|
Activity Number:
|
19
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Sunday, August 3, 2008 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
| Abstract - #300619 |
|
Title:
|
On Validating a Stepwise Variable Selection Algorithm with Another Stepwise Variable Selection Algorithm
|
|
Author(s):
|
Ryan E. Wiegand*+
|
|
Companies:
|
Centers for Disease Control and Prevention
|
|
Address:
|
Mailstop E48, Atlanta, GA, 30329,
|
|
Keywords:
|
Variable selection ; stepwise ; validation
|
|
Abstract:
|
The backward elimination, forward selection, and stepwise variable selection algorithms have been criticized in statistical studies, yet their use is still common in medical, social science and environmental literature. One such application of these methods is to use a stepwise algorithm to build a multivariable model and employ a different stepwise procedure to validate the model. This research attempts to characterize the circumstances for which this process is an advisable variable selection strategy. Data for linear and non-linear regression models were simulated varying the number and type of variables and correlation between those variables. The final models of the three stepwise variable selection algorithms were compared for each simulation and the proportions of agreement and agreement on the correct model will be presented.
|
