Steps Toward Creating a Fully Synthetic Decennial Census Microdata File

• Activity Number: 304
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Government Statistics
• Author(s): Martin Klein*+ and Robert Creecy
• Companies: U.S. Census Bureau and U.S. Census Bureau
• Keywords: Categorical data ; Confidentiality ; Decennial Census ; Disclosure ; Log-linear modeling ; Synthetic data

Abstract

Summary File 1 (SF 1) released by the U.S. Census Bureau provides counts based on the 100-percent data collected from the Decennial Census. These counts represent a subset of margins, at block level geography, from the fully cross-classified contingency table on the Census variables. In some cases however, data users have an interest in the full microdata file whose disclosure is prohibited by law. At the same time, there is growing interest on synthetic data, making the release of a fully synthetic Decennial Census microdata file an attractive proposal. In this paper we examine the feasibility of such a data synthesis and challenges that arise. Our technical approach employs log-linear modeling and iterative proportional fitting to create a synthetic microdata file whose margins closely match SF 1.