At least two studies have been made analyzing the results of IRV elections, and the individual ballot data, to see how well they perform regarding the promises of IRV: to elect the majority candidate even with more than two candidates in the race, to eliminate the spoiler effect, and to give voters the freedom to support third-party and independent candidates. One study is Song (this volume) that considered 172 IRV elections with more than two candidates. Another is Sarwate, et al. (2013) that considered 37 IRV elections. Both studies have found that the IRV method (Hare) of tallying the ballots and identifying the winner had elected the “Condorcet Winner” in nearly all elections. The sole exception[1] is the 2009 mayoral election in Burlington Vermont.
A Condorcet Winner (named after the Marquis de Condorcet, an 18th century French mathematician and philosopher) is the candidate who, from the ranking data expressed on ranked-order ballots, defeats every other candidate when paired with them in head-to-head runoffs. The head-to-head runoff is exactly what happens in the IRV final round.
It should not be surprising that the Condorcet Winner wins in almost all IRV elections, because the Condorcet Winner must have some base support just to achieve all those head-to-head wins. And all the Condorcet Winner needs to do to be the IRV winner is to get into the IRV final round; the Condorcet Winner will always win that final round.
For any candidate other than the Condorcet Winner to be elected (as in the sole exception of Burlington 2009), it is necessary that Principles 1 and 2 above (One-person-one-vote and Majority rule) be violated. Furthermore, I show that Properties 3 and 4 above must also be violated in such an election.
The 2009 Burlington IRV election illustrates this perfectly. I compiled the data in Tables 1 and 2 from public files obtained from the Burlington City Clerk. My analysis was replicated by Gierzynski et al. (2009) and Olson (2009). These analyses show that candidate Andy Montroll was preferred to Kurt Wright by a margin of 933 voters and that Andy Montroll was preferred to Bob Kiss by a margin of 588 voters, yet the IRV final round was between Wright and Kiss, with Kiss being preferred (to only Wright, not Montroll) by a margin of just 252 voters (out of 8374). Why was the final round contested between candidates Wright and Kiss when the ballot data indicates that candidate Montroll would have defeated either of them in the final round?
Candidate A (Andy) was preferred, as expressed explicitly on their ballots, by a simple majority of Burlington voters over Candidate B (Bob), yet Candidate B was elected to office. The 3476 voters that preferred Bob had votes that counted more than those of the 4064 voters that preferred Andy. These are not equally valued votes nor is this majority rule and as such is a failure of democracy.
Although he ran to win and was a serious contender, Kurt Wright was effectively a spoiler; a candidate who loses an election but by his presence changes who the winner is. Had Kurt Wright not run and the same Burlington electorate come to the polls and expressed their same preferences over the remaining four candidates, the election outcome would have been different, with Andy Montroll prevailing in the final round over Bob Kiss by a margin of 588 votes.
Of the voters preferring Wright in the semifinal round, the largest group were 1510 voters who marked Montroll as their second choice and preferred Kiss not at all. As Table 1 shows, if 371 (less than one in four) or more of these voters had anticipated that their guy was not going to win and had voted tactically, this voting tactic being “compromising,” they would have prevented the election of Bob Kiss, the candidate they disliked the most. Or if 587 of those voters, along with 154 preferring only Wright, had just stayed home and not come to the polls at all, they would have prevented the election of the candidate they disliked the most.
Except that Ranked-Choice Voting was used, this is hardly different than what happens with Progressive or Green Party voters who compromise and vote for the Democrat out of fear of helping elect the GOP candidate they loathe. IRV promised these voters that they could “Vote their hopes rather than vote their fears”.
But these conservative voters in Burlington found out otherwise: “In this liberal town I gotta choose between 'Liberal' or 'More-Liberal', because if I vote for the guy I really like then
'More-Liberal' gets elected!” That has got to make some people angry. Simply by marking their sincere favorite choice as #1, they literally caused the election of their most disliked candidate.
Recently, former Vermont governor, presidential candidate, and longtime Burlington resident Howard Dean, in promoting re-adoption of Hare RCV, mistakenly claimed “you can still get your second-choice vote” (Lamdin 2021). That promise was not kept with the 1510 Wright voters who disliked Kiss and caused the election of Kiss simply by marking Wright as #1. Their first choice was defeated and their second-choice vote was not counted. If those second-choice votes had been counted, a different candidate for mayor would have been elected. The following year, IRV was repealed in Burlington Vermont.
An additional desirable property of a voting system is:
5. Precinct summability: Transporting ballot information from precincts to the central tallying location (City Hall) in an opaque manner is viewed as less transparent than having each precinct report subtotals for each race to the media and to the campaigns for them to also tally and audit the election results. While plurality is precinct summable and precincts report their vote subtotals, IRV is not. However, a Condorcet consistent RCV method can, for each precinct, report vote subtotals for each pairing of candidates. These are then summable by outside parties. Condorcet consistent RCV methods preserve this salient property of process transparency that IRV does not.
IRV in Burlington in 2009 did not elect the candidate that was preferred by simple majorities over all of the other candidates. IRV did not protect against the spoiler effect and, after
Table 2: Six Pairings, Showing the Pairwise Tallies for the Four Significant Candidates,
The tallies for the Kiss-Wright pairing are consistent with the results of the IRV final round. Note that only candidate Montroll defeats every other candidate he faces. Because Condorcet-consistent RCV is precinct summable, these figures can be obtained by totaling the corresponding pair subtotals for all precincts.
|
Montroll
|
4064
|
M>K
|
Montroll
|
4570
|
M>S
|
Montroll
|
4597
|
M>W
|
Kiss
|
3476
|
K>M
|
Smith
|
2997
|
S>M
|
Wright
|
3664
|
W>M
|
|
̶ ̶ ̶ ̶ ̶ ̶ ̶ ̶ ̶
|
|
|
̶ ̶ ̶ ̶ ̶ ̶ ̶ ̶ ̶
|
|
|
̶ ̶ ̶ ̶ ̶ ̶ ̶ ̶ ̶
|
|
Margin
|
588
|
|
Margin
|
1573
|
|
Margin
|
933
|
|
|
|
|
|
|
|
|
|
|
Kiss
|
3944
|
K>S
|
Kiss
|
4313
|
K>W
|
Wright
|
3971
|
W>S
|
Smith
|
3576
|
S>K
|
Wright
|
4061
|
W>K
|
Smith
|
3793
|
S>W
|
|
̶ ̶ ̶ ̶ ̶ ̶ ̶ ̶ ̶
|
|
|
̶ ̶ ̶ ̶ ̶ ̶ ̶ ̶ ̶
|
|
|
̶ ̶ ̶ ̶ ̶ ̶ ̶ ̶ ̶
|
|
Margin
|
368
|
|
Margin
|
252
|
|
Margin
|
178
|
|
promising not to, IRV punished a large group of voters (one sixth of the electorate) for simply ranking their favorite choice first, thus discouraging sincere non-tactical voting. These facts are indisputable and are supported by the public record.
Furthermore, IRV is not precinct summable, thus requiring the opaque transporting of voting data from precincts to a central tallying location and the tallying of votes and identification of the winner to be done only at that central location.
It is this failure that happened in 2009 that we should not repeat. Particularly in the City of Burlington that had the only government election employing IRV in which IRV elected the wrong candidate, someone other than the Condorcet Winner.
[1] In August 2022, the Special General Election for U.S. Congress in Alaska demonstrated another example of an RCV election in which the Condorcet winner was not elected.