide greater
EFFECTS OF WELFARE REFORM 203 work incentives for those on welfare through smaller benefit reduction rates than had characterized AFDC immediately prior to reform. Reducing the benefit reduction rate increases the net income of women who work at the same time that they receive welfare, but it can also reduce incentives to leave welfare and thus jeopardize eligibility for federal assistance under time limits. Although earlier research on the AFDC program concluded that there was little evidence for effects on labor supply of more generous earnings disregards,2 1996 welfare reform included strong work require- ments, so employment response to incentives might be different than under AFDC (Moffitt, in press, 1992). The 1996 welfare reform act required increasing proportions of the caseload in each state to work or to participate in work-related activities. However, the act also provided a caseload reduction credit that allows states to reduce the percentage of the caseload subject to work requirements by 1 percentage point for each percentage point reduction in its caseload from fiscal year (FY) 1995 (for declines that are not attributable to changes in eligibility rules). Many states have met caseload work requirements through caseload reduction. By FY 1999, 38 percent of recipients were engaged in work or job-related activities (U.S. Department of Health and Human Services, 2000b), and nearly 30 percent were employed. According to Blank (2002), most states have used "work first" ap- proaches for work activities, which require participants to participate in limited job preparation services and then engage in job search. Little money has been spent on longer term education and training (Strawn et al., 2001). A total of 26 states have adopted the 60-month federal limit on assistance, whereas 8 states have no mandatory time limit to end all benefits (i.e., they may elect to provide benefits paid for with state funds to individuals who have exhausted eligibility for federal assistance). And 17 states have elected to impose limits shorter than 60 months, at least for some families. Some states attempt to divert new applicants from participating in TANF either by requiring search for work prior to eligibility (10 states), or by providing short-term emergency cash payments intended to allow would- be recipients to return to or to maintain a job. States and the federal government dramatically expanded work support expenditures such as child care and wage subsidies (through earned income credits) along with welfare reform. States have also increased spending on transportation and job search assistance. 2Earnings disregards allow welfare recipients to keep a proportion of their welfare benefits as they increase their earnings, rather than reduce recipients' benefits dollar for dollar as earnings rise.
204 WORKING FAMILIES AND GROWING KIDS Aggregate Trends Welfare Caseload and Employment Two patterns are prominent in the data for the 1990s: a sharp decline in the welfare caseload and a surge in employment among single mothers (Table 7-1 and Table 7-2). As a percentage of the population of the United States, the number of welfare recipients increased by about 1 percentage point from 1990 to 1993 (to over 5 percent of the population, or an increase of more than 2 million persons), declined modestly between 1993 and 1996, and then plummeted between 1996 and 2000, falling to less than 3 percent of the population by 2000 (Moffitt, in press). Time limits were not directly responsible for this decline, as it began and took place prior to the date that recipients began to exhaust eligibility for benefits. In all, the caseload fell by more than half between 1994 and 2000, and the decline was widespread across the states. The welfare caseload decline in many states began well before 1996. But in the late 1990s, caseloads declined at a faster rate than in any equiva- lent period. It is unlikely that the decline was merely the product of the strong economic recovery of the 1990s because the 1983-1989 recovery was not associated with notable reductions in the welfare caseload (Blank, 2002). Employment of unmarried women with children increased strikingly in the 1990s. The employment rate of single mothers with children increased over 12 percentage points between 1993 and 1998, peaking in 2000 at 75.5 percent for those with children under 19. For unmarried women with children under age 6 who traditionally made up the majority of welfare recipients, the increase in labor force participation was even more dramatic: the employment rate for single mothers with children under 6 increased from 52.5 percent in 1995 to 69.1 percent in 2000. There were also large increases for those with children under age 3 and under age 1 (see Table 7- 1). There were no corresponding increases in labor force participation this period for married women with young children (in fact, their participation rates declined somewhat over the second half of the 1990s). Although one might suspect that the surge in employment and labor force participation rates among single mothers with young children in the mid-1990s is solely a product of the economic recovery of the 1990s, single mothers saw little change in labor force participation and employment over the 1980s (Blank 2002). Furthermore, among those who received public assistance in the previous year, the proportion employed increased from 20 percent in 1990 to 44 percent in 2000 (Blank, 2002). Among adult TANF/AFDC recipi- ents, the proportion working increased from about 7 percent in 1990 to about 28 percent in 1999 (administrative data cited by Ron Haskins in
205 Employment Rate 56.0 50.6 59.5 55.0 51.3 67.5 59.2 48.5 population) Force of 2001 Labor Participation Rate 60.2 54.9 57.5 53.9 68.5 58.3 (% averages) Employment Rate 56.8 51.0 59.9 55.8 51.0 69.1 59.6 50.7 (annual Force 2000 Labor Participation Rate 60.4 54.6 58.0 53.3 67.9 58.8 Child Youngest Employment Rate 54.6 49.5 61.0 57.5 52.9 55.2 46.1 39.0 of Age Force by 1996 Labor Participation Rate 59.4 54.3 60.3 55.9 56.8 49.4 Mothers 2002). of Employment Rate 53.9 50.0 60.9 57.2 53.6 52.5 44.4 38.9 (1997, Status Force Statistics 1995 Labor Participation Rate 58.9 55.0 60.4 57.0 54.3 49.0 Labor of Employment 6 3 1 < < < <6 <3 <1 <6 <3 <1 Bureau 7-1 child child child child child child married child child child U.S. married not mothers spouse spouse Youngest Youngest Youngest Youngest Youngest Youngest Youngest Youngest Youngest TABLE Total Mothers with Mothers with SOURCE:
206 WORKING FAMILIES AND GROWING KIDS TABLE 7-2 Poverty Rates for Persons in Families Headed by a Female Householder, No Husband Present All Black Hispanic 2001 (new weightsa) 28.6 37.4 37.8 2000 (new weights) 28.5 38.6 37.8 2000 (old weights) 27.9 38.7 36.5 1999 30.4 41.0 40.7 1998 33.1 42.8 46.7 1997 35.1 42.8 50.9 1996 35.8 46.4 53.5 1995 36.5 48.2 52.8 1994 38.6 50.2 54.8 1993 38.7 53.0 53.2 1992 39.0 54.0 51.5 1991 39.7 54.8 52.7 1990 37.2 50.6 53.0 1989 35.9 49.4 50.6 1979 34.9 53.1 51.2 1969 38.2 58.2 53.5b 1959 49.4 70.6 NA aBeginning with data published for 2000, the Current Population Survey used new population weights. For 2000, figures were published using both old and new weights to indicate the effects of the revised weights. b1969 entry for Hispanics is data for 1972, the earliest available from the Current Population Survey. SOURCE: U.S. Census Bureau (2000). Blank and Haskins, 2001) (with some proportion of this increase attribut- able to changes in eligibility rules that affected eligibility for working fami- lies). Earnings, Poverty, and Income Since employment increased substantially at a time when welfare caseloads were falling, analysts have attempted to determine whether em- ployment improves the economic status of recipients and former recipients. Whether employment increases income depends importantly on wage rates and hours of work. More detailed analyses of wage rates of welfare leavers are discussed below. However, simple calculations indicate that half-time work at the minimum wage (plus the EITC) provides income that meets or
EFFECTS OF WELFARE REFORM 207 exceeds the maximum benefit in the median state (e.g., Ellwood, 1999; Jencks, 1997; Jencks and Swingle, 2000). In states with high benefit levels or generous earnings disregards, more than half-time work would be re- quired to improve income relative to welfare. These calculations also de- pend importantly on whether former recipients have major unreimbursed work expenses, particularly for child care. Combining employment and welfare unambiguously raises income for recipients in states that have adopted generous earnings disregards (see the section on state reactions to welfare reform), but it may not increase income in the absence of earnings disregards. Finally, although wages are low among the low-skilled women who make up the vast majority of welfare recipients, they do appear to increase with labor market experience (Gladden and Taber, 2000). Poverty rates among single mothers fell sharply in the 1990s, especially after 1995. Poverty rates fell 15 percentage points among black and His- panic single mothers from their early 1990s peaks to 2000. Perhaps more strikingly, more than half of this decline took place after 1996 (see Table 7- 2). Census tabulations of poverty rates for 2001 are not strictly compa- rable to figures for the late 1990s due to the introduction of new population weights based on the 2000 census. However, it is possible to use the newly weighted data to compute changes between 2000 and 2001. Between 2000 and 2001, the poverty rate for all persons increased 0.4 percentage points, but for persons in families headed by single mothers, the increase was only 0.1 percentage points (on a much larger base). More strikingly, despite the slowing economy, the poverty rate for individuals in families headed by a black single mother actually declined by 0.8 percentage points, and the corresponding figure for Hispanic single mothers was unchanged. (Al- though poverty reduction is a key goal and therefore an important outcome for assessment of welfare reform, it must be noted that, while they are receiving welfare, virtually all welfare recipients are very poor, so it is possible for their income to improve substantially and still be in poverty.) Poverty rates fell as income increased for single mother-headed fami- lies. Between 1993 and 1999, income increased across the entire distribu- tion (Blank, 2002:Table 5, citing data from the Center on Budget and Policy Priorities), whereas between 1997 and 1999, income increased for all but single mother families in the lowest quintile. Specifically, single moth- ers in the lowest quintile saw their average real income decline by about 5 percent between 1997 and 1999, after having risen by over 10 percent in the first half of the 1990s. Consumption expenditures increased over the 1990s, even among families with very low income headed by women (Blank, 2002, citing Meyer and Sullivan, 2001). Limitations of measures of income should be noted. Very few studies of changes in economic status use good measures of disposable income (Blank, 2002). Limitations include failure to measure accurately tax rates
208 WORKING FAMILIES AND GROWING KIDS and EITC benefits (which are ignored or imputed by analysts); poor or absent measurement of the value of in-kind public assistance (housing, food, medical insurance); lack of information on income sharing within and between households; and failure to measure work expenses, including out- of-pocket child care expenses. Causal Analysis A variety of approaches have been taken to estimate causal effects of welfare reform. Such attempts encounter several difficulties. Welfare reform was enacted in the mid-1990s and followed by a period of unusually strong economic growth. Other programs intended to aid low-income working families were expanded, and it is difficult to know whether such programs should be considered a component of or separate from welfare reform. Specifically, the federal EITC provided a 40 percent subsidy to earned income to low-income families with two or more children. The maximum benefit, at about $4,000 per year, is comparable to the maxi- mum AFDC benefit in many states. Because many related important fac- tors changed contemporaneously with nationwide implementation of fed- eral reform, isolating the causal contribution of federal welfare reform, either by cross-time comparisons within states or comparisons across states, is difficult. Some analysts have taken advantage of the cross-state variation in the nature of welfare reforms enacted and implemented by states as a result of the 1996 act to estimate effects of specific program components. Estimat- ing effects of specific welfare reform components requires an accurate and complete characterization of the components and the prereform policy en- vironment (and other state-specific characteristics), which has proven diffi- cult (Blank, 2002). Studies of the experiences of single states can provide better measures of both the welfare reform treatment and the counterfactual to welfare reform, but it is not clear that such results can be generalized to the nation. Analysts also attempt to extrapolate from studies of prefederal re- form statewide waiver evaluations that generally did conduct random assignment evaluations. However, for several reasons there are limits to generalizations to national welfare reform that can be drawn from evaluations of state waivers. First, waivers were not randomly assigned to states. For example, Blank (2002) notes evidence that waiver re- quests may have been more common among states with high unemploy- ment rates (perhaps because states attempted to use welfare reform to curb welfare expenditure growth), but federal welfare reform was imple- mented in a strong expansion. Limitations of waiver evaluations for causal inference and especially inference about effects of TANF include
EFFECTS OF WELFARE REFORM 209 (1) waivers often involved multiple components that are not all well measured, making it difficult to determine the specific component of reform responsible for estimated effects; (2) possible "contamination" of control groups due to awareness of treatment, including a general understanding that the welfare system may become time limited or require work (Blank, 2002:29); (3) waiver states may not be representa- tive (waivers were more common in Northern and Midwestern states); and (4) welfare reform may have affected behaviors related to entry, but such effects are generally not subject to experimental impact evalu- ation. A second approach to evaluating welfare reform is leaver studies, which describe the economic and social status of welfare families that exited welfare in the years following welfare reform. Although these studies do not provide direct estimates of the effects of welfare reform (because many welfare leavers would have exited welfare even in the absence of reform), they provide useful descriptive information on the well-being of recent welfare recipients. Problems with leaver studies include (Blank, 2002): low response rates, loss to follow-up, issues in the quality of administrative matching, lack of comparability across states, and lack of ability to study entry effects or other effects of welfare reform on those not on welfare. The third approach to the evaluation of TANF welfare reform uses econometric analyses of state panel data. A causal interpretation of esti- mates of effects of welfare reform or its components from such statistical analyses requires accurately specified state welfare policies and controls for characteristics relevant to the outcome that might be correlated with wel- fare reform or the specific reform measures taken. A key problem with econometric evaluations of welfare reform is insufficient variation in the timing of TANF implementation needed for identification of its effects, independent of other national policy and macroeconomic trends. As noted, it is also difficult to specify accurately state welfare policies. As a result of insufficient variation in the timing of implementation of TANF across states and widespread use of waivers prior to TANF, econometric models may be most appropriate for evaluation of the overall effect of policies benefiting low-income households in the 1990s, rather than effects of TANF specifi- cally (Blank, 2002). Research Findings Welfare Use/Caseload Decline The main concern of this literature has been to separate effects of the economy from those of welfare reform in explaining the dramatic (over 50 percent) decline in the welfare caseload in the 1990s, most dramatically in
210 WORKING FAMILIES AND GROWING KIDS the second half. This literature has relied on cross-state time-series econo- metric analysis almost exclusively. The typical study regresses monthly state caseload on state unemployment rate and a measure of welfare re- form, such as implementation of a statewide welfare reform waiver. Such estimates have attributed 26 to 35 percent of the decline prior to 1996 and 8 to 10 percent of the decline between 1996-1998 to the economy, leading to the conclusion that the economy is an important contributor to but cannot explain the majority of caseload movement in 1990s, particularly after 1996 (Blank, 2002). The impact of policy is gauged by the inclusion of an indicator variable that "switches on" in the month or year when reform is implemented. Evidence from such studies suggests that an early statewide waiver accounted for 13 to 31 percent of the caseload decline in the first half of the 1990s. Effects on welfare use appear strongest among the least educated. Few studies have looked at TANF implementation but those that do find that its effects on the caseload are larger than the effects of early waivers (Blank, 2002). Results of TANF studies tend to be sensi- tive to the inclusion of year fixed effects, which is not surprising since TANF was implemented in all states within a year of passage of the 1996 legislation. Criticisms of econometric studies of the caseload include that they focus on aggregate policy changes, not specific components of policy; that it remains very difficult to separate the effects of policy and the economy, especially in the case of TANF; and that these have been less useful in measuring additional impacts of related nonwelfare policy changes because national policies that take place in a given year (such as expansion of the EITC or increases in the federal minimum wage) are highly correlated with year effects. The impact of some related policy changes, such as those related to child support, child care, and health insurance, have rarely been accounted for in econometric studies of the caseload decline (Blank, 2002).3 Some studies have attempted to identify the specific policy components most responsible for the caseload decline. The welfare reform components most commonly investigated are time limits, family caps (which eliminate increases in benefits associated with additional births to families on wel- fare), the benefit reduction rate/earnings disregards, age-of-child work ex- emptions, timing of work requirements, and various sanctions (Blank, 2002). Although studies of components build credibility for causal infer- ence, only a limited set are measured. The most robust evidence of an effect of an individual component is time limits (Grogger, 2000, 2001). Grogger 3Exceptions are Huang et al. (2002a), who examine the effects of child support reforms on welfare caseloads, and Huang et al. (2002b), who find that stricter child support enforcement increases welfare exit rates and decreases entry rates.
EFFECTS OF WELFARE REFORM 211 (2001) reports that time limits account for about one-eighth of the decline in welfare use and about 7 percent of the large rise in employment of female household heads between 1993 and 1999, but they have little effect on earnings or income. Grogger (2001) also found a substantial effect of EITC expansions on labor supply and earnings. Labor Force Participation Implications of TANF, and the caseload decline in particular, for fam- ily work policies are most clearly related to induced changes in work behav- ior. Leaver studies find that approximately two-thirds of single mothers who have left welfare during the welfare reform period are employed at some future date (Cancian et al., 2002). The range of estimates of employ- ment among leavers is wide: 50 to 90 percent. This wide range is due to to differences across studies in the period of follow-up, in the period of em- ployment considered, and so forth. Studies typically find that about two- thirds of leavers are employed following exits from welfare; that they work about 30 hours per week on average; that their employment is fairly steady for about half of those leaving due to employment; and that their earnings, although low, grow with labor market experience at a rate comparable to the rate of other women (Cancian et al., 1999; Loeb and Corcoran, 2001; Brauner and Loprest, 1999). Wages of welfare leavers after 1996 typically average between $6.50 and $8.50 per hour (Loprest, 2001). The insights shared by a former welfare recipient illustrate the positive impact of work (Edin and Lein, 1997:140): I'm . . . happier now. . . . [When I was on welfare] I was kind of upset because I had nothing to do. . . . [J]ust thinking about the bad times . . . of all the problems. . . . [N]ow that I'm working . . . I feel good. As noted, leaver studies do not produce estimates of the effect of wel- fare reform on work behaviors. However, evidence from econometric stud- ies supports some impact of welfare reform. Studies that control for state unemployment and income changes find that welfare waivers are associated with increased employment of single mothers, but they have not made clear whether or how much TANF contributed to the very substantial increase in employment of single mothers after 1995.4 4 In contrast, Blank notes that there is "unambiguous agreement that EITC increased labor force participation among single parents" and concludes that "the lack of studies that effec-
212 WORKING FAMILIES AND GROWING KIDS Forming appropriate comparison groups for statistical analysis is prob- lematic. Effects of welfare reform and related policy changes on employ- ment have been estimated by comparing changes in employment of single mothers with and without children (Meyer and Rosenbaum, 2001; Kaushal and Kaestner, 2001), single mothers with older and younger children (Grogger, 2000), or single mothers with low and high levels of education (Schoeni and Blank, 2000; Kaushal and Kaestner, 2001). However, such characteristics as marital status, education level, and presence of young children are potentially influenced by welfare reform and therefore do not sharply delineate treatment and comparison populations. The causes of employment increase and caseload decline are disputed. Some employment increase is attributed to the EITC expansion (Meyer and Rosenbaum, 2001). Economic expansion is also important but employ- ment (and caseload) changes in the 1990s are unlikely to be due to the economic recovery alone, because caseload reduction and the employment response of single mothers are much greater in the 1990s than the earlier 1980 recovery. Reviews of this literature generally conclude that welfare reform played a role (Blank, 2002; Moffitt, 2002). Evidence from welfare-to-work experimental evaluations indicates sig- nificant positive effects of such policies on labor market participation, al- though estimated impact sizes vary (see the discussion below). Income and Poverty The impact of welfare reform on income, poverty, and other measures of well-being has been studied less frequently than effects on the welfare caseload and employment. Clearly, it is more difficult to measure dispos- able income and poverty than employment and welfare participation. Par- ticularly problematic has been the measurement of tax payments and trans- fer income, out-of-pocket work expenses, and access to income or in-kind assistance from family and household members and persons outside the household (e.g., child support or child care from a nonresident father). As noted, the aggregate data suggest increases in income and decreases in poverty for single mothers and their children over the 1990s. Leaver studies most often find that half or more of welfare leavers remain poor.5 The implication of these findings for understanding the tively include both welfare reforms and EITC changes makes it difficult to talk about the comparative impact of these two policy changes" (Blank, 2002:53). Blank also notes that there are few studies of the impact of welfare reform child care provisions on employment. 5 Blank (2002) cites the following results. Loprest (2001): 48 percent of welfare leavers are poor; Danziger et al. (2002): 50 percent are poor (annual income) two years after leaving; Moffitt and Roff (2000): 74 percent of leavers in three selected cities are poor; Cancian et al.
EFFECTS OF WELFARE REFORM 213 Jasmine, a 35-year-old divorcee and mother of two children, describes how she manages with "odd jobs" (Seecombe, 1999:146-147): I may have some extras, like my telephone, or cable bill, or something like that, but living in these days and time, children need those things. We need a phone for emergencies . . . To tell you the truth, I've been working little odd jobs on the side, like cleaning somebody's house or something like that to make ends meet. Also, there's a neighbor around the corner, and I've been taking him up to the grocery store, and I make money that way. . . . You just have to do that to survive. effects of welfare reform is unclear. On one hand, they clearly show that leaving welfare does not necessarily mean leaving poverty. On the other hand, the poverty rate of welfare recipients is near 100 percent when they receive welfare, and in many states only families with incomes far below the poverty line qualify for assistance. Therefore, it is possible that welfare reform could have improved the economic status of recipients even if it did not result in a reduction in the poverty rate among former or would-be recipients.6 Regression analysis of state panel data has also been used to study the effect of welfare reform on income and poverty. Moffitt (1999) found small effects of waivers on total income of single mothers; Grogger (2001) found that both waivers and TANF increased the income of single mothers; Schoeni and Blank (2000) found that waivers reduced poverty 2.4 points among less skilled women; and TANF reduced poverty 2 to 2.2 percentage points. Experimental Evaluations of Specific Policy Choices Experimental evaluations of specific policy components support the results of the general studies of welfare reform. Studies of welfare-to-work waiver programs of the 1980s and early 1990s (Gueron and Pauly, 1991; Friedlander and Burtless, 1995; Bloom and Michalopoulos, 2001; Michalopoulos and Schwartz, 2001) find significant increases in employ- (2002): 55 percent are poor in the year following exit, and 42 percent are poor five years later; Loprest (2001): leavers in 1997-1999 fared better than leavers in 1995-1997. 6Danizger et al. (2002) estimate that monthly net income increases $2.63 for each addi- tional hour of work effort associated with an exit from welfare to work.
214 WORKING FAMILIES AND GROWING KIDS ment and reductions in welfare usage and payments. Earnings increased approximately $200 to $600 per year. These studies found similar earnings increases for more and less advantaged single mothers. And, perhaps sur- prisingly (with the exception of those at high risk of depression), labor market effects among those with barriers to employment, such as child care problems, did not seem significantly worse, although these studies generally did not estimate the cumulative effects of multiple barriers. Expectations (e.g., Pavetti, 1997) were that a large fraction of the caseload had multiple employment barriers and would have great difficulty finding steady employment. However, a study of employment barriers (Danziger et al., 2000c; Danziger, 2001) among recipients of welfare in Michigan after a nearly 50 percent decline in the caseload found that al- though many had multiple barriers, the majority were nonetheless em- ployed half time or more at follow-up. Multiple barriers were associated with lower employment rates, but only with a very high number of barriers was employment substantially below 40 percent; only those with six or more barriers to employment (who made up 5 percent of the Michigan caseload) had very low employment rates (more than 20 percent). Earnings Supplements Other interventions tested the effects of earnings disregards and finan- cial incentives to work, some of which were combined with work require- ments. Studies of this type include those of Minnesota Family Investment Program (MFIP) (Miller et al., 2000; Gennetian and Miller, 2000), New York State's Child Assistance Program (NYCAP) (DeMarco and Mills, 2000), Milwaukee's New Hope (Bos et al., 1999), Vermont's Welfare Re- structuring Project (WRP) (Bloom et al., 1998), and Canada's Self Suffi- ciency Project (SSP) (Michalopoulos et al., 2000). The MFIP and SSP interventions tested the effects of combined disregards and services by ran- domly assigning some recipients to receive neither, one, or both work re- quirement and financial incentives (Blank, 2002, summarizes these results; see also Blank et al., 2000). For example, employment increased by 3.6 percent for MFIP recipients who were assigned to receive financial incentives only. In general, MFIP increased income and reduced poverty rates; earnings decreased slightly, but supplements resulted in increased income. MFIP did not reduce the cost or the use of public assistance, but shifted aid from nonworkers to workers. In reviewing the MFIP evaluation, Blank (2002:67) concluded that in "MFIP: employment effects depended on mandatory employment, anti-poverty effects depended on high earnings disregards." Canada's SSP, which "required" 30 hours of work in order to be eligible to participate (although participation was voluntary) and included
EFFECTS OF WELFARE REFORM 215 large earnings supplements, increased employment 7.2 percent relative to controls and reduced poverty rates by 9 percent. The Connecticut and Vermont programs also combined earnings disregards and strong work requirements. Both had strong earnings and employment gains but limited income gains. Many interventions also showed that gains in recipient income exceed $1 per $1 of government expenditure. Some have concluded that financial incentives are among the most important welfare reform components. However, more evidence on this point is needed. The evidence most supportive of this inference comes from programs implemented in Minnesota and Canada, which may not be repre- sentative of U.S. states. More generally, the program designs were not randomly assigned to states, even if within states recipients are randomly assigned to different treatments or control groups. Interactions between the treatment and state characteristics may limit generalization to other settings with different populations or different economic or policy environ- ments. As noted, welfare reform was intended to allow states to tailor programs specific to their economic, policy, and social environments as well as to their population's needs. In the absence of studies that jointly model states' policy choices and the interactions between state-specific char- acteristics and program outcomes, extrapolation of results to other settings is not justified. Time Limits Few recipients hit TANF time limits prior to 2001, so the direct effects of exhausting eligibility for federal assistance are not known. However, six states had early limits (Arizona, Connecticut, Delaware, Florida, Indiana, and Virginia). Studies of these states find no evidence for large effects of time limits on employment, although there is some evidence that recipients leave welfare faster to preserve their eligibility for assistance (Blank, 2002). Marriage and Fertility Changes to make welfare less generous reduced incentives for non- marital birth. Aggregate trends, however, do not support a strong effect of the 1996 welfare reforms on marriage or fertility. Marriage and divorce rates continued a downward trend in the 1990s. Births to unmarried women stopped increasing and may have fallen in the 1990s, but these developments appeared to begin before major welfare reform was enacted. There is evidence of a decline in the proportion of children living with a single mother and an increase in the proporation of children living with married parents in the late 1990s; this trend is particularly noticeable among non-Hispanic black children (Dupree and Primus, 2001; Primus, 2002).
216 WORKING FAMILIES AND GROWING KIDS Causal analyses of the effects of TANF on fertility and marriage have been limited. The extensive literature on the effects of AFDC on marriage and fertility suggest small effects on these behaviors (Moffitt, in press; National Research Council, 1998). However, it is unclear whether the results of studies of the AFDC program apply to TANF reforms because the latter represent much larger changes in the incentives for nonmarital child- bearing and include additional provisions targeted to reducing nonmarital fertility. Finally, experimental evaluations of welfare reform have found little effect on marriage and fertility, with one notable exception. The MFIP evaluation reported positive effects on marriage of welfare reform with financial incentives to support employment (Miller et al., 2000). Conclusion In the 1990s, welfare receipt declined and labor force participation increased markedly among single mothers. Even among single mothers with a child under 1 year of age, rates of labor force participation increased by nearly 10 percentage points in the five years following the passage of federal welfare reform (from 49 to 58 percent). Poverty rates for children fell dramatically in the late 1990s, especially in families headed by a single mother. The precise causal contribution of welfare reform to these devel- opments is uncertain and is likely to be debated for many years. However, implications of these changes for family work policy may not depend im- portantly on causes. Whatever the cause, the fact remains that many more single mothers with young children are employed (Bureau of Labor Statis- tics, 2002), and, relative to the average worker, these employees have little work experience, low levels of education, and few economic resources. EFFECTS ON THE WELL-BEING OF PARENTS, CHILDREN, AND ADOLESCENTS Processes Through Which Welfare Policies Might Affect Children and Adolescents Has the current wave of welfare reform affected child and adolescent cognitive or socioemotional development? If so, how? Many of the diffi- culties in inferring causality discussed in this report also apply to consider- ing developmental effects of welfare policy changes (Blank, in press; Blau, 2003). For example, there are no experimental data demonstrating effects of different policy approaches, after 1996, on children. Of the three kinds of studies that have been conducted to estimate causal effects of PRWORA (leaver studies, econometric studies of state policy contexts, and random
EFFECTS OF WELFARE REFORM 217 assignment studies of waiver and other programs prior to 1996), only one kind (the experiments) have collected data on child development. Thus the review here is limited primarily to those experiments, with some additional findings from nonexperimental studies that collected data on children both before and after 1996. Theories from developmental research and policy analysis have been brought to bear on the question of how welfare reform might affect chil- dren (Chase-Lansdale et al., 2001; Duncan and Brooks-Gunn, 2000; Huston, 2002; Johnson and Gais, 2001; Moore, 2001; Yoshikawa and Hsueh, 2001; Zaslow et al., 1995). These theories converge on five main mechanisms: changes in employment, family resources, family processes, family structure, and child care. As the parent behavior most directly targeted in the current legislation, increases in employment are most often cited as mediators of welfare policy effects on children. Longitudinal research on employment and develop- ment among low-income families, although largely nonexperimental, has shown small but consistent associations between increases in employment and better child school and cognitive outcomes (see Chapter 4 and Zaslow and Emig, 1997, for a review). There is other new evidence that shows no significant associations with mother's welfare and employment transitions for preschoolers or adolescents and, in fact, some evidence that mothers working may be related to improvements in adolescent mental health (Chase-Lansdale et al., 2003). However, these effects may vary by develop- mental period: recent evidence from national studies suggests that full-time maternal employment in a child's first year of life may be associated with decrements in cognitive abilities in middle childhood among non-Hispanic white families (Waldfogel et al., 2002; Brooks-Gunn et al., 2002). Few studies have examined the question of employment effects among adoles- cents, but these data suggest overall that variation in employment brought about by welfare reform approaches may have detectable, although gener- ally small, effects on children, and that such effects may depend on age and gender. Family resources, as potential mediators of the effects of welfare reform on children's development, have most frequently been measured as family income in developmental studies. Evidence from national studies, dating from the 1980s and 1990s, has shown that income poverty has negative effects on cognitive and socioemotional outcomes, particularly when expe- rienced chronically or in early childhood (Brooks-Gunn and Duncan, 1997; Duncan and Brooks-Gunn, 1997). These data suggest that approaches to welfare reform that result in decreases in family income and other resources may harm children and that approaches resulting in increases may benefit children. Family processes, such as parenting behaviors, family routines, and
218 WORKING FAMILIES AND GROWING KIDS Jervis keeps coming back to the power of character, to the dignity that real work bestows on people (Newman, 1999:254): What people do about going to work or going on welfare depends on two things: their self-esteem and what's giving them the greatest benefit. Your self-esteem will definitely make you want to work. But at the same time, the welfare's giving you the greater benefit in dealing with your bills. parent mental health, may be affected by increases in employment or changes in resources. Several hypotheses have been put forth regarding how such processes may be affected by welfare reform. Most of these have been hypothesized about welfare recipients making transitions to work. First, increases in employment brought about by TANF policies may result in higher levels of stress and family instability, due to the juggling of fre- quently shifting child care and work schedules and the pressures of low- wage employment. Second, increases in employment with regular hours may result in more regular family routines in the home. Third, engaging in employment may result in changes in self-concept, or self-esteem, or in improvements in mental health. Countervailing effects of these mecha- nisms, when considered in combination, may occur (London et al., 2001). Family structure is targeted in the language of the 1996 legislation indicating that "marriage is the foundation of a successful society." The balance of the evidence indicates that TANF did not bring about large changes in rates of marriage. Some nonexperimental evidence suggests small associations between TANF implementation and subsequent changes in rates of marriage, but in opposite directions (Bitler et al., 2002; Schoeni and Blank, 2000). These data are subject to the cautions about causality described earlier. There are no clear overall patterns in the experimental evaluations. In the Minnesota Family Investment Program, one of the state waiver programs that provided a generous earnings disregard coupled with a work mandate, mothers in the experimental group who were single at the outset of the evaluation were more likely to be married three years later than their control-group peers; two-parent families in the experimental group stayed married at a higher rate than those in the control group (Gennetian and Miller, 2000; Knox et al., 2000). However, in an experi- mental evaluation of the Iowa waiver program, which also included a relatively generous earnings disregard together with an employment man- date, a reduction in the proportion married occurred among new applicants who were single at the time of random assignment (Fraker et al., 2002; the follow-up spanned from 2.5 to 6 years). Other experimental programs
EFFECTS OF WELFARE REFORM 219 have shown some scattered effects, often in particular subgroups of the welfare caseload (Blank, in press). Another dimension of family structure that may have been affected by the 1996 legislation is the birth of additional children; the law allowed states to institute family caps, which deny additional benefits upon the birth of subsequent children. A study taking advantage of variations in the timing of implementation of family cap policies across the states, however, found no evidence of an effect of these policies on fertility rates (Kearney, 2002). Increases in employment have been associated with greater need for child care in many nonexperimental studies conducted both before and after passage of the 1996 legislation (Brady-Smith et al., 2001; Danziger et al., 2000a; Zaslow et al., 1998). The child care literature indicates, in turn, that type, stability, and quality of child care may affect cognitive and be- havioral outcomes among children in poverty, and more specifically among children of parents on welfare (Blau, 2001; Helburn, 1995; NICHD Early Child Care Research Network, 2000b, 2001c; Phillips et al., 1994; Vandell and Wolfe, 2000; Yoshikawa, 1999). It is likely, therefore, that one mecha- nism through which welfare reform may affect children is through variation in type, stability, or quality of care. Evidence of Effects Which of these hypotheses are supported in the emerging data on ef- fects of welfare policies on children and adolescents? No single study (or set of studies on the same data) has been able to test all of these rival hypotheses. Nonexperimental studies have begun to track the well-being of children and families across periods from just before 1996 through the end of the decade. In general, it appears that no strong trends have emerged, either negative or positive, in indicators of parent well-being or child devel- opment across the years just preceding and following the implementation of PRWORA (Cherlin and Fomby, 2002; Fuller and Kagan, 2002). For ex- ample, little change was found, before or after 1996, in rates of maternal depression, parent cognitive stimulation of young children, or developmen- tal delays in children in one study of single mothers with very young chil- dren on welfare in California, Connecticut, and Florida (Fuller and Kagan, 2002). In one large study of welfare recipients in three cities, transitions from welfare to work, post-1996, appeared to have had few negative effects on children of preschool or young adolescent age. The one exception was that adolescents of mothers who entered the workforce reported small but significant declines in levels of psychological distress, particularly anxiety, across a period of 16 months (Chase-Lansdale et al., 2003). Thus far it has not been possible to draw causal conclusions about effects of welfare re-
220 WORKING FAMILIES AND GROWING KIDS form from such trend data (i.e., that it was in fact the implementation of the 1996 legislation that caused these trends), due to difficulties capturing relevant policy variation, difficulties defining a consistent counterfactual condition, and lack of variation in timing of TANF implementation (Blank, in press). Experimental evidence provides preliminary support for the hypothesis that policies that raise parent income may benefit children in the primary grades. In a series of experiments conducted in the 1990s, a range of approaches to welfare policy drawn from responses to the Family Support Act of 1988 and the state waiver programs of the mid-1990s were tested, with random assignment of families to these policy approaches or to the existing state AFDC policy regimes. Although none of these experiments directly tested TANF programs post-1996, many incorporated aspects of policies that have become widespread in TANF programs, such as earnings disregards, mandated involvement in employment-related activities, or time limits. Data on adult economic outcomes, child school performance, parenting and family processes, and behavior problems were collected across follow-up periods of two to five years. The programs were divided into groups that represented three overall approaches to welfare and employ- ment policy: (1) four earnings supplement programs, which either provided cash supplements contingent on full-time work or generous earnings disre- gards; (2) six mandatory employment programs, which mandated employ- ment-related activities but did not include earnings supplements; and (3) two time-limit programs, which incorporated time limits on welfare receipt. In middle childhood (ages 6 to 12), a pattern emerged of consistent (though small) positive impacts for children in the earnings supplement programs. In these programs, experimental group members not only worked more than their control group counterparts, but also received more income overall, as a result of take-up of earnings supplements (income was measured as the combination of welfare, earnings, food stamps, any earn- ings supplements, and state and federal earned income tax credits) (Bos et al., 1999; Huston et al., 2001; Gennetian and Miller, 2000; Morris and Michalopoulos, 2000). Moreover, in contrast to two programs that called for time limits and six programs that mandated employment or education without earnings supplements, the earnings supplement programs brought about significant (though small) increases in school achievement and reduc- tions in externalizing (acting out) behavior problems among children in the early primary grades (Morris et al., 2001a). School achievement was mea- sured in these studies through a mix of parent reports, teacher reports, and standardized measures; problem behavior was measured through a mix of parent and teacher reports. It was pointed out earlier that the programs that increased both employment and income consistently were associated with improvements in these middle childhood outcomes.
EFFECTS OF WELFARE REFORM 221 A rival hypothesis is that other features of the larger policy contexts of these programs may have accounted for these effects, rather than the earn- ings supplement approach. The Minnesota program's experimental design allowed for the testing of earnings supplements with and without mandated employment as separate experimental conditions, in addition to a control group, subject to then-existing AFDC rules. The research design enabled a test of the rival hypothesis that it was mandated employment that ac- counted for positive effects on children. Data from that program show that the positive effects on children were essentially of the same size in the earnings-disregard-only condition, suggesting that it may have been that element, rather than any added effect of mandated employment activities, that brought about the positive effects. In addition, one follow-up study on the Minnesota program investigated whether it was the rise in employment or in income brought about by an earnings supplement program that was more strongly associated with the improvements in outcomes for children. This study examined the rival hypotheses in the Minnesota Family Invest- ment Program evaluation, using the two experimental conditions to con- duct an instrumental variables analysis teasing apart the influence of the two mediators. The researchers found that parent income appeared to more strongly mediate the effects of MFIP on child school performance than employment (Morris and Gennetian, 2001). Among the other hypotheses regarding mediators of welfare policy effects on children, data exist on employment, family processes, and child care. Although data on employment as a mediator of welfare policy effects is lacking (aside from the one study on MFIP just described), several ethno- graphic studies have examined experiences of welfare recipients as they made transitions to work after 1996. This qualitative evidence indicates that, for many welfare recipients, increased family stress has accompanied increased employment. Specifically, parents report worries stemming from spending less time with children, as well as worries about child care, follow- ing transitions to increased employment (Lowe and Weisner, 2002; Scott et al., 2001). This pattern extends to parents who experience rises in income following increased work; some of these parents find that as their earnings rise, they lose eligibility for federal programs and other supports for low- income parents (Lein et al., 2002; Scott et al., 2001). Some parents have also reported that they do feel better about themselves following transitions to increased work effort, citing benefits in terms of increased respect from children (London et al., 2001). The ethnographic work finds that parents typically use a blend of center-based, relative, and nonrelative home-based care. In studies of one experimental intervention that offered child care supplements (Gibson and Weisner, 2002; Lowe and Weisner, 2002), par- ents selectively took up child care supplements depending on beliefs about appropriate parenting, fears of "stranger care," child ages, ease of access to
222 WORKING FAMILIES AND GROWING KIDS child care services, and interference with existing child care arrangements with kin or partners. Also, if child care supports are tied to work and work is episodic and unpredictable, then child care arrangements change too often for some parents (three times over 18 months on average in the New Hope earnings supplement program, for example; Lowe and Weisner, in press). In general, no consistent impacts have emerged across welfare policy experiments on survey (self- or child report) measures of parenting, across such dimensions as monitoring, supervision, cognitive stimulation, control, or warmth (Bloom et al., 2000, 2002; Gennetian and Miller, 2000; Kisker et al., 1998; Morris and Michalopoulos, 2000; Quint et al., 1997). Very few of these studies assessed in-depth observational measures of parenting. One study that did was the New Chance demonstration, an intervention focused on human capital providing education, training, parenting, health and life skills classes, and child care to adolescents mothers on welfare and without a high school diploma. In a three-year post-program follow-up, researchers found small positive impacts of the program on measures of affective quality of mother-child interaction and cognitive stimulation in the home (Zaslow and Eldred, 1998). Another study examined observa- tional measures of parenting in the Teenage Parent Demonstration, a hu- man capital intervention for teenage mothers that provided relatively fewer support services than New Chance. No impacts on parenting were found (Aber et al., 1995). An analysis combining data from the 1977 to 1996 Current Population Surveys with state-level data on cases of child maltreat- ment, using a state-level, fixed-effects model, found that state welfare ben- efit levels were negatively related to rates of neglect cases (Paxson and Waldfogel, in press). More recently, Paxon and Waldfogel have analyzed the effects of welfare policies and welfare reforms on measures of child maltreatement over the period 1990 to 1998. The strongest evidence that welfare policies and welfare reforms matter comes from their results on the number of children in out-of-home care, which is negatively and signifi- cantly related to the level of welfare benefits and positively and significantly related to such welfare reforms as family caps, short lifetime limits, imme- diate work requirements, and tough sanctions for noncompliance (Paxson and Waldfogel, in press). A surprising set of emerging findings on family process effects of wel- fare policies concerns mothers' reports of domestic violence. Among the experimental evaluations of welfare policies that measured this outcome, five of nine programs have found evidence of significant decreases in re- ports of domestic violence at follow-up periods from three to five years (the Minnesota Family Investment Program and four of six programs in the National Evaluation of Welfare-to-Work Strategies [NEWWS]; the other four programs showed no change, or nonsignificant decreases; Gennetian
EFFECTS OF WELFARE REFORM 223 and Miller, 2000; Hamilton et al., 2001). It is not clear what mediating processes may explain these impacts on domestic abuse, although a set of nonexperimental analyses from the NEWWS evaluation suggests the roles of increases in employment and program caseworker attention to support services (Hamilton et al., 2001). However, another experimental waiver evaluation, in Iowa, testing a generous earnings disregard coupled with an employment mandate, was found to increase mothers' reports of domestic abuse (Fraker et al., 2002). A meta-analytic synthesis of experimental welfare policy effects on child care use, across 13 experiments, suggests that policies that increase employment also increase use of all types of out-of-home care, including center-based care and home-based care (Crosby et al., 2001; Gennetian et al., 2001). Welfare and employment policies that included elements aimed specifically at increasing the use of child care, through subsidies, vouchers, or services to help locate and obtain child care, increased the use of center- based care more than other types of care. Interestingly, this was true regardless of whether these child care assistance approaches supported cen- ter-based care over other kinds of care. These programs that provided child care assistance also, as intended, reduced out-of-pocket expenses for child care, increased use of child care subsidies, and reduced reports of child care as a barrier to employment (Gennetian et al., 2002a). A separate analysis of effects on Head Start use revealed that welfare and employment policies did not increase or decrease use of Head Start, on average (Chang et al., 2002). However, for the subgroup of adolescent and young mothers (under age 25), a parallel meta-analytic study indicated that these policies did significantly decrease Head Start use, by an average of 7 percentage points (Gassman-Pines, 2002). This may be because Head Start remains, in the majority of sites, a part-day program and thus may not meet the needs of low-income parents who work full-time, nonstandard hours, or shifting work schedules. Data have begun to emerge concerning subgroups differentially af- fected by welfare policies. Subgroups for whom data are available include those defined by developmental period (early childhood, middle childhood, adolescence), risk (particularly risk for being hard to employ and health risks), gender, and race/ethnicity. Impacts in the policy experiments fielded in the 1990s on early child- hood and adolescent outcomes differed from those found in middle child- hood. In early childhood, small samples of very young children in the experiments reduced the number of program impacts available for analysis. In addition, almost no programs were available that had sufficient numbers of infants in the first year of life. Among the available programs, no experimental effects were found, either positive or negative, among chil- dren younger than age 5 at follow-up (Morris et al., 2001a). In adolescents,
224 WORKING FAMILIES AND GROWING KIDS in contrast, a negative average effect on adolescent school performance (as rated by parents) was found across 10 experiments, as well as average increases in parent-reported grade repetition (12 experiments) and use of special education services (15 experiments). These findings were calculated using standard meta-analytic techniques across the available experiments; actual effect sizes ranged from zero to moderate and a very few large negative effects. However, no overall effects were found on rates of parent- reported dropout, suspensions, or high school completion (Gennetian et al., 2002b). The pattern of negative effects did not pertain to any particular kind of policy approach (time limit programs, earnings supplement pro- grams, or those that mandated employment without earnings supplements) but occurred across all of them. In examining a range of hypothesized mechanisms that might account for these effects (including intensity of parent employment, parenting measures, or parent resources), the only clear pattern pertained to adolescents with younger siblings. For these adolescents, larger unfavorable effects on parent-reported school perfor- mance and receipt of special education occurred, as well as some not found in the full sample: increases in dropout, suspensions, and expulsions. These adolescents were also more likely to care for younger siblings, a responsibil- ity that may have helped bring about the unfavorable pattern of impacts (Brooks et al., 2001). Ethnographic evidence from a study in four cities tracking urban welfare recipients, post-1996, has also suggested that sibling caregiving may be particularly harmful for adolescents in families affected by welfare reform (Gennetian et al., 2002b; Morris et al., 2001a). Among other subgroups, concern has emerged about the effects of welfare reform on the hard-to-employ, that is, those parents who are least work-ready and may have the greatest difficulties in making transitions to work from welfare and advancing in low-wage job markets (Danziger et al., 2000a,b; Kalil et al., 2001). Evidence from the earnings supplement pro- grams described above suggests that the positive impacts of these programs on middle childhood school performance and behavioral outcomes do not extend to children of the hardest to employ (Yoshikawa et al., in press). For the two U.S. earnings supplement programs, positive impacts on chil- dren were concentrated among the moderately hard-to-employ, with the very hardest to employ 25 percent of the samples experiencing neutral and negative effects (increases in behavior problems), despite equivalent and large increases in both employment and income. The unfavorable impacts on developmental outcomes among the children of the hardest to employ were accompanied by increases in maternal depression, decrements in the regularity of family routines, and smaller increases in center-based care, relative to the less at-risk other 75 percent of these samples. Although little research has used reliable measures of child health sta-
EFFECTS OF WELFARE REFORM 225 tus to examine how welfare policies affect children's health, some data exist on the experiences of welfare reform among families with existing health problems. Both survey and ethnographic research, post-1996, has found that families with health risks, whether of the parent or the child, appear to experience more difficulty responding to the work mandates of TANF pro- grams (London et al., in press; Romero et al., 2002). In many families (40 percent in one three-city ethnographic study), both primary caregivers and at least one of the children have been rated to be in poor health (Burton et al., 2002). Bernheimer et al. (in press) found that mothers with children with significant disabilities or other child problems (in school achievement or problem behaviors) in the New Hope program--a circumstance charac- terizing perhaps 15 percent of families--struggled with work and child care. Other studies have examined how subgroups defined by gender may be affected differently by welfare policies. The synthesis of primary grade outcomes of welfare policy experiments did not find consistent gender differences in school performance or behavior problems (Morris et al., 2001b). In the synthesis of impacts among adolescents, there was one notable difference in effects among boys and girls: across the welfare pro- grams, there was no significant average effect on grade repetition among female adolescents, but there was a significant increase in grade repetition among the male adolescents (a significant difference in average impacts; Gennetian et al., 2002b). Research on other important subgroups, such as different racial and ethnic groups, families with children with disabilities, and immigrant families, is in process, with few studies completed as of this date (Rosman et al., 2002). SUMMARY Evidence to date on how welfare policies affect children and adoles- cents is relatively strong for some of the policies that immediately predated PRWORA. Certain elements of these policies are well represented in TANF programs, post-1996 (earnings disregards and other earnings supplements; mandated employment activities without earnings supplements; time lim- its). However, the need is urgent to obtain experimental evidence on im- pacts of welfare policies in current policy contexts. Given that caveat, several messages emerge from the data. Policies that increase the incomes of low-income parents, through earnings disregards or supplements, were more likely to be associated with improvements in children's school perfor- mance and reductions in their behavior problems in middle childhood than those that simply mandated employment. However, welfare policies, whether of the mandated employment or earnings supplement types, ap-
226 WORKING FAMILIES AND GROWING KIDS peared to coexist with small but consistent decrements in parents' percep- tions of adolescents' school performance. Policies that increased employ- ment also increased use of out-of-home care, but increases in center-based care (the type of care associated most consistently with positive effects on low-income children) were brought about only by those policies that incor- porate services or subsidies to increase child care use.