A Prior-Robust Posterior Distribution for Empirical Bayes Analysis of a Large Number of Parallel Effects

• Activity Number: 236
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Jiangang Liao*+
• Keywords: large scale data

Abstract

In Liao, McMurry and Berg (2013, Biostatistics) we proposed a prior robust emprical Bayes inference for large scale data by conditioning on rank of the observed data instead of the data itself. In this talk, we introduce a new prior robust Bayes method by eliminating the influence of a mis-specified prior on part of the posterior distribution. The proposed method shares one unique chracteristic with the rank-conditioned method: it can significantly improve robustness against a mis-specified prior with little or no sacrifice of inferntial efficiency when the prior is correctly specified. The new method, however, is more generally applicable.