The Multicollinearity Illusion in Moderated Regression Analysis

Leaders and followers behaving badly: A meta‐analytic examination of curvilinear relationships between destructive leadership and followers’ workplace behaviors

“…Although we attempted to address multicollinearity by transforming the obtained correlations (see Appendix ), the correlation between the linear and curvilinear destructive leadership terms was still high across all of the overall analyses (i.e., .63 ≤ ρ DL & DL 2 ≤ .84). Thus, inferences from our findings should account for multicollinearity concerns and future research should explicitly address multicollinearity concerns that stem from examining linear and curvilinear terms simultaneously (Disatnik & Sivan, ; Echambadi & Hess, ; Shieh, ).…”

“…Prior to conducting the analyses, we used the means and standard deviations of the linear and curvilinear terms to transform the correlations we received from primary study authors to reflect the values we would have obtained if the linear terms were mean‐centered prior to estimating the curvilinear terms (see Appendix ). It was important to transform the correlations in our study due to multicollinearity concerns that arose because the linear and curvilinear terms were highly correlated, so we followed prior researchers’ recommendations (Disatnik & Sivan, ; Echambadi & Hess, ; Shieh, ) for examining curvilinear effects in a manner that facilitates the interpretation of curvilinear effects.…”

Strategic Talent Management in Emerging Markets and Its Impact on Employee Retention: Evidence from Brazilian MNCs

“…A generalization should thus only be undertaken with care. Furthermore, a bigger sample size may reduce the multicollinearity concerns in future studies (Disatnik & Sivan, ). Also, this study took only the employee perspective and did not consider HR employees’ or managers’ perspectives in general.…”

“…In order to test for multicollinearity, variance inflation factors (VIF), which measure the impact of collinearity among the variables in a regression model, were computed. Due to the fact that the product of predictor variables (moderator variable × independent variable) was added to an additive regression model (Disatnik & Sivan, ), the VIF for model 2 are above the acceptable level of five (Cohen, Cohen, West, & Aiken, ). It is argued that in “moderated multiple regression a high correlation between the product term and the independent variables does not imply a multicollinearity problem” (Disatnik & Sivan, , p. 3) because “the multicollinearity that is often obtained in moderated multiple regression is simply a matter of interval scaling and therefore does not create a multicollinearity problem and is irrelevant to estimating and testing the interaction” (Disatnik & Sivan, , p. 4).…”

Mean centering, multicollinearity, and moderators in multiple regression: The reconciliation redux

“…Doing so tends to reduce the correlations r(A,A × B) and r(B,A × B). We showed that the raw regression coefficient for the A × B term will not be affected (also see Disatnik and Sivan, 2016, on this point), 1 and this term may well be of primary focus for many researchers, yet other researchers may also care about the status of the main effects for A and B as well, and these regression coefficients will be clarified. In addition, the mean centering transformation will leave the overall model fit R 2 undisturbed.…”