Date of Completion

2-26-2014

Embargo Period

2-25-2014

Keywords

Missing data; Multiple imputation

Major Advisor

Dr. Ofer Harel

Associate Advisor

Dr. Dipak Dey

Associate Advisor

Dr. Jun Yan

Field of Study

Statistics

Degree

Doctor of Philosophy

Open Access

Campus Access

Abstract

This research presents a framework for generating multiple imputations when we wish to incorporate the uncertainty of prior distributions utilized in the imputation phase. Imputations are generated from multiple posterior predictive distributions based on different prior distributions. Parameter estimates under each imputation model are combined using the rules of nested multiple imputation. Through the use of simulation, we investigate the impact of prior distribution uncertainty on post-imputation inferences and show that incorporating this uncertainty improves the coverage of parameter estimates. We apply our method to several research studies with various parameters of interest where prevalence of missing data was a concern. We show that different assumptions of prior distributions can have substantial impact on inference.

Share

COinS