SNAP enrollment cycles: New insights from heterogeneous panel models with cross-sectional dependence

Abstract

The Supplemental Nutrition Assistance Program (SNAP) has grown rapidly over the past 2 decades. A large literature relies on state-level panel data on SNAP enrollment and implements traditional two-way fixed effects estimators to identify the impact of economic conditions on SNAP enrollment. This empirical strategy implicitly assumes slope parameter homogeneity and ignores the possibility of cross-sectional dependence in the regression error terms. The latter could feasibly arise in state-level panel data if the time-varying unobserved common shocks, such as national financial crises, have differential effects on SNAP participation across states in the United States. This study empirically evaluates the appropriateness of these two assumptions by adopting a more general common factor model, allowing for slope parameter heterogeneity and error term cross-sectional dependence both separately and jointly. We find that although assuming a common slope parameter across states does not seem problematic for identification, allowing for the error term cross-sectional dependence leads to a roughly 40% reduction in the estimated long-run impact of the unemployment rate on SNAP enrollment. This finding has important implications for policymaking decisions—even small biases could lead to suboptimal policy responses considering the program's size. Our counterfactual simulations support our main results, implying the importance of carefully accounting for time-varying unobserved heterogeneity when studying the cyclicality of SNAP enrollment using state-level panel data.