Creating Better Comparison Groups for Estimating the Effects of SNAP on Food Insecurity

The Supplemental Nutrition Assistance Program (SNAP; formerly the Food Stamps Program) is a federal program that provides funds for food to low-income households in an effort to alleviate food insecurity. Estimating the effects of participating in SNAP is challenging because gross income-eligible nonparticipants are not an effective control group. Eligible households not participating have selected out of the program, likely based on factors other than income. However, eligibility for SNAP is based not only on gross income but also net income, calculated as gross income less several household expenses, as well as asset requirements. Further, factors such as local retail prices may affect household stability and food insecurity. This paper investigates whether comparisons of SNAP participants and nonparticipants may be improved with the inclusion of household expenses, assets, and local retail prices.

I use restricted-use data from the USDA Economic Research Service’s National Household Food Acquisition and Purchase Survey (FoodAPS), collected from April 2012 to mid-January 2013. FoodAPS is a nationally representative survey of 4,826 households in 27 continental U.S. states. SNAP participating households were oversampled, providing a larger sample of these households with which to consider program factors, and SNAP participation is measured corroborated using state SNAP administrative data. FoodAPS contains detailed observations of household demographic and socioeconomic measures, household expenses, individual income, SNAP receipt, food insecurity, meals and snacks, and the food environment. I limit the sample to households with children at or below 130 percent of the federal poverty line.

I use a series of propensity score matching regressions to match SNAP participants and income eligible nonparticipants. First, I use typical sociodemographic controls found in previous SNAP food security literature. Second, I use typical sociodemographic controls and add net income, a factor used in eligibility for SNAP as well as benefit amount. The third matching algorithm includes other eligibility factors, such as assets owned, car ownership, and broad-based categorical eligibility. In specification four I include with all of the above local retail food prices. I then use each matching algorithm to measure the impact of SNAP participation on food insecurity. I then compare these results to each other and to existing literature on SNAP and food insecurity. In particular, I compare to the bounds on the effects of food security in households with children identified by Gundersen et al (2017). I expect that matching using more comprehensive eligibility criteria and household characteristics will provide a more reliable estimate of SNAP’s impacts on food insecurity.

This research does not offer a causal estimate. Other unobserved factors related to SNAP participation, such as informal social support through family, friends, or faith community, are still unaccounted for. We may still be missing some of the social support aspects. However, I provide information about which factors may better create control groups, which will strengthen food assistance and food insecurity research going forward.