Order Statistics--Based Shrinkage for Multiple Estimation Adjustment with Application in Genomewide Association Study

• Activity Number: 28
• Type: Contributed
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Jiangang Liao*+
• Companies: Penn State
• Keywords: empirical Bayes

Abstract

Statistical methods for handling multiple comparisons in large scale studies have received intensive attention in the past three decades. Methods for adjusting for multiple estimations, however, are much less well developed. This paper proposes an order based shrinkage for adjusting the potentially seriously inflated raw estimates. The proposed method builds on empirical Bayes shrinkage but differs in two important ways. First, it is designed to be the natural framework for multiple estimations. Second, it is much more robust to misspecification of the working prior distribution. it is asymptotically as efficient as standard empirical Bayes method when the prior is correctly specified. The method is illustrated by several genome-wide association studies. A ready to use R library is provided.