Identifiability and Falsifiability: Two Challenges for Bayesian Model Expansion

07/26/2023

∙

by   Collin Cademartori, et al.

∙

0

∙

share

We study the identifiability of model parameters and falsifiability of model predictions under conditions of model expansion in a Bayesian setting. We present results and examples suggesting a tendency for identifiability and falsifiability to decrease in this context and for the severity of these problems to trade-off against one another. Additionally, we present two extended examples that demonstrate how these difficulties can be partially overcome by inferential methods that leverage the joint structure of the posterior distribution.