We began with a univariate model of divorce in view of the limited heritability information for this measure from previous studies. Following this, we conducted a bivariate analysis of AUD and divorce, which is based on the idea of an underlying liability to AUD and divorce assessed as binary outcomes. We set up a bivariate model using the Cholesky decomposition where the first factor loads on both AUD and divorce while the second loads only on the latter. Although the method can handle missing items we excluded pairs where one twin or sibling was never married and thus missing information on divorce. Preliminary analyses indicated that the prevalence of AUD was different in these individuals owing to the fact that the missingness pattern is not completely at random. To account for the variation in age in the sample, associated with prevalence of AUD and divorce we tested whether to allow the threshold to linearly depend on birth year, which we refer to as the ‘age regression’. To account for potential quantitative sex differences, we tested whether to allow the paths
