U.S. flag

An official website of the United States government, Department of Justice.

NCJRS Virtual Library

The Virtual Library houses over 235,000 criminal justice resources, including all known OJP works.
Click here to search the NCJRS Virtual Library

Predicting Recidivism Risk: New Tool in Philadelphia Shows Great Promise

NCJ Number
240696
Journal
NIJ Journal Issue: 271 Dated: February 2013 Pages: 4-13
Author(s)
Nancy Ritter
Date Published
February 2013
Length
10 pages
Annotation
This article discusses the Philadelphia Police Department's use of a computerized system to predict with greater accuracy which probationers would be more likely to reoffend.
Abstract
This article discusses the random forest model, a computerized system developed by criminologists from the University of Pennsylvania and officials with the Philadelphia Department of Adult Probation and Parole to more accurately predict which probationers would be more likely to reoffend within two years of returning to the community. The system, which has been used successfully for over 4 years in Philadelphia, works by assessing each new probation case at its outset and assigning the probationer to a high-, moderate-, or low-risk category. The success of the system is due to the use of a sophisticated statistical approach that considers the nonlinear effects of a greater array of variables with complex interactions. The article presents information on how a jurisdiction can use the random forest model for its own police department. The steps needed to assess whether the model is appropriate include determining what data already exists in electronic form, determining when the forecasting begins and when the forecasting ends, and determining an acceptable error rate. The author notes that the Philadelphia Police Department has had a 66 percent accuracy rate with the system that is has been using. The final sections of the article discuss the benefits of random forest modeling, how it promotes the efficient use of resources, and the fairness and equitability that is inherent in the model's design. Recommendations for future research are discussed.