When Risk Factors Compete: A Machine Learning Approach to Identifying Antecedents of Adolescent Delinquency Across Domains, Development, and Gender
Abstract
Risk factors for adolescent delinquency have been identified at multiple levels, from individual dispositions through family and school processes to ecological context, yet their relative importance remains unestablished. Which factors carry greatest weight, and when, has not been tested in simultaneous modelling. This study assessed 111 predictors spanning five domains across four developmental periods, predicting self-reported delinquency in mid-adolescence. Predictors were teacher- and parent-reported and the outcome adolescent self-reported. Participants were a Swedish population-based cohort followed from early childhood into adolescence (N = 2,121; 46.6% girls; mean age 3.86 years at baseline; 97.8% born in Sweden). Using machine learning techniques, random forests were estimated across the full span (AUC = .787, 95% CI [.747, .826]) and at each developmental period for boys and girls. Parent-rated conduct problems ranked among the leading predictors at every period, while caregiver emotional functioning carried greatest weight in early and late childhood before being displaced by dispositional and behavioral characteristics. Parent-rated social competence was the strongest protective factor. Associations were threshold-shaped, with the shift to elevated risk falling within the ordinary range of the measures. Gender differences in ranking were not supported by formal testing, which was low-powered.
Related
- A Multi‐Task Learning Approach for Identifying Prognostic Variables Across Multiple Historical Clinical Trials
- Identifying ODEs from Unstructured Data with Causal Representation Learning
- A machine learning model for predicting ischemic and bleeding risk after percutaneous coronary intervention: Development and external validation
- Development of approximate scheduling-adaptive controllers for multi-products continuous chemical processes using deep learning techniques and model predictive control
- Development and external evaluation of an interpretable machine-learning model for early prediction of organ failure in higher-risk acute pancreatitis patients: A multicentre cohort study
BibTeX
@misc{cashman2026when,
title = {{When Risk Factors Compete: A Machine Learning Approach to Identifying Antecedents of Adolescent Delinquency Across Domains, Development, and Gender}},
author = {Matthew R. Cashman and Anna-Karin Andershed and Henrik Andershed and Frida Bood and Evelina Landstedt and Louise Frogner and Karin Hellfeldt},
howpublished = {PsyArXiv},
year = {2026},
month = oct,
doi = {10.31234/osf.io/pd9s2_v1},
url = {https://osf.io/preprints/psyarxiv/pd9s2_v1/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - When Risk Factors Compete: A Machine Learning Approach to Identifying Antecedents of Adolescent Delinquency Across Domains, Development, and Gender
AU - Matthew R. Cashman
AU - Anna-Karin Andershed
AU - Henrik Andershed
AU - Frida Bood
AU - Evelina Landstedt
AU - Louise Frogner
AU - Karin Hellfeldt
PY - 2026
DA - 2026/10/15
PB - PsyArXiv
DO - 10.31234/osf.io/pd9s2_v1
UR - https://osf.io/preprints/psyarxiv/pd9s2_v1/
AB - Risk factors for adolescent delinquency have been identified at multiple levels, from individual dispositions through family and school processes to ecological context, yet their relative importance remains unestablished. Which factors carry greatest weight, and when, has not been tested in simultaneous modelling. This study assessed 111 predictors spanning five domains across four developmental periods, predicting self-reported delinquency in mid-adolescence. Predictors were teacher- and parent-reported and the outcome adolescent self-reported. Participants were a Swedish population-based cohort followed from early childhood into adolescence (N = 2,121; 46.6% girls; mean age 3.86 years at baseline; 97.8% born in Sweden). Using machine learning techniques, random forests were estimated across the full span (AUC = .787, 95% CI [.747, .826]) and at each developmental period for boys and girls. Parent-rated conduct problems ranked among the leading predictors at every period, while caregiver emotional functioning carried greatest weight in early and late childhood before being displaced by dispositional and behavioral characteristics. Parent-rated social competence was the strongest protective factor. Associations were threshold-shaped, with the shift to elevated risk falling within the ordinary range of the measures. Gender differences in ranking were not supported by formal testing, which was low-powered.
ER -