Fall Seminar Series - November 11, 2004
University of Minnesota
School of Statistics
College of Liberal Arts

Multilevel models with applications in genomics

Brian Caffo
Department of Biostatistics
Johns Hopkins University

Thursday, November 11, 2004
3:30 PM, 115 Ford Hall
Minneapolis, East Bank Campus
Social at 3:00 PM, 300 Ford Hall

Abstract

In the simultaneous estimation of a large number of related quantities, multilevel models provide a formal mechanism for efficiently making use of the ensemble of the information in deriving individual estimates. In this talk we present a novel and flexible class of normal multilevel models, referred to as the ``power conjugate family''. This family overcomes some of the severe restrictions posed by standard conjugate normal models in describing the relationship between sources of variations at different levels of the model, while retaining attractive properties from the point of view of computations. We show that estimates based on this generalized family of conjugate distributions, outperform currently prevalent methods in a range of plausible simulated experiments. Our work was motivated by the analysis of data from high-throughput experiments in genomics. We illustrate the use of the power conjugate family on two such data sets, one of which gives an example where uncritical application of standard conjugate models can produce poor results.