Fairness Under Unawareness: Assessing Disparity When Protected Class Is Unobserved

Assessing the fairness of a decision making system with respect to a protected class, such as gender or race, is challenging when class membership labels are unavailable. Proxies, such as surname and geolocation for race, are sometimes used to impute these missing labels for compliance assessments. In this paper, the authors decompose the biases arising from this and propose an alternative.