Correcting for Misreporting of Government Benefits

Recent validation studies show that survey misreporting is pervasive and biases common analyses. Addressing this problem is further complicated, because validation data are usually convenience samples and access is restricted, making them more suitable to document than to solve the problem. I first use administrative SNAP records linked to survey data to evaluate corrections for misreporting that have been applied to survey data. Second, I develop a method that combines public use data with an estimated conditional distribution from the validation data. It does not require access to the validation data, is simple to implement and applicable to a wide range of econometric models. Using the validation data, I show that this method improves upon both the survey data and the other corrections, particularly for multivariate analyses. Some survey-based corrections also yield large error reductions, which makes them attractive alternatives when validation data do not exist. Finally, I examine whether estimates can be improved based on similar validation data, to mitigate that the population of interest is rarely validated. For SNAP, I provide evidence that extrapolation using the method developed here improves over survey data and corrections without validation data. Deviations from the geographic distribution of program spending are often reduced by a factor of 5 or more. The results suggest substantial differences in program effects, such as reducing the poverty rate by almost one percentage point more, a 75 percent increase over the survey estimate.