Given the complexity of the model, the large number of possible simplifications, our moderate sample size and evidence that full models can best capture subtle genetic and environmental effects (Sullivan & Eaves, 2002), we tested only two models. The first assumed no quantitative or qualitative sex effects by constraining path coefficients across males and females and constraining rg to 1.0 at all waves. The second model included both quantitative and qualitative sex effects. The Bayesian information criterion (BIC), which performs well with complex models (Markon & Krueger, 2004), was used to determine the best-fit model. Analyses were conducted using an ordinal, raw data approach in the statistical package Mx (Neale, 1997), which allows data from both incomplete and complete twin pairs to be used. Thresholds were unconstrained across sex and allowed to be freely estimated.
