New Approaches to False Discovery Control in Multiple Testing with Applications to Neuroimaging

• Activity Number: 9
• Type: Invited
• Date/Time: Sunday, August 3, 2003 : 2:00 PM to 3:50 PM
• Sponsor: IMS
• Author(s): Christopher Genovese*+ and Larry A. Wasserman
• Companies: Carnegie Mellon University and Carnegie Mellon University
• Keywords: multiple testing ; False Discovery Rates ; Functional Magnetic Resonance Imaging

Abstract

Multiple testing problems arise frequently in modern applications. In functional neuroimaging in particular, a common analysis involves performing a hypothesis test at every volume element in the brain to locate brain activity. Traditional multiple testing methods seek strong control of the familywise error rate. Benjamini & Hochberg (1995) introduced a new criterion--the False Discovery Rate--and put forward a procedure to control it. The False Discovery Rate is the expected proportion of false discoveries among the rejected null hypotheses. I will introduce new methods for controlling other features of the false discovery distribution, especially quantiles. A key to this approach is to study the proportion of false discoveries as a stochastic process in the rejection threshold. I will describe methods for constructing confidence envelopes for this unobserved process and techniques for constructing thresholds with desirable error control properties. The techniques extend in an interesting way to spatial and clustering problems on random fields.