|
Activity Number:
|
33
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 6, 2006 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #305962 |
|
Title:
|
Estimating Prediction Error in Linear Regression by Cross-Validation
|
|
Author(s):
|
Hui Shen*+ and William J. Welch
|
|
Companies:
|
The University of British Columbia and The University of British Columbia
|
|
Address:
|
333-6356 Agricultural Road, Vancouver, BC, V6T 1Z2, Canada
|
|
Keywords:
|
prediction error ; cross validation
|
|
Abstract:
|
Cross validation is commonly used to estimate prediction error. In this paper, we study the properties of the cross validation estimator of mean squared prediction error (CVMSE) in linear regression. We compare the prediction performance for different numbers of folds, v, in cross validation and find that the bias of CVMSE decreases with v increasing. We also propose a correction method to reduce the bias. We compare our correction with that of Burman (1993) through simulated and real examples. Our correction can reduce the bias of CVMSE significantly when the number of parameters in the model is not small relative to the number of observations in the data set. Moreover, we find that the bias correction can help in model selection.
|
