Additional considerations include controlling for potential confounders, selecting an appropriate statistical model, using multiple statistical methods to evaluate an effect, and following up significant cGxE effects with replication attempts in independent samples. With respect to controlling for potential confounders, Keller (2014) recommended statistical checks that the cGxE effect holds after including all potential Covariate × Environment and Covariate × Genotype interactions. With respect to model selection, recent simulation studies have indicated that the cross-product approach for testing cGxE for three-category single-nucleotide polymorphisms, where genotype is coded as zero, one, or two copies of a specific allele, may not be appropriate in many cases, and an alternative regression parameterization has been suggested to overcome these limitations (Aliev, Latendresse, Bacanu, Neale, & Dick, 2014).
