Can a Dynamic Risk Instrument Make Short-Term Predictions in "Real Time"? Developing a Framework for Testing Proximal Assessment of Offender Recidivism Risk During Re-entry

Abstract

Correctional psychologists have classified risk factors for criminal recidivism into static (non-changing, non-improving) and dynamic (theoretically changeable) domains. By definition, a dynamic risk factor is a variable that can change across time, and as it changes, offenders’ likelihood of recidivism must logically change in the same direction. Thus, a measure that assesses dynamic risk factors repeatedly across time is hypothesized to provide evidence that more proximal re-assessments have greater predictive validity for short-term reoffending outcomes, compared to prior, more distal assessments. This dissertation develops and proposes a framework for testing the proximity hypothesis with longitudinal, multiple re-assessment re-entry data, and argues that Cox regression survival analysis with time-varying predictors is an appropriate statistical model for comparing the relative value of ratings taken later in the follow-up period, compared to baseline ratings taken at the time of release from incarceration. In this dissertation, the proposed three-step framework was applied to a large dataset of paroled offenders in New Zealand who were assessed by their supervision officers on a regular basis (N = 3694 offenders represented by N = 97,185 assessments on a measure of theoretically dynamic risk factors). Results demonstrated consistent support for the conclusion that re-assessment improves predictive validity, specifically by showing (a) incremental prediction of re-assessments over baseline scores, and (b) incremental prediction of the most proximal assessment compared to averages of earlier scores. Effect size indices (time-dependent area under the curve [AUC] statistics) temper the interpretation of these results by showing that predictive accuracy was only sometimes enhanced to a statistically significant degree when including re-assessments in the model. Still, the results encourage community supervision correctional agencies to consider employing repeated measurement of offender clients using a dynamic risk instrument. It is also recommended that researchers consider using the proposed framework as a logical approach to verifying the underlying assumption that purportedly dynamic risk factors are both dynamic, and accurate markers of fluctuations in individuals’ risk to re-offend.