A secondary aim of this study was to test whether the association between CADM2 and risky behavior would be mediated by one or more indices of self‐control. Assuming that latent factors would be stronger measures of underlying risky behavior propensity than the separate phenotypes (and to limit the number of analyses), we used factor scores in the mediation analyses. Assuming that CADM2 is associated with risky behavior and reduced self‐control in general rather than specific behaviors or constructs per se, such latent factors might show stronger relationships with CADM2. We used a data‐driven approach without a priori specifying the nature of the factors or the number of factors to extract. We expect clustering due to the overlap in the measures, but the actual clustering could differ per sample. We used PC analysis with principal axis factoring (PAF/PFA) including oblique (oblimin) rotation; missing values were replaced with the mean. 46 The analyses were conducted separately for each cohort and factors with an Eigenvalue >1 that explained >10% of the variance were extracted from the dataset (see Table S3). Subsequently, individual factor scores were computed using regression.
