|
Activity Number:
|
346
|
|
Type:
|
Invited
|
|
Date/Time:
|
Wednesday, August 6, 2003 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Stat. Sciences
|
| Abstract - #301271 |
|
Title:
|
A Bayesian Approach to False Discovery Rate Control
|
|
Author(s):
|
Joseph G. Ibrahim*+ and Mahlet Getachew Tadesse and Robert Gentleman
|
|
Companies:
|
University of North Carolina and Texas A&M University and University of Auckland
|
|
Address:
|
Dept. of Biostatistics, Chapel Hill, NC, 27599-0001,
|
|
Keywords:
|
Bayesian methods ; microarray data ; multiple testing ; posterior probabilities
|
|
Abstract:
|
The false discovery rate (FDR) procedure is a promising solution to the problem of multiple hypotheses testing. The control of the FDR developed by Benjamini and Hochberg (1995) involves sequential p value rejection methods based on the observed data. Recently, Storey (2001, 2002) proposed a modified version, which he calls the positive false discovery rate (pFDR). We propose a Bayesian version of the FDR where the rejection rule is based on the posterior probabilities of the null hypotheses. The definition of the FDR lends itself well to a Bayesian interpretation; it is the expected proportion of null hypotheses falsely rejected, i.e., the probability that a hypothesis comes from the null given a measure of evidence for its falsity. The Bayesian approach has the attractive feature that it allows this measure of evidence to be computed based not only on the actual data but also on prior belief and external information. Simulated examples are used to compare the performance of this approach to the non-Bayesian versions. Correspondence between Bayesian and frequentist measures of evidence in hypothesis testing have been studied in several contexts.
|
