Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..

Strategic Costs of Perceived Bias in Fair Selection

Authors: L. Elisa Celis, Lingxiao Huang, Milind Sohoni, Nisheeth K. Vishnoi

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We simulate this dynamics under the setting p1 = Unif[0, 1], p2 = Unif[0, ρ = 0.8], pa = δ0, with c = 0.2, α = 0.5, and n = 20, 200, 600, 1200, running 500 iterations in each case. Figure 1 shows representative results; full plots are in Figures 5 8.
Researcher Affiliation Academia L. Elisa Celis Yale University Lingxiao Huang Nanjing University Milind Sohoni IIT Bombay Nisheeth K. Vishnoi Yale University
Pseudocode Yes Algorithm 1 Dynamics for Computing Finite NE Policies for the Uniform Distribution Case
Open Source Code Yes 5. Open access to data and code Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: See supplemental material.
Open Datasets Yes To demonstrate real-world applicability, we calibrate the model using genderdisaggregated data from JEE Advanced 2024, a highly competitive entrance exam for India s IITs. Of 180,200 candidates, 139,180 were male and 41,020 female; 40,284 males and 7,964 females qualified. This yields admit rates of 28.9% for males and 19.4% for females, giving an observed representation ratio of robs 0.671. The overall selection rate is c 0.268, and the female applicant fraction is α 0.228. These values anchor our analysis of strategic disparities and potential interventions. In Section G.1, under the uniform density setup in Proposition 4.1, we compute ρ 0.882 using the explicit form of robs = r R(A) = 1 (1 c)(1 ρ) [68] Press Information Bureau, Government of India. JEE (Advanced) 2024: Gender-wise statistics of candidates registered and qualified. https://pib.gov.in/Press Release Page.aspx?P RID=2023653, 2024. Accessed: 2025-05-07.
Dataset Splits No The paper uses gender-disaggregated data from JEE Advanced 2024, providing counts for male and female candidates and qualified individuals. This describes the observed data, but does not specify any training/test/validation splits for machine learning experiments.
Hardware Specification No The paper does not explicitly mention any specific hardware (e.g., GPU models, CPU types, or cloud computing instances with specifications) used for conducting its simulations or analyses.
Software Dependencies No The paper describes simulation parameters and an algorithm, but does not specify any particular software libraries, frameworks, or their version numbers (e.g., Python, PyTorch, CUDA, etc.) that were used.
Experiment Setup Yes In our simulations, we set the valuation resolution mv = 101, effort resolution me = 101, total number of iterations T = 500, and step sizes a(t) 1 = a(t) 2 = 1 10T. We always set n1 = n2 = n 2 , which means α = 0.5.