Modeling Risk at Sentencing: Addressing Theoretical and Methodological Concerns

• In Event: The Roles of Risk and Recidivism in Sentencing

Abstract

Actuarial risk assessments are increasingly being used at sentencing despite debates about the appropriateness of predictors in the risk scales. Opponents of risk assessment instruments fear that certain variables (such as geography) may be proxies for constitutionally suspect variables (such as race), and that the use of these tools will exacerbate disproportionalities in the criminal justice system. However, no studies have analyzed how different statistical methods may address concerns of disparate or unethical impacts on subgroups while maintaining the ability to effectively predict risk at sentencing. This paper focuses on the legal and theoretical concerns with using five controversial variables in sentencing risk assessment instruments: race, gender, age, prior arrests, and geography. Using data from the Pennsylvania Commission on Sentencing, this study analyzes how statistical methods can incorporate these controversial variables in order to maximize predictive validity while minimizing disparate impact. The results of this study indicate that unconstitutional and controversial variables (such as race and geography) must be statistically controlled for in order to maintain the use of important predictors (such as prior arrests) that may be correlated with these variables.

Authors

• Lauren Knoth, Pennsylvania State University

• Cynthia Kempinen, Pennsylvania Commission on Sentencing

• Leigh Tinik, Pennsylvania Commission on Sentencing

• R. Barry Ruback, Pennsylvania State University