Fairness Violations and Mitigation under Covariate Shift

TL;DR

This paper addresses the challenge of maintaining fairness in predictive models under distribution shifts by leveraging causal graphs and feature selection to ensure stability in both accuracy and fairness, with applications in healthcare.

Contribution

It introduces a causal graph-based approach for stable fair prediction under covariate shift, providing conditions for optimal fairness and accuracy preservation.

Findings

The method improves fairness in healthcare decision-making.

It guarantees worst-case optimality for certain fairness definitions.

The approach effectively estimates test set fairness metrics.

Abstract

We study the problem of learning fair prediction models for unseen test sets distributed differently from the train set. Stability against changes in data distribution is an important mandate for responsible deployment of models. The domain adaptation literature addresses this concern, albeit with the notion of stability limited to that of prediction accuracy. We identify sufficient conditions under which stable models, both in terms of prediction accuracy and fairness, can be learned. Using the causal graph describing the data and the anticipated shifts, we specify an approach based on feature selection that exploits conditional independencies in the data to estimate accuracy and fairness metrics for the test set. We show that for specific fairness definitions, the resulting model satisfies a form of worst-case optimality. In context of a healthcare task, we illustrate the advantages of the approach in making more equitable decisions.