Modeling the Changes in Delinquent Behavior of Adolescents Transitioning into Adulthood: Methods Comparison with Simulated Data and Pathways to Desistance Data

Various methods are available to model longitudinal data, for example, growth mixture modeling (GMM), latent class growth analysis (LCGA), k-means cluster analysis, and latent transition analysis (LTA). However, the extent to which different methods can adequately model different types of longitudinal data remains unclear. Using a set of simulated data, the current study evaluated how well the four methods perform under various simulation conditions. The extent to which the methods were able to i) accurately identify the number of latent classes, and ii) correctly assign individuals into their corresponding latent classes in the simulated data were compared with one another. Based on the simulation results, suggestions were made with respect to which method(s) were best applied to model the heterogeneity of longitudinal data. The present study further applies the methods of interest to model the changes in offending behavior as delinquent youths mature into young adults. Similarities and differences of modeling solutions from different methods are reported; recommendations are made for the exploration of inter- and intra-individual differences in longitudinal data.