All regression analyses were conducted in SAS v9.2. Logistic regression was used to estimate the association between rVUS and adolescent cannabis use – standard errors were adjusted for clustered family data using the Taylor method in PROC SURVEYLOGISTIC. Conditional logistic regression was used to examine within-pair associations, including in pairs of MZ twins discordant for adolescent cannabis use (27). Regression analyses were also conducted with adjustment for covariates. For within-pair analyses, only those covariates that were significantly associated with rVUS in the full sample were included. The extent to which additive genetic (A), shared (C) and person-specific (E) environmental factors influenced the variance in and covariance between rVUS and adolescent cannabis use was estimated by using data on all available twins and fitting them to a bivariate twin model in the statistical software package Mx (28).
