Food security continues to be an important public policy issue in the United States with almost 50 million people living in households that were food insecure at some point in 2013. A number of food assistance programs (FAPs) are available to help reduce food insecurity. These include the Supplemental Nutrition Assistance Program (SNAP), the National School Lunch Program (NSLP), and the School Breakfast Program (SBP). These programs generally serve different populations and provide different benefits. In spite of the fact that many families participate in more than one FAP at the same time, most research focuses on one program at a time.
Most existing research focuses on the SNAP program. Although one would expect participation in SNAP to reduce food insecurity, descriptive analysis typically finds that SNAP participation is associated with increased food insecurity. Accounting for selection on observables using methods such as linear regression, logistic regression, and propensity score matching typically reduces the positive association between program participation and food insecurity. Methods that allow for selection on unobservables, such as instrumental variables and fixed effects models using panel data find mixed results. There is less work on the NSLP or the SBP, and still less on combinations of programs.
Additionally, little research explores the role of program participation history on food insecurity. There is some evidence of a “dose-response” relationship where more benefits have a larger effect, and there is some evidence that a household’s income history affects food security (e.g., negative income shocks are associated with increases in food insecurity).
With this background in mind, this study seeks to answer three questions:
- What is the effect of participating in multiple FAPs on food insecurity?
- What is the effect of the history of participation (in SNAP) on food insecurity?
- What effects do other changes in household circumstances have on food insecurity?
To answer these questions, data from the 2008 Survey of Income and Program Participation (SIPP) are used. The 2008 SIPP was conducted in 12 waves and contains monthly information on over 40,000 households from May 2008 to March 2012. It contains information on participation in SNAP, the NSLP, and the SBP as well as information on individuals, families, and households. For the 2008 SIPP, questions about food security were asked during waves 6 and 9. Thus, household food security is measured at two points in time, which makes it possible to use panel data methods. The analysis is conducted at the household-wave level and, because of the focus on the school meals programs, omits households containing only elderly individuals.
For analysis of the SNAP, the sample consists of households with incomes less than 130 percent of poverty, and for analysis of the school meals programs, the sample consists of households with incomes less than 185 percent of poverty and that have school-aged children.
Linear and nonlinear panel data models with and without sampling weights were considered. Due to the similarity between the linear and nonlinear results and between the weighted and unweighted results, the analysis is conducted using linear panel data models without weights. Four models are used in the analysis: the linear probability model, random effects, fixed effects (the within estimator), and the between estimator.
The analysis began by considering the relationship between SNAP and food insecurity. Consistent with the existing literature, the linear probability (LPM) and random effects (RE) models reduce the positive relationship between SNAP and food insecurity. Additionally, the results from the fixed effects model remain positive but are much smaller and are statistically insignificant. The between model estimates show that, even though there is within variation in the data, most of the variation is between households.
The patterns are similar when examining the NSLP or SBP alone except that the fixed effects (FE) estimates are negative (but not statistically significant). The analysis considers two models of multiple program participation: both school meals programs and all three programs. Focusing on the model of the school meals programs, the LPM and RE effects are positive for the NSLP, negative for the SBP, and positive for the interaction; only the positive estimates for the NSLP are statistically significant. For the FE model, the two individual program estimates are negative and the interaction effect is positive but none of the estimates are statistically significant.
Next, a number of different specifications for SNAP participation during the past year were used, including the number of months of participation during the past year, the number of transition into or out of SNAP, and the number of entries into and exists out of SNAP. To account for other changes in household circumstances, the models include measures of the number of times household size has increased during the past year, the number of times household size has decreased during the past year, the number of times the household has moved during the past year, and an indicator for a negative income shock.
None of the results of the fixed effects model are statistically significant so results from the random effects model are discussed. The results suggest that food insecurity is positively related to the number of transitions onto or off of SNAP. More specifically, food insecurity is positively related to exits from SNAP and negatively related to entries onto SNAP. There is also evidence that food insecurity is positively related to negative income shocks and positively related to the number of moves.
The results highlight the challenges faced in understanding the effects of participation in multiple FAPs. Additional work to understand the reasons some households participate in different bundles of programs will be necessary. The roles played by past program participation and household changes suggest that underlying household instability plays a role in food insecurity beyond the household’s contemporaneous circumstances.