|
Activity Number:
|
379
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #303365 |
|
Title:
|
Randomized P-Values with Applications in Multiple Testing
|
|
Author(s):
|
Josh Habiger*+ and Edsel A. Pena
|
|
Companies:
|
University of South Carolina and University of South Carolina
|
|
Address:
|
Department of Statistics, Columbia, SC, 29208,
|
|
Keywords:
|
False Discovery Rate ; Two Group Model ; B-H procedure ; P-value ; Multiple Testing ; Microarray
|
|
Abstract:
|
The validity of many multiple hypothesis testing procedures for False Discovery Rate (FDR) control relies on the assumption that the p-value statistics for each pair of hypotheses are uniformly distributed (marginally) under the null hypothesis. However, this assumption is not satisfied if the test statistic has discrete distribution or if the distribution model for the observables is misspecified. This poster relates the significance level of a hypothesis test to the distribution of the p-value statistic under the null hypothesis and provides a framework for constructing a randomized p-value statistic that is uniformly distributed under the null hypotheses when test statistics have a discrete distribution. A nonparametric randomized p-value is presented and applied to some FDR controlling procedures and show through simulation to enhance the performance of some FDR procedures.
|
