The Limits of Ranked Choice Voting
In 2018, Maine became the first state to use the alternative voting method ranked choice voting1(RCV) in federal elections. That’s good because it meant that Maine stopped using our ubiquitous choose-one voting method — largely regarded as the worst single-winner voting method there is.
Keep in mind, however, that the current bar we’re pushing past — our choose-one voting method — is an especially low bar. That voting method breaks at the slightest hint of competition beyond a second candidate.
It’s important as well to not oversell the benefits of RCV, which many outlets are doing. Overselling RCV can create unrealistic expectations and sour voters to other alternative voting methods. And some of those alternative voting methods aren’t just simpler than RCV. They’re flatout better.
That said, it’s important to give the benefit of the doubt to media and those who advocate for RCV. Voting methods are complicated and not intuitive in the slightest (even when they appear to be). So, it can be easy to make mistakes.
A Primer on RCV
Let’s recap how RCV works. Voters are asked to rank the candidates on their ballot from most to least favorite. Sometimes voters are limited to three rankings, but other times there’s no limit. That’s the information component, just ranking, though it can be more demanding if you have a bunch of candidates.
The calculation component is more involved. Candidates with the fewest first-choice votes are sequentially eliminated and those next-preference choices transfer over in subsequent rounds. That continues until there’s a round where a candidate has more than half the remaining first-choice votes.
Among single-winner voting methods, this one is moderately complicated. For a visual, here’s a video from the Maine Secretary of State on how RCV works.
RCV’s Limits: Five Examples
1. What RCV Doesn’t Do: Guarantee A Majority
Let’s start with one of the bigger RCV claims that falls short, that it always elects majority winners. Now, RCV will elect a majority winner — so long as that majority winner actually exists in the election. But because a majority winner doesn’t always exist, no single-winner voting method can guarantee a majority in every election. That includes RCV.
There are a couple issues to consider here on the majority concept. One is ballot exhaustion. Ballot exhaustion is probably the more common criticism of RCV’s majority claim. Ballot exhaustion is when a voter’s ballot preferences get eliminated so that nothing carries over to later rounds. This means that the majority RCV arrives at is only within the remaining ballots rather than the original ballots. If you look closer at the Maine example video above, you’ll see that was the case there, too. The winner had fewer than half the votes of the original ballots.
To be fair, if we were to do a traditional runoff, a lot of those voters wouldn’t come back a second time. We might look at that as a kind of voluntary — and perhaps a bit literal — ballot exhaustion. Still, it’s important to be clear that RCV isn’t talking about a majority of all the voters who casted a ballot. It’s only talking about the remaining ballots.
A more pointed objection is that much of the time RCV isn’t getting any kind of majority at all. Rather, it’s contriving a majority by artificially narrowing down the candidate field. RCV knocks out candidates over each round, but sometimes it knocks out good candidates by mistake. (You’ll learn more about how this goes on in the RCV vote splitting section.) You see, any voting method that finds some way to knock out candidates until two remain will get a “majority”. But that “majority” is contrived and merely a byproduct of having two candidates remaining.
Don’t worry if that wasn’t immediately obvious. The concept of majority in voting theory is counterintuitive. We have a whole article on the concept of majority alone.
2. What RCV Doesn’t Do: Properly Address Vote Splitting
Vote splitting occurs when voters must decide between candidates with overlapping similarities. When this happens, vote splitting decreases the support among those similar individual candidates. On the surface, with all the ranking transfers that RCV does, it looks like RCV addresses the vote splitting issue. But it only does so a little bit.
To give RCV credit, it does address vote splitting in some cases where fringe candidates genuinely have little support. In RCV’s minor victory, it defends well against vote splitting from the edges of political support. But it largely fails at addressing vote splitting from the center. (Here’s a fun visualization tool to play with to see for yourself. RCV is noted as IRV.)
An example of RCV behaving well was the recent RCV election in Maine where a third-party candidate with little apparent support divided the vote between another liberal candidate. Once the third party candidate was eliminated for having the fewest first-choice preferences, those votes overwhelmingly went to the Democrat. So while some liberals got to vote for a third party, it didn’t cost them the election. That’s because the independent’s votes transferred over to a stronger candidate.
This scenario where an unpopular candidate shares support with a leading candidate is where RCV performs best and is clearly superior to the current choose-one way of voting. In competitive elections, however, more than two candidates can have significant support. And there, RCV has a tougher time.
Here’s an example of one such election in Louisiana that included a runoff. And though it was a runoff, the mechanics played out the same as in an RCV election2.
The 1991 Louisiana gubernatorial election was a close three-way race between Edwin Edwards (34%), David Duke (32%), and Buddy Roemer (27%). Another candidate, Clyde Holloway, was in fourth with 5%, while the remaining candidates had less than one percent.
Roemer was the incumbent governor. Edwards was suspected (and later convicted) of corruption charges, while Duke was the Grand Wizard of the Klu Klux Klan. The vote mainly split between these three candidates. This vote splitting caused Edwards and Duke to go to the second round instead of the more tame candidate, Buddy Roemer.
It’s also clear that Roemer was the preferred candidate. In polling that was able to compare candidates head-to-head, Roemer was preferred in 2-way races with both Edwards and Duke.
So what happened here?
Voters split their first-choice preferences, and that caused Roemer to get eliminated. The same happens in rankings. First-choice preferences get split and so Roemer would have still gotten eliminated. And in an RCV election, that would have continued to let the corrupt Edwin Edwards go to the next round with clansman David Duke.
When we see close elections between three or more candidates under ranked choice voting, we can expect this randomness to carry out again because of this first-preference vote splitting. Primaries would be particularly dangerous territory where you tend to have many competitive candidates with overlapping ideology. Further, with competitive elections there’s a tendency to squeeze out the center candidate, as described in the figure below, which would favor more extreme candidates (as was the case in the Louisiana election). This is called the “center squeeze effect”.
That’s not how you want your primary elections decided, which makes RCV a particularly bad choice for those elections. RCV does mitigate these scenarios ever so slightly compared to our choose-one voting method, but RCV is still highly vulnerable to the center squeeze effect.
3. What RCV Doesn’t Do: Always Let You Honestly Rank Your Favorite Candidate First
Admittedly, RCV does let voters get away with ranking their favorite first in two situations. But it can get dicey if you’re unsure whether you’re in one of those two situations.
One safe situation is when your favorite candidate is able to beat the other leading candidate in the last round. Then you’re fine. The other safe situation is when your favorite candidate has no chance against either the leading candidate or the candidates who could beat the leading candidate. You’re good there, too.
It becomes risky, however, to rank your honest favorite when that candidate is in between those two spots (or you’re unsure). To reiterate, this dangerous in-between spot is when your favorite candidate is neither a clear loser nor a clear winner. In those in-between cases, ranking your favorite as first risks getting a bad candidate elected. And that bad candidate gets elected by RCV eliminating a superior compromise candidate too early.
If a voter is either in this in-between state or is unsure, then they have two options. They can either rank their favorite first and risk a terrible candidate winning. Or, they can not rank their favorite first at the cost their favorite candidate’s much-needed support. Both of these outcomes are bad.
Need a visual? No worries. RCV can get complicated.
This scenario can happen in competitive elections. And it does happen. One famous example was in 2009 in Burlington, Vermont. There, conservatives ranked their favorite candidate first and it got them their least favorite candidate as the winner. Had these conservative voters instead tactically placed their favorite candidate as second, then they would have gotten a much better outcome.
Burlington voters have since chosen to repeal RCV.
Unfortunately, RCV’s complexity can make this favorite betrayal issue difficult to understand. But RCV’s complexity does not actually keep this issue at bay. It just makes it difficult to explain when it does happen.
And this scenario will happen again.
4. What RCV Doesn’t Do: Work In Presidential Elections
Some folks even want to use RCV in presidential elections already. This would be a terrible mistake and a logistical nightmare, with or without a national popular vote.
With the Electoral College, it’s challenging to implement any new voting method for a presidential election that doesn’t match all the other states. That’s because the discordance between states can mean that you don’t actually want electoral votes going to your favorite candidate if your favorite candidate isn’t competitive. Then, you’re not even just throwing away your vote. You’re throwing away electoral votes. And electoral votes are much more valuable.
Even under a national popular vote, RCV faces enormous technical hurdles. For one, the nature of RCV tabulation requires that all the ballot data be centralized for tabulation. This creates both security and logistical concerns. Just try getting all the raw ballot data together for all 50 states with all their individual precincts.
Realistically, you’d have to deal with holdout states still using our choose-one method, even if you had a national popular vote. But you can’t add RCV and regular choose-one ballots together. It just doesn’t work. You really realize RCV’s technical limitations when you consider the most viable approach to a national popular vote, which is through an interstate compact (though perhaps in some iteration other than its current form).
5. What RCV Doesn’t Do: Help Third Parties
RCV’s complex tabulations don’t show support for candidates’ rankings once a candidate is eliminated. This can hide a lot of support for third parties. Below is a very common tabulation table for RCV.
Compare this to other methods like approval voting that show all the data at once.
Also, according to Duverger’s Law (a political science concept), a voting method needs to have at least one of two features to encourage third parties. It needs to either (1) have a lower vote count threshold for a candidate to be elected or (2) allow voters to honestly support their favorite. RCV does neither of these things.
Sometimes people get confused about RCV’s support for third parties. In these cases, they tend to be thinking about single transferable vote, a different multi-winner voting method that is proportional. These two separate voting methods behave very differently.
Don’t Forget Your Other Options
These big RCV claims were all made by either large media outlets or by advocacy organizations themselves. These false claims are common and rarely questioned.
Mind you, this is not a call to get rid of RCV. Where RCV has been implemented, it’s moved us past our choose-one method, and that’s a good thing.
This is a call, however, for outlets to be careful of the claims they make and to evaluate their options beyond RCV. We need to take a closer look at other options. The mere fact that RCV elected someone is not enough cause to celebrate. If we place the bar at merely getting a clear result, then we can hurdle the same bar merely drawing names from a hat. We must set higher standards.
We have other voting method options, and they too need exploring. Approval voting is an obvious candidate. Following decades of research within academia, Fargo became the first city to implement approval voting in 2018. Approval voting does have its criticisms (like any method). But there’s strong evidence that in addition to being far simpler, it also performs substantially better than RCV. It’s much better about vote splitting, you can always support your honest favorite, it helps third parties, it’s workable for presidential elections, and it’s fine for primaries (presidential and otherwise).
Our default away from our terrible choose-one method shouldn’t be a needlessly complex method that has its own issues. This essay didn’t even go into RCV’s monotonicity failure where ranking a candidate as better can hurt that candidate and ranking a candidate as worse can help that candidate. This nonmonotonic behavior happened in Burlington’s wild election.
The takeaway here is that we have better and simpler options. Let’s look at them before going all in with RCV.
— — — — — — — —
1 Many voting methods have the confusing feature of multiple names. Ranked choice voting was called instant runoff voting, but its advocates chose to rename it. This is unfortunate because there are many other different ranking methods (ex// Condorcet methods, Borda, Bucklin). Consequently, ranked choice voting is often confused with an entire class of voting methods.
2 Because this was largely a three-way race, voters would provide virtually the same information in both scenarios (traditional runoff and RCV). Both voting methods do the same process when there are three main candidates. For this to not be true, the vote totals from each remaining candidate would have had to go almost perfectly to Roemer. That kind of unanimity would have been extremely unlikely.
Originally published at www.electionscience.org on February 7, 2019.