In this section we explore how RCV has worked in practice in the United States. We summarize research on how many voters rank candidates and how those rankings are organized, on how RCV has worked to remove the spoiler effect and on the tendency of RCV in single-winner seats to elect candidates with majority support. We also assess RCV in practice on some more technical grounds, namely monotonicity and its tendency to elect Condorcet winners.
Experience shows that, when American voters are given the option, they prefer to rank candidates rather than merely lodge a single preference for one candidate. In the vast majority of RCV races contested by more than three candidates (the minimum number of candidates needed before ranking is meaningful), a significant majority of candidates rank at least two candidates.
For example, in 2014, three-quarters (74%) of Oakland voters ranked three different mayoral candidates (the maximum allowed). Another 11% of voters ranked two. In the 11 Alameda County RCV races that had three or more candidates in 2014, 63% of voters ranked three candidates, and 76% ranked at least two. Similarly, in the five-way contest for San Francisco’s 10th Supervisor District with a strong incumbent, only a third of voters ranked just one candidate – and only 8% of these were bullet voters (2% of all voters in the 10th District race) who voted for a candidate who failed to reach the final round.
Using ballot image data, we can do more than summarize how many candidates voters ranked. In multi-winner RCV, we can study how voters used their preferences. For example, in 2014, FairVote's Andrew Douglas used ballot image data to show a pattern of racially and ethnically cohesive voting among Cambridge, MA, city council voters. FairVote will soon release more research, showing the relative influence of candidate ideology, incumbency status, race, gender and place of residence on the choices made by Cambridge city council voters.
One of the chief potential advantages of single-winner RCV over plurality is that it mitigates the spoiler effect. With the mounting experience with single-winner RCV in the United States, empirical analysis of the spoiler effect in American RCV elections is now possible. Stay tuned for our findings.
The use of single-winner RCV increases the proportion of candidates winning with a majority of votes cast and the likelihood of the winner being the Condorcet winner. Since San Francisco started using ranked choice voting in 2004, the cities of Berkeley, Oakland, San Leandro and San Francisco have held 164 RCV elections for their combined 53 elected offices.
Of these contests, 53 of the winning candidates did not obtain a majority (50% + 1) of the vote in the first round, meaning rankings were used to determine the winner. Seven of these candidates trailed in first choices, but went on to win the election, defeating the likely plurality winner. The winner of these races has been the “Condorcet winner” — the candidate who would defeat all others in a head-to-head contest — every single time.
If we assume that voters’ rankings reflect sincere (rather than strategic) views, a plurality system would have elected the non-Condorcet winner in seven elections. Additionally, under a runoff system, the Condorcet winner would have failed to reach the runoff in one instance — at the end of the first round the winner was in third place by a very narrow margin — and other Condorcet winners might have lost in runoffs that had smaller, less representative electorates due to a likely decrease in turnout.
In other words, our data from the Bay Area show that RCV is more likely to elect a Condorcet winner than either plurality voting or runoff elections.
Finally, the number of RCV elections with three or more candidates is substantially higher than those with just one or two, meaning voters have a significant range of choices at the polls without having to worry about a split vote. Candidates also benefit from clear mandates as eliminating plurality outcomes gives winners well over 50% majorities on average.
Monotonicity is a technical property of a voting system in which it is impossible for a voter to prevent the election of a candidate by ranking them higher on their ballot while also being impossible to elect an otherwise unelected candidate by ranking them lower. In the absence of monotonicity, a voter may hurt the chances of their most favorite candidate winning if they ranked the candidate first rather than second on their ballot. RCV is non-monotonic but in practice should rarely behave non-monotonically. FairVote is currently using ballot image data to explore whether single-winner RCV has ever behaved non-monotonically in the American experience.