Biometric models were fit to the raw data using full information maximum likelihood as implemented in the Mx software system (Neale, Boker, Xie, & Maes, 1999), an “all data” estimation procedure that corrects for potential statistical biases due to missing data. Four indices produced by Mx were used to evaluate model fit: (1) the likelihood-ratio test (-2LL); computed as the difference in -2 log-likelihood values between models tested, with 6 degrees of freedom corresponding to the 6 moderating parameters constrained to 0 in the first model, against the χ distribution; (2) the Akaike Information Criterion (Akaike, 1987); (3) the Bayesian Information Criteria (BIC; Raftery, 1995) ; and (4) Draper’s Information Criterion (DIC; Draper, 1995). AIC recognizes both accuracy of model fit and parsimony. BIC is conceptually similar to AIC but penalizes more for model complexity. DIC is similar to both BIC and AIC, but is thought to provide a better balance between parsimony and fit (Markon & Krueger, 2004). In all cases, the model with the lowest value of the fit statistics is preferred.
