TL;DR
This paper investigates how allowing voters to abstain from elections affects the "distortion value," a measure of how close the elected winner is to the ideal outcome in spatial voting models. The authors introduce a new abstention model and show that allowing abstention can significantly reduce distortion, particularly in large elections, and they provide bounds on this reduction.
What Problem Does This Paper Address?
In social choice theory, a key challenge is designing voting mechanisms that aggregate individual preferences into a collective decision. A common way to evaluate these mechanisms is through the lens of "distortion," which quantifies how far the winning candidate is from the socially optimal candidate (the one minimizing total voter cost). Previous work has shown that even in simple scenarios (like voters and candidates on a line), no deterministic voting mechanism can guarantee a distortion better than 3. This paper asks: what happens to this distortion if voters are allowed to abstain? Abstention is a realistic aspect of real-world elections, and this work aims to understand its impact on the efficiency of voting systems, specifically in a spatial model where voters and candidates are points in a metric space.
Methods
The authors model abstention using a probabilistic approach based on a voter's perceived distance to candidates. Their model generalizes existing abstention models by incorporating both indifference-based abstention (when voters are equally distant from candidates) and alienation-based abstention (when voters are far from their preferred candidate). They define "expected winner" and "expected distortion" to analyze election outcomes under this probabilistic abstention. A core part of their methodology involves "valid displacements," a technique to transform elections by moving voters while preserving the expected winner and not increasing distortion, allowing them to simplify election structures for analysis. They then derive upper bounds on the distortion value for various scenarios, including general metric spaces.
Datasets
Not discussed in the paper. The paper focuses on theoretical analysis and does not use empirical datasets.
Key Insights and Ideas
- Abstention as a Distortion Reducer: The most significant insight is that allowing voters to abstain can substantially decrease the distortion value, especially in larger elections. This suggests that abstention, often seen as a problem, can actually improve the efficiency of voting outcomes in certain contexts.
- A Generalized Abstention Model: The paper introduces a flexible abstention model that captures both indifference and alienation, providing a more nuanced understanding of voter behavior than previous models.
- The Role of the "Participation Parameter" (β): The parameter β in their abstention model quantifies how much relative distance matters for a voter's decision to vote. Lower β values mean voters are more likely to participate, while higher β values reflect greater sensitivity to relative distances, leading to more abstention.
- Structural Simplification for Analysis: The use of "valid displacements" is a clever technique to reduce complex election scenarios to simpler forms (e.g., voters located at only two points), making the analysis of distortion more tractable.
- Metric Space Generalization: The findings for the line metric are shown to extend to more general metric spaces, indicating the robustness of their results.
Key Results
- For the specific case of Kirchgässner's abstention model (β=0), the distortion is upper-bounded by 1.522.
- For large elections and any participation parameter β, the expected distortion is upper-bounded by (1 + 2α)D, where D is the distortion of the expected winner and α is a parameter related to the ratio of expected votes.
- The paper establishes that the upper bounds derived for the line metric also hold for arbitrary metric spaces.
- The analysis reveals that the worst-case distortion initially decreases as the participation parameter β increases from 0, then increases, with a minimum distortion around β ≈ 0.705.
Limitations and Future Work
- Small Number of Voters: The paper acknowledges that analyzing the expected distortion for a small number of voters remains an open question, with a counter-example suggesting an upper bound of 2 for β=1.
- Closed-Form Expression for Expected Winner Distortion: Finding a precise closed-form expression for the distortion of the expected winner is still an open problem, as current methods involve solving a non-convex mathematical program.
- More Than Two Candidates: The current work focuses on two candidates. Future research could extend the analysis to scenarios with more candidates and explore various voting mechanisms (Borda, k-approval, etc.) under abstention.
- Candidate Representativeness: The paper suggests investigating how abstention affects distortion in societies where candidates are highly representative of voters, drawing parallels to existing work showing improved distortion in such cases without abstention.
Summary updated