Data for this study come from the Wisconsin Department of Workforce Development CARES system, which contains information collected through the administration of AFDC and other means-tested programs. These data were matched to earnings and employment data from the state's UI system. All persons in the data used in this study received AFDC benefits in Wisconsin in July 1995. These cases were tracked with linked administrative data from January 1989 until December 1997, providing up to 9 years of data for each case.
Who Is in the Data Set?
Every observation in the data set received AFDC-Regular (for single-parent families) in July 1995. The entire caseload at the time numbered 65,017. The following types of cases were eliminated from the data, with the number of cases eliminated (nonsequentially) with the restriction in parentheses:
- Cases that were open in July 1995 but did not receive any benefits (n=397).
- Cases where there were no children 18 or younger in July 1995 (n=843).
- Cases where all eligible children in the case are being cared for by a not-legally responsible relative (n=6,101).
- Cases where there are two parents (n=907).
- Cases where a case head is a teen mom--meaning there is an eligible adult under the age of 18 (n=47), or there is no eligible adult and a child is the caretaker (n=254).
- Cases involving a large family or two conjoined families where a single case head is unidentifiable (n=138).
- Cases for which UI data were not requested (n=47).
- Cases where the case head is over 65 years old (n=83).
- Cases with a male case head (n=1,888).
After eliminating these cases, the data set contained 54,518 cases; this is the data set used by Cancian et al. (1999). We further eliminated cases under the age of 21 in 1995. Because we were able to obtain data on AFDC receipt back to July 1989 and UI earnings reports back to January 1989, those under age 21 were eliminated because they were under the age of 15 in 1989 and not reasonably expected to be on AFDC or working. After eliminating these cases, our final number of observations is 48,216.
Definition of a Leaver
A welfare "leaver" is defined as a case that received AFDC in July 1995 and, over the course of the next year (until August 1996), stopped receiving benefits for 2 consecutive months.(3) "Stayers" are those who did not stop receiving benefits for 2 consecutive months during the August 1995-August 1996 period. This period is referred to throughout the paper as the "exit period." The "preexit period" is between January 1989 and July 1995. The "postexit period" for a leaver begins in the quarter the leaver exited welfare and continues until the last quarter of 1997. For a stayer, the postexit period is between July 1996 and the last quarter of 1997. Stayers may have left welfare after August 1996 but did not do so during the exit period.
Two alternative definitions of leavers were explored; first, only those who stopped receiving benefits for 3 consecutive months from August 1995 to September 1996 were considered leavers, and a more stringent definition of a leaver considered only those who stopped receiving benefits for 6 consecutive months from August 1995 to December 1996 to be leavers. Caseload composition and outcomes using these definitions are reported in Appendix 13-A. In general, we find only small changes in the demographic composition of the group of leavers under a more restrictive definition of a leaver, that is, one who has stayed off of welfare for 6 consecutive months. The differences in demographic composition between 2-month and 3-month leavers are negligible. Outcomes of leavers change slightly with the more restrictive definition of leavers, as 6-month leavers are less likely to return to welfare and have modestly higher earnings than 2-month and 3-month leavers.
Welfare History Variables
The cases were categorized into groups based on each case's past welfare receipt history. This was done as a means to characterize the welfare caseload at the time the sample of leavers was drawn and as a means to standardize comparisons of outcome measures across different types of leavers. Leavers were stratified into groups using monthly AFDC receipt data from July 1989 through December 1997.(4) From these data, spells of receipt were counted. A spell began with 1 month of receipt (preceded by a month of no receipt) and ended with 1 consecutive months of nonreceipt. Those enrolled in AFDC in July 1989 were counted as starting a spell, even though they may have already been enrolled in months prior to that. No adjustment was made for this censored data. A month of nonreceipt surrounded by two months of receipt was not counted as an end of a spell. Rather, it was counted as if the spell continued. We implemented this strategy to ensure that a spell actually ended and that the break in receipt was not the result of administrative churning or erroneous reporting. Some cases continued spells after July 1995 and are right censored. No adjustments for these censored data were made.
The total number of months on AFDC, the total number of spells, and the average spell length in months (total months of receipt divided by number of spells) were calculated for each observation. Using these measures, all leavers and stayers are classified as short-termers , long-termers , or cyclers . Short-termers have average spell lengths of less than 24 months and fewer than three total spells throughout the preexit period; long-termers have average spell lengths of 24 or more months and fewer than 3 total spells; and cyclers have three or more spells, regardless of average spell length. The exact cutoff points of these classifications are somewhat arbitrary, however, under this definition, long-termers are those who have spent at least a third of the time we observe them on welfare and short-termers are those who have spent less than one-third of the time on welfare.(5)
In general, we expect that short-termers face the fewest barriers to self-sufficiency. We expect that long-termers have the most barriers to self-sufficiency. Cyclers are expected to be somewhere between them. Therefore, we expect that short-termers will be less dependent on assistance and have better labor market outcomes after leaving than long-termers and we expect outcomes of cyclers to be somewhere between them.
The AFDC receipt data only include administrative records from the state of Wisconsin. Some cases may have moved to Wisconsin just before the exit period and started spells then. These may include a mix of long-term, cycler, and short-term welfare users. However, because we cannot track welfare receipt in other states, these cases are classified as short-termers. Similarly, the definitions do not account for the age of the case head (except that all were at least 15 in 1989). Those who are younger have fewer years of "exposure" to welfare and are likely to have fewer and shorter spells compared to older recipients.
Work History Variables
Earnings information from Unemployment Insurance records from first quarter 1989 to fourth quarter 1997 are used in this study. A variable for the percentage of quarters with any earnings in the preexit period was created and used to stratify outcomes (number of quarters from 1989 to 1995 with positive earnings divided by total number of quarters between first quarter 1989 and third quarter 1995). The percentage of quarters with earnings was divided into the following categories to make comparisons feasible: (1) those who had never worked in the preexit period; (2) those who had worked at least one quarter but no more than 25 percent of the quarters in the preexit period; (3) those who had worked more than 25 percent of the quarters but not more than 50 percent of the quarters; (4) those who had worked more than 50 percent of the quarters but not more than 75 percent of the quarters; and (5) those who had worked more than 75 percent of the quarters. Each outcome of interest is also stratified by these categories of work history. Again, earnings records from other states are not available for those who move into Wisconsin. Also, no standardization for the age of the case head was made in this measure. The youngest welfare recipients in July 1995 are likely to have worked fewer quarters than older recipients. Thus, we expect the average age of groups with less work experience to be lower than the average age of groups with more work experience.
Postleaving Outcome Measures
Three types of outcomes for welfare leavers were examined: (1) public assistance receipt, such as whether the case returned to welfare and whether the case received other public assistance benefits (food stamps and medical assistance); (2) earnings and employment after leaving; and (3) total income, from earnings and public assistance benefits after leaving. The entire sample was tracked through administrative records through December 1997. For each leaver, there are at least five quarters of data on earnings and public assistance receipt after leaving.
Outcomes of both leavers and stayers are reported.(6) Some outcomes are reported relevant to the quarter the leaver stopped receiving AFDC, such as earnings in the first quarter after exit. For leavers, the actual calendar year quarter of these earnings will vary according to when the leaver stopped receiving welfare. For stayers, the first quarter after initial exit is the third quarter, 1996, the second quarter after exit is the fourth quarter 1996, and so on.
This study relies solely on administrative records from the CARES system and matched UI records from the state of Wisconsin. These data have important limitations. First, only records from Wisconsin are included in this study. If a case moved into or out of Wisconsin, information about the case when not in the state is not available. Second, good information on how many of these movers might be in the data file at some point is not available. Administrative data are available on those in the case unit and not on others who might be living in the same household as the unit. For example, earnings of a cohabitating partner are not available, nor are data on living arrangements. Third, errors may occur during the process of matching the CARES data to the UI data may occur if Social Security numbers are reported erroneously or if there are duplications in the data reported to the UI system from employers. Finally, with specific regard to UI data, not all jobs are covered in the Unemployment Insurance system (for example, self-employed persons or federal government employees) or recorded when they legally should be. As a result, some cases that appear to have no earnings may in fact have earnings from jobs. Hotz and Scholz (this volume: Chapter 9) review studies of underreporting in the UI system.
In the Wisconsin data, some cases cannot be tracked with the administrative records from the postexit period (for example, those who move into or out of the state as described). These cases, "disappearers," make up 3.7 percent of the total of 54,518 cases. Other cases appear in some but not all quarters. These "partial disappearers" make up 13.6 percent of the total caseload. Cases that disappear are used in the analysis unless otherwise noted. Cases not appearing in UI records for a quarter are assumed to have zero earnings for that quarter. Cases not appearing in public assistance records were assumed to not be receiving benefits.