Food Security Outside a Vacuum: Identifying Hidden Causes of Persistent Food Insecurity Among SNAP Households

• In Event: Understanding Food and Transportation Security and Policies to Address Them

Abstract

Despite the Supplemental Nutrition Assistance Program (SNAP) providing food assistance for low-income households, a significant portion of households enrolled in the program remain food insecure. While SNAP has beneficial secondary impacts on the ability to pay rent and utilities, its primary effect on reducing food insecurity is relatively low. This raises questions about which households have persistent food insecurity even after receipt of food assistance and the underlying reasons why SNAP is ineffective at increasing food security among these households.

While previous research has identified contributing factors to food insecurity among SNAP-enrolled households, these models often only examine one dimension (e.g., income) at a time and fail to capture the complex intersecting characteristics of low-income households. Current research only explains a portion of the variability in food security among SNAP recipients. These methods leave a gap in understanding which multidimensional household profiles remain persistently food insecure under varying socioeconomic conditions. 

This paper introduces a new approach to the problem by combining machine learning techniques with nationally representative panel data from the Survey of Income and Program Participation (SIPP). Specifically, we use random forest classification and logistic principal component analysis (PCA) to identify latent clusters of SNAP-enrolled households who remain food insecure after six months. These methods allow the inclusion of a large number of household characteristics and allow us to model the impact of intersecting characteristics rather than the impact of individual characteristics on outcomes. For example, measuring the differences in outcomes for Black, female-headed households as a distinct group compared to Black and female-headed households. While some studies examine some combinations of characteristics using interaction variables, this method uses machine learning to aid in identifying the most commonly occurring and distinct combinations in the data.   

Our preliminary results reveal distinct household clusters with persistently high food insecurity despite enrollment in SNAP. The clusters differ in demographic and socioeconomic challenges, which could benefit from targeted concurrent interventions across multiple social safety net programs. For example, a black, female-headed household with full-time employment and children may have greater time poverty and find SNAP's limits on the purchase of pre-made food a barrier to food security. Whereas a black, female-headed household with full-time employment and extended relatives temporarily living at home may find the combination of time poverty and additional adult dependents unenrolled in SNAP to reduce the overall effectiveness of the benefit. Our results also find that persistent food insecurity varies by region and state, which indicates that SNAP's federal policies are impacted by local variation.

The paper's findings contribute to the literature by providing a framework for using machine learning methods to identify at-risk households and target them for intervention. Persistent food insecurity among SNAP recipients is linked to intersecting challenges that food assistance cannot address in a vacuum. These results highlight the need for collaborative policymaking to improve and strengthen social safety net outcomes.

Author

• Lauren Elizabeth Davis, Princeton University
  Presenting Author