In community-based corrections, reassessment of dynamic risk factors improves the prediction of recidivism relative to initial risk assessment at the time of release. However, there is less evidence for predictions of violent recidivism. We examined whether reassessment proximity or aggregation of reassessments improved the prediction of imminent violence in a sample of paroled individuals on community supervision. We hypothesized that reassessment of dynamic risk would better predict violent recidivism than initial risk assessment at the time of release. Examination of aggregation and individual risk-factor domains was exploratory. In a prospective study of violent recidivism in a sample of individuals on community supervision in New Zealand (75,917 assessments from 3,421 participants; 92.8% men), we used supervision officers' ratings of dynamic risk (assessed using Dynamic Risk Assessment for Offender Re-entry [DRAOR]) and static risk scores (using the Risk of ReConviction × Risk of Imprisonment) to predict imminent violence (within 2 weeks). Individuals who recidivated violently had higher initial risk ratings (DRAOR Stable d = 0.36, 95% CI [0.17, 0.55]; DRAOR Acute d = 0.45, 95% CI [0.26, 0.64]) and showed more week-to-week fluctuations in risk ratings (DRAOR Stable d = 0.21, 95% CI [0.04, 0.41]; DRAOR Acute d = 0.26, 95% CI [0.06,0.46]). Total averages of faster-changing acute risk factors best predicted violence (c-index = 0.68), with changes in these factors incrementally predicting violence over well-established predictors (criminal history) and initial scores (Δχ2 = 15.54, df = 3). The constructs that best discriminated violence were consistent with social cognition explanations of violence. Because client consistency as determined through score aggregation was more important than current presentation, supervision officers should consider overall patterns of interpersonal hostility and reactivity rather than assuming the emerging presence of these factors will signal imminent violence among previously violent individuals. (PsycInfo Database Record (c) 2021 APA, all rights reserved).
Read full abstract