Comparison of Hot Deck and Multiple Imputation Methods Using Simulations for HCSDB Data
*Donsig Jang, Mathematica Policy Research, Inc.
Xiaojing Lin, Mathematica Policy Research, Inc.
Amang Sukasih, Mathematica Policy Research, Inc.
Thomas V Williams, Center for Health Care Management Studies
Keywords: Item nonresponse, within-class nearest neighbor hot-deck, sequantial regression multivariate imputation
A wide variety of imputation methods exist to compensate for item nonresponse. In this study, we compare two methods: a single imputation method--the within-class nearest neighbor hot-deck imputation, and a multiple imputation method--the sequential regression multivariate imputation (SRMI). We perform a simulation study by creating test data sets that vary the response rates and missing data mechanisms and use the two imputation methods to compensate for item nonresponse. Comparisons of the statistical properties of the imputed data sets will illustrate how to decide which method should be implemented in the Health Care Survey of DoD Beneficiaries (HCSDB.