Ranked-Choice Failure Modes
Ranked-choice voting can punish a candidate for receiving extra support. That is not a slogan against it; it is a theorem-shaped failure mode in instant-runoff counting. Burlington's 2009 mayoral election is the stress test because the winner survived the elimination order while the candidate who would have beaten every major rival head-to-head was eliminated before the final round.
How the failure happens
Instant-runoff voting counts first choices, removes the last-place candidate, transfers those ballots, and repeats until one remaining candidate has a majority of non-exhausted ballots. The rule feels like a series of local repairs: remove the weakest option, then ask those voters for their next preference.
The trap sits in the word "weakest." A broadly acceptable compromise candidate can be everyone's second choice and still sit third on first choices. If that candidate is eliminated before the final pairing, the method never asks the head-to-head question that would have revealed the wider preference.
A monotonicity failure is stranger. In some ballot profiles, moving a candidate higher on some ballots can change the elimination order and make that candidate lose. The vote has the right direction at the ballot level, but the aggregate procedure bends it into the wrong direction.
Burlington, 2009
The Burlington count had three competitive candidates after smaller candidates were removed. First-choice totals were 2,951, 2,585, and 2,063 for the three major contenders. The 2,063-vote candidate was eliminated, even though pairwise records later showed that candidate would have beaten each major rival one-on-one.
| Test | What it asks | Burlington 2009 result |
|---|---|---|
| First-choice plurality | Who starts ahead? | 2,951-vote candidate |
| Instant runoff | Who survives transfers? | 2,585-vote candidate |
| Condorcet test | Who beats every rival head-to-head? | 2,063-vote candidate |
| Repeal vote | Did the city keep the method? | Repealed in 2010, 52% to 48% |
That is why the case keeps returning in voting-method debates. It was not a hypothetical profile from a textbook. It was a city election with published ballot data, a visible center squeeze, and a repeal vote one year later.
Failure modes worth separating
Ranked-choice voting is often sold as one fix for the spoiler problem, but the word "spoiler" hides several different defects. Some are reduced by instant runoff. Some remain. Some move to a different part of the count.
| Failure mode | Plain-English version | Does instant runoff avoid it? |
|---|---|---|
| Spoiler effect | A third option changes which major option wins | Sometimes |
| Condorcet failure | The head-to-head winner loses | No |
| Monotonicity failure | More support can hurt a candidate | No |
| Ballot exhaustion | A ballot stops counting after all ranked candidates lose | No |
| Center squeeze | A compromise candidate gets eliminated early | No |
The clean lesson is not "ranked-choice voting is bad." The clean lesson is narrower: instant runoff optimizes for majority among survivors, not majority preference across all pairwise contests.
What's contested
The dispute is partly empirical. Joseph Ornstein and Robert Norman estimated monotonicity failures under spatial models in Public Choice in 2014; later empirical surveys found observed failures to be uncommon in real American instant-runoff elections. "Uncommon" does not mean "irrelevant" when the failure can decide a mayoral race, but it does matter if the comparison is first-past-the-post rather than an ideal method.
The interpretive dispute is sharper. Supporters value majority support after transfers, fewer separate runoff elections, and less obvious vote-splitting. Critics value Condorcet consistency, monotonicity, and auditability. Arrow's impossibility theorem explains why this argument does not end cleanly: with three or more options, every voting rule gives up something.
Why this has to do with other realms
This page belongs beside concept arrow impossibility theorem because the Burlington case is Arrow with street names and ward-level ballots. A voting method is not a neutral pipe that carries public will from voters to winner; it is a compression algorithm that decides which information survives.
That also makes it a cousin of concept information theory. A ranked ballot contains more information than a single mark, but instant runoff discards pairwise information during elimination. The live design question is not whether more data exists. It is which data the counting rule is allowed to remember.
An open question
If every voting rule throws away some information, which loss should a city choose before the first ballot is cast?
Key Sources
- Eivind Stensholt, "What Happened in Burlington?" 2015. A mathematical reconstruction of the 2009 election.
- Kenneth J. Arrow, Social Choice and Individual Values, 1951. The impossibility frame behind the family of tradeoffs.
- City of Burlington, Vermont, official 2009 annual city election results. The ballot-count source behind the case.
- Joseph T. Ornstein and Robert Z. Norman, "Frequency of monotonicity failure under Instant Runoff Voting: estimates based on a spatial model of elections," Public Choice, 2014. DOI: https://doi.org/10.1007/s11127-013-0118-2
See Also
- concept arrow impossibility theorem
- concept information theory
- concept condorcet method
- concept first past the post
Abhishek's take
What grabs me here is not that ranked-choice voting fails. Every rule fails somewhere. The useful move is to name the failure before adopting the rule, because election design is a choice about which disappointment the public can understand after a close result.
Tags: #ranked-choice-voting #voting-systems #social-choice #election-design #condorcet