THE PROBLEM OF MULTIPLE INFERENCE IN STUDIES DESIGNED TO GENERATE HYPOTHESES

摘要: Epidemiologic research often involves the simultaneous assessment of associations between many risk factors and several disease outcomes. In such situations, designed to generate hypotheses, multiple univariate hypothesis-testing is not an appropriate basis for inference. The number true positive in a collection can be estimated by comparing observed distribution p values theoretical uniform distribution, or negative associations, empiric randomization distribution. None these approaches, however, will distinguish from false associations. Various criteria selecting subset report are considered authors, including Bonferoni adjustment values, splitting sample searching testing, Bayesian inference, decision theory. authors prefer approach which all data reported, whether significant not, followed ranking order priority investigation using empirical Bayes techniques. Methods illustrated application preliminary study aimed at identifying hitherto unsuspected occupational carcinogens.