Bayesian Estimation of Disclosure Risks for Multiply Imputed, Synthetic Data

Many national statistical agencies, survey and research organizations, and businesses— henceforth all called agencies—collect data that they intend to share with others. These agencies strive to release data that (i) protect the confidentiality of data subjects’ identities and sensitive attributes, (ii) are informative for a wide range of analyses, and (iii) are relatively straightforward for secondary data analysts to use. Most strategies for meeting these three criteria involve altering data values, such as suppressing values, aggregating variables, swapping values across records [10], and adding random noise to data values [20]. An alternative to releasing datasets is to release perturbed results of user-specified queries [16]; we consider only dataset releases here.