Significance testing is one of the main objectives of statistics. The Neyman-Pearson lemma provides a simple rule for optimally testing a single hypothesis when the null and alternative distributions are known. This result has played a major role in the development of significance testing strategies that are used in practice. Most of the work extending single testing strategies to multiple tests has focused on formulating and estimating new types of significance measures, such as the false discovery rate. These methods tend to be based on p-values that are calculated from each test individually, ignoring information from the other tests. As shrinkage estimation borrows strength across point estimates to improve their overall performance, I show here that borrowing strength across multiple significance tests can improve their performance as well. The "optimal discovery procedure" (ODP) is introduced, which shows how to maximize the number of expected true positives for each fixed number of expected false positives. The optimality achieved by this procedure is shown to be closely related to optimality in terms of the false discovery rate. The ODP motivates a new approach to testing multiple hypotheses, especially when the tests are related. As a simple example, a new simultaneous procedure for testing several Normal means is defined; this is surprisingly demonstrated to outperform the optimal single test procedure, showing that an optimal method for single tests may no longer be optimal in the multiple test setting. Connections to other concepts in statistics are discussed, including Stein's paradox, shrinkage estimation, and Bayesian classification theory.


Bioinformatics | Computational Biology | Microarrays | Statistical Methodology | Statistical Theory