Title
On the Control of False Discovery Rate in Multiple Testing with no Assumption on Dependency
Speaker
MB Rao, University of Cincinnati
Abstract
Multiple hypotheses arise naturally in micro-array data and epidemiology. There are essentially two components in a multiple testing problem. 1) What constitutes as Type I error rate in the context of multiple hypotheses testing? 2) Choice of a multiple testing procedure controlling the error rate of your choice. A rudimentary introduction will be provided covering both the components. Our choice of error rate is the False Discovery Rate. Most multiple testing procedures available in the literature controlling the False Discovery Rate assume either independence or some specific type of dependence among the underlying statistics. We propose a new sequential step-down procedure, which controls the false discovery rate at the desired level no matter what the joint distribution of the underlying statistics is. We use the optimization technique in knapsack problems to demonstrate that the new procedure does the job it is claimed to do. If time permits, I will spend some time on Closed Testing Principle.
Meet the speaker in Room 212 Cockins Hall at 4:30 p.m. Refreshments will be served.
