Better subset regression using the nonnegative garrote
Technometrics
| Hi-index | 0.00 |
The number of potential variables included into a regression model is often too large and a more parsimonious model may be preferable. Selection strategies are widely used, but there are few analytical results about their properties. To investigate problems as replication stability, model complexity and selection bias we use bootstrap and cross-validation methods. For stepwise strategies, we discuss the importance of the predefined selection level. The methods are illustrated by investigating prognostic factors for survival time of patients with malignant glioma in the framework of a Cox regression model.