Abstract

Leveraging current legal standards, we define bias through the lens ofmarginal benefits and objective testing with the novel metric "ObjectiveFairness Index". This index combines the contextual nuances of objectivetesting with metric stability, providing a legally consistent and reliablemeasure. Utilizing the Objective Fairness Index, we provide fresh insights intosensitive machine learning applications, such as COMPAS (recidivismprediction), highlighting the metric's practical and theoretical significance.The Objective Fairness Index allows one to differentiate between discriminatorytests and systemic disparities.

 

Quick Read (beta)