COMPAS Recidivism Bias Study
In the aftermath of the death of George Floyd in Minneapolis, MN on May 25th, 2020, policing and the criminal justice system as a whole are under unprecedented scrutiny. It is more important now than ever before to examine the processes used within that system. The goal is to study one part of the system to determine whether or not it is impartial and therefore fair to all Americans.
COMPAS (Correctional Offender Management Profiling for Alternative Sanctions)
COMPAS is an algorithm used in the American criminal justice system for judges and parole officers to make judgements on a defendant's likelihood of reoffending - risk of recidivism. ProPublica set about to determine if the COMPAS algorithm is biased specifically against minorities. The resulting article concluded that there is significant bias against African Americans. Since its publication, there have been many other studies done with some confirming ProPublica’s conclusions while others hotly contest them
Data
Due to proprietary reasons, it is not possible to analyze the algorithm itself. All of the data used for this project was obtained from ProPublica’s github repository as linked from a Kaggle competition page: Propublica GitHub. There are five .csv files available for analysis though not all were used.
Recidivism rate vs COMPAS score
COMPAS score by age category
COMPAS Score by Marital Status
Conclusions
- One observation is clear, African Americans are arrested at a higher rate in general, regardless of other variables
- This points to the need to examine the racial bias of the system beyond the COMPAS algorithm
- If the criminal justice system is, itself, biased against minorities then the variables used by the algorithm would be corrupted by that bias and without access to the algorithm it is impossible to know if that baseline bias is accounted for when assigning COMPAS scores to defendants.