[Paper Review] Intersectional Affirmative Action Policies for Top-k Candidates Selection

Giorgio Barnabò, Carlos Castillo+2 more|arXiv (Cornell University)|Jul 29, 2020

LGBTQ Health, Identity, and Policy38 references4 citations

TL;DR

This paper proposes a novel algorithmic framework for top-k candidate selection that integrates intersectional affirmative action to reduce disparities among disadvantaged subgroups defined by overlapping attributes (e.g., race, gender, socio-economic status), while maintaining high utility in selected candidates. Using dynamic programming and greedy optimization, the method adjusts admission thresholds to achieve near-parity in acceptance rates across intersectional classes with only a minimal decrease in average candidate aptitude—demonstrated via real university admissions data from an OECD country.

ABSTRACT

We study the problem of selecting the top-k candidates from a pool of applicants, where each candidate is associated with a score indicating his/her aptitude. Depending on the specific scenario, such as job search or college admissions, these scores may be the results of standardized tests or other predictors of future performance and utility. We consider a situation in which some groups of candidates experience historical and present disadvantage that makes their chances of being accepted much lower than other groups. In these circumstances, we wish to apply an affirmative action policy to reduce acceptance rate disparities, while avoiding any large decrease in the aptitude of the candidates that are eventually selected. Our algorithmic design is motivated by the frequently observed phenomenon that discrimination disproportionately affects individuals who simultaneously belong to multiple disadvantaged groups, defined along intersecting dimensions such as gender, race, sexual orientation, socio-economic status, and disability. In short, our algorithm's objective is to simultaneously: select candidates with high utility, and level up the representation of disadvantaged intersectional classes. This naturally involves trade-offs and is computationally challenging due to the the combinatorial explosion of potential subgroups as more attributes are considered. We propose two algorithms to solve this problem, analyze them, and evaluate them experimentally using a dataset of university application scores and admissions to bachelor degrees in an OECD country. Our conclusion is that it is possible to significantly reduce disparities in admission rates affecting intersectional classes with a small loss in terms of selected candidate aptitude. To the best of our knowledge, we are the first to study fairness constraints with regards to intersectional classes in the context of top-k selection.

Motivation & Objective

• To address algorithmic unfairness in top-k selection processes where individuals from multiple disadvantaged groups face compounded discrimination.
• To design a data-driven policy that reduces disparities in admission rates for intersectional classes without significantly lowering the average aptitude of selected candidates.
• To model and solve the combinatorial challenge of managing fairness across exponentially growing subgroups defined by intersecting attributes.
• To evaluate the trade-off between fairness (representation parity) and utility (candidate aptitude) in real-world university admissions systems.
• To provide a principled, scalable algorithmic framework applicable to high-stakes selection processes such as college admissions or job hiring.

Proposed method

• Formulates a constrained optimization problem that balances utility maximization and fairness across all intersectional subgroups defined by protected attributes.
• Introduces a weighted objective function that combines total utility and discrepancy in admission rates across intersectional classes, with a tunable trade-off parameter λ.
• Develops a dynamic programming algorithm to compute the optimal threshold adjustment for achieving fairness-utility trade-offs with optimality guarantees.
• Proposes a greedy heuristic as a scalable alternative to the dynamic programming approach, trading off optimality for significant speed improvements.
• Adapts the algorithms to exclude intersectional classes with fewer than three applicants to ensure statistical robustness in small subgroups.
• Employs real-world university admission data from an OECD country, including standardized test scores, high school grades, and demographic attributes, to simulate and evaluate the algorithms.

Experimental results

Research questions

• RQ1 Can intersectional affirmative action policies significantly reduce disparities in admission rates for subgroups defined by overlapping protected attributes?
• RQ2 To what extent can fairness be improved across intersectional classes without incurring a substantial loss in the average aptitude of selected candidates?
• RQ3 How do different algorithmic approaches (dynamic programming vs. greedy) compare in terms of fairness, utility, and computational efficiency in real-world settings?
• RQ4 What is the impact of applying affirmative action at the program level versus a centralized single-track system on fairness and utility trade-offs?
• RQ5 How stable are the fairness outcomes when intersectional classes have small sample sizes in the applicant pool?

Key findings

• The proposed algorithms achieve near-parity in admission rates across all 12 intersectional classes—defined by socio-economic status, high school type, and regional development—with only a 1-2 point average decrease in admission scores.
• In the single-track setting, a utility decrease of just one or two points leads to a dramatic reduction in average admission rate discrepancy, with the y-axis in Figure 4 showing a logarithmic decline in disparity.
• The greedy algorithm is orders of magnitude faster than the dynamic programming approach, though it only failed to find the optimal solution in one configuration (single-track, 30% admission rate, λ=0).
• For individual programs, the required utility sacrifice to achieve fairness is smaller than in the centralized setting, due to lower baseline disparities in admission rates.
• In individual program applications, only the most privileged (e.g., 1Aa) and most disadvantaged (e.g., 3Bb) groups typically see changes in admission scores and discrepancies, indicating targeted fairness adjustments.
• The method remains robust even when excluding intersectional classes with fewer than three applicants, ensuring stable and meaningful fairness outcomes across diverse real-world scenarios.