Studies attempting to identify main effects of measured genes are limited by low statistical power, and this is even more of a concern in studies of GxE (Kraft & Hunter, 2005). In twin models, there is significantly less power to detect moderation of genetic or common environmental influences than moderation of individual specific environmental effects, and studies are likely underpowered to detect moderation of raw genetic variance when GxE effect sizes are small (Neale, Eaves, & Kendler, 1994). Power to detect GxE in a molecular genetic study depends on a variety of factors, including genotype frequencies, frequency of exposure to the environmental factor, magnitude of the interaction effect, whether the dependent variable is categorical or continuous, and the amount of measurement error (Wong et al., 2003). An adequately powered molecular genetic GxE study with a binary outcome (e.g., AUD diagnosis) might require several thousand cases and controls. Measurement error is particularly noteworthy, as many studies of alcohol-related GxE relied on retrospective self-reports of environmental factors and drinking outcomes. Ways to reduce measurement error in GxE models include using repeated measures, longitudinal measurement, and latent variable models (McArdle & Prescott, 2010).
