Drug testing—either random, scheduled, or for-cause—is a common condition of community supervision. This study investigates the utility of drug testing measures in predicting arrest or revocation outcomes for individuals on probation. Results were generated during the development of a set of new risk classification algorithms for individuals on probation or parole in Georgia. Multiple analytic approaches and predictive factors were tested in the development of these models that were built using data for approximately 154,000 individuals who began probation or parole between January 1, 2016, and December 31, 2019. Accuracy at predicting rearrest was improved by implementing time-period-specific models; using recent rather than lifetime criminal history; and incorporating dynamic measures of both protective and risk factors, including drug test results and absence of testing. Models were estimated separately for men and women, for different supervision statuses (probation-only, split-sentence probation, and parole), and for three consecutive time periods (first quarter, next three quarters, one-plus years). Positive test results were predictive of future arrest or revocation, while negative results and absence of testing in latter-period models were associated with reduced risk. Results are presented for the probation-only models for men and women, but similar results obtained for the split-probation and parole models.

This content is only available via PDF.