Further Improvements in Local FDR Based Grouped Hypotheses Testing

• Activity Number: 242 - Multiple Testing and Feature Selection
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 2:00 PM to 3:50 PM
• Sponsor: IMS
• Author(s): Shinjini Nandi* and SANAT SARKAR
• Companies: Temple University and Temple University
• Keywords: Grouped hypotheses testing; Local False Discovery Rates; Multiple Hypotheses Testing; Two way classified hypotheses

Abstract

In this article, we further explore application of multiple testing methods to grouped hypotheses. Hypotheses classified into several groups due to a single factor have been widely studied and testing methods have been adapted to control false discoveries in such scenarios. However, in many applications, hypotheses may be classified into groups created according to more than one criterion. We continue along the line of research in Liu et al 2016 and Sarkar et al 2017 and suggest improvements in their model to account for hypotheses that can be distributed to groups simultaneously defined on the basis of two different criteria. The significance of an individual hypothesis can then be explained in terms of each of the two grouping effects and its own intrinsic properties. We suggest Local FDR based testing methods to optimally control the total number of false discoveries and also the number of falsely discovered groups according to each of the specified criteria.