TOTAL SBlUNKAGE VERSUS PARTIAL SHRINKAGE IN MULTIPLE LINEAR REGRESSION

Abstract

The paper discusses the merits of partial shrinkage of the
'ordinary least square estimator of the coefficients of the mUltiple regression model of full rank. Theoretical comparisons of scalar and matrix-valued risks of the partially sbnmken and totally shrunken estimators are given. The strategy of partial shrinkage is applied to two data sets.