“…This is due to the fact that the quadratic term is the value on the original predictor variable squared, and the correlation between X and X 2 is typically very high, especially in narrow ranges of X. Centering procedures in which the mean value is subtracted from each observed value of the original predictor variable can be applied to quantitative predictor variables to minimize these multicollinearity problems without changing any statistical inference decisions (Aiken and West, 1991;Neter, Wasserman, and Kutner, 1989). Five of the case studies reviewed examined or recommended examining curvilinear relations (Borjas, 1983;Ferber and Green, 1982;Haberfeld, 1992;Raymond, Sesnowitz, and Williams, 1988;Weiler, 1990).…”