Bayesian Data Editing for Continuous Microdata

• Activity Number: 391
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Korean International Statistical Society
• Author(s): Hang Joon Kim*+ and Jerome P. Reiter and Alan F. Karr and Quanli Wang and Lawrence H. Cox
• Companies: Duke University/NISS and Duke University and NISS and Duke University and National Institute of Statistical Sciences
• Keywords: Census of Manufactures ; Dirichlet Process ; Fellegi-Holt ; Multiple Imputation ; Nonparametric Bayes ; Survey Sampling

Abstract

We present a full Bayesian, joint modeling approach to simultaneous editing and imputation for continuous microdata under the linear constraints. Several automatic editing systems in statistical agencies are based on the Fellegi-Holt method which uses a two-stage process of (i) finding the erroneous data items by logical conditions, called edits, and (ii) imputing new values for those detected items, usually using simple imputation methods. Our approach replaces the two step process with a single probability based, data-driven approach in which we (i) specify a flexible joint probability model for the continuous variables, which can capture more complex associations, (ii) stochastically identify erroneous items unlike Fellegi-Holt routines, and (iii) impute new values from the model in ways guaranteed to satisfy all edits without deriving the complete set of edits, implied and explicit edits. In this talk, we describe this integrated edit-imputation approach with simulation study and application to the 2007 U.S. Census of Manufactures data. We compare the approach against the Fellegi-Holt approach, shown how the joint model-based approach can offer improved accuracy.