Since no detectable systematic bias was observed, the patterns of missingness in this data were assumed to have occurred randomly (MAR; Schafer & Graham, 2002). As such, all analyses were conducted in Mplus version 5.0 (Muthén & Muthén, 2007), where missing data were accommodated in via robust maximum likelihood estimation, taking advantage of all available data rather than deleting cases with partially missing data. A standard set of indices was used to assess relative model fit, quality of classification, and direct comparisons between models: Bayesian information criterion (BIC; Kass & Raftery, 1995; Keribin, 1997), Akaike information criterion (AIC; Akaike, 1974), entropy coefficients (Muthén, 2004), and the Lo-Mendel-Rubin likelihood ratio test (LMR-LRT; Lo et al., 2001). Finally, given the strategies described above for examining conditional associations among Caucasian adolescents after further partitioning the sample into developmentally heterogeneous subgroups, all primary analyses were conducted irrespective of sex.
