Abstract
Instant Runoff Voting (RCV) is designed to capture nuanced voter preferences, yet many jurisdictions impose arbitrary limits on the number of candidates a voter can rank. This paper examines the impact of ballot length ($h$) on election outcomes. We demonstrate that ballot length is a significant variable: theoretically, changing $h$ can result in up to $k-1$ different winners from the same preference profile. Empirically, we find that 25% of real-world elections analyzed could have seen different outcomes if ballot lengths were restricted.
Introduction: The Arbitrary Limit of Ranked Choice
Instant Runoff Voting (IRV), or Ranked-Choice Voting, is gaining momentum across the United States, implemented in jurisdictions from Alaska to New York City. While the ideal IRV model assumes voters rank all available candidates, practical constraints often lead election officials to limit the ballot length ($h$). For instance, Minneapolis restricts voters to three rankings, whereas New York City allows five.
This paper addresses whether these arbitrary limits fundamentally alter election results. We explore the "messy middle" between plurality voting ($h=1$) and full IRV ($h=k$), investigating the structural constraints and volatility of winner sequences as ballot length increases.
Methodology
The study employs a three-pronged approach:
- Theoretical Construction: Proving the existence of preference profiles where the winner changes nearly every time $h$ increases.
- Linear Programming (LP): Identifying the minimal number of voters required to produce these pathological scenarios.
- Empirical Analysis: Analyzing 168 real-world elections from the PrefLib database, including APA presidential elections and local government races. Researchers "truncated" these historical ballots post-hoc to observe how outcomes would shift at different lengths.
Key Insights: The Mechanism of Volatility
The primary driver of winner shifts is ballot exhaustion. When a voter's ranked candidates are eliminated, their vote is removed from the count. Shorter ballots increase exhaustion, preventing votes from being reallocated to later-round frontrunners.
Key findings include:
- Theoretical Volatility: For $k$ candidates, it is theoretically possible to have $k-1$ different winners simply by adjusting the ballot length.
- The $k/2$ Stability Threshold: In spatial models (1D ideological spectrums), outcomes tend to stabilize once the ballot length exceeds half the number of candidates ($h > k/2$).
- Structure Matters: While "single-peaked" preferences reduce volatility, they do not eliminate it, still allowing for $\Omega(\sqrt{k})$ distinct winners.
Results and Real-World Impact
The theoretical bound for $k-1$ distinct winners requires at least $2k^2 - 2k$ voters. While such extreme cases are rare in practice, the impact of truncation is significant:
- 25% of real-world elections studied would have changed winners if the ballot length had been shorter.
- In the 2009 Burlington Mayoral election ($k=6$), a ballot length of $h=2$ would have resulted in a toss-up between the plurality winner and the eventual IRV winner.
- The maximum number of distinct winners observed in real-world data across different lengths was three, occurring in only one instance.
Conclusion
Ballot length is a consequential degree of freedom in IRV design. While pathological cases of extreme volatility are rare, the choice of $h$ can and does flip election outcomes. Future research should focus on identifying an "optimal" length that balances administrative simplicity with outcome stability, and how strategic voter behavior might respond to these constraints.
Summary updated