Heuristic Reasoning Distorts Police Predictions of Domestic Abuse

Abstract
Police in England and Wales are asked to predict the likelihood of serious recidivism in domestic abuse cases, without being given much support. We find variation in their skill levels, but at the same time, their predictions are generally poor. We ask how they formulate those predictions. We find substantial evidence of heuristic reasoning, including salience effects, representativeness bias, and implicitly, correlation neglect. These issues are greater for officers with lower skill levels. Analyzing decisions in prediction problems requires a means of adjusting the observed outcome for the censoring that may arise as a result of the prediction. We propose a method for dealing with such censoring which may be useful in other settings where workers charged with a prediction problem are not randomly assigned to cases. Joint paper with Andrew Jordan and Tom Kirchmaier.