Can ranking candidates fix elections?

Tuesday might have been the last traditional Election Day of my life in Washington, DC, where I’ve been voting for the past 12 years. 

The ballot included Initiative 83, a measure adopting ranked choice voting (RCV); it passed overwhelmingly. While it’s possible that the DC government could just refuse to implement the measure (they’ve done it before), it’s more likely that from now on, I’ll be ranking candidates for the DC Council and mayor — not just voting for one candidate per post.

Ranked choice is an electoral reform that felt like a pipe dream only a few years ago, but has been becoming mainstream over the past decade or so. Alaska, Hawaii, and Maine, have adopted it for some elections to Congress or statewide office. While a small handful of municipalities like San Francisco and Minneapolis have used it for decades, they were recently joined by New York City, Seattle, and Portland, Oregon. Alongside DC, the states of Colorado, Idaho, Nevada, and Oregon all voted Tuesday on adopting the system, and Alaska voted on whether to keep it. Interestingly, only in DC did the system pass.

Full disclosure: I voted yes on the DC initiative. I think it probably does more good than harm in the context of our city. First-past-the-post voting clearly has deep flaws, which is why so many places are jumping on the RCV bandwagon. But I also think RCV’s benefits have been oversold and that we should experiment with other ways to make our elections more proportional.

Ranked choice voting, explained

In ranked choice voting (also called “instant runoff”), voters rank candidates in order. All the first-choice ballots are counted. If no candidate has a majority of first-choice votes, then the candidate with the smallest share is eliminated; their votes are then redistributed based on who their supporters ranked second. This continues until a candidate has an outright majority.

I first encountered the idea after the 2000 election. In Florida, 97,488 people voted for Ralph Nader; of whom only 537 would have had to vote for Al Gore to give him the win in the state and thus the presidency. What if those Nader voters — who were overwhelmingly liberal — had been able to rank Gore second? Then this would’ve happened naturally, and the failure of left-of-center voters to coordinate wouldn’t have resulted in George W. Bush’s presidency, the war in Iraq, etc.

This rationale is also why I support the idea in DC. Here, like a lot of coastal cities, almost all the political competition occurs in the Democratic primary, which is often incredibly crowded. Every four years, good-government folks here try to unseat Anita Bonds, our notoriously ineffective and incompetent at-large city councilor, and every time, multiple challengers wind up dividing the anti-Bonds vote. Two years ago, she won renomination with 36 percent of the vote, while two challengers each got 28 percent. RCV would make it harder for unpopular incumbents to get renominated by dividing the opposition.

As a narrow tool to avoid spoiler effects, RCV works quite well. But its supporters also have grander ambitions. 

Katherine Gehl, a wealthy former CEO who has bankrolled many recent RCV initiatives, argues that her particular version (called “final five” voting) will almost single-handedly make politicians work together again. Gehl wrote two years ago:

Barriers to cooperation fall. Senators and representatives are liberated from the constraints of negative partisanship. They are free to enact solutions to complex problems by reaching across the aisle, innovating and negotiating.

The theory is elegant. In final five voting, all candidates — regardless of party — participate in a primary. The top five contenders are then placed on the general election ballot, where voters can rank them. 

The hope is that this eliminates the dynamic where partisan primaries push party nominees to ideological extremes, and where fear of such a primary prevents incumbents from compromising or defying their party (see the 10 House Republicans who voted to impeach Trump, of whom four lost renomination when challenged by a pro-Trump Republican). Then, ranked-choice voting in the general election means candidates compete for No. 2 and No. 3 votes, reducing the incentive to negatively campaign.

The case(s) against RCV

Sounds great! So why would someone oppose RCV?

One possible reason is the finding by political scientist Nolan McCarty that under RCV, precincts with more ethnic minorities see more “ballot exhaustion” (failing to rank as many candidates as one is allowed to). That means, McCarty has argued, that the reform tends to “reduce the electoral influence of racial and ethnic minority communities.” 

Work by Lindsey Cormack, an associate professor at Stevens Institute of Technology, has similarly found that “overvoting” (using the same ranking more than once, which means ballots can’t be counted accurately) is more common in minority communities, while University of Pennsylvania’s Stephen Pettigrew and Dylan Radley have found that ballot errors in general are much more common in ranked choice than traditional elections.

Anything that raises the specter of reducing electoral influence for minority communities in the US is worth worrying about. That said, I’m not sure this case is disqualifying either. Ranked choice is a significant change that takes time for an electorate to understand and adjust to. I’m not sure that higher error rates for a newly adopted approach to voting indicate these error rates will persist as the practice becomes normalized.

To me the more compelling counterargument is that RCV seems unlikely to do anything to reduce partisanship and encourage cross-party compromise. The reason why has to do with the classic case against instant-runoff voting, which you might have heard if you’re friends with social choice theory nerds (as, alas, I am). 

One thing you’d want a voting system to do is elect the person who would win in a one-on-one race against every other candidate. This is called the “Condorcet winner,” and while there isn’t always one in an election, when there is one, it seems like a good election system should give them the win, as the person the electorate prefers to all alternatives.

Ranked choice voting does not always pick the Condorcet winner, and we’ve now seen multiple real-world elections in which the Condorcet winner (which you can figure out from ranked-choice ballot records) lost. In Alaska’s US House special election in 2022, which used ranked choice, the Condorcet winner was Republican Nick Begich, but Democrat Mary Peltola won. Something similar happened in the 2009 Burlington, Vermont, mayoral race.

Importantly, in both cases the Condorcet winner was the most moderate of the three main candidates. Begich was to the right of Peltola, but to the left of Sarah Palin (!), the third candidate. In Burlington, the left-wing Progressive Party nominee beat both the Democratic and Republican nominees, though the Democrat (a centrist in Burlington terms) was the Condorcet winner.

RCV advocates note that these are two cases out of thousands of RCV elections, and that in practice, Condorcet failures are rare. I’m not so sure about that. 

Research from Nathan Atkinson, Edward Foley, and Scott Ganz used a national ranked choice survey of American voters to simulate what elections would look like under the system nationwide. For each state, they simulate 100,000 elections with four candidates. They find that in 40 percent of cases, the Condorcet winner loses, which suggests that the rarity of Condorcet failures in practice may just be an artifact of RCV being relatively new, and that such outcomes would become more common in time as the method spreads.

Worse, the simulation paper finds that the system results in much more extreme winners (that is, winners who are farther away from the median voter) than one that picks the Condorcet winner. Indeed, “the states where [the system] performs worst (including Arizona, Nevada, and Georgia) are among the most polarized, whereas the states where [it] performs the best (including Massachusetts, North Dakota, and Vermont) are among the least polarized.” The system seems to actually encourage polarization, not avoid it.

DC aside, voters themselves on Tuesday mostly opposed RCV. In Colorado, Idaho, Nevada, and Oregon, electorates all voted down ballot initiatives that would have introduced some form of RCV. In Alaska, a ballot initiative to repeal that state’s RCV is currently slightly ahead.

New America political scientist Lee Drutman was once such a great fan of RCV that he wrote a book calling for it, but has in recent years come to think it’s hardly the cure for polarization and dysfunction he once viewed it as, in part due to findings like Atkinson, et al. A better solution, he argues, is to strengthen parties and encourage more of them to form. 

States should allow “fusion voting,” in which candidates can run on multiple parties’ lines (New York already does this), and for legislatures, seats should be allocated proportionally: If there are 100 seats, and Democrats and Republicans each get 45 percent of the vote and Greens and Libertarians each get 5, then they should get 45, 45, 5, and 5 seats, respectively.

This is a much more radical change than ranked choice voting, and requires a real rethinking by politicians. It’s hard to imagine a DC with multiple functional political parties, or where anyone important isn’t a Democrat. But it’s worth trying it and experimenting. We have learned a lot from trying RCV, and we can learn even more.

Update, November 6, 11:19 am ET: This story was originally published on November 6 and has been updated to include the results on numerous state ballot initiatives around ranked choice voting.