This study examined predictive inequities in the Wisconsin Risk Needs Assessment tool across Black, Hispanic, and White probationers. It looked beyond overall predictive accuracy to focus on bias through examination of classification errors.
- The tool had higher error rates for under-classified Black and over-classified White probationers.
- It was more accurate for higher risk probationers than lower risk ones.
- Black probationers saw higher failure rates at lower risk levels than Whites.
- The tool explained only 19.4% of variation in probation failure overall.
- Presenting offense level provided additional predictive value beyond the risk score.
- Examining error and bias in risk tools beyond overall predictive accuracy is critical.