Unveiling Disparities in Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) Algorithm

Prepared by: Shubham Vats

Introduction

Across the nation, judges, probation and parole officers are increasingly using algorithms to assess a criminal defendant’s likelihood of becoming a recidivist – a term used to describe criminals who re-offend. The dataset used more than 10,000 criminal defendants in Broward County, Florida, and compared their predicted recidivism rates with the rate that occurred over a two-year period. In our analysis, we used two statistical models: Logistic Regression and Cox Proportional Hazards Model.

Key Objectives:

• Exposing Disparities: Investigating the distribution of COMPAS scores among demographic groups.
• Demographic Predictors: Identifying key predictors, such as age and race, influencing higher COMPAS scores.
• Predictive Accuracy: Evaluating the correlation between high COMPAS scores and the likelihood of recidivism.
• Gender Disparities: Examining variations in recidivism rates between high-risk men and women.

Major Findings:

Score Distribution Disparities:

• General Recidivism: Black defendants show similar score distributions, while white defendants' scores favor lower-risk groups.
• Violent Recidivism: Discrepancies persist in the distribution of scores between white and black defendants.

Predictive Models:

• Logistic Regression (General Recidivism):
  • Age is a robust predictor, with defendants under 25 having a 2.5-fold increased chance of higher scores.
  • Black defendants are 45% more likely than white defendants to receive higher scores.
• Logistic Regression (Violent Recidivism):
  • Age is a significant predictor, with youthful defendants having a 6.4-fold higher chance of higher scores.
  • Black defendants have a 77.3% higher chance of receiving higher scores.
• Cox Proportional Hazards Model:
  • High COMPAS scores correlate with a 3.5 times higher likelihood of general recidivism.
  • Concordance score of 63.6% falls below industry standards, raising concerns about reliability.

Gender-Based Analysis:

• High-Risk Men vs. Women:
  • High-risk men recidivate more (61.2%) compared to high-risk women (47.5%).

Race-Based Analysis:

• Predictive Accuracy:
  • COMPAS predictive accuracy remains consistent between white (62.5%) and black defendants (62.3%).
  • Slight variation in concordance scores by race (69% for white, 67% for black).
• Race-by-Score Interaction:
  • Racial differences in hazard ratios: 2.99 for high-risk black defendants, 3.61 for high-risk white defendants.

Limitations:

• Despite predictive value, COMPAS scores fall below industry standards, casting doubt on the system's overall reliability.
• Significant disparities exist in COMPAS predictions across racial and gender subgroups, raising concerns about potential biases and challenging law enforcement interpretation.