Opinion: Instant-runoff voting – right reasons, wrong reform

This piece is a response to Steven Hill’s editorial, “Ranked choice voting making a difference in the Bay Area.” The opinions expressed in this commentary do not necessarily reflect those of KALW News.

In 2002, I enthusiastically phone banked and voted for San Francisco’s IRV initiative – but I was mistaken. IRV promised to solve important problems, but after nine years, it’s hasn’t. In fact, despite what supporters or opponents with a vested interest will tell you, it’s elected much the same people at much the same cost and turnout as before. But now that we understand the flaws which kept IRV from fulfilling its promise, we should take the next step in voting reform, moving to simpler and better rated voting systems such as Approval, Range Voting, or Majority Judgement. These enhance IRV’s advantage of more-expressive ballots, without being subject to its flaws: more spoiled ballots; a fragile, centralized counting system; and results which are sometimes bizarrely extremist, due to premature elimination of viable compromise candidates.

IRV was an attempt to solve real, serious problems with prior voting systems. Traditional vote-for-one, also called plurality voting, effectively ignores votes for anyone but the top two. Most voters learn to shun potential “spoilers” – even if they honestly prefer them – and so candidates and parties grow complacent and corrupt. Meanwhile, occasionally some candidate still manages to act as a spoiler, and then system gives undemocratic, anti-majority results. Traditional runoffs help reduce these problems, but sometimes they’re just low-turnout wastes of money. Supporters claimed IRV would solve these problems more cheaply.

Although some people, such as those with a vested interest for or against IRV, still insist it’s turned out as good or bad as they said it would, the truth is IRV hasn’t changed much. On the one hand, there’s enduring opposition from the people who benefited under the prior system: above all, lobbyists who had become expert at exploiting the way things were. After their bogus talking points about “one man one vote” didn’t stop IRV’s passage, they backed a lawsuit. In Dudum v. San Francisco, a losing supervisor claims that the three-rank ballot unfairly restricts his ability to rank all candidates. Dudum is probably right that he would have won under traditional runoffs, and certainly this would have spared SF the shame of seeing Ed Jew, who beat him, have to leave office to go to jail. But if Dudum’s legal argument against IRV were true, it would go double against traditional plurality – something no U.S. judge is likely to rule.

On the other hand, people who have built their career on backing IRV claim it’s given all the benefits that they promised. But their cherry-picked success stories don’t change the fact that there’s no evidence for savings or improved results from IRV, while it has made things worse in small but measurable ways, such as increases in voter confusion and spoiled ballots. And for every example where IRV may have helped diversity or reduced mudslinging, there’s a counterexample. For instance Australia, which has used IRV nationwide for decades, is worse off in both of these regards than the UK, a comparable country without IRV.

In fact, IRV has elected much the same people, at much the same cost, with much the same average turnout as before. In San Francisco, every IRV race except one has been won by the candidate who led after the first count – the same person who would have won under plurality. And in the one exception (2010 Supervisor, District 10), over 60% of the balIots were spoiled or exhausted, so the winner ended up with just 21% of the vote! That’s right: contrary to professional IRV supporters’ claims, and even the claims in the original ballot language itself, that it guarantees a majority, only once since IRV was enacted in SF has the winner had 50% (2005 assessor-recorder). Even IRV’s “poster election” – Jean Quan’s dramatic 9th-round win in Oakland – was a non-majority result (she got 44% in the final count) which probably only chose the same person who would have won a pre-IRV runoff.

What about the savings? Well, IRV was adopted after a runoff (among other factors) led to a record year of spending for the SF elections department. But since then, two years have broken that record, about the number of runoffs we could have expected; and every single year has been above the prior non-runoff-year average, even adjusting for inflation and population.

IRV doesn’t significantly increase turnout. In the main elections, turnout since IRV hasn’t significantly changed its steady decline. And even compared to the supposedly “low turnout” runoff elections, IRV isn’t significantly better on average. Some runoffs have low turnout, but in important ones turnout can actually increase, which would be impossible under IRV. That’s why, in SF, runoffs averaged only 2.58% lower turnout. If traditional runoffs gave us more-democratic results when people cared, but less-so when they didn’t, that might have been a fair deal after all.

With the benefit of experience, it’s easy to see why IRV hasn’t fulfilled its promise. Though IRV is better than plurality without runoffs, it does have three interconnected flaws:

  • As with plurality, any ballot which rates two candidates at the same level can count as spoiled.
  • The ballot-counting process requires special programming, can only be run centrally, and is difficult to audit or recount.
  • Perhaps most-seriously, IRV can prematurely eliminate a compromise candidate, even when the ballots themselves prove that the eliminated candidate could win a runoff against any other. That’s what happened in Burlington, Vermont, a liberal town where the Democrat was eliminated and a Green was elected mayor, prompting a majority of voters to repeal IRV. (Greens should have a fair chance, and win if they’re actually preferred by a majority. But the ballots show that that wasn’t the case in Burlington; the Democrat would have won any two-way race.)

But since IRV was first adopted in SF, systems without those flaws have achieved new prominence. As sites like YouTube and Yelp have popularized ratings, researchers have found that (when spam isn’t a factor) these can actually be the basis for sophisticated, highly-democratic voting systems. The basic difference is that unlike ranked systems like IRV, which expects voters to strictly rank candidates with no ties, rating systems let voters put candidates into absolute rating categories, with zero, one, or many candidates at any given level. Because of that added flexibility, rated voting avoids IRV’s problems with spoiled ballots. Because the final rating for each candidate doesn’t depend on other candidates, these systems can be run cheaply, on any voting machines in the world, and precinct-level counts, recounts, or audits can be used to catch any fraud in the centralized count. And most importantly, because rated voting doesn’t rely on an elimination process, it never gives crazy results due to premature elimination as happened in Burlington (see above).

Several rated systems are better than IRV. Simplest is Approval Voting, with just two rating levels: approved (thumbs up) or unapproved (thumbs down). Essentially, as with plurality, most people would vote for their favorite major candidate; but unlike plurality, they could also vote for anyone they liked even better, thus breaking the two-party death grip on power. This is actually simpler than plurality, because there’s no need to detect and discard “over votes”.

If voters want to express finer distinctions, Range Voting is an option. As with 5-star ratings on Yelp, a candidate’s overall score under Range is their average score across the ballots. Computer-simulated elections show that, under a broad range of circumstances, Range would result in greater voter satisfaction than any other system. Its only defect is that it can give an advantage to a faction which exaggerates its opinions, unless other factions follow suit. Though it’s unlikely that this would happen in practice, the possibility could either lead Range to devolve into Approval, or foment accusations from sore losers.

To reduce problems with uneven exaggeration, the candidates can be ranked by their median score – a measure less affected by extreme values than the mean. In other words, a candidate would get the highest score that a majority of voters agree on. One median-based system is Majority Judgement, proposed by researchers Michel Balinski and Rida Laraki in their bookof the same title. Under MJ, exaggerating your rating does not help move a candidate’s score toward the score you think is fair. This minimizes opportunities for strategic voting. Field trials have shown that voters adapt happily to this and other rated voting systems. Systems like MJ are probably the best way, without runoffs, to guarantee that the winner has assent from a majority of voters. (Or MJ-like systems can absolutely guarantee majorities, if runoffs are used in some very rare, highly-contested situations.)

How could we get more experience with these improved systems? Luckily, Berkeley’s Measure I, which enabled IRV, did not specify which system should be used. With that authorization, the Berkeley City Council could implement one of these improved systems – I’d suggest Majority Judgement, but I’d strongly support any of them – whenever they choose. SF and Oakland could follow suit with an initiative if they liked the results.

Make no mistake. Lobbyists will oppose these changes for the same disingenuous reasons they opposed IRV: because they profit from being able to manipulate democracy. And, sadly, professional IRV advocates will probably also be opposed. But all three systems I’ve described are simpler, cheaper, and more-democratic than IRV. It’s time we in the Bay Area gave them a try.

Jameson Quinn, a programmer and voting reform advocate, is a founding member of the Center for Election Science, a group which supports rated voting systems. His opinions are his own. Find out more at the CES website, in William Poundstone's popular-science book Gaming The Vote, or in Majority Judgement, the book mentioned above.