Do Offenders Specialize? A Multilevel IRT Analysis of Offending Patterns in Adolescence and Early Adulthood



Journal Title

Journal ISSN

Volume Title



Research examining specialization in violence, or whether certain offenders commit violent crimes at a higher rate relative to their individual rate of offending, has important implications for policy makers and scholars alike. Despite good evidence for predictors of violence, most of these prior analyses focus on the frequency of violence, which is confounded by overall rate of offending and does not distinguish factors uniquely related to violent versus nonviolent offenses. Osgood and Schreck (2007) introduced an item response theory (IRT) measurement approach that is nested within a multilevel model, which overcomes many of earlier methods’ shortcomings. Several studies using this method have found evidence of specialization in violence and stability in measurement among adolescents, but longitudinal samples have been limited to five years or less; differences between local environments have yet to be examined within this framework. The current study utilized a multilevel IRT method of analysis to (a) determine whether individuals differ systematically in their pattern to commit violent versus nonviolent offenses; (b) determine whether there is stability (i.e., correlation) in the measurement of specialization and overall offending across our two measurement points; (c) examine demographic covariates, neuropsychological factors, peer risk factors, and environmental criminogenic risk factors; and (d) examine differences in the pattern of relationships between explanatory variables with overall offending compared to specialization. Altogether, our results provided several points in support of the existence of specialization as a phenomenon that is measurable, separate from the individual rate of offending and population base rates, and endures over time.



Specialization in violence, National study, Longitudinal study, Adolescent health, Hierarchical logistic regression, Multilevel item response model, Criminal offending