Bayesian methods for statistical disclosure control in microdata.

Abstract

The fundamental tension in statistical disclosure control (SDC) of microdata is the trade-off between the protection of individual respondents and the release of enough information for statistical inferences. We consider microdata that include key variables that contain identifying information and target variables that include sensitive information. Most of the current SDC techniques release a single data set modified from the original to the public and result in biased statistical inferences in the modified data. I propose two model-based Bayesian SDC methods for disclosure control in microdata, namely, selective multiple imputation of key variables (SMIKe) and multiple stochastic swapping of keys (MASSK). Both techniques release multiple independently modified data sets. The multiplicity of released data allows the incorporation of modification uncertainty into statistical inferences; disclosure risk in released data sets can be controlled to low levels; information loss is limited by the fact that the modification is restricted to the key variables for only a fraction of the total cases. Simulation studies and real data applications are used to evaluate these SDC techniques with respect to disclosure risk, information loss and quality of statistical inferences.