A multiple risk factor approach for predicting DWI recidivism.

Abstract

A sample of DWI (driving while impaired) offenders was studied to compare various approaches for predicting reoffenses over a 4-year period. Logistic regression yielded multivariate predictor equations that were significant statistically, but were not helpful to clinicians in assessing risk for reoffending. As a different approach, five predictor variables that were consistently correlated with reoffense status were examined to determine the cut score at which the repeat offense rate exceeded the base rate. These were combined to yield the number of risk factors (from 0 to 5) for each offender. This method, used for the original and a hold-out sample, yields results as accurate as those derived from a logistic regression model that includes all the risk variables, and allows clinicians to classify offenders into low and high risk categories in a straightforward manner. Nearly half of offenders with four or five risk factors (age, years of education, arrest blood alcohol concentration (BAC), score on the receptive area scale of AUI and raw score on the MacAndrews scale of MMPI-2) were rearrested compared to the base rate (25%). However, this method is not sufficiently precise to accurately predict which individuals will and will not be rearrested. Although generalizability of specific algorithms across populations needs to be examined, this method appears promising as a clinically accessible way to classify, in a given offender population, those who are most likely to repeat the offense.