Proceedings of the 3rd International Conference on Public Management and Big Data Analysis, PMBDA 2023, December 15–17, 2023, Nanjing, China

Research Article

Fairness Under Unawareness: A New Estimator Reduces Disparity When Protected Class Is Unobserved

Download23 downloads
  • @INPROCEEDINGS{10.4108/eai.15-12-2023.2345325,
        author={Kaixin  Du and Yanhao  Ji and Fanxin  Sun and Yuanli  Zhu},
        title={Fairness Under Unawareness: A New Estimator Reduces Disparity When Protected Class Is Unobserved},
        proceedings={Proceedings of the 3rd International Conference on Public Management and Big Data Analysis, PMBDA 2023, December 15--17, 2023, Nanjing, China},
        publisher={EAI},
        proceedings_a={PMBDA},
        year={2024},
        month={5},
        keywords={fairness protected class switch estimator demographic disparity probablistic proxy model},
        doi={10.4108/eai.15-12-2023.2345325}
    }
    
  • Kaixin Du
    Yanhao Ji
    Fanxin Sun
    Yuanli Zhu
    Year: 2024
    Fairness Under Unawareness: A New Estimator Reduces Disparity When Protected Class Is Unobserved
    PMBDA
    EAI
    DOI: 10.4108/eai.15-12-2023.2345325
Kaixin Du1,*, Yanhao Ji2, Fanxin Sun3, Yuanli Zhu3
  • 1: Shanghai Tech University
  • 2: University of California
  • 3: University of Nottingham
*Contact email: dukx@alumni.shanghaitech.edu.cn

Abstract

Fairness has become an important topic in recent years. Since the boosting development of the society and economy, discrimination against certain disadvantaged groups or communities has been considered a great issue that needs to be detected and prevented seriously. Some real-world examples that might be affected by fairness could be job offers, university admissions and loan approvals. The absence of fairness towards a specific protected class, such as when an advantaged group is more likely to receive favorable outcomes than a disadvantaged one, can lead to significant societal issues. Thus, for both ethical and legal reasons, it is important to conduct estimation methods based on an individual's protected class information and the outcome of a binary decision to measure the fairness of such decisions.