Abstract:
|
Sociologists, criminologists, and behavioral scientists have a long-standing interest in using statistical models for analysis of longitudinal individual trajectories. Analytic approaches of this type include group-based trajectory models, growth mixture models, latent class transition models, and grade of membership models, and unimodal curve registration models, among others. In the last several years, an increasing number of simulations studies have been conducted attempting to study merits and flaws of various procedures for the analysis of longitudinal individual trajectories. We critically re-examine the logic and limitations of using simulations to shed light on alternate trajectory models. We emphasize the importance of considering the crucial assumptions that distinguish models when applying them to a substantive process. Finally, we present a new model that is aligned with the data generative mechanism from a recent simulation study by Warren et al (2015). We argue that simulation studies have limitations often unacknowledged by researchers.
|