Maximising existing assets: the potential of Bayesian hierarchical approaches to improve the risk assessment process in youth justice

Bayesian approaches provide a fresh lens to consider the interaction between various characteristics and circumstances of children who have offended, and their likelihood of committing further offences over time. This has been done here by revisiting data captured as part of the risk assessment process used until the mid-2010s across England and Wales to consider how the probability of further offending behaviour differs depending upon whether the child has a prior history of offending or not. Whilst Asset has now been replaced by AssetPlus, the structure of the earlier risk assessment tool lends itself to mimicking rapid changes in the lives of some children and the evolving nature of youth offending. This has been done by treating the assessments conducted with 87 children in the formal youth justice system in a single Welsh local authority area whose supervision started in either 2012/13 or 2023/14 as a longitudinal dataset. The likelihood of further offending behaviour at different time points has then been estimated using additive binary logistical regression models based upon practitioner ratings for 12 domains of ‘risk’, along with a range of time-varying and non-time varying variables reflecting facets of their criminal career. This study demonstrates the utility of conducting the analysis in a Bayesian framework. Notably it highlights the potential to incorporate new ideas be they emerging theoretical perspectives, interventions or additional variables into existing models to increase understandings of the complex relationship between ‘risk’ and ‘protective’ factors for different subgroups.