This project is focused on Bayesian model inference, such as model selection or evaluation, under misspecification. Statistical modeling is often performed conditional on a particular model being true, even when there is a multiplicity of potential models that could have been used and which would lead to different conclusions. The Bayesian solution to this problem has historically been either to select one of the potential models based on some criteria (such as Bayes factors) or to combine the potential models using techniques like Bayesian model averaging. This project aims to expand on the work in "When are Bayesian model probabilities overconfident?" (https://arxiv.org/abs/2003.04026).