While both the discovery analysis in the DNS sample and the replication analysis in the HCP sample showed that alcohol use was significantly correlated with reduced GMV, the voxels of strongest association only partially overlap. Post hoc analyses found that the effect sizes of the association with atlas-defined ROIs were nearly identical in the 2 samples (Figure S2 in Supplement 1), further supporting the interpretation that identified GMV correlations with alcohol use are replicable. The limited overlap of peaks between the samples likely reflects the lower power in the HCP sample, which is a result of its smaller sample size and the family structure of the data, which resulted in even fewer independent observations. Several GMV findings in the DNS did not replicate, an outcome that may be attributable to differences between the samples (e.g., age), though the possibility that they are false-positives, or that null findings in the HCP sample are false-negatives, cannot be ruled out.
