摘要：In multiple testing, standard procedures for false discovery rate control are mostly derived assuming test statistics follow continuous distributions. For discrete tests, such procedures were found to be often conservative. As the uniformly most powerful tests for discrete problems are randomized tests, multiple testing based on the randomized tests have shown improved performance. One downside of this idea is that the decision outcomes are random and hence inconvenient to be adopted by practitioners. We propose a method using the marginal critical function and avoids the randomness in the decision.