A0217
Title: High-dimensional partially linear additive quantile regression
Authors: Ben Sherwood - University of Kansas (United States) [presenting]
Abstract: Quantile regression is a semiparametric approach to estimating conditional quantiles that can provide a more complete picture of the conditional distribution than focusing only on the conditional mean. Modeling non-central quantiles can capture heteroscedastic relationships, while modeling the condition median is a robust alternative to least squares. These properties are particularly appealing when dealing with large and noisy data sets. We will present the partially linear additive quantile regression model and a penalized estimator that simultaneously performs estimation and variable selection. The method will be evaluated using simulations and modeling birth weight using gene expression data.
