A Bayesian Analysis For The Homogeneity Testing Problem Using Epsilon-Contaminated Priors

Abstract

In this article the problem of testing if r populations have the same distribution from a Bayesian perspective is studied using r × s contingency tables and ϵ–contaminated priors. A procedure to build a mixed prior distribution is introduced and a justification for this construction based on a measure of discrepancy is given. A lower bound for the posterior probabilities of the homogeneity null hypothesis, when the prior is in the class of ϵ–contaminated distributions, is calculated and compared numerically with the usual p-value. Examples show that the discrepancy between both is more acute when the mass assigned to the null in the mixed prior distribution is 0.5