Asymmetric Loss Functions for Forecasting in Criminal Justice Settings

The statistical procedures typically used for forecasting in criminal justice settings rest on symmetric loss functions. For quantitative response variables, overestimates are treated the same as underestimates. For categorical response variables, it does not matter in which class a case is inaccurately placed. In many criminal justice settings, symmetric costs are not responsive to the needs of stakeholders. It can follow that the forecasts are not responsive either. In this paper, we consider asymmetric loss functions that can lead to forecasting procedures far more sensitive to the real consequences of forecasting errors. Theoretical points are illustrated with examples using criminal justice data of the kind that might be used for “predictive policing.”