Abstract:
|
Survey misreporting is pervasive and biases common statistical analyses. I use validation data on SNAP to show that underreporting in survey data severely affects measures of the nature and impact of transfer programs. I develop a method to combine information from the validation data with public use data and show that it drastically improves estimates. Contrary to the validation data, the required information can be released to the public, making it possible to correct estimates without access to the validation data. The method is simple to implement and applicable to a wide range of econometric models. It performs better than survey based estimates and improves upon common corrections, particularly for bi- or multivariate relationships. Using the method to extrapolate across time and geography improves over survey data and corrections without validation data. Deviations from administrative aggregates are often reduced by a factor of 5 or more. The results suggest substantial differences in program effects, such as an additional reduction of the poverty rate by almost one percentage point, increasing the poverty reducing effect captured by the survey by 75 percent.
|