Abstract
Several authors have proposed centering or transformation procedures to alleviate problems associated with the extreme multicollinearity often found in interactive models. Some of the claims made for these procedures are valid; others are not. Specifically, one of the transformations is very useful for improving coefficient stability. Conversely, arguments that centering will improve substantive interpretation or provide an orthogonal partition of explained variance for the original uncentered terms are not valid.

This publication has 6 references indexed in Scilit: