Nonstochastic Techniques for Selecting Ridge Parameter Values.
Abstract
Biased regression estimators are increasingly being utilized as alternatives to least square parameter estimators in multiple linear regression when the predictor variables are multicollinear. One popular biased estimator is the ridge regression estimator. Ridge estimators are known to have smaller mean squared errors than least squares for suitably small nonstochastic choices of the ridge parameter. To date, however, most of the practical applications of ridge regression employ stochastic techniques to select the ridge parameter. In this paper we examine three non stochastic procedures for choosing ridge parameters and compare their performance with another stochastic method. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1978
- Accession Number
- ADA064838
Entities
People
- Richard F. Gunst
- Tsushung A. Hua
Organizations
- Southern Methodist University