The Case Against Ranked-Choice Voting: Part II
This is part II of my previous post laying out the case against ranked-choice voting. In this post, I look at a real election and see if we can find some of the failure cases we saw in the previous post. I'm going to use the 2009 election for mayor of Burlington, Vermont as my test case. I chose this because the individual votes have been carefully tabulated and made available online (kudos to Juho Laatu and Warren D. Smith for doing this). Like the last post, the code required to reproduce this post is available online.
Note: As I was researching ranked-choice voting, I came across a site about range voting where I learned a lot about voting systems. Some of the examples below I first found out about on their site.
In this election, there were the following five candidates and their parties:
Bob Kiss - Progressive
Andy Montroll - Democrat
Kurt Wright - Republican
Dan Smith - Independent
James Simpson - Green
I'm going to start by simulating the original election to ensure that I get the same result as the real election. There's a technical detail in how ranked-choice voting is implemented that we haven't seen yet that I mentioned. I said before that at the end of each round, the person with the lowest votes is eliminated. That works, but in large elections, it's quite slow, so in practice, all candidates who are mathematically eliminated are removed in a single round (I won't go into the details here but there are a variety of techniques for determining this). I'm going to use that approach here so the rounds match the real election.
ROUND 1
Candidate Votes Status
------------- ------- --------
Kurt Wright 2949 Active
Bob Kiss 2585 Active
Andy Montroll 2062 Active
Dan Smith 1306 Rejected
Write-in 36 Rejected
James Simpson 35 Rejected
ROUND 2
Candidate Votes Status
------------- ------- --------
Kurt Wright 3292 Active
Bob Kiss 2981 Active
Andy Montroll 2553 Rejected
Dan Smith 0 Rejected
Write-in 0 Rejected
James Simpson 0 Rejected
Blank votes 147 Rejected
FINAL RESULT
Candidate Votes Status
------------- ------- --------
Bob Kiss 4313 Elected
Kurt Wright 4058 Rejected
Andy Montroll 0 Rejected
Dan Smith 0 Rejected
Write-in 0 Rejected
James Simpson 0 Rejected
Blank votes 602 Rejected
Bob Kiss wins the election with 4,313 votes, which is what happened in the real election. Let's go through the different failure modes of ranked-choice voting and see if we can find them here.
Note: There are minor disagreements in how certain invalid ballots were counted, so the results here do not exactly match all the tabulations online. However, they all seem to be within 5 votes of each other and the discrepancies don't affect the results.
Ranked-Choice Voting Encourages Strategic Voting
There is clear evidence that strategic voting would have been beneficial in this election. Let's look at the election from the perspective of voters who voted for Wright. To get a sense of what their preferences were, let's look at all the voters who listed Wright at the top and see who would have won the election had just those voters voted (and Wright not been on the ballot).
ROUND 1
Candidate Votes Status
------------- ------- --------
Dan Smith 882 Active
Andy Montroll 849 Active
Bob Kiss 281 Rejected
James Simpson 78 Rejected
Write-in 19 Rejected
FINAL RESULT
Candidate Votes Status
------------- ------- --------
Dan Smith 1002 Elected
Andy Montroll 984 Rejected
Bob Kiss 0 Rejected
James Simpson 0 Rejected
Write-in 0 Rejected
Blank votes 123 Rejected
We can see that 882 people would have voted for Dan Smith, 849 for Andy Montroll, and only 281 for Bob Kiss. It's clear that Wright voters preferred Montroll, the Democrat, over Kiss, the Progressive. With the promise of ranked-choice voting, they should be able to express those preferences. In theory, if they vote for Wright and then Montroll, that can't hurt Montroll's chances against Kiss, because Montroll is ranked first and Wright was ranked next. But let's look at what really happened.
So let's see what would have happened had some of those voters been strategic. Let's say they get together and all the voters who voted for Wright, then Montroll or Wright, then Smith, then Montroll, decide to switch Montroll with Wright. So now they list Montroll first and Wright either second or third (behind Smith).
Let's say all the voters who voted for Wright then Montroll switched the order, so now they put Montroll first. We'll also include those voters who had Dan Smith in between them, so now ballots that looked like:
Wright > Montroll
Wright > Smith > Montroll
Look like:
Montroll > Wright
Montroll > Smith > Wright
ROUND 1
Candidate Votes Status
------------- ------- --------
Bob Kiss 2585 Active
Kurt Wright 2571 Active
Andy Montroll 2440 Active
Dan Smith 1306 Rejected
Write-in 36 Rejected
James Simpson 35 Rejected
ROUND 2
Candidate Votes Status
------------- ------- --------
Bob Kiss 2981 Active
Andy Montroll 2931 Active
Kurt Wright 2914 Rejected
Dan Smith 0 Rejected
Write-in 0 Rejected
James Simpson 0 Rejected
Blank votes 147 Rejected
FINAL RESULT
Candidate Votes Status
------------- ------- --------
Andy Montroll 4063 Elected
Bob Kiss 3476 Rejected
Kurt Wright 0 Rejected
Dan Smith 0 Rejected
Write-in 0 Rejected
James Simpson 0 Rejected
Blank votes 1434 Rejected
Now we have Andy Montroll as the election winner. This is a preferred outcome for the Wright voters that they achieved by not voting their preferences and instead gaming the system.
Ranked-Choice Voting Doesn't Always Result in the Best Head-to-head Candidate
Now let's look at whether the election results were the same as the head-to-head matchup. For the head-to-head election, I'll go through every ballot and assign it to whichever of the two candidates is listed higher. This means that the numbers won't match the vote counts from before because I'm ignoring the other candidates.
Head-to-head Election Results:
Kurt Wright: 3662
Andy Montroll: 4596
Head-to-head Election Results:
Bob Kiss: 3476
Andy Montroll: 4063
This shows that more people wanted Andy Montroll than Bob Kiss, and, in addition, more people wanted Andy Montroll than Kurt Wright. And not by a little - these are substantial margins. But, unfortunately, ranked-choice voting thwarted the will of the voters in this case and gave them a mayor that fewer people wanted.
Ranked-Choice Voting Allows for Election Spoilers
Was there a spoiler in the election? What would have happened if Wright wasn't in the race at all? Let's run that simulation.
ROUND 1
Candidate Votes Status
------------- ------- --------
Andy Montroll 2911 Active
Bob Kiss 2866 Active
Dan Smith 2188 Rejected
James Simpson 113 Rejected
Write-in 55 Rejected
Kurt Wright 0 Rejected
Blank votes 840 Rejected
FINAL RESULT
Candidate Votes Status
------------- ------- --------
Andy Montroll 4063 Elected
Bob Kiss 3476 Rejected
Dan Smith 0 Rejected
James Simpson 0 Rejected
Write-in 0 Rejected
Kurt Wright 0 Rejected
Blank votes 1434 Rejected
Montroll would have won instead of Kiss. Thus Wright was a spoiler.
Ranked-Choice Voting Allows for Candidates to Receive Higher Ranked Votes but Have Worse Outcomes
Let's look at the weird case, where doing better could have caused a candidate to perform worse in a ranked-choice voting election. This gets a little complex so to simplify things I'm going to remove candidates that aren't the top three. This happens anyway after the first round, so it doesn't affect the results at all, it just makes things easier to explain.
Top Three Candidates Only
We'll run the election again, just to show the numbers between the three main candidates haven't changed.
ROUND 1
Candidate Votes Status
------------- ------- --------
Kurt Wright 3292 Active
Bob Kiss 2981 Active
Andy Montroll 2553 Rejected
Blank votes 147 Rejected
FINAL RESULT
Candidate Votes Status
------------- ------- --------
Bob Kiss 4313 Elected
Kurt Wright 4058 Rejected
Andy Montroll 0 Rejected
Blank votes 602 Rejected
OK. So let's say that Kiss campaigns extra hard and convinces more voters that he is the best candidate. Let's say that he convinces all the Wright > Kiss > Montroll voters to put him first instead, thus changing their ballots to Kiss > Wright > Montroll. In addition, imagine he convinces another 400 of the voters who only voted for Wright to vote for him instead. Let's see how this affects the results.
342 ballots with Wright > Kiss > Montroll were converted to Kiss > Wright > Montroll (ignoring the other candidates).
400 ballots with Wright only were converted to Kiss only.
Note: All the code and original data are available, so feel free to check the original data source or read through the code if you want to double-check the numbers.
ROUND 1
Candidate Votes Status
------------- ------- --------
Bob Kiss 3723 Active
Andy Montroll 2553 Active
Kurt Wright 2550 Rejected
Blank votes 147 Rejected
FINAL RESULT
Candidate Votes Status
------------- ------- --------
Andy Montroll 4063 Elected
Bob Kiss 3876 Rejected
Kurt Wright 0 Rejected
Blank votes 1034 Rejected
Suddenly, Bob Kiss gets many more votes, starts with a huge lead after the first round, and loses the election.
Did this actually happen in the real election? It's impossible to know because it depends on comparing the existing state with a counterfactual. That's why in the previous post with simulated elections we had to run the election twice to make a comparison.
My guess is this is an unlikely scenario, but it's certainly not impossible. It didn't take long to find the criteria necessary to make it happen.
Voters' Reactions to Ranked-Choice Voting
The election used in this post was the 2009 election for the mayor of Burlington, Vermont. Burlington Voters had approved of the system only a few years earlier. But after two elections, they had had enough. In 2010, those same voters voted to repeal it. Burlington isn't the only city to experiment with ranked-choice voting. In 2009, voters in Aspen, Colorado implemented ranked-choice voting, only to repeal it the following year. The same thing happened in Pierce County, Washington (the south part of the Seattle metropolitan area). Voters approved the change in 2006 before discarding it in 2009 by a 71-29 vote. That's quite the margin!
I think this comes down to the fact that ranked-choice voting sounds very promising but ultimately falls short. I was in favor of the system until a few things led me to change my mind. First, I read stories, like those above, of cities trying and out and reverting to it. I also read about some of the unfortunate characteristics that we've seen in this and the last posts. But it wasn't until I really looked at the numbers - both real and simulated - myself before I saw how it really doesn't hold up to the promised that have been made about it.
For another source of information on ranked-choice voting, Alaska Policy Forum has also analyzed it and found several ways not discussed here in which the system falls short.
Alternatives to Ranked-choice Voting
In the next post, I'll talk about approval voting and why I believe it is a good alternative to ranked-choice voting.