Association for Public Policy Analysis & Management

A host of authors have documented high and growing rates of under-reporting of receipt of public benefits in large US survey data sets. However, little research has assessed whether this matters for the types of applications for which these data are used. In this paper, we will test whether estimates of the effect of various factors on use of public programs vary importantly if estimated with survey data or with administrative data in a standard state panel data setting. We will test this using both survey self-reports of use of safety net programs in the Annual Social and Economic Supplement of the Current Population Survey and using either administrative counts of participants at the state level, or, where possible with administrative micro data. We will compare estimates of the responsiveness of these participation rates to measures of the strength of the local labor market and to other policy changes across these data sets, while controlling for time invariant and state invariant unobservables.