Adjusting risk of recidivism: do judicial departures worsen or improve recidivism prediction under the Federal Sentencing Guidelines?

A limited amount of research exists examining the ability of clinical or intuitive adjustments of formalistic methods of decision-making to improve upon predictive accuracy beyond that of the original measure. Using receiver operator characteristic (ROC) curves and the departure decisions of federal judges, the recidivism predictive utilities of two measures were compared. The two measures were the Criminal History category, based on the Federal Sentencing Guidelines (a formalistic procedure), and a re-coded measure of criminal history derived from the sentences actually imposed by judges after they departed from the Guidelines upwards or downwards for recidivism reasons (an adjustment to the formalistic procedure). Both a 10 year post-sentencing recidivism follow-up and a 6 year uniform follow-up period suggested that the Criminal History category performed poorly in predicting recidivism for this offender population, and that judicial departures not only failed to improve, but actually worsened, the predictive accuracy of pre-departure judgments. Policy implications of the findings are discussed.