Liya Mulugeta

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2023-294

December 1, 2023

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2023/EECS-2023-294.pdf

A key concern of policymakers who use student-assignment algorithms is increasing diversity and fairness. Drawing on two years of research into the student assignment algorithmic mechanism of Ethiopia, I studied how the collective values of stakeholders can shape the parameters of algorithmic mechanisms. I found that prioritizing the collective values of stakeholders in the initial stages of the algorithm design process enhances university diversity, empowers students to submit truthful rank-order lists, and establishes accountability mechanisms for universities. One of the main contributions in this thesis is a novel assessment of mechanism fairness, as defined by having: (1) an absence of justified envy, (2) an absence of lack of information, and (3) an absence of misalignment of values. Ethos, the core technical contribution in this thesis, consists of a machine learning-backed acceptance rate quiz for Ethiopian public universities, an informative portal for Ethiopian students, and a generalized student-university matching algorithm. My process for designing and evaluating Ethos consists of two in-person user studies (Study 1, n=33; Study 2, n=40) in which I identified and assessed the real-world impact of my system and algorithm parameters. I argue that listening to and prioritizing the collective values of stakeholders is critical to building diverse and fair algorithmic mechanisms and offer generalizable methods for doing so.

Advisors: Niloufar Salehi

---

BibTeX citation:

@mastersthesis{Mulugeta:EECS-2023-294,
    Author= {Mulugeta, Liya},
    Editor= {Salehi, Niloufar and Wagner, David A.},
    Title= {Ethos: Designing Algorithmic Mechanisms for Increased Fairness, Stakeholder Empowerment, and Systemic Accountability},
    School= {EECS Department, University of California, Berkeley},
    Year= {2023},
    Month= {Dec},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2023/EECS-2023-294.html},
    Number= {UCB/EECS-2023-294},
    Abstract= {A key concern of policymakers who use student-assignment algorithms is increasing diversity and fairness. Drawing on two years of research into the student assignment algorithmic mechanism of Ethiopia, I studied how the collective values of stakeholders can shape the parameters of algorithmic mechanisms. I found that prioritizing the collective values of stakeholders in the initial stages of the algorithm design process enhances university diversity, empowers students to submit truthful rank-order lists, and establishes accountability mechanisms for universities. One of the main contributions in this thesis is a novel assessment of mechanism fairness, as defined by having: (1) an absence of justified envy, (2) an absence of lack of information, and (3) an absence of misalignment of values. Ethos, the core technical contribution in this thesis, consists of a machine learning-backed acceptance rate quiz for Ethiopian public universities, an informative portal for Ethiopian students, and a generalized student-university matching algorithm. My process for designing and evaluating Ethos consists of two in-person user studies (Study 1, n=33; Study 2, n=40) in which I identified and assessed the real-world impact of my system and algorithm parameters. I argue that listening to and prioritizing the collective values of stakeholders is critical to building diverse and fair algorithmic mechanisms and offer generalizable methods for doing so.},
}

EndNote citation:

%0 Thesis
%A Mulugeta, Liya 
%E Salehi, Niloufar 
%E Wagner, David A. 
%T Ethos: Designing Algorithmic Mechanisms for Increased Fairness, Stakeholder Empowerment, and Systemic Accountability
%I EECS Department, University of California, Berkeley
%D 2023
%8 December 1
%@ UCB/EECS-2023-294
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2023/EECS-2023-294.html
%F Mulugeta:EECS-2023-294