ACF OPRE: TANF Caseload Composition and Leavers

III. Understanding Changes in the Circumstances of TANF Recipients

One of the major changes in welfare since reform is the unprecedented decline in the size of the caseload. The caseload fell by half between 1996 and 2000 and has continued to fall at a slower rate since, from 2.3 million families in 2000 to 1.9 million in the first part of 2006 (figure 1).23 The dramatic reduction in the size of the caseload in the early period of reform has implications for understanding changes in outcomes over time. In this section, we address four sets of questions that relate to the broad questions of how the characteristics and outcomes of welfare recipients and former recipients have changed over time. The four questions addressed are

In what ways have the characteristics of families receiving cash assistance changed over time, and what do we know about the relationship of these changes to caseload decline? To what extent have caseloads declined because needy families are moving off the program faster (increased exit rates) or because fewer families are taking up benefits (decreased entry rates)?
Does the caseload of TANF recipients include greater percentages of families with serious barriers to work over time? Are the most "able" or "work-ready" recipients the most likely to leave TANF? Has the relationship between barriers and work changed over time?
What do we know about the economic progress of TANF recipients and leavers over time? Are employment rates of TANF recipients continuing to rise in the later years of reform? Are families better off after leaving TANF than they were on welfare? Are those leaving TANF working steadily and progressing toward self-sufficiency so that they can no longer require government supports (such as food stamps and housing assistance, for example) to meet their expenses?
What do we know about those leaving welfare without work or advantageous changes in family structure? Are there changes in the reasons families are leaving welfare over time, (e.g., leaving for better opportunities through work, marriage, or other positive changes in living arrangements) as opposed to reaching time limits, sanctioned for failing to meet program requirements, or finding program requirements burdensome? What do we know about groups of families that leave welfare but are not working?

[D]

We address these questions by presenting a synthesis of relevant existing research that helps us to understand and interpret the changes in data over time supplemented with evidence from our own analysis of three national data sets: the National Survey of America’s Families (NSAF), the Survey of Income and Program Participation (SIPP), and the Current Population Survey (CPS). Before addressing these questions, we discuss data issues for our analyses of the three national data sets, including sample definitions and timing of comparisons.

Data Issues for Analysis of Three Data Sets

To present a clearer picture of how the welfare caseload and the status of welfare leavers have changed over the ten years following federal welfare reform and to see how the choice of data sets and specific years studied may influence one’s conclusions, we present our own tabulations from three nationally representative data sets: the NSAF, the SIPP, and the CPS. We attempt to use as similar sample definitions, variable measures, and time periods as possible across the three data sets. However, there are some differences in our sample definitions (unit of analysis, definition of welfare recipient and leaver) due to the idiosyncratic way each data source collects information, and there are differences in the calendar years of data available across the three data sets.24 For this reason, we emphasize the results over time within each data set, rather than the findings across data sets. We do, however, discuss differences in the trends within data sets over time and potential reasons for differences we find across data sets.

Below, we provide a complete discussion of the definitions we use for our analysis from each data set (summarized in table 1). We then discuss the time periods we use for our analysis from each data source (also summarized in table 1) and the implications of these differences for interpreting results.

NSAF. When weighted, the NSAF survey provides information on a nationally representative sample of nonelderly households with children. The NSAF was fielded in three largely independent rounds (1997, 1999, and 2002) and surveyed approximately 40,000 households in each round.25 The NSAF collects a broad set of information on the income, earnings, living arrangements, and well-being of families. With large samples of low-income families, it is an excellent source of information for studying welfare and welfare reform.

Our unit of analysis is a “social family”—this includes children and adults living together who are related by blood, marriage, or romantic attachment. As such, cohabiting partners are counted as social family members. This is the unit of analysis that is used as the basic unit in the NSAF survey (including collection of information about previous year’s income). The interviews are conducted with the adult most knowledgeable about the children (specifically the focal child or children) in the family. This is most commonly the child’s mother and is referred to as the MKA (most knowledgeable adult).

Table 1 - Analysis Sample Definitions for NSAF, SIPP, CPS
	NSAF	SIPP	CPS
Data Type, Years	cross-sectional: 1997, 1999, 2002	longitudinal: 1996 and 2001 panels	cross-sectional: 2000 and 2005
Unit of Analysis	"Social family" - children and adults living together related by blood, marriage, or romantic attachment (includes cohabiting partners). Adult most knowledgeable adult about the child is considered head.	"Family" - all persons related by blood, marriage, or adoption; excludes cohabitors. Head is designated parent or guardian of at least one of the children receiving welfare. In married couple families, head is mother; in multiple generation families youngest parent/guardian is head.	"Family" - all persons related by blood, marriage or adoption; excludes cohabitors. Head is designated parent or guardian of at least one of the children receiving welfare. In married couple families, head is mother; in multiple generation families youngest parent/guardian is head.
Definition of Recipient	Head reports family receiving TANF income at the time of the interview. We exclude families where parents of the children are not present or the head is receiving SSI. Family income at the time of the interview or in the prior year cannot have exceeded 250 percent of poverty and must be below 200 percent of poverty in current or prior year.	Families with children that report receiving TANF, general assistance, or "other welfare".	At least one person in family reports receiving TANF or "other public assistance" in the past 30 days.
Definition of Former Recipient	Head reports not receiving TANF at time of interview but did receive TANF in the past two years. Same sample exclusions as for TANF recipients.	Families receiving welfare in first wave of panel and stopped receiving welfare (no one in family received) for at least two consecutive months.	Families reporting received welfare in previous calendar year but has not received welfare in past calendar month.
Time periods analyzed for current recipients	early reform: 1997-1999 late reform: 1999-2002	early reform: 1996-2001 late reform: 1999-2003	late reform: 2000-2005
Time period analyzed for former recipients	1997-2002	1996-2001	2000-2005

We exclude from our sample families where the parents of the children in the family are not present and where the MKA is receiving Supplemental Security Income (SSI). These restrictions are used to limit our sample to families where adults are most likely to be affected by welfare work policies.

We define recipients as families that report receiving TANF income at the time of the interview. We classify the MKA as the head of the assistance unit. Recent welfare leavers are those families that do not receive TANF at the time of the interview but report having left TANF at some point over the last two years. We also use an income test to define our NSAF samples of welfare recipients and leavers. A family’s income cannot exceed 250 percent of poverty in either the past or current year and must fall below 200 percent of the poverty line in one of the years. The income screen eliminates about 3 percent of current welfare recipients and 12 percent of recent welfare recipients in 1997. This exclusion has little impact on the trends in outcomes over time (see Loprest and Zedlewski 2006).26

Standard errors for the NSAF are computed using a jack-knife technique to account for the NSAF’s complex sample design.27

As described earlier, the NSAF suffers from the standard survey data issues of survey nonresponse (addressed through ex-post weighting), item nonresponse (addressed through imputation), and under-reporting of income and welfare receipt. In addition to these potential problems discussed above, research on the 1999 NSAF data suggests that the joint distribution of welfare receipt with respect to race “may be inconsistent with joint distributions of roughly comparable variables in other data sources such as the CPS and administrative data.” Therefore, we do not report results for race for 1999 in the NSAF.28, 29

SIPP. The SIPP is a series of national panels with sample size ranging from approximately 14,000 to 36,700 interviewed households. The duration of each panel ranges from 2 1/2 years to 4 years. The core SIPP data track the employment, income, and program participation of families over time, interviewing them every four months, ascertaining information about the previous four months. Each four-month period is referred to as a wave. The coverage of low-wage individuals is quite good and is one of the primary strengths of this data source. The data are nationally representative in the first wave of data collection and through attrition and changes in the correspondence of the sampling frame to the population become less representative over time.

We use data from the 1996 and 2001 panels of the SIPP. The 1996 panel is 12 waves (four years) long; the 2001 panel is 9 waves (three years) long. Our unit of analysis in the SIPP data is the family and includes all persons related by blood, marriage, or adoption; unrelated cohabitors are excluded. Unlike the NSAF, the SIPP does not ascertain whether an adult unrelated to the children present is the current romantic partner (i.e., the cohabiting partner) of the children’s parent, so we cannot easily separate cohabiting partners from unrelated other adults.

Families are considered to be on welfare if they have children and report receiving TANF, general assistance, or “other welfare.” The family head is the designated parent or guardian of at least one of the children receiving welfare. In married couple families, we designate the mother as the head;30 in multiple parent/guardian families (e.g., a multigeneration family with a grandmother, adult child, grandchild, and the grandmother’s own minor child) the younger(est) parent/guardian is considered the head. In the first eight waves of the 1996 SIPP, if a child has welfare income, the parent/guardian is coded by the SIPP as receiving welfare. In later waves of the 1996 and in the 2001 SIPP, the parent is not coded as a welfare recipient. To keep comparability across the two panels of SIPP data, we include families where anyone in the family is receiving welfare.

Welfare leavers are families that were on welfare in the first wave of the SIPP and then stopped receiving welfare (i.e., no one in the family received welfare) for at least two consecutive months. Characteristics are measured as of the month of exit. In order to track welfare leavers over time, we restrict our sample to those who left welfare during waves two through four of each panel of the SIPP. This ensures that we can follow a substantial sample of leavers for at least 12 months (subject to sample attrition) in the SIPP data, allowing us ample time to observe circumstances after exit.

In the SIPP, our data for any given calendar year reflect the average monthly caseload characteristics of families receiving TANF in any month during that year. To assess which changes in the characteristics of TANF recipients are statistically significant, we only make comparisons between panels (these can be considered independent cross sections), not between waves or interviews within one panel. For example, we do not compare results for TANF recipients in 1996 to results for recipients in 1999. The 1999 results would be for the same group of recipients, later on in the same panel and therefore could not be considered a completely separate sample. This would also be comparing a panel without attrition (early waves) to a group with attrition (in later waves).31

We use a jack-knife technique to account for the complex sample design when computing standard errors. This same process is used for computing standard errors for leavers.32^,33

CPS. The Current Population Survey (CPS) is a monthly survey of about 50,000 households conducted by the Bureau of the Census for the Bureau of Labor Statistics. The CPS, when weighted, is representative of the civilian noninstitutionalized population and is the government’s main source of information about the labor force characteristics of the U.S. population. The Annual Social and Economic Supplement (ASEC) to the core CPS is fielded every March and collects detailed information about family income from the prior calendar year. Official estimates of poverty are based on these data.

Our unit of analysis is the family and includes all persons related by blood, marriage, or adoption residing in the household.34 As with the SIPP, we designate the mother as the family head in married couple families.35

Beginning in March 2000, the CPS added questions about the receipt of cash assistance in the past 30 days. The basic CPS data files released to the public do not contain this information; however, it is available upon request from the U.S. Census Bureau. A family is considered to be a welfare recipient if at least one person in the family reports receiving TANF or “other public assistance” in the past 30 days.36 This definition of welfare recipients varies from most past welfare literature based on the CPS that uses receipt of welfare in the past year to define recipients. Using the new questions allows for creation of a point-in-time sample of welfare recipients more comparable to the SIPP and NSAF data.

CPS information on welfare receipt in the past 30 days is used in conjunction with the ASEC supplement to identify welfare leavers. A welfare leaver is someone who reports receiving welfare in the past calendar year but has not received welfare in the past calendar month. Standard errors are estimated using conventional corrections for weighted data.37^, 38

Challenges making comparisons across data sets. It is very hard to use existing research to draw out a clear consistent picture of how the status and well-being of current and former recipients has changed in the TANF era because the existing research uses noncomparable definitions of recipients and leavers and their characteristics and studies different time periods. Even when a single research team endeavors to standardize definitions and time periods across multiple data sets, as we do here, many differences remain, and these differences can affect the levels and trends observed across the data. Below, we discuss specific differences in our approaches to analyzing the data from the NSAF, SIPP, and CPS and how these differences may influence comparisons across them.

Time period analyzed. Data for the same sets of years are not available in all three data sets. We examine changes in the characteristics and outcomes of welfare recipients for the early welfare reform period (roughly between 1996 and 2001) and then for the late reform period (roughly 1999 through 2005). In the NSAF, we assess changes in the characteristics of TANF recipients between 1997 and 1999 (the early reform period) and between 1999 and 2002 (the late reform period). As described above, in SIPP, we only make comparisons across the two panels, not over time within the same panel. Thus, we compare TANF recipients in 1996 to TANF recipients in 2001 to examine changes in the caseload during the early reform period, and we compare TANF recipients in 1999 and 2003 to examine changes during the late reform period. The CPS can only be used to study welfare recipients and leavers (using the above definitions) from 2000 forward because the CPS did not ask about welfare receipt in the month before the interview in earlier survey years. We therefore analyze changes in the characteristics of current recipients between 2000 and 2005 (the most recent CPS year available at the time of this study). These time periods are shown graphically in figure 2.

To assess changes in the characteristics and outcome of welfare leavers, we are even more restricted by data availability and the need to observe families over time. In the SIPP, we can only compare 1996 and 2001 because leavers identified in the later years of any given panel are not followed for enough subsequent months to support this type of analysis. To use a comparable time period for the NSAF, we show changes over the period from 1997 through 2002. For the CPS, we again show results for 2000 and 2005.

Although one would expect some differences in findings simply because the years analyzed differ, it is important to note how the economic slow-down of 2000–2002 can influence our findings. The early reform period captured by the NSAF is 1997 to 1999, and the economy was quite strong in 1999. The early reform period in the SIPP, however, extends to 2001, when the economy was weaker. As such, one would expect to find more positive trends (in income, for example) during the early years of reform when using NSAF rather than SIPP data. It is also important to remember that when examining a trend over time by looking at changes between two years, there is the possibility that trends within the time period are missed. For example, a flat trend from 1999 to 2003 could mask a decline from 1999 to 2001 and an increase from 2001 to 2003.

Definition of a TANF case. Ideally, one would want to analyze TANF cases, the group of people that a TANF grant is meant to support. In secondary data, it is virtually impossible to consistently identify cases, especially in households in which there may be adults and children who are not supposed to benefit from a TANF grant made to other household members. In our three data sets, our unit of analysis is a "TANF family" as defined above, and our TANF families may include individuals who are not technically in a TANF case.39 The composition of TANF families varies across the three data sets, and this raises issues of comparability discussed below.

Figure 2: Time Periods for TANF Recipient Analysis

[D]

In the NSAF, we focus solely on families with children in which at least one parent is present who is not receiving disability benefits—as such, the welfare case is headed by an adult who generally would be subject to all TANF program rules (e.g., work requirements, sanctions, time limits).40 In the CPS and SIPP, we consider all families with children that receive welfare to be TANF cases. As such, any family with a child that reports receiving welfare income is considered a welfare family, even if the child does not live with his or her parents and the benefits are meant for the child alone. In other words, child-only cases are more likely to be included in our SIPP and CPS samples than in our NSAF samples. We can see this difference in the size of our samples of TANF recipients (table 2). The 1996 SIPP sample represents approximately 3.3 million TANF families, while the 1997 NSAF sample represents only 1.9 million families. In 1999, our SIPP sample of TANF recipients represents 1.7 million families, compared with 1.1 million in the NSAF and 1.4 million in the 2000 CPS.

Adults in child-only cases are not subject to the same program rules (work requirements, sanctions, time limits) as parents in TANF families. Further, the adults’ income in a nonparent, child-only household may not be deemed available to the child. Consequently, the heads of TANF families in the SIPP and CPS data may have stronger human capital characteristics and higher incomes than TANF families in the NSAF.

Table 2 - Weighted Sample Sizes of All Datasets
	NSAF		SIPP		NSAF		SIPP		CPS
	EARLY PERIOD				LATER PERIOD
	1997	1999	1996	2001	1999	2002	1999	2003	2000	2005
Stayers (in millions)	1.85	1.11	3.25	1.51	1.11	0.84	1.74	1.28	1.42	0.94
Leavers (in millions)	1.05	(N/A)	1.04	0.66	(N/A)	0.72	(N/A)	(N/A)	0.95	0.81

Notes:
Unit of Analysis - "TANF Family" as defined below.
NSAF - children and adults living together related by blood, marriage, or romantic attachment (includes cohabiting partners). Adult most knowledgeable about the child is considered head.
SIPP - all persons related by blood, marriage, or adoption; excludes cohabitors. Head is designated parent or guardian of at least one of the children receiving welfare. In married couple families, head is mother; in multiple generation families youngest parent/guardian is head.
CPS - all persons related by blood, marriage, or adoption; excludes cohabitors. Head is designated parent or guardian of at least of the children receiving welfare. In married couple families, head is mother, in multiple generation families youngest parent/guardian is head.

Another difference between the NSAF samples and the samples from the other two data sets for current recipients is that the NSAF samples are restricted to exclude families with the highest incomes.41 Only three percent of current recipient families are excluded by this criterion, so it is likely to have only minor impacts on differences between the NSAF, CPS, and SIPP results for recipients.

Finally, while the NSAF income cutoff and the greater inclusion of child-only cases in the CPS and SIPP samples may tend to drive up incomes in the CPS and SIPP samples relative to the NSAF samples, another difference has the opposite effect. The NSAF samples capture cohabiting partners who are not related to the parent’s child. These unrelated individuals are excluded from the SIPP and CPS samples. A priori, it is impossible to know the net effect of these offsetting differences; however, given that relatively few TANF families have unrelated cohabiting partners and the ones that do tend to have low incomes, we suspect that average incomes measured in the SIPP and CPS samples will tend to be higher than those measured in the NSAF samples.

Definition of a welfare leaver. In the NSAF and the CPS samples, welfare leavers are identified based on those who are currently not receiving welfare but report receipt at some point in the past (for CPS in the prior calendar year and for NSAF at some point in the two years prior to the interview). As such, families that have remained off welfare for over a year will be included as welfare leavers in both these samples. In the SIPP, because participation is reported down to the month, we can capture leavers in the month they leave and measure their characteristics at the time of exit. Consequently, the NSAF and CPS leavers will disproportionately represent “successful” leavers—those who are able to remain off TANF for extended periods of time—while the SIPP leavers will include families that may transition back onto TANF in a short period of time. Thus, we would expect the work experience, educational attainment, wages, incomes, and perhaps other characteristics to look better among the NSAF and CPS leavers than among the SIPP leavers.42

Other differences in the samples discussed above remain. The different criterion across data sets for which leaver families to include remains the same. However, for the samples of leavers, this difference should be much smaller. Because child-only cases are generally less likely to exit welfare than other cases (e.g., they are generally not subject to time limits or work requirements and related sanctions), the samples of leavers in the SIPP and CPS data should include fewer child-only cases.

Finally, the difference in income-based exclusions and inclusion of cohabiting partners in the NSAF remain for the leaver sample. The income screens have a larger impact on TANF leavers than recipients, excluding 12 percent of all leaver families in the NSAF in 1997. This means very successful leavers are excluded from the sample, leading to lower absolute mean income and wage measures for leavers in the NSAF relative to the SIPP and CPS. The effect on medians is likely to be lower, and the impact on trends over time in the NSAF is limited. Again, inclusion of cohabiting partners would tend to increase income in the NSAF relative to the SIPP and CPS.

These differences in leaver definition and sample criteria work in opposite directions in their implications for the relative size of samples across the data sets (table 2). The broader definition of leaver in the NSAF and CPS would tend to increase relative sample sizes while the sample exclusions are likely to lower NSAF leaver sample sizes relative to the SIPP and CPS. In 1996, the SIPP sample represents approximately 1.1 million families, while the NSAF represents 1.2 million leavers in 1997. In 2001, the SIPP sample represents approximately 670,000 leavers, compared with 750,000 in the NSAF in 1999 and 950,000 in the CPS in 2000.

Given all of these differences in definitions and time periods across the three data sets, we focus on making comparisons over time within data sets and note the potential influence of the differences discussed here when making comparisons between data sets.

Question 1: In what ways have the characteristics of families receiving cash assistance changed over time, and what do we know about the relationship of these changes to caseload decline?

The welfare caseload is not a constant population—people move on and off the welfare rolls all the time. For the size of the caseload to fall, either fewer people must be entering the program or those who have entered the program must be exiting after shorter stays (or some combination of both phenomena). Ultimately, changes in who enters welfare and how long they stay influence the composition of the caseload and, as a result, the characteristics of welfare leavers.

For example, if women with high school degrees stop entering welfare, it would not take long before the average educational attainment of women on welfare started to fall. In due time, the vast majority of women leaving welfare would have less than a high school education (because they were the only ones coming on), and so the average educational attainment of leavers would also fall. Further, because less educated women have poorer labor market prospects than more educated women, it would not be surprising to find that the average wages of welfare leavers also declined. As such, changes in welfare dynamics can change the composition of current and former recipients and this in turn can influence the average status and outcomes for these groups. Below, we assess research on changes in welfare entry and exit (dynamics) under TANF and examine its implications for current and former welfare recipients.

Most of the literature related to caseload dynamics has focused on determining the extent to which the initial decline in caseload size was due to welfare reform, opposed to the strong economy of the late 1990s. The first generation of these studies focused on the aggregate caseload—that is, the number of families on welfare at different points in time. Blank (2002) and Grogger, Karoly, and Klerman (2002) both provide extensive reviews of these studies, which generally find that both welfare reform and the economy played significant roles in the decline of the welfare caseload, with the weight on the importance of each varying across studies. Among studies that focus on the TANF program that do not try to isolate the effects of any particular program provision and find significant caseload reduction effects, the estimated size of the effects ranges from 18 to 34 percent (Grogger, Karoly, and Klerman 2002). These studies have little to say about the implications of caseload decline for caseload composition.

A second group of studies make the important contribution of decomposing caseload decline into changes in entry and exit. They point out the importance of decreases in entry to welfare as a reason for caseload decline in addition to increased exits (see Klerman and Haider 2001; Oellerich 2001; Bavier 2002; Acs et al. 2001).43 Most of these studies find that changes in entry were an important part of caseload decline, although increases in exits from welfare played a relatively larger role. For example, Grogger, Haider, and Klerman (2003) find that reduced entry was responsible for 39 percent of caseload decline from 1986 to 1999 using data from the SIPP. Mueser et al. (2000) use administrative data from five urban areas and find that about one-third of caseload decline in the mid-1990s was due to reduced entry and two-thirds was due to increased exits.

Several studies examine the impacts of welfare reform and the economy on entry and exit separately with varying results. For the most part, studies suggest that welfare reform had a significant impact on exits (Hofferth et al. 2001; Grogger 2004; Moffit and Winder 2003). The impact of reform on entry is less clear. Several studies find weak or no effects of waiver policy on caseload entry and returns to welfare (Ribar 2005; Gittleman 2001; Hofferth, Stanhope, and Harris 2005). However, Acs, Ross Phillips, and Nelson (2005); Bavier (2002); Grogger (2004); and Moffit and Winder (2003) find that TANF policies influenced entry.

One reason that studies of welfare reform find such mixed results is that the packages of policies that make up “reform” vary from state to state, and it can be difficult to measure these policies and disentangle their effects. In addition, states may adopt policies that have contradictory effects. For example, Moffitt (1996) uses a microsimulation model to demonstrate that mandatory employment and training programs could reduce welfare entry rates while voluntary programs might increase them in the long run. Examining state policies pursued during the early 1990s prior to federal reform, when states were beginning to implement specific welfare reform policies, Hofferth, Stanhope, and Harris (2005) find that generous earnings disregards reduce the probability that a woman leaves welfare and extend the average welfare spell length, while state requirements that mothers of very young children work increase the probability that a woman leaves welfare.

Whether reforms affect exit and entry differentially for families with different characteristics has implications for the caseload. For example, if more work-ready individuals are less likely to enter welfare after reform, then, all else equal, this would change the composition of the caseload over time. Most of these studies do not specifically address this issue.44

Our analysis of the changes in demographic and family characteristics of TANF recipients and leavers over time provides some evidence on the changing composition of the caseload.

Personal and Family Characteristics of Recipients. During the earliest years of welfare reform, 1997 to 1999, the characteristics of recipients remained fairly stable according to tabulations from the NSAF (table 3). Opening up a wider window on the early reform period—1996 to 2001—using the SIPP, shows evidence of a caseload that has been changing over time. For example, there is no significant change in the age profile of recipient family heads between 1997 and 1999, but between 1996 and 2001, the caseload has become somewhat older. According to SIPP data the average age of recipients increased from 33.4 to 35.0 years and the share over 35 increased by 6.6 percentage points. In general, the age of TANF family heads is slightly higher in the SIPP than in the NSAF, although the differences between the two are quite small when comparing the earliest years (1996 for SIPP to 1997 for NSAF).45 Other researchers report that the average age of TANF family heads is in the early 30s during the early TANF period (e.g., Grogger 2004 and Kim 2000). Administrative data indicate that the mean age of adults receiving TANF is about 31 years during the late 1990s.46

Table 3 - Personal Characteristics of Welfare Recipients (Family Head) - Early and Late Reform Periods
		NSAF			SIPP			NSAF			SIPP			CPS
		EARLY PERIOD						LATER PERIOD
		1997	1999	Change	1996	2001	Change	1999	2002	change	1999	2003	Change	2000	2005	Change
Age (%)	<25	24.0	25.8	1.9	22.7	21.3	-1.4	25.8	28.4	2.6	17.8	19.8	2.1	27.2	24.9	-2.4
	25-34	44.1	44.7	0.6	38.0	32.8	-5.2 *	44.7	36.1	-8.6 *	33.2	34.8	1.6	34.4	32.5	-1.9
	35+	32.0	29.5	-2.5	39.3	45.9	6.6 *	29.5	35.5	6.0	49.0	45.3	-3.7	38.3	42.6	4.3
	Mean (in years)	31.3	30.8	-0.4	33.4	35.0	1.6 *	30.8	30.9	0.0	35.8	35.4	-0.4	33.1	34.3	1.2 *
Education (%)	<12th grade	39.8	46.1	6.4	38.9	43.3	4.4 *	46.1	41.5	-4.7	42.0	40.5	-1.5	40.6	41.3	0.7
	12th, HS dip, GED	36.9	31.8	-5.2	37.6	32.2	-5.4 *	31.8	39.2	7.4	37.2	33.2	-3.9	40.3	35.4	-5.0 *
	Some college, AA, Voc Tech	20.7	17.7	-3.0	21.3	21.7	0.5	17.7	15.8	-1.9	18.7	24.0	5.3 *	19.0	23.0	4.0
	4 year college or more	2.2	2.9	0.7	2.3	2.7	0.5	2.9	2.9	0.1	2.1	2.2	0.1	0.0	0.3	0.3 *
Race/Ethnicity (%)	White, other Non-Hispanic	43.5	N/A		44.5	41.9	-2.7	N/A	33.8		35.5	42.1	6.6 *	39.7	43.3	3.6
	Black Non-Hispanic	32.3	N/A		35.1	32.1	-3.0	N/A	35.4		36.5	32.2	-4.3	36.0	30.3	-5.6
	Hispanic	24.3	N/A		20.4	26.0	5.7 *	N/A	30.8		28.0	25.7	-2.3	24.3	26.4	2.1
Gender (%)	Male	4.5	6.3	1.9	3.9	4.8	1.0	6.3	6.9	0.6	4.1	4.4	0.3	4.0	6.0	2.0
Gender (%)	Female	95.6	93.7	-1.8	96.1	95.2	-1.0	93.7	93.1	-0.6	95.9	95.6	-0.3	96.0	94.0	-2.0

* indicates change is significant at the 90 percent confidence level.

Notes:
Unit of Analysis -- "TANF Family" as defined below
NSAF - children and adults living together related by blood, marriage, or romantic attachment (includes cohabiting partners). Adult most knowledgeable about the child is considered head.
SIPP - all persons related by blood, marriage, or adoption; excludes cohabitors. Head is designated parent or guardian of at least one of the children receiving welfare. In married couple families, head is mother; in multiple generation families youngest parent/guardian is head.
CPS - all persons related by blood, marriage, or adoption; excludes cohabitors. Head is designated parent or guardian of at least one of the children receiving welfare. In married couple families, head is mother; in multiple generation families youngest parent/guardian is head.

Next, consider education. The SIPP data show that the education level of recipients declined between 1996 and 2001. The share of TANF families headed by an individual who failed to complete high school rose by 4.4 percentage points over the five-year period. This trend is evident as well between 1997 and 1999 in the NSAF data, although the 6.4 percentage point change is not statistically significant. The distribution of educational attainment is quite similar across the two data sources with a little over 75 percent of TANF case heads having a high school degree or less. This is similar to the findings of other researchers examining data from the same period (Bavier 2001 and Kim 2000). However, administrative data suggest that over 90 percent of adults receiving TANF have a high school degree or less education.

The 1996 SIPP and the 1997 NSAF show very similar racial/ethnic distribution among welfare recipients; however, for the early reform period, information on racial and ethnic trends in the caseload comes only from the SIPP.47 The SIPP shows a substantial rise in the share of the caseload that is Hispanic between 1996 and 2001. Bavier (2001) also uses the SIPP and finds a similar rise in the proportion of the caseload that is Hispanic. The rising share of Hispanics in the caseload can be observed in administrative data as well. Finally, both the SIPP and the NSAF surveys indicate that about 19 out of every 20 adults heading welfare families are women and that this has remained true over time.

The later reform period extends roughly from 1999 to 2005, with our three data sources covering slightly different years within this period. The NSAF data from 1999 to 2002 show that there are fewer recipients in the 25 to 34 age range with a small and insignificant trend toward older recipients. This is similar to the early trend observed in the SIPP data. However, between 1999 and 2003 the SIPP shows no increase in the age of TANF recipients. In contrast, using the CPS and taking the analysis from 2000 through 2005, we find the average age of TANF caseload heads increased by 1.2 years. Again when comparing the base years across the data sets, we find the SIPP caseload to be slightly older than the NSAF, and estimates from the CPS fall between the two.

In contrast to the earlier years of welfare reform where we find some evidence that the education levels of TANF caseload heads declined, data from the later reform years indicate that caseload heads are becoming more educated. Although the NSAF shows no significant changes between 1999 and 2002, the SIPP finds that the share of heads with some college increased by 5.3 percentage points (from 18.7 to 24.0) between 1999 and 2003 and the CPS shows a significant drop in the share with only a high school degree, a notable but not significant rise in those attending college, and a small but significant increase in the share with college degrees between 2000 and 2005. The base year education distributions are quite similar across the data sets.

The SIPP and CPS data sets show slightly different trends in the racial/ethnic composition of TANF caseload heads in the later period. The SIPP shows that during the late reform period (1999-2003), the caseload became increasingly white, while the CPS shows virtually no change in the racial/ethnic composition of the caseload between 2000 and 2005. Other than differences in the years considered, there is no clear reason why the trends should differ across the data sets. Nevertheless, the 1999 SIPP and 2000 CPS race/ethnic distributions are quite similar. Finally, caseload heads continue to be overwhelmingly female.

Taken together, there is some evidence that TANF caseload heads have become slightly older on average over time, with the trend beginning towards the end of the early reform period and continuing through the late reform period. Trends in education are quite interesting. During the early reform period, education levels appeared to drop, but this trend reversed itself during the late reform period.48 The gender profile remains overwhelmingly female. And while there have been shifts in the racial/ethnic composition of the caseload, the data reveal no consistent trends.

Data from the NSAF and the SIPP show different trends in family composition during the early reform period (table 4). Between 1997 and 1999, the NSAF indicates a substantial 9.1 percentage point rise in the share of TANF cases made up of single parents living with other adults. The SIPP shows no such trend between 1996 and 2001. Further, in the base years (1996 for the NSAF and 1997 for the SIPP), the SIPP data indicate that 20.1 percent of TANF families are married couple families while the NSAF indicates that only 14.6 percent are married. No doubt, this discrepancy in levels influences the observed trends. Further, the discrepancies likely reflect differences in the way surveys define families and identify other adults in the household. In particular, the NSAF uses a “social family” concept that captures a broader array of adults in the household (such as unrelated cohabiting romantic partners) than the CPS and SIPP family concept. Thus, we would expect to see a higher percentage of TANF families classified as single, living with other adults in the NSAF than in the SIPP. And recall that the SIPP data include children living in families without their parents—the adults heading these families in which the child receives welfare are not necessarily in the TANF case even though we consider them to be the head of the welfare family. Consequently, one would expect to see a higher proportion of married TANF families in the SIPP data than in the NSAF data.49 Nevertheless, our estimates are broadly similar to the range of results found by other researchers examining the same historic era. For example, Kim (2000) uses CPS data and finds that 18.6 percent are married using the CPS, and administrative data indicate that 16.2 percent are married in 1997. Because of the detailed household roster available in the NSAF in which all family relationships are considered, the NSAF data may be more useful for studying issues related to family composition and living arrangements than the SIPP or the CPS.50

Table 4 - Family Characteristics of Welfare Recipients - Early and Late Reform Periods
		NSAF			SIPP			NSAF			SIPP			CPS
		EARLY PERIOD						LATER PERIOD
		1997	1999	Change	1996	2001	Change	1999	2002	change	1999	2003	Change	2000	2005	Change
Family Type (%)	Married	14.6	12.0	-2.6	20.1	20.8	0.7	12.0	11.8	-0.2	17.4	23.1	5.7 *	20.4	17.5	-2.9
	Single, no other adults	54.5	48.2	-6.3	56.0	55.2	-0.8	48.2	46.3	-1.9	52.3	48.8	-3.5	60.1	59.0	-1.1
	Single, living with other adults	30.6	39.8	9.1 *	23.9	24.0	0.1	39.8	41.9	2.2	30.4	28.2	-2.2	19.5	23.6	4.1 *
Family Size					4.0	4.0	0.0				4.2	4.2	0.0	3.9	3.8	-0.1
Number of Children Under 18 (%)	1	23.0	26.6	3.6	32.0	32.0	-0.1	26.6	25.9	-0.7	32.1	31.9	-0.2	32.3	32.9	0.6
	2	34.8	28.7	-6.1	31.1	30.2	-0.9	28.7	28.4	-0.3	29.2	31.1	1.8	31.3	31.4	0.0
	3+	42.3	44.7	2.5	36.9	37.9	1.0	44.7	45.7	0.9	38.7	37.1	-1.6	36.3	35.7	-0.6
Mean Number of Children		2.6	2.6	0.0	2.3	2.3	0.0	2.6	2.6	0.0	2.4	2.4	-0.1	2.3	2.2	-0.1
Age of Youngest Person (%)	<1	15.8	17.5	1.6	13.9	14.9	1.0	17.5	18.3	0.8	9.5	12.2	2.7 *	16.7	17.8	1.1
	1-5	50.6	48.3	-2.3	49.0	42.4	-6.6 *	48.3	48.0	-0.3	41.1	43.9	2.8	44.9	41.4	-3.4
	6-11	22.2	28.7	6.5	24.5	28.3	3.8 *	28.7	21.2	-7.5 *	31.7	28.8	-3.0	24.9	24.4	-0.6
	12+	11.4	5.6	-5.8 *	12.7	14.4	1.8	5.6	12.5	6.9 *	17.6	15.1	-2.5	13.5	16.4	2.9
Mean (in years)		4.7	4.4	-0.3	5.1	5.5	0.4	4.4	4.8	0.4	6.1	5.7	-0.5	5.1	5.4	0.4

* indicates change is significant at the 90 percent confidence level.

Family size and the number of children remain fairly stable over the early reform periods captured by both the NSAF and the SIPP. The age distribution of the youngest child in the base year is fairly similar across the data sets, and both show an increase in the share whose youngest child is elementary school aged (6 to 11). However, while the NSAF finds a decrease in the share with adolescents, the SIPP shows a decrease in the share with young children (ages 1 to 5).

During the later reform period, the NSAF shows no change in the types of families receiving TANF while the SIPP shows an increase in married couple families and the CPS shows an increase in the share of single parents living with other adults (table 4). Again, the base year differences in the family types across the three surveys are quite substantial. These differences may be related to the differences in sample definition. None of the data sets indicate any change in family size or the number of children in TANF families during the late reform period. Finally, the share of TANF families whose youngest child is a teenager increased during the 1999 to 2002 period in the NSAF but not in the 1999 and 2003 period covered by the SIPP nor the 2000 to 2005 period covered by the CPS. In fact, the SIPP shows a moderate but significant 2.7 percentage point increase in the share of TANF families with an infant. It is worth noting that the age of youngest child and the proportion of youngest children who are teens are substantially lower in the 1999 NSAF than in the 1999 SIPP or the 2000 CPS. Again, this could be due to differences in sample definitions.

This review of data across the early and late reform periods highlights the sensitivity of findings about the TANF caseload to the precise years considered, the definitions employed, and the data sets used. This is particularly true for trends in the family structure of TANF families and the age of the youngest child.

The most striking finding may be that, despite the implementation of federal welfare reform, the massive decrease in welfare caseloads, and the very different economic climate during the early and late reform periods, data on the demographic characteristics of families on welfare show few statistically significant changes. Further, there is little consistent evidence of changes over time and across data sets. Perhaps this reflects the fact that major one-time shifts in the composition of the caseload began occurring well before federal reform under state waivers to AFDC. It may also be the case that what differentiates families that came onto welfare and/or stayed on welfare during the reform period and the families that would have come on/stayed on under AFDC but not under TANF is not captured by the measures we use to describe TANF recipients—indeed, the differences may be largely unobservable. Such differences include attitudes towards work and dependence, undiagnosed health issues, and self-esteem.

Personal and Family Characteristics of Leavers. The characteristics of families leaving TANF are inextricably linked to the characteristics of TANF recipients—after all, only the families that actually go on to TANF can become TANF leavers. Data from the NSAF and SIPP reflect the early to mid-reform period while data from the CPS reflect more recent changes between 2000 and 2005. It is important to remember that welfare leavers as measured by the NSAF and CPS are families that received welfare in the past year (or two for the NSAF) but are not receiving welfare at the time they are interviewed. As such they may have been off TANF for several months or even longer. In contrast, welfare leavers in the SIPP are families that have left welfare in the past month.51

The personal characteristics of former welfare recipients have remained quite stable over time for most attributes (table 5). Neither the NSAF nor the CPS show any significant changes in the age distribution of TANF leavers. The SIPP shows a stretching out of the age distribution of welfare leavers, with a significant 6.8 percentage point decline in the share between the ages of 25 and 34 fell and insignificant increases in both the share older and younger. On net, however, the average age of family heads leaving TANF is unchanged. As is the case with current recipients, the heads of NSAF families leaving welfare are somewhat younger than those in the SIPP and CPS; again, this is likely due to differences in the definition of the welfare family unit.52

Table 5 - Personal Characteristics of Former Welfare Recipients (Family Head)
		1997	2002	Change	1996	2001	Change	2000	2005	Change
		NSAF			SIPP			CPS
Age (%)	<25	27.6	26.4	-1.3	23.1	27.5	4.3	23.3	25.8	2.5
	25-34	45.0	45.0	0.0	38.3	31.6	-6.8 *	34.8	34.3	-0.5
	35+	27.4	28.7	1.3	38.5	41.0	2.4	41.9	39.9	-1.9
Mean (in years)		29.8	30.6	0.8	33.2	33.5	0.3	33.3	33.3	0.1
Education (%)	<12th grade	30.1	33.5	3.5	33.5	30.4	-3.1	35.1	33.5	-1.6
	12th, HS dip, GED	39.3	41.8	2.5	39.3	39.1	-0.1	39.4	37.3	-2.0
	Some college, AA, Voc Tech	23.9	22.6	-1.3	24.6	27.8	3.2	25.1	28.2	3.2
	4 year college or more	6.1	1.2	-4.9 *	2.7	2.7	0.1	0.5	1.0	0.5
Race/Ethnicity (%)	White, other Non-Hispanic	54.5	49.4	-5.1	55.0	48.4	-6.6	45.3	44.5	-0.8
	Black Non-Hispanic	28.6	35.5	6.9	29.1	32.1	2.9	32.4	36.7	4.3
	Hispanic	16.9	15.1	-1.8	15.9	19.6	3.7	22.3	18.8	-3.5
Gender (%)	Male	5.3	9.1	3.9	7.7	6.6	-1.0	3.8	3.5	-0.3
Gender (%)	Female	94.7	90.9	-3.9	92.3	93.4	1.0	96.2	96.5	0.3

* indicates change is significant at the 90 percent confidence level.

Next, considering education, only the NSAF data show any significant changes in the educational attainment among the heads of welfare leaving families. The NSAF indicates that the share of leavers with four or more years of college dropped from 6.1 to 1.2 percent between 1997 and 2002. Given that the share of TANF family heads with four or more years of college hovered between 2 and 3 percent between 1997 and 2002, this suggests that the most educated welfare recipients under AFDC were the first to leave TANF in its early years. However, we do not observe this change in the other data sets.

Finally, none of three data sets show statistically significant changes in the race/ethnic or gender composition of welfare leavers, and the distribution of leavers across race/ethnic groups and gender categories is fairly similar across the data sets.

There is also little consistent evidence that the family characteristics of welfare leavers have changed over time (table 6). For example, the SIPP data indicate that, between 1996 and 2001, the share of leavers that are single parent families living with other adults increased by 6.7 percentage points while the share that are married decreased by 6.3 percentage points. Neither the CPS nor the NSAF show evidence of this trend. Although none of the data sets shows significant trends in family size or the number of children, the CPS data show an increase in leavers whose youngest child is between the ages of 1 and 5, offset by a decrease in leavers with youngest children ages 6 to 11. The mean age of the youngest child in welfare leaving families, however, is unchanged.

Table 6 - Family Characteristics of Former Welfare Recipient
		1997	2002	Change	1996	2001	Change	2000	2005	Change
		NSAF			SIPP			CPS
Family Type (%)	Married	28.4	26.6	-1.8	31.8	25.5	-6.3 *	25.2	24.5	-0.6
	Single, no other adults	41.9	43.9	2.0	50.3	49.9	-0.4	54.0	53.7	-0.2
	Single, living with other adults	29.7	29.5	-0.2	17.9	24.6	6.7 *	20.9	21.7	0.9
Family Size					3.9	4.0	0.1	3.9	3.8	-0.1
Number of Children Under 18 (%)	1	27.8	26.8	-1.1	35.2	33.9	-1.3	36.1	34.6	-1.5
	2	37.1	33.3	-3.9	30.9	31.5	0.6	31.1	32.1	1.0
	3+	35.1	40.0	4.9	31.5	32.1	0.6	32.8	33.3	0.4
Mean Number of Children		2.3	2.5	0.2	2.1	2.2	0.1	2.2	2.2	0.0
Age of Youngest Person (%)	<1	11.1	14.4	3.3	13.7	12.5	-1.2	13.8	14.1	0.3
	1-5^	53.0	48.5	-4.5	42.3	45.1	2.7	36.9	45.0	8.1 *
	6-11^	28.0	27.5	-0.4	25.4	24.7	-0.7	31.8	23.6	-8.2 *
	12+	7.9	9.6	1.7	18.6	17.9	-0.8	17.5	17.4	-0.1
Mean (in years)					6.3	5.8	-0.5	6.1	5.6	-0.5

* indicates change is significant at the 90 percent confidence level.
^ For the NSAF, the age categories are 1-4 and 5-11.

Overall, data on different cohorts of welfare leavers over time indicate that in most respects, the personal and family characteristics of leavers are fairly stable. Given that we find few notably significant shifts in the characteristics of families receiving welfare, perhaps the stability in demographic characteristics of leavers over time is not surprising.

Question 2: Does the caseload of TANF recipients include greater percentages of families with serious barriers to work over time?

Since the first discussions of welfare reform, questions have been raised about “hard-to-employ” recipients or those who face multiple barriers to work. Debates continue about whether and to what extent these recipients can find work, leave welfare, and become self-sufficient. Related concerns were expressed as to how these families would meet work requirements and whether they would face sanctions and time limits that would ultimately lead them to exit welfare without finding work. Another concern was that the caseload would become more disadvantaged over time, requiring more intensive services and resources. Although the law allows states to exempt up to 20 percent of the caseload from the time limit, initially some argued that a greater percentage of the caseload would be unable to meet work requirements.

Prevalence of Barriers among TANF Recipients. Numerous studies have documented the prevalence of barriers to work among welfare recipients, both before and after welfare reform. Some of the barriers measured include physical health, mental health, domestic violence, substance abuse, criminal history, education levels, and work history. The studies with the most detailed set of measures are based on data from state or local areas that conducted surveys of recipients. 53 Results on the prevalence of barriers among TANF recipients from 12 location-based studies are reported in table 7. The specific definition of a barrier can vary across studies. However, six of these state and local area studies were sponsored by the U.S. Department of Health and Human Services (DHHS) and used the same survey instrument to measure barriers.54 Similar measures were also used by the WES in Michigan and Nebraska, so eight of these studies have almost identical measures.55

Table 7 - Barriers to Work Among TANF Recipients in Location-Based Studies
Barrier	Alameda County, CA 1999	New Jersey 1999	Three Cities 1999	Milwaukee, WI 1999	Michigan WES 1997	Nebraska 2000	Illinois 2001	Colorado 2002	Wash. DC 2002	Maryland 2002	Missouri 2002	South Carolina 2002
Less than a HS diploma/GED	38	45	39	57	31	22	44	33	38	42	40	38
Low work experience		51			15	21	24	22	27	21	21	18
Substance Abuse	6	5	--	5	5	17	3	6	3	5	4	1
Mental Health problem	14	11	11	52	35	17	25	41	21	22	36	29
Physical Health problem	32	36	27	21	19	12	21	26	16	29	20	22
Domestic Violence	6	7	--	14	15	13	13	21	15	15	13	15
Caring for child w/ special needs	23	--	--	9	22	29	30	36	26	29	28	27
Criminal record	9	--	--	--	--	10	8	17	7	14	14	10

Shaded studies all use the same definition and survey instrument. The Michigan and Nebraska studies use very similar survey instruments as well.
Definitions of barriers are as follows with exceptions noted below:
      Low work experience is defined as working less than 50 percent of the years since age 18.
      Physical health problem includes those who self-report fair or poor health and score in the lowest age-specific quartile of physical functioning.
      Mental health problem includes those who experienced major depression in past year or experienced serious psychological distress in past 30 days.
      Domestic violence includes those who reported experiencing severe physical domestic violence in the past year.
      Substance abuse includes those likely to be "chemically dependent" according to the CIDI-SF for substance dependence.
      Criminal record is self-reported by respondents.

Study data sources and exceptions to above definitions:
Alameda County, CA : Dasiger et al (2002). Physical health is reports of fair to poor health. Mental health is defined as depression in last 7 days.
Substance abuse is alcohol dependence. Daily use of illegal substance is similar (7 percent).
Criminal record is involvement with criminal justice system in last 90 days or arrested in past year.
New Jersey : Rangarajan and Wood (1999).
Three Cities study : Moffitt and Cherlin (2002). Physical health is poor or fair health. Mental health is depression score above clinical cutoff.
Milwaukee, WI : Courtney and Dworsky (2006). Physical health is poor or fair health. Mental health problem is those who scored in clinical range on CESD.
Substance abuse is problem with alcohol or drugs during past year. Caring for child with special needs includes are for other family member with disability.
WES : Danziger et al (2000). Low work experience is defined as working less than 20% of years since age 18.
Nebraska: Ponza et al (2002). Caring for child with special needs includes care for other family members with health problems.
Illinois through South Carolina : Hauan and Douglas (2004).

For many barriers, the range of cross-state variation, particularly in the studies using the same measures, is relatively low. In most study areas, 30 to 45 percent of recipients lacked a high school diploma or GED, with the exception of Milwaukee, where 57 percent had lower education levels. In all areas except Nebraska, less than 10 percent had serious substance abuse dependence issues. There is more substantial variation across geographic areas in health problems. Among studies using the same measure of mental health, the percent with problems ranges from 17 percent in Nebraska to 41 percent in Colorado. The percent of recipients with physical health problems ranges from 12 to 36 percent, although among studies using the same survey instrument, the range is smaller: from 16 to 29 percent. It is possible that differences in welfare policies across areas affect caseload composition and explain some of the differences in levels of barriers.

All of these studies conclude that barriers are relatively common among recipients. Most find that the vast majority of the caseload has at least one barrier, and a substantial minority of the caseload has multiple barriers. The specific percentage varies with the number of barriers being measured. For example, in Michigan, 85 percent of the sample in 1997 had at least one barrier while 37 percent had two or three barriers (Danziger et al. 2000).

Loprest and Zedlewski (2006) compare the prevalence of barriers among recipients, former recipients, and low-income women with children who have never received welfare using the set of barriers available in NSAF.56 They find that the prevalence of barriers is generally similar or higher among current recipients compared with leavers. In particular, leavers have more work experience. However, they do not have significantly lower levels of health or mental health problems than current recipients in 2002. Current recipients also have similar or higher rates of barriers than women who have never received welfare, with the notable exception that a much higher percentage of women who have never received welfare are primarily Spanish speaking.

Is the caseload more disadvantaged over time? We now turn to evidence on whether a growing percentage of the caseload faces significant barriers to work since welfare reform. This concern stems from the idea that those most ready for work (i.e., the least disadvantaged) would exit welfare quickly, leaving behind those who would have more trouble finding work. We have seen some evidence that TANF increased exits across all families. However, a number of factors could offset increased exits of those most job ready. One is the characteristics of new entrants to the program. While entry to the program declined after reform, there continued to be new entrants. If new entrants have fewer barriers on average than longer-term recipients, entry would tend to dampen a trend of growing disadvantage. In addition, exits may increase among those with more barriers to work due to time limits or full-family sanctions. This would have the opposite implication for caseload composition over time, reducing barriers to work among the caseload.

Relatively few studies have examined directly whether the caseload is becoming more disadvantaged over time. There are several ways studies examine this question.57 The most direct evidence is from studies that use national data to compare characteristics of multiple cross-sections of welfare recipients. However, because changes in the composition of the caseload might lead to offsetting trends in disadvantage, some studies also examine changes in subgroups of the caseload.

Moffitt and Cherlin (2002), extending work in Moffitt and Stevens (2001), use data from the CPS and compare the time period 1990–93 with 1996–98. They show a significant increase in the share of the caseload that has 12 years or more of education, as well as increases in annual weeks worked, indicating a decrease in disadvantage. Bavier (2001) examines data from the 1st, 12th, 24th, 36th, and 48th month of the 1996 SIPP panel, roughly representing caseloads from 1996 through 2000.58 He shows that the share of the caseload with low educational levels, who have never married, and who use rental assistance remained relatively steady over this period, while the share who were long-term recipients (more than 60 months) had actually fallen substantially. However, he also finds that an increasing share of recipients have a health condition that limits or prevents work. In subsequent work, he extends this analysis using data from the SIPP up to 2002 and includes 1993 to 2001 March CPS data with similar results (Bavier 2003). Both the later SIPP and CPS data show increases in the share of recipients with work-limiting conditions.

Bavier (2003) also separately looks at measures of entry and exit in the SIPP for groups with different observable barriers. He finds that exit rates increased from 1993 to 2001 for those with less than 10th grade education, four or more children, never married, receiving rental assistance, or with a work-limiting condition. Over the same period, entry rates tended to decline for these groups, with the exception of an increase in entry for most groups in 2001. In addition, entry rates for those with a work limitation have tended to increase over the entire time period. These changes in entry and exit are consistent with relatively unchanged prevalence of barriers in the caseload, with the exception of an increase in those with a work-limiting condition.

Table 8 - Barriers to Work of Welfare Recipients - Early and Late Reform Period
	NSAF			SIPP			NSAF			SIPP			CPS
	EARLY PERIOD						LATER PERIOD
	1997	1999	Change	1996	2001	Change	1999	2002	change	1999	2003	Change	2000	2005	Change
Less Than HS Degree (%)	39.8	46.1	6.3	38.9	43.3	4.4 *	46.1	41.4	-4.7	42.0	40.5	-1.5	40.6	41.3	0.7
Child Under Age 1 (%)	15.8	17.5	1.7	13.9	14.9	1.0	17.5	18.3	0.8	9.5	12.2	2.7 *	16.7	17.8	1.1
Child on SSI (%)	7.9	5.9	-2.0	7.7	6.2	-1.5	5.9	7.6	1.7	8.0	5.2	-2.9 *	na	4.1	na
Health Condition Limits Work (%)	22.7	30.4	7.7 *	21.3	29.2	7.9 *	30.4	25.2	-5.2	27.5	26.6	-0.9	22.1	24.7	2.6

* indicates change is significant at the 90 percent confidence level.

Our analysis of the NSAF, SIPP, and CPS data, focusing on four barriers, shows some evidence of an increase in barriers over the early period of reform, but no change in more recent years.

Changes in Barriers among Recipients. Data on barriers among TANF recipients from NSAF, SIPP, and CPS for different years are shown in table 8. The barriers presented include having less than a high school degree, having a child under age 1, having a child with a disability (measured as receiving SSI disability benefits), or having a health condition that limits work. Focusing on the 1997 NSAF and 1996 SIPP results, we find very similar levels of these barriers nationally. For example, a little less than 40 percent of recipients have less than a high school degree and about 8 percent have a child receiving SSI.

During the early years of welfare reform, there is some evidence that the typical TANF recipient faced increasing barriers to work. Data from the NSAF indicate that the share of TANF family heads with a health condition that limits work increased from 22.7 to 30.4 percent between 1997 and 1999. SIPP data covering the period between 1996 and 2001 show a similar trend with the share of case heads with a work-limiting health condition rising from 21.3 to 29.2 percent.

Both the NSAF and the SIPP find that the share of TANF heads that have failed to complete high school grew during the early reform period. Although the change in the NSAF is larger than the change in the SIPP, it is not statistically significant. The SIPP data show that the share of TANF families without high school degrees rose by 4.4 percentage points, from 38.9 to 43.3 percent. Neither data set shows any significant change in the share of TANF families with infants or with children on SSI.

During the late reform period, there are few significant changes in barriers to work. In fact, neither the NSAF from 1999 to 2002 nor the CPS from 2000 to 2005 shows any significant change in the prevalence of barriers among TANF recipients. The SIPP shows two significant but offsetting changes in barriers between 1999 and 2004: the share of families with an infant rose by 2.7 percentage points from 9.5 to 12.2 percent, but the share with a child on SSI fell by 2.9 percentage points from 8.0 to 5.2 percent.

In addition to these results, other evidence shows that long-term recipients, presumably representing those who are more disadvantaged, are making up a smaller share of the caseload over time, not a larger share as might be expected from growing disadvantage. Moffitt and Stevens (2001) analyze administrative data in Baltimore City, Maryland, from 1985 to 2000, and find a sharp decline in long-term recipients in the later 1990s compared with earlier years. Loprest and Zedlewski (2006) show that in the NSAF data, longer-term recipients (those on welfare for two years or more) made up 68 percent of the TANF caseload in 1997 (excluding child only cases) but fell to 42 percent of the caseload in 2002. This decline in the share of recipients that are long-term could be a result of time limits.

There is also some evidence that the level of disadvantage among longer-term recipients is not that much greater than among new entrants. These results could occur if new entrants are more disadvantaged as a group than in the past or if long-term recipients are less disadvantaged. Loprest and Zedlewski (2006) show that in 2002, long-term recipients are significantly more likely to not have worked in the past two years and have less than a high school education than new entrants, although there is no difference in having a young child, poor health, or poor mental health. Similarly, Moffitt and Winder (2003) using data from the Three Cities study show only small differences between stayers (on welfare in both waves of their data) and new entrants in the percent lacking a high school degree or GED and rates of depression.

Changes in Barriers among Former Recipients. Information on the same set of barriers among former recipients for the three national surveys is shown in table 9. In the 1997 NSAF and 1996 SIPP data shown, we find very similar prevalence of most barriers, including having low education level, an infant child, and a child on SSI. However, the NSAF finds significantly fewer former recipients have a health condition that limits work (8.8 percent) compared to the SIPP (16.1 percent). This difference could stem from differences in the definition of leavers between the two datasets, particularly if leavers with health problems are more likely to return to TANF or return to TANF sooner.59 Our analysis also shows that, generally, current recipients are more likely to have barriers than those who left TANF.

Table 9 - Barriers to Work of Former Welfare Recipients
	1997	2002	Change	1996	2001	Change	2000	2005	Change
	NSAF			SIPP			CPS
Less Than HS Degree (%)	30.1	33.5	3.4	33.5	30.4	-3.1	35.1	33.5	-1.6
Child Under Age 1 (%)	11.1	14.4	3.3	13.7	12.4	-1.3	13.8	14.1	0.3
Child on SSI (%)	5.5	5.2	-0.3	6.5	3.1	-3.4 *	na	1.6	na
Health Condition Limits Work (%)	8.8	18.7	9.9 *	16.1	19.8	3.7	13.1	15.9	2.8

* indicates change is significant at the 90 percent confidence level.

The prevalence of barriers to work among TANF leavers has not changed appreciably over time with a few exceptions (table 9). Indeed, the CPS data indicate no significant changes in barriers to work among leavers between 2000 and 2005. Starting with the earlier base year, the SIPP and NSAF find some changes in barriers. The SIPP data find a decline in the share of leavers with a child on SSI from 6.5 to 3.1 percent between 1996 and 2001. Interestingly, the data from the NSAF suggest that the share of leavers with a disability that limits work increased from 8.8 to 18.7 percent between 1997 and 2002. The CPS and SIPP also find a rise in leavers with work-limiting conditions but these changes are not statistically significant.

In summary, this evidence on current and former recipients suggests that although there may have been some increases in prevalence of specific barriers in the early years of reform, in the more recent period, there is little evidence of the caseload becoming more disadvantaged.

How can this result be reconciled with anecdotal evidence from individual welfare offices and advocates that there are many more hard-to-employ welfare recipients? First, there are several important caveats to this result. It is based on the available national level data that measure a limited set of barriers to work, not the much broader set measured by some state and local studies. It is possible that the barriers not measured here have increased among the caseload. These include, for example, experience of domestic violence or substance abuse. Possibly more important is the consideration of “unobservable” barriers. Researchers can only report on factors that they can measure. While great strides have been made in identifying and measuring barriers that recipients may not want to voluntarily report (criminal history and domestic violence) or may not even be aware of (depression or other mental health problems), there are factors that are even more difficult to observe and measure. Factors such as motivation, self-esteem, and ability to cope with complex systems can all impact work and exit from the caseload and are difficult for researchers to measure. However, case managers and front-line staff may be finding these barriers among more of their caseload over time as they work with their clients.60 It is also possible that it is not that more of the caseload has barriers, but that a growing number of the clients actively engaged in work activities have barriers. This could occur if welfare programs are “going deeper” into their caseload, that is, trying to engage a greater percentage of clients in work activities than in the past, including some who were formerly exempt due to barriers. This may be occurring in some states as they move toward universal participation and try to meet higher work requirements and as they work more intensively with clients nearing time limits.

Barriers to Employment and Movements on and off Welfare. The high level of barriers itself provides some evidence of need for services among welfare recipients, but many point out that these are really potential barriers to work, and it is not clear to what extent they actually limit the work of TANF recipients. Therefore, in addition to documenting the prevalence of barriers among the caseload, many studies examine the relationship between barriers and employment at a point in time. The evidence is mixed, but most of the barriers measured are associated with lower rates of employment, although the nature of the relationship depends on the specific study methods. Because some barriers are co-occurring (for example, mental health problems and substance abuse), inclusion of multiple barriers in one specification leads to fewer having a significant association with employment than univariate estimates suggest (for examples, see Danziger et al. 2002, Hauan and Douglas 2004, and Loprest and Zedlewski 2006). However, one can generally conclude from these studies that those with barriers on welfare are less likely to be employed than those without barriers. The evidence is strongest for recipients with multiple barriers to work—they have significantly lower employment rates than recipients with one or no barriers.

The negative relationship between barriers and work suggests that those with barriers may be less likely to exit TANF. A few studies have examined the association between barriers and exiting TANF directly. This requires longitudinal data to observe individuals on and off welfare. Hofferth, Stanhope, and Harris (2001) using the PSID find that, in the prewaiver period, lower education levels, work experience, and having a physical or nervous disability that limits work are associated with lower rates of exit for any reason and lower rates of exit to work. However, only disability was associated with a lower rate of exit for nonwork reasons. The United States General Accountability Office (2003), using SIPP data, finds that recipients with health impairments (including mental health) were half as likely to exit as those without impairments between 1997 and 1999. Acs et al. (2001), also using SIPP data, find that in 1996, the probability of exit is significantly lower for long-term recipients, those with a disability, and those with a child under age 1 than for those without these barriers. These differ from the rates for 1990, where the probability of exit for those with a disability is not significantly different than those without a disability.

Although the studies discussed above examine the association between barriers and employment or barriers and exit among TANF recipients at a point in time, for the most part they do not attempt to examine changes in these relationships over the time period after reform.

Whether welfare reform would tend to increase or decrease the work of recipients with barriers or tend to increase or decrease exit and entry from the caseload among individuals with barriers is unclear. Specific policies have offsetting potential impacts. Some states continued policies that formally exempted those with work barriers from work requirements (or put them in a separate state program) or informally concentrated work efforts on those without barriers, making exits to work less likely for those with barriers. Many states instituted more generous earnings disregards that would tend to keep on the rolls those with fewer barriers who found jobs. Work requirements and mandatory sanctions could cut both ways. They could “smoke out” those who are already working or who could easily find work and for whom the hassle of meeting requirements outweighed the value of the welfare check. However, full-family sanctions could mean higher exits among those with barriers who have trouble meeting work requirements. These policies also have impacts on entry and the composition of entrants. For example, Acs et al. (2005) find lower TANF entry rates in states with full family sanctions.

There is some indirect evidence on the impact of welfare reform on disadvantaged recipients over time. A few studies suggest that exit rates among those with disadvantage (measured in various ways) were higher after reform than before reforms. Grogger (2004), using data from the SIPP, finds that TANF has large positive impacts on exits from ongoing (longer-term) spells of welfare, which he argues likely represents recipients with greater levels of disadvantage. However, he finds little impact of reform policies on exit from “fresh” spells of welfare, that is, newer recipients, whom we expect to have fewer disadvantages. Also using SIPP data, Bavier (2002) finds that average outcomes for leavers, including income loss after exit, are worse for those who left welfare after July 1996 than for those who left earlier, controlling for a standard array of demographic, economic, and policy variables. This could be due to TANF policies increasing exit rates of individuals with less ability to succeed in the labor market (i.e., those who would have been less likely to exit welfare in the prereform period). However, he does not find greater negative impacts of observable disadvantages (such as low educational attainment or disability) on outcomes in the post reform period relative to the earlier period. Bavier suggests that if TANF policies did lead to greater exits for disadvantaged families, these disadvantages must be in characteristics that are not observed in the data.

Loprest and Zedlewski (2006) also provide some information about changes in the relationship between work and barriers over time. They use NSAF data to show the change in probability of work among welfare recipients with barriers between 1997 and 2002. They find a significant increase in work among recipients with less than high school education, a child receiving SSI, and those who were interviewed in Spanish because of limited English proficiency. The percentage with two or more barriers working more than doubled from 10 to 26 percent. However, those in poor health were relatively less likely to work over time.

These results provide some suggestions that work and exit among those with barriers has increased over time, although work is still lower among those with barriers than among those without barriers. While Bavier (2002) suggests outcomes of leavers after reform were initially worse than the pre-TANF period, we know little about whether any increases in work or exit among recipients in the later reform period were linked to positive or negative outcomes. Increased work and exits among those with barriers over time is consistent with welfare programs increasing their focus on helping these individuals move successfully into the labor market. However, increased exit of those with barriers is also consistent with increases in these families losing benefits through sanctions or time limits, because of inability to meet requirements.

Question 3: What do we know about the economic progress of TANF recipients and leavers over time?

One of the major goals of welfare reform in the 1990s was moving families from welfare to work and increasing work among welfare recipients. States used varying combinations of requirements and incentives to move recipients into work and off welfare (Rowe and Giannarelli 2006). States increased work requirements but varied in what they counted as work activities and who they exempted from the requirements. They also varied in the triggers and severity of sanctions for noncompliance with work requirements, including eliminating benefits for the family. And to varying degrees, states increased the amount of earnings that can be retained while still receiving cash assistance and increased funding for child care and other work supports.

While welfare reform was being implemented, employment rates among single women with children boomed. Between 1994 and 1999, the labor force participation rate of single mothers rose 10 percentage points, compared with almost no change over the previous decade and a half (Blank 2002). Work among TANF recipients has likewise increased markedly in the time since the passage of welfare reform. Our analysis of the NSAF, SIPP, and CPS data shows this increase in employment.

Employment among Welfare Recipients. During the early reform period, the share of TANF family heads working increased (table 10). The NSAF data indicate that the share of TANF case heads who worked rose from 20.9 to 31.5 percent between 1997 and 1999, while the SIPP data show an increase from 22.8 to 27.8 percent between 1996 and 2001, both statistically significant increases. Both data sets show a similar increase in work among single parents on TANF. These trends mirror that reported in the

administrative data, which show an increase in employment among adult TANF recipients from 11.3 percent in fiscal year 1996 to 27.6 percent in fiscal year 1999.61

Table 10 - Employment Characteristics of Welfare Recipients - Early and Late Reform Periods
		NSAF			SIPP			NSAF			SIPP			CPS
		EARLY PERIOD						LATER PERIOD
		1997	1999	Change	1996	2001	Change	1999	2002	change	1999	2003	Change	2000	2005	Change
Current Status (all families %)	Employed	20.9	31.5	10.6 *	22.8	27.8	5.0 *	31.5	29.2	-2.3	28.2	24.9	-3.4	31.0	24.5	-6.5 *
	Not Working but Looking				13.1	13.3	0.2				10.2	12.3	2.1	12.6	15.0	2.4
	Not Working or Looking but in School				8.4	5.6	-2.8 *				6.3	7.3	1.0	56.4	60.4	4.0
	Not Working, Looking or in School				55.7	53.3	-2.4				55.3	55.5	0.2
Currently Employed (single families %)		20.3	30.9	10.6 *	21.3	26.1	4.8 *	30.9	33.3	2.4	28.1	21.2	-6.9 *	32.2	23.1	-9.1 *
Usual Weekly Hours (head %)	<20	35.8	22.2	-13.6 *	24.2	20.6	-3.6	22.2	13.4	-8.8	16.0	19.6	3.6	6.8	10.1	3.3
	20-34	24.1	21.7	-2.4	30.0	21.5	-8.6 *	21.7	29.9	8.3	32.4	23.9	-8.5 *	26.2	29.4	3.2
	35+	40.2	56.1	15.9 *	45.7	57.9	12.2 *	56.1	56.7	0.6	51.6	56.5	4.9	67.0	60.5	-6.5
	Mean (in hours)				30.3	34.3	4.0 *	33.1	34.1	0.9	31.7	32.1	0.4	34.8	33.3	-1.5
Median Hourly Wage ($2005)		$6.69	$7.05	$0.36	$5.84	$6.71	$0.88 *	$7.05	$7.60	$0.55	$6.99	$7.52	$0.53	$6.74	$7.75	$1.01
Has Own-Employer Health Insurance (%)		4.3	5.1	0.8	15.0	17.1	2.0	5.1	9.9	4.8	18.5	19.1	0.6	13.4	14.6	1.2

* indicates change is significant at the 90 percent confidence level.

Notes:
Unit of Analysis -- "TANF Family" as defined below

NSAF - children and adults living together related by blood, marriage, or romantic attachment (includes cohabiting partners). Adult most knowledgeable about the child is considered head.
SIPP - all persons related by blood, marriage, or adoption; excludes cohabitors. Head is designated parent or guardian of at least one of the children receiving welfare. In married couple families, head is mother; in multiple generation families youngest parent/guardian is head.
CPS - all persons related by blood, marriage, or adoption; excludes cohabitors. Head is designated parent or guardian of at least one of the children receiving welfare. In married couple families, head is mother; in multiple generation families youngest parent/guardian is head.

In addition, among those who work, there was a significant increase in the share working full-time (35 or more hours per week)—from 40.2 to 56.1 percent between 1997 and 1999 in the NSAF and from 45.7 to 57.9 percent between 1996 and 2001 in the SIPP. The SIPP also reports a rise in mean hours worked per week from 30.3 to 34.3.

Data on compensation in the early period show median hourly wages for those working grew by about 90 cents in the SIPP from $5.84 to $6.71.62 There was some increase in wages from 1997 to 1999 in NSAF, but the increase is not statistically significant.63 It is not clear why absolute median hourly wage levels for working current recipients in the NSAF are somewhat higher than in the SIPP. The percent of working recipients who are covered by employer-sponsored health insurance (ESI) is quite low—about 1 in 20 in the NSAF and about 1 in 6 in the SIPP. The higher rates of coverage in the SIPP likely stem from differences in sample composition discussed earlier.64 This compares to about 60 percent of the general population covered by ESI (U.S. Census Bureau 2006). We find no significant change in this rate over time.

Our analysis indicates that the rapid increase in employment in the early period begins to level off and even decline in the later years after reform. Between 1999 and 2002, the NSAF data indicate no significant change in employment for all TANF heads and single parent TANF heads. Over the years 1999 to 2003, the SIPP shows a statistically insignificant decline in employment for all TANF family heads, but a significant decline of 6.9 percentage points for single parent heads. Between 2000 and 2005, the share of TANF heads that are employed falls from 31.0 to 24.5 percent in the CPS data, and the share of single parent cases employed declines 9.1 percentage points. The administrative data also show declining employment rates in the later years after reform. Reported employment rates for TANF recipients fell from a high for the 1990s of 27.6 percent in fiscal year 1999 to 22.0 percent in 2004 (the most recent data available).

In this later period, the trend toward full-time work among those who are employed also flattens out. There are no significant changes in mean usual hours worked. Trends in the distribution of hours worked vary across the three data sources, but the only statistically significant change appears in the SIPP data: the share of TANF family heads working 20 to 34 hours per week declined, but this decline was accompanied by statistically insignificant increases in both the shares working less than 20 hours and 35 or more hours. Finally, neither wages nor employer-provided health benefits grow significantly over the late reform period.

Overall, the data indicate that employment and, to some extent, wages for those who were employed increased during the early reform period. During the late reform period, particularly after 2002, employment rates for TANF families fell. Whether this is due to a slowdown in the economy relative to the late 1990s or other factors is not clear.

Research on the reasons for the observed increases in work in the early period of reform suggests that welfare reform played a role. Several excellent reviews have summarized research on the impacts of welfare reform on employment, earnings, and income. Blank (2002) concludes that the research finds that welfare policy (including waivers) had a significant impact on labor force participation. This research also suggests that other factors also had significant impacts on the increases in employment for women with children over this time period including the expansion of the EITC and the strong economy. In their synthesis, Grogger, Karoly, and Klerman (2002) consider the impacts of specific welfare policies (primarily from waiver evaluations) as well as TANF reform as a bundle. They conclude that there is strong evidence that mandatory work-related activities lead to increases in employment and earnings (mainly through increased hours) and moderate evidence that financial work incentives increase employment. Considering TANF reforms as a bundle, they conclude there is moderate evidence that these combined policies led to a significant increase in employment and earnings. All of these research studies measuring the impacts of TANF examine changes in outcomes in the broader group of (usually single) women who are at risk for participating in TANF, not just the TANF caseload changes that our tables report. Most of this research on TANF impacts focuses on the early period of reform.

Employment among Former TANF Recipients. A substantial number of studies address the issue of employment rates among former recipients. A review of a large number of these studies finds that a majority of former recipients are employed in the first few months after exiting TANF, with the median employment rate of 57 percent in the areas studied (Acs and Loprest 2004). Similar findings have been reported from national data. Acs et al. (2001), using the SIPP, find that the employment of single mothers in the first four months after exiting TANF was 64 percent. Hofferth, Stanhope, and Harris (2001), using data from the PSID, find that about two-thirds of all exits are associated with work, as opposed to other reasons for exit, such as new marriages, changes in living arrangements, or a child turning 19. All of these results come from the early years after reform, 1996 though 1998.

Relatively few studies have examined the changes in employment for leavers over time. Our analysis of three national data sets shows that employment rates for leavers declined during the later years after reform. The NSAF and the CPS measure employment at the time of interview, potentially some months after exit. In the SIPP data, employment is measured at the month of exit. All three data sets find decreases in employment at exit for later groups of leavers, although the NSAF results are not statistically significant (table 11). Between 1996 and 2001, the SIPP indicates that the share of TANF leaver family heads who are employed at exit fell from 56.7 to 49.3 percent, a 7.3 percentage point drop. Similarly, the CPS data indicate that employment of TANF leavers dropped by 15.2 percentage points, from 54.5 to 39.3 percent between 2000 and 2005. The share of families leaving TANF with any employed adult also dropped significantly in both SIPP and CPS data.

Among TANF leavers that do work, there is little change in hours worked. The NSAF and SIPP find no significant changes, but the CPS data indicate that working leavers are less likely to work full-time in 2005 than in 2000. Hourly wages for working leavers are somewhat higher over time in all three data sets, but the only significant increase appears in the NSAF data. Between 1997 and 2002, wages climbed from $7.61 to $8.41 an hour.65 As in the case of current recipients, wage levels are somewhat higher in the NSAF than in the SIPP.66 We should note that hourly wage rates of working leavers in NSAF and SIPP are consistently higher than those of current recipients, suggesting that those who can earn higher wages are more likely to exit or less likely to continue to be eligible for TANF. Finally, there are no significant changes in employer health insurance coverage among employed welfare leavers over these time periods.

Table 11 - Employment Characteristics of Former Welfare Recipients
		1997	2002	Change	1996	2001	Change	2000	2005	Change
		NSAF			SIPP			CPS
Current Status (all families %)	Employed	63.4	59.4	-4.1	56.7	49.3	-7.3 *	54.5	39.3	-15.2 *
	Not Working but Looking				8.4	7.9	-0.5	8.3	14.3	6.0 *
	Not Working or Looking				2.1	3.2	1.1	37.3	46.4	9.1 *
	Not Working, Looking or In School				32.9	39.6	6.8
Any Adult in Family Employed (%)		88.2	81.0	-7.2	74.8	64.7	-10.1 *	69.5	57.6	-11.8 *
Currently Employed (single families %)		71.2	66.0	-5.3				54.5	38.6	-15.9 *
Usual Weekly Hours (head %)	<20	10.6	8.4	-2.2	13.9	16.4	2.5	9.5	9.2	-0.4
	20-34	20.7	25.6	4.9	17.8	19.8	2.0	22.8	35.6	12.8 *
	35+	68.7	66.0	-2.7	68.3	63.8	-4.5	67.6	55.2	-12.4 *
	Mean (in hours)				36.1	37.5	1.4	34.6	33.2	-1.4
Median Hourly Wage ($2005)		$7.61	$8.41	$0.80 *	$7.05	$7.58	$0.53	$7.89	$8.11	$0.22
Has Own-Employer Health Insurance (%)		17.5	23.5	6.0	23.3	21.7	-1.6	26.3	21.8	-4.5

* indicates change is significant at the 90 percent confidence level.

The trends in the employment and earnings of TANF leavers indicate that after the early years of welfare, fewer leavers are working at exit, but the quality of their jobs (measured by wages and employer health insurance) has not grown worse.

Income of Welfare Recipients. Although work was a major focus of welfare reform efforts, understanding changes in income among recipients and leavers provides a picture of families’ economic well-being over time. There are several ways to examine changes in income over time. Some studies have analyzed changes in the income of low-income, female-headed families with children, a population that is “at risk” for welfare receipt (Haskins 2001; Zedlewski 2002). Both of these studies find that the average income of female-headed households with children increased over the mid- to late 1990s largely due to increases in earnings. But both studies also find that there are families in the lower end of the distribution whose incomes have fallen, mainly due to a decline in benefits that was not offset by increases in earnings.

Another approach to understanding changes in incomes for TANF recipients and former recipients is to compare average incomes for these groups at different points over time, as we do below. We then review the evidence on income changes of individual welfare recipient families over time, especially as they leave welfare. These studies attempt to answer the question of whether families are economically better off after exiting welfare. This type of study requires longitudinal data that collects information for the same individuals over time, typically while on and off welfare.

The early period of reform saw increases in the incomes of some TANF recipients (table 12). TANF family income in all three data sets includes pretax cash income, including government cash benefits but excluding the value of food stamps.67 Because most TANF recipients receive food stamps and there is not much difference in the value of food stamps over time, its exclusion should not have much impact on the trend results, although absolute incomes including food stamps would be higher. The NSAF shows that between 1997 and 1999, both mean and median family income of TANF recipients rose by about $2,000.68 This is consistent with the large increase in employment for this group in the NSAF. The SIPP data show modest but statistically insignificant increases in annual family income of TANF recipients from 1996 to 2001.

In both the NSAF and SIPP data sets, we observe reductions in the percentage of TANF families in poverty. In the NSAF, the percentage of TANF families in extreme poverty (less than 50 percent of the poverty line) fell by 11 percentage points over this early period, with increases in the percentage of TANF families above the poverty line. The SIPP data show a significant decline in the percentage of TANF families between 50 and 100 percent of the poverty level by 5.6 percentage points with a statistically significant increase in the percent of families above 150 percent of the poverty level.

Table 12 - Income, Earnings, and Benefits of Welfare Recipients - Early & Late Reform Periods
		NSAF			SIPP			NSAF			SIPP			CPS
		EARLY PERIOD						LATER PERIOD
		1997	1999	Change	1996	2001	Change	1999	2002	change	1999	2003	Change	2000	2005	Change
Mean TANF Family Income ($2005)		$10,568	$12,763	$2,196 *	$20,469	$21,583	$1,113	$12,763	$14,846	$2,083 *	$23,366	$24,495	$1,129	$17,820	$17,535	-$285
Median TANF Family Income ($2005)		$7,629	$9,936	$2,307 *	$11,876	$12,593	$716	$9,936	$11,790	$1,854	$14,334	$14,662	$327	$11,957	$12,407	$449
TANF Family Income Relative to Poverty (%)	<50% poverty	60.5	49.5	-11.0 *	35.5	37.3	1.8	49.5	44.1	-5.4	31.6	35.0	3.3	33.3	36.8	3.5
	50-100% poverty	27.5	25.9	-1.6	34.3	28.8	-5.6 *	25.9	29.0	3.1	33.5	28.2	-5.4 *	36.7	30.2	-6.5 *
	100-150% poverty	8.5	18.7	10.2 *	13.0	13.6	0.5	18.7	17.5	-1.3	13.9	13.8	-0.1	16.4	16.2	-0.1
	150%+ poverty	3.5	5.9	2.4	17.1	20.3	3.2 *	5.9	9.4	3.5	20.9	23.0	2.1	13.6	16.7	3.1

* indicates change is significant at the 90 percent confidence level.

It is important to note here that the NSAF estimates of the level of family income are consistently lower than those reported in the SIPP and CPS. This is most likely due to differences in sample composition, the definition of income, and the way income is collected in each data set. As discussed earlier, SIPP and CPS data include a greater number of families where only children are receiving TANF. Adults in these families may have higher incomes than TANF recipient adults and in some cases their income may not be considered in TANF eligibility. The NSAF sample also has an income screen that excludes the highest income families, although only 3 percent of TANF families are excluded in 1997. In addition, the NSAF definition of income only includes earnings of the TANF recipient and spouse/partner. The lower income in the NSAF data could be stemming from not including earnings of other adults in the household. As we saw earlier, in the NSAF data, about 30 percent of TANF recipients in 1996 are living with other adults not their spouse. Unfortunately, NSAF does not gather current (at the time of the interview) earnings data for these other adults. Finally, the way income data is collected differs across the data sets. One of the advantages of the SIPP is the short recall period individuals have (four months), providing more accurate and potentially higher income reports as people are less likely to forget sources of income. CPS respondents are asked to recall income information from the prior year. Also, we define TANF recipients in the CPS as receiving in the month of the interview, but reported income is for the prior year, when they may or may not have been TANF recipients. This could bias the income reports upwards, as earnings while not on TANF may be higher than earnings in the month of TANF receipt.

During the late reform period, we continue to see growth in average incomes of TANF families in the NSAF for the years 1999 to 2002, although it is lower than the early period. Growth in median income is smaller and no longer statistically significant, and the total growth in mean income is lower over this longer time period. The SIPP shows no significant increase in average incomes over this later time period. However, it is important to note that median and mean income levels in both 1999 and 2003 are higher than income in 2001. This suggests that this later time period is likely masking a decline and recovery in TANF recipients’ incomes over this period. From 2000 to 2005, the CPS also shows no significant change in incomes of TANF recipients.

In the later period, there continue to be declines in the percentage of families in poverty, although the decline in extreme poverty in the NSAF is not statistically significant. Both SIPP and CPS show declines of more than 5 percentage points in families with incomes between 50 and 100 percent of the poverty line. At the same time, there were roughly even increases in the percent of families in deep poverty and the percent of families with incomes more than 150 percent of the poverty line, although these changes are not statistically significant. This suggests that although, on average, incomes of recipients remained steady over this entire later time period (in these data), there were changes in the distribution of income that indicate an increase in inequality among TANF recipients. Some families were moving up and out of poverty, but some families were moving down into extreme poverty.

Overall, the data sets show two different income pictures. The NSAF data, focusing on a more narrow group of welfare recipients, show income growth in both periods, although somewhat slower growth in the later period. These data show especially dramatic declines in deep poverty in the early period. The SIPP data show a fairly stagnant income picture on average over both periods of reform, although there is some evidence that the distribution of income was changing. In both periods, there are decreases in the percentage of families in poverty, but in the later period, the SIPP and CPS both suggest a more unequal distribution of income among TANF families after 2002.

Income of Former Recipients. All three data sets show no significant changes or declines in average income for different groups of welfare leavers over time (table 13). The SIPP and the CPS find substantial declines in mean and median income; however, only the decline of median income in the SIPP of almost $5,000 from 1996 to 2001 is statistically significant.69 Both of these data sets show a significant increase in the percent of leaver families in extreme poverty. Between 1996 and 2001, the SIPP shows that the share of welfare leavers with incomes below 50 percent of the poverty line climbed from 24.4 to 33.3 percent, an 8.8 percentage point increase. The CPS indicates that the share in deep poverty rose from 25.3 to 31.6 percent between 2000 and 2005. The NSAF data may be less likely to show changes in income for leavers because the sample includes those who left over the past two years and remain off welfare. Thus, leavers in 2002 represent in part leavers from earlier years and a greater number of successful leavers. SIPP and CPS include all leavers, including those who end up returning fairly rapidly to TANF.

Table 13 - Income, Earnings, and Benefits of Former Welfare Recipients
		1997	2002	Change	1996	2001	Change	2000	2005	Change
		NSAF			SIPP			CPS
Mean TANF Family Income ($2005)		$16,528	$16,593	$65	$26,189	$22,904	-$3,285	$25,528	$22,733	-$2,795
Median TANF Family Income ($2005)		$15,476	$15,006	-$470	$18,686	$13,835	-$4,851 *	$16,904	$15,405	-$1,499
TANF Family Income Relative to Poverty (%)	<50% poverty	27.9	31.3	3.5	24.4	33.3	8.8 *	25.3	31.6	6.3 *
	50-100% poverty	35.7	30.1	-5.6	25.4	27.0	1.6	31.2	27.6	-3.6
	100-150% poverty	22.8	23.8	1.0	19.6	12.9	-6.7 *	16.6	16.8	0.1
	150%+ poverty	13.6	14.8	1.1	30.6	26.9	-3.7	26.9	24.1	-2.8

* indicates change is significant at the 90 percent confidence level.

We also continue to find higher absolute incomes among SIPP and CPS leaver families than among the NSAF leaver families, despite the fact that NSAF leavers would likely represent more successful leavers. While there continue to be some differences in the sample composition due to how we define welfare receipt in these data sets, as discussed earlier, this difference is likely much more limited among leavers, because families where only the child receives benefits are less likely to exit TANF. This suggests that much of the difference in levels is coming from the income screen in the NSAF sample that excluded some higher-income leaver families.70

It is important to note that, on average, the incomes of former TANF recipients are higher than of those on TANF. This is in part related to the fact that family incomes of TANF recipient households must fall below certain eligibility thresholds. Median TANF family income is higher for leavers than for recipients for all the years presented in the three data sets we analyze. For example, in 1996 NSAF data, TANF recipients have median family incomes of $7,629, compared with $15,476 for former TANF recipients. However, all three data sets also show the difference between leaver and recipient family income declining over time. This follows from the increasing or flat income among recipients and flat or declining income among TANF leavers that we observe over time. It is also worth noting that while deep poverty among recipients remained stable or grew slightly during later reform period, it grew markedly for families that left welfare. This is consistent with families that left welfare who lost jobs finding it more difficult to find new jobs in a tighter economy in the later period. It is also consistent with a greater percentage of families leaving welfare due to not meeting the work requirement or hitting time limits (and having lesser job prospects) than in the earlier period.

Are Families Better Off after Leaving TANF? Although our analysis shows that leavers have higher incomes than recipients and many leavers are working after exit, this does not necessarily mean that families are better off after leaving welfare than while receiving benefits. This is because the groups of leavers and recipients we compare are not the same individuals, and the composition of these two groups varies greatly. To address the question of whether individual families leaving TANF are better off after exit, several studies have used longitudinal data to compare the income and circumstances of families before and after leaving TANF. Only one of these studies, Bavier (2001), uses national data (SIPP). A second study by Cancian et al. (2002) focuses on Wisconsin in the early years of reform under state waiver between 1995 and 1997. Two additional studies use data from limited geographic areas, the WES from a county in Michigan, and the Three Cities Study, from Boston, Chicago, and San Antonio. Because these latter studies’ findings are influenced by the specific policies and characteristics of the places studied, the results cannot necessarily be generalized to the nation.

Bavier (2001) uses SIPP data for 1996–1997 to compare household income before and after leaving AFDC/TANF. He finds that less than half of welfare leavers had household incomes that increased more than $50 per month. Comparing mean monthly post-exit household income (including the value of food stamps) of leavers over the year after exit with their mean monthly pre-exit income in the two months before exit, he finds that 44.3 percent have an increase of more than $50 and 48.9 percent have a decrease of more than $50 (the remaining 7 percent had stable incomes). On average, households that gained income had about 50 percent higher income than on welfare, while those that lost income had about two-thirds of their pre-exit income. However, those with income losses had on average much higher pre-exit income ($2,514) than those who gained income ($1,614). Bavier also shows that even among those who are employed at exit, 48 percent have increases in post-exit income of more than $50, although this is pretax income and does not include the EITC or exclude payroll taxes.

Cancian et al. (2002) use administrative data from Wisconsin, including AFDC benefits, food stamps, and posttax earnings, to examine the incomes of women who left AFDC in 1995 and 1997. They find that less than one-third of leavers had increased total income after exit relative to their income prior to exit. However, these results are limited because they do not include the earnings of other household members and other sources of income, biasing down the income of leavers.

Danziger et al. (2002) analyze the economic circumstances in 1999 of a sample of single mothers in a county in Michigan who received welfare in 1997 using longitudinal survey data from the WES. The study focuses on the question of “does it pay to move from welfare to work” by comparing earnings, income, and poverty across groups of women. They find that working welfare leavers (“wage reliant”) have higher earnings, total household income, and net income than either “welfare reliant” women, defined as nonworking welfare recipients or “combiners” who have earnings from work and receive welfare. In a later commentary and expansion of these results, Danziger and Wang (2005) revise their estimates downward based on newly available administrative data on TANF benefits and additional controls for heterogeneity across groups. The revised finding is that the net income difference between wage-reliant and welfare-reliant groups is $339 and between wage-reliant and combiners is $110.

Differences in income across groups may be due to women who leave welfare for work having different observed and unobserved characteristics than those who remain on welfare. The authors estimate a regression model that takes into account this potential adverse selection. The results show that for every additional hour of work, a women’s monthly net income increases by $2.63, suggesting that women are better off when they leave welfare for work than remaining reliant on welfare.

Moffitt and Winder (2004) commenting on the Danziger et al. (2002) study analyze data for women on welfare in 1999 from their Three Cities survey with a second interview in 2000/2001. They find that monthly net income for working leavers is $1,329, compared with $1,102 for nonworking welfare recipients, and $1,408 for those working and on welfare. The authors find a smaller difference in income between working leavers and nonworking recipients in this sample than in the Michigan study (even the revised downward numbers). They argue this is in large part due to higher levels of income from other family members among families leaving welfare in the WES data. In addition, unlike the Michigan study, Moffit and Winder find that incomes of working leavers are slightly lower than incomes of those combining work and welfare. This difference is due in part to greater contributions to income of other family members for those leaving welfare in the WES data and lower benefit reductions in the Three Cities sample for those combining work and welfare. The study shows that the gains to work are not exclusively enjoyed by those leaving welfare, and that in some circumstances, combining welfare and work can be more beneficial than leaving welfare for work.

The Three Cities Study also is able to deal with the adverse selection issue discussed above by directly measuring changes in income for families at two points in time.71 These data show that women who moved off welfare into work experience an increase in income of about $200 a month, while women who stay on welfare without working have an increase in income of $49.72 The difference between these two (about $150) is a measure of the gain to leaving welfare for work. This is about a 14 percent gain in income.

The authors go on to point out that another group of recipients leave welfare and are not working, and that it might be more correct to average the income for working and nonworking leavers if all leavers are uncertain about their employment prospects and/or face the possibility of losing employment. In the Three Cities data, this group of nonworking leavers has much lower incomes ($955) than the other groups. Combining groups of leavers, Moffitt and Winder find that the expected income gains from leaving welfare is either zero or very small.

The results from all of these studies suggest that there are at most modest gains in income from those leaving welfare to work, that there are those that gain income and those that lose income in this transition, and that perhaps much of the gains are due to movements from nonwork to work and not necessarily from “welfare and work” to “no welfare and work.” Offsetting losses in TANF benefits resulting from increased earnings serve to depress total income for those moving from welfare to work.

Moffitt and Winder (2004) point out that policies that provide incentives to work through increased earnings disregards will lead to higher incomes while on welfare. Of course, time limits on benefit receipt limit the extent to which this combining can occur, and for many, the goal of welfare reform was to decrease dependency on welfare while increasing incomes. The results also point to the differences in the benefits of moving from welfare to work across different geographic areas, in part due to differences in how different states reduce TANF benefits as earnings increase.

Another aspect of this question is how earnings and income of leavers changes over time after leaving. A number of location-based leaver studies report changes in earnings over time for leavers, typically based on unemployment insurance administrative records. A summary of these results in Acs and Loprest (2004) shows that, on average, quarterly earnings of employed welfare leavers generally increased by about $300, or more than 10 percent, from the first quarter after exit to the fourth quarter after exit. The studies report a fairly wide range of findings, from almost no increase (0.4 percent) in Cuyahoga County, Ohio, to about 30 percent in South Carolina. These results include individuals who returned to welfare by the fourth quarter. The few studies that report results for “continuous leavers” (those who were off welfare for the 12 months after exit) show higher dollar increases in earnings. Because these studies rely on administrative data for longitudinal analysis, they can only report earnings and not total household income.

Our own analysis using the 1996 and 2001 SIPP data allows us to track families after they have left welfare to see how trends in income have changed over time. We find that for the group of leavers that does not return to TANF in the year after exit, income increases for both our 1996 and 2001 cohorts. Average monthly TANF family income increases about $400 over the year for the 1996 leavers, from $1,557 to $1,982, and over $500 for the 2001 leavers, from $1,157 to $1,699.73 Both of these increases are statistically significant; however, the difference between these two periods is not statistically significant. Given that the 2001 leavers are starting with somewhat lower income at exit, the percentage increase in earnings in the later period is about 50 percent compared with 30 percent in the earlier period. This suggests that, although leavers in the later reform period have lower incomes than leavers in the early period, the average leaver in both periods who remains off welfare experiences income growth.

An important caveat to this finding is that this selects the sample of leavers who remain off welfare for a year. In our SIPP data, we find that 19 and 22 percent in 1996 and 2001, respectively, are back on TANF one year after exit. These estimates are similar to results on returns to welfare from the NSAF data (Loprest 2002). Assuming individuals with worse economic circumstances are more likely to return, these estimates of income gains may overstate the situation of all leavers taken together.

Question 4: What do we know about those leaving welfare without work or advantageous changes in family structure?

In the early period of reform, at least half of families left welfare with work, as we reported earlier. Many view this as a positive step in these families’ movement toward self-sufficiency. However, the employment of leavers has been falling over time, and families leave welfare for other reasons. Understanding these reasons and examining how they change over time provides information on changes in the group of former recipients and, potentially, their interaction with the welfare program.

Table 14 shows data from the NSAF on former recipients’ self-reported reasons for leaving welfare in 2002 and 1997.74 NSAF is the only national data set that directly asks individuals to report their reasons for exiting TANF. In 1997, 70 percent of recent welfare leavers gave work (either finding a new job or increased hours or earnings on a current job) as their reason for leaving welfare. This was by far the most common reason for leaving. The second most common reason (reported by 8.6 percent of respondents) was the respondent did not want or need benefits or continuing on benefits was “too much hassle.”

Over time, there were some significant changes in the reasons for leaving welfare. The percentage of former recipients reporting work as their reason for leaving welfare declined significantly from 70 percent in 1997 to 56 percent in 2002. Other significant changes include an increase to 17.0 percent in 2002 of those reporting they do not want or need benefits or it is “too much hassle,” potentially reflecting increased requirements for recipients. Receiving additional income from other sources or assets also increased over this time period. Also, an increasing number of recipients reported reaching the end of a time limit (0.5 percent in 1997 compared with 4.7 percent in 2002), although this is still a relatively small group of recipients. In part, this reflects that in many states, 2002 was the first year when recipients would begin to reach the federal 60 month time limit. Finally, leaving due to a change in family situation (which includes marriage) did not change over time.

Table 14 - Reasons for Leaving Welfare, 1997 and 2002 NSAF Data
	1997	2002
Earnings increased, found a job, or worked more on same job	70.0	56.0 *
Did not want or need benefits, not interested, too much hassle	8.6	17.0 *
Received additional income from other sources, assets	4.8	7.5
Did not follow program rules	6.7	6.4
Reached end of time limit	0.5	4.7 *
Change in family situation	3.8	3.8
Moved	5.4	1.4
Administrative problem, mix-up	0.2	1.3
Other	0.0	2.0

Source: Reproduced from Table 7 Loprest and Zedlewski (2006).

Note: In the first round of NSAF, 328 recent leavers were in error skipped out of questions regarding why they left welfare. These and 35 recent leavers who did provide answers in the 2002 NSAF are omitted from this table.

* Significantly different from 1997 at the 90% confidence level.

These changes in reasons, particularly the decline in work as a reason for leaving welfare, are consistent with the declining employment among leavers seen over this time period.

One concern over time has been what has happened to families that leave welfare without work or without a change in family situation such as marriage. There have been several studies of this group of former welfare recipients who are not working. Loprest (2003) and Loprest and Zedlewski (2006) use the NSAF data to define a group of “disconnected” recipients, those who are not working or living with a working spouse/partner, have not worked recently, and are not receiving cash assistance (either TANF or SSI). The authors find that 17.1 percent of recent leavers (those who had exited in the prior two years) were disconnected by this definition in 1997. Danziger et al. (2006) using the WES report that during the February 1997 and August 2003 period, 9 percent of those who left welfare in a county in Michigan were without both work and welfare for more than one-quarter of the time. As part of their evaluation of welfare reform in New Jersey, Wood and Rangarajan (2003) found that about one in four former recipients in that state were not working and off TANF in a given month. A smaller subset of this group, roughly one in ten leavers in a given month, do not have a working spouse, recent work experience, or other source of government income, such as disability benefits or unemployment insurance.

Families must have some alternate sources of income, since relatively few families actually report zero income. Bavier (2001) finds that 4 percent of leavers in the 1996 SIPP data report zero household income at exit. Zedlewski and Nelson (2003) conduct qualitative interviews with a set of families that report they have no current employment or cash government assistance and income less than half the poverty line in the 2002 NSAF. They show that a substantial percentage of these families either were in this status for a short period or should not have been included due to misreporting of income sources in the original interview.75 For families that were without employment or government cash assistance at the time of the qualitative interview, strategies for coping were complex and varied. Families put together multiple other sources of income to survive, including child support, in-kind government support (housing assistance and food stamps), help from family and friends, “side jobs,” and charity.

While these studies all provide slightly different definitions of a group of welfare recipients without work, they all find that this subset of former recipients have lower incomes, more barriers to work, and face substantially more material hardship than other welfare leavers. For example, in the NSAF in 2002, as would be expected, more than two out of five disconnected leavers have not worked in more than two years, compared with only 4.2 percent of other leavers. And well over half have not graduated from high school, more than double the rate for other leavers. In addition, these disconnected leavers are more likely to report having a health condition that limits work (29.4 percent) than other leavers (15.9 percent). The disconnected leavers also have significantly lower average incomes than other leavers, $6,178 compared to $17,681 in 2002.

Loprest and Zedlewski (2006) provide the only information on the change in the size of this group over time. They find a slight increase in the percentage of leavers that are disconnected, from 17.1 percent in 1997 to 20.8 percent in 2002, although the change is not statistically significant.76 These figures suggest a static share of leavers is disconnected over this period. However, given the decline in the number of leavers over time, these shares translate into fewer families being disconnected in 2002 than in 1997, 150,000 relative to 200,000.

Wood and Rangarajan (2003) are able to follow the same individuals longitudinally in their data for New Jersey and find substantial movement in and out of this disconnected status over time. Among leavers in this no work/no welfare group, one in four was back on TANF 12 months later, and a similar percentage was working. This indicates that families may not be in this precarious position for long periods, but may move to work or back onto welfare. But returning to welfare will not be an option forever, given time limits on benefit receipt.

Beyond the welfare population, Loprest and Zedlewski (2006) also point out that low-income women with children who have never been on welfare can also be defined as disconnected. In 2002, they find that 12.4 percent of all low-income women (with incomes less than 200 percent of poverty) not receiving welfare are disconnected by this definition. Since these women have not formerly received TANF, they cannot benefit from any programs or activities that welfare offices implement to reconnect with disconnected leavers. In this way, they are even more “disconnected.” Many of them are likely eligible for TANF.

In summary, about 20 percent of leavers are disconnected, although the absolute number of families in this situation is falling over time. This decline is tempered by the fact that many families that have never received welfare are also disconnected from the labor market and nonwelfare cash benefits. Those that are disconnected are economically worse off than other leaver families. There is some evidence that these families may remain disconnected only temporarily. However, returning to welfare is not an option for families that have hit their benefit time limit.

Summary of Findings

This section addressed the broad questions of how the characteristics and outcomes of welfare recipients and former recipients have changed over time in four issue areas. We provide a summary of our findings below.

1. In what way have the characteristics of families receiving cash assistance changed over time, and what do we know about the relationship of these changes of caseload decline?

The large declines in welfare caseload mean either fewer people are entering the program or more people are exiting or are exiting more quickly. Ultimately, changes in who enters welfare and how long they stay influence the composition of the caseload and, as a result, the characteristics of welfare leavers. In our data analysis, we find that the educational attainment of recipients seems to be sensitive to the economic cycle, with the typical recipient having more education when the economy is softer. Overall, however, we find that trends in the characteristics of families on TANF are sensitive to the precise years considered, the definitions employed, and the data sets used.

The most striking finding may be that despite the implementation of federal welfare reform, the massive decrease in welfare caseloads, and the very different economic climate during the early and late reform periods, data on the demographic characteristics of families on welfare show few statistically significant changes. Further, there is little consistent evidence of changes over time and across data sets. Similarly, data on different cohorts of welfare leavers over time indicate that in most respects, the personal and family characteristics of leavers are fairly stable. These results are similar to what has been found in existing literature.

2. Does the caseload of TANF recipients include greater percentages of families with serious barriers to work over time?

As caseloads declined, concerns grew that more of the TANF caseload would have serious barriers to work, requiring greater investment of resources to move off welfare and into the labor market. Numerous location-based studies of the welfare caseload show that significant percentages of TANF recipients have serious barriers to work, such as physical or mental health problems, recent experience of domestic violence, substance abuse, criminal history, low education levels, or learning disabilities. Our analysis of a more limited set of barriers in national data shows that current recipients generally have higher levels of barriers than former recipients.

Studies that have directly examined changes in the prevalence of barriers among recipients over time for the most part find little evidence of change in the percent of recipients with barriers in the years before 2000, with some conflicting findings on education levels and with several studies finding that the percentage of recipients with health issues has increased. Our analysis also finds increases in the percentage of family heads with a health condition that limits work and the percentage that failed to complete high school. The later period of reform shows little change in the percentage of recipients with barriers. We conclude that there is some evidence of increasing disadvantage relative to specific barriers, particularly in the early years of reform.

It is also important to note that a number of studies document a negative association between specific measured barriers and work, suggesting that those with barriers are less likely to be employed. The specific relationships vary across study locations and methods, but all find a strong negative relationship between having multiple barriers and employment. We know little about whether the relationship between barriers and work has changed over time since welfare reform, as there has been little direct analysis of this question. A few studies have found that work and exit from welfare among those with barriers on TANF have increased over time, consistent with welfare programs increasing their focus on helping these individuals move to work. However, we have little information about the outcomes for these recipients.

3. What do we know about the economic progress of TANF recipients and leavers over time?

Our analysis supports most of the literature in finding that employment increased substantially for welfare recipients during the early years after welfare reform. We also find some evidence that wages of employed recipients increased in the early time period. During the later years, particularly after 2000, we find employment rates for TANF recipients fell. Our analysis of former welfare recipients finds declines in employment between 1996 and 2001 in the SIPP as well as in later years between 2000 and 2005 in the CPS.

The preponderance of research on the reasons for the observed increases in recipients’ work in the early period of reform suggests that welfare reform played a role. This research finds that other factors such as the expansion of the EITC and the strong economy also had significant impacts on the increases in employment during this time period. There has been relatively little analysis of employment changes in the later years or changes in the employment of cohorts of former recipients over time. Whether these declines are due to a slowdown in the economy in the 2000s relative to the late 1990s or other factors is not clear. For both current and former recipients, the fact that we find little change in the personal and family characteristics of cohorts over time suggests that any compositional changes in this group may play less of role than may have been expected. However, detailed analysis of these relationships would be necessary to draw firm conclusions.

Our data analysis shows some evidence of an increase in the incomes of TANF recipients in the early period with stagnant incomes in the later period. The NSAF data, focusing on a more narrow group of welfare recipients, show income growth in both periods, although somewhat slower growth in the later period. The SIPP data show a fairly stagnant income picture over both periods of reform. However, it is important to note that income levels in the later period in the SIPP (both 1999 and 2003) are higher than income in the early period. Because the early reform period stretches from 1996 to 2001 in the SIPP rather than from 1997 to 1999 as it does in the NSAF, the SIPP early reform trend likely masks an increase in TANF recipients’ incomes through 1999 and subsequent decline. Like the SIPP, the CPS shows no significant change in the incomes of TANF recipients during the late reform period (which stretches from 2000 to 2005 in the CPS data). We also find some evidence that the distribution of income for recipients was changing over time. The NSAF data show especially dramatic declines in deep poverty (income less than 50 percent of the poverty line) in the early period. In the later period, the SIPP and CPS both suggest a more unequal distribution of income among TANF families.

On average, the incomes of former recipients are higher than current recipients. However, the three data sets we analyze show either no change or declines in the average incomes of former welfare recipients in the years after reform. Both the SIPP and CPS show significant increases in the share of leavers experiencing deep poverty.

There are a few studies that address the question of whether recipients are better off after leaving TANF than while on TANF. These studies show at most modest average gains in income for those leaving welfare and also show that there are many families whose income falls when leaving TANF. The studies point out that the key transition for raising household income seems to be moving from nonwork to work, and those families that are able to combine welfare and work may have as high incomes (or higher) than families moving off of welfare to work. These studies also show that there can be considerable difference in income changes across different localities, in part due to differences in welfare policies such as the rate at which TANF benefits are reduced as earnings increase.

Finally, there is some evidence from location-based leaver studies that, on average, in the early period after reform, earnings of leavers increased during the first year after leaving TANF, although there is a wide variation in specific results. Our analysis using SIPP data from 1996 and 2001 shows that for both cohorts of leavers, average incomes increased over the year after exit for those not returning to TANF. These results suggest that even though cohorts of welfare recipients leaving welfare in the later period of reform do not, on average, experience higher levels of income than those leaving in the early reform period, in both periods individual families that leave welfare and remain off for a year experience growth in income.

4. What do we know about those leaving welfare without work or advantageous changes in family structure?

Reasons for leaving welfare have changed over time since welfare reform, with fewer exiting for work and more exiting for other reasons. Both national and location-specific studies of the early years after welfare reform find that a majority of recipients leaving welfare exit with employment. Data from the NSAF show a significant decline in this reason for work between 1997 and 2002, consistent with our earlier results on employment of welfare leavers. These data also show a significant increase in the percentage reporting they left welfare because they do not want or need benefits or benefit receipt involves “too much hassle,” potentially reflecting increased requirements for recipients over time.

The subgroup of former recipients who leave welfare without work has been the focus of several studies using national and location-specific data sources. This group, sometimes referred to as “disconnected former recipients,” typically includes those who are not working or living with a working spouse or partner and not receiving any other cash assistance. While the size of this group varies across studies, national estimates put the size of this group as about one-fifth of those who left welfare between 2000 and 2002 and remained off welfare in 2002. Both national and location-based studies find that this group is significantly more disadvantaged than other leavers in terms of household income, barriers to work, and experience of material hardship. The only national information on change in size over time for this group is NSAF data, which find no increase in the size of this group as a percentage of all leavers. While this share remains steady over time, the absolute number of disconnected families has fallen as the number of leavers has declined (nationally, from 200,000 in 1997 to 150,000 in 2002). Evidence from New Jersey finds substantial transitions in and out of this group over time, both back to welfare and to work. Additional examination of outcomes for this group as more recipients hit their time limits and are unable to return to welfare is needed.

While the findings summarized here provide a great deal of information about the characteristics and outcomes of TANF recipients and leavers over time, they also highlight many areas where additional research is needed. The next section further describes these gaps and recommendations for future research.

23 Welfare caseloads peaked in 1994, and the decline in caseloads coincides with a growing economy and changes in state welfare policies implemented under waivers to federal AFDC rules—thus, welfare reform, in many states, predates PRWORA and the TANF program. (back to footnote 23)

24 For the most part, these are unavoidable differences or would have required additional resources beyond the scope of this project to create comparability. (back to footnote 24)

25 To enhance the power to detect statistically significant changes over time, the NSAF did re-interview some respondents from earlier rounds in later rounds of data collection. The effect of the small overlap sample is accounted for when computing standard errors using appropriate adjustments for sample design and sample weights. The data are not, however, appropriate for longitudinal analysis because there are relatively few re-interviewed families and they are not representative of the population or any particular population subgroup. See Triplett (2005) for details. (back to footnote 25)

26 The NSAF data in this report are drawn from another analysis, Loprest and Zedlewski (2006) to keep within the financial constraints of this project. Thus we are using the sample definitions used in that work. Appendix A of that paper has a detailed discussion of the impact of these income cuts on outcomes. (back to footnote 26)

27 Computing standard errors using a jack-knife technique involves obtaining a series of point estimates for the statistic in question (say, mean family income) using strategically selected subsamples (replicates) and then computing the standard error around the mean of the replicate means. Although not as precise as conventionally computed standard errors, the jack-knife estimates are not subject to problems created by unknown or highly complex correlations between observations which are difficult to completely address when constructing conventional (parametric) standard errors. For more information, see Tripplet (2005). (back to footnote 27)

28 The source for this research is an internal methodology report from the Urban Institute’s Assessing the New Federalism project titled “1999 NSAF Benchmarking Report.” (back to footnote 28)

29 Unweighted counts of welfare recipients in the NSAF are 1,458, 601, and 530 for the 1997, 1999, and 2002 rounds, respectively. Unweighted counts of welfare leavers are 1,049 and 537 for 1997 and 2002, respectively. (back to footnote 29)

30 About 20 percent of the TANF cases in the SIPP are married couple families. Because mothers tend to be the head of most TANF assistance units, we deem the mother to be the head of married couple families to make the data on characteristics more comparable across family types. (back to footnote 30)

31 We use sample weights to make samples from different panels representative, correcting for nonresponse and attrition. We assume that the quality of the attrition adjustment is similar across the different panels. If the quality of the attrition adjustments grows worse in later panels, then the trends in late panel estimates, for example 1999 to 2002, would be impacted. However, we have no reason to believe the quality of the weighting adjustments differs over time. (back to footnote 31)

32 The standard errors are computed using replicate weights, available by request from the U.S. Census Bureau. (back to footnote 32)

33 Unweighted counts of welfare recipients in the SIPP are 1,691, 646, 703, and 455 for a single month in 1996, 1999, 2001, and 2003, respectively. Unweighted counts of welfare leavers are 429 and 240 for 1996 and 2001, respectively. (back to footnote 33)

34 A family in the CPS consists of all persons residing in a household who are related to the householder/reference person through blood or marriage. The householder is the resident homeowner or leaseholder. (back to footnote 34)

35 As is the case with the SIPP, about 20 percent of the TANF cases in the CPS are married couple families. Because mothers tend to be the head of most TANF assistance units, we deem the mother to be the head of married couple families to make the data on characteristics more comparable across family types. (back to footnote 35)

36 “Other public assistance” is a catchall category that refers to the receipt of aid not reported in response to questions about specific, named programs like TANF, Food Stamps, SSI, and so on. Because state TANF programs may have different names and survey respondents may be receiving TANF and not know it, we include families reporting receipt of “other welfare” as TANF recipients. (back to footnote 36)

37 The degrees of freedom used in computing standard errors are based on the unweighted sample size. (back to footnote 37)

38 Unweighted counts of welfare recipients in the CPS are 1,084 and 1,278 in 2000 and 2005, respectively. Unweighted counts of welfare leavers are 476 and 654 in 2000 and 2005, respectively. (back to footnote 38)

39 Our TANF families are more inclusive groupings than the official Census definition of a family (all persons living with and related to a reference person by blood, marriage, or adoption) but more restrictive than the Census definition of a household (all persons sharing a housing unit, excluding group quarters). (back to footnote 39)

40 The main exception is an immigrant parent who may not be eligible to receive TANF even though her child does receive TANF. Technically, this family is a child-only case where the parent is likely not subject to welfare rules. We cannot identify these families in the data so they are included in our NSAF sample. (back to footnote 40)

41 To be included in the sample, families must have current or last year income less than 200 percent of poverty and must not have current or last year income above 250 percent of poverty. (back to footnote 41)

42 Because most of the research on welfare leavers focuses on families that left welfare in the early years of welfare reform (the 1990s), there is no satisfactory way to benchmark our statistics against prior research. The work on families that left welfare in the later years of welfare reform (post-2000) largely is based on data from the 2002 NSAF (e.g., Loprest and Zedlewski 2006), and our tabulations of the NSAF data are drawn from this work. (back to footnote 42)

43 Futher, Klerman and Haider (2001) show that direct models of the aggregate, as opposed models of entry and exit, are misspecified. (back to footnote 43)

44 Bavier (2003) examines entry and exit by different barriers to work. We discuss this in a later section. (back to footnote 44)

45 Differences in the family definitions between the SIPP and NSAF samples may account, at least in part, for this difference in age. Recall that SIPP welfare families include families in which the child receives welfare but the child’s parent is not in the family; thus, the head of the family could be a grandparent. In the NSAF sample, the child’s parent is always in the family. (back to footnote 45)

46 All administrative cited in this report can be reached through the following website: http://www.acf.hhs.gov/programs/ofa/character/indexchar.htm (last accessed January 30, 2007). The tabulated administrative data available here are not consistently broken out across demographic categories in the same way we break out data from the survey data sets. Also, the survey data sets can only approximately replicate the TANF cases that are the basis for the administrative data. Further, administrative data report information for adults and children in the assistance unit, but we focus on the characteristics of the adult we deem is the head of the TANF family in our survey data sets. Finally, as noted in section 2, the quality of administrative data is not necessarily better (and may even be worse) than the quality of survey data for elements that are not germane to the computation of eligibility and benefits. (back to footnote 46)

47 Race/ethnicity data are available on the NSAF and can be used for 1997 and 2002; however, the race/ethnicity information for the 1999 round of the NSAF is problematic. See the earlier discussion for more details. (back to footnote 47)

48 Although these tabular data comparisons cannot tell us why the educational attainment of welfare recipients falls and then rises, one might speculate that more educated individuals were less likely to enter welfare and more likely to leave welfare than less educated individuals when the economy was particularly strong during the 1990s and that as the economy cooled after 2000, even the more educated came on to or remained on the welfare rolls. There may well be other explanations for this trend. Understanding the reasons for the trend in educational attainment among welfare recipients would be an interesting research topic. (back to footnote 48)

49 See earlier discussion and Table 1 for the differences in definitions of families in the SIPP and NSAF. (back to footnote 49)

50 The SIPP and CPS designate a reference person and consider relationships to that reference person; consequently, these data cannot be used to consistently identify cohabiting families. (back to footnote 50)

51 To be considered a TANF leaver in the SIPP, a family had to remain off TANF for two consecutive months; once leaver status has been determined, the characteristics of SIPP TANF leavers are measured in the month of exit. (back to footnote 51)

52 See table 1 and earlier discussion for details. (back to footnote 52)

53 Few location-specific studies report in-depth information on the prevalence of barriers among former TANF recipients. (back to footnote 53)

54 The six are Illinois; Colorado; Washington, D.C.; Maryland; Missouri; and South Carolina. (back to footnote 54)

55 Variation from these definitions by other studies is noted in the table. (back to footnote 55)

56 In Loprest and Zedlewski (2006) the list of barriers includes low work experience, less than high school education, child less than age one, child receiving SSI, Spanish-speaking, without a car living outside of an MSA, poor health, and poor mental health. (back to footnote 56)

57 Moffitt and Cherlin (2002) have a good discussion of these issues. (back to footnote 57)

58 Because the first interview in the 1996 panel occurred between December 1995 and March 1996, these month estimates do not fully represent the caseload for the year. (back to footnote 58)

59 See Table 1 for details on the differences in definition of leavers across these data sets. (back to footnote 59)

60 In fact, some welfare programs are trying to specifically address issues such as low self-esteem through their work programs and case managers. (back to footnote 60)

61 Administrative data are taken from http://www.acf.hhs.gov/programs/ofa/character/FY2000/analysis.htm#trends. (back to footnote 61)

62 All wage data are shown in constant 2005 dollars. (back to footnote 62)

63 We report median rather than mean wage given the skewed nature of the distribution of wages. Calculation of standard errors for medians is not straightforward. For NSAF, we report tests of significance from Loprest and Zedlewski (2006) that used a bootstrapping technique to develop standard errors around the median. Significance tests on medians from the SIPP and CPS are computed using standard errors around the mean as a proxy for the median standard errors. Because mean wages are always higher than median wages, this is a conservative test that understates potential significance. (back to footnote 63)

64 Specifically, the SIPP data include family heads that are not the parent of the child and therefore are not necessarily meeting income-eligibility thresholds. These heads may have higher incomes and greater access to employer insurance. (back to footnote 64)

65 Wage data is in constant 2005 dollars. (back to footnote 65)

66 We would expect wages of working leavers in the NSAF to be higher than working leavers in the SIPP because NSAF leavers have been off of welfare for some time and those that have returned to welfare are excluded. Working leavers’ wages in the SIPP are measured at the time of exit, and the sample includes those who later return to welfare, which is likely to include many with lower wages. However, this difference in sample does not exist for TANF recipients, where we also see difference in wage levels between NSAF and SIPP. (back to footnote 66)

67 Reported monthly income is multiplied by 12 to create an estimate of annual household income in the SIPP data and the NSAF data. Monthly income at the time of the interview in NSAF includes current monthly earnings for the MKA and spouse or partner (if relevant), current TANF receipt, and an imputation of other income based on last year’s sources of income. This assumes that relatively permanent sources of income in the prior year continue into the current month. Since the vast majority of income for current recipients is made up of TANF and earnings, this is a good approximation of current monthly income (Loprest and Zedlewski 2006). The CPS data are reports of annual prior year income. (back to footnote 67)

68 Income data are in constant 2005 dollars. Calculation of standard errors for medians is not straightforward. For NSAF, we report tests of significance from Loprest and Zedlewski (2006) that used a bootstrapping technique to develop standard errors around the median. Significance tests on medians from the SIPP and CPS are computed using standard errors around the mean as a proxy for the median standard errors. Because mean income is always higher than median income, this is a conservative test that understates potential significance. (back to footnote 68)

69 Income data are in constant 2005 dollars. (back to footnote 69)

70 The exact income sample criteria are given in the first part of this section. (back to footnote 70)

71 The WES data do not collect a complete income profile at the time families are on welfare and selected into the sample. The first survey data are collected three months later, when some families have already exited welfare. (back to footnote 71)

72 This increase could be due to changes in nonwork income sources, such as earnings of other family members. (back to footnote 72)

73 This includes all leavers who remain off welfare for a year, whether or not they are working at exit. (back to footnote 73)

74 Survey respondents gave open-ended answers that were then coded into categories. About 5 percent of leavers reported multiple reasons for leaving. These respondents are categorized using a hierarchy of responses with employment first. Administrative data on reasons for leaving are of more limited use than survey data because often families leaving welfare do not report the reason for leaving (such as work or marriage) to the welfare office. These families often appear in administrative data as case being closed due to lack of contact or not providing paper work for recertification. For more discussion, see Acs and Loprest (2004). (back to footnote 74)

75 Of the families interviewed, 13 percent had experienced a change in income circumstances since the original NSAF interview, largely due to work by the respondent or spouse. Another 31 percent were found to have had earnings or government cash assistance that was misreported in the original survey (Nelson and Zedlewski 2003). (back to footnote 75)

76 Loprest (2003) finds an increase in the percentage of leavers that were disconnected between 1997 and 1999 using NSAF data and this same definition. (back to footnote 76)

Table of Contents | Previous | Next