Rashomon-Seeded Annealing for Robust Bayesian Inference in Factorial Designs

arXiv:2606. 02589v1 Announce Type: cross Abstract: Integrating over model uncertainty in factorial designs via Bayesian model averaging is hindered by the combinatorial explosion of interpretable interaction effects, often yielding a multimodal posterior, where standard Markov chain Monte Carlo algorithms encounter significant convergence issues.