The biggest question for a voting method

“How will this affect political culture?”

I recently wrote an evaluation of how much good we can do by working on voting methods reform as an entry for Open Philanthropy's cause exploration prizes. Two takeaways: First, adopting better voting methods matters a lot. Second, far more of the benefits of better voting methods come from reducing political polarization than from having better/more representative leaders; by my calculations, the latter is literally hundreds of times more important.

I’ve become convinced that political polarization is a grave and worsening threat to American democracy. Both Democrats and Republicans accuse the other side of stealing elections. Metaculus puts the probability of a second American civil war by 2031 at 5%, which is terrifyingly high when you consider how bad one could be; by my calculations, this “small” chance imposes an expected cost of over $300 billion per year. And civil war seems less likely than democratic backsliding, in which freedom of the press is lost and elections become rigged, not as the result of a coup or other violence, but due to the actions of elected officials.

Voting methods have a lot to do with political polarization; the two-party system wouldn’t exist if there was proportional representation, and both Plurality and partisan primaries are sufficient to leave Democratic candidates without any incentive to appeal to reliably Republican voters and vice versa.

Why electing good winners isn’t (that) valuable

As I’ve become convinced that depolarization matters a lot more than I originally believed, I’ve turned to the opinion that electing optimal winners matters significantly less than I originally assumed.

A basic point: whether an elected official has broad support has no intrinsic moral significance. (At least according to my ethics; others may differ.) What matters intrinsically is people living happy, healthy, and meaningful lives; what matters about our leaders is their effectiveness at improving people’s lives. (I intend this in a way that also encompasses harm; Stalin’s effectiveness at improving people’s lives was extremely negative.) The importance of electing leaders who properly reflect the will of the people is determined by the correlation between reflecting the will of the people and how effective a person would be at improving people’s lives. There is reason to think that this correlation is quite weak.

Voters do want plenty of qualities in a leader that are indisputably good. They want leaders to be competent. They want leaders not to be corrupt. For these reasons I do think that leaders who were more popular when they got elected will tend to be better leaders. But this isn’t where most of the electorate’s optimization power is going.

Much of what voters care about is partisanship, and good voting methods won’t favor one major party over another (on average). Take the recent Alaska election, in which Democrat Mary Peltola won even though she would have lost by 5 points to Republican Nick Begich head-to-head. Why was Begich more popular? Because he’s a Republican and Alaska is a red state. You could just as easily have a Republican win a House seat in a blue district because of a suboptimal voting method, and the average effect is basically zero.

Voters care about a lot of stuff that doesn’t matter; charisma, seeming relatable, and “sounding presidential” all affect voters’ opinions. (I’ve certainly been guilty of voting based on such things in the past and I expect to be guilty of it in the future. Voting well is hard.) Voters on both sides of the aisle also want a lot of things that are bad for them.

Another point is that the quantitative difference in who different voting methods elect is not all that big. Take this graph of Voter Satisfaction Efficiency (VSE):

Usually, when I talk about VSE I’m comparing voting methods. But here I want to talk about the scale of the x-axis. VSE is defined such that a hypothetical voting method that always elected the best possible winner would have a VSE of 100% and electing a winner completely at random has a VSE of 0%. But electing a winner at random in the context of a VSE simulation is less terrible than it might seem. In a high-profile real-world election you’ll get plenty of candidates who aren’t remotely qualified, don’t campaign with the intent of actually winning, and attract virtually no support. These don’t really exist in VSE simulations. With 6-candidate Plurality elections and the voter model used to make the chart above, the least popular candidate gets an average of 4.5% of the vote. Compare this to the performance of the Presidential candidate I personally know (Blake Huber of the Approval Voting Party), who got 0.0109% of the vote in Colorado. Electing random winners in VSE corresponds to electing a bunch of Kirsten Gillbrands, Jay Inslees, and Bobby Jindals for President; not to electing Blake Hubers.

Back to the chart: The difference between using Plurality with viability-aware voters and using something top-notch like STAR or Minimax is around 1/6 the difference between electing Biden-esque candidates and electing Gillibrand-esque candidates. The difference between IRV and a top-tier method isn’t even half as big.

How should we evaluate voting methods?

To recap: When it comes to comparing different reasonable voting methods — not just assessing the harms of using Plurality — who a voting method elects is not all that important. Differences in how effective voting methods should be at reducing polarization matter much, much more. More generally, “How would this voting method affect political culture?” is the question we should focus on when assessing a voting method. Other considerations, such as whether the voting method favors centrists or extremists and the role of strategic voting, still matter. But they mainly matter indirectly, by how they and the incentives they create shape political culture.

• A pro-extremist bias matters because voting methods with a tendency to elect extremists (like IRV and Plurality) necessarily incentivize extremism. The fact that Mary Peltola’s voters preferred Nick Begich over Sarah Palin by more than a 10-to-1 margin didn’t help him; Palin still knocked him out before the final round. As a result, he has no reason to reach out to Peltola voters the second time around.
• An extreme vulnerability to strategic voting matters for several reasons. Consider the political culture induced by strategic voting (and strategic candidate behavior) in Borda Count. Candidates want their supporters to rank their viable competitors last — ahead of their non-viable competitors, even if those non-viable competitors are utterly terrible. To make their supporters willing to engage in such skullduggery, candidates are incentivized to drum up extra hatred among their supporters for the other viable candidates and to create the perception that the other campaigns are cheating (perhaps by using this very same strategy). (Note that this problem only occurs in certain versions of Borda Count.)
• VSE may not matter that much in and of itself, but it’s still a good heuristic for evaluating voting methods because it’s correlated with stuff that matters. When a voting method has a VSE substantially below 100%, it means that the equally-weighted preferences of all voters aren’t doing so much to determine the winner — so some other factor must be playing a role. This other factor is most likely quite harmful; for instance, some voters may be far more influential than others.

What deserves more attention?

I mainly focus on tabulation algorithms for voting methods and their consequences, but a focus on political culture suggests other considerations that deserve a lot more scrutiny than I’ve been giving them:

• Primaries and runoffs: These mean that there are multiple phases in an election in which politician have different incentives, different voters are paying attention, etc.
• Requirements for ranking every candidate: In Australia’s federal IRV elections, voters are required to rank all the candidates or their ballots are invalidated. No American jurisdiction requires this.

There are probably other neglected considerations I haven’t thought to mention. I haven’t ever bothered to mention those that fall outside the umbrella of voting methods, such as compulsory voting rules, money in politics, and eligibility to participate in Presidential debates. In no case am I remotely confident as to what is optimal, but some possibilities would argue heavily for some voting methods over others. For instance, if compulsory ranking is beneficial, this argues for ranked voting methods in which we can disallow tied rankings and against STAR and Approval.