A recent study by Frank and Friedman (1993) indicated that cross-validated ridge regression performed well when compared to partial least-squares regression and crossvalidated principal components regression. Thorpe and Scharf (1995) consider a number of uncross-validated ridgetype estimators from an engineering point of view. In this paper we examine a variety of estimators to see if we can do as well as or nearly as well as fully cross-validated ridge regression. We conclude that when the number of parameters does not exceed the number of observations, it may be possible to avoid cross-validation.
展开▼