|
Activity Number:
|
110
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 9, 2004 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #301559 |
|
Title:
|
Improved Control and Estimation of False Discovery Rate When Substantial Dependence among the Tests Exists
|
|
Author(s):
|
Cheng Cheng*+ and Stanley B. Pounds
|
|
Companies:
|
St. Jude Children's Research Hospital and St. Jude Children's Research Hospital
|
|
Address:
|
Dept. of Biostatistics, Memphis, TN, 38015-2794,
|
|
Keywords:
|
false discovery rate ; multiple comparison ; dependent tests ; smoothing ; microarray ; data-mining
|
|
Abstract:
|
The fast growth of data collection capabilities in science has stimulated statistical research in analysis of large datasets and massive multiple tests. It seems to be a consensus now that the false discovery rate (FDR) is an error assessment quantity more preferred than the familywise Type I error rate. Following the pioneering work of Benjamini and Hochberg, there has been much theoretical and methodological development for FDR control and estimation, for the applications in which the tests are "weakly dependent." In certain applications however, more substantial dependency may exists among the tests. We compare the control and estimation performance of several procedures, including the adaptive FDR control and the q value methods, when substantial dependency exists among the tests. We demonstrate that an accurate but slightly up-biased estimate of the proportion of the null hypotheses is important and propose a smoothing-based method to improve the existing procedures.
|
