A series of biometric models was fitted directly to the raw twin data by the method of robust weighted least squares (for the analysis of categorical phenotypes) or maximum likelihood (for the analysis of continuous phenotypes) using the Mplus program version 7 (Muthén & Muthén, 1998–2012). To account for missing data, including incomplete twin pairs, analyses were conducted using full information maximum likelihood (FIML; Enders & Bandalos, 2001). Models were iteratively fit to all phenotypes to determine the best-fitting reduced models to carry forward for the final parameter estimates. As described below, variance due to the shared family environment was negligible in univariate models (i.e. for each phenotype) and in a quadrivariate model (i.e. including all phenotypes), resulting in reduced models estimating only genetic and unique environmental contributions. Bivariate model fitting was conducted with (1) MDD and CON, (2) MDD and SR, (3) AUD and CON, and (4) AUD and SR, in order to partition the covariation between each pair of variables into genetic and environmental factors.
