Criminal justice researchers often develop prediction instruments as a practitioner tool for improving the allocation of resources in community corrections administration. Although best practices have emerged for developing predictions, those best practices lead to predictions that fail to distinguish risk factors from control and correctional responses to risk. The consequence is that predictions fail to predict what they purport to predict, and this limits the utility of those predictions for public policy. This note argues that when properly done, predictions pertain to a latent, unobservable population. Given that perspective, some best practices advocated for prediction should be abandoned, and new best practices should be adopted.

You can purchase access to the full article here.