Approximating Interval Hypothesis: p-values and Bayes Factors

Abstract

Abstract In this paper we study Bayes factors and special p-values in their capacity of approximating interval null hypotheses (or more generally neighbourhoods of point null) hypotheses by point null or eventually parametric families in the case of the goodness of fit test of a parametric family. We prove that when the number of observations is large Bayes factors for point null hypotheses can approximate Bayes factors for interval null hypotheses for extremely small intervals. We also interpret the significance of calibration goodness of fit tests using a p-value in terms of width of neighbourhoods of the point null hypothesis. Finally we study the consistency of Bayes factors for goodness of fit tests of parametric families, which enables us to shed light on the behaviour of the Bayes factors.