Updated on 9/28/21 to add image via Yes! Media and mention of ranked-choice voting. I had originally intended this to be part of a four-part series that also covered Approval Voting and the Condorcet Method, which was my favorite method at the time, but political science professor Matthew Shugart convinced me that ranked-choice was better. Originally posted at my old blog.

In Part I: The Problem with Plurality, I illustrated the serious issues with our current voting system, plurality. If you haven’t already read it and are not familiar with Duverger’s Law, I recommend you read it first before continuing.

Here, I’ll discuss one of the better-known alternatives to plurality, Instant Runoff Voting, or IRV (commonly referred today as ranked-choice voting or RCV). One of its bigger attractions is that it’s relatively simple. There are only two things you have to understand: runoffs and ranked ballots. I’ll start with the latter.

Ranked ballots

Instead of voting for just one candidate, you simply rank your preferences in order. It’s as simple as that. We’ve all seen top ten lists (although some of us are a little bit more obsessive about listing things). You also go through this process when you rent a video or borrow a library book. After all, you start by looking for your first choice, and if it’s not there you look for your second choice, etc.

Note, you can rank as many or as few candidates as you want. You can even pick just one person, just like in plurality. Of course, the advantage to ranking candidates is that your preference will still be taken into account if your first-choice doesn’t win, but I’ll get back to that later. First, let me start with an example. To keep this apolitical, let’s say you’re voting for your favorite… um… COOKIE! The choices are oatmeal, vanilla wafers, Oreos, white chocolate macadamia, and, of course, CHOCOLATE AND PEANUT-BUTTER!! My ballot would look like this:

1. CHOCOLATE AND PEANUT-BUTTER (duh!)
2. Oreos
3. white chocolate macadamia
4. vanilla wafers
5. oatmeal

I guess you could say I have a bias towards chocolate. My wife, on the other hand would have a ballot like this:

1. white chocolate macadamia
2. oatmeal
3. Oreos
4. vanilla wafers
5. chocolate and peanut-butter

She’s not so fond of peanut-butter. So it’s pretty simple from the voting side. Now, the intuitive way to tally them has some very serious issues, and is not how IRV tallies them. Nonetheless, let me explain it first so that there’s no confusion.

Borda is bad, mmm-kay?

The intuitive way to tally them is to assign a point value for each ranking, say 4 points for a 1st-place vote, 3 for 2nd place, etc. As I said, this is not how IRV does it, but instead is known as the Borda method. It’s actually fairly commonly used, for example in Major League Baseball’s awards, like Cy Young and MVP. In the above example, the winner would then be white chocolate macadamia with 6 points (2 from me, 4 from the wife) edging out Oreos with 5 points (3 from me, 2 from the wife). You might say, “What’s wrong with that? Doesn’t your wife always get her way anyway?”

Well, to be honest, yes she does, because she is always right. And in truth, there’s actually nothing wrong with this outcome (any time you end up with cookies, you can’t really complain). Instead, the problem with Borda is that it’s extremely susceptible to tactical or strategic voting. This is when voters take advantage of how the voting system works by not voting according to their true preferences.

For example, consider the case where I correctly anticipated that my top pick had no chance of winning and that white chocolate macadamia would be the biggest challenger to my second choice, Oreos. I could then move Oreos to first and white chocolate macadamia to last. This would bump white chocolate macadamia down from 6 to 4, and up Oreos from 5 to 6, handing the election to Oreos (yay!). Essentially, my vote resulted in a 4 point difference between the top two contenders, whereas my wife’s vote only resulted in a 2 point difference. So my vote ended up being worth twice as much as my wife’s, and she would smack me upside the head after stealing my cookie, and nobody would be happy (except the cat, who couldn’t care less).

Obviously, this goes against the ideal of one person, one vote (and yes, so does the Electoral College, but that’s an issue for another post). Also, in order to have a bigger say, voters must anticipate who will be the likely contenders and vote differently from their true preferences. Now, almost all voting systems are susceptible to some sort of tactical voting, but IRV is much less susceptible than Borda because it counts the votes differently.

Runoffs

While Borda is pretty intuitive, IRV is not really that hard to understand either. After all, most people are already familiar with the concept of runoffs. You have an election involving a bunch of candidates and count the votes. Instead of stopping there, you eliminate the candidate(s) with the fewest votes and hold a runoff election. Rinse and repeat until you have two candidates left, and then hold a final runoff election, awarding it to the winner.

The purpose of this is to avoid that spoiler effect I discussed in Part I by removing potential spoilers before the final runoff. The problem, of course, is that multiple elections take a lot of time, money, and effort. Not only that, but voter turnout tends to be much lower in the earlier runoffs, which can result in an extremist and fringe candidate winning them (indeed, this is exactly how Jean-Marie Le Pen ended up being the finalist against Jacques Chirac in France a couple years ago). IRV doesn’t have this problem because there’s only one actual election, and the runoffs are immediately tallied using the extra information gathered by the ranked ballots (thus the name, Instant Runoff Voting).

This is how it works. You count all the first place votes, and then you eliminate the candidate with the fewest votes. Next is the key step. All the ballots that had marked that candidate as their first choice now get transferred to their second choice. Note, this is exactly what they would have wanted. If an actual runoff election had been held where their first choice was no longer available, they’d have voted for their second choice.

Then you just keep repeating this process until one candidate is left. Note, any ballots whose first and second choices have both been eliminated get moved to their third choice, etc. This way, you get the benefit of runoffs (eliminating possible spoiler candidates early on) without the drawbacks (low voter turnout in early runoffs).

Effect on third parties

As you might expect, most third parties support IRV because it removes a big obstacle that prevents most voters from voting for them. Namely, that your vote will be “wasted,” and furthermore, you might help your last-choice candidate win, as Nader voters found out in 2000. Indeed, many disgruntled Democrats are also jumping on board. To see why, recall my grossly-simplified diagram of the 2000 election situation from Part I:

With IRV, Nader is eliminated in the first round, and his votes will get transferred to their second-place choice, presumably Gore:

And then Gore would win, 11-10, which is why Democrats are starting to support IRV. But note that IRV is not biased towards liberals. It merely addresses the spoiler effect, regardless of where it happens. It would have had the same effect on Perot as it had on Nader, which might have helped the elder Bush against Clinton (although Perot may have drawn support from both sides).

Also note that many Gore voters would have liked to vote for Nader, but anticipated the spoiler effect, and decided to vote for their second choice instead. Under IRV, they would be freed to vote according to their true preferences.

Spoiler effect lessened, not eliminated

Of course, with the spoiler effect addressed, third parties also have less incentive to move to the fringes, and indeed will likely become more moderate to try and draw broader support. While this is normally a good thing, it becomes a problem for IRV.

Let me illustrate. To try and avoid comparisons to a specific race, let me move to a hypothetical race between Dan Democrat and Reg Republican, with a third-party candidate emerging on the Dan’s left. Let’s just call him Joe, not for anybody in particular.

Thanks to the fact that people are no longer dissuaded from voting third parties, Joe is no longer at the extremes, and is strong enough to survive the first round, knocking off Dan. What are we left with in the second runoff?

Joe, being further from the center, doesn’t have much of a chance of beating Reg, even though he picks up most of Dan’s ballots, losing 9-12. In essence, Joe and Dan have again split the vote, handing the election to Reg (as I said, IRV is not biased towards liberals).

You see, IRV is basically an attempt to patch the third-party spoiler effect by simply throwing out third-party votes. However, it only works if the votes it discards are actually third-party votes. If not, you still see the same spoiler effect you see in plurality. So it doesn’t fix the spoiler issue, you just see it in fewer situations, namely when the third party is no longer irrelevant. Of course, that’s the exact situation I’m trying to promote so that voters aren’t limited to two bad choices.

And the spoiler effect still happens when you have more than three parties. Consider the case where we have candidates across the political spectrum:

Although Cal would be the best representative candidate of the country (and shares the name of a darn good school! Go Bears!), he would lose in the first runoff. Although he probably appears on a lot of ballots as a 2nd or 3rd choice, none of that would be taken into account by IRV because he’d be eliminated before any of those ballots could be transferred to him.

Still an improvement

However, note that IRV still handles this situation better than plurality, as most of the ballots will eventually be transferred to the moderates, Bob and Deb, who would probably win instead of the extremists, Art or Ed (who might win under plurality). This is in sharp contrast to the impression of Professor Bainbridge, who displayed a remarkable ignorance of political science with the following remark in February, back when Nader announced his candidacy:

In the United States, the Electoral College makes it almost impossible for a third party candidiate[sic] to win the Presidency. Countries in which that is not true are not demonstrably better off. Look at the last Presidential election in France: In the first round of voting, Chirac led — but got less than 20% of the vote. Worse yet, nationalist nut-job and perrenial[sic] fringe party candidate Jean-Marie Le Pen qualified for the run-off with a mere 17%.

Hopefully, you’ve already spotted his misconceptions. First of all, it’s not the Electoral College that makes it impossible for third-party candidates, but the plurality system because of Duverger’s Law. Secondly, I’ve already explained that the problem in France was due to low voter turnout in earlier runoffs. IRV would not have had this problem. There’s more discussion on that case at Center for Voting and Democracy.

In addition to the continued existence of the spoiler problem, IRV does have some other issues which are more complicated to explain, like its failing of the monotonicity criterion, which I won’t get into here. This is why I’m more supportive of the Condorcet Method, which I’ll discuss in Part III (albeit probably after November due to NaNoWriMo). It’s actually not as complicated as it appears. Just think of a round-robin tournament.

Of course, Kenneth Arrow has proven that no voting system can be perfect, and Condorcet is no exception, but some voting systems are still better than others. Hopefully, I’ve shown that IRV is better than plurality or Borda. Indeed, since IRV is relatively easy to explain and is already in use in places like Australia and now in San Francisco (and, for you geeks, in the voting for the Hugo Awards), I am fully supportive of implementing IRV for now. After all, we can always improve on it further in the future, and indeed, Condorcet should be easier sell after voters have gotten used to ranked ballots.

For more information, check out FairVote – The Center for Voting and Democracy and InstantRunOff.com (hat tip to Zepfanman). Also, the Washington Post has an article about San Francisco’s recent adoption of IRV (via Chris Lawrence).

And many thanks to Donna, whose interest in the topic helped me get off my arse and finally continue this series.