Top-Coding and Public Use Microdata Samples from the U.S. Census Bureau

Abstract

Public Use Microdata Samples (PUMS) released by the U.S. Census Bureau and other data providers undergo various privacy protection transformations prior to public release of the individual records. We briefly review these methods but focus our attention on "top-coding" as implemented by the Census Bureau. In particular, we provide a brief analysis of the method used for top-coding of records within a hierarchy. We also show that top-coding artificially moves the correlation between two variables (at least one of which is top-coded) closer to zero by the transformation. We then discuss our attempts to recover the un-transformed data, or at least the original correlations, which all failed. In the final section we briefly discuss methods of disclosure avoidance in PUMS files which preserve joint probability distributions.