Title: Asymptotic Theory of a Non-Marginal Multiple Testing Procedure and Comparison With Existing Methods

Abstract: Recently, Chandra and Bhattacharya (2016) proposed a novel and general Bayesian multiple comparison method such that the decision on any hypothesis depends upon the joint posterior probability of the hypotheses on which the current hypothesis is strongly dependent. Here we investigate the asymptotic properties of their methodology, establishing in particular rates of convergence to zero of several versions of Bayesian false discovery rate and Bayesian false non-discovery rate associated with the non-marginal approach, as the sample size tends to infinity.
We also establish convergence properties of several other established multiple testing methods, each representing a class of methodologies, and show that the non-marginal method is as good as the existing ones in that the associated versions of Bayesian false non-discovery rate converge to zero at the same rate, compared to the other methods, when versions of Bayesian false discovery rate are asymptotically controlled.
Moreover, we also propose a novel method to retrospectively compare the decision configurations obtained by the non-marginal method and the other methods in small samples, by conceptually extending the available finite sample to a sufficiently large sample. In this regard, we show that the decision configuration obtained from the non-marginal method minimizes the Kullback-Leibler divergence from the true decision configuration, and hence, is optimal in this sense, compared to all existing methods. In fact, we are able to prove that the non-marginal method retrospectively yields the Bayes oracle with respect to a loss function based on the Kullback-Leibler divergence measure.