Evaluating Causal-Based Fairness Metrics and Statistical Fairness Metrics ⚖️
This work aims to explore established causal-based frameworks and evaluate them under causality-based fairness notions, comparing these measures to statistical fairness metrics. A major limitation of existing causal literature is that it assumes that a causal model to work from, which is rarely the case. For this reason, part of our work will be analyzing ways to discover such a causal model through which we will evaluate the fairness notions specifically the direct and indirect effect of the sensitive attribute on the outcome. In this paper, we developed a casual fairness pipeline for observational data. This pipeline can be applied to analyze classification outcomes and give insight into the effect of statistical fairness mitigation algorithms.
