Nested Dirichlet Process Model for Household Data Set Synthesis

• Activity Number: 346
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): Jingchen Hu*+ and Jerome P. Reiter
• Companies: Duke University and Duke University
• Keywords: synthetic datasets ; latent class model ; disclosure risk

Abstract

This project is focused on generating partial synthetic datasets for households, with the application for decennial census household synthesis. Extensions of nested Dirichlet Process model are developed to allow two-level clustering of households and individuals in households. Both household-level variables and individual-level variables can be modeled, and the model provides good data utility in terms of recovering the marginal, bivariate distributions of the variables in the original dataset, as well as within household structures. A data augmentation method to account for the relational structural zeros in a household dataset is developed. Risk measure computation methods based on computing the posterior probability of one record being identified given the synthetic datasets and other information available to the intruder, are developed.