To test whether CADM2 SNPs were associated with separate risk behavior outcomes, association analyses were firstly conducted in PLINK (version 1.9). 39 For dichotomous phenotypes, logistic regression was used; for continuous variables, we used linear regression. Covariates included sex, age, and highest level of education, as we aimed to capture the influence of CADM2 on risk behavior and self‐control that was independent of these factors (e.g., education has shown to be associated both with CADM2 and risk behavior). 40 Furthermore, principal components (PCs) for ancestry were included. PCs are used to control for possible stratification effects that arise when a genetic factor and a trait show a spurious correlation due to systematic differences in allele frequencies between groups of different genetic ancestry. We used the PCs as calculated by the institute we received the data from, following their recommendations on how many PCs were appropriate to control for ancestry stratification effects within their specific sample. Because S4S participants were recruited at university, parental rather than own education level was included as a covariate in this sample. In 25Up, S4S, and
