B1336
Title: An extension of the Benjamini and Hochberg procedure with optimal data-driven weights
Authors: Guillermo Durand - Universite Pierre et Marie Curie (France) [presenting]
Abstract: The BH procedure is a well-known FDR-controlling procedure which power can be improved by putting weights to the $p$-values. One way of doing has previously been proposed by designing optimal weights based on the distribution of the $p$-values under the alternative. Unfortunately this distribution is rarely known, so the weights cannot be computed. In a context of grouped $p$-values sharing the same distribution, we propose data-driven weights which converge to the optimal weights when the number of tests tends to infinity. The resulting step-up procedure is also shown to asymptotically control the FDR.
Journal