The analyses were conducted first for sexes combined from the published data of 14 studies and then for men and women separately for each of the 10 studies with individual-level data. A random-effects meta-analysis was used to obtain weighted averages across studies. There was minimal change in the coefficients or their standard errors when ethnicity and age were included in the models, and we conservatively included only 1 random member of each twin pair from twin study data. The statistical models for the formal meta-analysis included the actual number of S alleles to approximate the dosage model applied by Caspi et al,10 although the findings were comparable for a recessive model (SS vs SL plus LL) and a dominant model (SS plus SL vs LL).
