Growth mixture modeling was conducted in Mplus 5.21 (Muthen and Muthen, 1998–2007), using the five internalizing sum scores described above as manifest variables, and estimating an intercept (I), slope (S), and non-linear (quadratic, Q) growth term for each class. Growth mixture models were conducted with the sexes combined, as previous studies have found that sex differences in internalizing typically do not emerge until after the onset of puberty (e.g., Angold et al., 1998; Ge et al., 1994). However, in an effort to account for potential sex differences, sex was used as a covariate in subsequent analyses. As the goal of the mixture modeling was to characterize early internalizing problems, the last age for which data were included is 11.5 years, at which point the majority of the sample was prepubescent (Joinson et al., 2011). Fit and feasibility were assessed using Akaike’s Information Criterion (AIC), Bayesian Information Criterion (BIC), entropy, the Lo-Mendell-Rubin adjusted likelihood ratio test, and general interpretability (see Results). Due to difficulties with model convergence in conditional models, trajectory identification and subsequent analyses were conducted separately, as opposed to in a single stage analysis, within Mplus.
