False Discovery Rates: Algorithms and Estimations

Abstract:

This study surveys existing research results concerning false discovery rates (FDRs), aiming to give readers a picture of the development of multiple comparison procedures (MCPs) controlling FDRs and their properties as well as how these MCPs can be used. We first discuss the motivation for FDRs and then present MCPs including the Benjamini and Hochberg’s procedure, an adaptive method from Liu and Sarkar (2011), MCPs using the q-value, and an MCP using the local FDR. The emphasis of the study is on the estimations of FDRs under independent test statistics. A brief discussion on dependent test statistics is given as well. Applications are discussed with a concentration on a labor study.