Massachusetts Daily Collegian

IRV, a voting spoiler alert

By Eli Gottlie

Hang on for a minute...we're trying to find some more stories you might like.

Email This Story

Nowadays it turns out that a lot of people want to switch the American voting system to something called instant runoff voting. Under this new system, you would fill out your ballot by ranking the candidates from those you most prefer to those you least prefer. Votes would be counted as normal, except that if no candidate had the majority the candidate with the least votes loses outright, and their votes are redistributed according to the next preference down to everyone else. This repeats until the election has a winner. 

Unfortunately, this system is what essayist Richard P. Gabriel would have called “New Jersey,” or an example of “worse-is-better” philosophy. It values the simplicity of implementation over the correctness of the results. But, governing something according to a few apparently simplistic rules provides no guarantee whatsoever of simple or desirable behavior. 

So yes, IRV does definitely provide a way of reducing the “spoiler effect” over First Past the Post voting by allowing losing third-party candidates’ votes to go support the “lesser evil” rather than spoiling things for the most-desired candidate. IRV provides an extremely simple ballot interface that doesn’t require voters to understand much about the voting method itself. I suppose if Americans want to be anti-intellectual, we could support such a new voting system. 

However, its mathematical properties as a voting system carry a downside. IRV does not have the voting-system property known as monotonicity, which means that under IRV, ranking a candidate higher can actually cause them to lose. Experts have calculated that the probability of any particular election resulting in a monotonicity failure is about 5 percent to 15 percent according to Nicholas Miller from the University of Maryland in a 2002 publishing for the Public Choice Society. It actually occurred in the 2009 mayoral election for Burlington, Vt., according to IRV can genuinely result in people electing a candidate that they most definitely didn’t want. 

Instead, I propose that when we need to elect a single-seat office, we use Schulze Method Condorcet Voting. Schulze does things the MIT way; It does the right thing. It uses the exact same ranked preference ballot format as IRV, except that it allows voters the give the same preference to more than one candidate, skip preference numbers and leave some candidates unranked. So where IRV will result in more spoiled ballots due to human error, Schulze Method voting lets people mess up their ballot a little, or just not care about certain things, while still counting all the preferences that they choose to express. It also not only possesses the monotonicity property, but it’s a Condorcet Method The only downside to the Schulze Method is that it requires a rather complex algorithm to count the ballots and decide the winner of the vote. On the upside of the downside that algorithm runs in polynomial time. 

If we want something simpler, we could sacrifice the Condorcet Method’s property and use a Borda Count. This would involve the same balloting procedure as IRV and the Schulze Method, but it counts the votes by assigning a point value to each preference rank and then simply adding up the points given for each candidate. The candidate with the most points then wins. 

This system only sacrifices the majority criterion: the candidate preferred at rank one by an absolute majority of voters might fail to get elected under a Borda Count if ranked lowly by the minority voters. Even this comes with a hidden benefit: Borda Count voting leads to the election of consensus candidates who please everyone and defeats the tyranny of the majority. 

In a country so rampantly partisan that many people on the right now consider the government to have gone fascist, I think we could do with an electoral balloting system that reduces partisanship and builds consensus by nature. 

Or, on the other hand, we could always just go with a proportional-representation Parliament like a proper democracy. Still, even those can get too democratic at times, leading to the election of unsavory politicians from parties with few total votes to high. Besides, I doubt we could get such a measure through the United States Congress or a Constitutional Convention. Passing a consensus-building Borda Count into electoral law will require only plain legislation and will have the direct benefit of electing candidates who fit a consensus not between the various echo chambers created by the media or the parties themselves but between the voters’ actual views. 

Instead of getting excited about IRV, a system that would in fact take away our ability to elect who we want just to save us from the dreaded “spoiler effect”, let’s do that. 

 Eli Gottlieb is a Collegian columnist and can be reached at [email protected]

Click here to read the other side of this week’s Point-Counterpoint: “Let’s get instant runoff running”


2 Responses to “IRV, a voting spoiler alert”

  1. Dale Sheldon-Hess on November 12th, 2010 6:10 pm

    Proportional representation would be great… but you’re looking at constitutional amendments for that, and you still can’t apply to singular, executive offices.

    Borda, as its inventor begrudgingly stated, is a system “meant for honest men.” Borda performs quite poorly when the electors figure out how to game the system. (Hint: get all your friends to run for the same office.)

    He made that admission in the face of arguments from Condorcet. And while Condorcet’s method is quite admirable in principle–elect the candidate that would defeat every other in a one-on-one contest–because of a lack of “honest men”, Condorcet methods are not as likely to elect the true Condorcet winner as some other methods are.

    What method is it that is most-likely to elect true Condorcet winners than any Condorcet method when faced with dishonest voters? Which also elects consensus candidates at a higher rate than the consensus-seeking Borda count? That is monotonic and completely immune to spoilers?

    Approval voting.

    It gets ignored because it’s not a ranking-based method. (You may have heard about Arrow’s Theorem. It only addresses rank-based methods, and so that’s all many people talk about.)

    But simulations suggest that it, and a slightly-more-complex method called range voting (or score voting) are the best voting methods available.


  2. Clay Shentrup on November 16th, 2010 1:07 am

    Score Voting (aka Range Voting) and its simplified variant, Approval Voting, are better according to essentially every metric. Particularly as judged by Bayesian regret.

    The most sophisticated calculations of non-monotonicity failure I’m aware of are by a Princeton math PhD named Warren D. Smith.

    It actually turns out that, among elections in which IRV makes a difference vs. Plurality Voting, non-monotonicity is actually extremely common.


If you want a picture to show with your comment, go get a gravatar.