especially when some combinations of the independent variable have small sample sizes and when control covariates are included in the model. Because apparent cGxE effects can be due to stratification, genotyping artifacts, and even gender (Keller, 2014), careful attention must be paid to the variables under study in order to bolster confidence that the moderation effect is due to the specific gene/environment under study. For example, if whites are more sensitive to environmental trauma, leading to Posttraumatic Stress Disorder (PTSD), in a mixed ethnicity sample examining environmental trauma-by-gene interactions, any SNP that differentiates whites from blacks will also show an apparent cGxE interaction. In this case, the gene isn't the true moderator - race is - the gene was merely correlated with race. To increase confidence in cGxE findings and to eliminate alternative explanations for them, researchers need to include all relevant gene-by-covariate and environment-by-covariate interaction terms in their models. Researchers should consider including quadratic transformations of the environmental term as well as covariates, as these may change the interpretation of the interaction depending on multicollinearity. Finally, the assumptions that are imposed by modeling interactions using cross-product terms in linear regression and/or by the use of logistic regression (as discussed
