Models of family transmission were fit using Mplus 5.0. We used the MLR estimator that adjusts for the non-independence of the family-level data and the non-normal symptom count variables, and accommodates missing data. Model fit was evaluated using the mean-adjusted χ2, 30 the Bayesian information criterion (BIC; χ2 – df[ln N])31 and the root mean-square error of approximation (RMSEA)32. The mean adjusted χ2 provides an estimate of overall fit for non-normal data. The BIC balances overall fit with parsimony such that model fit is penalized for the inclusion of unnecessary parameters. Negative BIC values indicate a good fit with lower (more negative) values indicative of better fit. The RMSEA provides an estimate of discrepancy in model fit per degree of freedom; values less than 0.080 and 0.050 indicate an adequate fit and very good fit, respectively. Nested models can be compared by conducting a likelihood ratio test calculated as the difference in the χ2 values and degrees of freedom between the two models. Such tests, however, are overpowered for large samples, especially with non-normal data. Therefore, we used differences in
