A Different Method of Predicting Risk: Unpacking the Potential of a Statewide Sentencing Risk Assessment

As states attempt to reduce prison populations, they are increasingly turning to sentencing risk assessments (RAs) to identify serious offenders eligible for non-incarcerative punishments. Despite a recent proliferation of pro-sentencing RA legislation, many questions remain about
their effectiveness, predictive validity, and potential effects on racial disproportionality. This project seeks to expand on a previously funded RA study to answer three main questions: 1) Which offender characteristics best predict post-release recidivism (i.e., arrest, violent arrest, felony conviction, and re-incarceration) in a group of serious offenders? 2) Does grouping offenders according to their combination of risk factors predict who recidivates? and 3) Does
using a formalized risk-based sentencing structure increase racial disproportionality in incarceration - compared to the current sentencing structure?

These questions will be answered in three phases using a longitudinal sample of 10,002
serious adult offenders sentenced in Pennsylvania between 2001-2005 and followed for up to 11.6 years. Offender characteristics and outcomes measures are collected from the Pennsylvania Department of Corrections, the Pennsylvania Commission on Sentencing, and the Pennsylvanian
State Police. First, we use a Cox proportional hazards model to identify which factors and offender characteristics best predict risk of recidivism. Along with standard demographic and offense characteristics, we expand the potential risk-predictive measures to include pro-criminal attitudes, peer contacts, and socio-economic status (among others) – many of which are largely missing from previous sentencing RA development studies. Second, we use a Burgess equal weight linear model to create a sentencing RA tool based on the findings. Receiver Operator Characteristics (ROC) are used to provide a measure of how well the tool balances true positives
and true negatives and Area Under the Curve (AUC) estimates are used to compare predictive power across the validation and test samples. Third, we follow a propensity score approach to assesses how likely it is for a minority offender to get an incarcerative sentence, compared to a
similarly situated white offender, under both the current sentencing guidelines and under the new risk-based sentencing structure.

Results from this study will benefit the criminal justice field by informing scholars and practitioners about the relationship between offender characteristics, theoretical concepts, and criminality. States considering adopting an empirically-based RA will be interested in the process behind creating a validated tool. Finally, this study addresses a critical question missing from previous RA studies about the potential effects of using a formalized method of predicting risk on racial disproportionality.
ca/ncf