Sociodemographic Factors, Supplemental Nutrition Assistance Program Participation, and Health: The Case of Low-Income Individuals in Tennessee

Year: 2008

Research Center: Southern Rural Development Center, Mississippi State University

Investigator: Yen, Steven T., Donald J. Bruce, and Lisa Jahns

Institution: University of Tennessee

Project Contact:
Steven T. Yen
Department of Agricultural Economics
The University of Tennessee
Knoxville, TN 37996-4518
Phone: 865-974-7474


There is continued interest in the effects of food assistance programs on consumer welfare. The outcome variables of interest include food expenditures, nutrient intakes, and direct and indirect measures of health. This study addresses the effects of participation by individuals in the Supplemental Nutrition Assistance Program (SNAP) on self-assessed health (SAH) status—a widely used indicator of health-related quality of life. SNAP participation can affect health outcomes in many ways. First, to the extent that SNAP benefits represent effective income increases, the additional purchasing power can allow individuals to consume more and better health care. SNAP participation can improve health outcomes in other ways—for instance, by reducing the severity of food insecurity—which in turn can lead to better health. The role of SNAP participation in health has largely been investigated with national datasets. This study focuses on the South by using survey data for a sample of low-income individuals in Tennessee.

The primary data source is the Family Assistance Longitudinal Study (FALS), a collaborative effort of the Tennessee Department of Human Services and research organizations at the University of Tennessee and the University of Memphis. The FALS collected data from a large random sample of individuals from Tennessee who participated in Families First as of January 2001. Included in the survey are questions regarding food stamp participation and SAH outcomes, as well as sociodemographic characteristics of the survey participants. The study sample was drawn from the 9th and 10th waves of the survey collected in 2007 and 2008, respectively. While nearly all respondents are SNAP eligible, the FALS data were supplemented with administrative records from the Admiral database to enhance the accuracy of eligibility determination.

An ordered probability model with binary regime switching was developed to accommodate endogeneity of the SNAP participation variable and discrete (ordinal) nature of the SAH variable. The model is an extension of conventional switching regression models in that the primary outcome equation is an ordered probability model versus a continuous regression model. Further, in contrast to conventional models based largely on the bivariate normal distribution of the error terms, the error distribution was specified as a non-Gaussian distribution using the copula approach. Specifically, the marginal distributions of error terms in both the participation and SAH equations are specified as the generalized log-Burr distribution, which nests the extreme value and the logistic distributions. These marginal distributions are linked by the copula function to produce a bivariate distribution, which accommodates both skewness in and correlation between the two error terms. Three alternative copula functions (Gaussian, Frank, and Clayton) are considered along with two marginal distributions (Gaussian and generalized log-Burr).

The Clayton-Burr model was selected as the preferred specification based on nonnested specification tests. The Gaussian-Gaussian model, used extensively in models with sample selection, is rejected at the 1-percent level of significance, favoring the Clayton-Gaussian model. Endogeneity of regime switching was found, as was skewness in the error distribution, suggesting that conventional statistical models based on the Gaussian distribution or ones not accommodating such endogeneity would have produced biased and misleading effects of SNAP participation and sociodemographic variables on SAH.

Two sets of marginal effects are calculated, which allow further examination of the roles of explanatory variables in SNAP participation and SAH. Echoing findings reported in the literature, the number of children in the household contributes to SNAP participation. Previous-year income and current employment have negative effects on SNAP participation, whereas health insurance coverage contributes to SNAP participation.

Differentiated effects on SAH are found for many variables between SNAP participants and nonparticipants. Participants who are White and employed are more likely to be in the good, very good, and excellent SAH categories, whereas having a child with a health condition(s), being in the younger age (30-44) category, and having been hungry have the opposite effects. None of these variables has a significant effect on health among SNAP nonparticipants. Among both SNAP participants and nonparticipants, being divorced, separated or widowed, having an adult with a health condition(s), and being in the older age (45-64) category all have positive effects on the probabilities of poor or fair health and negative effects on the probabilities of good to excellent health, whereas the number of children have the opposite effects. The effects of age are substantial. For instance, SNAP participants ages 45-64 are 10.36 percent more likely to have poor health, 10.71 percent more likely to have fair health, and 5.05 percent, 9.18 percent, and 6.84 percent less likely to be in the good, very good, and excellent health categories, respectively, than are their younger (age < 45) counterparts. The effect of age on SAH is also very notable among the nonparticipants.

The calculated average treatment effects (ATEs) suggest that the effect of SNAP participation on health is negative overall. Specifically, participation in SNAP increases the probabilities of being in the low (poor or fair) SAH categories and decreases the probabilities of being in the higher (very good and excellent) categories. On a 1-5 scale, participation in SNAP decreases SAH level by 0.97. In sum, differentiated effects of sociodemographic variables on SAH are found between SNAP participants and nonparticipants, and SNAP participation has a negative effect on SAH. Given our use of survey data for a sample of current and former participants in Tennessee’s welfare program, we view these negative effects of SNAP on SAH as suggesting that the neediest families (such as those with the lowest SAH) might be more likely to participate in SNAP. Alternatively, those on (or recently on) Families First but not participating in SNAP are likely to be the least needy families. This inverse relationship between SNAP participation and SAH reveals that, while SNAP does not necessarily lead to improved SAH, the SNAP program is appropriately targeted to the neediest families in the State of Tennessee.

Direct inquiries about this study to the Project Contact listed above.