|
Activity Number:
|
339
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 8, 2006 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #306452 |
|
Title:
|
Robust Winsorized Regression Using Bootstrap Approach
|
|
Author(s):
|
Deo Kumar Srivastava and Jianmin Pan*+ and Ila Sarkar
|
|
Companies:
|
St. Jude Children's Research Hospital and St. Jude Children's Research Hospital and Louisiana Health Care Review, Inc.
|
|
Address:
|
332 N. Lauderdale Street, Memphis, TN, 38105,
|
|
Keywords:
|
linear regression ; winsorization ; robustness ; bootstrap
|
|
Abstract:
|
In linear regression the explanatory variables are customarily assumed fixed and one models the relationship between the response and explanatory variables. However, in practice, one collects information on several factors and models one as a function of others. The least squares (LS) estimation forms the backbone of classical regression analysis. However, this approach is highly sensitive to "outliers" in both response and explanatory variables and several methods to safeguard against them have been proposed. In this paper we propose Winsorized regression using bootstrapping for estimating and testing the parameters when the data are from symmetric populations and evaluate their asymptotic properties. The simulation results indicate that the new method provides significant improvement over LSE if the data are from non-normal models with minimal loss in power if normality holds.
|
