Computer Model Predicts at Arraignments the Likelihood of Offending

A recent study examined “machine learning application to forecasts of domestic violence, defined broadly as in the governing statute, to inform release decisions at a preliminary arraignment.” The study, “Forecasting Domestic Violence: A Machine Learning Approach to Help Inform Arraignment Decisions,” by Richard A. Berk, Susan B. Sorenson, and Geoffrey Barnes, examined 8,646 domestic violence arraignments between January 1, 2007 and October 31, 2011, and included a two-year follow-up from the point of release after arraignment.  The “observational units” were the cases, and not individuals, because some individuals might appear more than once in the data.  The study found that almost twenty percent of those arrested and arraigned for domestic violence will again be arrested for domestic violence within the two-year period.

The input factors for the machine learning algorithm included age, gender, high-crime zip codes, the number of prior DUI, domestic violence, animal mistreatment, prior serious, property, weapons, assaultive crime and drug charges, age at first arrest, prior jail or prison sentences, and prior failures to appear; and the number of instant criminal charges.

The researchers concluded that using their forecasts, the re-offending rate could be reduced by one-half, or with a re-arrest rate of about 10%.  “If magistrates only released offenders our forecasts identified as good bets, approximately 10 percent of those offenders would be arrested for domestic violence within two years. Failures could be cut in half.”

Sources:  Faye Flam, “The Crime You Have Not Yet Committed,” bloombergview.com, March 8, 2016: http://www.bloombergview.com/articles/2016-03-08/the-crime-you-have-not-yet-commit ted. Richard A. Berk, Susan B. Sorenson and Geoffrey Barnes, “Forecasting Domestic Violence: A Machine Learning Approach to Help Inform Arraignment Decisions,” Journal of Empirical Legal Studies, Vol. 13, Issue 1, March, 2016: http://onlinelibrary.wiley.com/doi/10.1111/jels.12098/abstract and http://onlinelibrary.wiley.com/doi/10.1111/jels.12098/full

by Neil Leithauser
Associate Editor