Make. All. Votes. Count. (Part III: Nonpartisan multiwinner)
I live in Cambridge, MA. Our ballot for city council typically looks something like this:
We have this crazy grid of a ballot because we use STV, single transferable vote, for our local elections.¹ This is a proportional representation (PR) method, which ensures that X% of the voters can (if they’re not totally miscoordinated) elect at least around X% of the winners.
PR’s a great idea.
PR methods have a host of advantages over the method used by most US elections, known as plurality or FPTP (first past the post):
- PR minimizes wasted votes. This means a drastically lower chance of “spoiled” elections; and more stakeholders involved in decisionmaking.
- PR tends to improve gender and ethnic balance, especially if voters care about their identity groups being represented.
- By improving voter power, PR has the potential to increase turnout.
- PR encourages positive, issue-focused campaigns, rather than mudslinging.
I’m somebody who cares deeply about improving election methods; I’m a board member for electology.org (aka the Center for Election Science), a relevant nonprofit. Check us out! So I should be overjoyed to be one of the few people in the US who gets to use this kind of voting method. And in large part, I am.
But.
STV’s too complicated.
Look at that ballot again. And think about the place where you live. Imagine being asked to rank every single one of the candidates for city/town council in order of preference. How many aspiring candidates could you even name?Can you even name one incumbent?
Unless you’re a 99.9th percentile politics geek, ranking all candidates isn’t easy. And that’s a problem. Sure, voting shouldn’t be a no-brainer; but it also shouldn’t take more homework than a college class. That STV ballot is just too much work.
And STV has another flaw, too: it requires all the ballots (or at least, all the data from each unique ballot) to be transmitted to a centralized location in order to find the winner. That’s not just cumbersome; it also reduces election security.
Newer methods to the rescue?
If you read the first two articles in this sequence, you know that it’s possible to design voting methods to be easy and robust. Part I dealt with partisan legislative elections (eg, congress) and advocated a method called GOLD voting. Part II dealt with single-winner elections (eg, president) and advocated a method called 3–2–1 voting. In both cases, the methods I discussed use simple ballot formats and relatively strategy-resistant counting processes to make the voters’ job as easy as possible. (Also, both systems are “summable”, meaning you can do all the ballot-counting at the precinct level.)
But neither of those methods would work for Cambridge. 3–2–1 is for electing just one winner, not a full council; and GOLD requires candidates to have party labels.
3RD voting
So, to round out the suite of methods, we need a third one. Or, in this case, a 3RD one: 3 Ratings Delegated voting, aka 3RD voting. This method runs on 5 basic principles:
- As a voter, you can rate each candidate using one of 3 ratings: “Good”, “OK”, or “Bad”.
- If you rate more than one candidate “Good”, your vote is split into fractions. For instance, if you rate the 3 candidates A, B, and C “good”, candidate D “OK”, and the rest “bad” or blank, your vote is split into thirds. The first third would be counted as rating A “good”; B, C, and D “OK”; and the rest “bad”. The next third would be similar, but with B instead of A as “good”; and the final third would do the same with C. Note that each ballot fraction rates exactly one candidate as “good”.
- If you leaves a candidate’s rating blank, that rating is filled in using the public predeclared ratings from the candidates whom that voter rated “good”. For instance, in the example above, if you’d left a blank for candidate Z, then Z’s rating on the first third of your ballot would be either “OK” or “bad” depending on what A had said; and similarly, Z’s rating on the second and final thirds of your ballot would depend on what B and C had said respectively.
- All the ballot fractions with a given candidate as “Good” are averaged together. Each precinct reports the number of fractions rating each given candidate X as “good”, and also for each given pair of candidates X and Y, reports what percent of the ballots which rate X as “good” also rate Y as at least “OK”.
- The resulting totals are used in an STV-like process to find a proportional set of winners. Vote fractions start out as counting for the candidate they rate as “good”; then, if that candidate is eliminated, they’re transferred, split among all the candidates in proportion to how often those candidates were rated “OK” on ballots that rank the original candidate as “good”.²
This definitely sounds a bit complicated as a counting process, though in practice, it’s not much harder than STV. But the advantage is how simple it makes things for voters. Candidates rate each other either “OK” or “bad”, and voters can use those ratings as a fallback. If you like one candidate, and you trust them to do a good job rating the other candidates, you can just rate that one candidate “good” and then go home. But if you don’t trust any of the candidate’s predeclared ratings to “fill your ballot out for you”, you can just fill it all out yourself. And even then, it’s pretty simple; you don’t have to put all of them into a strict order, but can just class them into 3 general ratings categories.
As the Gibbard-Satterthwaite theorem shows, no democratic voting method is entirely immune to strategic voting — people tweaking their ballot in order to get an advantage in voting power. However, strategy in this method, while possible, is not a serious issue. Most strategies are self-limiting; that is, the more people use them, the less effective they become. And all strategies are fragile; that is, in order to be used successfully, they require an extremely detailed and accurate forecast of how other people will vote.
Conclusion
This concludes my series on newer voting methods. All three methods I’ve discussed — GOLD voting, 3–2–1 voting, and 3RD voting— are in the class of “GOOD” (Graded Or Optionally Delegated) methods. That means they’re based on these simple ideas:
- Ballots should be simple.
- The vote-counting method should minimize the incentives for strategic voting.
- “Lazy” voters should have the option to delegate their ballot to their favorite candidate.
- But highly-engaged voters shouldn’t be forced to delegate to any candidate; there should be an option for those who don’t trust any candidate.
Footnotes
¹ These days, an election reform organization called FairVote is trying to rebrand STV, along with the corresponding single-winner method IRV, as “RCV”, ranked choice voting. I don’t like that name; it deliberately blurs the important distinction between PR and single-winner reforms, and also tries to hide the fact that other ranked choice methods (e.g., Ranked-Pairs Condorcet, Borda, etc.) exist.
² Actually, when transferring votes, the proportion that originally went to X and now go to Y are not solely based on the number of ballots for X that rate Y at least “OK”; but on the product of this number with the number of ballots that originally went for “Y”. This helps increase the impact of a “good” rating.
