FOOD STAMP AND WIC TAKE-UP AND THE RELATIONSHIP BETWEEN TAKE-UP AND TANF RECIDIVISM AMONG ILLINOIS TANF LEAVERS

Transcription

1 H A R R I S SCHOOL WO R K I N G PAPER S E R I E S 05.6 FOOD STAMP AND WIC TAKE-UP AND THE RELATIONSHIP BETWEEN TAKE-UP AND TANF RECIDIVISM AMONG ILLINOIS TANF LEAVERS Mairead Reidy, Meejung Chin, Duck-Hye Yang, Robert M. Goerge

2 Food Stamp and WIC Take-Up and the Relationship between Take-Up and TANF Recidivism Among Illinois TANF Leavers Mairead Reidy, Ph.D. Meejung Chin, Ph.D. Duck-Hye Yang, Ph.D. Robert M. Goerge, Ph.D. Chapin Hall Center for Children at the University of Chicago Comments Welcome February 2005 This research was conducted with support from the Small Grants program ( ) at the Economic Research Service (ERS) of the U.S. Department of Agriculture (USDA). The grant is coordinated by the Harris School of Public Policy at the University of Chicago. We are particularly grateful to Mark Prell, Branch Chief of the Food Assistance Branch in the ERS Food and Rural Economics Division; David Gruenenfelder, Chief, Bureau of Program Development and Evaluation, Division of Human Capital Development at the Illinois Department of Human Services (IDHS); Penny Roth, Chief, Bureau of Family Health, Division of Community Health and Prevention at IDHS; Robert Lalonde at the Harris School of Public Policy, and David Betson, of Notre Dame University for valuable input and comments, and to participants in the Small Grants Program Conference, and in the two workshops organized at the Harris School for grantees. We are also grateful to Matt Reading at Chapin Hall for helpful comments and to Anne Clary, Heather McGuire, and Emilie Schrage for their editorial assistance. Chapin Hall gratefully acknowledges funding from ERS. The views expressed are those of the authors and not necessarily those of ERS or the IDHS.

3 CONTENTS OVERVIEW...1 RESEARCH QUESTIONS AND STUDY RATIONALE...2 THE FOOD STAMP AND WIC PROGRAMS...5 PREVIOUS RESEARCH...7 STUDY POPULATION, RESEARCH DESIGN, AND DATA SOURCES RESULTS Our Study Sample and Program Take-Up The Relationship Between Program Take-Up and Return To TANF CONCLUSION NEXT STEPS References ii

4 OVERVIEW Food assistance programs are an integral component of the public assistance safety net for the working poor, but many families do not use these programs when eligible to do so. In this paper, we use linked administrative data from the Illinois Integrated Database (IDB) to examine the patterns of nonparticipation in two food assistance programs the Food Stamp Program (FSP) and the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) among Illinois families with young children that leave cash assistance but continue to be eligible for these programs. 1 We examine how participation in one of these programs is correlated with the decision to participate in the other and begin to explore how participation is correlated with TANF recidivism. We restrict our analysis to a group of TANF leavers with children under age 5 (ages 0-4) a group that is likely eligible for both programs at TANF exit. We begin by using simple descriptive statistics to identify nonparticipation rates in both programs. We then use logistic regression analyses to understand the socioeconomic and demographic characteristics of low-income families who do not take up these programs when eligible to do so. Finally, we begin to explore how nonparticipation is correlated with recidivism to cash assistance, using a series of hazard analyses to begin to identify the importance of both the independent and combined effects of FSP and WIC participation on return to cash assistance. A significant body of research indicates that nonparticipation in both FSP and WIC is widespread, and we are beginning to learn more about the characteristics of those who decline participation in either program. We also know that those who continue to use food stamps when exiting cash assistance have a lower probability of returning to cash assistance relative to those 1 Nonparticipation is used here to mean the failure to participate when eligible to do so. 1

5 who lose benefits at exit. However, existing research does not typically distinguish those who are eligible for Food Stamps from those who are not. The focus of research on the effects of WIC participation has typically been on child health and nutritional outcomes, and very little work to date has examined WIC s role as a financial stabilizer for families. Furthermore, there has been no systematic effort to study the interaction of the FSP and WIC among a population likely to be eligible for both programs or to examine the effects of multiple program participation on the selfsufficiency pathways of families. This paper begins to fill this gap. To do so, we follow a series of new TANF entry cohorts from the time of entry between 1995 and 1997 over time through December For those who exit TANF with children aged 0-4 (a criteria for eligibility for WIC), we use Unemployment Insurance (UI) wage records to estimate eligibility for the FSP (130 percent of federal poverty level) and WIC (185 percent of federal poverty level). Using Food Stamp and WIC administrative data records, we distinguish between those who are eligible and take up services (program participants) and those who are eligible but do not take up services (nonparticipants), and we examine how TANF recidivism varies across these groups. RESEARCH QUESTIONS AND STUDY RATIONALE Food assistance programs are an integral component of the public assistance safety net for the working poor. At low incomes, TANF leavers are often eligible for both FSP and WIC, and their receipt can add significantly to household resources. Families can use the food assistance benefits to buy the food they would otherwise have purchased and then use the money released to buy other goods (Currie, 2003). It is estimated that about 70 percent of Food Stamp benefits are used to divert other household cash income to nonfood expenditures (Devaney et al., 2

6 1997). Although the primary objective of these programs is to improve food security and provide nutritional support, the added resources these non-cash programs offer to households may also help facilitate the transition from dependence on the state to economic independence. In a study of the importance of transitional benefits, researchers in Illinois, for example, found that even after controlling for employment status at TANF exit and other factors, families that lost both Medicaid and Food Stamps were nearly three times as likely to return to TANF as those who kept both benefits (Illinois Families Study, 2001). Despite the argued importance of these programs, we know that FSP take-up among TANF leavers remains low. A considerable amount of research has also recently emerged that focuses on the socioeconomic and demographic characteristics of those who take up the Food Stamp Program. We know much less, however, about nonparticipation patterns in WIC, and no work has been completed to date that examines the correlates of take-up across the programs at the time of TANF exit. Furthermore, despite the argued importance of FSP and WIC as an aid to self-sufficiency, we know very little about how use of these programs individually and in combination is linked to families remaining off TANF. The purpose of this paper is to further our understanding of those among the population eligible at TANF exit who do not take Food Stamps or WIC benefits, and to explore how both programs aid TANF recipients in their quest for economic independence. Specifically, we are interested in understanding who uses these programs when eligible to do so; whether the two programs are used as substitutes for each other or are typically used together (if the decision to participate in one is correlated with the decision to participate in the other); and the role of program use in predicting return to TANF. We ask the following questions: 3

7 1. What are the patterns of program participation in WIC and/or FSP among families leaving cash assistance with children aged 0-4 who are eligible for these programs? 2. Do TANF leavers with young children who are eligible for these programs use FSP and WIC as complements to or as substitutes for each other? 3. What are the socioeconomic and demographic characteristics of FSP and WIC nonparticipants? 4. How is the decision to participate in one or both programs related to recidivism to cash assistance among families with young children? This work is significant for several reasons. First, it will further our understanding of the distributional consequences of FSP and WIC. Put simply, we will uncover who uses either or both of these supports. Program nonparticipation clearly reflects system ineffectiveness and indicates that policymakers original intentions for the system are not being fully met (Van Oorschot, 1995). However, if nonparticipation is predominant among the most able of poor families (who, for example, believe that their economic conditions will soon improve), policymakers may have fewer concerns than if it is concentrated among the most economically disadvantaged. By identifying the socioeconomic and demographic factors that affect take-up rates in either or both programs at TANF exit, we believe that it will be possible to better target underserved populations when fiscal resources are available. The second contribution of this study is our examination of the role of FSP and WIC in preventing return to TANF. Given time limits on welfare receipt, it is critical for those exiting TANF to avoid what is often described as welfare churning the return to the welfare rolls after a short-term spell of employment. This paper, by exploring the independent and combined 4

8 effects of FSP and WIC take-up on TANF recidivism, begins to fill an important gap in the research on the role of food assistance programs in welfare trajectories of families with very young children. THE FOOD STAMP AND WIC PROGRAMS The Food Stamp Program (FSP), managed by the Food and Nutrition Service (FNS) of the U.S. Department of Agriculture, helps low-income households buy the food they need for a nutritionally adequate diet. 2 The rules are complex, but the most important factors that determine eligibility for Food Stamp benefits are income, the number of persons who live and eat together, and the amount of available liquid assets, such as money in checking and savings accounts. Food Stamps serve households with gross incomes less than 130 percent of the federal poverty guidelines (FPG) for the household size. 3 Although most program rules are set at the federal level, the FSP is usually operated through the same state or county welfare agency and staff that run the TANF program. Although the value of the FSP benefit varies across states, the average benefit to families from the FSP is sizable. In fiscal year 2002, for example, monthly benefits averaged $79.60 per person and about $186 per household (USDA, 2003). During fiscal year 2001, the Food Stamp Program served a monthly average of 17.3 million low-income individuals, and program expenditures totaled about $15.5 billion (GAO, 2002). At any point in time, approximately one- 2 Food stamp benefits can be used to buy any food or food product for human consumption, plus seeds and plants for use in home gardens to produce food. 3 Federal poverty guidelines are established by the Office of Management and Budget, and are updated annually by the Department of Health and Human Services. Gross income includes all cash payments to the household, with a few exceptions specified in the law or the program regulations. 5

9 half of the caseload is children (GAO, 1999). Unlike cash assistance, the Food Stamp Program was essentially preserved under welfare reform for families with children. WIC provides not only food assistance but also nutritional counseling and health and social service referral services to low-income pregnant women, postpartum women with a child 6 months old or younger, women who are breastfeeding an infant aged 7-12 months, infants (0-1 year) and children 1-4 years of age. To be WIC-eligible, a family must have income below 185 percent of the poverty level or be receiving AFDC/TANF, Medicaid, or Food Stamps, and be considered nutritionally at risk. 45 Those receiving AFDC/TANF or Medicaid are eligible even if income exceeds 185 percent of FPL. Virtually all income-eligible households are thought to be nutritionally at risk (Bitler et al., 2002; Institute of Medicine, 2002). WIC provides eligible families with monthly vouchers to purchase a specified nutrient-rich package of food as well as nutritional and behavioral counseling and referrals to health care or other social service providers. Unlike the FSP, WIC is not an entitlement program, and participation is limited by appropriated federal funding. During the period under study, however, no official waiting lists for WIC existed in Illinois or in other states (Currie, 2003). WIC is administered by the Food and Consumption Service (FCS) of USDA and by state WIC agencies. The monthly value of the food package provided varies across states. In 1994, for example, this value varied from $40 in the Southeast to $52 in the West (Currie, 2003). In fiscal year 2001, the program served a monthly average of 7.3 million persons. Approximately 52 percent of participants are children, 25 percent infants, and 23 percent women (U.S. House 4 In determining income eligibility, cash income from social security, welfare, or other public assistance is counted, but in-kind transfers in the form of FSP are excluded (Currie, 2003). 5 To determine nutritional risk applicants may be weighed and their height measured. Additionally, a blood sample may be taken to test iron levels, and applicants may be interviewed by a nutritionist. 6

10 Committee on Ways and Means, 1998). Like its FSP counterpart, WIC saw no changes under welfare reform. PREVIOUS RESEARCH Nonparticipation is widespread across many means-tested federal, state, and local social welfare programs (Bendrick, 1980). A significant body of research indicates that nonparticipation in both FSP and WIC is high, particularly with regard to FSP. A recent Report by the USDA (2003) estimated that there are 2.6 million nonparticipating individuals who are eligible for a relatively high monthly Food Stamp benefit of $200. Overall, using figures from the U.S. Committee on Ways and Means (various years), Currie (2003) estimates that 41 percent, 49 percent, and 39 percent of the population with income less than 130 percent of the federal poverty level participated in the FSP in 1990, 1995, and 1998, respectively. FSP participation rates vary by work and TANF status of the family. Research on working households consistently shows participation rates of approximately 50 percent (see Shavians, 1997; McConnell & Ponza, 1999; Castner, 2000). Recent studies of former FSP families show participation rates two years after leaving the FSP of between 35 and 44 percent (see for example, Rangarajan & Gleason, 2001; Richardson et al., 2003). We know that the use of Food Stamps plummets upon exit from cash assistance (Blank & Ruggles, 1996; Loprest, 1999, 2001; Rangarajan & Gleason, 2001), and that participation rates at the time of TANF exit declined during the first few years after welfare reform implementation (GAO, 1999; Zeslewski & Brauner, 1999). Although efforts have been underway to reach out to former welfare recipients and inform them of their Food Stamp eligibility (GAO, 2002), a review of the U.S. Department of Health and Human Services, TANF Leaver Studies by Isaacs and Lyon (2000), 7

11 found that Food Stamp Participation among families 12 months after leaving TANF ranged from 20 to 40 percent 6 (also see Loprest, 1999; Dion & Pavetti, 2000). In contrast, WIC participation rates in the population with incomes below 185 percent of the federal poverty level (FPL) have been rising steadily over the past 20 years. Currie (2003) estimates that although only 28 percent of income-eligible children aged 0-4 participated in WIC in 1985, the participation rate rose to 39 percent by 1990 and to 62 percent in Using the SIPP data, Bitler et al. (2002), for example, estimate that of those who are eligible for WIC, 73 percent of infants, 67 percent of eligible pregnant and postpartum women, and 38 percent of children ages 1 through 4 participate in the program. Burstein et al. (2000) also show that infants are more likely to participate in the program than are older children. In 1996, 25 percent of those receiving WIC were also receiving AFDC/TANF. A considerable amount of research has recently emerged that focuses on the socioeconomic and demographic characteristics of those who take up the FSP; again, much less exists on the correlates of WIC participation. The research generally indicates that nonparticipation is more prevalent among the economically advantaged in both the FSP (for example, Bartlett & Burstein, 2004; Reidy et al., 1998; Reidy et al., 2003; Miller et al., 2002) and WIC (for example, Swann, 2003; Bitler et al., 2002). Bartlett and Burstein (2004) found that, compared with the active Food Stamp caseload, eligible nonparticipant households had higher average household income, and that individuals in families with higher incomes were also less likely to remain on Food Stamps. When other 6 In 1998, the U.S. Department of Health and Human Services Assistant Secretary for Planning and Evaluation (ASPE) awarded $2.9 million in grants to states to study the effects of welfare reform on TANF leavers. This group of studies is known as the TANF Leaver Studies. 8

12 background demographics were controlled for, black and Hispanic TANF leavers were more likely to stay on Food Stamps than White/non-Hispanic TANF leavers (U.S. Committee on Ways and Means, 1998; Miller et al., 2002), and leavers who were public housing residents were more likely to remain on Food Stamps than were their counterparts in private housing (Miller et al., 2002). Similarly, Cancian et al. (2001) found education, family composition, and location to affect Food Stamp enrollment. Those who lacked a high school degree, had larger families with very young children, and lived in an urban setting were more likely to be enrolled in the FSP, and families that declined to use Food Stamps in 1999, despite having a poverty-level income, were more likely to have owned a car (Zedlewski & Gruber, 2001). Furthermore, those in rural households have higher take-up than urban households (McConnel & Ohls, 2000). Studies looking at FSP receipt among AFDC/TANF exiters generally do not find that welfare families who left FSP differed significantly from those that stayed in terms of education, health status, and presence of a working husband or partner (Zedlewski & Gruber, 1999, 2001). Heflin (2004), however, finds that among mothers receiving cash assistance in an urban Michigan county, the rate of exit from the FSP is negatively associated with the number of children in the household, but positively associated with the number of adults in the household and being married or cohabiting. Using SIPP data, Bitler et al. (2002) find that WIC participation is positively correlated with being married and with Hispanic ethnicity, and negatively correlated with residence in a central city and with Asian ethnicity. They also find that it is the less well-educated who are most likely to participate in the program. Swann (2003), focusing on pregnant women only, shows that socioeconomic and demographic characteristics such as low education, Hispanic ethnicity, low income, and participation in other welfare programs are all correlated with a higher likelihood of 9

13 participation in WIC. Younger pregnant women are also more likely to participate. Burstein et al. (2000) found that poor and minority women were more likely to be enrolled, as were high school dropouts and single mothers. Research has also recently begun to suggest the importance of continued Food Stamp use, in that those who continue to use Food Stamps after exiting cash assistance have a lower probability of returning to cash assistance than those who do not use benefits at exit (see, for example, Lee & Lewis, 2001; Loprest, 2001); but for the most part, these studies do not take account of eligibility, and when they do (see, for example, Zedlewski & Brauner, 1999), no extensive multivariate analysis has been completed to examine the effects of the participation decision on self-sufficiency pathways. Even less is known about the relationship between WIC participation and selfsufficiency outcomes. Although previous research on WIC has focused on the effect of WIC participation on positive birth outcomes (Kennedy et al., 1982; Kotelcheck et al., 1984; Rush, 1988; GAO, 1992; Gorden & Nelsen, 1995; Brien & Swann, 1999a; Kowaleski-Jones & Duncan, 2000; Burstein et al., 2000) and child health outcomes (Gorden & Nelsen, 1995, Rose et al., 1998; Brien & Swann, 1999b; Bitler et al., 2002), we know little about how WIC participation affects the self-sufficiency outcomes of families. We have recently begun to learn about the interaction of FSP and WIC participation. Recent research, commissioned by the USDA (Cole and Lee 2004), examines the interaction of FSP and WIC participation using within-state linked administrative data in Florida, Iowa, and Kentucky between 2000 and While not taking into account income eligibility for both programs, the report finds that the percent of FSP infants and children who participated contemporaneously in WIC in a single month ranges from 84 to 94 percent for infants and from 10

14 50 to 57 percent for children. This difference reflects declining WIC participation with age. By contrast, the percent age of WIC infants and children who are also participating in the FSP ranged from 22 to 38 percent for infants and 29 to 50 percent for children. These rates are lower in part because only a subset of WIC participants is income-eligible for FSP. The study further examines the characteristics of FSP infants and children who also participate in WIC. While the results from the infant model did not produce consistent results across states, for FSP children, the characteristics related to WIC participation showed consistency across States. The likelihood that a FSP child also participates in WIC increases with both the number of adults in the household, and the number of children under age 5. WIC participation is also positively associated with a married head of household and receipt of TANF. By contrast, the likelihood that FSP children participate in WIC declines with age, and is negatively correlated with residence in a metropolitan area. Hispanic children in Florida and Kentucky were more likely to participate in WIC than other racial groups, and African American children were less likely to participate than other racial groups in Iowa and Kentucky. There has been no systematic effort to study the relationship between multiple program participation and the self-sufficiency pathways of families. STUDY POPULATION, RESEARCH DESIGN, AND DATA SOURCES Our research design is to follow a series of new TANF entry cohorts from point of entry between 1995 and 1997 through December A new TANF entrant is defined as one who has not received TANF in the previous two-year period. Because having a child under 5 years old is a criterion for WIC eligibility, we focus our analysis on TANF exiters whose youngest child is between 0 and 4 (up to age 5) years of age at the time of TANF exit. Our study 11

15 population therefore is those families who exit TANF with young children (aged 0-4 years). For this population, we estimate family eligibility for the Food Stamp Program and WIC, and examine both family take-up of these programs and the relationship between the participation decision and TANF recidivism. We note that those who exit TANF with a child under 5 are an atypical group of TANF exiters, which has not been studied in detail to date. We rely exclusively on state-level linked administrative data in Illinois. By linking individual-level TANF records with Unemployment Insurance (UI) wage records, we identify the employment and earnings patterns of TANF exiters. Data on earnings are used to estimate who among TANF exiters is eligible for the FSP (130 percent of FPL) and WIC (185 percent of FPL). Food Stamp and WIC administrative records are used to distinguish between those who are receiving Food Stamps or WIC (participants) and those who are not (nonparticipants). TANF records are then used to examine subsequent patterns of return to TANF among WIC and FSP participants and nonparticipants. It is important to note that we are concerned with program eligibility and take-up at the family/household level. Although FSP eligibility is estimated based on household income, and program receipt is also recorded at the household level, eligibility and participation in WIC can vary across family members. For our analysis, we allow any family with a child aged 0-4 years at TANF exit whose quarterly income falls below 185 percent of the federal poverty level to be WIC-eligible. We assume that all are nutritionally at risk. If WIC receipt is recorded for any member of the household, we consider the household to be in receipt of WIC 7. Furthermore, we are not looking at all eligible groups at TANF exit just those who, at TANF exit, have children aged 0-4. We noted above that WIC is also available to pregnant and 7. We recognize that families may not enroll all age-eligible children in WIC (Cole and Lee, 2004). 12

16 postpartum women, as well as women who are breastfeeding children between the ages of 6 and 12 months. The administrative data does not allow us to identify those women who are eligible for WIC because of pregnancy at TANF exit or because they are breastfeeding children between the ages of 6 and 12 months. Because eligibility is observed on a quarterly basis (using UI quarterly wage records), we use the quarter as the unit of analysis. We count the family as taking up a program in a quarter if the family uses FSP or WIC during any months of the given quarter. Note that we consider the family to be WIC or FSP participants at TANF exit if we see program take-up in the quarter after the family exits TANF. First, using simple descriptive statistics, we explore the take-up of the FSP and WIC among TANF leavers both at the point of exit and over time. Second, using a succession of logistic and event history hazard regression models, we examine the socioeconomic and demographic characteristics of nonparticipants. Third, we focus on the role of the FSP and WIC on TANF recidivism rates. We use event history analysis to examine the effects of individual socioeconomic and demographic characteristics, and the take-up of the FSP and WIC on rates of TANF recidivism. The independent variable of interest is service receipt status of WIC and/or FSP, and our concern is to begin to explore the importance of both the independent and combined effects of FSP and WIC participation on return to cash assistance. Throughout the analyses, available demographic and economic characteristics, which have been shown to be important in previous research, are included as explanatory variables. Although some of these characteristics, such as race, education, work experience, or marital status as recorded at entry to TANF, do not change across time in our data, others, such as participation in the non-cash programs of interest, do. Event history analysis can account for both types of independent 13

17 variables (Coleman, 1981; Tuma & Hannan, 1984; Allison, 1984; Heckman et al., 1985; Lillard, 1993; Lillard & Panis, 1998). Data Sources. The data used are drawn from the Illinois Integrated Database on Child and Family Services in Illinois (IDB). Built and maintained by researchers at the Chapin Hall Center for Children, the IDB is a state-level, longitudinal database constructed from administrative data gathered by public agencies that serve children and families in Illinois (Goerge et al., 1994). The IDB allows researchers to track children and families across human service data systems. Specifically, for the purposes of this research, we combine records from monthly cross-sectional AFDC/TANF and Food Stamp data from the Illinois Department of Human Services, Client Database, monthly WIC data received from the Illinois Department of Human Services, Cornerstone System; and quarterly Unemployment Insurance wage records (total quarterly earnings reported by employers to state UI agencies for each employee) to create a cross-program record for each household. Most employers who pay $1,500 in wages during a calendar quarter to one or more employees are subject to a state UI tax and must report the quarterly amount paid to each employee. Data Linking. Because departments maintain separate databases, it was necessary to link records across departments. The linking process is complicated by the fact that no single variable, even Social Security Number (SSN), can always be relied on to completely establish the identity of a client from the records of various agencies. We thus use a technique called probabilistic record matching. Probabilistic record matching assumes that no comparison between fields common to the source databases will link an individual s records with complete confidence. Instead, the method calculates the likelihood that two records belong to the same person by matching as many pieces of identifying information as possible from each database 14

18 (Jaro, 1985,1989; Newcombe, 1988). We use identifying information including first and last name, birth date, gender, race and ethnicity, Social Security number, and county of residence to link those in receipt of AFDC/TANF, Food Stamps, WIC, with unemployment insurance records. Once a match has been determined, a unique number is assigned to the matched records so that each record can be uniquely identified. The end result of computer matching is a linkfile, which contains the unique identifier, the dates of TANF, FSP, and WIC receipt, as well as the date and levels of earnings found in the UI wage data. Data Confidentiality. The confidentiality of administrative databases is a key issue in this research. We implemented extensive procedures to ensure data security, protect confidentiality, and control access to data. These procedures include inventorying confidential records when received, storing data tapes in a locked facility, and maintaining passwords. Once the recordlinkage phase of the process is complete using identifying information from the source data, most identifying information (especially SSN and name) is removed to a separate file, accessible only to authorized personnel. Data Limitations. The primary data limitations are those inherent in the use of UI wage records to estimate program eligibility, as well as the limited number of socioeconomic and demographic characteristics available in administrative data. The problems associated with use of UI wage data to estimate eligibility for Food Stamps have been noted by others (see, for example, Cancian et al., 2001; Miller et al., 2002). First, UI wage data allow us to observe quarterly income from employment only, yet both FSP and WIC eligibility are dependent on additional factors. For example, Food Stamp eligibility is based on income from a range of sources, the amount of available liquid assets, such as money in checking and savings accounts, and the number of persons who live and eat together. Income other than earnings, including, for 15

19 example, Social Security or Supplemental Security Income (SSI), are important omissions, and although household liquid assets may not be large among TANF exiters, we are unable to factor them into our eligibility estimations. 8 Second, we recognize the limitations of the UI wage data as a source of information on income from employment. First, UI does not cover all jobs. Major types of employment not covered include federal government civilian and military employment, U.S. Postal Service employment, railroad employment, employment by some philanthropic and religious organizations, and employment as an independent contractor. Hotz and Scholz (2002) suggest that between 86 and 90 percent of the employed population is included in UI data (Baj et al., 1991; Blakemore et al., 1996; Kornfeld & Bloom, 1999). Furthermore, all states have a minority of workers who are employed outside the state, and these out-of-state workers earnings will not be reflected in their state UI wage data. In addition, even when wages are found in UI records, they may be understated (Kornfeld & Bloom, 1999). 9 Because we cannot perfectly (accurately) identify eligibility, our regression results are likely to be tainted by some amount of measurement error. See Appendix A for a discussion of how these limitations may affect our results. 8 Most households are ineligible for food stamps if they have resources greater than $2,000 ($3,000 if a household member is 60 years old or older). It is important to note, however, that some common items, such as a home, jewelry, and other personal items, do not count toward the resource limit. Our study population is TANF leavers, and we know that TANF leavers have relatively few assets. A study of TANF leavers in Massachusetts found that although 29.3 percent of respondents in time-limited closings and 35.2 percent of respondents in nontime-limited closings had a savings account in a bank or credit union, over four-fifths in each category had $500 or less (Massachusetts Department of Transitional Assistance, 2000). Furthermore, a recent study of Illinois TANF leavers finds that increased assets account for only a very small proportion (0.1 percent) of TANF case closings (Institute for Public Affairs, 2000). We are confident, as proposed by Zedlewski & Brauner (1999), that the TANF leavers in our study do not have sufficient available liquid assets to significantly skew our FSP eligibility estimates. However, they may subsequently accumulate such assets from employment. 9 Comparisons of UI wage records with Internal Revenue Service data by Kornfeld and Bloom (1999) suggest that wage estimates based on UI records may be understated by approximately 11 to 14 percent. 16

20 Finally, we are limited in the range of socioeconomic and demographic variables we can include in our models to those collected in the administrative data records. Additionally, the socioeconomic and demographic characteristics are observed in our study population either at TANF entry or at TANF exit, and we cannot take account of changes that occur over time in some of these variables, including number of children, education, marital status, and work experience. These changes may be important to the non-cash program take-up decision as well as the likelihood of returning to TANF. RESULTS Our Study Sample and Program Take-Up In Table 1, we look at selected characteristics of the study population, mothers whose youngest child is under age 5 at TANF exit. We note first that almost equal numbers are African American and White (44% and 43%). On average, our sample has been in receipt of TANF for a little over twelve months in their current spell. This is a young population; on average, mothers are less than 25 years of age. A majority have never married, and approximately 30 percent are high school dropouts. The average number of children is two, and the age of the youngest child is just over 1 year. Table 1 also outlines Food Stamp and WIC eligibility and take-up rates at TANF exit. Based on the eligibility criterion of 185 percent of federal poverty level for WIC, and 130 percent of the federal poverty level for Food Stamps, we find that 96 percent are eligible for WIC and 87 percent are eligible for FSP. Among the eligible at TANF exit, 54 percent take up WIC, and 45.5 percent take up FSP. 17

21 Table 1. Selected Characteristics of the Study Population at TANF Exit (Study Population: Those who were new TANF Entrants and at first TANF exit had a Child under 5) Characteristics Percent Total Number of Mothers 1995 TANF Entrants TANF Entrants TANF Entrants 27.8 Race/Ethnicity White/Non-Hispanic 43.2 African American 44.3 Hispanic 10.8 Percent High School Graduate 49.2 Percent More than High School 19.9 Average Number of Months of TANF Spell 12.3 Average Age of Mother at TANF Exit 24.7 Percent Ever Worked for Pay 90.5 Percent Never Married 70.0 Average No. of Children at Ttime of TANF Exit 2.1 Average Age of Youngest Child at TANF Exit (Years) 1.1 Percent Living in Cook County 49.3 Year of TANF Exit Percent Eligible for Food Stamps at TANF Exit 87.2 Percent Eligible for WIC at TANF Exit 96.1 Percent Take-Up of Food Stamps Among Eligible 45.5 Percent Take-Up of WIC Among Eligible 54.3 Percent of Food Stamps and WIC Among Those Eligible for Both

22 In Table 2, we begin to explore take-up patterns across programs among those who are eligible for both programs (at TANF exit). We distinguish three categories among the eligible: those who take up the programs when they leave TANF, those who do so at a later date, and those who never participate in the programs. First, we consider the Food Stamp Program. We noted above that among those eligible at TANF exit, only 45 percent use the program initially. However, an additional 30.1 percent take up the program at a later date, implying that only a quarter (24.5%) of all of those eligible for FSP at TANF exit never use it. Table 2 also shows that initial take-up of WIC is higher (52.6%), but not many families subsequently use it (15.6%), so that the proportion of families that never use WIC is slightly higher than for FSP at 31.8 percent. Table 2. Food Stamp and WIC Take-Up Patterns Among Those Eligible for Both Programs at TANF Exit (Percent) WIC Receipt Take-Up at Exit Later Never Total FS Receipt Take-Up at Exit Later Never Total Total Given the similarity in take-up rates for the two programs, an interesting issue is whether, among those who are eligible for both programs at TANF exit, it is the same families that are using both programs. Table 3 provides the raw numbers for people falling into the three take-up categories (at TANF exit, later, or never). These data give us some insights into whether the food assistance programs are used together as complements or typically used as substitutes for each other. We note that among those who take-up WIC at TANF exit, only 51 percent (16719/32538) 19

23 take-up FSP. Likewise, among those who take up FSP at TANF exit, only 59 percent (16,719/28,128) take up WIC. A substantial proportion of those who do not take-up these programs will do so later, but it still remains the case that of those who take up WIC at TANF exit, 19 percent (6,317/32,538) never take up FSP. Similarly, among those who take up FSP at TANF exit, 25 percent (7,031/28,128) never take up WIC, despite being eligible to do so. We conclude, therefore, that among those who are eligible for both programs at TANF exit, it is not the same people who by and large use both. Table 3. Food Stamp and WIC Take-Up Patterns Among Those Eligible for Both Programs at TANF Exit (Number) WIC Receipt Take-Up at Exit Later Never Total FS Receipt Take-Up at Exit 16,719 4,378 7,031 28,128 Later 9,502 4,308 4,767 18,577 Never 6, ,832 15,111 Total Total 32,538 9,648 19,630 61,816 In this paper, we identify the determinants of two outcomes program take-up rates and rates of subsequent return to TANF and much of the multivariate analyses to follow will focus on identifying how socioeconomic and demographic characteristics correlate with these outcomes. We begin here by providing some simple descriptive statistics on how the characteristics of individuals differ on these outcome variables. In Table 4, we consider the socioeconomic and demographic characteristics of those who are eligible for FSP at TANF exit, and compare the socioeconomic and demographic make-up of those who take up and those who do not take up FSP at TANF exit. We note in accordance with the literature that it is likely the more needy who use the FSP, i.e., take-up of the FSP is more 20

24 focused on the economically disadvantaged. For example, those who use the FSP have had longer TANF spells than those who do not use the program. They are also more likely to have never been married and to have slightly more children than those who do not use the program. Those who take up FSP are more likely to have dropped out of high school than those who do not take up the program at TANF exit. Additionally, 48.2 percent of participants are African American compared with 38.3 percent among the non-users. Table 4. Eligibility, Take-Up, and Non-Take-Up of Food Stamps at TANF Exit by Selected Socioeconomic and Demographic Characteristics (Percent) Characteristics Eligible for FS Take-Up of FS Non-Take-Up of FS Total Number of Mothers 1995 TANF Entrants TANF Entrants TANF Entrants Race/Ethnicity White/Non-Hispanic African American Hispanic Percent High School Graduate Percent More than High School Average Number of Months of TANF Spell Percent Ever Worked for Pay Percent Never Married Average Age of Mother at TANF Exit Average No. of Children at Time of TANF Exit Average Age of Youngest Child at TANF Exit (Years) Percent Living in Cook County Year of TANF Exit

25 A less clear pattern emerges when we examine the characteristics of eligible participants and nonparticipants in WIC (Table 5). As expected, the average age of the youngest child is lower (0.7 years) among those who take up WIC than those who do not (1.6 years). This is in accordance with the literature that indicates that WIC participation is higher among infants than children. We note that 45.7 percent of WIC users are White compared with 41.5 percent of nonusers. By contrast, African Americans comprise 42.3 percent of users but 45.3 percent of nonusers. If anything, the descriptive statistics suggest that take-up is higher among the less economically disadvantaged. For example, those with more children are less likely to take up the program. Non-users have an average of 2.2 children at TANF exit, while users have an average of 2.0 children. WIC users are slightly more likely to have dropped out of high school than non- WIC users. While 29 percent of users have less than a high school education, 33 percent of nonusers are high school dropouts. This suggests that the users of WIC may not be drawn from the most needy among the eligible, and explains at least in part the divergence found in Table 3, where it was shown that users of the programs have relatively little overlap.. 22

26 Table 5. Eligibility, Take-Up, and Non-Take-Up of WIC at TANF Exit by Selected Socioeconomic and Demographic Characteristics (Percent) Characteristics Eligible for WIC Take-Up of WIC Non-Take-Up of WIC Total Number of Mothers 1995 TANF Entrants TANF Entrants TANF Entrants Race/Ethnicity White/Non-Hispanic African American Hispanic Percent High School Graduate Percent More than High School Average Number of Months of TANF Spell Percent Ever Worked for Pay Percent Never Married Average Age of Mother at TANF Exit Average No. of Children at Time of TANF Exit Average Age of Youngest Child at TANF Exit (Years) Percent Living in Cook County Year of TANF Exit Simply looking at descriptive statistics in this way, however, can be misleading because many of these variables are correlated with each other. As a result, we turn now to a multivariate analysis of our data by providing logistic regressions predicting take-up of each program. We begin in Table 6 by predicting Food Stamp take-up at TANF exit. First consider Model 1 in Column 1. In line with the descriptives in Table 4, African Americans are significantly more likely to take up FSP than are Whites. Hispanics are less likely to take up FSP when eligible to 23

27 do so than are their White counterparts. On many dimensions, it appears that Food Stamp takeup is more prevalent among those with greatest need. For example, those with more education (with a high school graduation or beyond) are less likely to take up FSP than their high school drop-out counterparts. Furthermore, the more children one has, the more likely one is to use Food Stamps, and similarly, families with younger children are more likely to take up the program. However, contrary to the need-based take-up theory, we find that the older the mother, the more likely she is to take up FSP, although we note that the coefficient is small. We further note that take-up of FSP is declining across TANF exit years through 1999 and increasing from 2000 onward, and this is in line with the broader literature that Food Stamp use in the aggregate was on the decline and subsequently rose again toward the end of our study period (see, for example, McKean, 2003, for caseload numbers in Illinois). We know that many families use both programs simultaneously, and in order to consider the possibility that the decision to take up FSP is related to the WIC participation decision, we add a control variable of whether or not the family took up WIC at TANF exit. The results of this second model are shown in the second column of Table 6. Inclusion of whether or not a family takes up WIC at TANF exit does not change our results on the socioeconomic and demographic characteristics of FSP users. 24

28 Table 6. Logit of Take-Up of Food Stamps at TANF Exit Variables Race/Ethnicity Parameter Estimate Parameter Estimate Model 1 Model 2 African American 0.39 ** ** Hispanic ** ** Educational Level High School Graduate ** ** More than High School ** ** Number of Months of TANF Spell 0.01 ** 0.01 ** Ever Worked for Pay Marital Status Never Married Divorced or Separated ** ** Age of Mother at TANF Exit 0.00 ** 0.00 * Number of Children Two 0.28 ** 0.30 ** Three ** 0.56 ** Age of Youngest Child at TANF Exit ** 0.04 ** Living in Cook County ** ** Year of TANF Exit ** ** ** ** ** ** ** 0.22 ** ** 0.31 ** WIC Take-Up ** *<.05 **<0.01 A similar exercise is carried out to predict WIC take-up in Table 7. In our first model (Column 1), we confirm the descriptive findings of Tables 4 and 5 that take-up patterns are different across WIC and FSP. We first note that, in line with the literature, that WIC take-up is rising across TANF exit years. Similar to FSP, but as expected for WIC, the older the youngest 25

29 child (observed at TANF exit), the less likely one is to take up WIC. Families know they will age out of the program when the youngest child reaches age 5, and are less likely to take up the program when children are older. On many dimensions, the most socioeconomically needy would appear to be least likely to take up WIC immediately at TANF exit. Those with more education are significantly more likely to take up WIC than high school dropouts. Additionally, those with longer welfare spells are less likely to use WIC. Likewise, the greater the number of children, the less likely one is to take up WIC. However, unlike the descriptive statistics, African Americans and Hispanics are more likely to use the programs than Whites. The fact that WIC is not linked to cash assistance in the way FSP is may explain some of these counterintuitive differences. One interpretation is that it takes an additional effort to learn about WIC locate the office, and access the program. Those with more human capital are more likely to be able to surmount these difficulties, and families with more children may find it harder to find the time to negotiate the application process, especially if the perceived pay-off from the program is low. Again when we add Food Stamp Program take-up as an additional explanatory factor in our model (Model 2 in Table 7), we see a positive and significant coefficient indicating that takeup of WIC is higher for those who also take up FSP. It does not, however, change the effects of the socioeconomic and demographic variables included in both models. To test to see if the kinds of families who take-up WIC and FS with children under one are different to those who take it up with older children, we run separate regression analyses for those who whose youngest child is under age one and those whose youngest child is between1 and 4. The results, which we do not show, are qualitatively very similar to those found in Tables 7 and 9 with the more socioeconomically needy taking up the FS program but not WIC. 26

30 Table 7. Logit of Take-Up of WIC at TANF Exit Variables Race/Ethnicity Parameter Estimate Parameter Estimate Model 1 Model 2 African American 0.14 ** 0.10 ** Hispanic 0.18 ** 0.20 ** Educational Level High School Graduate 0.17 ** 0.18 ** More than High School 0.23 ** 0.27 ** Number of Months of TANF Spell ** ** Ever Worked for Pay 0.10 ** 0.10 ** Marital Status Never Married ** ** Divorced or Separated ** ** Age of Mother at TANF Exit 0.00 ** 0.00 ** Number of Children Two ** ** Three ** ** Age of Youngest Child at TANF Exit ** ** Living in Cook County ** ** Year of TANF Exit ** 0.08 ** ** 0.25 ** ** 0.26 ** ** 0.41 ** ** 0.37 ** ** 0.47 ** Food Stamp Take-Up 0.47 ** *<.05 **<0.01 These results clearly point to the fact that these two programs are being used by different populations. To address this more systematically, in Table 8 we compare the characteristics of those who, at TANF exit, only use WIC and those who use only the FSP. Here the dichotomy between the more needs-based take-up of FSP compared with WIC can be seen more clearly. For example, 28.9 percent of the WIC-only recipients have less than a high school education 27