Impartial Predictive Modeling and the Use of Proxy Variables

TL;DR

This paper explores how to define and achieve fairness in predictive models by understanding the role of proxy variables, emphasizing the importance of full-data scenarios and analyzing partial-data estimation methods.

Contribution

It introduces a framework for impartiality considering different data-generating perspectives and clarifies the distinction between explainable variability and discrimination using proxies.

Findings

Fairness requires full-data knowledge of all covariates.

Regression decomposition reveals differences between explainable variability and discrimination.

Proxy variables can be analyzed to distinguish between bias and explainability.

Abstract

Fairness aware data mining (FADM) 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 distinction hinges on a rigorous understanding of the role of proxy variables; i.e., those variables which are associated both the protected feature and the outcome of interest. We demonstrate that fairness is achieved by ensuring impartiality with respect to sensitive characteristics and provide a framework for impartiality by accounting for different perspectives on the data generating process. In particular, fairness can only be precisely defined in a full-data scenario in which all covariates are observed. We then analyze how these models may be conservatively estimated via regression in partial-data settings. Decomposing the regression estimates provides insights into previously unexplored distinctions between explainable variability and discrimination that illuminate the use of proxy variables in fairness aware data mining.