Modeling conditional dependencies between recidivism and cognitive–emotion regulation strategies among prisoners using a Bayesian network with interpretable summary indexes

Emotion regulation is widely recognized as a critical factor in offender rehabilitation and recidivism reduction. However, empirical research has not yet clarified how distinct patterns of cognitive emotion regulation strategies (CERS) relate to recidivism across different types of criminal offenses. We therefore constructed a Bayesian network to model the probabilistic relationships between CERS and recidivism, conditioning on offense type. To obtain a practical, parsimonious network, we first derived interpretable summary indexes from multiple CERS indicators using convex generalized structured component analysis. These indexes were then used as network nodes. The resulting network suggests that CERS differ systematically across offense types and are differentially associated with the likelihood of recidivism. These findings underscore the importance of incorporating both individual CERS profiles and offense type into the design of correctional interventions aimed at reducing recidivism.