As described above, a supraordinate substance use lifestyle factor, built from the eight indicators representing adolescent alcohol use and friends’ support of substance use, was used due to high level of multicollinearity between adolescent alcohol use and friends’ support of substance use (r = 0.89). This alternative model (Fig. 2) had better (i.e., lower) Bayesian (BIC) information criteria (BIC = 18159.05) than did the first model (BIC = 18160.28). Except for a significant χ2 value, fit indices for the alternative model were adequate: χ2(81) = 308.01, p<0.001, CFI = 0.91, RMSEA = 0.05 (90% CI = 0.05 to 0.06), SRMR = 0.05. In this model, all indicators were significantly related to the latent variable to which they were assigned at p<0.001. Substance use lifestyle in adolescence was a strong, significant predictor of problematic alcohol use in early adulthood (p<0.001), but effortful control was not a significant predictor.
