On the Distortion Value of the Elections with Abstention

Mohammad Ghodsi, Mohamad Latifian, Masoud Seddighin

Abstract: In Spatial Voting Theory, distortion is a measure of how good the winner is. It is proved that no deterministic voting mechanism can guarantee a distortion better than 3, even for simple metrics such as a line. In this study, we wish to answer the following question: how does the distortion value change if we allow less motivated agents to abstain from the election? We consider an election with two candidates and suggest an abstention model, which is a more general form of the abstention model proposed by Kirchgässner. We define the concepts of the expected winner and the expected distortion to evaluate the distortion of an election in our model. Our results fully characterize the distortion value and provide a rather complete picture of the model.

Introduction: The Problem of Distortion

In Social Choice Theory, we often evaluate voting mechanisms using a metric called distortion. This concept assumes that voters and candidates are points in a metric space (such as a 1D political spectrum or a multi-dimensional map of issues). A voter’s "cost" for a candidate is the distance between them. The socially optimal candidate is the one who minimizes the total distance to all voters (the social cost).

However, most voting rules only have access to ordinal preferences (e.g., "Voter 1 prefers Candidate A over Candidate B") rather than the exact distances. This loss of information leads to a theoretical gap: deterministic voting rules can produce winners that are up to 3 times worse than the optimal choice in the worst-case scenario. This study challenges the realism of the classic "distortion = 3" result by accounting for voter abstention.

Modeling Voter Abstention

The authors argue that the classic model is unrealistic because it assumes every voter participates, regardless of how indifferent they are or how far they feel from the political process. To address this, they employ a generalized version of the Kirchgässner model, which introduces a participation parameter $\beta \in [0, 1]$.

The participation function for different values of β, showing how indifference and alienation affect voting probability across the spectrum.

As shown in the figure above, the probability of a voter casting a ballot depends on their relative distance to the candidates. This captures two critical real-world behaviors:

  1. Indifference: If both candidates are roughly equidistant from a voter, that voter is less likely to vote.
  2. Alienation: If both candidates are very far away from a voter's own position, the voter feels alienated and is less likely to participate.

When $\beta=0$, the model reverts to the classic case where everyone votes. As $\beta$ increases, the "dips" in the participation curve become more pronounced, meaning ideological distance plays a larger role in the decision to stay home.

Methodology: Valid Displacements

To find the worst-case distortion across all possible voter distributions, the authors utilize a technique called "valid displacements." Analyzing every possible configuration of voters in a metric space is mathematically daunting. However, the authors prove that any complex distribution can be simplified without decreasing the distortion.

Illustration of valid displacements, a technique used to simplify complex voter distributions into specific points to find worst-case distortion.

By moving voters within specific regions (labeled A, B, C, and D in the figure) toward the candidates or the midpoint, the researchers can reduce the problem to a solvable mathematical optimization program. This approach demonstrates that the worst-case scenarios on a simple 1D line are actually representative of the worst-case scenarios for any metric space that satisfies the triangle inequality.

Key Results

The study reveals that abstention acts as a natural filter for "noisy" voters. By removing those who lack a strong preference, the remaining electorate drives the outcome toward a more socially optimal winner.

  • Significant Improvement: While distortion is 3 when everyone votes ($\beta=0$), it drops significantly as abstention increases. At $\beta=1$, the worst-case distortion is approximately 1.522.
  • The Sweet Spot: The authors identified a minimum possible worst-case distortion of $\sqrt{2} \approx 1.414$, occurring when the participation parameter $\beta$ is approximately 0.705.
  • Scalability: In large elections, the "expected distortion" is shown to be very close to the distortion of the "expected winner," suggesting these bounds are highly reliable for real-world political systems.

Limitations and Future Work

While these results provide a complete picture for two-candidate elections, several challenges remain. The current model assumes "sincere" voting—where voters choose the closest candidate if they vote at all—and does not account for strategic tactical voting. Furthermore, generalizing these findings to multi-candidate elections (using rules like Borda count or Copeland) remains an open and complex problem, as the geometry of indifference and alienation becomes significantly more intricate with three or more candidates.

Summary updated

Abstract

In Spatial Voting Theory, distortion is a measure of how good the winner is. It is proved that no deterministic voting mechanism can guarantee a distortion better than $3$, even for simple metrics such as a line. In this study, we wish to answer the following question: how does the distortion value change if we allow less motivated agents to abstain from the election? We consider an election with two candidates and suggest an abstention model, which is a more general form of the abstention model proposed by Kirchgassner. We define the concepts of the expected winner and the expected distortion to evaluate the distortion of an election in our model. Our results fully characterize the distortion value and provide a rather complete picture of the model.
BibTeX
@inproceedings{ghodsi2019distortion,
  author = {Mohammad Ghodsi and Mohamad Latifian and Masoud Seddighin},
  title = {{On the Distortion Value of the Elections with Abstention}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2019},
  doi = {10.1609/aaai.v33i01.33011981},
}
Figure