A question that sometimes arises is what can and should be done with studies whose small sample size is necessitated by the prohibitive costs associated with measurement of the phenotype, for example, with behavioral or neuroimaging studies, multi-wave longitudinal studies or studies of special populations. Regardless of the nature of the variables (e.g., self-report vs. neuroimaging), a small sample size nearly always implies reduced power. In fact, Button and colleagues (K.S. Button et al., 2013) reported that the median statistical power across 49 meta-analyses of neuroscience studies was typically less than 20%, with the estimate plummeting to 8% when examining neuroimaging studies alone -- and that is without estimation of G or GxE! Some have argued that genotypic effect sizes associated with endophenotypes (such as neuroimaging outcomes) are likely to be higher because the outcome is more proximal to the biological substrate (E. J. Rose & Donohoe, 2013), yet there is no demonstrable evidence that we understand the genetic underpinnings of threat-related amygdala reactivity and habituation, any better than we do the etiology of depression and anxiety. In other words, endophenotypes may be as polygenic as self-report measures (Flint & Munafo, 2007).
