Perceived Algorithmic Fairness using Organizational Justice Theory: an Empirical Case Study on Algorithmic Hiring

Abstract

Growing concerns about the fairness of algorithmic decision-making systems have prompted a proliferation of mathematical formulations aimed at remedying algorithmic bias. Yet, integrating mathematical fairness alone into algorithms is insufficient to ensure their acceptance, trust, and support by humans. It is also essential to understand what humans perceive as fair. In this study, I therefore conduct an empirical user study into crowdworkers’ algorithmic fairness perceptions, focusing on algorithmic hiring. I build on perspectives from organizational justice theory, which categorizes fairness into distributive, procedural, and interactional components. By grouping participants based on the type of information they receive about several hypothetical recruitment algorithms, I find that algorithmic fairness perceptions are higher when crowdworkers are provided not only with information about the algorithmic outcome but also about the decision-making process. Remarkably, this effect is even observed when the decision-making process can be considered unfair, when gender, a sensitive attribute, is used as a main feature. By showing realistic trade-offfs between fairness criteria, I find a preference for equalizing false negatives over equalizing selection rates amongst groups. Moreover, I discover a negative effect of selection rate differences and false negative rate differences on fairness perceptions. These findings contribute to the literature on the connection between mathematical algorithmic fairness and perceived algorithmic fairness, and highlight the importance of considering multiple components of algorithmic fairness, rather than solely treating it as an outcome distribution problem. Importantly, this study highlights the potential benefits of leveraging organizational justice theory to enhance the evaluation of perceived algorithmic fairness.