Releasing Multiply Imputed, Synthetic Data Generated in Two Stages To Protect Confidentiality

• Activity Number: 191
• Type: Topic Contributed
• Date/Time: Monday, July 30, 2007 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Joerg Drechsler*+ and Jerome P. Reiter
• Companies: Institute for Employment Research and Duke University
• Keywords: Confidentiality ; Disclosure ; Multiple Imputation ; Synthetic Datasets

Abstract

In 1993 Rubin suggested to generate synthetic datasets from existing survey data if a release of the unaltered original data is impossible for confidentiality reasons. However, generating datasets that preserve inferences for a large number of possible estimates is a difficult and labor intensive task. Furthermore, considering the partially synthetic data framework, the release of too many different imputations for sensitive variables that vary only slightly might enable the intruder to guess the true values for these variables very closely. To reduce this risk and the burden of generating synthetic datasets, a two stage imputation process can be used. In this paper, we introduce this new approach, first developed for a German establishment panel. We derive new combining rules for the synthetic datasets necessary in this context and evaluate the new estimates with a simulation study.