A Bayesian Procedure to Control Family-Wise Error Rate

• Activity Number: 563
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Zijiang Yang*+ and Yanping Liu
• Companies: Merck and Temple University
• Keywords: Family Wise Error Rate ; Multiple Testing ; False Discovery Rate ; Local FDR

Abstract

Multiple testing procedures are designed to control an error rate, such as the family-wise error rate (FWER) or the false discovery rate (FDR). Introduced by Benjamini and Hochberg (1995), FDR gathered most research interests in the past 20 years. Traditional frequentist methods usually rely on p-values and can control FWER or FDR by a single-step, stepwise, or fixed rejection region procedures. More recently, Bayesian methods emerged as an alternative to provide FDR control through local FDR, a term introduced by Efron (2002). So far, all procedures involving local FDR are for FDR control. This presentation provides a step-back aiming to control the FWER using the local FDR. Simulations show the procedure could outperform Hochberg procedure in certain scenarios.