A big selling point Ranked Choice Voting (RCV) is that it’s tried and true; it’s used by major US cities like San Francisco and New York, and in Maine, Alaska, Australia, and Ireland. By contrast, Approval Voting is used in just St. Louis and Fargo, STAR Voting isn’t used by anything larger than a regional political party, and Condorcet methods see some use in a few European cities and Pirate parties but little use in the US. Intuitively, this seems like an advantage of RCV, but I argue that for someone who wants to improve democracy as effectively as possible, the fact that a voting method is widely used is actually a reason to avoid advocating for it.
Imagine you’re one of thirty study participants in a room together. The experimenter shows all of you two decks of cards; one has 20 red cards and 10 green cards, and the other has 20 green cards and 10 red cards. She then chooses a deck at random (none of you can tell which one) and has the first participant take a card from the deck, look at it, and say aloud which deck she thinks was chosen — the knowledge of which card it is remains private. The first participant looks at the card and says she thinks it’s the red deck. Then the next participant looks at his card and says that he thinks it’s the red deck. Then you take your card, and it’s a green one. Do you say that you think it’s the green deck or the red deck?
Your card is green, but it sure seems like it’s in the minority; why would the first two people have said “red” if their own cards weren’t red? So you’re most likely to be correct if you say “red” as well. Then if the next person’s card is green they’ll be even more likely to say “red” — it looks like 3–0 in favor of red from their perspective! But maybe the first two cards were just a fluke and it was two red cards in a green deck. Maybe the first person got a red card and the second person just went along with it because he was colorblind and couldn’t tell his card was green. This phenomenon is called an information cascade: one or two misleading observations early on can result in everyone being wrong down the road.
When someone says that it’s better to implement RCV than Approval or STAR Voting because RCV is more widely used, this person is telling you to go along with an information cascade. If you want to personally be right about something and don’t want to spend time investigating it, just going with what most people think is a reasonable approach. But it’s a disastrous approach for making decisions collectively. If you want everyone in the room to converge on the truth, you have to say “green”. Otherwise, mistakes will be compounded indefinitely.
The Significance of Momentum
There’s an important flip side to decision-makers’ susceptibility to information cascades. Most people prefer to adopt an already-popular and widely used voting method. This means that victories in enacting a voting method, like mistakes in guessing which deck you’re drawing from, compound. If you put a major effort into getting a particular voting method adopted in one city, this makes it more likely that another city will adopt it later without any additional effort from you. Simply being able to point to a city that uses the voting method you’re advocating is a big help. Moreover, having a first example is a bigger deal than having a second example; the marginal effect of serving as an example is larger the fewer places that use a voting method. The lesser-known a voting method is, the more a single adoption of it will raise its profile.
Value of Information
Of course, most people advocating RCV aren’t just saying, “You should use this because other cities do”, they’re saying “You should use this because other cities do and it works well for them.” And this appears to be true! There is (weak) evidence that it results in more victories for women and people of color than occur under Plurality, and it probably improves campaign civility. But this tells us next to nothing about how it compares to other voting methods. Perhaps STAR will outperform both RCV and Plurality — but we can’t know unless we test it.
Suppose 50 cities use STAR for a decade and 50 use RCV, such that we have solid datasets for each and can properly evaluate their effects. There are several possibilities for what happens:
- RCV outperforms STAR. In this case, these STAR cities should switch to RCV, and other cities should also adopt RCV. The cost of anyone ever adopting STAR instead of RCV is that these cities used a suboptimal voting method for a decade (plus the cost of switching from STAR to RCV).
- STAR and RCV perform about the same. In this case, it doesn’t really matter that some cities adopted STAR instead of RCV.
- STAR outperforms RCV. In this case, it’s entirely beneficial that STAR got adopted in places instead of RCV. The first benefit is that 50 cities use a superior voting method for a decade. The second benefit is that states and other cities know to adopt STAR instead of RCV, so they will reap all the benefits that STAR has over RCV in this scenario.
The asymmetry between the first and third cases is due to the value of information: knowing what is most effective improves our decisions going forward, and even in advance of actually having this knowledge we can see how it will be useful. To be fair, there are plenty of real-world considerations that reduce the value of information compared to what it looks like in the rosy picture I’m painting of perfect decision-making. It’s safe to say that some jurisdictions will decide to adopt voting methods that are demonstrably inferior no matter how strong the evidence is, and it’s not like everyone would switch overnight once the evidence is in. But the basic observations remain true: Adopting a currently-unused voting method will give information on how well it works, and this information is valuable.
There is ample reason to think we can do better
On paper, STAR and Condorcet methods look better for single-winner elections than RCV. (Approval is a bit more ambiguous.) They perform better in tricky scenarios, and, according to computer simulations, do a better job of satisfying the preferences of the electorate in general. Both incentivize candidates to have greater concern for the preferences of voters who don’t like them than RCV, which can reproduce the incentive structure of partisan primaries. Both are easier for computers to tabulate since they’re precinct summable; RCV requires centralized tabulation. Scoring methods like STAR display more third-party support in election results than RCV does; it’s less clear with Condorcet methods, but at the very least they shouldn’t do any worse.
But my point is not that STAR and Minimax are clearly better than RCV, it’s that they have a significant chance of being better. The thesis of this post is that “This voting method has been used a lot in the real world” is a disadvantage of adopting that voting method at the level of a city or a state, rather than an advantage. Whether or not STAR and Minimax actually outperform RCV is irrelevant to this thesis.
A final note: all of these arguments apply to more than just voting methods; they also apply to alternative forms of government (such as sortition or a futarchy), most kinds of public policy, and are arguments for doing science more broadly. Trying new things is important in just about any arena.