Impartial Predictive Modeling: Ensuring Fairness in Arbitrary Models

Fairness aware data mining aims to prevent algorithms from discriminating against protected groups. The literature has come to an impasse as to what constitutes explainable variability as opposed to discrimination. This stems from incomplete discussions of fairness in statistics. We demonstrate that fairness is achieved by ensuring impartiality with respect to sensitive characteristics. As these characteristics are determined outside of the model, the correct description of the statistical task is to ensure impartiality. We provide a framework for impartiality by accounting for different perspectives on the data generating process. This framework yields a set of impartial estimates that are applicable in a wide variety of situations and post-processing tools to correct estimates from arbitrary models. This effectively separates prediction and fairness goals, allowing modelers to focus on generating highly predictive models without incorporating the constraint of fairness.