Trajectories of impulsivity from ages 18 to 35 were classified using finite mixture models (McLachlan and Peel, 2000) rather than the default standard of growth mixture modeling (e.g., Muthén, 2004). The primary difference between these frameworks is that the former does not explicitly include time in the modeling process, resulting in a benefit of not having to specify a structure for the relationship between time and the outcome. That is, the mean profiles that are extracted from the data are equivalent to a model that includes time in the model but does not impose a shape on the curve (e.g., comparatively, if viewed from a structural equation modeling perspective, the loadings corresponding to the time construct would be “free”). To allow for the greatest flexibility for discovering the most appropriate mean profiles, nine different correlation structures (see Martinez and Martinez, 2004) for 1 to 9 class solutions (resulting in 81 potential solutions) were investigated with Matlab version 7.1.0.124 (The Mathworks, 2005). Model fit was evaluated by comparing Bayesian information criteria (BIC) among the 81 potential mixture model solutions. To allow
