Best‐practice recommendations for estimating interaction effects using moderated multiple regression

Abstract

SummaryAn interaction effect indicates that a relationship is contingent upon the values of another (moderator) variable. Thus, interaction effects describe conditions under which relationships change in strength and/or direction. Understanding interaction effects is essential for the advancement of the organizational sciences because they highlight a theory's boundary conditions. We describe procedures for estimating and interpreting interaction effects using moderated multiple regression (MMR). We distill the technical literature for a general readership of organizational science researchers and include specific best-practice recommendations regarding actions researchers can take before and after data collection to improve the accuracy of MMR-based conclusions regarding interaction effects.

Serving Others at the Expense of Self: The Relationship between Nonprofit CEO Compensation and Performance in Trade and Professional Associations

“…Moderation was tested by creating interaction terms (Baron & Kenny, 1986), and we centered interaction terms following suggested best practices (Aguinis & Gottfredson, 2010) to minimize multicollinearity concerns and aid in the interpretation of the results (Aiken & West, 1991). We tested for the presence of multicollinearity in our analyses by examining variance inflation factors (VIF); however, no evidence of multicollinearity was found.…”

The Role of Difficult Childhood Behaviors on Intimate Partner Violence

“…Factors known to impact the marital relationship or to predict IPV; caregiver depression, substance abuse in the household, and low socioeconomic status [39][40][41] served as control variables. Using a statistical model of moderator [42][43][44], the hypothesis was supported if and only if the family stress x child difficulty interaction was statistically significant at the p<0.05 threshold.…”

“…Based upon the recommendations of Aguinis and Gottfredson [42], the four moderator procedure steps that were used in this study were:…”

“…Perform a regression analysis entering the predictors in two blocks, with the first block containing the centered moderator and centered independent variable and the second block containing the product term. For the present study, the moderator (child problems) and independent (family stress) variables were centered before computing their product to make the interaction term, consistent with step 1 and step 2 [42]. Consistent with step 3, multiple regression predictors were entered in two blocks, first with the independent and moderator variables in the first block (model), then with the interaction product term added in the second block (model 2).…”

“…Consistent with step 4, the change in R 2 from block (model) 1 to block (model) 2 was assessed for statistical signify cance, which would be interpreted as evidence of moderation. [42] term "block 1"), Table 1 shows that the combination of child gender, child problems (CBCL) and family stress (ESI) accounted for 9% of the variance in IPV (CTS) (R 2 =0.09). For model 2 ("block 2"), Table 2 shows that adding the child problems x family stress interaction to the combination of child gender, child problems (CBCL) and family stress (ESI) improved the R 2 to 0.12.…”