Partial identification methods for evaluating food assistance programs: A case study of the causal impact of snap on food insecurity

Craig Gundersen, Brent Kreider, John V. Pepper

Research output: Contribution to journalArticlepeer-review

Abstract

We illustrate how partial identification methods can be used to provide credible inferences on the causal impacts of food assistance programs, focusing on the impact that the Supplemental Nutrition Assistance Program (SNAP, formerly known as the Food Stamp Program) has on food insecurity among households with children. Recent research applies these methods to address two key issues confounding identification: missing counterfactuals and nonrandomly misclassified treatment status. In this paper, we illustrate and extend the recent literature by using data from the Survey of Income and Program Participation (SIPP) to study the robustness of prior conclusions. The SIPP confers important advantages: the detailed information about income and eligibility allows us to apply a modified discontinuity design to sharpen inferences, and the panel nature allows us to reduce uncertainty about true participation status. We find that SNAP reduces the prevalence of food insecurity in households with children by at least six percentage points.

Original languageEnglish (US)
Pages (from-to)875-893
Number of pages19
JournalAmerican Journal of Agricultural Economics
Volume99
Issue number4
DOIs
StatePublished - Jan 1 2017

Keywords

  • Classification error
  • Food insecurity
  • Food stamp program
  • Nonparametric bounds
  • Partial identification
  • Supplemental nutrition assistance program
  • Survey of income and program participation
  • Treatment effects

ASJC Scopus subject areas

  • Agricultural and Biological Sciences (miscellaneous)
  • Economics and Econometrics

Fingerprint Dive into the research topics of 'Partial identification methods for evaluating food assistance programs: A case study of the causal impact of snap on food insecurity'. Together they form a unique fingerprint.

Cite this