Predicting arrest probability across time: An exploration of competing risk perspectives

Abstract

ObjectivesCriminal involvement is non-randomly distributed across individuals and across groups, resulting in differential probabilities of arrest. Thus, various predictors of arrest probability across time were examined for different groups. MethodsUsing data from the National Longitudinal Survey of Youth 1997, the current study examined latent class membership in the probability of arrest over a 15-year time span starting when participants were 12–16 years-old and ending when they were 28–31 years-old. Latent class regressions were employed to prospectively investigate whether demographic and criminological risk factors from the base wave could predict class membership. ResultsResults from the latent class growth analyses resulted in three to four classes characterized by an abstainer group, a stable, low-level group, an adolescent-limited group, and a stable moderate-level chronic group. In general, race, poverty, and other risk factors exhibited weak and inconsistent effects in predicting class membership. In contrast, being male and self-reported delinquency were consistent predictors of class membership. ConclusionAcross model comparisons and net of other predictors, self-reported delinquency was a strong predictor of class membership, suggesting that differential arrest probability is predominantly explained by time-stable delinquent behavior.