Gender, Race/Ethnicity, and Prediction: Risk in Behavioral Assessment

This study explores the predictive validity of the Wisconsin Risk Needs Assessment Instrument across racial and ethnic groups of female probationers, considering additional contextual variables such as location and offense severity. The primary focus is on understanding whether the instrument equitably predicts probation outcomes for female minority probationers, addressing existing gaps and inconsistencies in prior research on gender, race, and risk assessment in correctional settings.

BACKGROUND

Evaluations of risk assessment instruments are critical in predicting reoffending and guiding probation decisions. However, previous studies have shown conflicting results regarding the predictive accuracy of these instruments across racial and gender lines. While some studies report no significant differences in predictive accuracy among racial groups, others highlight disparities, particularly for minority probationers. Similarly, findings on gender-specific predictions and probation outcomes remain inconclusive, with some research suggesting the need for gender-responsive assessment tools.

KEY FINDINGS

  1. Predictive Inequities: Risk assessment tools like the Wisconsin Risk Needs Assessment show varying accuracy and error rates across racial/ethnic groups and genders, particularly underpredicting needs for Hispanic female probationers.
  2. Racial Disparities: Black probationers experience higher misclassification rates and overclassification errors, while Hispanic probationers often miss out on needed interventions due to high underclassification rates.
  3. Gender Differences: Female offenders, especially minorities, are frequently overclassified as high-risk, leading to unnecessary interventions. However, outcomes can vary widely, showing inconsistencies in gender impacts on probation success.
  4. Location and Offense Severity: Probation outcomes are heavily influenced by location and offense severity, often more than individual risk scores, pointing to the role of external factors like jurisdictional policies.
  5. Interaction Effects: Race, location, and risk scores interact significantly, with Black and Hispanic female probationers in specific counties facing higher rates of probation failure, highlighting the complex factors influencing outcomes.

RECOMMENDATIONS

– Revise Risk Assessment Tools: Adjust the Wisconsin Risk Needs Assessment Instrument to reduce biases, particularly for Black and Hispanic female probationers, ensuring more equitable predictions across racial and gender groups.

– Incorporate Contextual Variables: Integrate location and offense severity as standard factors in predictive models to improve the accuracy and fairness of probation outcome predictions.

– Develop Gender-Responsive Assessments: Consider gender-specific risk factors and develop assessment tools that address the unique needs of female probationers, particularly those from minority backgrounds.

– Expand Research on Predictive Inequities: Encourage further research on the intersection of race, gender, and location in probation outcomes to identify and address systemic biases in risk assessments.

– Implement Error Analysis in Validation: Include error analysis as a standard component in the validation of risk assessment instruments, focusing on understanding and reducing predictive errors for minority groups.

CONCLUSION

The study underscores significant predictive inequities in the Wisconsin Risk Needs Assessment Instrument for minority female probationers, with Black women particularly affected by lower predictive accuracy and higher error rates. These findings highlight the need for comprehensive revisions to risk assessment tools, incorporating contextual variables and focusing on reducing biases. Addressing these disparities is crucial for ensuring fair and effective probation outcomes for all individuals, regardless of race or gender.

Sample Characteristics

The study used data from 34,238 female probationers from a large southwestern state's probation database, including Black (43.8%), Hispanic (11.7%), and White (44.4%) individuals. Most were on probation for misdemeanors (81.3%), and ages ranged from 23 to 93 years.

Predictive Accuracy and Errors

The Wisconsin Risk Needs Assessment Instrument's predictive accuracy varied across racial groups. The AUC was highest for Hispanic probationers (0.75) and lower for Black probationers (0.68), indicating less predictive accuracy for Black individuals. Overclassification errors were similar between Black (56.9%) and White (53.6%) probationers but significantly lower for Hispanic probationers (16%).

Impact of Contextual Variables

Location, race, offense severity, and risk score significantly impacted probation outcomes. Specifically, being Black, serving probation in County A, having a felony charge, and higher risk scores increased the likelihood of unsuccessful probation completion. Notably, location was a more predictive factor than offense severity, race, or risk score alone.

Interaction Effects

The interaction between race and location revealed that Black and Hispanic probationers in County A were more likely to fail probation than their White counterparts. Additionally, the risk score was a significant predictor of probation failure for Black probationers but not for other racial groups, indicating the importance of considering racial interactions in predictive models.

Predictive Bias

The study highlighted potential biases in the risk assessment tool, particularly for Black female probationers, where predictive accuracy was lower, and error rates were higher. This finding calls for closer examination of the factors contributing to these disparities and suggests the need for adjustments to enhance the tool's equity.