Pointwise Testing of Functional Data

• Activity Number: 407
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Shubing Wang*+ and Jia Cao
• Companies: Merck & Co., Inc. and Columbia University
• Keywords: Pointwise testing ; functional data ; false discovery rate ; resampling ; random field theory ; cluster-based

Abstract

In this paper, we explore the advantages and shortcomings of the three most widely used multiple testing methods: False discovery rate, resampling and random field theory. A thorough simulation study demonstrate that each single method of the three has serious drawbacks when applied to pointwise testing of functional data. False discovery rate simply ignore the temporal or spatial correlation of functional data. Resampling is computationally time-consuming, which prevent it from general applications. Random field theory is indeed conservative. We propose a cluster-based false discovery rate method using random field theory. This two-stage method uses random field to find the proper clusters of the given data set; the false discovery rate is applied within-clusters and between clusters. Simulations show this method significantly improve pointwise testing of functional data.