Speaker: Tyler Bonnett, Graduate Student, UW Biostatistics
Abstract: Previous work has shown that popular measures of evidence such as p-values are frequently misinterpreted and, even given the correct interpretation, can greatly overestimate the level of evidence provided for different hypotheses in some settings. On the other hand, there are also settings in which p-values and other measures such as Bayesian posterior probabilities can be reconciled. We show how one and two-sided p-values, Bayes factors, and the distinction between hypothesis and significance testing can be motivated by developing tests in a decision theoretic framework with carefully chosen loss functions. By developing all of these methods in a single setting, we hope to help users understand these analyses and be able to choose relevant analysis methods based on their questions of interest rather than their preferred statistical framework.