Wednesday, November 04, 2009

Instant Runoff Voting

Minnesota's mayor was just (re)elected by Instant Runoff Voting, on the same day that spoilers threw the Gubernatorial elections in New Jersey and Florida and New York's 25th Congressional District (and actually won in New York.)

So, thanks, Minneapolis, for showing us how it's done. And, rest of the country, aren't you sick of spoilers? If New Jersey voters had been allowed to show their displeasure with Governor Corzine without voting for Chris Christie, wouldn't the world be a better place?

I think the iron is hot, now, and it's time to strike.


broken ladder said...

Here's an open question to supporters of Instant Runoff Voting

I expect that the results of this poll will support my belief that score voting and approval voting are better than IRV, even if we use the LOGIC of the IRV supporters themselves.

But I could be wrong. We'll see what the IRV proponents actually say.

malechem said...

I was happy to be the second person to respond to your poll. I think approval voting is where you say 'any of these would be fine,' and score voting is similar, but let's you give out partial votes. Here's the comment I put on the poll itself:

Here's the thing, though. Looking at the 1992 Presdential Election -- Perot, GHW Bush and Clinton. Say, I'm against NAFTA, and don't suspect the first thing Bubba is going to do when elected is sign it. I just don't want Bush. So, I give the maximum score to Clinton and Perot.

It was a pretty tight race. Let's say 67 % approve (or give maximum score to) Perot, 66 % approve Clinton, and 33 % approve Bush. Perot wins, and the nation enters a very dark time.

If we had IRV, we could have 35 % Clinton/Perot, 31 % Perot/Clinton, 1 % Perot and 33 % Bush. So, Perot has the first vote of only 32 %, and we move on to the second votes. Clinton wins 66 % to 33 %, with 1 % NOTA.

Wouldn't that be better?

broken ladder said...

I replied to your comment on the poll:


Here's my take on your voting scenario in which Perot wins.

The basic strategy with score voting is to give a max score to your favorite front-runner, plus everyone you like better, and a minimum score to your least favorite front-runner, and everyone you like less. I presume you're okay with treating Bush and Clinton as the front-runners.

So if most voters were strategic, then in your example we have two main groups of voters, who voted like this:

66% Bush=10, Perot=?, Clinton=0
33% Clinton=10, Perot=?, Bush=0
1% Perot=10, Bush=?, Clinton=?

This means that in order for Perot to have won with the 67% you describe, he would have had to be preferable to both Clinton AND Bush by 67% of the voters, which means that in that case he would have obviously been the favorite overall choice of the voters.

Moreover, you can always contemplate particular situations in which a "bad" or counter-intuitive result happens with any voting method, including IRV (not that this was one of them -- it wasn't). The issue is what the TYPICAL or average behavior of these systems will be over the course of several realistic election scenarios (not atypical ones specifically designed to make one system or another look bad or good). When you do that with computer simulations of elections, you find that score and approval voting vastly outperform IRV. Here are Bayesian regret figures which show that: (more in light of tactical voting)

malechem said...

It's a pretty interesting assumption that "monarchs, on average, were as least as good rulers as a random candidate, since, e.g. they were trained from birth to rule" (as per your link.) I'll assume the rest of the work was as carefully thought out.

I don't recall ever having thought about these voting methods before. I'm a proponent of IRV, but really in the interest of improving our electoral outcomes from what plurality gives us. If I start to hear credible voices promote one of these other methods, I'll think harder about what I prefer. But, on first blush, IRV makes more sense.