TL;DR
This paper introduces a new framework for understanding "spoilers" in party elections, moving beyond single-winner systems to analyze multi-district scenarios. They propose a quantifiable measure of spoiler impact and experimentally compare common electoral rules, finding that more proportional systems are less susceptible to spoiler effects.
What Problem Does This Paper Address?
In elections, a "spoiler" is typically a candidate whose presence changes the election outcome, often to the detriment of voters who would have preferred a different winner. While this concept is well-studied in single-winner elections, it's less clear how it applies to party elections, especially in multi-district systems where parties, not individuals, are allocated seats. The authors aim to generalize the definition of a spoiler to party elections and to quantify how susceptible different electoral rules are to this phenomenon. This is crucial for understanding the fairness and stability of electoral systems, particularly in contexts where multiple parties compete and seats are allocated across various districts.
Methods
The paper introduces a generalized definition of a spoiler for party elections: a party is a spoiler if its elimination causes another party's seat share to decrease. They then formalize this by defining "excess electoral impact," a measure of how much a party's removal distorts the payoff distribution beyond simply reallocating its own share. This allows for a nuanced understanding of "spoilership" as a matter of degree rather than a binary outcome.
For analyzing spoiler susceptibility, they develop probabilistic models for generating multi-district party elections, extending existing models used in computational social choice. These models simulate various voting behaviors and district configurations. They then experimentally evaluate several common apportionment methods (FPTP, Jefferson-D'Hondt, Webster-Sainte-Laguë, and Largest Remainders) by simulating elections under these models and calculating the average excess electoral impact for each rule.
Datasets
The paper does not use specific, pre-existing datasets. Instead, it generates synthetic election data using probabilistic models described in Section 7.1. These models simulate various voter preferences and district structures, allowing for controlled experiments to evaluate electoral rules.
Key Insights and Ideas
- Spoilers are a matter of degree: The paper moves beyond a binary definition of spoilers to a quantifiable measure of "excess electoral impact," allowing for a more nuanced analysis of how much a party's presence distorts outcomes.
- Proportionality is key to spoiler resilience: More proportional electoral rules (like Webster-Sainte-Laguë) are significantly less susceptible to spoiler effects than winner-take-all systems (like FPTP).
- Borda scores improve resilience: Using Borda scores (which consider ranked preferences) instead of just first-choice votes for apportionment substantially decreases spoiler susceptibility.
- District magnitude matters: Increasing the number of seats per district (committee size) generally reduces spoiler susceptibility.
- Identity is the only spoiler-proof rule: For "seats-votes" elections (where vote shares directly determine seat allocations), only the identity function (perfect proportionality) is spoiler-proof.
Key Results
The experimental results consistently show that spoiler susceptibility is negatively correlated with proportionality. The Webster-Sainte-Laguë rule, being the most proportional among those tested, exhibited the lowest spoiler susceptibility. Conversely, the First-Past-the-Post (FPTP) rule, a winner-take-all system, showed the highest susceptibility, with excess electoral impact sometimes reaching over 1.2 (meaning more than half the total payoff was redistributed). Switching from first-choice apportionment to Borda scores significantly reduced spoiler susceptibility for most rules, nearly eliminating differences between the more proportional methods. Spoiler susceptibility generally increased with the number of parties.
Limitations and Future Work
The paper acknowledges that their probabilistic models are simplifications of real-world elections and that some models might not perfectly capture all nuances of voter behavior. Future work is suggested to explore relaxations of spoiler-proofness, such as approximate spoiler-proofness, and to investigate spoiler susceptibility in non-seats-votes elections, including ordinal elections and participatory budgeting systems. They also plan to test their findings on other electoral rules like STV and k-Borda.
Summary updated