When Ranked Choice Voting Goes Bad: The 2009 Burlington Mayoral Election
Election season is once again upon us, though unfortunately many Americans seem to be rather unhappy with their choices. Indeed, Americans have rather consistently expressed their desire for a third party to break the two party system, but such a party has so far been unable to materialize.
So what causes this gap between the demand for more options and the failure for any of those other options to actually be elected? Many blame the system of voting America uses, known as “First Past the Post”. In First Past the Post voting (abbreviated as FPTP) whichever candidate wins the most votes wins the election as a whole.
On the surface, FPTP may seem reasonable. However, it creates perverse incentives, forcing voters to choose strategically. After all, if two candidates are neck and neck and your preferred option is languishing in the single digits, you may be tempted to vote for one of the main candidates who actually has a chance of winning. After all, you would be “wasting your vote” if you voted for your preferred candidate.
For instance, imagine you and your friends are deciding on what food to get: 4 want pizza, 4 want burgers, and 2 want sushi. You really want sushi, but you absolutely hate burgers. Knowing sushi won’t win and wanting to avoid burgers, you might feel compelled to vote for pizza, even though it’s not your first choice.
But thankfully, FPTP isn’t the only way to vote. There is a better way.
Enter Ranked Choice Voting
So if we don’t decide elections based simply on who gets the most votes, how can we decide them? Why, by letting people rank them, of course!
Imagine a world where instead of voting for one person, you can instead rank all the candidates according to your preferences. So in our previous scenario, instead of choosing between voting honestly (for sushi) and voting strategically (for pizza) you could instead just rank your preferences as such:
- Sushi
- Pizza
- Burgers
So how do we use ranked ballots to determine a winner? Well, there’s a couple of different methods, but in the United States the term “Ranked Choice Voting” (RCV) has become essentially synonymous with a method known as “Instant Runoff Voting” (IRV).
The idea behind Instant Runoff Voting is fairly simple: eliminate the person who got the least votes and redistribute their votes. Repeat this process until one candidate has a majority of the votes
So for our sushi-pizza-burger example, you could simply rank sushi in first place knowing that when it is eliminated as an option, your vote would be reallocated to pizza. You can now vote honestly without feeling the need to change your vote strategically.
Indeed, Instant Runoff Voting is quickly catching fire across America. It has been adopted by states like Maine and Alaska as well as many cities, such as New York City. Additionally, it will be on the ballot this November in Nevada.
But while there is a rush by many municipalities to adopt, many voting reform advocates have raised concerns about IRV. While there are a variety of (fairly nerdy) concerns, one of the biggest is the fact that IRV fails the so called “Condorcet Criterion”, which we will discuss later on in this article.
Enter The 2009 Burlington Mayoral Election
Burlington, Vermont is a city with an interesting political scene. Unlike most of the country, they do not have two major parties, but rather three with the left wing “Progressive Party” serving as a third force in Burlington elections. It is because of this reason that Burlington became one of the first cities in the nation to adopt Ranked Choice (specifically IRV) in 2005.
The first ranked choice election in Burlington was the 2006 mayoral election. Progressive Bob Kiss would use the system to prevail over Democrat Hinda Miller and Republican Kevin Curley. Notably Kiss won both the first round vote as well as the final count, so there wasn’t very much controversy in this election.
The 2009 mayoral election however was a different story (and the the main subject of our article), with the results proving to be much more controversial.
First, let’s meet the five candidates running in this election:
- The Progressive Candidate Bob Kiss, the incumbent mayor seeking a second term
- The Republican Candidate Kurt Wright, a city councillor and state representative
- The Democratic Candidate Andy Montroll, a city councillor
- An Independent Candidate Dan Smith, a local lawyer
- The Green Candidate James Smith
With the candidates introduced, let’s dive into the details of the election itself.
Visualizing the Election
Note: if you would like to see the full code for how I generated these visualizations, please check out this Kaggle Notebook
As you may have been able to guess by now, Burlington is a fairly left wing city. The party which managed to break the usual American two party system after all wasn’t some sort of centrist force, but rather the Progressives who were to the left of the Democrats.
Usually such a split would be discouraged by partisans. After all, splitting the vote of only one side of the political spectrum (in case the left) would simply increase the chances of the united opposition winning.
Indeed, in the first round of voting, the Republican Kurt Wright did indeed end up winning the most votes with a 368 vote lead over second place Bob Kiss.
Indeed, had Burlington still been using a simple First Past the Post electoral system, Kurt Wright would have simply won here due to the opposition being divided. Thankfully, the Burlingtonians of 2006 lived in more enlightened times and had adopted RCV. So it did not simply end here.
In the second round, James Simpson, Dan Smith and all write ins were eliminated simultaneously due to their mathematical impossibility of winning.
Their votes were of course now redistributed to the remaining three candidates: Kurt Wright, Bob Kiss and Andy Montroll.
If you’re wondering what the new “discarded” section is, people aren’t required to rank all candidates. So if someone only ranked Dan Smith or James Simpson and no one else, they would be moved to discarded.
For most voters, this was not the case as they instead headed towards one of the major three candidates. The ranking of the top three candidates remained the same though, with Wright on top, Kiss in second and Montroll in last place.
So obviously, the next candidate to be eliminated would be the last place Montroll, with his votes being redistributed to Wright and Kiss for the final round.
And there we go, Montroll’s votes were redistributed.
As you can see, most of Montroll’s voters went to Bob Kiss, which put Kiss ahead of Wright to win the election. This makes sense. Montroll’s center left voters mostly shifting to the left wing Kiss over the right wing Wright makes sense.
Indeed, the fact that ranked choice changed the winner from Kurt Wright (whom the plurality preferred) to Bob Kiss (whom a majority preferred over Wright) could be seen as a win. It is after all clear case after all of Ranked Choice helping stop vote splitting from determining a race.
However, if we dig a bit deeper, we can see that things went a bit wrong in this election.
Much to say about Condorcet
Taking a step back for a moment, let us now return to the concept of the Condorcet Criterion.
So, what exactly is the criterion? Well, it’s fairly simple. It’s just a fancy term for the idea that if one candidate would beat every other candidate in a head to head match up, then they should be the winner.
Going back to our Sushi-Pizza-Burgers example, let’s say we compare each option against the others directly. If a majority of people would pick pizza over sushi, and a majority would also pick pizza over burgers, then according to the Condorcet Criterion, pizza should be the winner. This means that pizza is the consensus choice, as it is preferred by the majority in all one-on-one comparisons. In other words, most people would be happiest if pizza won.
Instant Runoff Voting does not meet the Condorcet Criterion. It is perfectly possible in IRV for the “Consensus Candidate” which voters prefer over every other candidate to simply lose.
The graphic above helps illustrate this phenomenon, specifically through the idea of a “Center Squeeze.” Candidate B would beat Candidate A 64–36 and would beat Candidate C 66–34. They are the consensus candidate. But because they are in third place in terms of first-choice voters, in IRV they will simply be eliminated during the first round, and voters will again be forced to choose between two polarizing candidates.
Burlington 2009: A Condorcet Failure
As you may have been able to guess by now, the 2009 Burlington Mayoral Election failed the Condorcet Criterion.
While the application of Instant Runoff Voting did help avoid the greatest weakness of First Past the Post (vote splitting), in this situation it elected someone who very much was not the consensus candidate.
Instead of taking a look at the instant runoff calculations, let us instead look at the pairwise preferences of voters:
Now the problem starts to emerge. While Kiss and Wright both beat Montroll in the IRV, in terms of actual head to head preferences, voters prefer Montroll over both Wright (by 11.2%) and Kiss (by 7.8%).
Put another way, here is how Montroll would fare against each candidate
So what exactly is happening? Well, it’s a pretty classic example of the Center Squeeze phenomenon in action. Right wing and Left wing voters each preferred their own candidate the strongest, but would rather have the moderate Montroll than the candidate all the way on the other side of the spectrum.
But despite being the Condorcet Winner and being the consensus candidate Montroll lost the election due to not having enough supporters in the first round. That is to say, he lost due to the quirks of IRV.
The Aftermath
The 2009 election had largely soured Burlington on the concept of Instant Runoff Voting. Both Republicans (who had been the plurality winners) and the Democrats (who had been the Condorcet winners) had felt like they had been robbed. In 2010, a coalition of Democrats and Republicans started an effort to repeal IRV, an effort which would end up successful, with 52% of Burlingtonians voting for the repeal in a referendum.
However, as IRV started to become popular once more in the rest of the nation, Burlington would soon take another look at it as well.
In 2021, Burlingtonians approved a referendum to reintroduce IRV specifically for city council elections. This would soon be extended for all offices, with the 2024 mayoral election being the first since 2009 to again use IRV.
Conclusion: Is there a better way?
How important you think the Condorcet Criterion to be ultimately depends on your own views. Fairvote, the country’s largest ranked choice advocacy group for example explicitly argues that the Condorcet Criterion isn’t very important, because it is more important to elect the candidate most voters like instead of a consensus candidate.
Personally, I do not agree with this logic. If all that matters is which candidate is the most liked and nothing else, then there would be nothing wrong with plurality after all. It is important to elect a consensus candidate where possible to maximize the utility of voters. If voters prefer a candidate to all others, why should that candidate not be elected?
So are there any alternatives to both FPTP and IRV? Yes!
One possibility to ensure a Condorcet winner while retaining rankings is to simply use a Condorcet Method. There are quite a few different versions of these to handle the edge case of a “Condorcet Cycle”, but in practice such edge cases are fairly rare. Probably the easiest Condorcet method would be to just pass IRV with the addendum that if there is a Condorcet Winner, they will be elected.
Alternatively, many people also support voting systems from the “Cardinal Family”. These are voting systems where instead of ranking candidates, you score them independently. There are some different implementations of this, including Approval Voting (where you can either approve or disapprove of each candidate), Score Voting (where you can rate each candidate on a scale) and STAR voting (which combines score voting with a runoff).
Of course, at the end of the day, regardless of whether it’s Condorcet Voting, Approval Voting or even Instant Runoff Voting, the vast majority of election systems are probably better than FPTP. While it is important to recognize the pros and cons of each of these systems, it is for the best to support any of them as a replacement for FPTP.
If you’re interested in learning more about different voting systems, I highly recommend this interactive by Nick Casey. Additionally, if you’re interested in the visualizations which I created for this, I have the code for them published on this Kaggle notebook.
Thank you for reading!
