We also encourage researchers to take a rigorous and transparent approach to statistical tests for cGxE interaction, which includes checks for robustness of the cGxE effects following nonlinear transformation of the dependent variable. Interaction effects are dependent, in part, on the scale of the outcome variable (Mather & Jinks, 1982). The scales of many of the outcomes in the behavioral sciences (e.g., internalizing and externalizing behavior) are arbitrary in the sense that they have no true zero, and the differences between scores on the scales cannot be interpreted as ratios (i.e., the magnitude of the difference between people scoring 1 and 2 points on a depressive symptom inventory may be not be the same as the magnitude of the difference between people scoring 9 and 10 points). Accordingly, checking for the robustness of cGxE effects following nonlinear transformations of scale (e.g., logarithmic or square root transformations) is important. Neither the transformed nor the untransformed version of the outcome variable is “right”; however, nonlinear transformations of scale reduce heteroscedasticity that may masquerade as cGxE and thus represent a key discriminant test.
