Psychometric Racial and Ethnic Predictive Inequities
This study investigates the extent to which a commonly used offender risk needs assessment instrument equitably predicts probation success across racial and ethnic groups, specifically Black, Hispanic, and White probationers. The analysis focuses on the predictive accuracy and errors associated with the Wisconsin Risk Needs Assessment Instrument, highlighting the impact of these errors on probation outcomes. The findings underscore significant predictive inequities, particularly in the classification and supervision of Black and White probationers.
BACKGROUND
Behavioral assessments are critical tools in the American criminal justice system, influencing decisions for over 7 million offenders. Previous research has produced ambiguous results regarding the predictive equity of risk assessment tools across racial and ethnic groups, with some studies suggesting minimal bias and others highlighting significant disparities. This study builds on prior research by examining not just the regression slopes but also the distribution of predictive errors across different racial and ethnic groups, providing a more nuanced understanding of the equity of these instruments.
KEY FINDINGS
- Inequitable Predictions: The Wisconsin Risk Needs Assessment Instrument predicts probation success more accurately for higher-risk probationers but shows racial biases, particularly under-classifying Blacks and over-classifying Whites.
- Racial Bias: The instrument’s predictive accuracy is lower for Black probationers compared to Whites and Hispanics, with significant disparities noted especially among lower-risk offenders.
- Error Rates: The instrument over-classifies high-risk status for 45% of Whites, 41% of Hispanics, and 34% of Blacks, and under-classifies 33.5% of Blacks, 21.7% of Whites, and 24.1% of Hispanics.
- Risk Score Interaction: Black probationers are more likely to fail probation at lower risk scores, though this gap decreases as risk scores increase, indicating variable bias across risk levels.
- Recommendations: The study urges policy makers, practitioners, and researchers to address predictive biases in risk assessment tools to ensure fair and equitable probation outcomes across all racial/ethnic groups, aiming to reduce systematic discrimination.
RECOMMENDATIONS
– Adjust Risk Assessment Instruments: Revise and validate risk assessment instruments like the Wisconsin Risk Needs Assessment to reduce biases, particularly for Black probationers who are frequently under-classified and thus receive inadequate supervision.
– Focus on High-Risk Groups: Implement targeted strategies to address predictive errors among high-risk and low-risk probationers, ensuring appropriate levels of supervision and rehabilitative services that align with their actual risk and needs.
– Incorporate Offense Severity: Enhance the predictive models by consistently including offense severity, which has proven to significantly affect probation outcomes, thereby refining the assessment’s accuracy.
-Conduct Further Research on Bias: Promote more in-depth studies into the specific factors that contribute to racial and ethnic predictive inequities in risk assessments to guide future policy and practice improvements.
– Collaborate on Solutions: Encourage collaboration between researchers, practitioners, and policymakers to develop evidence-based solutions that mitigate predictive biases and support equitable probation outcomes.
CONCLUSION
The study identifies significant racial and ethnic predictive inequities within the Wisconsin Risk Needs Assessment Instrument, particularly affecting Black probationers. While the instrument predicts more accurately for high-risk probationers, it fails to equitably distribute errors across racial groups. These findings call for comprehensive adjustments to risk assessment tools and processes to enhance their fairness and effectiveness, ensuring that all probationers receive appropriate levels of supervision and support.
Sample Characteristics
The study analyzed data from 117,071 Texas probationers, comprising 40.3% Black, 18.5% Hispanic, and 41.3% White individuals. Probation success rates were 62.4%, with a notable disparity in failure rates among racial groups, especially for high-risk probationers.
Predictive Accuracy:
The Wisconsin Risk Needs Assessment Instrument showed varied predictive accuracy across racial groups. AUC scores were highest for Hispanic (.727) and White (.719) probationers, while Black probationers had the lowest AUC (.700), indicating less predictive accuracy for this group.
Error Distribution
The analysis revealed that the instrument over-classified White probationers and under-classified Black probationers, with 45% of Whites, 41% of Hispanics, and 34% of Blacks being over-classified. Conversely, 33.5% of Black probationers were under-classified compared to 21.7% of Whites.
Risk Scores and Predictive Bias
Higher risk scores correlated with higher rates of probation failure, but Black probationers were more likely to fail at lower risk scores compared to their White counterparts. This difference narrowed as risk scores increased, highlighting a critical intersection between race and risk level.
Role of Offense Severity
Adding offense severity to the predictive model improved its accuracy, with each increase in offense severity making an offender 1.24 times more likely to fail probation, even after controlling for risk score. This finding suggests that offense severity should be a crucial consideration in assessing probation outcomes.