Close Races and Similarity Bias: A Bayesian Analysis of the Ranked Ballots for the Ballon d'Or
Abstract
The Ballon d'Or ranks players whose contributions have no common unit, so its results are often disputed. Regressions of award points describe associations but not how complete ballots arise or how likely a different winner would be. We propose a Bayesian mixture-of-experts model for partial rankings that learns nonlinear performance relationships, the number of voting styles and journalists' style memberships. A shared soft regression tree with style-specific leaf values lets styles evaluate the same season records differently on a common basis. Region-dependent mixing probabilities and journalist--nominee similarity terms separate geographic differences in voting styles from affinity within a style, and journalist- and rank-specific precisions capture remaining variation. We establish identification and consistency for the number of occupied styles under explicit conditions; simulations examine recovery, calibration and robustness to misspecification. Applied to all 176 published ballots from 2019 and 99 from 2024, the model compares award outcomes under changes in affinity and jury composition. For 2019, Messi's probability of winning with a new jury from the same countries rises from 0.49 to 0.76 when similarity terms are removed. For 2024, the direction of this change depends on whether affinities differ by region. Regional representation changes winner probabilities more than jury size.
Related
- Anchor-proofness in Voting with Ranked Ballots
- Semantic Similarity‐Informed Bayesian Borrowing for Quantitative Signal Detection of Adverse Events
- Divergence based Ranked Preference Alignment for LLMs
- Ranked Right, Sized Wrong: Ensemble Spread Fails Silently at Streamflow Extremes
- Single-Temperature Ranked Greedy Cluster Exchange for Max-Cut on Two-Dimensional Tori
BibTeX
@misc{hu2026close,
title = {{Close Races and Similarity Bias: A Bayesian Analysis of the Ranked Ballots for the Ballon d'Or}},
author = {Guanyu Hu and Xiao Wang},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.36175},
eprint = {2610.36175},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.36175},
note = {CC BY-NC-SA 4.0},
}
RIS
TY - UNPB
TI - Close Races and Similarity Bias: A Bayesian Analysis of the Ranked Ballots for the Ballon d'Or
AU - Guanyu Hu
AU - Xiao Wang
PY - 2026
DA - 2026/10/26
PB - arXiv
DO - 10.48550/arxiv.2610.36175
UR - https://arxiv.org/abs/2610.36175
AB - The Ballon d'Or ranks players whose contributions have no common unit, so its results are often disputed. Regressions of award points describe associations but not how complete ballots arise or how likely a different winner would be. We propose a Bayesian mixture-of-experts model for partial rankings that learns nonlinear performance relationships, the number of voting styles and journalists' style memberships. A shared soft regression tree with style-specific leaf values lets styles evaluate the same season records differently on a common basis. Region-dependent mixing probabilities and journalist--nominee similarity terms separate geographic differences in voting styles from affinity within a style, and journalist- and rank-specific precisions capture remaining variation. We establish identification and consistency for the number of occupied styles under explicit conditions; simulations examine recovery, calibration and robustness to misspecification. Applied to all 176 published ballots from 2019 and 99 from 2024, the model compares award outcomes under changes in affinity and jury composition. For 2019, Messi's probability of winning with a new jury from the same countries rises from 0.49 to 0.76 when similarity terms are removed. For 2024, the direction of this change depends on whether affinities differ by region. Regional representation changes winner probabilities more than jury size.
ER -