Predicting Criminal Charges in Adolescents Using Multi-Informant Assessments at Child Welfare Intake

Technical Summary: Predicting Criminal Charges in Adolescents Using Multi-Informant Assessments at Child Welfare Intake

Problem Statement
Despite a general decline in youth crime in recent decades, rates have recently plateaued, with self-reported offending increasing since 2015. A critical public health challenge is identifying adolescents within Child Welfare Services (CWS) who are at elevated risk of progressing to life-course-persistent offending trajectories. Current risk assessment tools, such as the Level of Service scales, are primarily validated on juvenile justice populations with established offense histories. Their applicability to CWS populations—where youths are referred for family dysfunction or disruptive behavior rather than criminal acts—remains uncertain. Furthermore, interventions are often reactive, occurring only after justice system involvement. There is an urgent need for validated, feasible screening tools tailored to CWS contexts that utilize multi-informant data (parents, self, teachers) to predict future criminal charges and inform targeted preventive interventions before justice system contact occurs.

Methodology
The study analyzed data from a two-arm randomized controlled trial (RCT) conducted in routine CWS settings across four sites in Norway between 2013 and 2017.

• Sample: 129 adolescents (mean age 15) and their families were recruited. Inclusion criteria included youths aged 11–18 with violent, criminal, or aggressive behavior, risk of out-of-home placement, or school-related problems. Notably, only 5% of the sample had been referred specifically for a criminal offense.
• Design: Participants were randomized to Functional Family Therapy (FFT) or Treatment as Usual (TAU). TAU included various interventions such as Multisystemic Therapy (MST), parent training, or family counseling.
• Predictors: Pre-treatment assessments (intake) included multi-informant measures:
  • Parent/Teacher Reports: Rule-breaking behavior, aggressive behavior, social skills, and callous-unemotional (CU) traits (Inventory of Callous Unemotional Traits - ICU).
  • Self-Reports: Delinquency (Self-Reported Delinquency - SRD) and CU traits.
  • Teacher Reports: Hyperactivity/impulsivity (SNAP-IV), oppositional-defiant behavior (IOWA), and CU traits.
  • Intake Staff: Youth Level of Service/Case Management Inventory (YLS/CMI).
• Outcomes: The primary outcome was the occurrence of criminal charges (CC), linked via national registry data from Statistics Norway at 18 and 36 months post-intake.
• Analysis: The study employed Welch's t-tests to identify initial predictors, followed by logistic regression models controlling for treatment condition, age, sex, parental education, and income. Interaction effects between predictors and treatment condition were tested. Predictive accuracy was evaluated using Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC) statistics.

Key Results

1. Predictive Validity of Multi-Informant Data:

   • Rule-Breaking/Delinquency: Parent-reported rule-breaking, self-reported delinquency, and teacher-reported rule-breaking significantly predicted criminal charges at both 18 and 36 months. These factors remained significant after controlling for covariates and treatment condition.
   • Callous-Unemotional (CU) Traits: Teacher-reported CU traits were a robust predictor of criminal charges at both 18 and 36 months. A one standard deviation increase in teacher-reported CU traits was associated with a substantial increase in the estimated probability of being charged (e.g., from ~16% to ~52% at 18 months).
   • Non-Predictors: Aggression (parent and teacher reports), hyperactivity/impulsivity, self- and parent-reported CU traits, and the YLS/CMI risk score did not significantly predict future criminal charges in this sample.
   • Time-Specific Predictors: Teacher-reported oppositional-defiant behavior and low social skills predicted charges at 18 months but lost significance at 36 months.

2. Treatment Interaction:

   • While the overall study found no significant main effect of FFT versus TAU on criminal charges across the total sample, interaction analyses revealed that FFT was associated with reduced offending specifically among school-attending youths with higher risk profiles (those scoring high on rule-breaking or CU traits).
   • Significant interactions were found between treatment condition and both teacher-reported rule-breaking and teacher-reported CU traits. Youths with higher scores on these traits in the FFT group had lower odds of receiving criminal charges compared to similar youths in the TAU group.

3. Predictive Accuracy:

   • ROC analysis indicated moderate to good predictive accuracy for the significant variables. AUCs for rule-breaking and delinquency ranged from 0.70 to 0.79 at 18 months.
   • Teacher-reported CU traits showed particularly strong discrimination at 18 months (AUC = 0.81), though this decreased to moderate levels at 36 months (AUC ≈ 0.68).

Significance and Claims
The paper claims that integrating multi-informant assessments at CWS intake can effectively identify adolescents at heightened risk of future criminal charges, even in populations where prior criminal history is rare. The study demonstrates that specific traits—particularly rule-breaking behavior and teacher-reported callous-unemotional traits—are superior predictors in this context compared to standard justice-derived tools like the YLS/CMI, which rely heavily on prior offense history.

The authors argue that these findings support the development of practical assessment strategies at CWS intake that synthesize caregiver, youth, and teacher reports. Such strategies could enable the timely deployment of evidence-based preventive services (like FFT) to high-risk youth before they enter the juvenile justice system. The study modestly suggests that while FFT may not reduce offending across all CWS populations, it shows promise for reducing charges among those with elevated risk profiles (high rule-breaking or CU traits), highlighting the importance of matching intervention intensity to specific risk factors. The authors conclude that further research is needed to evaluate and adapt justice-derived instruments for CWS samples and to test screening protocols that utilize these multi-informant reports.