Hot-Deck Multiple Imputation via Predictive Moment Matching

• Activity Number: 251
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Chia-Ning Wang*+ and Roderick Joseph Little
• Companies: University of Michigan and University of Michigan
• Keywords: missing data ; imputation ; Hot-Deck imputation ; predictive mean matching ; heteroscedastic

Abstract

Imputations for missing data are often draws from a predictive distribution of the missing values, estimated from the observed data. The Hot-Deck method creates the distribution from "similar" responding units. In predictive mean matching, the similarity is measured by the closeness of predicted means of the incomplete variable regressed on the observed variables. Since this approach selects the matching sets that have similar predicted means, other key features of the predictive distribution such as variances are not taken into account. This may lead to bias when data are heteroscedastic, and there are covariates related to variance but not mean. We propose a generalization of predictive mean matching, Predictive Moment Matching, where the matching set is selected based on the predicted mean and variance simultaneously, which ensures better predictive distributions and imputations.