Additive Partially Linear Quantile Regression in Ultra-high Dimension
Ben Sherwood
School of Statistics
University of Minnesota
Engr. Room 1602
4400 University Drive, Fairfax, VA 22030
Time: 12:30 P.M. - 1:30 P.M.
Date: Thursday, Mar 6, 2014
Abstract
As high-dimensional data become common in diverse fields, tremendous efforts have recently been devoted to sparse regression problems. Most of the existing work have focused on estimating the conditional mean of the response variable. High-dimensional data can be heterogeneous, for which focusing on the mean function alone may be misleading. Also, it is often assumed that the covariates and response have a linear relationship. To accommodate heterogeneous data and non-linear relationships I will consider a partial linear quantile regression model. The non-convex SCAD penalty is used for simultaneous variable selection and estimation of the linear components. Asymptotic properties of this method are presented along with Monte Carlo simulations and an application to a microarray study using birth weight as a response.