Monday, October 18, 2010

Want to know something about Ranked Choice Voting?
You've come to the right place.

So I was walking across the Laney campus a few weeks back, minding mah own bidness, when a fellow steps up to me and asks if I would like the school the votin' populace of Oakland about Ranked Choice voting, also known as Instant Run-Off.

Hey, it's about combinations of stuff! It produces numbers which can be analyzed! I git PAID to 'splain stuff! What part of this is not fun, fun, fun to Matty Boy?

So this election day, I will be a Ranked Choice Voting facilitator in a precinct near my house. People will come to me confused and in need of knowledge, and they will depart in confidence that their vote will count.

I've taught the stuff to both of my stats classes and did a run through of how it might work in an election like the Oakland mayor's race, where there are ten, count 'em ten candidates.

It should be lots of fun.

Looking around the Internets for more info,I found this graphic showing how voters in Minneapolis felt about RCV.

Notice how Democrats and Independents, who we will lump together in a new category called Right Thinking People, prefer Ranked Choice Voting to a primary system by wide margins, while Republicans, who felt bad they didn't get a new category name so I will call them The Unwashed Masses just to keep them from whining, pretty much hate the idea in a nigh-exact mirror image of the Democratic numbers.

Weird, huh?

I get the feeling anything that takes more than 20 seconds to 'splain and doesn't have a catchy jingle is just not their cup of tea.

So to speak.


Abu Scooter said...

If you must have first-past-the-post election, ranked-choice voting provides a great fix. I like multi-party parlimentary schemes better, but that's not going to happen in the U.S. anytime soon.

Minneapolis implements RCV a little differently. In San Francisco and Oakland, the losing candidates are eliminated one at a time. Minneapolis eliminates all but the top two contenders after the first round, so there are at most two rounds.

That said, I'm pretty sure Republicans hate both forms.

Anonymous said...

Abu Scooter,

You are incorrect about Minneapolis, they do not eliminate all the candidates except the top two.

They do "multiple elimination's" whereas if the low voters getter do not have enough votes combined to that equal to the next highest candidate (or something like that) they eliminate them and redistribute the votes.

Consider with 10 candidates, 4 might have 1% - 3% of the vote, so they are all eliminated (there's your wasted vote) and your second choice come into play.

So if these candidates had 400 votes and the others had several thousand, then it makes sense to eliminate them all, rather than first round, eliminate candidate with 400 votes, and distribute them, then eliminate the candidate with 450 votes, and distribute them, then eliminate the candidate with 475 votes, and redistribute..... The results will be the same, it's just more efficient.
Most Minneapolis races were won by incumbents in the first round. Others went a couple of rounds, but still the top vote getter in the first round won. It's just like plurality.

A report on Minneapolis' RCV can be found here:

It seems RCV did not deliver on its promises in Minneapolis.

Matty Boy said...

Most municipal elections in California are 50%+1, so run-offs are necessary. One of the candidates I'm supporting in November, Michael Nava, was the top-vote getter for municipal superior court judge in the June primary, but did not get the absolute majority, so he has to be in a run-off. If they had RCV in San Francisco muni elections, the voters would be spared the expense of the second election and Nava would have been spared the expense of a second campaign.

Dale Sheldon-Hess said...

It's not that *Republicans* are opposed to RCV/IRV/whatever-name-it's-going-by-this-week; it's that the *smaller of the two dominant parties* in any given region is opposed to RCV/IRV.

For instance, when IRV was proposed in Alaska, it was the Democrats who were opposed.

This is because RCV/IRV, "returns" all the third-party votes back to the major-party that they "rightfully belong" to.

The larger of the two parties knows their chance to win is increased in this way, while the smaller party, already at a disadvantage, sees a glimmer of hope in having a small third party pop up as a spoiler and split the vote with their larger opponent.

Of course, if third parties get enough of a foothold, they go back to being spoilers, but growing to that size (within striking distance of 25% of first-place preferences) usually takes some time; time which the #2 party doesn't have the patience for.

I wrote about the Minneapolis RCV report back in July, and noted that it: didn't help with turnout, didn't help with costs, and quadrupled ballot-spoilage rates.

It's not an effective system. It is, in fact, the worst-possible electoral system we could use other than plurality. I think this image captures that clearly.

Much better solutions would be approval voting and range (AKA score) voting.

Matty Boy said...

I followed your links, Dale. I followed the arguments and I'm not impressed by the results.

The Oakland mayoral race is non-partisan, though most of the front runners are probably Democrats if party affiliation was asked. The big question in this one is whether Don Perata, who has the biggest name recognition, most money and very high negative ratings, will be able to win on the first ballot and if he does not, whether he will maintain his likely early lead given how many people would rather have cancer than vote for this guy.

Your favorite system, score voting, would confuse many voters, as will RCV.

The random samples run on a computer using made up numbers are the least impressive link in your set of links. I programmed for a living, I have a master's in math, and I teach. Trust me when I call bullshit on that bullshit. Using the word "Bayesian" doesn't impress me much.

sfmike said...

Dear Matty: I'm so happy I could be of service over at "Civic Center" with the advice from the election lady: "this is all you really need to know about ranked-choice voting, Put Down Three Fucking Names not One Name Three Times," and there you have it.

Dale Sheldon-Hess said...

Actually, my favorite is approval voting, since it gets almost all the benefits of (the more complex) score voting method, but is astonishingly simple.

It's pretty easy to draw a connection between voter confusion and ballot spoilage rates, so it's important to highlight not only that RCV quadruples ballot-spoilage, but that approval reduces ballot spoilage below the level of plurality ballots. (That's in real-world testing, not just computer simulations.) I would argue, then, that approval is not confusing to voters in the slightest.

As for your concern with simulation in general: there are good reasons for using simulations.

The take-away quote: "Of course... we have to admit... computerized "voters" and "candidates" are not the real thing. They are only as good as the (oversimplified) models we program. The best response to this criticism is to program lots of models with lots of parameters, and have the computer simulate elections in all of them. That way, you can be reasonably confident you've captured a lot of kinds of situations in your data, and at least some of them are realistic. Range voting was the best (i.e. lowest Bayesian Regret) voting method in our experiments in all of 720 different models."

By the way, the guy who wrote most of this stuff that I'm linking and quoting? He has a PhD in math, and teaches at a university. So you can take your appeal to certification and kindly shove it. Engage the work on its academic merits. If you want to attack his simulations as being unrepresentative of real-life elections, you are welcome to do so; you wouldn't be the first, and he's produced several different models to incorporate various people's concerns. They all show pretty much the same results: approval and score significantly out-perform all ranked-order-based methods, especially RCV/IRV. That suggests to me that it's at least worth trying.

My prediction for Oakland: neither Perata nor Quan reach 50%, as too many ballots will be exhausted before the final elimination.

Matty Boy said...

I'm going to agree with you on one thing, Dale. RCV must be difficult to understand, since your final prediction CANNOT BE TRUE.

Someone in the last round WILL reach 50% of the remaining ballots. If you compare the exhausted ballots to the lower voter turnout that would happen in a single issue two-person runoff sometime in December or January, the cost of RCV and the voter participation in the final instant runoff, even if it is the ninth round, RCV is clearly more cost-effective and very likely enfranchises more people than a run-off after the general.

Dale Sheldon-Hess said...

Don't pretend I wouldn't notice that you added a very important word: "remaining"; when I explicitly mentioned exhausted ballots.

Interestingly, hotly contested races (ones where the top-two get into the mid to high 40s) tend to see an increase in voter-participation for a traditional, delayed runoff; but these are precisely the races that RCV/IRV will end with less than 50% of all valid ballots going to the eventual winner. Call that "self-disenfrancisement;" people, after all, chose not to include one of the top-two candidates anywhere on their ballot. But don't pretend that that sort of self-disenfrancisement is any different than a voter choosing to not go out and vote in a traditional runoff. If you count a drop off in voters in a traditional runoff as "lost turnout", then you can't turn around and count the exhausted IRV ballots toward "increased turnout" or, more importantly, disregard them entirely when you want to claim a 50%+1 majoritarian decision.

But my complaint against IRV isn't for failing to get to it's (artificially enhanced) 50% threshold; my complaint is, as measured by Bayesian regret in Dr. Smith's simulations, it does astoundingly poorly, in the worst case performing precisely as poorly as plurality. And the real-world backs up the simulation. They've used IRV for nearly a hundred years in Australia, and they have the same tired two-party dominated dynamic in their IRV-elected seats as the US does in its plurality-elected seats. IRV just doesn't deliver on its promises.

As for cost, as I noted with Minneapolis: the cost of the IRV election (even ignoring one-time costs like new voting machines and also all voter education cost) was 20% higher than holding a single election. This means that, unless more than 20% of elections go to a runoff, IRV is a net increase in costs. Also, in that particular election, only 15% of seats were not awarded in the first IRV round. You'll note that 15% is less than 20%.

And, again: quadrupled ballot spoilage rates. You want to talk about disenfranchisement? (And Minneapolis was a particular good performance; other regions have seen six or seven fold spoilage increases.)

Look, I didn't want this to get hostile. But calling what I'm saying "bullshit" and then purposefully mis-reading my prediction in order to insult my ability to do basic arithmetic IN ALL CAPS... if I apologize for the "PhD" and "at a university" thing, can we return this to a civil discussion?