A related concern is that studies that produce negative evidence of GxE are less likely to be submitted and published, resulting in the well-known publication bias of larger effect sizes and overestimation of evidence for alcohol-related GxE (Flint & Munafό, 2008). The likelihood of spurious GxE can be reduced by: i) investigating whether a monotone transformation (i.e., taking the logarithm or square root of G or E) removes the interaction (Dempfle et al., 2008; Moffitt, Caspi & Rutter, 2005), ii) differentiating between a priori and exploratory hypotheses and accurately reporting the number of statistical tests conducted and the effect sizes of all GxE tested (including non-significant tests for all environments, genotypes, and alcohol-related outcomes) (Dempfle et al., 2008; Sullivan, 2007), iii) reporting descriptive statistics separately by genotype (Moffitt, Caspi & Rutter, 2005; Uher & McGuffin, 2008), iv) replicating studies with different samples and in other settings (Moffitt, Caspi, & Rutter, 2005), and v) including nonreplications in meta-analyses.
