Smart Decarceration: Is It Possible to Detain Fewer People and Reduce Arrests?

This paper provides the results of a thought experiment to see what would have happened if a jurisdiction made release decisions solely based on risk assessment predictions of new arrests. Random forest imputation is used with data from all admissions to a large county jail system ( n = 28,188) to forecast new arrests for individuals detained by the court. After imputing outcome rates for the detained, we rank order everyone by their predicted probability of future arrest from lowest to highest probability and compare release and new arrest rates between the predicted outcomes and observed release decisions. The results show the risk-based release approach has the potential to reduce the detained population by 7% and reduce new arrests by 13% compared to current practices.