Perpetual Voting: Fairness in Long-Term Decision Making

Martin LacknerAAAI 2020 · DOI

Standard voting systems often leave minorities empty-handed, especially when decisions are made repeatedly over time. This research introduces perpetual voting rules, which leverage the history of past decisions to ensure that every group—even a small minority—exerts a proportional influence on long-term outcomes.

The Challenge of Temporal Fairness

In a typical majoritarian vote, a 51% majority can win 100% of the time. For example, if a group of five friends chooses a restaurant every week and three prefer French while two prefer Indian, a simple majority vote means they always eat French. Intuitively, a fair outcome would reflect the group's composition: French 60% of the time and Indian 40% of the time.

This paper addresses the challenge of achieving "temporal fairness" in complex scenarios where voters remain constant, but available choices and preferences fluctuate. The goal is to move beyond "winner-takes-all" dynamics and create a stable process where minorities remain incentivized to participate because they are guaranteed eventual satisfaction.

Methodology and Axiomatic Framework

The study formalizes perpetual voting using approval-based ballots. The core mechanism involves assigning evolving weights to voters based on decision history. Two primary rules are highlighted:

  1. Perpetual Consensus: Treats the process as a virtual consensus. After a decision, the weight of satisfied voters decreases significantly while the weight of all voters increases slightly, ensuring those who haven't won recently gain influence.
  2. Perpetual Quota: Tracks a "quota" for each voter—a running tally of influence they should have had—and prioritizes those whose actual satisfaction has fallen furthest behind.

To evaluate these rules, the author employs axiomatic analysis to prove mathematical adherence to fairness principles.

Table 1: Axiomatic properties of various perpetual voting rules. Perpetual Consensus is the only rule satisfying all three core axioms.

Simulation and Key Results

The rules were tested using a 2D Euclidean model with 20 voters (a 14-person majority and a 6-person minority) across 10,000 simulated instances. The results distinguish between mathematical rigor and average-case performance:

  • Perpetual Consensus is the theoretical "gold standard," being the only rule to satisfy simple proportionality, independence of unanimous decisions, and bounded dry spells.
  • Perpetual Quota emerged as the practical winner in simulations, achieving superior scores in lower quota compliance and the Gini influence coefficient, which measures the inequality of influence.
Figure 1: Gini influence coefficients across different rules. Lower values indicate more equal distribution of influence; Perpetual Quota and Perpetual Consensus significantly outperform standard Approval Voting (AV).

Key Insights

  • Memory is Essential: By tracking historical wins, rules can provide "dry spell" guarantees, mathematically ensuring a voter is satisfied within a specific number of rounds.
  • Robustness Against Manipulation: Rules must satisfy "Independence of Unanimous Decisions" to prevent groups from resetting their weights by adding uncontroversial items to the agenda.
  • Deterministic Superiority: Simulations showed that deterministic perpetual rules often outperform Random Serial Dictatorship (RSD) in the short-to-medium term.

Limitations and Future Work

The current model assumes a static voter population. Future adaptations must account for fluctuating populations (newcomers and leavers) without allowing strategic abstention. Additionally, while the paper focuses on selecting a single winner, future research could explore how perpetual voting might encourage middle-ground compromises rather than simple turn-taking.

Summary updated

BibTeX
@inproceedings{lackner2020perpetual,
  author = {Martin Lackner},
  title = {{Perpetual Voting: Fairness in Long-Term Decision Making}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2020},
  doi = {10.1609/aaai.v34i02.5584},
}
Figure