ABSTRACT
“‘Predictive parity' actually corresponds to ‘optimal discrimination,'” said Nathan Srebro, associate professor of computer science at the university of Chicago and the Toyota technological institute at Chicago. That's because predictive parity results in a higher proportion of black defendants being wrongly rated as high-risk. The need to look at the harms that arise when a test is inaccurate arises frequently in statistics, particularly in fields like health care. When researchers weigh the merits of exams like mammograms, they want to know both how often they correctly detect breast cancer and how often they falsely indicate that patients have the disease. In the criminal justice context, false findings can have far-reaching effects on the lives of people charged with crimes. Judges, prosecutors and parole boards use the scores to help decide whether defendants can be sent to rehab programs instead of prison or be given shorter sentences.
