Observed Population Heterogeneity, Model Selection and Group-Based Trajectory Models

• In Event: Extensions to Group Based Trajectory Modeling

Abstract

Group-based trajectory modeling is a common method used to detect population heterogeneity in longitudinal patterns of criminal justice contact, and trajectory groups have proven to yield important value in post-estimation analyses, especially as moderators of key, distal criminal justice outcomes. While scholarship has largely moved beyond debating disagreements over reification of trajectory groups, the nature of the complex estimation involved in generating group-based trajectories nonetheless presents analysts with a wide array of variable parameters to consider, including but not limited to the length of longitudinal observation, the duration of each observation period, the interaction between the duration of each observation period and the density of the outcome within each bin, and the distribution of the independent variable between periods that have differential temporal coarseness. It is possible and perhaps even likely that such choices, which go beyond the traditional model selection criteria, will materially affect the nature of the observed population heterogeneity. In this paper, we examine state-level criminal records data to investigate how the less often discussed analytic choices that comprise the construction of trajectory models contribute to our conclusions about population heterogeneity in patterns of arrests over time. We also explore implications for future analyses.

Authors

• Sarah Tahamont, University of Maryland

• Benjamin W. Pheasant, University of Maryland

• Thomas A. Loughran, Pennsylvania State University