Bayes Sensitivity with Fisher-Rao Metric

Abstract

We propose a geometric framework to assess sensitivity of Bayesian procedures to modeling assumptions based on the nonparametric Fisher-Rao metric. While the framework is general in spirit, the focus of this article is restricted to metric-based diagnosis under two settings: assessing local and global robustness in Bayesian procedures to perturbations of the likelihood and prior, and identification of influential observations. The approach is based on the square-root representation of densities which enables one to compute geodesics and geodesic distances in analytical form, facilitating the definition of naturally calibrated local and global discrepancy measures. An important feature of our approach is the definition of a geometric $\epsilon$-contamination class of sampling distributions and priors via intrinsic analysis on the space of probability density functions. We showcase the applicability of our framework on several simulated toy datasets as well as in real data settings for generalized mixed effects models, directional data and shape data.