Methodological Sensitivities to Latent Class Analysis of Long-Term Criminal Trajectories

Abstract

A recent and growing body of research has employed a semiparametric group-based approach to discover underlying developmental trajectories of crime. Enthusiasm for such latent class models has not been matched with robustness and sensitivity analyses to determine how conclusions from the method vary according to fundamental methodological problems that inhere in criminological data. Using a sample of 500 delinquent boys and their official crime counts from ages 7 to 70, this paper systematically addresses how three concerns in longitudinal research—(a) length of follow-up, (b) the inclusion of exposure time (incarceration), and (3) data on involuntary desistance through death—influence our inferences about developmental trajectories. While there is some evidence of stability, a comparison of group number, shape, and group assignment across varying conditions indicates that all three data considerations can alter trajectory attributes in important ways. More precisely, longer-term data on offending and the inclusion of incarceration and mortality information appear to be key pieces of information, especially when analyzing high-rate offending patterns.