Calibrating general posterior credible regions

Authors:

Abstract
An advantage of statistical methods that base inference on a posterior distribution is that uncertainty quantification, in the form of credible regions, is readily obtained. Except in perfectly-specified situations, however, there is no guarantee that these credible regions will be calibrated in the sense that they achieve the nominal frequentist coverage probability, even approximately. To overcome this difficulty, we propose a general strategy---applicable to Bayes, Gibbs, and variational Bayes posteriors, among others---that introduces an additional scalar tuning parameter to control the spread of the posterior distribution, and we develop an algorithm that chooses this spread parameter so that the corresponding credible region achieves the nominal coverage probability, exactly or approximately. Simulation results demonstrate that the proposed algorithm yields highly efficient credible regions in a variety of applications compared to existing methods.

This publication has 17 references indexed in Scilit: