Nonparametric Bayesian Models for Generating Synthetic Household Data

• Activity Number: 301
• Type: Topic Contributed
• Date/Time: Tuesday, August 6, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Jingchen Hu*+ and Jerry Reiter
• Companies: Duke University and Duke University
• Keywords: synthetic dataset ; nonparametric Bayes

Abstract

When releasing microdata to the public, data disseminators typically are required to protect the confidentiality of survey respondents' identities and attribute values. To satisfy these requirements, removing direct identifiers such as names and addresses generally is not efficient to eliminate disclosure risks, so that data must be altered before release to limit the risks of unintended disclosures. Statistical agencies can release the units originally surveyed with some values, such as sensitive values at high risk of disclosure or values of key identifiers, replaced with multiple imputations. So far such research has been done mainly focuses on individual respondents, i.e. simulating attributes of each individual respondent for protecting their privacy. In this study we developed nonparametric Bayesian models for generating household data, i.e. simulating people nested in households. Dependence structures among variables at both the individual level and the household level need to be preserved in such models.