Wait, not IRV, umm, that other thing...

[dougo]

In the past I've mentioned Instant Runoff Voting (IRV) as a way to fix the totally broken voting situation that's led to "lesser of two evils" contests for as long as I've been a voter. Well, thanks to a link in novalis's journal, I am reminded that IRV is rather broken itself. I had looked into this a while ago, but forgotten the details; the Election Methods Education and Research Group (EMERG) has a good collection of information, as does Wikipedia. The first thing to know is that Arrow's impossibility theorem proves that no voting system is perfect, which is almost as depressing as Gödel's incompleteness theorem. Fortunately there is somewhat of a loophole, in that if you relax the definition of "perfect" a little bit, there are some voting systems that satisfy all reasonable criteria; in particular, the EMERG advocates a form of Condorcet voting. However, the Center for Voting and Democracy (CVD) (who advocates IRV) has its own complaints about Condorcet (and other alternatives). Oh dear, just when it seems like the media (e.g. The Nation) might finally be getting the message that plurality voting sucks, there's some serious dissent about which alternative is best. I'm currently more swayed by EMERG's arguments towards Condorcet than by CVD's arguments against it, though I need to do some more reading, but I'm really afraid that this kind of confusion may mean we're stuck with plurality voting (and the corrupt two-party system it engenders) for the forseeable future. Bah!

Comments

yer welcome!

[nosrialleon.livejournal.com]

mmmmmmmjapanesestonerrock...

[mrmorse.livejournal.com]

Another less broken voting system is approval voting. From a list of multiple candidates, each voter votes for all acceptable candidates. Whichever candidate ends up with the most votes wins. In this Presidential election, many voters would be happy with either Kerry or Nader, but not Bush. With approval voting, they would vote for both. With IRV voting, they would rank the two candidates.

Supporters of IRV, approval, and Condorcet voting like to argue about which system is best. I'm pretty agnostic, but approval voting appeals to me because it's simple.

[dougo.livejournal.com]

I'm uncomfortable with approval voting; it doesn't capture the fact that I prefer A to B, but also prefer B to C. Either I vote solely for A, in which case it's the same as a plurality vote, or I vote for A and B both, in which case I'm hurting A's chances if it's a close vote—same problem as IRV. Even if I vote naively, just following the instructions: do I find B acceptable? Well, I dunno... relative to C, sure, but I'm not really sure what that means as an absolute measure. I might not even consider A to be acceptable, so maybe I should just leave the ballot blank!

[mrmorse.livejournal.com]

I found a page about approval voting while poking around at Wikipedia. It makes the claim that if everyone follows a "good" voting strategy, then approval voting will give the same result as Condorcet voting.

It also has a response to your problem. The fear is that approval voting will tend to result in compromise candidates being elected. Consider an election between Kucinich, Lieberman, and Bush. If most supporters of Kucinich or Bush also vote for Lieberman on the basis that Lieberman is at least better than the third candidate, then Lieberman will win, which may not be the desired outcome.

The claim is that IRV is actually more likely to produce this result than approval voting. In IRV voting, Kucinich voters are likely to vote 1. Kucinich 2. Lieberman 3. Bush, which may actually overstate their approval of Lieberman. If Kucinich gets the fewest votes, his votes will be kicked over to Lieberman, resulting in Lieberman winning. Kucinich voters may be unhappy with this outcome.

In approval voting, Kucinich supporters may be more likely to vote only for Kucinich, therefore not supporting the compromise candidate. This is equivalent to voting for Kucinich first in IRV and leaving the other slots blank, but a study has shown that voters are less likely to throw additional support to compromise candidates in approval voting than in IRV, making it less likely that a compromise candidate will be elected.

This year's Democratic primary offers a practical example of the benefits of approval voting. I personally thought that several of the candidates would be good Presidents, while others would not. Furthermore, of the candidates I supported, I was fairly indifferent as to which one got nominated. (The differences between Kerry and Edwards are negligible, and I don't think Dean or Clark were that far off.) Under the current voting system, I could only choose one. I felt unsatisfied about the choice I made and have been questioning it since. Under IRV, I would be able to rank the candidates, but that wouldn't really capture the sharp difference I feel between the good candidates and the bad candidates while it would exaggerate the differences between the good candidates. Under approval voting, I could have voted for all of the candidates I liked and feel confident that I was increasing the likelihood of any of them winning.

[fin9901.livejournal.com]

I've always been a fan of approval voting, and my position was only encouraged upon finding out that the American Mathematical Society uses it for their voting.

[dougo.livejournal.com]

Check out EMERG's discussion of the Strategy-Free Criterion: "The Strategy-Free criterion goes further and shows that, under certain reasonable conditions, a majority of voters have no incentive to vote insincerely." Is that what you were looking for, or are you disagreeing with the "reasonable conditions" clause?

[mrmorse.livejournal.com]

This discussion of voting systems raises an interesting point. Arrow's impossibility theorem lays out a bunch of reasonable expectations about voting and then goes on to prove that no voting system can satisfy the entire set. The conclusion is that no voting system is ideal and we should attempt to develop a voting system that comes as close as possible to meeting the reasonable expectations while recognizing that any system will have some undesireable outcomes.

Maybe a better way to approach voting is to throw out Arrow's expectations and develop a new set. The place to start is probably monotonicity and independence of irrelevant alternatives. (Just because I find it helpful, I'm going to restate both of them. Monotonicity states that voting for a candidate will never make that candidate less likely to win. Independence of irrelevant alternatives states that removing any particular candidate from the set of candidates doesn't change the ordering of the remaining candidates. Plurality voting fails on irrelevant alternatives. IRV fails on monotonicity. (Condorcet fails on generating a unique ranking, one of the other expectations.))

What I want in a voting system is for third parties to be viable. As I see it, a good voting system would both provide that elections with more than two major parties get resolved reasonably while ensuring that minor parties can't act as spoilers. (I suspect that these two statements are basically equivalent to monotonicity and irrelevant alternatives, but the focus is different.) Plurality voting fails on spoilers. IRV and Condorcet fail on multiple major parties.

I wonder about the merits of a hybrid system. Have each voter rank the candidates. Look at the percentage of first place votes and eliminate everyone below a certain threshold. Have all voters who voted for an eliminated candidate fall back to their first remaining candidate, and then have a plurality vote.

Small third parties would not function as spoilers, but the plurality vote among the major parties prevents the strategic voting weirdness of IRV. The only question is what happens to candidates whose popularity is right around the threshold. I fear that might be ugly enough to make the whole system break down, but I can't really tell.

[mrmorse.livejournal.com]

I've already mentioned approval voting once, but I feel like I should mention it again. It actually meets both the multiple major party criteria and the spoiler criteria, assuming that resulting with a compromise candidate is considered a reasonable outcome.