|
Activity Number:
|
260
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 5, 2008 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #301954 |
|
Title:
|
An Asymptotic Analysis of the Stepwise Correlation Pursuit Variable Selection Method
|
|
Author(s):
|
Tingting Zhang*+ and Wenxuan Zhong and Jun S. Liu
|
|
Companies:
|
Harvard University and University of Illinois at Urbana-Champaign and Harvard University
|
|
Address:
|
1 Oxford Street 608, Cambridge, MA, 02138,
|
|
Keywords:
|
variable selection ; stepwise regression ; single index model
|
|
Abstract:
|
Classic nonparametric methods for regression analysis break down quickly as the dimension of predictors increase. As an alternative approach, a stepwise correlation pursuit variable selection procedure for single index model has been proposed recently by Zhong et. al. This article aims to study the asymptotic behavior of this stepwise variable selection procedure. More specifically, we have analyzed the convergence rate of the test statistics under the null hypothesis of no effect for selected predictors and the power of each testing step as sample size goes to infinite. We have also compared the new variable selection procedure with the classic stepwise forward addition backward deletion procedure for linear regression models. The possibility of generalizing the method to multiple index model is also discussed in the paper.
|
