Measuring Disclosure Risk for Multimethod Synthetic Data Generation

Government agencies must simultaneously maintain confidentiality of individual records and disseminate useful microdata. We propose a method to create synthetic data that combines quantile regression, hot deck imputation, and rank swapping. The result from implementation of the proposed procedure is a releasable data set containing original values for a few key variables, synthetic quantile regression predictions for several variables, and imputed and perturbed values for remaining variables. To measure the disclosure risk in the resulting synthetic data set, we extend existing probabilistic risk measures that aim to imitate an intruder attempting to match a record in the released data with information previously available on a target respondent.