Geographic Variation in Food Stamp and Other Assistance Program Participation Rates: Identifying Poverty Pockets in the South

Transcription

1 Geographic Variation in Food Stamp and Other Assistance Program Participation Rates: Identifying Poverty Pockets in the South Final Report submitted to the Southern Rural Development Center, Mississippi State University, Mississippi State, MS in completion of a grant funded by the Economic Research Service of the U.S. Department of Agriculture. William H. Hoyt Gatton Professor of Economics Department of Economics University Of Kentucky Lexington, KY Lexington, KY (859) office (859) office (859) fax (859) fax whoyt@uky.edu fscott@uky.edu Frank A. Scott, Jr. Gatton Professor of Economics and DGS Department of Economics University Of Kentucky December 2004

2 Project Summary: A result of the 1996 welfare reforms was that TANF/AFDC rolls declined dramatically, while SSI enrollment increased slightly. Since Food Stamp participation rates are considerably lower among SSI recipients than among TANF recipients, Food Stamp enrollment was significantly affected by welfare reform even though it was not directly targeted by the legislation. Understanding changes in Food Stamp participation thus requires a simultaneous analysis of participation in TANF and SSI. Along with welfare participation, Food Stamp participation declined in the late 1990 s. While the reductions in Food Stamp participation are generally not of the same magnitude as those in welfare participation, they are, nonetheless significant. Given relatively insignificant changes in the Food Stamp program as a result of PWROA of 1996, particularly when compared to those made with respect to welfare, the large reductions in Food Stamp participation seem puzzling. As a result of the observed reductions in participation in both programs, evaluating the link between AFDC/TANF participation and Food Stamp participation may be helpful in explaining why Food Stamp participation declined so sharply during this period. In this project we have directly examined the link not only between Food Stamp participation and AFDC/TANF participation but also between Food Stamp participation and SSI participation using county-level data on participation in these three programs. Our results from estimation using both a sample of states throughout the U.S. and only southern states indicate that a strong relationship exists between the level of Food Stamp participation and both welfare and SSI participation, even when controlling for numerous demographic, economic, and program characteristics that are likely to affect both eligibility and participation. If anything, the link 1

3 between TANF participation and Food Stamp participation in 2001 appears stronger than the link between AFDC participation and Food Stamp participation in Food Stamp participation responds in predictable ways to the different options states now have as a result of welfare reform to treat income and resources, eligibility, and assets for the purposes of determining Food Stamp eligibility and income. County administration as opposed to state administration has a significant negative impact on participation, while exempting child support from income and expanded categorical eligibility increase participation. State-required training and employment appear to have a significant negative on Food Stamp participation, particularly in the Southern states. While One-Stop Centers appear to have little impact on participation, at least in the South, the longer the time span for certification the higher the Food Stamp participation rate. We have also directly examined the change in Food Stamp participation, both in absolute terms (change in the number recipients per 1000 residents) and percentage terms, while controlling for the levels and changes in both AFDC/TANF and SSI participation. Absolute changes in Food Stamp participation between 1995 and 2001 are generally negatively related to the levels of both AFDC and SSI participation in However, changes in Food Stamp participation are positively related to changes in AFDC/TANF participation, suggesting that counties with large reductions in welfare case loads have large reductions in Food Stamp caseloads as well. This result, however, is not robust. When the change in Food Stamp participation is measured as a percentage change, we find that larger (percentage) reductions in welfare participation result in smaller (percentage) reductions in Food Stamp participation, indicating that Food Stamp and welfare participation might be better described as substitutes than complements. 2

4 Introduction and Problem Statement The obvious, and frequently striking, demographic and geographical variation in welfare recipiency has been a topic of research for both economists and sociologists. 1 The usual starting point for this research is the assumption that there are no inherent differences among eligible individuals in their likelihood to participate in welfare programs. Any differences in observed participation among demographic groups or across regions would thus be explained by differences in eligibility rates. Invariably, differences in welfare recipiency are not fully explained by differences in eligibility alone. Instead, among different demographic groups and across regions participation by eligible individuals and households can vary dramatically. Maps 1 and 2 give participation rates for Food Stamps for Kentucky and Arkansas in As these maps suggest, there is a great deal of geographical variation in Food Stamp participation among counties in these two states. Maps 3 and 4 give poverty rates for Kentucky and Arkansas in Again, there is a great deal of variation among counties. Of particular interest to us is to what extent and why there are differences in the variation across counties in Food Stamp participation and the poverty rate. While there is clearly a strong correlation between heavy or light Food Stamp participation and high or low poverty rates in a county, the correlation is not perfect. Some counties with high participation in Food Stamps are not among the ranks of the poorest counties and some of the poorest counties in these two states do not have the highest participation in food stamps. If poverty does not fully explain the variation in Food Stamp participation among counties, what other factors might explain it? 1 Hoyt and Scott (2002a) provide a summary of this research. 2 Maps were generated using the ERS Food Stamp Map Machine, located at 3

5 Several different patterns among welfare participants have attracted the attention of researchers, including intergenerational correlations, social network correlations, and physical location correlations. The fundamental question is whether spatial concentrations of poverty represent more than just collections of poor people who happen to live in the same place. If there are identifiable transmission mechanisms, the answer is yes. As others have suggested, possible mechanisms include positive and negative information flows, role modeling, attitude or stigma effects, and institutional infrastructure that supports welfare participation. We are thus interested in spatial concentrations of poverty and the existence of neighborhood effects. We propose to study the variation in participation rates across geographic locations in Food Stamps, Supplemental Security Income (SSI), and Aid to Families with Dependent Children (AFDC)/Temporary Assistance for Needy Families (TANF) using county-level data. By using county-level data, we can analyze differences in program policy parameters at the state level in combination with population eligibility characteristics at the county level to see how these differences influence variation in participation rates. And the use of county-level data enables us to explicitly consider the influence of geography on the transmission of welfare participation. While this study is national in scope, we focus on areas where welfare participation rates are highest. We pay particular attention to the South, and specifically to four regions within the South with high poverty rates: the Mississippi Delta Region, the Rio Grande Valley, the Black Belt, and the Appalachian Region. 3 Our primary emphasis is participation in the Food Stamp program. But since Food Stamp participation is closely connected to TANF and SSI participation, we simultaneously analyze participation in those two programs as well. We first analyze welfare participation in 1995, prior to the 1996 passage of major reforms to the AFDC 3 See Table 1 for a list of the counties included in each of these high poverty areas. 4

6 program (creating TANF) and minor reforms to the SSI program. We then analyze welfare participation in 2001, after the program changes have been in effect for awhile. This approach enables us to assess the use over time of the nation s major food assistance program. We are able to compare rural vs. urban and metro vs. non-metro participation in Food Stamps, as well as in TANF and SSI. We are able to analyze the effects of macro factors, such as program interactions, eligibility criteria, and benefit levels, and of micro factors, such as economic and demographic characteristics of the eligible populations, on Food Stamp participation. By examining food stamp participation both before and after welfare reform we are also able to analyze how welfare reform may have differentially affected food stamp participation in different regions and settings. Objectives This research project has a number of specific objectives: o To identify the determinants of program usage for Food Stamps, SSI, and AFDC/TANF. Explanatory variables fall into three broad categories. First are characteristics of individuals that by legislative design determine program eligibility. Second are variables describing program structure, specifically measures of benefits levels and the allocation of funding responsibilities. Third are characteristics of the population that serve as proxies for environmental or cultural factors that are unrelated to eligibility but may, nonetheless, influence welfare participation. o To compare changes in the determinants of program usage for Food Stamps, SSI, and AFDC/TANF before and after the welfare reforms of We specifically look at the interaction between AFDC participation and Food Stamp participation in the pre-reform period and compare it to the interaction between TANF participation and Food Stamp parti- 5

7 cipation in the posreform period to determine whether the relationship has changed. We also examine the interaction between SSI participation and Food Stamp participation pre- and posreform to determine the extent to which changes in the SSI program and the recipient population have impacted the use of Food Stamps. o To investigate county-by-county participation rates throughout the South in Food Stamps, SSI, and TANF. We first look at raw participation rates in each of the three programs. Then we control for differences in (estimated) eligible populations to calculate an adjusted participation rate. It is this variation in participation due to non-eligibility factors that is most relevant for identifying neighborhood or network effects, because this residual variation may yield clues about other factors that serve as transmission mechanisms for welfare dependence. We then compare the raw participation rates and the adjusted participation rates. In this report, for obvious reasons, we focus on participation in the food stamps program but consider this participation in the context of a model in which participation in other programs (AFDC/TANF and SSI) as well as the characteristics of these programs are considered. The determinants of participation in AFDC/TANF and SSI are not reported here. Research Methodology Since our interest is in explaining and understanding geographical variation in Food Stamp recipiency rates, particularly variation beyond that explained by variation in the eligible population, we estimate a participation, or more precisely recipiency, equation for Food Stamps. Since we believe the structure of and participation in other welfare programs, particularly, AFDC/TANF and SSI, may be important in understanding participation in Food Stamps, characteristics of these programs are included in the Food Stamp participation equation. Of 6

8 course, it follows that if the structure of and participation in AFDC/TANF and SSI influence Food Stamp participation, then Food Stamp participation will likely have an influence on participation in AFDC/TANF and SSI. For this reason we estimate a system of equations explaining participation in all three programs. 4 Before discussing more completely both the estimation procedure and explanatory variables, we turn our attention to measures of program participation. As previously explained, we examine participation in three major welfare programs: Food Stamps, AFDC/TANF, and SSI. Our focus is on participation. For Food Stamps and SSI we measure this in terms of participants per 1000 persons at the county level, and for AFDC/TANF we measure as cases per 1000 households at the county level. There are several advantages to the use of county-level data instead of state-level data as is more typically done. First, use of state-level data masks both the considerable variation in welfare program participation within states and what we consider some of the reasons that welfare participation may vary within states. Second, to the extent that regional characteristics influence Food Stamp and other program participation, the county, not the state, is probably a better approximation of the relevant region. If, for example, supply factors such as the extent of legal or medical services influence program participation, these factors are probably only relevant within a county or contiguous counties and certainly not throughout the state. If population density or concentration of ethnic groups lead to networks that increase information about programs, again this is probably relevant at the county or lower level, and not at the state level. Given both the limited geographical information and the limited geographical representation found in individual or households surveys on program participation (for example, SIPP, CPS, NLSY), a focus on participation aggregated to the county level ensures 4 This simultaneous system can be estimated either as a system or equation by equation using instrumental variables for each equation. We use the latter approach. 7

9 both identification at the county level and measures of participation in counties where populations are small and therefore representation in surveys might be very limited. As mentioned before, we believe that one explanation of geographical variation in Food Stamp participation may be differences in participation in other welfare programs, specifically AFDC/TANF and SSI. Given the significant structural changes that occurred with the transition from AFDC to TANF as well as the dramatic declines in participation, we wish examine geographical variation in Food Stamp participation as well as in AFDC/TANF and SSI both before welfare reform and after welfare reform. While SSI has not had the same major programmatic changes as AFDC/TANF, there has been a significant change in the composition of SSI cases during this period due to earlier changes in the program. Beginning in the early 1990 s but continuing throughout the decade, there have been significant increases in the percentage of persons receiving SSI who are classified as disabled, specifically with mental disorders and under eighteen years of age. Since the mix of SSI participants may influence participation in Food Stamps, we might expect different Food Stamp participation pre- and poswelfare reform for this reason as well. For these reasons, we examine geographic variation in Food Stamps, AFDC/TANF, and SSI pre- and poswelfare reform. While our focus in this project is on counties for states within the South, for completeness we have obtained county-level data for the entire U.S. Data on AFDC, SSI, and Food Stamp participation aggregated to the county-level were obtained for 1995 and data reflect the last full year prior to welfare reform data represent the posreform sample. We chose 2001 rather than an earlier year to allow more complete adjustment to the changes in welfare policy. 5 SSI participation rates are actually for January

10 Data on the number of Food Stamp cases at the county level are collected by the United States Department of Agriculture. These are referred to as the FNS-101 data. Their use has been rather limited. Wilde and Dicken (2003) and Goetz et al. (2002) are the only studies using these data of which we are aware. The data are now readily available electronically from the Bureau of Census ( which uses them as inputs in its simulation model for determining small area income and poverty estimates. Data are available for July of the years 1989, 1993, 1995, 1997, 1999, 2000, and Data on aggregate SSI recipients within a county are available electronically from 1990 to 1996 on the compact disc, USA Counties. Data on county-level SSI participation for 2000 and 2001 are also available electronically at the Social Security Administration website: Other years are available in hard copy in SSI recipients by state and county, an annual publication of the Social Security Administration. One advantage that this publication has over the data obtainable from USA Counties is that, in addition to the total recipients in a county, it provides a breakdown of recipients by category (aged and blind or disabled) as well as age (under 18, 18-64, 65 or older). Since participation in Food Stamps is likely influenced not only by the number of SSI participants but also the composition of SSI participants with respect to category and age, we have entered these data manually. County-level data on AFDC/TANF are much more difficult to obtain. In contrast to Food Stamps and SSI which are administered federally, AFDC/TANF is administrated by state governments. Since federal assistance is given to state governments based on state caseloads, no reporting of caseloads at the county level to federal agencies is required and none is federally available. While caseloads for some counties are available in federal publications or data 9

11 collections such as USA Counties or The Metropolitan Area Data Book, there is no federal source for data on AFDC/TANF for all counties. For earlier work (Hoyt and Scott, 2002b) we obtained county-level AFDC participation for 45 states for the year This was done by contacting and requesting the data from each of the states agencies for administering AFDC. We have had to use the same labor-intensive approach to get TANF participation data for We contacted each state and obtained TANF participation data for each county for the year To insure consistency, we also obtained AFDC participation data for In addition to the structure of and participation in AFDC/TANF and SSI, Food Stamp participation is influenced by a number of other factors for which we also obtained data. With welfare reform, several changes occurred with the Food Stamp program as well as the change from AFDC to TANF. One of the changes was expanded flexibility by states in administering Food Stamps in their state. While benefit levels are federally set, states have significant flexibility in defining income and resources, asset limitations, and expanding the eligible population by using TANF funds if they so desire. Table 2 lists some of the options available to the states with a summary of what each state does. One distinction among states is whether the program is administered by the state or county government. Another option states have, which eight currently use, is to exclude child support as income, thereby increasing Food Stamp income. Sixteen states currently have Employment & Training Pledges that place a three-month limit on Able-Bodied Adults Without Dependents (ABAWD) to obtain qualifying education, training or employment in order to continue receiving food stamps. Thirteen states offer Transitional Benefits enabling TANF recipients, upon the loss of TANF, to continue to receive Food Stamp income equal to or above the level they received on TANF for up to five months. Thirty-nine states have Expanded Categorical Eligibility that essentially eliminates any asset 10

12 tests for households receiving TANF money. In addition, states have a great deal of flexibility in the treatment of automobiles as an asset. 6 While the alternative options are reported in Table 2, the complexity of these options makes it difficult to operationalize their use in our empirical analysis. In addition to differences among the states in the options they employ, attention has been given to how states might vary in certifying or re-certifying that a household is eligible for food stamps. Table 2 reports summary data on time for recertification for a variety of cases calculated from the Economic Research Service (ERS) data on quality control as reported in Bartlett, et. al. (2002). We create an average time for recertification for all categories of recipients based on these data. First and most obvious are variables that can be considered as determinants of or proxies for determinants of eligibility. These include measures of poverty and income within the county as well as measures of family size. Annual estimates of poverty rates and median income were obtained from the Census Bureau ( In addition to giving the poverty rate for the entire county, the Census estimates the poverty rate for households with children under 18 and elderly households as well. Analogously, AFDC/TANF and SSI participation depend on the eligible populations for these programs. While poverty and income are obviously relevant for these means-tested programs, other factors are important in determining eligibility. As either disability or age is necessary to qualify for SSI, we have obtained measures of both. Annual estimates of the elderly population for each county are available from the Census Bureau. A measure of disability (selfreported) is available from the decennial Census. For AFDC/TANF, we include as proxies for 6 Information on state food stamp options are from Food and Nutrition Service, Food Stamp Program Options: State Options Report, United States Department of Agriculture, September

13 the eligible population in a county the percentage of households with children under 18 and the percentage of family households with a single parent. Again, both are obtained from the U.S. Census Bureau. In addition, following other studies of program participation, we include demographic factors unrelated to eligibility for the program. Thus we include measures of the ethnic and racial composition of the county s population, the age distribution for the county, and the education background of county residents. These factors should not affect eligibility but may affect informational networks about programs or affect participation for other reasons. Again, data on these factors were obtained from the U.S. Census, many of which are provided as annual estimates. As factors explaining participation in all three programs we include measures of locational characteristics of the county (population density, urban/rural, MSA/non MSA), employment and economic conditions (earnings, employment rates), and region. Data on employment and earnings were obtained electronically from the Department of Commerce Bureau of Economic Activity via REIS (Regional Economic Information System). In addition to general employment and earnings levels, we include measures of earnings in industries which we characterize as suppliers of services related to participation in welfare programs. Specifically, following our earlier study on AFDC and SSI participation (Hoyt and Scott, 2002b), we include county-level earnings (percentage of total earnings) in the health, legal, and social service sectors as well as government spending at the local, state, and federal level. Among the three programs, Food Stamps is unique in having uniform benefits and eligibility requirements across all the states. While SSI is also federally administered, some states do supplement benefits and have different criteria for eligibility for Medicaid for SSI recipients as 12

14 well as different benefit levels for Medicaid recipients. Benefits and to a lesser extent eligibility for AFDC/TANF vary significantly across states. Data on benefit levels for AFDC/TANF for the different states were obtained from the Statistical Abstract of the United States. Data on state supplements and the Medicaid criteria for SSI were obtained from the Social Security Administration. A comprehensive list of other differences among states in eligibility, work, and time limits for AFDC and TANF was obtained from the Urban Institute. Data Analysis As our discussion of the data suggests, we estimate a system of participation equations for the three programs in which participation depends on a number of different (sets of) factors. Formally we can express the model as F ijt = 1 t ATjt + α2ssiijt + α3tspjt + α4tefijt + α5teatijt + α6tesijt + α7tcijt + α8tsijt + α9tdijt + α10t α L + π + ε ijt F rjt F ijt AT ijt = 1 tfi jt + + β2tssii jt + β3t ATPjt + β4t EFijt + β5t EATijt + β6t ESijt + βtcijt + β8t Sijt + β9t Dijt + β10t β L + π + ε ijt AT rjt AT ijt and SSI ijt = 1 tatj + γ2tfi j + γ3tspji + γ4tefijt + γ5teatijt + γ6tesijt + γ7teijt + γ8tsijt + γ9tdijt + γ10t γ L + π + ε ijt S rjt S ijt where the subscript i refers to county; j to state; t to year; and r to region (cluster of counties). The terms F ijt, AT ijt, and SSI ijt refer to the participation rates in Food Stamps, AFDC/TANF, and SSI in county i in state j in year t. The terms ATP j and SP j represent the set of variables characterizing the program structure and benefits for AFDC/TANF and SSI in state j. The terms E Fijt, E ATijt, and E Sijt are characteristics of the county population that serve as measures of the fraction of individuals (SSI) or households (AFDC/TANF, Food Stamps) in the county eligible for the programs. These measures would include the poverty rate and characteristics of family structure. The subscript AT refers to AFDC/TANF, F is for Food Stamps, and S is for SSI. The term C ijt refers to economic and employment conditions in the county and the term S ijt 13

15 refers to supply factors discussed in the preceding section. Demographic characteristics of the county population unrelated to eligibility, such as race and ethnic composition, are included in D ijt and locational characteristics such as population density and whether the county is in an MSA are included in L ijt. The term, k=f,at,s represents a spatial element to participation that might indicate, for example, spatial autocorrelation in participation among adjacent counties. The term k ε rt k π rt, k=f,at,s is a county-specific error term. Note that the coefficients, α, β, and γ are not assumed to remain constant over time. Specifically, we do not believe that the relationships between participation in the programs and program and eligibility characteristics (AT jt, SP jt, E kjt, k=f,at,s) are the same before and after welfare reform. The inclusion of other program participation rates in the equation explaining participation of a program will result in inconsistent estimates using ordinary least squares if the error terms of the three equations are correlated. For this reason we estimate these equations separately using an instrumental variable technique. As discussed in more detail in our results sections, our instruments are variables included in one equation but not the others and are composed of eligibility and program characteristics of the three programs that vary among states and the programs. Estimation of the parameters of these participation equations enables us to address the issues in which we are most interested: 1) the reasons for geographical variation in Food Stamp participation among counties; 2) what impact welfare reform has had on the determinants of geographical variation; and 3) what impact welfare reform has had on Food Stamp participation. Empirical Results Participation by Location 14

16 Before turning to regression results it is informative to examine the raw data on program participation before and after welfare reform. Figures 1a-j contain summary data on participation rates in the Food Stamp, AFDC/TANF, and SSI programs for 1995 and 2001 for the entire U.S. (lower 48 states). Figure 1a illustrates Food Stamp participation. Between 1995 and 2001 Food Stamp participation declined everywhere, but the decline was sharpest in the West. Figure 1b, which also illustrates Food Stamp participation, reveals that the decline in Food Stamp participation was greatest in high poverty areas within MSA s. Figures 1c and 1d illustrate SSI participation. There were only modest changes in SSI participation between 1996 and The biggest decline in SSI utilization occurred in the South. Figures 1e and 1f illustrate AFDC/TANF participation. The obvious conclusion is that between 1995 and 2001 participation in AFDC/TANF declined sharply everywhere, in all regions of the country and in MSA and non- MSA areas. Figure 1g contains the ratio of Food Stamp recipients to AFDC/TANF recipients in 1995 and 2001 by region of the country. Food Stamp usage declined the least relative to AFDC/TANF usage in the Midwest and the South, and, as Figure 1h shows, in non-msa regions. Figure 1i contains the ratio of SSI recipients to AFDC/TANF recipients in 1995 and 2001 by region. There has been a significant decline in the utilization of AFDC/TANF relative to SSI between 1995 and 2001, and it is most evident in the South. Figure 1j shows that SSI participation relative to AFDC and TANF is higher in non-msa regions, whether the county is in a high poverty area or not. Figures 2a-j contain the same information for the South. An additional breakdown is to add the high poverty areas of the Mississippi Delta, Appalachia, the Black Belt, and the Rio Grande Valley. Our definition of these areas is more restrictive than, for example, the definition 15

17 of Appalachia by the Appalachian Regional Commission or the definition of the Mississippi Delta by the Mississippi Delta Authority. Our definition of Appalachia is restricted to what is referred to as central Appalachia by the ARC. However, as with these other poverty areas we require that all counties that are included have poverty rates in 2000 exceeding 18% and not be within an MSA. In addition, included counties are coterminous with at least one other county in the poverty region. A list of the counties in these poverty regions is found in Table 1. It is especially important to understand program utilization and program interactions in these specially designated regions. Pre and post welfare reform Food Stamp recipiency rates for the southern states are illustrated in Figure 2a. There is considerable variation across states in both the utilization rates and changes in utilization. Food Stamp participation declined sharply between 1995 and 2001 in Georgia, Maryland, Mississippi, Texas, Virginia, and West Virginia. There was relatively little decline in Arkansas, Florida, North Carolina, and South Carolina. Figure 2b contains the same information according to MSA and Poverty status, and for designated poverty areas. Food Stamp utilization was relatively high and declined more sharply in the Delta and Rio Grande Valley regions, as compared to the Appalachia and Black Belt regions. Figure 2c contains AFDC and TANF recipiency rates for southern states in 1995 and Again there is considerable variation across states, but there is one common theme participation declined sharply everywhere between 1995 and Figure 2d illustrates AFDC/TANF participation for MSA, Poverty, and specific poverty areas. The proportionate changes from 1995 to 2001 were similar in the Appalachia, Black Belt, and Delta regions. In each case the relative decline in recipients was more pronounced than in the Rio Grande Valley region. Figure 2e shows SSI recipiency rates in 1996 and SSI participation was almost 16

18 unchanged or actually increased in the District of Columbia, Delaware, Florida, Kentucky, Maryland, and West Virginia. Declines were only slight in the other states. As can be seen in Figure 2f, SSI participation actually increased in Appalachia, and declined the most in the Delta region. The ratios of Food Stamp recipients to AFDC/TANF recipients in 1995 and 2001 for the southern states are illustrated in Table 2g. In some states, such as Florida and South Carolina, the decline in AFDC/TANF participation was much more pronounced that the decline in Food Stamp participation. In other states, such as Tennessee and Texas, the relative declines were closer in magnitude. Similarly, the decline in AFDC/TANF participation was higher relative to the decline in Food Stamp participation in the Black Belt region than in the Rio Grande Valley region, as is shown in Table 2h. Figure 2i illustrates the increase in SSI participation relative to AFDC/TANF participation between 1995 and This measure combines the effects of small declines or even increases in SSI participation and large declines in AFDC/TANF participation. Large relative shifts occurred in Florida, Louisiana, Maryland, and Mississippi. Figure 2j shows that large relative shifts occurred in Appalachia, the Black Belt, and the Delta, but not in the Rio Grande Valley. Estimation Results While much can be gleaned from examining raw participation rates, further insights can be obtained by using multiple regression analysis to investigate participation rates. As our examination of participation rates indicates, throughout the U.S. there were significant decreases in welfare participation from 1996 to At the same time, Food Stamp participation also decreased significantly. Of course, these reductions in the U.S. overall are well known, as well as reductions, though not uniformly, in every state. Our contribution to the analysis of welfare 17

19 reform is the ability to track how these changes varied among rural and urban regions and in areas particularly prone to poverty. Given that we have geographical variation and identification, we also have the opportunity to examine how differences among states in the structures of their programs, particularly in Food Stamps, affect Food Stamp participation. Previous studies have been unable to perform this type of analysis because of the availability of data available at or only identified at the state level. Food Stamp Participation While the examination of mean participation rates is useful, it does not offer much insight into how and why reductions in program participation, particularly for food stamps, occurred. As discussed earlier, we believe that examining participation in Food Stamps in a framework that simultaneously considers participation in both welfare programs (AFDC in 1995 and TANF in 2001) and SSI may offer some insights into these reductions in participation. We begin our examination of Food Stamp participation with a cross-sectional analysis of the level of Food Stamp participation in both 1995 and As discussed earlier, we are interested in the influence of a number of different factors on Food Stamp participation. Some of these factors can be expected to be positively related to Food Stamp participation (at the county level) because they will determine the fraction of the county population that is eligible (poverty measures, unemployment rates, and mean income, for example). Other factors not directly related to eligibility might be correlated with participation conditional on eligibility. That is, these factors might be related to characteristics of the eligible population that make participation more or less likely (demographic characteristics such as age, race, non-native fraction of the population, education). Locational characteristics (rural vs. urban, for example, or population density) also may influence participation. We also include measures of supply factors, i.e., 18

20 sectors of the economy that might influence participation such as legal, health, or social services. Finally, characteristics of the programs that vary among states are also included. Variable descriptions and summary statistics for 1995 and 2001 both for the U.S. and the South are found in Table 3. We estimate structural rather than reduced-form equations. In our analysis, in the structural estimates we include as explanatory variables AFDC (1995) or TANF (2001) participation and SSI participation. As we argued earlier, these participation rates are endogenous and are themselves functions of other factors such as welfare and SSI program characteristics that vary across states and characteristics of the county population. Estimating structural rather than reduced-form equations requires instruments to independently identify the endogenous variables, which are in our case participation in AFDC/TANF or SSI. For AFDC/TANF we use as instruments characteristics of the programs, including maximum benefits, work requirements, child care, and time limits, as well as measures of eligibility, including the fraction of households headed by single parents for AFDC/TANF. For SSI we have as instruments whether the state offers a supplement and if so how much, the link with Medicaid eligibility in the state, and work disability rates for the county as reported in the Census of Population. These regression results for 1995 are contained in Table 4 and for 2001 in Table 5. Columns (a) and (d) in Table 5 are directly comparable to the results reported for all states and for southern states in Table 4. Columns (b) and (c) in Table 5 contain results using additional Food Stamp program characteristic variables for all states, and columns (e) and (f) contain similar results for southern states. In 1995 for all states and for just the southern states Food Stamp participation rates were significantly related to AFDC and SSI participation rates. That 19

21 relationship was still true in 2001 for all states, but in the southern state sample there was a tight connection with TANF participation but no significant connection between Food Stamp participation and SSI participation. In both years and in both samples the proportions of the population with incomes between 0% and 50% and between 50% and 100% of the poverty level were positively and significantly related to Food Stamp participation. In 1995 there was a significant connection between Food Stamp participation and per capita income, but that relationship does not show up in the 2001 regressions. The unemployment rates among both males and females did not seem to affect Food Stamp participation in 1995, but are significant in the 2001 regressions, with the exception of the female unemployment rate in southern states. The one consistent age distribution effect on Food Stamp participation seems to be the fraction of the population under age 5 years old, which is consistently positive and significant. In the 1995 sample the African-American fraction of the population is a positive and significant predictor of Food Stamp participation. The Hispanic fraction is positive and significant in the southern state sample. Those results change in the 2001 sample. Percent African-American is not significant in the southern state sample, and percent Hispanic is negative and significant in the southern state sample. The location variables, MSA vs. non-msa and population density, do not show any consistent relationship with Food Stamp participation, once all of the other control variables are included in the analysis. Within the southern state sample, the poverty regions are characterized by higher Food Stamp participation rates, even after controlling for poverty rates, unemployment rates, income levels, education levels, racial and ethnic composition, and other factors. 20

22 The effects of specific Food Stamp program characteristics that were in place in 2001 can be seen in the regressions reported in columns (b), (c), (e), and (f). In the all states sample, each of the program variables has the expected effect on Food Stamp participation rates. Expanded eligibility and exempted child support increase participation, and county administration and employment and training requirements decrease participation. In the southern state sample, increasing the number of months between recertification significantly increases Food Stamp participation. One final item of interest is One-Stop Centers. We collected data on the number of One-Stop Centers in each county in the southern states. When this variable is included in the southern state sample, there is no significant effect on Food Stamp participation. Changes in Food Stamp Participation Tables 6 and 7 analyze the absolute and the percentage change in Food Stamp participation rates between 1995 and Results are reported for both the all states sample and the southern states sample. Three different specifications are run for each sample. The first includes the AFDC participation rate and the SSI participation rate in 1995 as an explanatory variable. The second includes the change in the AFDC/TANF participation rate and the change in the SSI participation rate between 1995 and The third includes both the 1995 level and the change in both AFDC/TANF and SSI participation. As can be seen in Table 6, in the all states regressions the decline in Food Stamp participation was greater the higher the level of AFDC participation in 1995 (column (a)) and the greater the decline in AFDC/TANF participation between 1995 and 2001 (column (b)). When both variables are included in the regression as reported in column (c), neither the level of AFDC participation in 1995 nor the change in AFDC/TANF are associated with greater declines in Food Stamp participation between 1995 and In contrast, while the change in SSI 21

23 participation between 1995 and 2001 had no significant impact the change in Food Stamp participation, greater participation in SSI in 1995 had resulted in a statistically-significant reductions in Food Stamp participation between 1995 and Food Stamp program characteristics affect participation in predictable ways. Administration of Food Stamps at the county level is associated with significant declines in participation. Exempting child support is associated with smaller declines in participation, while imposing an employment and training requirement is associated with larger declines in participation. The fraction of the county s population with incomes below the poverty level does not seem to be connected with changes in Food Stamp participation. The greater the change in unemployment, both among males and females, the greater the change in Food Stamp participation between 1995 and As inspection of columns (d) (f) indicate, the coefficients, in sign and significance, for these variables is quite similar for the sample of only Southern States. Counties with larger fractions of the population below age 5 or greater than age 64 experienced larger declines in Food Stamp participation between 1995 and Where the African-American or Hispanic proportion of the county s population was greater, the greater was the decline in Food Stamp participation. The reverse is true for counties where the Native American proportion of the population was greater. Controlling for other factors, counties with larger fractions of the population living in rural areas experienced larger declines in Food Stamp participation. In the all states sample there were significantly greater declines in Food Stamp participation in the South and the Northeast. In the southern states sample, the designated high poverty regions did not appear to experience greater or lesser changes in Food Stamp participation than other areas, after controlling for other factors. In the sample of Southern states, the 22

24 coefficients on our categorical variables indicating a county located in one of our four poverty pockets indicates that only for counties in the Black Belt region is the change in Food Stamp participation significantly different than that in counties with similar demographic and economic characteristics, with the decrease in Food Stamp participation in the Black Belt counties approximately 9-18 recipients per 1000 less than similar counties. It may not be too surprising that counties with the highest AFDC rolls in 1995 exhibited the greatest absolute decrease in Food Stamp participation, as these counties will also exhibit the highest Food Stamp rolls as well. For this reason we also estimate the change in Food Stamps participation between 1995 and 2001 in percentage terms, with changes in AFDC/TANF and SSI participation measured in percentage terms as well. Results of this set of estimates are found in Table 7. In addition, inference using the absolute change in AFDC/TANF is made difficult because of the high correlation between the level of AFDC in 1995 and the absolute change in AFDC/TANF participation between 1995 and Focusing on estimates using all the states (columns (a) (c)) and the interaction of the programs, we find results that are somewhat different than those examining the determinants of the absolute change in Food Stamp participation found in Table 6. In column (a) of Table 7, we see that the level of AFDC cases in 1995 has a positive, but insignificant coefficient contrasting with the significant negative coefficient on this variable in the analogous estimate of absolute changes in participation. The coefficient on SSI participation in 1995 continues to be negative and significant. Perhaps more interesting is the coefficient on the percentage change in AFDC/TANF participation in both the specifications reported in column (b) and column (c). The negative coefficient on this variable indicates that, ceteris paribus, Food Stamp participation 7 The correlation coefficient between the two variables is In contrast, the correlation coefficient between the 1995 level of AFDC and the percentage change in AFDC/TANF is only

25 increased more (or decreased less) in counties where there were greater percentage decreases in AFDC/TANF rolls. The results for the sample of Southern states are reported in columns (d) (f). Unlike the results with all states, the results for the South indicate that the higher the level of AFDC participation in 1995, the greater the percentage decrease in Food Stamp participation. The level of SSI participation in 1995 has a positive impact on Food Stamp participation when no changes in participation are included (column (d)) but no significant impact when changes in participation are included (column (f)). Similar to the results with all states, the greater the percentage reduction in AFDC/TANF, the smaller the (percentage) reduction in Food Stamp participation (columns (e) and (f)) though this effect is statistically insignificant when the level of AFDC participation is included (column (f)). With a few exceptions, the impacts of the other variables on the change in Food Stamp participation are similar to those found in Table 6. The magnitudes of coefficients obviously change as the magnitude of the dependent variable is significantly smaller. In general, the impacts of the Food Stamp program variables seem to be insignificant and perhaps less stable with respect to changes in our specifications. Summary and Conclusions Along with welfare participation, Food Stamp participation declined in the late 1990 s. While the reductions in Food Stamp participation are generally not of the same magnitude as those in welfare participation, they are, nonetheless significant. Given relatively insignificant changes in the Food Stamp program as a result of PWROA of 1996, particularly when compared to those made with respect to welfare, the large reductions in Food Stamp participation seem puzzling. As a result of the observed reductions in participation in both programs, evaluating the 24

26 link between AFDC/TANF participation and Food Stamp participation may be helpful in explaining why Food Stamp participation declined so sharply during this period. In this project we have directly examined the link not only between Food Stamp participation and AFDC/TANF participation but also between Food Stamp participation and SSI participation using county-level data on participation in these three programs. Our results from estimation using both a sample of states throughout the U.S. and only southern states indicate that a strong relationship exists between the level of Food Stamp participation and both welfare and SSI participation, even when controlling for numerous demographic, economic, and program characteristics that are likely to affect both eligibility and participation. If anything, the link between TANF participation and Food Stamp participation in 2001 appears stronger than the link between AFDC participation and Food Stamp participation in Food Stamp participation responds in predictable ways to the different options states now have as a result of welfare reform to treat income and resources, eligibility, and assets for the purposes of determining Food Stamp eligibility and income. County administration as opposed to state administration has a significant negative impact on participation, while exempting child support from income and expanded categorical eligibility increase participation. State-required training and employment appear to have a significant negative on Food Stamp participation, particularly in the Southern states. While One-Stop Centers appear to have little impact on participation, at least in the South, the longer the time span for certification the higher the Food Stamp participation rate. We have also directly examined the change in Food Stamp participation, both in absolute terms (change in the number recipients per 1000 residents) and percentage terms, while controlling for the levels and changes in both AFDC/TANF and SSI participation. Absolute 25

27 changes in Food Stamp participation between 1995 and 2001 are generally negatively related to the levels of both AFDC and SSI participation in However, changes in Food Stamp participation are positively related to changes in AFDC/TANF participation, suggesting that counties with large reductions in welfare case loads have large reductions in Food Stamp caseloads as well. This result, however, is not robust. When the change in Food Stamp participation is measured as a percentage change, we find that larger (percentage) reductions in welfare participation result in smaller (percentage) reductions in Food Stamp participation, indicating that Food Stamp and welfare participation might be better described as substitutes than complements. 26

28 References Bartlett, Susan, Nancy Burstein, and William Hamilton, Food Stamp Program Access Study Final Report E-FAN November 2004 Rebecca Blank, Evaluating Welfare Reform in the United States, Journal of Economic Literature, December 2002, Janet Currie and Jeffrey Grogger, Explaining Recent Declines in Food Stamp Program Participation, Brookings-Wharton Papers on Urban Affairs: Food and Nutrition Service, Food Stamp Program Options: State Options Report, United States Department of Agriculture, September William Greene, Econometric Analysis (4 th edition), Upper Saddle River, New Jersey: Prentice Hall, Stephan Goetz, Anil Rupasingha, and Julie Zimmerman. Food Stamp Program Participation Dynamics in US Counties and States: Final Report. University Park, PA: The Pennsylvania State University. Thomas Hirschl and Mark Rank, The Effect of Population Density on Welfare Participation, Social Forces, September 1991, William Hoyt and Frank Scott, Geographic Variation in Welfare Participation Rates: Identifying Poverty Pockets, working paper, University Of Kentucky, April 2002a. William Hoyt and Frank Scott, Participation in AFDC/TANF and SSI: Analyzing Substitution between Programs by Recipients and State Governments, working paper, University Of Kentucky, August 2002b. Jonathan Jacobson, Nuria Rodriguez-Planas, Loren Puffer, Emily Pas, and Laura Taylor-Kale, Mathematica Policy Research, Inc., The Consequences of Welfare Reform and Economic Change for the Food Stamp Program Illustrations from Microsimulation, Economic Research Service, January Robert Kornfeld, Abt Associates Inc., Explaining Recent Trends in Food Stamp Program Caseloads, Economic Research Service, March Park Wilde, The Impact of Race and Ethnicity on County Level Food Stamp Program Participation Measures in California, Economic Research Service, December Park Wilde and Chris Dicken, Using County-Level Data to Study the Race and Ethnicity of Food Stamp Program Participants in California, working paper, Economic Research Service for presentation at the American Agricultural Economics Association Meetings, July 27-30,

29 Sheila Zedlewski and Sarah Brauner, Declines in Food Stamp and Welfare Participation: Is There a Connection? Urban Institute Working Paper 99-13, October Jim Ziliak, Craig Gundersen, and David Figlio, Food Stamp Caseloads over the Business Cycle, Southern Economic Journal, April 2003, Jim Ziliak, David Figlio, Elizabeth Davis, and Laura Connelly, Accounting for the Decline in AFDC Caseloads: Welfare Reform or the Economy, Journal of Human Resources, Summer 2000,

30 Map 1: Food Stamp Participation in Kentucky, 1999 Map 2: Food Stamp Participation in Arkansas, 1999 Legend Per Capita FSP Participation, 1999 Less than 4.8% 4.8% - 8.7% 8.7% % 14.3% % 29

31 Greater than 23.1% Map 3: Poverty Rate in Kentucky, 1999 Map 4: Poverty Rate in Arkansas, 1999 Legend Per Capita Poverty, 1999 Less than 10% 10% % 14.6% - 20% 20% % Greater than 28.5% 30