Multiple Imputation for Disclosure Limitation

• Activity Number: 440
• Type: Invited
• Date/Time: Wednesday, August 1, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Trivellore Raghunathan*+
• Companies: University of Michigan
• Keywords: Synthetic data ; Missing data ; Combining Data ; Small area estimation ; Administrative data ; Bayesian Framework

Abstract

Demands for survey data collected, especially funded by tax payers, is increasing for the analytical purposes. At the same time administrative data sources with personal identifying information is also becoming available. An intruder linking the such data sources may be able identify a respondent and glean information collected in the survey under the pledge of privacy and confidentiality. Multiple imputation technique was proposed to handle such situations whereby some or all variables on some or all subjects be synthesized based on the observed data to allow valid statistical analysis but preserve the privacy and confidentiality. This technique could greatly expand the information that could be provided to obtain national and sub national estimates, linked survey and administrative data files and combined data from multiple surveys. This talk will focus on some methodological issues and challenges in implementing this method and strategies for obtaining efficient estimates from synthetic data minimizing concerns about identification of respondents to the survey from the released multiply imputed synthetic data sets. Parametric and nonparametric strategies will be discussed.