Shrinkage in adaptive procedures for multiple testing: structure and synthesis

Debashis Ghosh, Penn State University

Abstract

There has been much interest in the study of adaptive estimation procedures for controlling the false discovery rate (FDR). In this article, we take the q-value approach of Storey (2002) and show how it can reexpressed as a particular type of shrinkage estimator. This representation leads to natural conditions on finite-sample FDR control for a general class of shrinkage estimators. In addition, many previous proposals from the literature can be unified under this framework for which finite-sample FDR results can be developed. Some asymptotic results are also provided.