Understanding Patterns of Change in Group-Based Trajectory Modeling Using Latent Transition Analysis: Valid Approximations of Development or Statistical Artifacts?

The objective of this study is to validate how well aggregate change is approximated using group-based trajectory modeling and latent transition analysis. The Pathways to Desistance dataset was analyzed. Group-based trajectory modeling was used to identify patterns of street time data. Analyses were carried out for the full dataset from start to finish and separate analyses were carried out for the early half and late half of the study period. The data was split and latent transition analysis was used to determine how well trajectory groups in the full-data approximated change observed in the early-data to late-data models. Regression was used to determine the robustness of these change effects with covariates controlled for. A five-group model was identified in the early-data that was very similar to the model identified in the full-data. An eight-group model was identified in the late-data though. The change trajectories characterized by declines in recidivism were predictive of abstaining in the late-data model in a manner consistent with the change predicted by the declines observed in the early-data model. Latent transition analysis may help validate change observed in group-based trajectory modeling. Group-based trajectory modeling may provide more accurate approximation of aggregate change than stability.