Using Dynamic Risk to Predict Violent Recidivism in "Real Time": Applying a Framework for Proximal Assessment of Risk of General Recidivism to Predict Violent Outcomes
In correctional psychology, risk factors are offender characteristics and contexts that increase the likelihood of reoffending. Risk is generally conceptualized as being either static or dynamic (Andrews & Bonta, 2010). Static risk factors are variables that cannot change, such as one’s criminal history or gender. Dynamic risk factors must, by definition, be able to change across time. Perhaps more importantly, changes in dynamic risk factors must correspond to changes in the likelihood of an offender committing a new offense. Although static risk is a more robust predictor of recidivism, dynamic risk is important, in that it (a) has clearer theoretical significance (e.g., the risk factors examined in the proposed study are based on cognitive and social psychology), and (b) can be changed (i.e., is a potential target for rehabilitation). The present study sought to determine whether certain dynamic risk factors could be used to predict violent recidivism. The study utilized a dataset comprised of an entire jurisdiction of paroled offenders in New Zealand during a two-year period (N = 3,421 offenders), reassessed approximately weekly or fortnightly over a period of up to two years. Using Cox regression survival analysis with time-linked covariates, I analyzed how theoretically important risk variables predicted violent reoffending in “real time”. Results indicated that offenders who went on to recidivate violently had higher levels of both static and dynamic risk. Additionally, their risk levels were significantly more erratic than those of offenders who did not recidivate violently, demonstrating greater fluctuation week-to-week. Results also indicated that repeated assessment of dynamic risk improved prediction of violent recidivism. Including updated re-assessments of dynamic risk factors into the statistical model significantly improved prediction incremental to both static risk and to baseline ratings of dynamic risk. Testing different strategies for averaging re-assessments indicated that while model fit was optimal without averaging for stable risk factors and protective factors, model fit for acute risk factors was best when the eight most recent assessments were aggregated. Finally, it was found that certain specific risk factors (antisocial attachment, sense of entitlement, unemployment, anger/hostility, and access to victims) differentially predicted violence over technical violations. These results provide unequivocal support for re-assessment of risk across the re-entry process. Additionally, they indicate that there may be important differences that distinguish violent offenders from non-violent offenders, regarding overall levels of risk, how risk changed across the follow-up period, as well as increased salience of specific risk factors. These findings have important implications for supervision practices, approaches to rehabilitation, and assessment of treatment success. As future studies clarify the temporal relationship between changes of risk levels and violent recidivism, the field moves closer to being able to predict—and prevent—imminent violence.^
Stone, Ariel Grace, "Using Dynamic Risk to Predict Violent Recidivism in "Real Time": Applying a Framework for Proximal Assessment of Risk of General Recidivism to Predict Violent Outcomes" (2017). ETD Collection for University of Texas, El Paso. AAI10688987.