Latent Root Regression: An Alternate Procedure for Estimating Parameters in the Presence of Multicollinearity

Abstract

Marketing data are almost always plagued by multicollinearity. The most popular approach to addressing the problem is ridge regression which is one of the biased estimation procedures. The authors introduce to marketing a relatively new biased estimation procedure called “latent root regression.” This procedure, unlike other biased estimation procedures, not only provides stable estimates in the presence of multicollinearity, but also provides a measure for determining whether or not a biased estimation procedure is appropriate. The use of latent root regression is illustrated with previously published marketing data.