Analyzing Criminal Trajectory Profiles: Bridging Multilevel and Group-Based Approaches Using Growth Mixture Modeling

• NCJ Number: 221977
• Journal: Journal of Quantitative Criminology Volume: 24 Issue: 1 Dated: March 2008 Pages: 1-31
• Author(s): Frauke Kreuter; Bengt Muthen
• Date Published: March 2008
• Length: 31 pages
• Publication Format: Article
• Publication Type: Report (Study/Research)
• Language: English
• Country: United States of America

Annotation

Using the well-known Cambridge data and the Philadelphia cohort study, this article compares the two "classical models"--conventional growth curve model and group-based trajectory--for the statistical modeling of longitudinal data; two growth mixture models are introduced to bridge the gap between conventional growth models and group-based trajectory models.

Abstract

For the Cambridge data, substantive researchers can be encouraged by the essentially equivalent results of the models compared. For the Philadelphia cohort study, the results differ across the models and substantively different conclusions are likely to be drawn if only one of the modeling approaches is used. The authors identify three issues that should be considered in applying any of the models discussed. First, an important part of mixture models is the prediction of class membership probabilities from covariates. This gives the profiles of the individuals in the classes. If theories differentially relate auxiliary information in the form of covariates of class membership and growth factors, those should be included in the set of covariates in order to correctly specify the model, find the proper number of classes, and correctly estimate class proportions and class membership. Second, similar to the examination of covariates, the predictive power of different trajectory types for later outcomes should be considered in the modeling. Third, researchers should also be aware that a model comparison can lead to different results for different data, as well as differently scaled outcome variables. Modeling comparison for ordered categorical and continuous outcomes can be found in Muthen (2001a, 2001b, 2004). More research on binary outcomes and counts is required. This article provides guidance on how such modeling comparisons can be approached. 11 tables, 9 figures, 47 references, and appended Zero-inflated Poisson Model