Second Order Regression with Two Predictor Variables Centered on Mean in an Ill Conditioned Model

摘要

It has been recognized that centering can reduce collinearity among explanatory variables in a linear regression models. However, efficiency of centering as a solution to multicollinearity highly depends on correlation structure among predictive variables. In this paper, simulation study was performed in a polynomial model to examine the effect of centering at various level of collinearity. The results empirically verify that centering first dramatically reduces the collinearity whereas under severe collinearity, centering provides only a small improvement over no centering at all. Therefore application of centering as a solution to multicollinearity problem should be discouraged under severe collinearity.