Childhood Behavior Severity Tied to Adult General Offending in Justice-Involved Youth

A new study of justice-involved youth finds that how serious a child’s behavioral problems are before age 11 is associated with whether that person keeps offending well into their twenties, but with a striking twist: the measure significantly predicted general offending in emerging adulthood, yet showed no significant association with violent offending. The research, published in the American Journal of Criminal Justice, used a statistical technique borrowed from educational testing to place more than 1,300 formerly adjudicated youth on a hidden continuum of childhood behavioral seriousness, then examined their offending years later.

The study drew on the Pathways to Desistance dataset, which followed 1,354 young people who had been adjudicated for a serious offense, recruiting participants from Philadelphia, Pennsylvania and Maricopa County, Arizona between 2000 and 2003. Qualifying offenses included all felonies as well as misdemeanor sexual assault and weapons-related charges. Participants were aged 14 to 19 at baseline and were interviewed at six-month intervals for the first three years, then annually, producing eleven waves of data spanning the period from 2000 to 2010. Attrition peaked at the final wave, when 16.2 percent of the original sample no longer provided data.

The theoretical backdrop comes from criminologist Terrie Moffitt’s dual taxonomy, which distinguishes adolescence-limited offenders, who start and stop offending during their teenage years, from life-course-persistent offenders, whose behavioral problems begin early and whose offending escalates in seriousness as they age. Moffitt proposed that neuropsychological vulnerabilities interacting with adverse social environments create a negative feedback loop that makes it difficult for this small subset of youth to transition away from antisocial lifestyles. Early childhood behavioral problems, in this framework, serve as observable markers of chronic offending risk, which is why researchers have long sought better ways to measure and interpret them.

What makes the new study distinctive is its measurement approach. Rather than simply counting how many problem behaviors a child displayed, the researchers used item response theory, or IRT, the same latent-variable framework used to design standardized exams. In educational testing, IRT identifies which exam questions best discriminate between students of higher and lower ability, distinguishing, for example, a simple addition problem from a long-division question. Applied here, the latent trait being probed is not academic ability but childhood behavioral severity, with each participant assigned a theta score locating them on a continuum of seriousness based on their pattern of responses across five binary items.

The five childhood behavioral indicators were drawn from self-reports in which participants indicated whether they had gotten in trouble for each behavior before age 11: fighting, stealing, cheating, disturbing class, and getting drunk or stoned. This framing matters, because the items capture not merely whether a child engaged in a behavior, but whether an adult detected and sanctioned it. Highly visible behaviors such as fighting and disturbing class are far more likely to come to adult attention than covert acts like cheating, stealing, or early substance use. The prevalence figures in the sample reflect this asymmetry: more than 60 percent of participants reported getting in trouble for disturbing class, while only 56 reported being drunk or stoned before age 11. That rarity is precisely what pushed the substance use item to the highest position on the seriousness continuum, since behaviors endorsed by fewer respondents sit at the more extreme end of the latent scale.

The two-parameter model estimated in the study yields two distinct pieces of information for each item. Difficulty locates the behavior along the seriousness continuum, marking the point at which a participant has a fifty percent probability of endorsing it. Discrimination describes how sharply an item separates participants with higher seriousness from those with lower seriousness, reflected in the steepness of the item characteristic curve. In this sample, disturbing class combined the lowest difficulty with by far the highest discrimination, and the authors interpret this as evidence that it functions less as a marker of extreme seriousness and more as a threshold indicator: endorsing it strongly signals that a child reports at least some behavioral issues, while failing to endorse it aligns with reporting none at all. Roughly a quarter of the sample reported none of the five behaviors, and disturbing class did much of the work of separating that group from the rest.

This pattern carries practical weight for screening. Because classroom disruption is conspicuous to teachers and school administrators, it offers an easily observable entry point for identifying children who may warrant deeper assessment. The study found that only about one in five participants who reported getting in trouble for disturbing class described it as their sole childhood behavioral issue, suggesting that visible classroom disruption rarely occurs in isolation. Channeling such children into additional screening could surface more covert and, by the study’s metric, more serious problems, including early substance involvement. At the same time, the authors caution that school discipline is not distributed evenly across racial groups, with prior research documenting elevated disciplinary rates for racially minoritized youth. This raises the possibility that endorsement of the disturbing class item partly reflects differential detection and sanctioning rather than differential behavior alone, a concern the authors flag for future work using differential item functioning analysis.

The predictive results add nuance to expectations derived from developmental taxonomies. Greater seriousness on the childhood behavioral continuum significantly increased the odds of general offending at the final wave, when participants averaged about 23 years of age. The general recidivism measure was broad, encompassing property offenses, drug sales, weapon carrying, and a range of violent acts, and nearly 39 percent of the sample reported at least one such offense during the final observation period. Yet the same continuum showed no significant association with violent offending specifically, which about 24 percent of participants reported. The authors note that only one of the five childhood items, fighting, tapped violent behavior, and they speculate that asking about fighting may evoke a mutual-combat interpretation that obscures one-sided forms of early aggression such as bullying or shoving. A richer battery of childhood violence items might improve prediction of violent outcomes, and expanding item coverage more generally could also reduce the near-perfect correlation observed between the IRT scores and a simple count of behaviors.

That correlation, at .986, is one of the study’s most instructive findings. With only five binary items, the sophisticated latent scoring added essentially no predictive information beyond a straightforward variety count, and sensitivity analyses using the count measure produced analogous results. The authors are candid about this, framing it as evidence that little is lost by using simpler count-based measures when item pools are small. Still, the IRT approach retained value in characterizing the individual items themselves, revealing which behaviors are rare and serious and which are common and discriminating. The longitudinal analyses reinforced the main findings while adding a temporal dimension: when participants were grouped into low, moderate, and high seriousness categories, the moderate and high groups showed consistently elevated general recidivism compared with the low group across most of the study period, with that gap persisting into the twenties, whereas elevated violent recidivism rates among the higher groups largely converged with the low group during the twenties. Notably, the moderate and high groups did not differ significantly from each other at several ages for either outcome, suggesting that the presence of any childhood behavioral issues, rather than gradations of seriousness along the continuum, may carry much of the signal.

The analytic strategy illustrates the practical demands of longitudinal criminological research. Data were not missing completely at random, so the authors used multiple imputation by chained equations, generating twenty imputed datasets and conditioning imputation on fully observed variables including gender, race, and the number of waves completed. Models controlled for an unusually thorough set of covariates, including impulsivity, deviant peer association, exposure to violence, hostility, socioeconomic status, age, and even the proportion of the final observation period spent in secured facilities, since incarceration limits opportunities to offend. Controlling for observation period length addressed the fact that participants followed for more days had more opportunities to report offenses. Several limitations nonetheless temper generalization: the sample consisted entirely of youth adjudicated for serious offenses, so findings may not extend to community samples; only five childhood behavioral measures were available, omitting behaviors such as weapon carrying; and the reliance on getting in trouble rather than on self-reported engagement introduces detection bias that likely varies across behaviors. For prevention, the implications are targeted rather than sweeping. Early substance involvement before age 11 stands out as the most serious indicator, while classroom disruption offers a visible screening trigger. The divergence between general and violent outcomes suggests that distinct developmental pathways may lead to different forms of continued offending, and the authors argue that identifying the underlying risk factors driving each pathway, rather than the behavioral symptoms alone, should guide future intervention research for youth flagged as high risk.

Subject of Research: Predicting recidivism in emerging adulthood from childhood behavioral issue severity among justice-involved youth using item response theory

Article Title: Understanding the Relevance of Early Childhood Behavioral Issues for Predicting Continued Recidivism in Emerging Adulthood Among Justice-Involved Youth: An Item Response Theory Approach

Article References: Wojciechowski, T. (2026). Understanding the Relevance of Early Childhood Behavioral Issues for Predicting Continued Recidivism in Emerging Adulthood Among Justice-Involved Youth: An Item Response Theory Approach. American Journal of Criminal Justice. https://doi.org/10.1007/s12103-026-09957-5

Image Credits: AI Generated

DOI: 10.1007/s12103-026-09957-5

Keywords: item response theory, juvenile justice, recidivism, childhood behavior problems, Pathways to Desistance, life-course persistent offending, emerging adulthood, violent offending, logistic regression, risk assessment, criminology, longitudinal study