Primary Submission Category: Causal Fairness, and Bias/Discrimination
Individualized Inference for Causal Fairness through Conformal Mediation Analysis
Authors: Cheng Yu, Zhimei Ren,
Presenting Author: Cheng Yu*
Ensuring causal fairness is a critical concern across a wide range of applications. However, assessing fairness at the individual level and identifying units that have experienced unfair outcomes remain challenging tasks. In this paper, we focus on the direct effect of a sensitive attribute on an outcome, potentially mediated by other variables, and propose a novel framework for individualized statistical inference. Our approach integrates causal mediation analysis with conformal prediction to enable inference on causal fairness at the individual level. To control the false discovery rate (FDR) in selecting individuals subjected to unfair treatment, we further develop a multiple testing procedure based on the conformalized e-values with conditional calibration. We formalize the notion of individual causal fairness and demonstrate the utility and novelty of our methodology through extensive simulations and two real-world applications.