A0267
Title: Design of posterior analyses with sampling distribution segments
Authors: Luke Hagar - McGill University (Canada) [presenting]
Nathaniel Stevens - University of Waterloo (Canada)
Abstract: To design trustworthy Bayesian studies, criteria for operating characteristics of posterior analyses - such as power and the type I error rate - are often defined in clinical, industrial, and corporate settings. These operating characteristics are typically assessed by exploring entire sampling distributions of posterior probabilities via simulation. A scalable method is proposed to determine optimal sample sizes and decision criteria that maps posterior probabilities to low-dimensional conduits for the data. The method leverages this mapping and large-sample theory to explore sampling distributions of posterior probabilities in a nonuniform, targeted manner. This approach, based on exploring segments of sampling distributions, prompts consistent sample size recommendations with fewer simulation repetitions than standard methods. The posterior probabilities are repurposed, computed in that approach to efficiently investigate various sample sizes and decision criteria using contour plots.
