We showed the limited power of our study to detect a diplotype and comorbid disorder interaction, with the power varying as a function of the prevalence rates of comorbid disorders used as predicting variables and the effect size (Figure 1b). For a comorbid disorder with 10% prevalence such as cannabis dependence, power is very limited. Even when the comorbid disorder covariate is more common (e.g., 30%, the prevalence of panic attacks in our sample), we only had 80% power to detect a large interaction effect size, such as an OR of 3.6 or higher. In other words, a much larger sample size is needed to detect an interaction of smaller effect size. In addition, we confirmed our results by computing the power to detect the reported estimates from our actual analyses with all covariates in our sample.
