Its advocates claim that "With Ranked Choice Voting, you can honestly rank candidates without having to worry about how others will vote and who is more or less likely to win" because if your first choice can't win, your second choice will be counted.
In truth, Ranked Choice only counts the secondary preferences of some of the voters who put a non-winning candidate in first position, and this can and does lead to non-representative, non-majoritarian outcomes. The Burlington 2009 Mayoral election is a textbook example of such a sideways election, and its fallout contributed to Burlington's repeal of RCV the next year.
The election summary
Three strong candidates, Andy (A), Bob (B) and Kurt (C), faced off after the first elimination round. The full preference table* on the right shows that a plurality of the voters (4,067 vs 3,477) preferred A to B, and a majority (4,597 vs. 3,668) preferred A to C. Because A was preferred head-to-head over both B and C, A was the "beats-all" candidate, also known as the Condorcet Winner.
8,980 ballots were cast in total, 151 of which were exhausted after the first round. Despite the electorate's preference for A over both B and C, the instant runoff counting algorithm eliminated the "beats-all" candidate A before either B or C, and elected B in the final round without a majority (4,313 of 8,980 total ballots).
Why did this happen?
IRV only counts the second and subsequent preferences of the voters whose prior higher preference choices have been eliminated.
This means that the subsequent preferences of the voters who chose B or C first, highlighted in red, were not counted by the IRV tabulation algorithm. Those voters preferred A second by strong super majorities over the alternative. This counting flaw was particularly bad for those who placed C first: they had A as a second choice by more than a 3:1 ratio (1,513 to 495).
IRV maintains the spoiler effect
For the voters who picked C as a first choice, this is an example of IRV's spoiler effect at work. IRV advocates make the inaccurate claims that with IRV, "you can honestly rank candidates in order of choice without having to worry about how others will vote and who is more or less likely to win" and that IRV "allows more than two candidates to compete without fear of splitting the vote." Burlington voters who preferred C most were told they could honestly express their preferences on the ballot. As a result, they marked their preferences honestly and 3/4 of them got their worst possible outcome.
Had even a quarter of those voters strategically placed A above C, their outcome would have improved; those voters' 2nd choice would have won instead of their 3rd. Sincerely ranking their favorite candidate in 1st place hurt them. Thus IRV maintains the spoiler effect that can give you an incentive to vote against your true favorite.
The Fallout
Burlington voted to repeal IRV the next year.
What would have happened if Burlington had used STAR Voting?
We can't know for sure, because rank-only ballots don't contain any notion of level of support, just preference order. However we think it is very likely that one of two things would have happened:
-
A would have achieved a high enough score to make it to the automatic runoff with either B or C, and then gone on to win the runoff since A was preferred head-to-head over both B and C.
Given the actual votes, it seems likely that A would land in the top two, score-wise. A large proportion of both B and C first choosers put A second: A received 3,556 second-choice votes in the ABC contest, versus 1,827 for B and 1,138 for C. Assuming voters gave their first choices a 5, and that voters rated second choices similarly between factions, the average of second choice scores would have to be ~1.3 or greater in order to propel A into the runoff over B. Thus we think it likely that STAR would have elected the "beats-all candidate" in Burlington. Alternatively,
-
A might not have scored highly enough to edge either B or C out of the runoff, and B would have won just as in IRV. If A fails to score highly enough to make STAR's automatic runoff, the electorate knows that they, as a whole, rated A as such a weak Condorcet winner that he failed to out-score both B and C.
The C voters who placed A second, who would have had to have given very low scores to A in order for B to win under STAR, would be inclined to be more willing to compromise in subsequent elections. The basic sense of unfairness that registers from the IRV result, and a clear motive for repeal, would not be present under STAR.
STAR Voting produces results that make sense
The Burlington election looks like an extraordinary failure for IRV, since there was a clear "beats-all candidate" who didn't win the actual election, a clear spoiler effect for C voters, and the system was repealed by Burlington voters the next year.
Under STAR, all the voters can offer full honest support to their favorites, and the notion of overall scores from the public fully justify either election outcome. Advantage: STAR.
Compare and Contrast
FairVote, the leading Instant Runoff advocacy organization, has a different interpretation of the 2009 Burlington Mayoral race:
Before: "... Under old plurality election rules (the "winner" is the one who gets the highest vote total, even if far less than 50%), common in most U.S. elections, whenever there are more than two candidates, there is a danger of a "spoiler" scenario. There have been absolutely no such concerns in Burlington's multi-candidate mayoral races with IRV... Burlington is clearly being well-served by its instant runoff voting system."
Immediately After: "Burlington's instant runoff voting (IRV) election went off without a hitch in 2009... IRV clearly worked as intended to avoid the "spoiler" dynamic... While Sore losers in Burlington are complaining about sour grapes, instant runoff voting has proven itself again as a bulwark of democracy."
After Repeal: "The case for IRV remains as strong as ever, and appreciation for its value keeps expanding. Losses hurt and lessons from them must not be forgotten, but our nation’s shift to “rank the vote” continues."
*Note that this data set, pulled directly from RangeVoting.org, shows minor discrepancies between the official tally and the unofficial preference matrices computed from the full ballot data. Those discrepancies have been carried over.