having 0–1 AUD symptoms, or 0–2 in the case of more narrowly defined AUD, set to missing because, by definition, only those with AUD can remit from them. Under this two-stage model, twin pairs discordant for alcohol problems still contribute information about genetic influences on AUD and about shared genetic and environmental variance between AUD and remission (Heath et al. 2002). A test of bivariate normality confirmed that a single dimension of liability underlay the AUD variable, so that a model using full information maximum likelihood estimation could be used. A bivariate Cholesky model was then fit to the data using the MX statistical package (Neale et al. 2003). The Cholesky model estimates sources of variance unique to AUD and remission in addition to variance common to the phenotypes. Information from opposite-sex twin pairs was estimated by allowing separate thresholds for males and females within opposite-sex pairs. First, a saturated model was fitted that estimated parameters separately for males and females; subsequent models were compared to this model using change in log likelihood relative to change in degrees of freedom and Akaike's Information Criterion (AIC). A model that constrained male and female parameters to be equal did not provide a
