Biased power regression: a new biased estimation procedure in linear regression

Abstract

In order to circumvent the effects of multicollinearity on the quality of a multiple linear regression, a new strategy of analysis is proposed. It is based on a biased estimation of the vector of coefficients. Properties of this approach of analysis are shown. Moreover, the link between this new strategy of analysis and existing strategies are discussed, particularly Ridge and Generalized Ridge regression. Illustrations on the basis of two datasets are also outlined and the outcomes are compared to those of Ridge regression.