Pendulum Or Plumb?

An engineer, a mechanic, and a statistician go on an African hunting safari. They spot a lion in the distance, and the engineer aims, and fires... but misses five feet to the left. The mechanic scoffs, takes aim, and fires... but misses five feet to the right. So the statistician jumps up and shouts "We got 'em!"

---

Since about WWII, congressional polarization has steadily increased. Over the last election cycle we've seen another wave of incumbent moderates forced from office, either by successful primary challenges (e.g. Richard Lugar), or via replacement with a more partisan candidate that's more in-line with the area's political leanings (e.g. Scott Brown or Ben Nelson). As a result, this month's election continues that trend and makes the US Congress yet more polarized than it already was.

With moderation virtually impossible to find in any one member of congress, many Americans have resigned themselves to the hope that, while political power will swing like a pendulum between too-far-left and too-far-right, the results will average out to be somewhere in the middle. (Or worse, they've resigned themselves to the despair that politics has become hopelessly and irrevocably polarized.)

But what if there's a better way? Plurality voting (as well as instant runoff voting) suffers from a problem known as "center squeeze", where if three candidates are running for office, the one in the center is at a significant disadvantage. And so overtime, despite the honest intentions of every voter, the political center is emptied of elected representation. But if we use a voting method that does not suffer from "center squeeze", such as approval voting, then we would see the opposite effect: Over time, more members of congress would be found in the center.

Two New French Voting Studies

In the last couple days, two new voting studies have come out of France, following the Presidential elections there. One (translation to English) and two (and also in English.) The first included a look at approval voting, and the second score voting, with a range of -2 to 2, and both suggest that France would have gotten a different, and probably better, result if they had used either of these methods.

Specifically, the first study found that, if approval voting had been used in the first round, that the two candidates to advance would have been Hollande (the Socialist leader who advanced in the real election, and went on to defeat incumbent center-right President Sarkozy) and the original fourth-place finisher, François Bayrou. Bayrou is an interesting character; he came in third in the previous election, and his Democratic Union party is considered a centrist group. The study also showed that, in a head-to-head match up, Bayrou would have beaten Hollande. This is some real-world data supporting the theory that approval voting does a better job of electing centrist candidates than plurality. They examined instant runoff voting as well, but got the same result as the plurality election, supporting that theory as well.

The second study looked at score voting. Again, Bayrou and Hollande led (and were the only candidates with net-positive scores!) while Sarkozy slipped back to fourth place. Their head-to-head challenges also found Bayrou to be a beats-all winner. They also found very little bullet voting.

These studies build on previous, smaller ones from the 2002 and 2007 elections, which found similar results. All told, this is some excellent real-world data adding further evidence to our theoretical expectations of approval and score voting.

IRV a Uniquely Poor Choice for Centrists

The other day, I noticed that the number-one referrer to this site recently has been Rise of the Center. So, first of all, thanks to them and everyone who's found their way here through them. Second, I have an important observation to share with you: instant runoff voting is a uniquely poor voting reform option for centrist. I was reminded of this fact while playing with Ka-Ping Yee's Voteline simulator, in order to respond to points made in the comments of my previous post on the Independence of Irrelevant Alternatives. And while it's something I've alluded too before, I don't think I've ever stated it this bluntly.

Model On Line

The Voteline simulator uses a very simple model. Voters are smoothly distributed along a left-to-right spectrum. Candidates occupy one point on that spectrum. Voter preferences are based on the linear distance from themselves to the candidates; closer candidates are preferred. The meanings of the colors may not be obvious though. The represent who the winner of the vote would be if the entire electorate were shifted, together, so that the median voter was at that point on the line. The relative distances between the candidates don't change, nor does the distribution of voters.

If you believe the winner of an election should be the candidate closest to the center of the political distribution, you would want an election method that always has a nice broad swath of matching color around each candidates' place in line, and 3 of the 5 methods used in Voteline do precisely that. The two that don't are plurality and "Hare (IRV)", AKA instant runoff voting. Instead, a candidate's place in line can easily be taken up with other colors, meaning that even if they are the most-centrist candidate, a different candidate would win the election.

It's not even that hard to squeeze a centrist candidate completely out of the running.

I've talked before about why we use models in these arguments. All models have their shortcomings though, and the one used here certainly contains some exaggerations; not all voters fall neatly on a single axis for instance (but see Yee's two-axis simulations) and we only have symmetrical voter distributions to play with here and so on. And there are other models in which IRV doesn't look quite as bad. But those models don't have any way of defining a center, so if you're buying into the idea of centrism, you don't have many other places to turn to for data.

It is certainly the case that voting method reform could help centrist. But, what these models suggest is that IRV is probably the least helpful voting method reform for that.