Predictions of Dangerousness in Sentencing: Déjà Vu All Over Again

Abstract

Predictions of dangerousness are more often wrong than right, use information they shouldn’t, and disproportionately damage minority offenders. Forty years ago, two-thirds of people predicted to be violent were not. For every two “true positives,” there were four “false positives.” Contemporary technology is little better: at best, three false positives for every two true positives. The best-informed specialists say that accuracy topped out a decade ago; further improvement is unlikely. All prediction instruments use ethically unjustifiable information. Most include variables such as youth and gender that are as unjust as race or eye color would be. No one can justly be blamed for being blue-eyed, young, male, or dark-skinned. All prediction instruments incorporate socioeconomic status variables that cause black, other minority, and disadvantaged offenders to be treated more harshly than white and privileged offenders. All use criminal history variables that are inflated for black and other minority offenders by deliberate and implicit bias, racially disparate practices, profiling, and drug law enforcement that targets minority individuals and neighborhoods.