GCTA indicated that the measured variants explained a low proportion of variance of the alcohol factor score (h2=0.052, SE=0.093, p=0.3). We next tested whether genetic risk variants identified in one half of the sample predicted alcohol problem factor scores in the other half of the sample. We applied a series of p-value cutoffs, as described by Purcell and colleagues [44]. Polygenic risk scores derived from Plink accounted for a very small proportion of the variance – up to r2=0.6% (p’s>0.05) – with markers meeting a p<0.10 threshold providing the highest estimate. This estimate did not differ significantly from 0, and varied widely depending on which subsample was used for discovery versus replication.
