Making SNAP Work: A Longitudinal Analysis of SNAP Receipt and Employment in Oregon
*Names in bold indicate Presenter
This paper investigates patterns of SNAP participant and employment in Oregon using a unique administrative data linkage. Specifically, we link SNAP administrative records to UI employment and earnings data for two SNAP ‘intake’ cohorts: one that entered SNAP in 2005 and another that entered in 2009, each of which we are then able to follow—in both the UI and SNAP administrative data—for at least five years. With this data linkage, we are able to examine how the relationship between SNAP and work may have changed from the pre- to the post-recessionary period. Moreover, by combing the administrative data with information on local labor market indicators from the Bureau of Labor Statistics (BLS), we also estimate the differences in SNAP recipients’ responsiveness to local labor market conditions across the business cycle.
Using longitudinal administrative data for measuring both SNAP participation and UI-covered employment and wages confers several advantages. One is that we obtain more accurate measures of our two key measures of interest—SNAP and wages—which are often reported with considerable error in household survey data. Another is that, unlike studies that have used administrative program records combined with geographically aggregated labor market information, the availability of linked administrative microdata on both SNAP and employment outcomes allows us to observe the decisions individuals actually make about both program participation and labor supply in response to local labor market conditions.
This study will provide both descriptive analysis of employment and SNAP participation patterns as well as multivariate analysis, employing discrete-time hazard models to estimate the effect of employment status on the likelihood of program exit for the two cohorts. For SNAP participants who begin program spells without employment, we are also able to estimate hazard models of entry into employment. Our models account for individual unobserved heterogeneity so as to better isolate duration dependence effects.