Title: Bayesian hypothesis testing with highly unbalanced designs
Authors: Victor Pena - Baruch College, City University of New York (United States) [presenting]
Abstract: The behavior of Bayes factors is explored in situations where there is a categorical predictor with some levels that are hard to observe. Some prior specifications exhibit pathological behavior which is reminiscent of what has become to be known as Lindley's paradox, with the key difference that, in this instance, it is proper priors that are problematic. We identify classes of prior specifications that are well-behaved and describe potential avenues for future work.