Feature Screening for Time-Varying Coefficient Models with Ultrahigh Dimensional Longitudinal Data

• Activity Number: 455
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Wanghuan Chu*+ and Runze Li and Matthew Reimherr
• Companies: Penn State and Penn State and Penn State
• Keywords: Feature screening ; time-varying coefficient model ; ultrahigh dimensional longitudinal data

Abstract

This paper is concerned with feature screening for time-varying coefficient models with ultrahigh dimensional longitudinal data. We propose a new screening method that identifies important predictors after accounting for within-subject correlation and time-varying variance of the longitudinal response. In special cases where certain predictors are potentially responsible for the variation across time, we propose a method that is able to select both fixed and random predictors under the varying coefficient mixed model framework. We examine their finite sample performance by comparing with other existing methods via Monte Carlo simulations. In the real data example, Childhood Asthma Management Program (CAMP) datasets are analyzed, where SNPs of genes that affect children's asthma measurements are selected after accounting for baseline predictors. We advocate a two-stage approach by first reducing the ultrahigh dimensionality to a moderate size using the proposed procedure, and then applying variable selection techniques to make statistical inference on the coefficient functions and covariance structure.