[ Main page of Report | Contents of Report ]
[ Go to Contents ]
The data used in the report come from five MDRC evaluations of welfare reform initiatives, including two non-experimental analyses. Three experimental (random assignment) evaluations tested several key policies that are now part of the states' Temporary Assistance for Needy Families (TANF) programs. In these evaluations, ongoing recipients or new applicants to welfare were assigned to either a program group that received the new services or financial incentives or were subject to new regulation on eligibility for welfare benefits or to a control group that was subject to the existing welfare system in the state at the time of the evaluation. New applicants for welfare were randomly assigned at the time they were applying for welfare, while ongoing recipients were randomly assigned at their eligibility re-determination interviews.
The two non-experimental analyses studied the effect of welfare reform on caseloads and recipient behavior as part of MDRC's Project on Devolution and Urban Change (hereinafter, Urban Change). These analyses utilized data on the universe of recipients who received welfare benefits at any time during the early- to late-1990s.
A brief description of the five evaluations and a summary of key findings follow below:
The Connecticut's Jobs First Program implemented a 21-month time limit on welfare receipt. The program also included very generous financial incentives to encourage work - all of the recipient's earnings were disregarded when calculating her welfare and food stamps benefits until her earnings reached the poverty line. The evaluation tested the effects of Jobs First among new applicants for assistance and ongoing recipients in the offices of Manchester and New Haven (although the program was run statewide).
Jobs First increased the employment and earnings of program group members above control group levels throughout the evaluation's four-year follow-up period. During the early part of the follow-up period, before recipients began reaching time limits, the program also increased welfare receipt and, because of the generous disregards, increased incomes. About half of the recipients in the program group reached the time limit during the period, and two-thirds of them subsequently received a 6-month extension. Those who left welfare because of a time limit had higher earnings than those who left earlier, most likely because of the extension policy. See Bloom et al. (2002) for more information.
The Florida's Family Transition Program imposed a 24-month time limit in any 60-month time period for most recipients and a 36-month time limit in any 72-month period for the least job-ready. The program offered financial incentives as well as enhanced services designed to help recipients find jobs. FTP operated as a pilot program in Escambia County (which includes the City of Pensicola) from 1994 to 1999. Many of FTP's features were then incorporated into Florida's statewide TANF program. There were important differences, however, between the two programs. For example, FTP did not have full-family sanctions, whereas Florida's statewide TANF program did. The FTP evaluation tested the program's effects among single-parent applicants for welfare and ongoing recipients. FTP increased program group members' employment and earnings and reduced their welfare receipt, compared with the control group. The majority of those in the program group left welfare before reaching their time limit. Most of the recipients who did reach a time limit were not given extensions: about 40 percent had adequate earnings levels that did not warrant an extension, while most of the rest were deemed non-compliant. The group that reached a time limit and had benefits cancelled was somewhat more disadvantaged than other leavers, but they did not, on average, experience more hardship, partly because they relied more on other sources of support. See Bloom et al. (2000) for the final report on the program's effects.
Vermont's Welfare Restructuring program implemented a 30-month work trigger that required most single parents to work once they had received welfare for 30 cumulative months. The program also included financial incentives in the form of an enhanced earned income disregard that was somewhat more generous than under Vermont's old AFDC program. WRP operated statewide from July 1994 through June 2001.
The evaluation tested the effects of WRP in six welfare districts. The research sample included single parents and adult members of two-parent families who were applying for welfare benefits or were ongoing recipients at their time of random assignment. The evaluation used a three-group research design, in which one program group received the enhanced financial incentives with the work trigger and a second program group received just the incentives. This design tested the effects of incentives alone compared with the effects of incentives combined with the work trigger.
The full WRP program produced modest increases in employment and earnings and decreased welfare payments over a six-year follow-up period, although WRP had little effect on welfare receipt prior to the 30-month point when families were subject to the work requirement. The WRP Incentives Only program had little impact on employment rates and welfare payments, indicating that the program's work requirements were needed. See Bloom et al. (2002) for more detail on the program's effects.
The Project on Devolution and Urban Change (Urban Change, hereafter) is a five-year, non-experimental multi-component study of PRWORA's implementation and of its effects on poor families with children, the communities in which they live, and the institutions that assist them. The study takes place in and around four major urban centers: Cleveland, Los Angeles, Miami, and Philadelphia.(2)
a. Urban Change - Cleveland (Cuyahoga County)
Cuyahoga County remade its welfare system in response to TANF by shifting to a neighborhood-based delivery system and dramatically increased the percentage of recipients who participated in work activities. It also launched a major initiative to divert families from going on welfare. The county firmly enforced the statewide 36-month time limits, starting in October 2000, but it ensured that families were aware of their cutoff date, and it offered short-term extensions and transitional jobs to recipients who had employment barriers or no other income.
The Cleveland sample includes 536,256 recipients, the universe of all people (adults and children) who ever received Medicaid or food stamps from July 1992 through December 2000. The study found that between 1992 and 2000, welfare receipt declined in the county, and employment among welfare recipients increased. A longitudinal survey of former and ongoing welfare mothers in Cleveland's poorest neighborhoods showed substantial increases in the percentage that were working and had "good" jobs between 1998 and 2001.
The study's findings counter the notion that welfare reform would lead to service retrenchment and a worsening of conditions for families and neighborhoods. To the contrary, there were many improvements in Cleveland - though the favorable economy played a major role, and time limits had just been implemented when the study ended. See Brock et al. (2002) for more information.
b. Urban Change - Philadelphia County
The Urban Change Philadelphia project evaluated the effect of Pennsylvania's welfare reform on welfare receipt, employment, material hardship, and neighborhoods. The state focused its welfare-to-work program on employment, expanded and simplified the provisions that allowed welfare recipients to keep part of their welfare checks if they worked, and instituted two time limits: a 24-month limit that requires recipients to work or participate in a work activity for 20 hours per week and a 60-month lifetime limit on welfare receipt. In Philadelphia (3), implementation of the law was lenient in some respects. During the first two years on welfare, recipients were asked to conduct an eight-week job search but otherwise were not held to a strict work requirement. At the 24-month limit, many parents who were not working were placed in subsidized jobs. In addition, families received extensions to the lifetime limit if they participated in assigned activities.
The Philadelphia sample includes 778,510 recipients, all people who received cash assistance, food stamps, or Medicaid between January 1993 and July 1999. In Philadelphia, welfare receipt declined and employment increased between 1992 and 2000. TANF seems to have encouraged long-term recipients to leave the rolls faster, to have increased employment (but mostly unstable employment), and to have raised the likelihood that some families would return quickly to welfare.
This study's findings are consistent with the above Urban Change report on Cleveland, again countering the notion that welfare reform leads to service retrenchment and a worsening of conditions for families and neighborhoods. See Michalopoulos et al. (2003) for more information.
[ Go to Contents ]
Each evaluation provides three data sources for this report. First, demographic data, including age, race, and number and ages of children, were collected for all sample members at the point of random assignment, or baseline.(4) Second, state and county administrative records provide information on sample members' quarterly earnings, monthly welfare receipt, and monthly food stamps receipt. The earnings data come from each state's Unemployment Insurance (UI) system, to which most employers must report employee earnings. Earnings from self-employment, informal or "off-the-books" jobs, or employment from the federal government or the military are not reported, however. The welfare and food stamps data come from each state's automated benefit payment system. One limitation of these statewide administrative data is that they do not capture earnings or benefit receipt for sample members who have moved out of state.(5) Third, each study administered a survey to a subset of the sample about three to five years after sample entry. The surveys capture employment and earnings not reported to the UI system, and they also provide more detailed information on family well-being, including household composition, income and income sources, material hardship, and barriers to employment.
[ Go to Contents ]
Table 2 presents for each site the sample size and the "sample intake period" - that is, the dates from which the samples were drawn. The sample intake period for the three evaluation sites consists of the months in which welfare applicants and recipients were randomly assigned to program and control groups. We selected intake periods for the Urban Change sites to mirror those from the random assignment evaluations in order to make the analyses across these disparate samples more comparable. That is, for the Urban Change sites, the sample intake months correspond roughly to the months of random assignment for the three evaluation sites. In these sites, the month of sample intake is the first month within the designated intake period in which the individual received a welfare payment.
The analyses focus on a large subgroup of the five evaluation samples. Specifically, the sample for this report includes single parents (usually, mothers) aged 18 to 59 at sample intake who received at least one welfare payment during or after their month of sample intake. Welfare recipients excluded from this analysis include: (1) single parents younger than 18 or older than 59 years of age at sample intake; (2) members of two-parent welfare cases; and (3) adult caretakers of Child-Only or Foster Care assistance cases.(6) Together, these excluded groups made up between 20 and 30 percent of the welfare caseload in these sites.(7)
In the two Urban Change sites, Cleveland and Philadelphia, all welfare recipients who met the criteria listed above were included in the research samples. In contrast, the samples in the three evaluation sites are somewhat less representative of their respective caseloads in that they also exclude welfare recipients who were determined by welfare agency staff members to be exempt from their welfare program's requirement to participate in pre-employment activities. Exempted single parents included (1) incapacitated or disabled adults; (2) parents of a child under the age of one year (6 months of age in Florida FTP); and (3) adults caring for a disabled child or other dependent relative. These groups represent another 12 to 20 percent of the caseload.(8) Finally, most analyses in this report exclude sample members randomly assigned to the control group, which made up a quarter of the sample in Vermont WRP and half in Connecticut Jobs First and Florida FTP.(9)
| Sites | Sample Site | ||||
|---|---|---|---|---|---|
| Location | Full Sample | Five-Year Follow-Up Sample |
Survey Sample | Sample for Testing Effects of PRWORA |
|
| Evaluation sites | |||||
| Connecticut Jobs First | Manchester and New Haven | ||||
| Sample intake period | 1/96 - 2/97 | -- | 4/96 - 1/97 | -- | |
| Sample size | 2,184 | -- | 1,157 | -- | |
| Florida FTP | Escambia County | ||||
| Sample intake period | 5/94 -2/95 | 5/94 -09/94 | 8/94 - 2/95 | -- | |
| Sample size | 1,150 | 535 | 711 | -- | |
| Vermont WRP | State-wide | ||||
| Sample intake period | 7/94 -6/95 | 7/94 -6/95 | 10/94 - 6/95 | -- | |
| Sample size | 4,051 | 4,051 | 781 | -- | |
| Urban Change sites | |||||
| Cleveland | Cuyahoga County | ||||
| Sample intake period | 7/94 - 12/96 | 7/94-01/96 | 7/94 - 10/95 | 1/93-12/96 | |
| Sample size | 55,764 | 50,217 | 887 | 26,365 | |
| Philadelphia | Philadelphia County | ||||
| Sample intake period | 1/94 - 12/97 | 1/94-12/96 | 1/94 - 7/95 | 1/93-12/97 | |
| Sample size | 97,858 | 91,348 | 749 | 49,067 | |
| Total sample size | 161,007 | 146,151 | 4,285 | 75,432 | |
| Sources: MDRC calculations from
state and county administrative records, Background Information Forms, and
survey responses. Notes: Sample sizes for the evaluation sites include program group members only. Control group members number: Connecticut Jobs First: 2,102; Florida FTP: 1,124; and Vermont WRP: 1,004. Control group members are excluded from all calculations except those displayed on Table 12. The study sample is limited to single-parent adults, ages 18 to 59, meeting study criteria. |
|||||
Table 2 shows reasonable sample sizes for the random assignment evaluations and large sample sizes for the non-experimental Urban Change sites. Sample intake took place during 1994 through 1997 and varied in duration from 10 months in Vermont WRP to 48 months in Philadelphia.
As shown in Table 2, the report calculates the incidence of cycling and other outcomes for several different subsamples. The largest sample, called the full sample, is used for most analyses in this report. The full sample includes 161,007 members from the five sites who met the sample selection criteria described above. Each member of the full sample has employment and welfare data, collected from administrative records, from two years before through at least four years after her month of sample intake.
The five-year follow-up sample includes the 146,151 members (about 91 percent) of the full sample who have employment and welfare data for at least five years following their month of sample intake. This sample is used for the analyses of employment and welfare outcomes in year 5 after sample intake.(10) Table 2 shows that all members of the full sample in Vermont WRP have administrative data for five years after sample intake, as do more than 90 percent of the full sample from Cleveland and Philadelphia. In contrast, the five-year follow-up sample includes only 535 full sample members (about 47 percent) from Florida FTP and no full sample members from Connecticut Jobs First.
The survey sample consists of the 4,285 members of the full sample who completed survey interviews between 36 and 60 months after their date of sample intake. The survey sample includes respondents from all five sites and is used for analyses of outcomes such as household income, household composition, job characteristics, and material hardships, which were unavailable from administrative data.
The strategies for fielding a survey sample varied somewhat in the three evaluation sites.(11) However, in each evaluation site, survey respondents were chosen from nearly all months of sample intake. This strategy increased the likelihood that results for survey respondents could be generalized to the full samples in these sites.
In contrast, for the Urban Change sites, a more specialized sampling design was followed in which respondents were selected from among all recipients of welfare or food stamps benefits in each site during a single month, May 1995. In both Cleveland and Philadelphia, this month falls toward the beginning of the sample intake period for this report. By definition, this sampling strategy excludes members of the full sample who did not receive welfare or food stamps benefits in May 1995 - either because they had already left assistance by that date or because they had not yet applied. As a result, the survey samples in Cleveland and Philadelphia are somewhat less representative of the full samples, compared with the samples in the three evaluation sites.
Finally, we use a special sample from the Urban Change sites to evaluate the effects of PRWORA on the likelihood of becoming a cycler.(12) This sample for testing the effects of PRWORA includes full sample members who had received their first welfare payment between January 1993 (the first month of available data)(13) and their month of sample intake. That is, they were either first-time applicants for welfare benefits at their time of sample intake or they had received welfare for the first time between January 1993 and their sample intake date. This sample for testing the effects of PRWORA also includes recipients not in the full sample. This additional "early cohort" was included to better detect the effect of welfare regulations on cycling during the years before PRWORA. The members of this early cohort first received welfare benefits on or after January 1993, but before the first month of sample intake for the full sample. For example, the early cohort includes those recipients who received their first welfare benefit between January 1993 and January 1994(14), but did not receive welfare benefits during the sample intake period. These samples are used since the longitudinal nature of the samples allows a quasi-experimental estimate of the effect of welfare reform. The early cohort members are not part of the full sample and are excluded from all other analyses in this report.
We analyze the effect of PRWORA separately in each Urban Change site. Table 2 shows that there are 26,365 and 49,067 members of this special sample in the Cleveland and Philadelphia, studies respectively.
For the majority of the analyses in this report, sample members from all five sites are combined into a single "pooled" sample weighted equally by site. This means that even though Philadelphia represents nearly 61 percent of recipients in the unweighted sample, they represent only 20 percent of the weighted sample. Other times, we present results separately for each site.
Appendix Table 1 displays the characteristics of sample members from each site, as well as for the pooled sample, equally weighted. As can be seen from the Full Sample column, the sample includes primarily females, around 30 years of age, with one or two children. Most sample members had at least one child under the age of 6. The sample was relatively evenly divided among whites and African Americans. Hispanic sample members comprised less than 10 percent of the sample. About 60 percent of sample members were ongoing recipients at their time of sample intake. A similar percentage worked for pay during the two years prior to sample intake, but their work history was limited. This subgroup of sample members with prior employment averaged about one year (4 quarters) of employment (2.2 quarters divided by 0.555) in the two years before sample intake.
The five welfare samples in this report do not represent a random sample of the national caseload. Most notably, all sites are located in the eastern half of the nation. In other ways, however, the sites encompass much of the variation of welfare populations across the U.S. The full sample includes welfare recipients from one of the nation's largest urban centers (Philadelphia), several medium-sized cities (and surrounding suburbs), and some rural areas (especially, Vermont). Sample members also come from states with relatively high welfare grants (Connecticut and Vermont), low welfare grants (Florida and Ohio), and grants near the national average (Pennsylvania).
To measure the representativeness of the full sample, we compare the background characteristics of the full samples to published data on the characteristics of single adult welfare recipients in the national caseload. For this comparison, we use data from FY 1996 (15), which falls within the sample intake period in three sites and begins shortly after the last month of intake in the other two. Members of the full sample closely resemble adults in the national caseload in sample members' gender, age, and average number of children, but had a somewhat higher percentage of children below the age of six. The full sample contains a larger percentage of whites and African-Americans and a smaller percentage of Hispanics than adults in the national caseload. Finally, members of the full sample were much more likely to be new recipients and to have entered assistance with a recent work history compared with adults in the national caseload. These similarities and differences should be kept in mind when making generalizations of the findings of this report.
For most analyses in this report, we track sample members' welfare and other outcomes from their month of sample intake through the end of year 4. These 48 months are referred to as the observation period. The fifth-year follow-up period includes months 49 through 60, following the month of sample intake and is used to measure outcomes for the five-year follow-up sample. In contrast, the period for tracking survey outcomes is less exact. It ranges from 36 to 60 months after sample intake and reflects differences among the five studies in when interviews were scheduled. Thus, some members of the survey sample reported outcomes that occurred during years 3 or 4 of the observation period, whereas others reported outcomes that took place during the fifth-year follow-up period.
For this report, we define a welfare spell as a series of consecutive months of welfare receipt. In keeping with previous research, we allow for interruptions of one month when measuring the length of welfare spells, but end welfare spells after two consecutive months without a payment. For this analysis, we focus only upon welfare spells that include the sample member's month of sample intake or began later during the observation period.
For several analyses, we will present findings separately for sample members who were just starting a welfare spell around their time of sample intake (hereafter referred to as new recipients) and sample members who had been receiving benefits for some time before the sample intake period (hereafter referred to as ongoing recipients). More specifically, a new recipient is either (1) a sample member who received welfare during her month of sample intake and began her welfare spell no earlier than 3 months prior to sample intake; or (2) a sample member who received no payment during her month of sample intake but started a welfare spell within the next 1 or 2 months. (16) New recipients who entered the research sample during the same month of sample intake may be thought of as entry cohorts. (See Tables 4 and 12 and Figures 3 through 6 for results for this subgroup.) Ongoing recipients were receiving welfare during their month of sample intake and began their welfare spell at least four months prior to sample intake.
New recipients are of particular interest for the study of cycling because we can accurately measure the duration of their first spell on welfare. For this reason, the sample for testing the effects of PRWORA is comprised solely of new recipients. In contrast, some ongoing recipients have welfare spells that began before the data collection period for this report, and estimates of their total number of months on assistance are necessarily truncated. Still, ongoing recipients represent a large portion of the welfare caseload - between 55 percent and 66 percent of the sample members in each site. Their outcomes, when combined with those of new recipients, make the findings on welfare cycling more representative of the welfare recipients in the sites included in this study.
[ Go to Contents ]
It is important to get an idea of the economic environment present over time in each site, as this may affect the outcomes of cyclers. Table 3 shows selected environmental statistics for each site over an eight-year period. In general, employment growth over the years of 1994 to 2001 ranged from around 2 to 3 percent in four sites and by more than 9 percent in Vermont. Growth in employment was greatest in Vermont and slowest in Cleveland.
The table also shows that unemployment rates generally declined over the period of 1994 to 2000. This decline was largest in Connecticut, where the unemployment rate decreased by 3.6 percentage points over the period. In 1994, Connecticut, Cleveland, and Philadelphia experienced the highest levels of unemployment, an average rate of 6.6 percent across these sites. However, by 2001, unemployment in Connecticut declined to levels comparable to Florida and Vermont.
Consistent with national trends, welfare caseloads declined dramatically (between 44 and 77 percent) in all sites, with the greatest decrease occurring in Florida, followed by Cleveland. (17) This large change in Florida may be explained by the extraordinary changes in both the state and federal welfare policy during the implementation years. For example, in addition to the Florida legislature voting to expand FTP from one county to several other Florida counties, it then passed the state's welfare reform act in May 1996, and the federal welfare reform act was passed just 3 months later. All of these changes were widely publicized. Another possible explanation for Florida's relatively large caseload decline may be its very high caseload level in the early 1990s. Perhaps because the state's caseload was so high, the rate of decline since that time has been much greater in Florida than most other states (the national caseload declined by 49 percent during the same period). From the Urban Change report, we know that the dramatic decline in caseloads in Cleveland occurred prior to the implementation of welfare reform and may be partially due to the strong economy or other factors.
| Characteristic | Evaluation Sites | Vermont WRP | Urban Change Sites | ||
|---|---|---|---|---|---|
| Connecticut Jobs First | Florida FTP | Cleveland | Philadelphia | ||
| Total employed (a) | |||||
| 1994 | 306,110 | 115,294 | 42,674 | 632,589 | 604,573 |
| 1995 | 301,037 | 114,850 | 43,556 | 639,911 | 594,381 |
| 1996 | 302,451 | 114,327 | 44,273 | 641,649 | 596,100 |
| 1997 | 305,349 | 114,448 | 45,091 | 648,191 | 597,975 |
| 1998 | 309,957 | 115,464 | 46,007 | 643,281 | 593,325 |
| 1999 | 309,826 | 115,612 | 46,873 | 645,339 | 592,102 |
| 2000 | 320,244 | 114,813 | 46,425 | 646,281 | 623,460 |
| 2001 | 317,015 | 119,104 | 47,449 | 643,402 | 625,031 |
| Employment growth, 1994 - 2001(%) | 3.4 | 3.3 | 9.5 | 1.7 | 3.4 |
| Unemployment rate (%)(b) | |||||
| 1994 | 6.0 | 4.7 | 4.8 | 6.0 | 8.0 |
| 1995 | 5.8 | 4.3 | 4.3 | 5.0 | 7.7 |
| 1996 | 6.1 | 4.1 | 4.7 | 5.1 | 7.1 |
| 1997 | 5.4 | 4.2 | 4.1 | 4.9 | 7.0 |
| 1998 | 3.5 | 3.9 | 3.5 | 4.5 | 6.2 |
| 1999 | 3.3 | 3.6 | 3.1 | 4.6 | 6.1 |
| 2000 | 2.4 | 4.0 | 3.0 | 4.6 | 6.1 |
| 2001 | 3.6 | 5.0 | 3.7 | 4.6 | 6.4 |
| Welfare caseload (c) | |||||
| 1994 | 59,200 | 254,032 | 9,917 | 251,037 | 208,260 |
| 1995 | 61,000 | 241,193 | 9,789 | 232,574 | 208,899 |
| 1996 | 58,124 | 215,512 | 9,210 | 209,830 | 192,952 |
| 1997 | 56,095 | 182,075 | 8,451 | 192,747 | 170,831 |
| 1998 | 51,132 | 121,006 | 7,591 | 147,093 | 140,446 |
| 1999 | 35,481 | 89,674 | 6,717 | 121,142 | 110,567 |
| 2000 | 28,095 | 67,355 | 6,043 | 97,969 | 89,899 |
| 2001 | 25,650 | 58,849 | 5,524 | 85,005 | 82,644 |
| Change in welfare caseload, 1994 - 2001 (%) | -56.7 | -76.8 | -44.3 | -66.1 | -60.3 |
| Maximum welfare grant for a family of 3 during first month of sample intake($) | 636 | 303 | 638 | 341 | 403 |
| First month of sample intake | 1/96 | 5/94 | 7/94 | 7/94 | 1/94 |
| Sources: MDRC calculations from
data collected from U.S. Department of Labor, Bureau of Labor Statistics
website; U.S. Department of Health and Human Services, Administration for
Children and Families website; and U.S. House of Representatives, Committee
on Ways and Means, Green Books,1994-2000. Notes: (a): Employment totals are monthly averages, not seasonally adjusted. (b): Unemployment rates are monthly averages, not seasonally adjusted. (c): Welfare caseloads totals are state monthly averages. |
|||||
The bottom of Table 3 reports the maximum welfare grant for a family of three in 1994. Connecticut is the most generous state, while Florida is the least generous with a maximum benefit of just $303 for a family of three.(18)
[ Go to Contents ]
For most analyses in this report, we divide the samples into three key outcome groups, based on each sample member's pattern of welfare receipt: cyclers, short-term recipients, and long-term recipients, this grouping reflects definitions used in the literature (e.g., Ver Ploeg, 2002), combined with an examination of the full sample. In this report we define a cycler as someone who had 3 or more spells of welfare receipt during the 4-year observation period. (19) A short-term recipient is defined as someone who had 1 or 2 spells and a total of up to 24 months of welfare receipt during the observation period. Long-term recipients are defined as sample members with 1 or 2 spells and a total of 25 to 48 months of welfare receipt during the observation period.
Table 4 presents the percentage of cyclers, short-term recipients, and long-term recipients by site for the full sample (top panel) and for new recipients (bottom panel). The table shows results for each site as well as pooled (5-site) averages.
| Welfare Outcome(%) | Evaluation Sites | Vermont WRP | Urban Change Sites | Total | ||
|---|---|---|---|---|---|---|
| Connecticut Jobs First | Florida FTP | Cleveland | Philadelphia | |||
| A. Full Sample | ||||||
| Cyclers | 4.5 | 13.7 | 9.4 | 10.8 | 3.8 | 8.5 |
| Short-term recipient | 52.1 | 59.2 | 42.4 | 45.0 | 34.5 | 46.6 |
| Long-term recipient | 43.4 | 27.0 | 48.2 | 44.1 | 61.7 | 44.9 |
| Sample size | 2,184 | 1,150 | 4,051 | 55,764 | 97,858 | 161,007 |
| B. New Recipients | ||||||
| Cyclers | 6.1 | 11.7 | 11.4 | 11.6 | 4.9 | 9.1 |
| Short-term recipient | 60.7 | 76.0 | 56.5 | 59.0 | 49.6 | 60.3 |
| Long-term recipient | 33.3 | 12.4 | 32.2 | 29.4 | 45.5 | 30.5 |
| Sample size | 860 | 420 | 1,383 | 23,657 | 44,164 | 70,484 |
| Sources: MDRC calculations from
state and county administrative records. Notes: The samples were weighted equally by site when calculating percentages for the Full Sample and for New Recipients. The Full Sample includes 6.5 percent cyclers, 38.7 percent short-term recipients, and 54.8 percent long-term recipients, when samples are pooled without weighting. The corresponding percentages among New Recipients are: cyclers: 7.4 percent; short-term recipients: 53.4 percent: and long-term recipients: 39.2 percent. |
||||||
For the full sample, rates of cycling range from a low of 3.8 percent in Philadelphia to 13.7 percent for Florida FTP. Differences in maximum grant levels, earnings disregards, and the imposition of time limits on eligibility to receive welfare benefits are potential reasons for the variation across sites in the incidence of cycling. Connecticut, for example, has much higher benefit levels and more generous earnings disregards than Florida, which may explain its lower rate of cycling. However, Vermont's benefit levels and earnings disregards are similar to Connecticut's, but the WRP's rate of cycling is twice at high (9.4 percent), which may be due to the strong economy in Vermont during this time period. Still, it is apparent that cyclers represent a small fraction of the full samples in each site. In Connecticut Jobs First and Florida FTP short-term recipients comprised the largest portion of the sample, whereas in Vermont WRP and, especially, Philadelphia, long-term recipients predominated. In Cleveland, about the same percentage of sample members were short-term and long-term recipients.
The bottom panel shows the same results for new recipients. For this subgroup, rates of cycling were roughly similar to those for the full sample, ranging from 4.9 percent in Philadelphia to 11.7 percent in Florida FTP. In four sites, the incidence of cycling among new recipients exceeded the rate for the full sample by about 1 to 2 percentage points. The exception was Florida FTP, where the incidence of cycling was 2 percentage pointers lower. New recipients differed more dramatically from the full sample in their relative proportions of short-term- and long-term recipients. In all sites except Philadelphia, a large majority of new recipients became short-term recipients during the four-year observation period - especially in Florida FTP (76 percent). Furthermore, even in Philadelphia, short-term recipients comprised the largest group.(20)
To place the results for full sample is broader context, we also calculated the incidence of cycling over four years among program group members in six of the seven sites evaluated in the National Evaluation of Welfare-to-Work Strategies (NEWWS) project. (21) We compare the sites in this report to the NEWWS sites because NEWWS employed a random assignment design and tracked sample members' welfare and employment outcomes over a similar observation period.(22) Figure 1 shows that the rates of cycling in the NEWWS sites ranged from 4.4 percent in Detroit to 11.7 percent in Grand Rapids, similar to results for the five sites in this study (also shown in the figure).
The incidence of cycling in the sites in Figure 1 is lower than the rates calculated in other studies that defined cycling as receipt of welfare during multiple spells. For instance, Moffitt (2002) found that 20 percent of individuals who had ever been on welfare were cyclers. Ver Ploeg (2002) found that cyclers represented about 14 percent of the sample. Moffit and Ver Ploeg examined cyclers over ten and nine year periods, respectively.
The four-year follow-up period for this study accounts for at least part of this difference in measured rates of cycling. Figure 2 shows cycling rates over this period compared with a longer follow-up period. When measured over five years (for the five-year follow-up sample), the incidence of cycling increased by 2 to 4 percentage points. Around 15 percent of sample members in Florida FTP, Vermont WRP, and Urban Change-Cleveland became cyclers by the end of year 5, compared with fewer than 6 percent for Urban Change-Philadelphia. (23)
Sources: MDRC calculations from state and county administrative
records.
Notes: Calculations for National Evaluation of Welfare-to-Work Strategies
(NEWWS) sites and for Evaluation sites are for program group members only.
A single (combined) program group was created in sites that randomly assigned
individuals to two or more program groups.
Figure 2.
Percentage of Sample Members Who Became Cyclers by Site
Sources: MDRC calculations from state and county administrative
records.
Notes: The full sample was included in the calculations for years
1-4, where as the five-year follow-up sample was used in the calculations
for years 1-5. The Connecticut Jobs First sample lacked five years of follow-up
data and was excluded from the calculations for years 1-5. See
table 2 for the sample sizes and intake dates for
each sit.
The percentage of cyclers during years 1-4 for the 5-year follow-up is as
follows: 13.8: Florida FTP; 9.4: Vermont WRP; 10.9: Cleveland; and 3.6
Philadelphia.
A single (combined) program group was created in sites that randomly assigned
individuals to two or more programs groups.
[ Go to Contents ]
(2) The Los Angeles and Miami sites were excluded from this report because the administrative data for these sites were not available at the time of this analysis.
(3) Philadelphia County is coterminous with the city of Philadelphia and therefore the terms are used inter-changeably throughout this report.
(4) Education information is also available, but only for the random assignment evaluation sites.
(5) The Vermont WRP evaluation collected UI Wage records from neighboring New Hampshire.
(6) In contrast to single-parent TANF recipients, the welfare grants only covered the financial needs of the children. Therefore, these adults are considered to be administrators of the grant, but not welfare recipients.
(7) This finding is based on published data for the five states in U.S. Department of Health and Human Services, Administration for Children and Families 1996, Tables 8 and 11.
(8) U.S. Department of Health and Human Services, Administration for Children and Families, 1996, Tables 11 and 15.
(9) The exception occurs in Table 12, in which we analyze the program impacts (the difference between program group outcomes and control group outcomes) on the percentage of recipients who became cyclers. It should be noted that the exclusion of control group members does not affect the generalizability of the findings, because sample members were assigned to program and control groups at random.
(10) For the evaluation sites, year 5 after sample intake represents five years after random assignment.
(11) For additional details on selection of the survey samples in the three evaluation sites, see Bloom et al., 2002, pp. 24-27 (Connecticut Jobs First); Bloom et al., 2000, pp. 20-21 (Florida FTP); and Bloom et al., 2002, pp. 9-12 (Vermont WRP).
(12) See Michalopoulos et al., 2000, for a detailed discussion and evaluation of this method, which is called "multiple cohort design."
(13) In Philadelphia, the administrative records data extend back to January 1992. However, to lower the possibility of mislabeling recipients in a current spell of welfare receipt as a new recipient, we drop those recipients who received a payment in 1992. In Cleveland, the first month of available data is July 1992. Again, to lower the chances of including ongoing recipients, we drop the recipients who received any payments during July 1992 to December 1992 and begin analyzing welfare receipt in January 1993.
(14) The start of the sample intake period for Cleveland is July 1994.
(15) U.S. Department of Health and Human Services, Administration for Children and Families 1996, Tables 2, 8, 11, 12, 22, 23, 25, 26.
(16) New recipients may have received welfare payments during one or more previous spells. For the Urban Change sites these respondents received welfare benefits for the first time since July 1992 and January 1992 in Cleveland and Philadelphia, respectively.
(17) Note that the samples are drawn during a general downturn in the welfare caseloads of each site. However, the intake dates are sufficiently early such that the samples should be fairly representative of the average caseload and not overly representative of those with the most barriers to leaving welfare.
(18) For Connecticut, the grant level in 1996 is given since this is the earliest year for which we use data. Nevertheless, Connecticut's maximum grant level in 1994 was $680, substantially greater than the remaining sites.
(19) Recall that in this report, we consider a welfare spell to have ended after two consecutive months without a payment.
(20) Not shown, the rate of cycling among ongoing recipients is slightly lower than the full sample rates in all sites except Florida FTP.
(21) NEWWS examined the long-term effects on welfare recipients and their children of 11 mandatory welfare-to-work programs, operated in seven sites that took different approaches to helping welfare re-cipients find jobs, advance in the labor market, and leave public assistance. The effects of the NEWWS programs were estimated based on a wealth of data on more than 40,000 single-parent families, making NEWWS the largest study of welfare-to-work programs ever conducted. Parents and their children were tracked over a five-year follow-up period, which, depending on the site, spanned different parts of the 1990s. In the study's innovative and rigorous research design, each parent was randomly assigned to a program group (in some sites, there were two program groups), whose members were eligible for pro-gram services and subject to the mandate, or a control group, whose members were not. See Hamilton, 2002, for more information. MDRC collected fewer than four years of welfare payments records for the seventh site, Oklahoma City, and therefore that site is not included in Figure 1.
(22) The NEWWS samples are not used in this report because too little of the observation period occurred after the signing of PRWORA.
(23) Not shown, for the five-year follow-up sample, the incidence of cycling over four years was nearly identical to the rate for the full sample, ranging from 3.6 percent in Philadelphia to 13.8 percent in Florida. Thus, the additional incidence of cycling in year 5 represents a real increase over time and does not result from dif-ferences in sample composition. Furthermore, we observed a similar increase for the NEWWS sites-not shown.
Main Page of Report | Contents of Report
Home Pages:
Human Services Policy
(HSP)
Assistant Secretary for Planning and Evaluation
(ASPE)
U.S. Department of Health and Human Services
(HHS)