and age at opportunity to use cocaine in STATA (Stata Corp, 2003) using stcox. The Huber–White variance estimator that adjusts standard errors for familial clustering was implemented. Univariate twin analyses were conducted in Mx (Neale, 2004) using raw ordinal data and full-information maximum-likelihood (FIML) estimation which makes use of incomplete twin pairs. Variance was attributed to additive genetic (A), shared environmental (C), and unique/nonshared environmental (E, also including measurement error) factors, with no evidence, based on MZ–DZ correlations for nonadditive genetic (D) influ-ences. Sex differences in prevalences or multiple response categories were modeled by allowing thresholds to differ between males and females. Even if these are the same between sexes, it does not necessarily follow that causes of variation in underlying liability will be the same in males and females and we also modeled this sex limitation by allowing the A, C, and E parameters to vary across sexes, including allowing for qualitative sex differences (Rg ≠ 0.5) in DZ opposite sex pairs.
