New Lessons from Problems with Approval Voting in Practice

Posted by Rob Richie on December 14, 2016

Ranked choice voting (RCV, also called instant runoff voting) is a proven way to open up elections to give voters more voice and greater choice. RCV and traditional two-round runoff elections are the only single winner systems proven to uphold majority rule in meaningfully contested elections with real campaigns and reported results -- not only in governmental elections, where RCV and runoffs are widely used, but in literally thousands of private group and campus elections.

Yet there is a persistent group of online critics who espouse other voting methods and attribute to them a range of untested virtues. Because FairVote is the preeminent American research and advocacy group on voting methods, we are sometimes asked why we support RCV and stay neutral on these other methods. Our concerns focus on viability and workability.

Approval voting is a good example. In this system, every voter can indicate approval for as many candidates as they want, and the one with the highest rate of approval wins. There are uses of approval voting that can make sense -- like when a group of people deciding on what movie to watch. While mathematicians can like this system and its alleged likelihood of electing a consensus winner, my colleagues and I are highly skeptical of its use in candidate elections. Two factors stand out:

  • Viability and the issue of majority rule: If voters truly are free with their approvals in an approval voting election, it’s quite possible two or more candidates could earn more than half the vote. Indeed, it’s possible that a candidate whom well over half of voters see as a top choice could lose to someone who nobody sees as their top choice. Approval voting advocates defend such outcomes as fair, but it remains to be seen what voters would say.

  • Workability in the real world: In approval voting elections, you can’t indicate support for more than one candidate without support for a lesser choice potentially causing the defeat of your first choice. This transparent dilemma for voters trying to cast a smart vote has immediate consequences. Because most voters as a result of this problem will refrain from approving of more than one candidate, the system in practice ends up looking far more like a plurality voting election system than a majority system.

Approval voting has never faced voters on the ballot -- although it was repealed in 2009 by a vote of 81% of Dartmouth alumni after it was tried for electing trustees to the alumni board and contributed to a perception that tactical voters were getting an advantage over other voters. But we have evidence that suggests the problem of workability is very real, which is the focus of the rest of this analysis.

Reviewing the major single winner voting systems

Only three basic electoral systems are used in elections around the world to elect single winners in governmental elections -- for offices like president, governor, mayor and legislator in a single-winner district.

  • Plurality voting is a method where voters cast one vote, and the candidate with the most votes wins even if that candidate might be strongly opposed by a majority of voters.  This system often is termed “first past the post,” but that doesn’t describe the system nearly as well as “top of the heap.” It is used for most single winner elections in the United States, although for many elections we have primaries which narrow the field before the general election.

  • Runoff elections are methods where the winner must surpass a certain threshold of support  to win -- usually a majority of the vote. Voters have a single vote, and there is a second election in the event no candidate reaches the threshold. In the second round, only two candidates advance. Runoffs are used in most presidential elections around the world, in a handful  of congressional races, and in many primary and local elections in the United States. Generally, the “math-based” arguments used against ranked choice voting - like “non-monotonicity”  and not always electing the “Condorcet winner”  -- also apply to runoff elections.

  • Ranked choice voting combines features of plurality voting and runoff elections. Like plurality voting, there is a single election. Like runoff elections, a candidate must surpass a certain threshold of votes to win in the first round. Rather than ask voters to return for a second election, however, they are asked to rank candidates from their first choice to last (stopping when they are indifferent to the remaining candidates), and the those rankings can simulate an “instant runoff” between the two strongest candidates in the final round of counting. Ranked choice voting is used in several nations for high offices, in 12 cities and counties in the United States, and in hundreds of meaningfully contested non-governmental elections based on it being recommended by Robert’s Rules of Order as a backup to repeated voting in person.

There  are some minor variations in how these systems are used, but they collectively are the only systems used in public single winner elections for any governmental office at any level of government anywhere in the world.

Approval Voting Trials and Disappointments

Two recent uses of the approval voting method by the Independent Party of Oregon and the Independent Voters Network show that it has value in certain uses, but is a far more problematic electoral method than ranked choice voting for electing candidates in meaningfully contested elections. These uses show that tactical voters can flip outcomes, and, due to that reason, most voters treat the election like a traditional plurality voting election and vote for only one. While in competitive RCV elections for important offices some nine in ten voters will rank more than one candidate, in approval voting elections that number plunges to less than half indicating support for more than one candidate, and often far less.

One reason FairVote prefers ranked choice voting to approval is that with RCV, voters can freely rank back-up candidates without worrying that they will hurt their first choice. In our experience, fear about helping a second-choice defeating a first choice is often the most common worry among those less familiar with RCV. When RCV elections are taking place, it is therefore very important to emphasize that with RCV, ranking a second or later choice will never hurt your first choice; ranking more candidates instead gives you more power, because it means you will still be heard even if your first choice is too weak a candidate to have a chance to win.

Unfortunately, this important quality isn’t true with approval voting. Because votes are equally weighted and counted simultaneously, voting for a second candidate means you’re helping them just as much as your first choice. In an election that might be close, that means that indicating you “approve” of a second choice might lead to your favorite candidate losing to your second choice. For that reason, approval works well when voters are more likely to accept that possibility (like in a non-decisive straw poll or deciding among several generally acceptable options). But it works far less well in closely contested and decisive elections with multiple rival candidates campaigning. In those contexts, most people vote only for their favorite, raising many of the same issues with vote splitting and low plurality outcomes as ordinary vote-for-one plurality.

Approval voting is not used for any governmental elections anywhere in the world, and has limited history in elections where the results are reported and campaigns seriously debated – primarily a few student elections, internal elections within minor parties and NGO elections (although the latter often don’t report results, which means voters don’t learn anything  from the results beyond who wins.)

We’ve analyzed the results of approval voting elections before, and elections that have similar problems like the Bucklin system. Here’s an update based on two relatively high-profile uses this year.

Independent Party of Oregon Preference Poll

Last summer, the Independent Party of Oregon held an online primary among major presidential candidates using approval voting. It was a potentially important contest because the party would have offered the winner its ballot line in the November election if able to secure more than 50% support in the approval voting contest. But, as is the norm in meaningfully contested approval voting elections, most voters only cast a vote for only one candidate. With 1,549 total voters, the results were:

  • Bernie Sanders 31.38 % (488 votes)

  • Donald Trump 30.16 % (469 votes)

  • Hillary Clinton 23.92 % (372 votes)

  • Gary Johnson 16.01 % (249 votes)

  • John Kasich 12.22 % (190 votes)

  • Jill Stein 09.45 % (147 votes)

  • None of the Above 09.20 % (143 votes)

  • Ted Cruz 05.66 % (88 votes)

That's a total of 2,146 votes, meaning voters cast an average of 1.38 votes. The party reports that more than 70% of voters cast bullet votes, with varying bullet voting strategies reported:

  • Trump: Of 466 voters who backed him, 354 (76%) voted only for him.

  • Sanders: Of 488 voters who backed him, 193 (39.5%) voted only for him.

  • Clinton: Of 372 voters who backed her, 198 (53.2%) voted only for her.

The party reported additional data on voters for Sanders, Clinton and Trump. For example, of Sanders’ 488 voters:

  • 135 (27.7%) also voted for Clinton

  • 108 (22.1%) also voted for Stein

  • 76 (15.6%) also voted for Johnson

  • 44 (9.0%) also voted for Kasich

  • 24 (4.9%) also voted for Trump

  • 23 (4.7%) also voted for Cruz

  •  4 (0.8%) also (oddly) voted for “none of the above”

The bottom line is that approval voting did not come close to identifying a majority winner, with a key reason being that voters withheld preferences they almost certainly felt about other candidates. In contrast, consider that when FairVote and the College of William and Mary partnered on a YouGov poll of the views of 1,000 likely Republican voters on the 11 leading Republican candidate at the time of the Iowa caucuses, more than nine in ten respondentschose to rank every single candidate and hardly anyone voted for only a single person. The YouGov poll also showed a change in outcome, with Ted Cruz coming from well behind to defeat Donald Trump head-to-head.

Similarly, in ranked choice voting mayoral elections in the United States where there have been several candidates in a contested race and full data sets provided, nearly nine in 10 voters will rank a second choice, and about three in four will rank at least three. That participation provides a far richer set of information that allows contrasts between the top two candidates where the great majority of voters have indicated support for one of them.

For example, the Oregon primary result with approval voting makes this contest seem closer than it really is. Donald Trump lost by only 19 votes, but when looking at how few people were willing to approve of both Trump and another candidate compared to how many approved of  both Sanders and another candidate, it becomes clear that most of the nearly 70% of voters who did not cast a vote for Trump likely preferred several other candidates over him. If a runoff had been  held - or if the race had been conducted by ranked choice voting - Sanders would almost certainly have defeated Trump head-to-head easily.

Approval voting gave Trump a chance to win, but it also may have hurt Trump in one way - and one that the prospect of which would lead to even more bullet voting over time in approval voting elections. Among primary voters, 24 of them backed both Trump and Sanders. If 19 of  those 24 voters preferred Trump to Sanders, then those voters effectively helped defeat their favorite candidate.

It is exactly such fears that lead voters to withhold preferences when approval voting is tried in contested elections. We have seen that trend when approval voting is tried in contested non-governmental elections. Students at the University of Colorado and Dartmouth have been using approval voting, and winners in multi-candidates races for student body president now rarely receive more than 40% support. Here’s an excerpt from a September 2015 analysis from my colleague Drew Spencer Penrose.

Approval voting elections conducted at Dartmouth and the University of Colorado both frequently show most voters only voting for one, and candidates regularly winning with less than 40% approval. Dartmouth was particularly instructive, as it went straight from using ranked choice voting in six student body president elections in 2005-2010 to using approval voting in those elections from 2011-2015. In the six RCV elections, the average number of votes cast for the winner in the final round was 1,073, and only once did the RCV winner's vote total fall below 1,000.

In contrast, in the five approval voting elections at Dartmouth, the most votes a winner ever has received is 966 (last spring when only two people ran), and the average has dropped by more than 20% to 808. The great majority of voters keep voting for only one person, explaining why victors in multi-candidate races regularly win with less than 40%. One student body president won with just over 30%. We've seeing the same kind of results at the University of Colorado: every race with more than two candidates ends up with winners having less than half the votes, just as if a plurality vote system were in place.

Notably, the Independent Party also used approval voting for other ballot choices that showed where the system can be useful. The party asked voters to indicate their support for a range of potential measures relating to government transparency, again presenting an approval ballot. This time, with the same ballot design, voters cast nearly four votes per voter, with at least 66% support for all five proposals. In other words, once taken out of the candidate frame and any concern about a second choice hurting a first choice in a way that really mattered to them, voters used the exact same approval voting system in a way that allowed identification of the option that had the closest to consensus support (more than 87% supporting more lobbyist disclosure.) For less highly charged uses like this, approval voting can indeed have value -- but to assume it will carry over to candidate elections is an unproven and unlikely leap.

Similar Cautionary Results from Independent Voter Network Elections

The Independent Voter Network (IVN) also used approval voting in a less formal poll on July 15-24. 2016 among independent and independent-minded voters. They had nearly 32,000 responses to a survey in which participants first voted by plurality, then with approval voting.

Once again, the great majority of people cast only one vote in the approval voting contest, with about 1.25 votes per voter in the approval vote. In this case, the approval voting contest flipped the outcome in a way that shows why the number of approvals would almost certainly decline in a real election where voters really thought about how they might vote. Here were the results as reported by IVN.

Plurality Vote:

  • Gary Johnson35%

  • Jill Stein31%

  • Donald Trump13%

  • Hillary Clinton 9%

  • Write-in12%

Approval vote:

  • Jill Stein49%

  • Gary Johnson47%

  • Donald Trump16%

  • Hillary Clinton13%

  • Write-insNot allowed/reported

With plurality voting, the results add up to 100%, with one person per vote. With approval voting, they add up to 125%, given the average of 1.25 voters per voter. When compared to the first round, Trump only picked up 3% when voters had approval voting, and Clinton picked up only 4%. The big beneficiaries were Stein and Johnson, with Stein rising from 31% to 49% and Johnson 35% to 47%.

But here’s the rub -- and it’s one similar to the Trump-Sanders dynamic in Oregon.  If just over 2% of her increase came from Gary Johnson backers, that means their decision to approve of both candidates led directly to the victory of a Green Party candidate over a Libertarian. In a real election with polls or even just discussions of such a potential outcome, far fewer people would have approved of both Stein and Johnson – and indeed, overall likely even fewer people would have voted for more than one person.

As IVN points out, the approval voting contest would be a useful measure of which candidate should be considered to be in the presidential debates. Right now, the Commission on Presidential Debates limits participation to candidates with 15% support in the polls, and approval voting polling would give more candidates a chance to surpass that threshold. Approval voting can work in pre-election polls, but in an actual election, its downsides becomes highly problematic.

These examples help to illuminate why FairVote prefers ranked choice voting for decisive public elections. Voters can freely rank second and later choice candidates, and they do. It gives voters greater voice than plurality elections. The elections are more participatory and efficient than runoffs. Social scientists continue to experiment with new ideas for election methods, but for now the practical options for policymakers include plurality, runoffs, and RCV. Ranked choice voting is the best option.

 


 
Show Comments
comments powered by Disqus

Join Us Today to Help Create a More Perfect Union