Using Linked Data to Better Measure Poverty and to Evaluate Survey Accuracy

Abstract

This thesis demonstrates some ways in which researchers and policy makers can make use of this wealth of (often unstructured) information. Building on the recent success and availability of linked data, this thesis extends the methodological studies to make use of such data from my dissertation. It takes a more applied perspective by using linked data to demonstrate the extent and nature of survey error, to document its consequences and to propose simple ways to use data combination to monitor the problem of survey error. In the first chapter, we use linked data to analyze the extent and nature of misreporting in “Errors in Survey Reporting and Imputation and their Effects on Estimates of Food Stamp Program Participation” (forthcoming in the Journal of Human Resources), which is joint work with Bruce Meyer and Robert Goerge. The second chapter, “Using Linked Survey and Administrative Data to Better Measure Income” with Bruce Meyer, examines how these data errors affect common analyses in practice. It was published in the American Economic Journal: Economic Policy. In the third chapter, “An Empirical Total Survey Error Decomposition Using Data Combination” with Bruce Meyer (forthcoming in the Journal of Econometrics), we show how to use data combination to measure, monitor and thereby ultimately improve survey accuracy by estimating total survey error and decomposing it according to its source (e.g. coverage error, item non-response and measurement error). All three chapters are based on data sets we created by linking administrative records to survey data. Creating such validation data is necessary for the analyses we conduct, as it provides a unique way to validate survey responses. Obtaining, linking and working with these data sources requires the joint effort of policy makers (who supply the administrative data), statistical agencies (who supply the survey data and the tools to link them) and researchers who work with the data. The data are subject to strong confidentiality requirements and can therefore only be accessed from secure locations such as the U.S. Census Bureau and its Research Data Centers by researchers with special security clearance. In consequence, conducting studies such as the ones in this habilitation thesis require a team of researchers. Therefore, like most other studies that rely on linked data, all three chapters are joint work with co-authors, who have been essential to successfully create the data and conduct these studies.