Project:
Reporting Bias in Studies of the Food Stamp Program

Year: 2007

Research Center: The Harris School of Public Policy Studies, University of Chicago

Investigator: Meyer, Bruce D., and James X. Sullivan

Institution: University of Notre Dame

Project Contact:
James X. Sullivan
University of Notre Dame
Department of Economics and Econometrics
447 Flanner Hall
Notre Dame, IN 46556
Phone: 574-631-7587
E-mail: sullivan.197@nd.edu

Comparing s food stamp receipt as reported in surveys with administrative data on receipt indicates that receipt is underreported substantially in surveys. For example, more than 40 percent of months of food stamp receipt were not reported in the Current Population Survey (CPS) in 2004. This underreporting is evident in several large national surveys and, in some of these surveys, the extent of underreporting has grown over time. An important consequence of underreporting is that it may lead to significant bias in studies that examine the determinants of participation in the Food Stamp Program (FSP) or the distributional consequences of the program. Underreporting could also bias regressions with food stamp receipt as an explanatory variable, and instrumental variable methods may not be valid if the measurement errors are correlated with common explanatory variables, as is suggested by this study.

This study presents a new econometric method for estimating the determinants of reporting that uses two data sources with information on the same demographic characteristics but with samples of different individuals rather than matched individuals. This method compares the characteristics of those who report receipt in a survey with the characteristics of recipients in administrative data to determine the influence of those characteristics on reporting. To implement this procedure for the FSP, the study uses administrative microdata from the Food Stamp Program Quality Control (FSPQC) Database and survey data from the CPS.

Results from the two-sample estimation procedure used in this study indicate that observable characteristics, including education, gender, and region, are significantly related to underreporting. The study then demonstrates how these two-sample estimates can be used to adjust for reporting bias in studies of the FSP. A number of studies have examined how participation in the FSP is related to observable characteristics. Underreporting that varies by characteristic will bias estimates in such studies. The two-sample estimates described above can be used to correct for underreporting bias. Using estimates from a simple participation model, this study shows that the bias can be substantial. For example, adjusted estimates for the relationship between education and participation are about 30 percent larger than the unadjusted estimates.

Underreporting will also bias studies of the distributional consequences of the FSP. Studies that examine the extent to which food stamps increase the resources of poor families will understate the impact of the FSP due to underreporting of food stamps. This study shows how to correct for underreporting bias in such studies by using estimates from the two-sample procedure.

A better understanding of underreporting and how it may bias various studies of food stamps has important implications for both policymakers and researchers. Policymakers have long been concerned with low participation rates in the FSP and have recently taken steps to increase participation. In addition, a more accurate estimate of program take up provides better information about who is benefiting from the FSP, why families choose not to participate in the program, or how individual characteristics affect participation. Such information could be used to increase take up and better target the program. In addition, correcting for underreporting bias will yield better measures of the well-being of the disadvantaged and provide a clearer picture of the distributional consequences of the FSP. Lastly, the methods presented in this study could be used to analyze underreporting for other transfer programs that collect administrative microdata.

Direct inquiries about this study to the Project Contact listed above.