Time-Varying Signal Detection for Correlated Data

• Activity Number: 105
• Type: Invited
• Date/Time: Monday, July 30, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Lan Xue and Annie Qu*+ and Colin Wu
• Companies: Oregon State University and University of Illinois at Urbana-Champaign and National Heart, Lung, and Blood Institute
• Keywords: varying-coefficient model ; high-dimensional data ; longitudinal data ; NGHS data ; model selection ; nonparametric

Abstract

This talk is motivated by the National Heart, Lung and Blood Institute Growth and Health Study (NGHS) which evaluates the longitudinal effects of race, height and body-mass index on the levels of systolic blood pressure and diastolic blood pressure. Here the signals associated with relevant predictors could be time-dependent. For NGHS data, it has been observed that children's and adolescents' heights seem to be associated with high blood-pressure during certain age periods. Therefore it is also scientifically important to detect relevant covariate effects which could be time-dependent. We propose a varying-coefficient model selection and estimation which can capture relevant time- dependent covariates. Because of nonparametric components are involved for the varying-coefficient model, the parameters involved for nonparametric functions could be very high-dimensional due to local information identification. We will investigate model selection consistency and the asymptotic property for varying-coefficient estimators when the dimension of covariates exceeds the sample size, where the data are correlated and non-normal. The proposed method will be illustrated using the NGHS data.