Presentation: Fairness in Classification: A Look at Bias in Recidivism Prediction Instruments
Speaker: Alexandra Chouldechova, Ph.D., Assistant Professor of Statistics and Public Policy, Heinz College, Carnegie Mellon University
Abstract: Recidivism prediction instruments (RPI’s) provide decision makers with an assessment of the likelihood that a criminal defendant will reoffend at a future point in time. While such instruments are already widely used across the country, their use is also attracting tremendous controversy. Much of the controversy concerns potential discriminatory bias in the risk assessments that are produced. In this talk I will discuss several “fairness” or “parity” criteria that have been applied to assess the predictive bias of recidivism prediction instruments. I will present a simple impossibility result showing that these criteria cannot all be simultaneously satisfied when recidivism prevalence differs across groups. To conclude, I will illustrate some limitations of looking at aggregate-level parity metrics.