The False Promise of Instant Runoff Voting

By ballot initiative, Maine has just adopted instant runoff voting (IRV) for state elections.[1] IRV is also used in some local elections around the country. In principle, IRV has some desirable properties compared to the status quo electoral system in the United States (single-member-district plurality), but once we look at the actual political context in which it would be implemented in the United States, it may well make things worse for third parties, especially Libertarians. Moreover, there are alternative electoral systems that are superior both in the abstract and in the concrete to both the status quo and IRV.


Advantages of IRV

Rob Richie’s essay did a good job of laying out the potential benefits of IRV. Its single biggest advantage may be how it addresses the wasted-vote problem. If you like a third-party candidate best, you may be able to safely rank that candidate first on your IRV ballot, knowing that if and when that candidate is eliminated, your second-choice preference will count. For instance, if you liked Ralph Nader but preferred Al Gore to George W. Bush in 2000, under IRV you could have ranked Nader first and Gore second, and your vote would have counted for Gore once Nader was eliminated from the counting process. If every state had used IRV to allocate electoral votes in 2000, Al Gore would probably have won the presidency.

By reducing the wasted-vote problem, IRV also reduces major-party gamesmanship. In last month’s U.S. Senate election in my state, New Hampshire, there was a “conservatarian” independent candidate, Aaron Day, whose mission was to defeat Republican incumbent Kelly Ayotte by drawing away conservative and libertarian votes. He spent no money on his race, but in the closing days, a shadowy, Democratic-linked group sent a glossy mailer to Republican households touting Day’s conservative record and bashing Ayotte as too liberal. Ayotte lost by about a tenth of the votes Day won, and so he quite plausibly cost her reelection. In 2010, Democrats ran “Tea Party” independents in congressional races, and Republicans have also been caught out funding Green candidates in the past (William Poundstone’s book Gaming the Vote describes many such shenanigans).

When used in computer-simulated elections with voters having randomly assigned preferences, IRV outperforms plurality voting by the metric of interpersonally comparable utility (“Bayesian regret” is the technical term). In other words, voters are in aggregate happier with IRV than plurality, at least when they vote sincerely rather than tactically.


Disadvantages of IRV in the Abstract

Even in the abstract, though, IRV has some disadvantages. It doesn’t actually eliminate the wasted-vote problem, and it may reduce it only a little bit. Think about a case with a strong third-party candidate, like Ross Perot in 1992. Suppose 35% of voters prefer Bill Clinton to George H.W. Bush to Perot, 31% prefer Bush to Clinton to Perot, and 34% prefer Perot to Bush to Clinton. If everyone votes sincerely under IRV, Clinton wins after Bush is eliminated in the first round – even though 65% of voters prefer Bush to Clinton. But it gets worse. If just a small number of Perot preferrers (>3%) put Bush first and Perot second, then Perot would be eliminated first, and Bush – their second choice – would win. They’ll have a strategic incentive to falsify their preferences.

Now, every electoral system is subject to tactical voting like this. But IRV makes it easy and obvious how to vote tactically. In general, you “up-vote” your lesser-evil candidate and “bury” your lesser-evil candidate’s most viable opponent. This is just what voters do under plurality, voting tactically for the lesser evil instead of their preferred third-party candidate.

As the Bush-Clinton-Perot example above shows, IRV can also fail to select the Condorcet winner, the candidate that a majority of voters prefer to each other candidate. In fact, this happened in Burlington’s mayoral election in 2009, causing the city to end IRV for mayoral elections when the Progressive won over the Republican in the final round of IRV counting, even though the Democrat, eliminated earlier, was actually the Condorcet winner.

There are several other “in the abstract” disadvantages to IRV. IRV has a more complicated ballot, potentially confusing some voters, and a much more complicated counting process than plurality or approval voting. IRV requires counting at a centralized location rather than by precinct. IRV is also subject to a technical-sounding but important problem called “non-monotonicity,” which means that you can help your preferred candidate by ranking her lower. Finally, the Bayesian regret criterion suggests IRV is worse than other ranked-ballot alternatives like Condorcet-consistent methods and Borda count, and much worse than approval and score voting.


Disadvantages of IRV in the Real World

Getting away from the blackboard for a moment, we also need to think about how IRV would actually work in American elections. It’s no accident that IRV is almost universally a project of the ideological left here in the United States. Vermont started using it in local elections once the Progressive Party became a threat to the Democrats. Maine adopted it after two consecutive elections in which a Republican governor was elected because the left split its vote between a Democrat and a left-of-center independent. (Maine also has a strong Green Party.) Republicans face little third-party threat from their right flank (the Constitution Party is extremely weak), but Democrats do face such a threat in certain places. IRV helps them overcome that threat.

IRV actually neuters third parties, especially those with a strong ideological orientation. Third parties may get higher shares of first-preference votes under IRV, but it is still almost impossible for them to win seats, and they lose all the “blackmail power” that they enjoy under plurality. Currently, strategic third parties can choose to run candidates in races where they want to punish one of the major-party candidates (as Day did to Ayotte) and refrain otherwise. This possibility gives major parties an incentive to cater a bit to ideological minorities.

Is this blackmail power a good thing or a bad thing? It depends on one’s perspective, I suppose, but one way to defend it is to note that democracy’s institution of majority rule threatens to trample the rights and interests of passionate minorities. Third-party blackmail power under plurality rule gives passionate minorities some leverage; it is a way of incorporating intensity of preference into our otherwise majoritarian political system.

For libertarians, the Libertarian Party’s potential blackmail power is a valuable thing. We libertarians expect Democrats at least to be decent on civil liberties and Republicans at least to be decent on economic freedom. When they stray to the authoritarian side of the spectrum, the Libertarian Party can run a candidate to punish them by campaigning on those issues, drawing away conservatives upset by a Republican’s apostasy on economic freedom, for instance. The potential for this sanction should make the major parties govern in a more libertarian fashion than they would otherwise. If this is right, the adoption of IRV would result in less freedom.


Better than IRV: Approval Voting

Approval voting (AV) is a simple system that lets voters select more than one candidate in an election. The candidate with the most votes wins. There is no need for new ballots or new counting equipment, it doesn’t exhibit non-monotonicity (you can’t hurt a candidate by voting for her), it scores much better than IRV and plurality on Bayesian regret in experimental simulations, and although it is gameable like every electoral system, it is actually difficult and non-obvious how to vote tactically (you need good information about other voters’ preferences and to be able to calculate expected utilities).

For libertarians, AV is especially attractive. The Libertarian Party tends to be perceived as a party of the ideological center in the United States. For instance, Gary Johnson voters were about equally split between Trump and Clinton preferences. Moreover, the Libertarian candidate frequently garners double-digit percentages when there is only one major-party candidate on the ballot. Thus there are many Democrats willing to cast Libertarian votes when there is no Democrat on the ballot for a race, and many Republicans willing to do the same when there is no Republican. Under approval voting, then, these voters might well cast votes for both the Libertarian and their preferred major-party candidate in three-way races. Libertarians might actually have a chance of winning some elections.

Consider an election like the following:

Percentage of voters Ranking Approved-of
30% Trump>Johnson>Clinton>Stein Trump, Johnson
5% Trump>Johnson>Clinton>Stein Trump
7% Johnson>Trump>Clinton>Stein Johnson
3% Johnson>Trump>Clinton>Stein Johnson, Trump
7% Johnson>Clinton>Stein>Trump Johnson
3% Johnson>Clinton>Stein>Trump Johnson, Clinton
5% Stein>Clinton>Johnson>Trump Stein, Clinton
30% Clinton>Stein>Johnson>Trump Clinton, Stein, Johnson
10% Clinton>Stein>Johnson>Trump Clinton

Under plurality with sincere voting, Clinton wins narrowly, as she actually did. Under IRV, Stein is eliminated right away, and then Johnson. Clinton wins the election handily. Under approval voting, Johnson wins in a landslide.

Now, I am not actually claiming that Gary Johnson would have won the 2016 presidential election if it were held under approval voting, although it is possible given how unpopular the other candidates were, but rather I am making the point that approval voting helps candidates who can manage to be everyone’s second choice. Under IRV, everyone’s second choice is eliminated right away and has no chance. AV tends to pick consensus candidates, and a high-quality, centrist Libertarian could very easily manage to be just that candidate in the United States.

Now, of course we have to consider how the other parties would react to approval voting. Chances are they would run to the center and try to snag approval votes from libertarians. They would try not to alienate significant minorities with inflammatory language and discriminatory policies, which would motivate “disapproval votes” (ballots marked for every other candidate). In fact, the major parties would probably keep winning most elections using this strategy – but the policy outcome doesn’t sound so bad to me.



We know this for sure: had the Republican Party used approval voting in its primaries, it would never have nominated Donald Trump. Had it used IRV… who knows? In the real world United States, approval voting has massive advantages for libertarians, moderates, and the total satisfaction of voters that IRV lacks – and IRV might even be worse than the status quo because it neuters third parties, generally to the advantage of center-left Democrats.


[1] Using the term “ranked-choice voting” for IRV is not really correct, because there are many ranked-ballot systems of which IRV is just one.

Also from this issue

Lead Essay

  • Rob Richie of FairVote explains why he thinks ranked choice voting is better than the system that most of us now use. Rather than marginalizing all outsiders and shoehorning everyone into two big, unwieldy parties, a ranked choice system would tend relatively more toward the representation of multiple viewpoints and toward the enactment of all policies that could command a majority, even if they do not command a majority within a given party or division of the electorate. RCV isn’t just better for our politics, says Richie; it’s been tried successfully in a wide array of settings, and it deserves to be seen not as an experiment, but as a well-known alternative. Given the shortcomings of our current system, RCV looks better and better, he argues.

Response Essays

  • Thomas L. Knapp recommends the use of approval voting, in which candidates simply signal their approval of as many candidates as they find acceptable, without ranking them; the candidate with the most votes then wins. Knapp’s case for approval voting is that it is simple and easy to understand, and that it will produce fewer headaches for voters than Ranked Choice Voting. In the universe of possible systems, none are perfect, but simplicity is a factor that should not be discounted entirely: Simple elections encourage turnout and foster confidence in the legitimacy of the results. Whatever problems our democracy may have, Knapp concludes by arguing that voting method is perhaps not so high on the list.

  • Jason Sorens admits that Instant Runoff Voting has some advantages over our current plurality system. Yet he too recommends approval voting, and he supports his choice with reference to both the system’s mathematical appeal and certain real-world considerations. Approval voting will be easier to administer and count; parties will be discouraged from running divisive candidates; and even certain recent elections would, if run by the approval system, likely have returned results that he finds much better.

  • Jason McDaniel dives into the empirical literature on ranked choice voting. He finds that RCV tends to produce confusion rather than consensus. Many voters rely on party identifiers and a small number of candidates in order to make relatively informed decisions; these voters will be perplexed as parties and candidates proliferate. Rather than a rush to the amiable center, RCV is likely to provoke a mass exodus from the polls. Although RCV may have some upsides, the reality of an already polarized but not terribly well-informed electorate suggests that it will not likely achieve the outcomes that well-informed political scientists hope to see. New voting systems must be tested using experimental evidence drawn from real electorates, and by this measure, RCV is not as great as its proponents would have it.

The Conversation