Ending Access as We Know It: State Welfare Benefit Coverage in the TANF Era

Ending Access as We Know It: State Welfare Benefit Coverage in the TANF Era

Ending Access as We Know It: State Welfare Benefit Coverage in the TANF Era

Permalink: ‎

Much of the quantitative literature evaluating welfare reform focuses on caseloads. In order to contextualize caseload declines, the current study examines a closely related measure of welfare coverage: the ratio of children receiving welfare assistance to children in poverty. A multilevel model approach is employed to investigate state-level factors that have contributed to declines in coverage. The findings suggest that welfare coverage has fallen the most in states with higher levels of coverage prereform, ideologically conservative governments, Republican governors, and larger proportions of African American welfare recipients. In addition, this study identifies specific policies and administrative practices that are associated with falling coverage and reveals a substantial erosion of the traditionally countercyclical relationship between unemployment and welfare provision since reform. By the late 2000s, the policy choices that embody welfare reform have produced both historically low levels of welfare coverage nationally and unprecedented diversity in benefit accessibility across states.

In his speech accepting the Democratic nomination for President of the United States, William Clinton promised to “end welfare as we know it” (New York Times 1992). One of the main problems with the Aid to Families with Dependent Children program (AFDC), according to Clin- ton and others in favor of dramatic reform, was that it encouraged dependency (O’Connor 2000). Advocates of reform viewed welfare de- pendency as both the cause and effect of a variety of social ills, including teenage pregnancy, crime, and low labor-market participation among racial and ethnic minorities. In creating the Temporary Assistance for Needy Families program (TANF), the Personal Responsibility and Work

224 Social Service Review

Opportunity Reconciliation Act of 1996 (PRWORA; 110 Stat. 2105) cod- ified this rhetoric about the ills of dependency. “End[ing] the depen- dence of needy parents on government benefits by promoting job prep- aration, work, and marriage” is listed as one of the four main goals of the new program (110 Stat. 2113 [1996]).

With dependency framed as a problematic consequence of welfare provision, caseload reduction became the primary metric of welfare reform’s effectiveness. As caseloads declined dramatically following re- form, many media commentators, regardless of political orientation, viewed these declines as an indication that welfare reform did something right (Besharov 2006; Clinton 2006; Jencks, Swingle, and Winship 2006; Kim and Rector 2006; New York Times 2006). Academics also contributed to this debate, studying why caseloads fell so quickly after the institution of TANF (Council of Economic Advisors 1997, 1999; Martini and Wise- man 1997; Mead 2000; Schoeni and Blank 2000; Ziliak et al. 2000; Blank 2001; Danielson and Klerman 2008).

The current study explores the long-term consequences of reform for the adequacy and responsiveness of state welfare (TANF) programs. Access to cash assistance declined dramatically after reform. A 2008 Congressional Research Service report finds that, in 2007, one-third of single mothers in poverty were both unemployed and not receiving cash benefits, over twice the proportion in this situation in 1995 (Burke, Gabe, and Falk 2008). Studies examining levels or change in state case- loads can provide insight into these developments, but caseload mea- sures are not ideal indicators of welfare state adequacy. The primary issue is that it is difficult to interpret the meaning of a caseload decline without assessing whether need is declining as well. In the following, the authors hope to help shift the focus of the welfare reform debate toward questions of welfare state adequacy and away from discussions of dependency and caseloads. Following the work of Marcia Meyers, Janet Gornick, and Laura Peck (2002), this study employs a different measure as the dependent variable in the analyses: the number of state welfare child cases relative to the number of children in poverty, a measure of welfare coverage.

The research on caseload changes since welfare reform is dominated by debate about the extent to which caseload declines are a consequence of economic or policy changes. This framing, combined with an em- pirical focus on the uniquely strong economic growth following reform in the late 1990s, obfuscates important transformations in access to welfare services and enables, however unintentionally, the development of unqualified narratives about the success of welfare reform. Examining the performance of TANF through the lens of a coverage measure may suggest alternative narratives.

Coverage in the TANF Era 225

Fig. 1.—States’ average number of children receiving welfare and states’ average welfare coverage for children, 1995–2009.

Coverage versus Caseloads

In this study, the focus on welfare coverage over caseloads deserves some further elaboration. Figure 1 presents national trends for both measures since welfare reform. Specifically, this figure displays the mean of the number of children receiving welfare in each state (recipients of benefits from AFDC, TANF, and SSPs [Separate State Programs]) as well as the mean of state child coverage rates between 1995 and 2009. Separate State Programs are TANF-like programs funded by states and admin- istered by state TANF offices, but these programs were exempt from many federal TANF policies, such as time limits and work requirements, until TANF was reauthorized in 2006. Many states have used SSPs to varying degrees to provide assistance to families outside of the frame- work (and, some argue, the constraints) of TANF. The child caseload measure is the ratio of the average monthly number of children re- ceiving assistance to the total number of children in a state. Child cov- erage is the ratio of the average monthly number of children receiving assistance to the total number of children in poverty in a state.

If caseload decline is the sole measure of success, then welfare reform has been an extraordinary triumph. Nationally, the total number of children receiving welfare declined 65 percent between 1996 and 2007. At the state level, there is substantial variation in the magnitude of

226 Social Service Review

caseload decline. In a handful of states (Florida, Georgia, Idaho, Illinois, Louisiana, Mississippi, and Wyoming), child caseloads declined by over 80 percent between 1996 and 2007. Caseloads did increase in response to the 2007–9 recession; total child caseloads rose nearly 13 percent between 2008 and 2010. Similarly, average welfare coverage fell dra- matically nationwide, with individual states converging on historically low rates of coverage. In contrast to caseload trends, whether measured at the national level or as state averages, child coverage decreased every year since reform, even falling through the 2001 and 2007–9 recessions.

A simultaneous examination of the two measures is instructive. For example, declines in coverage between 1995 and 1998 appear to be driven largely by falling caseloads and not by reductions in poverty. The drop in caseloads continues from 1998 until the 2001 recession, but the decline in coverage moderates substantially in these years. This is a result of the considerable drop in poverty during the very late 1990s. However, the fact that coverage continues to decline in these years indicates that the decline in caseloads is more than that warranted by the declines in poverty and unemployment alone. Finally, although child caseloads sta- bilize for a few years during and following the 2001 recession, and even increase in 2009 and 2010, coverage falls through both of these reces- sionary periods. These trends indicate that caseloads did not keep pace with the increase in child poverty during either recession.

This is a key advantage of the coverage measure, as it enables the assessment that the dramatic caseload declines, especially in the very late 1990s, are not driven solely by falling poverty in the context of a tight labor market. Further, reliance on a caseload measure could lead one to overestimate the adequacy of state responses to postreform re- cessions. However, the differences between these two measures should not be overstated, as they are closely related, have identical numerators, and display similar overall trends. At the state level, the measure of child coverage and the ratio of child cases to child population are highly correlated (r p .85). It is important to stress that the use of a coverage measure is not intended to be a methodological contribution; the cov- erage ratio is not presented as a more accurate measure of some com- mon underlying concept than caseloads. Instead, the coverage measure is considered a better indicator of the adequacy and responsiveness of TANF. Consequently, this study is not a direct extension of research examining caseloads; rather, it focuses on the specific question of the determinants of change in program adequacy since reform. The authors expect that the factors influencing coverage are not necessarily identical to those that affect caseloads.1

1. While tangential to this study’s primary research questions, analyses were run ex- amining the determinants of change in the child caseload to child population ratio in

Coverage in the TANF Era 227

The coverage measure is desirable for several other reasons. On a descriptive level, it is more intuitively informative and accessible than the measure of caseloads. In 2009, nearly 4 percent of all children participated in TANF or an SSP. The child coverage ratio for the same year was .21. The coverage measure allows an immediate assessment of the extent of program use relative to need, something that is not possible with a caseload measure. This limitation of caseload measures is exac- erbated if one wishes to examine changes in caseloads or to make com- parisons over time. Caseload numbers are sensitive to the size of the population eligible for benefits, and that population fluctuates in re- sponse to changing economic conditions. Focusing on coverage allows one to partially control for the mechanistic changes in eligibility, and consequent changes in caseload volume, created by macroeconomic fluctuations.

Variation across States

While all states have experienced declines in coverage since 1995, within this national trend trajectories of change in coverage vary substantially across states. Figure 2 displays welfare coverage rates for children in five states from 1995 to 2009. Coverage declines substantially in California and Alabama, but both states maintain their customary positions at the extremes of a now compressed spectrum of welfare adequacy. Illinois, on the other hand, experienced dramatic reductions in coverage through the 2001 recession, and these declines substantially change its rank order in level of coverage. Cumulatively, these state-level changes constitute a trend of nationwide convergence upon lower levels of wel- fare coverage.

In the context of federal policy constraints, a broad mandate to reduce caseloads, and falling coverage nationwide, why have some states re- duced welfare coverage more substantially than others? The 2001 re- cession, the subsequent weak recovery, and the intensity and duration of the 2007–9 recession have provided dramatic tests of TANF’s re-

addition to the analyses of child coverage provided below. While many of the results are similar, the findings are not identical and differ in noteworthy manners. In particular, a number of key factors of interest are statistically significant in one analysis and not the other. Further, comparisons of standardized coefficients across models indicate that the magnitude of effects vary substantially between these different dependent variables. This may lead one to either overstate or understate the impact of a particular factor. For example, states with larger proportions of African Americans receiving welfare benefits experienced statistically significant and substantial declines in both child cases and child coverage. However, the estimate of the effect of caseload racial composition is nearly twice as large in the caseload analysis as the coverage analysis, even when controlling for child poverty and state unemployment rates. What this suggests is that a substantial portion of the reduction in caseloads in states with more African American welfare recipients is attributable to falling child poverty in those states.

228 Social Service Review

Fig. 2.—Welfare coverage for children, selected states, 1995–2009

sponsiveness to increases in poverty. Overall, the impact of these eco- nomic downturns on coverage has been surprisingly weak, although individual states exhibit significant variation in their responses to in- creases in poverty.

The current study seeks to explain this variation by examining the factors that have shaped state-level trajectories of change in coverage following reform. The manner in which state political and economic conditions as well as policy changes and changes in administrative prac- tices have influenced these trajectories is investigated using a form of hierarchical linear modeling for longitudinal analyses, the multilevel model for change. This study contributes to the debate over recent changes in welfare provision on a number of levels. First, the following exploration of the determinants of changes in coverage in the TANF era is the most extensive to date. The vast majority of research on caseload decline is confined to the 1990s. The period examined here covers both the 2001 and 2007–9 recessions, allowing the authors to assess how welfare reform affected coverage in both the late 1990s and during the economically turbulent 2000s. Second, the modeling ap- proach utilized here permits a more detailed examination of the effects of time-invariant factors, especially stable, state political and racial char- acteristics, than is possible in the approaches utilized in many studies.

Coverage in the TANF Era 229

Potential Determinants of Coverage

Little research specifically examines AFDC or TANF benefit coverage. Meyers and associates (2001, 2002) find that coverage declined dra- matically in the 1994–98 period, but they do not explore the causes of these changes. The literature on caseloads, broad studies of welfare generosity and retrenchment, and research examining states’ TANF pol- icy choices and administrative practices suggest additional factors that may influence welfare coverage.

Following an unusual rise in the early 1990s, AFDC caseloads began a dramatic and unprecedented decline in 1994 (Blank 2001, 2002). A 1997 Council of Economic Advisors report on this decline triggered the development of the caseload literature. The report concludes: “The estimates provided here suggest that over 40 percent of the decline in welfare receipt between 1993 and 1996 may be attributed to the falling unemployment rate and almost one-third can be attributed to the waiv- ers” (1997, 11); that is, to policy changes. Continuing in the mold set by the 1997 report, several studies (e.g., Wallace and Blank 1999; Blank 2001, 2002) find that the economy and policy are both important to explaining caseload decline. However, James Ziliak and colleagues (2000) find that policy has a negligible effect and that the strong econ- omy of the late 1990s was a primary driver of the caseload decline. Developing an index that characterizes the strength of state-level TANF sanctions, Robert Rector and Sarah Youssef (1999) find substantially larger declines in caseloads between 1997 and 1998 in states with stricter sanctions. Using this same index, Joe Soss and associates (2001) report similar findings based on their examination of changes in caseloads be- tween 1997 and 1999.

This literature suggests that economic factors likely play a central role in explaining caseload decline and that policy and political variables may also be important. Most of the scholars who study caseload decline do not include political variables, but those who do find statistically significant effects. Rebecca Blank (2001), for example, finds that the presence of Republican governors and partisan control of the state legislature by either party reduce AFDC and TANF caseloads. Political factors are prom- inent in research examining the determinants of state policy content under TANF. The wide range of punitive and disciplinary policy features incorporated in state TANF programs is directly relevant to explaining coverage, as states that implemented more stringent policies would be expected to be more likely to restrict access to TANF. Matthew Fellowes and Gretchen Rowe (2004) find that liberal citizen and government ide- ology as well as the proportion of Democrats in the state legislature all reduce the stringency of state eligibility requirements under TANF. Sim- ilarly, Soss and colleagues (2001) find that liberal government ideology reduces the strength of state sanctions under TANF.

230 Social Service Review

However, in more recent work, Soss, Richard Fording, and Sanford Schram (2011) find that party control and state government ideology provide no leverage in explaining whether states adopted a wide variety of TANF policies ranging from harsher sanctions and more rigid work requirements to restrictive eligibility standards. This stands in contrast to their extensive research, which indicates that partisan control of state governments is consistently a primary factor shaping changes in a variety of features of state welfare programs, including AFDC benefit levels and the adoption of AFDC waivers, in the decades preceding reform. This development leads Soss and colleagues (2011) to suggest that welfare reform may have fundamentally altered the forces shaping state welfare provision. In addition to state political context, the two other primary forces that have been central to shaping state action and policy choices in regard to welfare provision are the racial composition of states (and welfare recipients) and market wages for low-income workers.

The existence of multiple and pervasive effects of race on welfare provision, both historically and today, is one of the most consistent findings in research examining welfare benefits and state policy choices (Soss et al. 2011). In terms of TANF policies specifically, states with higher percentages of African American residents tend to implement more re- strictive policies (Soss et al. 2001; Fellowes and Rowe 2004). Further, Soss and colleagues (2011) find that across all dimensions of TANF policy choices examined, ranging from strength of sanctions to eligibility stan- dards, states with larger proportions of African Americans receiving ben- efits were more likely to adopt stringent or restrictive policies.

Local labor-market conditions, in particular the level of demand and wages for low-skilled labor, are also argued to be central to the character of welfare accessibility and generosity (Piven and Cloward 1971; Soss et al. 2011). In the decades preceding reform, changes in AFDC benefits were strongly associated with the ratio of benefits to average wages for low-skilled workers (Soss et al. 2011). Further, Soss and colleagues find that, in the early 2000s, patterns of TANF sanctions in Florida counties were strongly related to local unemployment rates and demand for low- wage labor. Broadly speaking, such labor market impacts are argued to operate on a “principle of less eligibility,” in which access to benefits and the generosity of benefits are limited in manners that ensure welfare remains less attractive or accessible than the lowest-paying jobs within local labor markets (Piven and Cloward 1971, 35).

Finally, a handful of studies examine the effects of changes in admin- istrative practice under TANF, in particular the rise in both formal and informal diversion practices. Formal diversion practices may take the form of the offer of one-time, lump-sum payments. In exchange for such pay- ments, recipients agree to forego TANF eligibility for a specified period. Other diversion programs assist applicants in utilizing publicly or privately provided services other than TANF (Ridzi and London 2006). While there

Coverage in the TANF Era 231

are no systematic figures on the number of applicants diverted nationwide, a number of case studies suggest that utilization of diversion strategies is widespread and in some cases aggressive. Drawing upon studies from four states, Rebecca London (2003) reports increases in the numbers of di- verted recipients and expansion of the use of one-time cash assistance, although in all locations, less than 10 percent of all cases were diverted. In a study of 2,400 low-income families living in Boston, Chicago, and San Antonio, Robert Moffitt (2003) finds that diversion experiences are extremely common. Finally, Frank Ridzi and Andrew London (2006) dis- cover that an overwhelming number of formal and informal diversion practices have been integrated into the TANF intake process in West County, New York.

Efforts to shift TANF recipients onto the caseloads of different gov- ernment programs parallel these diversion tactics and represent another change in administrative practice. Specifically, studies suggest that wel- fare reform has provided incentives for both individuals and state gov- ernments to make greater use of the Supplemental Security Income program (SSI) over TANF. The incentive for individual recipients is that SSI payments are higher than those from TANF, and SSI does not impose work requirements or time limits. For state governments, there are strong formal incentives to reduce TANF caseloads but not SSI caseloads. In addition, some argue that states have a financial incentive to en- courage movement from TANF to SSI, as SSI is financed entirely by federal funds (Nadel, Wamhoff, and Wiseman 2003–4; Schmidt and Sevak 2004; Wamhoff and Wiseman 2005–6).

Data and Hypotheses

The data set compiled for this study contains annual observations on 50 states for a 14-year period (1995–2009), and the various models discussed below examine change in coverage over three periods: 1995– 2009 (the entire period), 1995–2000, and 2000–2009. The dependent variable in these analyses is an annual measure of welfare coverage for children, which is the number of children, in an average month, re- ceiving AFDC, TANF, or SSP benefits relative to the number of children in poverty in that state. The primary reason for focusing on the number of children receiving assistance is to obtain an assessment of the ade- quacy of program participation relative, roughly, to the size of the pop- ulation served by the program. The vast majority of recipients of TANF funds are children, and the proportion of recipients who are children has increased over time with the rise in the number of child-only cases that do not have an adult recipient (US Government Accountability Office 2011). In 1995, child recipients constituted 68 percent of all AFDC recipients; by 2007, the proportion of child recipients of TANF had risen to 77 percent (SSA [Social Security Administration] 1997; US

232 Social Service Review

Department of Health and Human Services [USDHHS] 2009b). Op- erationalizing the coverage measure as the ratio of the total number of child recipients to the total number of poor children comes much closer to assessing the TANF caseload relative to the target population than a ratio of the total number of recipients to the total number of individuals under the poverty line. The data for the average monthly number of children receiving AFDC, TANF, and SSP benefits are drawn from two sources. The Annual Statistical Supplement to the Social Security Bulletin provides data for the years 1994–99 (SSA 1994–1999). For the years 2000–2009, TANF and SSP caseload data come from the USDHHS Ad- ministration for Children and Families (2009a–2009d, 2010a–2010p).

An ideal measure of coverage would be the ratio of the total number of child TANF recipients to the total number of poor children in single- parent households. Unfortunately, state-level estimates of the number of poor children in single-parent households suffer from measurement error as a consequence of focusing on such a small segment of the population. This issue is especially problematic in the context of less populous states, where sample sizes are small. Instead, for this study, the best available state-level estimates of child poverty, the Census Bu- reau’s Small Area Income and Poverty Estimates, are used as the de- nominator in the coverage ratio (US Census Bureau 2011). The Small Area Income and Poverty Estimates have the additional benefit of ac- counting for the influence of taxes and tax credits on household in- comes. Given multiple constraints and considerations, the authors feel strongly that this specific construction of the coverage variable is the best possible for assessing welfare adequacy over time and across states.2

States have responded to welfare reforms in two ways that complicate efforts to accurately characterize the extent to which states are providing assistance. The first involves state use of SSPs following the 1996 reform, which many states created in order to provide a broader level of assistance than was possible within the constraints of federal TANF guidelines. States could create SSPs that were funded solely by the state but administered by TANF agencies to meet federal maintenance-of-effort requirements. Despite the increased cost associated with creating and operating these programs, states had an incentive to utilize SSPs, as families and children receiving support through SSPs were not considered to be receiving TANF assistance, were not subject to a number of TANF requirements (including work participation requirements), and were not included in the calcu- lation of state work-participation targets (Cohen 2006; SSA 2008a).

Unfortunately, data on SSP caseloads are only available beginning in

2. The authors also considered examining caseloads as a proportion of the eligible population, because eligibility criteria are determined at the state level and vary widely across states. This approach is rejected, however, because eligibility criteria affect the extent to which assistance reaches the poor (in this case, poor children). Instead, a measure of eligibility thresholds is included as an independent variable.

Coverage in the TANF Era 233

2000. While the majority of states either did not use, or made only very limited use of, SSPs prior to 2000, there are a handful of states that did make use of SSPs before 2000. An examination of figure 1 suggests that, at the national level at least, the inclusion of SSP cases in the coverage ratio in 2000 does not produce a disruptive jump in coverage estimates. In order to control for any artificial increase in coverage due to the lack of data on SSP cases prior to 2000, a dummy variable for the year 2000 is included in the 1995–2009 period analyses.3

A second complication results from how state policy makers have re- sponded to additional extensions of TANF requirements contained in the Deficit Reduction Act of 2005 (120 Stat. 4 [2006]). While many states explicitly created SSPs in order to meet TANF work requirements, the Deficit Reduction Act reduces the capacity for states to utilize SSPs for this purpose by requiring states to include SSP cases in their work-par- ticipation calculations as of October 2006. In response, some states have shifted from using SSPs to solely state-funded programs (SSFs), which are not funded with maintenance-of-effort dollars and consequently are not included in states’ work participation calculations (Schott and Parrott 2009). As the programs are completely state funded, there are no federal reporting requirements and no systematic federal data on SSF caseloads.

This is potentially a serious problem for a study of state welfare ad- equacy, given that the creation of SSFs represents a direct effort by states to increase benefit access and that use of SSFs has increased since the onset of the 2007–9 recession. Danilo Trisi and LaDonna Pavetti (2012) collect data on total TANF, SSP, and SSF caseloads directly from state agencies, as opposed to from the USDHHS. These data are on total cases, and it is not possible to distinguish child cases. In order to assess the consequences of excluding child recipients of SSF funds in this study, an estimate of total child cases is generated for each year after 2005. The estimates use the degree of change in total cases in the Trisi and Pavetti (2012) data.4 These estimates are created for the 25 states that implemented SSF programs by 2009 (Schott and Parrott 2009). For the years 2006–9, the correlation is very high (r p .95) between the coverage measure used in the analyses below and the estimate of total coverage using the Trisi and Pavetti (2012) data. Further, the fact that the results of models using either measure are nearly identical (and do not differ in terms of any of the central findings) provides reassurance that the inclusion of SSF recipients would not alter this study’s conclusions.

3. In addition, for the states that made use of SSPs in 2000, estimates of SSP cases between 2000 and 1997 were created using a linear interpolation. The inclusion of these estimated SSP cases in the coverage ratio produces results that are identical to those presented below.

4. Trends in child cases are projected using 2005 child caseload numbers. For example, if total caseloads in a state increase by 5 percent between 2005 and 2006 in the Trisi and Pavetti (2012) data, then 2006 child caseloads are estimated to be 5 percent higher than their level in 2005.

234 Social Service Review

Fig. 3.—Average AFDC and TANF coverage vs. average AFDC, TANF, SSP, and SSF coverage for all states and selected states. AFDC p Aid to Families with Dependent Chil- dren program; TANF p Temporary Assistance for Needy Families program; SSP p separate state programs; SSF p solely state-funded programs. * Inclusion of SSP recipients increases AFDC and TANF coverage measure by 5 percent or more in 14 states: California, Con- necticut, Hawaii, Iowa, Maine, Maryland, Minnesota, Missouri, Nebraska, New York, Rhode Island, Vermont, Virginia, and Washington.

Figure 3 illustrates the contribution of state use of SSPs and SSFs to national coverage rates by comparing average state AFDC and TANF coverage to a measure of coverage that includes SSP and estimated SSF child cases in the numerator. In addition, this figure provides an illus- tration of the portion of coverage attributable to SSP and SSF cases in a subset of 14 states in which the inclusion of SSP recipients increases their coverage ratio by 5 percent or more in any year. Nationally, the use of either SSPs or SSF programs only increases coverage rates marginally. However, for a relatively small number of states, the use of SSPs or SSF programs has allowed these states to cover a significantly larger portion of the poor (e.g., coverage rates increase 10 percentage points or more with the inclusion of SSP recipients in Hawaii, Maine, Minnesota, Ne- braska, New York, Rhode Island, and Virginia).

Coverage in the TANF Era 235

The independent variables in the following analyses may be roughly grouped in four distinct categories: measures of economic factors, polit- ical and racial context, TANF policy content, and administrative practices. Definitions and sources for all variables are listed in table 1.

Economic Factors

One of the most consistent and robust findings from the caseload lit- erature is the crucial role that strong economic growth played in the reduction of caseloads during the late 1990s (Council of Economic Advisors 1997; Wallace and Blank 1999; Ziliak et al. 2000; Blank 2001). The composition of the coverage variable partially controls for caseload fluctuations driven by changes in the state of the local economy. How- ever, the authors expect that unemployment will still exert considerable influence on coverage, as high unemployment may increase the depth of poverty for poor families or push the working poor out of the labor market. These conditions are expected to increase application for and receipt of benefits but would be poorly captured by the poverty measure in the coverage ratio. Further, the authors expect that state welfare offices may fluctuate between leniency during economic downturns and more stringent approaches when unemployment is very low. As such, high unemployment is expected to increase coverage as both need and application for, and possibly receipt of, benefits increase.

This analysis examines a number of additional economic factors, in- cluding the female employment-to-population ratio, real per-capita in- come, real per-capita revenue, and average earnings in low-wage occu- pations. As low-income women increasingly enter the workforce, either pulled by the strong economy or pushed by welfare reform, the authors expect that a higher female employment-to-population ratio will either result in higher coverage rates as the size of the population in poverty (the denominator in the coverage ratio) decreases or will have no influ- ence on coverage as both cases and poverty decline simultaneously. It should be noted that this factor in particular is potentially endogenous given that TANF policies may affect both welfare coverage and the em- ployment-to-population ratio. There is also a possibility of reciprocal cau- sation between these two factors. To address the latter issue, the female employment-to-population ratio is lagged by one year.5 This issue of en- dogeneity is addressed in greater detail in the discussion of the modeling approaches below.

Following the literature on welfare benefit generosity, wealthier states, measured by real per-capita income and real per-capita revenue, are

5. The former issue is more difficult to address. Duncan and Raudenbush (1999) suggest that one way to deal with endogeneity in the context of multilevel models is to control, if possible, for the relevant omitted factor. In the analyses below, variables are introduced that characterize various TANF policy characteristics expected to affect welfare coverage.