Effects of Welfare Reform on Education Acquisition of Adult Women

Abstract

Education beyond traditional ages for schooling is an important source of human capital acquisition among adult women. Welfare reform, which began in the early 1990s and culminated in the passage of the Personal Responsibility and Work Opportunity Reconciliation Act in 1996, promoted work rather than education acquisition for this group. Exploiting variation in welfare reform across states and over time and using relevant comparison groups, we undertake a comprehensive study of the effects of welfare reform on adult women’s education acquisition. We first estimate effects of welfare reform on high school drop-out of teenage girls, both to improve upon past research on this issue and to explore compositional changes that may be relevant for our primary analyses of the effects of welfare reform on education acquisition among adult women. We find that welfare reform significantly reduced the probability that teens from disadvantaged families dropped out of high school, by about 15%. We then estimate the effects of welfare reform on adult women’s school enrollment and conduct numerous specification checks, investigate compositional selection and policy endogeneity, explore lagged effects, stratify by TANF work incentives and education policies, consider alternative comparison groups, and explore the mediating role of work. We find robust and convincing evidence that welfare reform significantly decreased the probability of college enrollment among adult women at risk of welfare receipt, by at least 20%. It also appears to have decreased the probability of high school enrollment among this group, on the same order of magnitude. Future research is needed to determine the extent to which this behavioral change translates to future economic outcomes.

Introduction

Major goals of the Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA) of 1996 were to reduce reliance on welfare and move recipients and potential recipients of cash assistance into the labor force. The legislation ended entitlement to welfare benefits under Aid to Families with Dependent Children (AFDC) and replaced AFDC with Temporary Assistance for Needy Families (TANF) block grants to states. Among the features of TANF and many pre-PRWORA state waiver programs,¹ which together constitute “welfare reform,” were lifetime limits on the receipt of welfare benefits, work requirements as a condition of receiving welfare, and sanctions for non-compliance with program rules. PRWORA also strengthened child support enforcement and made it easier for married and cohabiting couples to qualify for welfare benefits. These sweeping changes ushered in a new “work first” era that de-emphasized education for adult women. The PRWORA legislation granted considerable discretion to states in establishing welfare eligibility and program rules. As a result, there is substantial state policy variation within the broad national regime of time-limited cash assistance for which work is required.

In terms of reducing caseloads, welfare reform (including the pre-PRWORA waivers) has been successful; welfare rolls declined by over 50% since their peak in 1994 and at least one-third of the caseload decline can be explained by welfare reform (see Grogger and Karoly 2005). At the same time, employment rates of low-skilled mothers rose dramatically (Ziliak 2006), and at least some of that increase was a result of welfare reform (Schoeni and Blank 2000). The effects on family structure were less dramatic. A large literature on the effects of welfare reform on marriage and a smaller one on cohabitation reveal mixed findings, and the literature on non-marital childbearing and female headship indicates slightly negative but inconsistent effects of welfare reform (Blank 2002, 2007; Moffitt 1992, 1995, 1998; Grogger and Karoly 2005; Gennetian and Knox 2003; Peters et al. 2003; Ratcliffe et al. 2002).

The vast majority of mothers on welfare are adults (USDHHS 2000), and education beyond traditional ages for schooling is an important source of human capital acquisition among adult women. Almost 30% of adults in the U.S. engaged in some form of work-related education (broadly defined to include courses at the workplace or courses and degree programs in other locations) in 2004–2005, and those in their childbearing years were more likely than older adults to engage in such activities (O’Donnell 2006). In 2000, 7.7% of females enrolled in grades 9–12 were age 21 or older (U.S. Census Bureau 2000); in 2005, 80% of females taking vocational courses were age 25 years or older (U.S. Census Bureau 2005); and in 2006, 40% of all female college students were age 25 years or older (U.S. Census Bureau 2006).

Postsecondary education is not a rare event among welfare recipients. For the last school year before PRWORA (1995–1996), unpublished Department of Education reports indicate that about 650,000 welfare recipients were enrolled in post-secondary education (Price 2005). Using the 1979 National Longitudinal Survey of Youth (NLSY), a pre-PRWORA cohort, London (2006) found that almost 14% of all welfare recipients and 17% of high school graduate welfare recipients attended college during a spell of welfare receipt. According to the U.S. Census Bureau (2002), almost 23% of TANF recipients in 1998 had at least 1 year of college.

The “work first” approach under welfare reform has been targeted to adult mothers, for whom education and training generally do not count as required work activities. In contrast, minor mothers are subject to a broader human capital approach. In particular, PRWORA requires minor mothers to attend high school or training in order to receive TANF and does not impose time limits or work requirements if they are full-time students. On the surface, these differential incentives suggest that while welfare reform would have increased school enrollment among teenage girls, it may have decreased education acquisition among adult women.

Relatively few studies have investigated the effects of welfare reform on education acquisition, and all but one of those (Jacobs and Winslow 2003) focused exclusively on teenage girls. Exploiting variation in welfare reform across states and over time and using relevant comparison groups, we undertake a comprehensive study of the effects of welfare reform on formal education acquisition of adult women. We first estimate effects of welfare reform on high school drop-out of teenage girls, both to improve upon past research on this issue and to explore compositional changes that may be relevant for our primary analyses of the effects of welfare reform on the education acquisition of adult women. We then conduct comprehensive analyses of the effects of welfare reform on adult women’s school enrollment. A number of alternative models are estimated as specification checks, and the mediating role of work is explored. We find robust and convincing evidence that welfare reform significantly decreased formal school enrollment among adult women.

Background

Theoretical Framework

We base our analyses on Becker’s classic work (Becker 1975), in which schooling is an investment undertaken if the present value of the future stream of benefits exceeds the present value of the current and future costs. This model implies that individuals are more likely to invest in education when they are young, are likely to be successful in school, enjoy schooling, have a low opportunity cost of time, and have a low rate of time preference (i.e., tend not to “live for today”). Welfare reform could affect several of these factors. For adult women, work requirements would increase the time cost of attending school, which would decrease schooling. However, there are potential opposing effects that could increase schooling. Work requirements may provide access to opportunities for education or increase awareness about labor market payoffs from education, which would decrease the costs and increase the benefits of pursuing education. Lifetime limits on benefits would make welfare much less of a long-term option for potential recipients, which would increase expected returns to education.

The Evolution of Education Policy under Welfare

Traditionally, mothers on AFDC were not required to work and were allowed to attend school, including college for which tuition assistance was potentially available from Pell grants, scholarships, and student loans. The situation changed for some mothers under the Job Opportunities and Basic Skills Training (JOBS) program, which was created under the Family Support Act of 1988 and required states, to the extent resources allowed, to engage mothers with no children below age 3 in education, work, or training activities. However, many women were exempt from participation in JOBS, and between 1992 (just prior to the first statewide AFDC waiver program) and 1996 (enactment of PRWORA) only 10% of all welfare recipients in the U.S. participated in the JOBS program.²

Major statewide AFDC waiver programs, first implemented in late 1992, substantially altered the nature of welfare by imposing time limits, significantly reducing participation exemptions, imposing sanctions, increasing earnings disregards, imposing family caps, and/or implementing work requirements. Compared to JOBS programs, statewide waivers were broad-based in that they applied to large proportions of welfare recipients. While states were required to provide many specifics of their programs in their waiver plans, they were not required to report policies vis-à-vis education. Complicating the picture, states could change their policies without having to amend their waiver plans (U.S. Department of Health and Human Services 1997). The situation changed notably under TANF, which required states to file detailed program specifics (including education policies) at the outset as well as any intended changes to those policies. That is, under TANF, the extent to which education could count toward work requirements was more explicit. Because of the reporting issues under the waivers, it is difficult to compare education policies under AFDC waivers and TANF, even in a given state. However, it is clear that under both AFDC waivers and TANF, work and other requirements gave women less flexibility in deciding how to spend their time and many welfare recipients could attend college or other schooling only after fulfilling work requirements.

The “work first” paradigm may have emerged because evaluations of welfare-to-work experiments generally indicated that an education-based approach was less successful than the “work first” strategy in advancing self-sufficiency. However, some have questioned whether those results pertained to formal education, particularly college education, as programs generally provided basic skills, job-seeking skills, or vocational training rather than formal education (Shaw et al. 2006). Moreover, in a rigorous reevaluation of a major welfare-to-work program, Hotz et al. (2006) found that the short-run employment advantages had been over-estimated and that the education-based approach was more effective in the longer-term. The authors concluded that the “work first” approach appears to provide a “quick fix,” but is not necessarily the better long-term strategy for promoting self-sufficiency.

Returns to Education for Adults and Welfare Recipients

Corman (1983) pointed out decades ago that the human capital literature tended to (incorrectly) assume that individuals attend school as early as possible and that adult education is an anomaly. Since then, there have been several studies focusing on education of non-traditional aged students. Particularly relevant to our study, Boudett et al. (2000) found that almost two-thirds of female high school drop-outs receive some form of education in the decade following drop-out and that adult education has a positive effect on earnings for this group. Leigh and Gill (1997) found that the returns to community college education (both degree-granting and vocational programs) are about the same for returning adult students as for traditional-age students.

Welfare recipients are more likely to attend two-year rather than four-year colleges (London 2006). Recent work by Bound et al. (2010) finds that individuals with lower levels of college preparedness and individuals attending community colleges are far less likely to complete a college education. However, even non-completers may derive some positive returns. Kane and Rouse (1999) found that each year of credit at a community college yields, on average, a 5 to 8% increase in annual earnings—a return similar to that from 1 year of a four-year college. Kane and Rouse also found that evaluations of experimental programs supported by the Job Training Partnership Act tended to underestimate the returns to community college education, since many in the control groups also attended school. This finding is consistent with that of Hotz et al. (2006) in that it suggests that at least some experimental studies that were influential in shaping welfare policy underestimated the value of education to welfare recipients. In sum, the literature suggests that education has been and is a worthwhile human capital investment for many welfare recipients.

Effects of Welfare Reform on Education Acquisition

Hao and Cherlin (2004), Kaestner et al. (2003), Offner (2005), and Koball (2007) all estimated difference-in-difference models to assess the impact of welfare reform on high-school attendance or drop-out among female teens. Offner, using the March Current Population Surveys (CPS) from 1989 to 2001 and comparing female teens in low and higher income families, found that welfare reform resulted in increased high school attendance. The use of income to define target and comparison groups is potentially problematic, however, because welfare reform could have affected income and thereby changed the composition of the target and comparison groups over time. Kaestner and colleagues, using the NLSY to compare teens in 1979 and 1997, found that welfare reform reduced high school drop-out. Hao and Cherlin, concerned about cohort effects in the Kaestner study, compared older and younger teens (who had different exposure to the legislation) in the 1997 NLSY and found no effect of welfare reform on drop-out. Both studies used multiple sets of reasonable target and comparison groups, but the timing of the NLSY makes that dataset less than ideal for studying welfare reform, as it predates most reforms by almost 15 years (i.e., 1979) or takes place very soon after PRWORA (i.e., 1997). As a result, the comparison groups either have to be from very different cohorts as in the Kaestner study or from very similar cohorts as in the Hao study, likely “contaminating” the control group. Koball examined two cohorts closer in time, by combining a pre-welfare cohort from the National Education Longitudinal Study and a post-welfare cohort from the NLSY. However, she focused on teen mothers rather than all teenage girls. Her results were consistent with those of Offner and Kaestner et al. in that she found that welfare reform significantly reduced drop-out. Overall, the available evidence suggests that welfare reform had favorable effects on school enrollment of teenage girls.

The potential effects of welfare reform on education among adult women have been largely neglected by the research community. The lone (published) exception is a study by Jacobs and Winslow (2003) that examined the effect of welfare receipt and parenthood/marital status on college attendance at two points in time (1995 and 1999), using two different data sets—the March CPS and the National Household Education Surveys (NHES). The March CPS allowed an examination of only very young adult women (<25 years old), but included state TANF policies vis-à-vis education. The (public use) NHES allowed the authors to examine a wider age range but not to incorporate state TANF policies. In the CPS analyses, the authors found no changes in the effects of welfare receipt on college attendance between 1995 and 2000, but that holding welfare receipt constant, unmarried mothers were less likely to attend college in 2000 than in 1995. They also found that women living in states that were more supportive of college attendance for welfare recipients were more likely to attend college. In contrast, in the NHES analyses, the authors found that welfare recipients were less likely to go to college in 1999 than in 1995 but that, holding welfare receipt constant, unmarried mothers were more likely to attend college in 1999 than in 1995. These contradictory results make it unclear whether women at risk for welfare receipt were more or less likely to attend college after welfare reform.

A major problem in the Jacobs and Winslow study was the use of widely spaced cohorts, which make it likely that changes not related to welfare reform were attributed to welfare reform. For example, concurrent with the unfolding of welfare reform was a two-decade trend in state implementation of minimum standards for high school graduation. By 2005, almost half of states implemented “exit exams” for high school students. According to Dee and Jacob (2007) these exams resulted in some students dropping out of high school who would have otherwise received a diploma, while improving the academic performance of other students. A rigorous study of the effects of welfare reform on education must account for these and other relevant changes. In addition, using 1995 as the base year in the Jacobs and Winslow study was less than ideal since pre-PRWORA waivers had already been implemented in 19 states.

In sum, there is a dearth of existing work on the effects of welfare reform on adult women’s education despite the facts that: (1) the vast majority of mothers on welfare are adults, (2) education acquisition is commonplace among adult women and welfare recipients, (3) economic theory suggests that broad-based work requirements and the de-emphasis on education and training would alter the costs and benefits of pursuing education for this group, and (4) even modest increases in education raise earnings. Although past research has investigated enrollment in basic skills programs among long-term welfare recipients and effects of welfare reform on high school dropout among teen girls, only one published study has investigated the effects of welfare reform on formal education among adult women and that study yielded conflicting results. As such, this study addresses a clear and important knowledge gap.

Data

We use data from the Current Population Survey (CPS), a large-scale nationally representative monthly survey of approximately 57,000 households that is conducted by the Bureau of the Census for the Bureau of Labor Statistics. The October CPS contains information on current high school, current college (full- and part-time), and current trade school, General Education Development (GED) preparatory program, and other education enrollment. We use October CPS data from 1992 to 2001, which spans the period of welfare reform, to estimate the effects of the reforms on high school enrollment, college enrollment, full time college enrollment, and any education enrollment (high school, college, trade school, GED preparatory programs, and other education).³ For analyses of high school drop-out, high school enrollment is the relevant outcome. For analyses of adult women’s education acquisition, we focus on current high school enrollment and any school enrollment (for those with less than a high school degree) and current college enrollment and full-time college enrollment (for those who graduated high school but not college).

Our research design exploits the substantial variation across states in the timing of the enactment of pre-PRWORA waivers and TANF. We follow the convention in the prior literature with respect to the construction of the key independent variables capturing the shifts in welfare-related policies (reviewed in Blank 2002). The welfare reform measures can be classified into two phases. The first represents federal waivers granted to states to experiment with AFDC rules prior to PRWORA. We construct an indicator to reflect the fraction of the year that a given state in time period t had a statewide waiver in place that substantially altered the nature of AFDC with regard to time limits, JOBS work exemptions, JOBS sanctions, increased earnings disregards, family caps, and/or work requirements.⁴ The second welfare reform construct represents the implementation of TANF programs post-PRWORA. Similar to the construction of the indicator for an AFDC waiver, an indicator will reflect the fraction of the year that a given state had implemented TANF in time period t.⁵ For some analyses, we also construct indicators to exploit variation in state TANF plans with respect to the strength of the work incentives and the extent to which education is allowed as a stand-alone authorized work activity.

Since the welfare indicators are measured at the state level, we incorporate additional state-specific socioeconomic measures in the analyses to capture time-varying trends within states. State unemployment rate and per capita personal income are derived from figures provided by the Bureau of Labor Statistics. Welfare case-loads, defined as the total number of welfare recipients in a state, are obtained from the Department of Health and Human Services Administration for Children and Families Office of Family Assistance.⁶ All models further include indicators for whether a given state in a given year had a strict high school exit exam (testing material at or above the 9th grade level) or a less strict exam (below the 9th grade level), with the reference category being no high school exit exam. For women who completed high school, we use the existence of the exam when they are 18 years old. For those who did not complete high school, we use the contemporaneous existence of the exam in their state. These data are derived from the Appendix in Dee and Jacob (2007). Information on the maximum state-mandated age for compulsory school attendance is used to restrict the sample for the models of high school drop-out, and is obtained from the Education Commission of the States and the National Center for Education Statistics.⁷ Indicators for the maximum state-mandated age for compulsory school attendance are also included in all models to further capture any shifts in state policies relating to education.

Methods

We employ a difference-in-difference-in-differences (DDD) framework—akin to a pre-and post-comparison, with treatment and control groups, across states that have and have not implemented the policy—in conjunction with multivariate regression methods, which is standard in the economics literature on evaluating the effects of welfare reform and other state policies (e.g., Kaestner and Tarlov 2006; Bitler et al. 2005; Blank 2002). Under certain conditions, described below, this quasi-experimental research design will yield causal estimates of the effects of welfare reform on our outcomes of interest. We conduct various specification and robustness checks to assess the validity of the identification assumptions underlying this methodology.

Consider the following DDD model which relates changes in education outcomes to state and federal welfare policies for the target group relative to a comparison group:

$$\begin{array}{l}{E}_{\mathit{ist}}={\alpha }_0+\left({\alpha }_1-{\alpha }_1^{\ast }\right){\mathit{Target}}_i+\left({\pi }_1-{\pi }_1^{\ast }\right)({\mathit{AFDCWaiver}}_{st}\ast {\mathit{Target}}_i)\\ +\left({\pi }_2-{\pi }_2^{\ast }\right)({\mathit{TANF}}_{st}\ast {\mathit{Target}}_i)+{\pi }_1^{\ast }({\mathit{AFDCWaiver}}_{st})+{\pi }_2^{\ast }({\mathit{TANF}}_{st})\\ +X_{\mathit{ist}}\beta +Z_{st}\delta +{\mathit{State}}_s\lambda +{\mathit{Year}}_t\phi +({\mathit{State}}_s\ast t)\phi +{\eta }_{\mathit{ist}}\end{array}$$ (1)

Equation 1 posits that the education outcome (E), for the i^th woman residing in state s during year t, is a function of welfare policy, measured here by indicators reflecting the fraction of the year that a given state enacted a major AFDC Waiver and TANF. Target represents a dichotomous indicator for whether the individual is in the target group, which is the population at risk of welfare receipt and thus most likely to be impacted by welfare policy, as we describe below. In addition, education acquisition depends on a vector of individual characteristics (X) such as age, race, ethnicity, highest grade completed, and urban residence, a vector of time-varying state characteristics (Z) such as economic conditions and education policies, state fixed effects (State_s), year fixed effects (Year_t), and state-specific time trends (State_s* t). The parameter η represents an individual error term.

There are several benefits to estimating Eq. 1. It bypasses having to estimate the structural model relating welfare reform to welfare caseloads, which has been problematic in the literature (Kaestner and Tarlov 2006; Blank 2002).⁸ Equation 1 is also more policy relevant as it represents the reduced-form model directly linking welfare policy measures to key outcomes, and therefore accounts for any and all mechanisms through which welfare policy may be affecting education acquisition.

The direct focus on AFDC Waiver and TANF also underscores the point that the population of interest, individuals affected by welfare reform legislation, is all women at risk of being on public assistance, and not just current or former program participants (Kaestner and Tarlov 2006). Potential welfare recipients are shown to behave strategically in their use of welfare benefits when faced with time limits and other regulatory constraints (DeLeire et al. 2006; Grogger 2004). Thus, in order to identify the population effect of welfare reform on key outcomes, the appropriate sample is all women at risk of being on public assistance.

Traditionally, the welfare caseload has consisted primarily of low-educated, unmarried mothers. This at-risk population group is the target group, for whom welfare policy would be expected to have the largest behavioral effects. While interactions between the state indicators and linear time trends in general will control for systematically-varying unobserved state-specific factors, the possibility of omitted variables remains. This problem is addressed in the DDD framework by considering a comparison group—individuals who are similar in many ways to the target group but are unlikely to participate in public assistance programs, and therefore not likely to be affected by welfare reform policies. In the above equation, Target represents a dichotomous indicator equal to one if the individual is in the target group (population at risk of being on welfare) and zero if the individual is in the comparison group (population not at risk of being on welfare). The DDD estimates of the effects of welfare reform are the coefficients of the interaction terms between the policy measures (AFDC Waiver and TANF) and the Target group indicator.⁹ The impact of welfare reform, in this DDD framework, is identified using three key sources of variation. The first source is the pre- and post-policy comparison of individual outcomes within a given state. The second source results from the substantial variation in the timing and incidence of welfare reform across different states. Thus, states that had not yet implemented the policy provide a “control” group, allowing a comparison of outcomes pre- and post-policy between states that implemented the policy and those that had not yet done so. The third source of variation exploits target and comparison groups. This third “D” accounts for the possibility that unobserved time-varying state trends in education outcomes may be correlated with welfare reform.

The assumption necessary for the DDD effect to represent an unbiased estimate is that in the absence of welfare reform, unobserved state-varying factors would affect the target and comparison groups similarly. We implement several checks to assess the validity of the comparison groups. Ideally, the target and comparison groups would look similar prior to the policy shift, after controlling for all observed individual and state-level characteristics. The adjusted (conditional) difference in the level of education outcomes for the two groups prior to welfare reform is given by the coefficient (α₁–α₁*) of the Target indicator in Eq. 1. Checking the magnitude and statistical significance of this coefficient serves as a test of the similarity between the target and comparison groups prior to the policy shift. By definition, the comparison group should also not be affected by the welfare policies. Thus, the coefficients of AFDC Waiver and TANF (π*₁ and π*₂), which reflect the impact of welfare reform on individuals who are at low-risk of being on welfare, should also be relatively small in magnitude and insignificant.

The choice of target and comparison groups is integral to a valid implementation of the DDD methodology. Following the literature, we employ target and comparison groups that are conventionally defined. The exact specifications of the target and comparison groups that we use for various education outcomes are described below.

Our analyses consist of several steps. First, we consider how welfare reform affected high school drop-out among teenage girls. PRWORA likely had direct effects through its minor parent provisions requiring school attendance. In addition, the new regime may have encouraged teenage girls from disadvantaged families, who traditionally have been at risk for future welfare receipt, to complete high school in order to reduce their risk of needing cash assistance in the future (Kaestner et al. 2003). Finally, we allow for the possibility that welfare reform increased the probability that young women continued schooling even after the age of 18, as they may recognize that there is a lower return to not graduating from high school in the post-reform period. As such, we define the target group as unmarried females between the ages of 15 and 20 living in a non two-parent household (that is, one or no parent) where the highest grade attained by anyone in the household is less than a bachelor’s degree. ¹⁰ The comparison group consists of similar males.¹¹ As indicated earlier, while there is some literature on how welfare reform has affected teenage high-school drop-out rates, previous studies have used potentially problematic target groups (Offner 2005) or comparison cohorts that were very widely spaced (Kaestner et al. 2003), very narrowly spaced (Hao and Cherlin 2004), or excluded at-risk non-recipients (Koball 2007).

Next, we address the primary research question of this paper: What are the effects of welfare reform on adult women’s education acquisition? To investigate how welfare reform has affected adult women’s high school enrollment, we compare unmarried mothers ages 21 to 49 years with less than a high school education (target group) to unmarried women in the same age and education groups who have no children (comparison group). We focus on women ages 21 and older since, as discussed earlier, the incentives to attend and finish high school are very different for this group than for younger women.¹² To investigate how welfare reform has affected college enrollment among adult women who are at risk of being on welfare, we compare unmarried mothers ages 24–49 years with less than a college education (target group) to unmarried women in the same age range and education group who have no children (comparison group).¹³ The third step of the analysis involves an extensive set of model specification checks and an exploration of potential mechanisms.

Before undertaking any of the analyses, however, we examine whether our comparison groups are valid counterfactuals. Table 1 shows the mean education outcomes for the various target and comparison groups prior to welfare reform. For analyses of teen drop-out, we examine years prior to the enactment of TANF (1992 to 1995), since welfare policy to discourage teen drop-out was an explicit feature of PRWORA but not of most waivers. For analyses of adult women, we examine the first year of the sample period, 1992, which pre-dated welfare reform.¹⁴ Panel A compares high school drop-out rates for unmarried females and males ages 15–20 (and 15–19) living in non two-parent low-educated households. Prior to the implementation of major waivers or TANF, high-school drop-out rates were virtually identical between males and females in these households, with no statistically significant differences.

Table 1.

Baseline means—target and comparison groups

Panel A: high school drop-out, 1992–1995 (pre-TANF)
	Target Group	Comparison Group
	Unmarried Females	Unmarried Males
	Age 15–20	Age 15–20
	Non Two-Parent Household	Non Two-Parent Household
	Household Education: < College Graduate	Household Education: < College Graduate
Drop-out	0.215	0.227
	Unmarried Females	Unmarried Males
	Age 15–19	Age 15–19
	Non Two-Parent Household	Non Two-Parent Household
	Household Education: < College Graduate	Household Education: < College Graduate
Drop-out	0.213	0.213
Panel B: school enrollment (Adult women), 1992 (Non-waiver states)
	Target Group	Comparison Group
	Unmarried Mothers	Unmarried Women—No Children
	Age 21–49	Age 21–49
	< High School Graduate	< High School Graduate
Current high school enrollment	0.029	0.028
Any school enrollment	0.035	0.030
	Unmarried Mothers	Unmarried Women—No Children
	Age 24–49	Age 24–49
	High School Graduate	High School Graduate
	< College Graduate	< College Graduate
Current college enrollment	0.089	0.104*
Full-time college enrollment	0.047	0.048
	Unmarried Mothers	Unmarried Women—No Children
	Age 24–49	Age 24–49
	< College Graduate	< College Graduate
Any school enrollment	0.078	0.091**

Panel B presents means for high-school and college enrollment in 1992 for the relevant target and comparison groups for adult women. Differences in the prevalence of current high school enrollment and full-time college enrollment are insignificant between the target and comparison samples. While the difference in current college enrollment is statistically significant at the 10% level, the magnitude is relatively small. This small observed difference in the unconditional means is reduced to zero and becomes insignificant after adjusting for observed covariates (in auxiliary multivariate analyses, not shown). Panel B also presents the means for any school enrollment (high school or college) for the relevant target and comparison groups. The small, statistically significant difference is also eliminated when adjusting later for observed covariates.

Figure 1 presents pre- and post-welfare reform trends, defining welfare reform as the implementation of a major AFDC waiver or TANF, whichever occurred first. Trends in high school dropout (Fig. 1a), any college enrollment (Fig. 1b), and full-time college enrollment (Fig. 1c) are very similar for the target and comparison samples prior to welfare reform. Pre-welfare reform differences in trends in outcomes between the target and comparison groups are also found to be statistically insignificant.¹⁵ Even before any adjustment or controls, however, there is a clear discernible break in these trends post-implementation, adding a note of confidence to the validity of the assumptions underlying our methodology (approximately a 3–4 percentage points decrease in high-school dropout among female teens from disadvantaged families and a 1–2 percentage points decrease in both any and full-time college enrollment among low-educated adult women). In addition to this validation of our primary comparison groups, we later discuss supplementary analyses that utilize alternate comparison groups.

Fig. 1.

Pre- and post-welfare reform trends

Results

All models are estimated as linear probability models (though marginal effects are robust to estimation via probit or logit methods), and reported standard errors are adjusted for arbitrary correlation within each state.¹⁶ In all tables, the estimated effects of welfare reform are indicated by the (bolded) coefficients on the interaction terms between the welfare reform and target group indicators.

Table 2 presents estimates corresponding to Eq. 1, with respect to dropping out of high school. The sample is restricted to individuals who were above the maximum state-mandated age for compulsory school attendance in a given year.¹⁷ Specification 1 indicates that TANF implementation reduced the probability of dropping out of high school by about 3.4 percentage points (15% relative to the baseline mean for the target group) among unmarried females ages 15–20 living in low-educated non two-parent households. The effect is robust to controlling for state-specific trends (specification 2), lagged measures of the state’s economy (specification 3), and lagged measures of the state’s welfare caseload (specification 4). In additional specifications (not shown), redefining the relevant age range to 15 to 19 did not substantively change the results.¹⁸

Table 2.

Effects of welfare reform on high school drop-out, October CPS 1992–2001

Target group	Unmarried females
	Age 15–20
	Non two-parent household
	Household education: < College graduate
Specification	1	2	3	4
Target	−0.0170 (0.0136)	−0.0176 (0.0137)	−0.0176 (0.0136)	−0.0177 (0.0135)
AFDC Waiver	0.0049 (0.0207)	−0.0019 (0.0211)	−0.0036 (0.0218)	0.0005 (0.0228)
AFDC Waiver*Target	−0.0084 (0.0224)	−0.0082 (0.0221)	−0.0083 (0.0221)	−0.0079 (0.0220)
TANF	0.0128 (0.0339)	0.0114 (0.0384)	0.0107 (0.0363)	0.0141 (0.0326)
TANF*Target	−0.0338** (0.0167)	−0.0331* (0.0169)	−0.0330* (0.0168)	−0.0329* (0.0167)
Lagged state economic indicators a	No	No	Yes	Yes
Lagged state welfare caseload b	No	No	No	Yes
State-specific trends	No	Yes	Yes	Yes
Adjusted R-squared	0.040	0.043	0.043	0.043
Observations	18532	18532	18532	18532

Coefficient estimates from linear probability models are presented. Standard errors are adjusted for arbitrary correlation within each state and reported in parentheses. Target group is males ages 15–20 living in non two-parent low-educated (< college graduate) households. All models include age and age-squared, indicators for race and Hispanic ethnicity, indicator for no parents in household, age and age-squared of the oldest household member, state fixed effects and year fixed effects. State covariates include state-level unemployment rate, real state personal income per capita, indicators for maximum compulsory schooling age, and indicators for state high school exit exam requirements. Significance is denoted as follows:

^***p≤ 0.01,

^**0.01<p≤ 0.05,

^*0.05<p≤ 0.10

^aLagged state economic indicators include the one-year and two-year lags of the state-level unemployment rate and real personal income per capita

^bLagged state welfare caseloads include one-year and two-year lags of the number of welfare recipients in the state

The results indicate that, in contrast to TANF, state AFDC waivers had no significant effect on high school drop-out. This is empirically validating since the AFDC waivers generally did not shift incentives with regards to schooling. The validity of the maintained counterfactual assumption underlying the DDD analysis can be assessed by examining the coefficient of the Target indicator in the multivariate models. The fact that it is statistically insignificant and relatively small in magnitude in all specifications confirms that the target and comparison groups were similar prior to the policy shift conditional on observed covariates. Overall, we find that welfare reform, specifically TANF, reduced female teen drop-out by about 15%. This range is very similar to the estimates in Offner (2005), which used a potentially problematic target group based on income, but lower than those in the Kaestner et al. (2003) study, which used less problematic target groups but compared two cohorts very widely spaced over time.

Table 3 presents our main estimates of the impact of welfare reform on adult women’s school enrollment, and Tables 4 and 5 present estimates from alternative model specifications. The results from specifications 1 and 2 in Table 3 weakly suggest that, among unmarried low-educated mothers between the ages of 21 and 49, TANF reduced the probability of attending high school (by 0.8 percentage points) and the probability of any school enrollment, which in addition to high school includes trade school, GED preparatory programs, and any other type of schooling, (by 0.5 percentage point). While the effect magnitudes are somewhat sizeable—they represent a 29% and a 16% decline relative to the target group’s baseline mean—they should be interpreted with care since relatively large standard errors render them statistically insignificant.¹⁹ The estimated effects of TANF on current high school enrollment are on the margin of significance (p-value=0.11), however. These point estimates are not sensitive to controlling for state-specific linear trends (results not shown).²⁰

Table 3.

Effects of welfare reform on school enrollment of adult women, October CPS 1992–2001

Sample	Unmarried		Unmarried
	Age 21–49		Age 24–49
	< High school graduate		High school graduate, < College graduate
Outcome	Current high school enrollment	Any school enrollment	Current college enrollment	Full-time college enrollment	Two-year college enrollment	Four-year college enrollment
Specification	1	2	3	4	5	6
Target	0.0019 (0.0042)	0.0024 (0.0046)	0.0043 (0.0050)	0.0042 (0.0040)	0.0106*** (0.0033)	−0.0057 (0.0041)
AFDC waiver	0.0074 (0.0065)	0.0049 (0.0068)	0.0077 (0.0077)	−0.0046 (0.0046)	0.0153** (0.0068)	−0.0055 (0.0054)
AFDC Waiver*Target	−0.0045 (0.0076)	0.0006 (0.0081)	−0.0194*** (0.0058)	−0.0097* (0.0055)	−0.0149*** (0.0054)	−0.0085* (0.0051)
TANF	0.0029 (0.0116)	0.0047 (0.0176)	−0.0006 (0.0156)	−0.0011 (0.0084)	0.0078 (0.0099)	−0.0088 (0.0104)
TANF*Target	−0.0083 (0.0057)	−0.0049 (0.0068)	−0.0106* (0.0054)	−0.0122*** (0.0041)	−0.0087** (0.0043)	−0.0034 (0.0038)
Adjusted R-squared	0.019	0.022	0.097	0.075	0.038	0.082
Observations	13505	13505	56618	53721	53487	54337

Coefficient estimates from linear probability models are presented. Standard errors are adjusted for arbitrary correlation within each state and reported in parentheses. All target groups have children and comparison groups do not have children. All models include age and age squared, indicators for race and Hispanic ethnicity, indicators for highest grade attended, number of children in the household, residence in a metro area, residence in a center city within a metro area, residence in suburban area, state fixed effects and year fixed effects. State covariates include state-level unemployment rate, real state personal income per capita, indicators for maximum compulsory schooling age, and indicators for state high school exit exam requirements. Significance is denoted as follows:

^***p≤0.01,

^**0.01<p≤0.05,

^*0.05<p≤0.10

Table 4.

Effects of welfare reform on any formal school enrollment of adult women, October CPS 1992–2001

Sample	Unmarried
	Age 24–49
	< College graduate
Outcome	Any school enrollment
Specification	1	2
Target	0.0033 (0.0042)	0.0034 (0.0042)
AFDC Waiver	0.0050 (0.0062)	0.0062 (0.0055)
AFDC Waiver*Target	−0.0173*** (0.0054)	−0.0181***(0.0054)
TANF	−0.0023 (0.0131)	0.0020 (0.0141)
TANF*Target	−0.0082* (0.0047)	−0.0081* (0.0048)
State-specific trends	No	Yes
Adjusted R-squared	0.098	0.099
Observations	68378	68378

Coefficient estimates from linear probability models are presented. Standard errors are adjusted for arbitrary correlation within each state and reported in parentheses. All target groups have children and comparison groups do not have children. All models include age and age-squared, indicators for race and Hispanic ethnicity, indicators for highest grade attended, number of children in the household, residence in a metro area, residence in a center city within a metro area, residence in suburban area, state fixed effects and year fixed effects. State covariates include state-level unemployment rate, real state personal income per capita, indicators for maximum compulsory schooling age, and indicators for state high school exit exam requirements. Significance is denoted as follows:

^***p≤0.01,

^**0.01<p≤0.05,

^*0.05<p≤0.10

^aSample is restricted to individuals who were 21 years of age or older in the first full year that a state implemented an AFDC waiver or the first full year that a state implemented TANF if it did not have an AFDC waiver in place

Table 5.

Effects of welfare reform on high school drop-out and school enrollment of adult women, alternative comparison groups, October CPS 1992–2001

Sample	Unmarried Females	Mothers		Mothers		Mothers
	Age 15–20	Age 21–49		Age 24–49		Age 24–49
	Household education: < College graduate	< High school graduate		High school graduate, < College graduate		< College graduate
Target group	Non Two-Parent Household	Unmarried Mothers		Unmarried Mothers		Unmarried Mothers
Comparison group	Two Parent Household	Married Mothers		Married Mothers		Married Mothers
Outcome	High School Drop-out	Current High School Enrollment	Any School Enrollment	Current College Enrollment	Full-Time College Enrollment	Any School Enrollment
Specification	1	2	3	4	5	6
Target	0.1040*** (0.0092)	0.0134*** (0.0038)	0.0161*** (0.0038)	0.0361*** (0.0037)	0.0272*** (0.0031)	0.0340*** (0.0032)
AFDC Waiver	−0.0043 (0.0171)	−0.0047 (0.0056)	−0.0074 (0.0050)	0.0018 (0.0042)	0.0020 (0.0022)	−0.0001 (0.0038)
AFDC Waiver*Target	−0.0071 (0.0190)	−0.0029 (0.0071)	0.0034 (0.0062)	0.0021 (0.0069)	−0.0063 (0.0043)	−0.0003 (0.0056)
TANF	0.0400* (0.0217)	−0.0018 (0.0075)	0.0034 (0.0101)	0.0052 (0.0086)	0.0070* (0.0042)	0.0039 (0.0073)
TANF*Target	−0.0234** (0.0100)	−0.0050 (0.0054)	−0.0022 (0.0059)	−0.0108** (0.0045)	−0.0102*** (0.0033)	−0.0088** (0.0038)
Adjusted R-squared	0.060	0.017	0.017	0.051	0.035	0.050
Observations	18733	18613	18613	98457	95004	115370

See notes to Tables 2, 3, and 4. All models also include lagged state economic indicators and lagged welfare caseloads, as noted in Table 2

There is stronger evidence that welfare reform had an adverse effect on college enrollment among high-school educated unmarried mothers (ages 24–49). Of this group, women in states that implemented major waivers to AFDC had a two percentage point (22% relative to the baseline target group mean) decrease in the probability of current college enrollment, on average, relative to those in states that did not introduce welfare experiments into their AFDC programs. Similarly, the implementation of TANF reduced the probability of current college enrollment by 1.1 percentage points (12%) among the target group. The next set of specifications considers full-time college enrollment (with the reference category being not enrolled in college at all). Full-time college enrollment, being a time-intensive pursuit, should be especially responsive to time and work constraints imposed by welfare reform. Estimates suggest that AFDC waivers and TANF reduced full-time college enrollment by about one percentage point (21% relative to the baseline mean) and about 1.2 percentage points (26%), respectively, among low-educated unmarried mothers.²¹

Specifications 5 and 6 explore effects for 2-year and 4-year college enrollment separately. While welfare reform is negatively associated with enrollment in both, larger effects are found for 2-year enrollment. Specifically, AFDC waivers and TANF reduced enrollment in a 2-year college by 1.5 percentage points (28%) and about one percentage point (17%), respectively, among low-educated married mothers; this compares to a 9–20% reduction in 4-year college enrollment. The larger adverse effects for 2-year college enrollment are to be expected given that the majority (60%) of low-educated single mothers who attend college do so at a 2-year institution.

The estimates from Tables 2 and 3 underscore the differential schooling incentives embedded in the welfare reform policies. The results indicate that welfare reform significantly increased the probability of staying in high school among teenage girls from disadvantaged families, by 15%, while it has decreased the probability of school enrollment, especially college enrollment, among low-educated adult mothers, by 12–25%. In separate analyses (not reported), we stratified the sample based on race. In absolute terms, the effect magnitudes are statistically similar for whites and blacks. However, given the lower prevalence of college enrollment and other schooling among less-educated single black mothers, the relative decline is somewhat larger for this group (19% decrease in any school enrollment compared to 15% among whites).

Specification Checks

We estimated a series of supplemental models to assess the robustness of our results vis-à-vis adult women’s school enrollment. All results not shown are available upon request.

Compositional Selection

When the target and comparison groups are defined according to ducation, parental status, and marital status, which may themselves be affected by welfare reform, potential bias due to compositional selection is a concern. We confirm that key characteristics used to define the target and comparison groups have not changed significantly over the sample period.²² For instance, 9.9% of all women are classified into the target group in 1992 (being low-educated, unmarried, and with children) compared with 9.8% in 2001. Similarly, the prevalence of marriage among low-educated women and the number of children among low-educated unmarried and married women have remained relatively stable over the sample period. However, there may be selective composition with respect to classifying an individual as low- or high-educated if educational attainment is affected by shifts in welfare policies, which is the key question addressed in this study. The results from Table 2 suggest that welfare reform decreased high school dropout among teens. In this case, an analysis of college enrollment among low-educated (high school, but not college, graduate) unmarried mothers may impart a negative bias to the estimated effect of welfare reform on college enrollment due to the selection of women whose high school completion was positively impacted by the reform.

We address the sensitivity of the estimates to selective sample composition by conducting two specification checks. First, we restricted the sample to individuals who are at least 21 years of age in the year their state implemented an AFDC waiver or TANF, whichever came first. In these specifications (not shown) the coefficient magnitudes for the effect of AFDC Waiver decline slightly (as expected), by about 8–12%, but the conclusions are not materially affected. The effects of TANF decline somewhat more than that²³, although the effects generally remain significant for full-time college enrollment and suggest that TANF implementation reduced full-time college enrollment by about 16%.

Second, we examine enrollment in any type of education (high school, college, or other training programs) among low-educated women, defined broadly as those with less than a college education. This broader definition of low-educated women, based only on college graduation, alleviates compositional selection as a result of the reforms having increased the probability of completing high school (as shown in Table 2).²⁴ While these models, shown in Table 4, do not separately estimate the effect on high school and college enrollment, the relative magnitudes are robust; AFDC waivers reduced any school enrollment by a statistically significant 1.8 percentage points among low-educated adult women with children, which translates into a relative 23% decline (evaluated at the baseline target group mean), and TANF reduced school enrollment by a statistically significant 0.8 percentage point (11% decline). These alternate specifications suggest that overall the inferences are not sensitive to selective sample composition.²⁵

Finally, we explored potential compositional selection by marital status using weighted data from the National Household Survey on Drug Abuse/National Survey on Drug Abuse and Health. Specifically, we compared receipt of public assistance among women who would be in our target group (unmarried mothers aged 24 to 49 with less than a college education) to that of their married counterparts (mothers, same age and education categories) and found that: In 1992, about 30.8% of unmarried mothers received some form of public assistance, whereas only 2.7% of married mothers received assistance during that year, and in 2001, about 14.3% of unmarried mothers received public assistance, whereas only 1.6% of married mothers received public assistance during that year. The very small fraction of married women who received public assistance prior to welfare reform and the lack of increase in that fraction after the full implementation of welfare reform, together, validate the use of marital status to delineate our target and comparison groups.

Policy Endogeneity

It is possible that the specific timing of the implementation of welfare reform is dependent on the state’s economic conditions and welfare history. We address this possibility by expanding the baseline specifications, including one- and two-year lags of the state-level unemployment rate, real personal income per capita, and welfare caseload.²⁶ We also include state-specific linear trends in these models to capture any remaining systematically-varying unobservables that are unique to a given state. The marginal effects of welfare reform on education remain robust with respect to both magnitudes and statistical significance (results not shown).

We further check for potential bias due to policy endogeneity by assessing the importance of lead effects of welfare reform. This set of models includes interactions between one-year lead values of the welfare reform measures and the target group indicator in addition to all of the other variables in Table 3. Significant lead DDD policy effects may not only suggest that changes in outcomes could have driven the implementation of welfare reform policies within states over time (policy endogeneity), but also that there could have been anticipatory effects wherein individuals modified their behaviors in anticipation of welfare reform. In all specifications, the main DDD effects remain robust to adding the leads (results not shown).

Lagged Effects

It is possible that there is a lag between welfare reform and the full response on education outcomes due to adjustments over time in family inputs, binding time limits over time, and learning within the family. To address this possibility, we estimate alternative DDD specifications in which the contemporaneous welfare reform measures are substituted with their one-year lags. Overall, the estimates in the lagged-response models remain robust in terms of statistical significance and direction (results not shown).

Alternative Comparison Groups

The estimates of welfare reform on school enrollment are based on comparison groups that most validated the DDD research design. To further assess the robustness of our findings, we estimated our main models using alternative comparison groups, with results presented in Table 5. For the analysis of high school drop-out, we use unmarried females (ages 15–20) from low-educated two-parent households as the alternative comparison group and find that TANF implementation reduced high school drop-out among the target group by 2.3 percentage points (11%). This figure compares to the 15% effect estimated with our primary comparison group of males from low-educated non two-parent households (from Table 2). For the analyses of college and school enrollment among low-educated unmarried adult mothers, we use married mothers as our alternative comparison group and find that although the effects of AFDC waivers are not statistically significant and are generally close to zero, the effects of TANF are similar to those in Table 3 in terms of both magnitudes and significance. The robustness of the welfare reform effects using alternate assumptions regarding the comparison group, including no comparison group,²⁷ reinforces our findings that welfare reform reduced high school drop-out among girls but adversely affected school enrollment, especially college enrollment, among low-educated adult women.

Earned Income Tax Credit

The Earned Income Tax Credit (EITC) is a federal program that provides a refundable tax credit for low- and middle-income individuals and couples, especially those with children. The credits were substantially expanded in 1990, 1993, and 2001, and unmarried mothers with a high-school degree and fewer than two children have comprised the largest group of EITC recipients. Research has shown that among this group EITC raises employment rates by increasing the returns to work (Eissa and Hoynes 2006), suggesting that the program may have disproportionately decreased school enrollment among the target population. To address this concern, we estimated models with interactions between the year effects and the target indicator to account for all national unobserved factors that may have differential effects on the target group relative to the comparison group. In other models, we additionally interacted the year fixed effects with the number of minor children in the household since the amount of the tax credit is based partially on the number of qualifying children. Results remain robust. For instance, welfare reform is associated with a 14–24% decline in college enrollment (compared to a 12–22% decline reported in Table 3), and a 10–23% decline in any school enrollment (compared to an 11–23% decline reported in Table 4).²⁸ Similar results (noted earlier) based on an alternate comparison group (where both the target and comparison women have children in the household) are also validating.

In addition, ten states had their own EITCs that supplemented the federal credit at some point over the sample period, and an additional five states had EITC programs at all points over the sample period (these latter states would be captured by the state fixed effects). In additional analyses, we controlled for the presence of a state EITC program and its interaction with the Target group. We also control for a larger vector of state-specific factors including state child support expenditures, minimum wage, and poverty rates. Results are highly robust with respect to both magnitudes and significance. This robustness to additional controls for state-varying factors is validating since in a well-specified model, the comparison group would account for time-shifting unobservable factors; thus, results should not be sensitive to parametric controls for state-specific factors.

The Role of Work

The estimates discussed thus far represent the “reduced-form” effects of welfare reform—that is, the total effect of welfare reform on education acquisition, operating through a variety of potential (and possibly competing) mechanisms. Work is the centerpiece of the policy shift and there is strong consensus that welfare reform has indeed increased employment and decreased caseloads as intended. We broadly test the work mechanism through stratification analyses which exploit (1) how states specifically differ in their treatment of education post-PRWORA and (2) the strength of work incentives in states’ TANF plans.

Adverse effects of welfare reform on education that appear to operate through work should be stronger in states that do not permit or limit education as an authorized work activity for adults. To explore this hypothesis, Table 6, Columns 1 through 5, present estimates for states stratified by the degree to which education is permitted to fulfill work requirements. Panel A limits the sample to states that either never allow post-secondary or other education to substitute for work requirements, or impose substantial time limits on the duration of education.²⁹ Across all measures of school enrollment, the adverse effects of welfare reform policies on education are strongest for this sample of restrictive states. For instance, specifications 1 and 4 suggest that TANF lowered the probability of high school enrollment by a statistically significant two percentage points (65%) and lowered the probability of full-time college enrollment by 1.4 percentage points (29%). For states that support schooling as an authorized work activity in some form (Panel B), the adverse effects of welfare reform on high school, college, or other schooling are much mitigated. Among unmarried mothers with less than a high school education, there is no adverse effect of welfare reform on high school enrollment in the more supportive states. With respect to college enrollment, the marginal effects continue to be negative but are much smaller in magnitude than in Panel A. While these education policies are defined based on TANF, they may be reflective of a state’s sentiment towards work versus education incentives in general. Indeed, the impact of AFDC waivers is also lower for the education-supportive states. For instance, AFDC waivers reduce the probability of any school enrollment by 1.4 percentage points (18%) in these states, versus 2.4 percentage points (31%) in states that do not support schooling as a work requirement. These results provide support for the hypothesis that welfare reform reduced schooling among adult women through policies which emphasized work at the expense of education. However, they do not rule out the possibility that stricter rules increase the demand for education over a longer time horizon.

Table 6.

Effects of welfare reform on school enrollment of adult women, stratified samples by TANF education policy, benefit generosity, and strictness of time limits, October CPS 1992–2001

Panel A	School Enrollment Not Supported as Work Requirementa					Low Benefit Generosity c	Strict/Medium Time Limits c
Sample	Unmarried		Unmarried			Unmarried	Unmarried
	Age 21–49		Age 24–49			Age 24–49	Age 24–49
	< High School Graduate		High School Graduate, < College Graduate			< College Graduate	< College Graduate
Outcome	Current High School Enrollment	Any School Enrollment	College Enrollment	Full-Time College Enrollment	Any School Enrollment	Any School Enrollment	Any School Enrollment
Specification	1	2	3	4	5	6	7
Target	0.0073 (0.0050)	0.0043 (0.0064)	0.0083 (0.0085)	0.0083 (0.0072)	0.0075 (0.0074)	0.0088 (0.0071)	0.0063 (0.0047)
AFDC Waiver	0.0271* (0.0148)	0.0215 (0.0215)	0.0077 (0.0117)	−0.0108 (0.0098)	0.0028 (0.0102)	0.0308 (0.0177)	0.0134* (0.0072)
AFDC Waiver* Target	−0.0200 (0.0174)	−0.0093 (0.0179)	−0.0317** (0.0121)	−0.0113 (0.0088)	−0.0244* (0.0126)	−0.0361** (0.0145)	−0.0224*** (0.0080)
TANF	0.0156 (0.0127)	0.0342* (0.0178)	0.0023 (0.0155)	0.0106 (0.0097)	0.0038 (0.0120)	0.0512* (0.0253)	0.0281** (0.0122)
TANF*Target	−0.0197*** (0.0065)	−0.0140 (0.0097)	−0.0123 (0.0091)	−0.0137** (0.0066)	−0.0098 (0.0078)	−0.0148 (0.0084)	−0.0071 (0.0051)
Adjusted R-squared	0.027	0.029	0.098	0.075	0.098	0.099	0.094
Observations	6811	6811	27742	26372	33616	11757	49665
Panel B	School Enrollment is Supported as Work Requirement					High or Medium Benefit Generosity	Lenient Time Limits
Sample	Unmarried		Unmarried			Unmarried	Unmarried
	Age 21–49		Age 24–49			Age 24–49	Age 24–49
	< High School Graduate		< College Graduate			< College Graduate	< College Graduate
Outcome	Current High School Enrollment	Any School Enrollment	College Enrollment	Full-Time College Enrollment	Any School Enrollment	Any School Enrollment	Any School Enrollment
Specification	1	2	3	4	5	6	7
Target	−0.0028 (0.0062)	0.0015 (0.0070)	0.0001 (0.0057)	−0.0013 (0.0041)	−0.0015 (0.0042)	0.0026 (0.0049)	−0.0052 (0.0063)
AFDC Waiver	0.0006 (0.0096)	−0.0003 (0.0102)	0.0122 (0.0095)	0.0031 (0.0063)	0.0086 (0.0088)	0.0068 (0.0065)	0.0090 (0.0083)
AFDC Waiver* Target	0.0007 (0.0097)	0.0007 (0.0104)	−0.0149** (0.0060)	−0.0060 (0.0067)	−0.0141** (0.0054)	−0.0172*** (0.0060)	−0.0070 (0.0072)
TANF	−0.0104 (0.0179)	−0.0180 (0.0176)	0.0114 (0.0256)	0.0019 (0.0122)	0.0087 (0.0210)	0.0006 (0.0142)	0.0229* (0.0116)
TANF*Target	0.0024 (0.0083)	0.0016 (0.0093)	−0.0084 (0.0063)	−0.0087* (0.0049)	−0.0059 (0.0055)	−0.0070 (0.0055)	−0.0103 (0.0095)
Adjusted R-squared	0.029	0.030	0.099	0.078	0.100	0.098	0.111
Observations	6694	6694	28876	27349	34762	56621	18713

See Tables 3 and 4. In addition, we include state-specific trends and lagged state economic indicators. Lagged state economic indicators include 1-year and 2-year lags of state-level unemployment rate and real personal income per capita

^aSample is limited to states that do not completely allow post-secondary education to count toward the state’s work requirement

^bSample is limited to states that have high earnings disregard, as defined in Blank and Schmidt (2001)

^cSample is limited to states that have strict or medium time limits as defined in Blank and Schmidt (2001)

The final specifications in columns 6 and 7 of Table 6 stratify the sample based on work incentives as reflected in components of state TANF programs. Specifically, Model 6 in Panel A includes only states that have strong work incentives as measured by low benefit generosity and Model 7 is restricted to states that have strong work incentives by virtue of having strict time limits (Blank and Schmidt 2001).³⁰ Individuals in both sets of states face a much stronger push towards work (compared to states with relatively generous benefits and lenient time limits, respectively), which may have come at the expense of education. The marginal effects confirm this scenario and suggest that welfare reform (in the form of AFDC waivers) decreased any school enrollment by 2.2 to 3.6 percentage points (28–46% relative to the baseline target sample mean). Similarly, TANF decreased any school enrollment by 1.5 percentage points (19%) among residents in states with less-generous welfare benefits, though this effect is imprecisely estimated. For individuals who reside in states with weaker work incentives, the adverse effects of welfare reform on education are also weaker. Specifically, estimates from specifications 6 and 7 in Panel B indicate that welfare reform (AFDC waivers) decreased any school enrollment by 9–22% in those states and the effect is not statistically significant for the sample of states with lenient time limits.

The push to work is the key hypothesized pathway by which welfare policies may affect education, and the stratified analyses in Table 6 are consistent with this hypothesis. Greater time in market work reduces other available time and exacerbates time constraints, which in turn may reduce the demand for time-intensive activities such as schooling. Associated with changes in employment are changes in personal and household income, though total income post-welfare reform may increase or decrease as earnings substitute for cash benefits. The effect of income is ambiguous depending on the sign of the income effect, the source of income (labor vs. non-labor), and whether total income has increased for the woman post-welfare reform. For instance, if the income effect with respect to non-labor income is positive (for instance, see Kodde 1986), then the reduction in welfare benefits may reduce the demand for schooling. If education, on the other hand, is an overall normal good with respect to total income, and total income increases post-welfare reform, then this would increase the demand for schooling. In order to shed light on these different pathways, Table 7 presents estimates from models that incorporate the role of time (as proxied by hours worked) and income constraints as mediating factors.

Table 7.

Effects of welfare reform on school enrollment of adult women, assessing hours worked and income as potential mediators, October CPS 1992–2001

Sample	Unmarried		Unmarried		Unmarried
	Age 21–49		Age 24–49		Age 24–49
	< High school graduate		High School Graduate, < College Graduate		< College graduate
Outcome	Current high school enrollment	Any school enrollment	Current college enrollment	Full-time college enrollment	Any school enrollment
Specification	1	2	3	4	5
Hours worked (1–19)a	−0.0057 (0.0102)	0.0024 (0.0116)	0.0401*** (0.0134)	0.0296*** (0.0103)	0.0303** (0.0118)
Hours worked (20–31)	−0.0240*** (0.0061)	−0.0276*** (0.0068)	−0.0280*** (0.0070)	−0.0403*** (0.0068)	−0.0272*** (0.0055)
Hours worked (32–40)	−0.0254*** (0.0039)	−0.0287*** (0.0042)	−0.0615*** (0.0065)	−0.0754*** (0.0060)	−0.0574*** (0.0056)
Hours worked (>40)	−0.0251*** (0.0057)	−0.0343*** (0.0059)	−0.0733*** (0.0073)	−0.0809*** (0.0061)	−0.0705*** (0.0059)
Household incomeb	−0.0006*** (0.0002)	−0.0007*** (0.0002)	0.00002 (0.0002)	−0.0001 (0.0001)	−0.0002 (0.0002)
Targetc	0.0204*** (0.0056)	0.0248*** (0.0057)	0.0234** (0.0094)	0.0075 (0.0082)	0.0245*** (0.0079)
AFDC waiver	0.0016 (0.0084)	−0.0005 (0.0100)	0.0061 (0.0077)	−0.0051 (0.0053)	0.0024 (0.0062)
AFDC Waiver*Target	−0.0017 (0.0114)	0.0019 (0.0121)	−0.0171*** (0.0061)	−0.0073 (0.0053)	−0.0151*** (0.0051)
TANF	0.0020 (0.0141)	0.0072 (0.0199)	−0.0013 (0.0141)	−0.0065 (0.0097)	−0.0008 (0.0119)
TANF*Target	−0.0056 (0.0062)	−0.0032 (0.0080)	−0.0066 (0.0052)	−0.0067* (0.0038)	−0.0037 (0.0047)
Lagged state welfare caseloade	Yes	Yes	Yes	Yes	Yes
Adjusted R-squared	0.030	0.034	0.116	0.120	0.115
Observations	12423	12423	51529	48854	62336

See notes to Table 5. Models also include the one-year and two-year lags of the number of welfare recipients for the state

^aReference category is individuals who are not working. Marginal effects for hours worked are reported for the target group. All models also include interaction terms between the indicators for hours worked and the comparison group indicator to allow the marginal effects to differ between the target and comparison groups

^bHousehold income is measured in thousands of dollars, adjusted by the national consumer price index. Marginal effect for income is reported for the target group. All models also include an interaction term between household income and the comparison group indicator to allow the marginal effect to differ between the target and comparison groups

^cIn this context, the marginal effect of the target group indicator represents the difference in pre-welfare reform school enrollment between individuals in the treatment and comparison groups who are not currently working and have zero household income

The specifications in Table 7 include indicators of hours worked in the past week (reference category being non-employed) and a continuous measure of household income, as well as interactions between these labor supply measures and the target group indicator, to allow the effects of labor supply to differ between the target and comparison groups. The models indicate a significant non-linear relationship between hours worked and school enrollment. Up to about 19 h of work per week, intensity of labor supply has positive or no effects on school enrollment relative to individuals who do not work.³¹ This may be due to potential simultaneity between the labor supply decision and school enrollment. In addition, increases in personal income associated with higher hours worked may, up to a point, lead to a positive income effect on schooling. Individuals who work more than about 19 h weekly have a lower propensity to attend high school and college. The negative link between work and schooling becomes stronger with more hours worked, consistent with an exacerbation of time constraints with greater labor supply. There may be measurement error in reported hours worked. To the extent that this is the case, the estimated impacts of hours of work are potentially downward-biased. Household income generally has a negative impact on the probability of being currently enrolled in school for the target group.³²

More importantly, hours worked substantially mediate the impact of welfare reform on education.³³ When including hours worked, there is no longer a statistically significant effect of either AFDC Waivers or TANF on school enrollment among unmarried mothers with less than a high school education, with effect magnitudes declining by 36–69% relative to those reported in Table 3. It is plausible that work constraints are particularly consequential for the group with less than a high school education. Although it is possible that some of these women may have substituted from formal high school programs to less time-intensive schooling such as other training or GED preparatory programs, the diminution of the marginal effect of TANF on any schooling suggests that this substitution was not a significant outcome.

Among high school graduates with less than a college education, hours worked also mediates the negative effect of welfare reform on college enrollment. In contrast to the less-than-high school educated women, the diminution of the effect magnitudes is smaller (on the order of 15–42%). Some of the largest diminution of effects of welfare reform (on the order of 27–42%) occurs for full-time college enrollment, a relatively time-intensive educational activity. These results should be interpreted with caution due to potential simultaneity between the work and schooling decisions. However, the diminution of the effects (when labor supply is included) suggests that at least part of the negative impact of TANF and AFDC waivers on adult women’s education acquisition was due to added time constraints resulting from increased work requirements.

Conclusion

This study makes three major contributions. First, it contributes to the sparse literature on adult education by identifying welfare policy as a strong determinant of education acquisition among poor adult women and underscores that policies not specifically focusing on education may be important determinants of educational attainment. That said, it is important to underscore that we have looked at enrollment, not attainment, and it is therefore possible that our effects reflect a slowing down of schooling rather than decreased educational attainment.

Second, we produced the most compelling evidence to date on the effects of welfare reform on high school drop-out. Specifically, we found that welfare reform significantly increased the probability that teens from disadvantaged families stayed in high school, by about 15%. The opposing effects for teens and adult mothers underscore the differential educational incentives for the two groups that are built into welfare reform.

Third, and most importantly, we found robust and convincing evidence that welfare reform significantly decreased the probability of college enrollment among adult women, by at least 20%. It also appears to have decreased the probability of high school enrollment of adult women on the same order of magnitude, although high school enrollment, unlike college enrollment, is a relatively uncommon outcome among adult unmarried mothers with less than a high school education. These negative effects may have become even larger under the Deficit Reduction Act of 2006, which raised states’ work participation targets and narrowed the range of welfare-to-work activities that can be counted toward those targets.

The finding that welfare reform decreased formal education enrollment may have negative implications for poor mothers’ ability to attain self-sufficiency and experience upward mobility, given the evidence of substantial earnings gains from even 1 year of community college. While not every potential welfare recipient would benefit from a high school or college education, it cannot be assumed that the large negative aggregate effects of welfare reform on formal school enrollment among adult women are an economically efficient or socially desirable outcome. A comprehensive assessment of the costs and benefits of welfare reform would need to account for the impact on educational attainment. In addition, it is important to further explore the extent to which adult women may have substituted vocational education and training or on-the-job training for formal education, as well as the long term effects of welfare reform on household income and other economic outcomes. Preliminary results on vocational education and training based on the National Household Education Surveys (NHES)³⁴ suggest that welfare reform reduced vocational education and training and therefore that a substitution of vocational education for formal education did not take place (Dave et al. 2011).