Metric Distortion in Ordinal Social Choice

477 concepts200 papers1 proposal

The Information Bottleneck in Voting

Humans are excellent at ranking things ("I prefer candidate $A$ over candidate $B$"), but they are notoriously bad at assigning precise numerical values to those preferences ("I like candidate $A$ exactly 2.34 times more than candidate $B$"). Because of this cognitive burden, democratic and algorithmic systems almost exclusively rely on ordinal preferences (rankings) rather than cardinal utilities (numerical values), as noted by Gerdus Benadè et al. in Low-Distortion Social Welfare Functions.

However, this convenience introduces a profound information gap. Imagine an election with three voters and two candidates, $A$ and $B$. All three voters rank $A$ above $B$. An ordinal voting system sees a unanimous mandate for $A$. But what if the voters are nearly indifferent, sitting ideologically close to both candidates, while a third unconsidered option would have served them vastly better? Or worse, what if they prefer $A$ by a hair, but passionately detest $B$? By flattening preference intensities into rigid ranks, the system becomes blind to the true scale of voter utility.

The Metric Framework and Distortion

To quantify what is lost in this gap, social choice theorists use a spatial model: the metric preference framework. In this model, voters and candidates are embedded as points in a hidden metric space (such as a multi-dimensional political or preference landscape). The "cost" a voter incurs from a candidate is simply the geometric distance between them. The ultimate utilitarian goal is to select a candidate that minimizes the total social cost—the sum of distances from all voters.

Because a voting mechanism only observes the ordinal rankings induced by these distances, it cannot know the exact geometry of the space. The worst-case performance penalty of this blindness is called distortion: the ratio of the social cost of the chosen winner to the social cost of the true, omnisciently optimal candidate.

As formalized by Elliot Anshelevich et al. in Approximating Optimal Social Choice under Metric Preferences, distortion acts as a competitive ratio against nature. If a voting rule has a distortion of 3, it guarantees that no matter how adversarial the underlying metric space is, the selected candidate will never inflict more than three times the optimal social cost.

Why Some Voting Rules Fail (and Others Succeed)

Not all ordinal rules handle this information gap equally. Common positional scoring rules like Plurality (which only looks at top choices) or Borda Count (which assigns linear weights to ranks) perform poorly in the worst case. Piotr Skowron et al. demonstrated in Social Choice Under Metric Preferences: Scoring Rules and STV that no standard scoring rule can achieve a constant distortion bound as the number of candidates grows. Fixed weights cannot adapt to the arbitrary stretching and squeezing of distances in a metric space.

In contrast, tournament-based or iterative rules perform remarkably well. Head-to-head matchups implicitly exploit the triangle inequality inherent to metric spaces: if a majority prefers $A$ to $B$, the geometric properties of the space dictate that $A$ cannot be too far from the center of the voter distribution. Utilizing this property, Elliot Anshelevich et al. proved that the Copeland rule—which matches candidates head-to-head and selects the one with the most wins—achieves a tight, constant distortion of exactly 3.

To systematically analyze these bounds, David Kempe et al. introduced a framework based on linear programming duality in An Analysis Framework for Metric Voting based on LP Duality. By treating the worst-case metric embedding as a dual flow problem, the authors provided a unified toolkit to bound distortion and evaluate complex, multi-stage rules like Single Transferable Vote (STV).

Bridging the Gap

How can we bypass the fundamental limits of purely ordinal voting without reintroducing heavy cognitive burdens? Researchers have uncovered several powerful levers:

  • Representative Candidates: In many real-world scenarios, candidates aren't adversarial outliers; they are drawn from the same population as the voters. Yu Cheng et al. demonstrated in Of the People: Voting Is More Effective with Representative Candidates that when candidates are representative of the electorate, the expected distortion drops significantly, allowing even simple majority rules to approximate the social optimum effectively.
  • Eliciting a Drop of Passion: We do not need full cardinal utilities to fix the gap. Ben Abramowitz et al. showed in Awareness of Voter Passion Greatly Improves the Distortion of Metric Social Choice that injecting just a single bit of preference intensity data—asking voters if they prefer candidate $A$ over $B$ by more than a factor of two—drastically collapses the worst-case distortion.
  • The Communication Frontier: This points to a fundamental trade-off between communication complexity (how many bits voters send) and efficiency. As mapped out by Debmalya Mandal et al. in Efficient and Thrifty Voting by Any Means Necessary, we can design novel, "thrifty" voting mechanisms that sit on a Pareto frontier, minimizing distortion while keeping the elicitation process highly manageable.

Beyond Single Winners

The ordinal-cardinal gap becomes even more intricate when selecting committees or full rankings rather than a single winner. In The Metric Distortion of Multiwinner Voting, Ioannis Caragiannis et al. showed that multiwinner distortion exhibits a sharp trichotomy: if voters only derive utility from a tiny fraction of a selected committee, ordinal data becomes completely blind, causing the metric distortion to become unbounded.

Furthermore, researchers are rethinking how we measure distortion itself. While traditional work focuses on multiplicative ratios, Anson Kahng et al. argued in Worst-Case Voting When the Stakes Are High that multiplicative metrics can be overly pessimistic in low-stakes scenarios where all candidates are reasonably close. By shifting the analytical lens to additive distortion (regret), the authors designed rules that prioritize high-stakes decisions, ensuring robust social welfare when the cost of making a wrong choice is genuinely severe.

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 210 papers
  2. Version 1 · 2026-10-04 · 123 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_ordinal_cardinal_information_gap_7b627a02,
  title = {ordinal-cardinal information gap},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/ordinal-cardinal-information-gap/rdr_7b627a02c1e441189f63a0f4579162a8},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 2, frozen text):

@misc{lacuna_ordinal_cardinal_information_gap_7b627a02_v2,
  title = {ordinal-cardinal information gap},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/ordinal-cardinal-information-gap/rdr_7b627a02c1e441189f63a0f4579162a8/v/2},
  note = {Version 2, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake