Latent growth curve models (LGCMs) were used to characterize change in girls’ propensity for alcohol use from ages 12–15 years. The models were estimated using a weighted least squares estimator in Mplus 6 (Muthén and Muthén 2010) to allow for non-normally distributed data. Missing data on dependent variables were handled using the expectation maximization (EM) algorithm. Model fit was evaluated using the χ2 goodness of fit test, comparative fit index (CFI), Tucker-Lewis index (TLI), and root-mean-square error of approximation (RMSEA). For CFI and TLI, we used the conventional cutoff ≥. 90 for acceptable fit, and ≥. 95 for good fit. RMSEA values between .05–.08 represent acceptable fit, whereas values <.05 indicated good fit (McDonald and Ho 2002).
