DISTINGUISHING BETWEEN MODERATOR AND QUADRATIC EFFECTS IN MULTIPLE-REGRESSION

Moderated regression analysis is commonly used to test for multiplicat ive influences of independent variables in regression models. D. Lubin ski and L. G. Humphreys (1990) have shown that significant moderator e ffects can exist even when stronger quadratic effects are present. The y recommend comparing effect sizes associated with both effect types a nd selecting the model that yields the strongest effect. The authors s how that this procedure of comparing effect sizes is biased in favor o f the moderated model when multicollinearity is high because of the di fferential reliability of the quadratic and multiplicative terms in th e regression models. Fortunately, levels of multicollinearity under wh ich this bias is most problematic may be outside the range encountered in many empirical studies. The authors discuss causes and implication s of this phenomenon as well as alternative procedures for evaluating structural relationships among variables.