Metanet Forums

http://en.wikipedia.org/wiki/Single_transferable_vote
I like this idea a lot regarding the election of Congress and the state congresses in the United States. It satisfies roughly every criterion I can think of:
- you vote for candidates, not parties
- it has been shown to produce a multiparty legislature (I submit for your consideration the Australian Senate)
- it's easy to understand

It's basically proportional representation, except it allows independent candidates to run.

What I don't like is Instant Runoff Voting, STV's single-winner equivalent. There's an excellent critique of it here. Basically:
- ranking someone higher can cause them to lose (!!!)
- it generally doesn't "give third parties a chance" (Duverger's law applies, as it's still logical to go with the "sensible" first choice rather than the one you want)
I don't think IRV or any single-winner method is a good choice for electing a lot of people. In the case of the House of Representatives, as an example, STV would require electing all candidates at once, which would have the additional beneficial effect of killing gerrymandering.

Rather, I like range voting, at least for single-winner elections. I like range voting with a small range, preferably three (maybe four) levels of support, not 99 as advocated at http://rangevoting.org/. Using a range of 3 easily dispatches both examples on fairvote that are used to argue against range and approval voting, and in my opinion is simple enough to actually be implemented. I don't like the idea of a range higher than three because it leads to 'sniping' (as seen on the popular website NUMA) and I think 3-level beats approval (though approval is my second choice) because it allows people to give some support to a 'sensible' choice and thus not feel like they're 'wasting' their vote without the issue of strategically denying votes.

Ideas?

Having gone through all of the links you gave, and any of the links on those pages that were interesting, I agree that range voting with 3 levels holds some promise, but it still seems to be vulnerable to voter exaggeration (that is, tactical voters will have more sway than honest voters), which IMO sounds like a big problem.

Also, the anti-IRV arguments were not that convincing, as they relied upon somewhat contrived circumstances with small numbers of voters. It seems to me that:

(1) The probability of such situations arising decreases with the number of voters, the number of candidates, and the amount of voter variability. In an election with 100,000 voters (normal humans) and 5 candidates, are FBLE upsets really that likely?

(2) It's interesting to see after the fact how an election could have been screwed up by dishonest voting, but voters don't have that information going into an election. Even with knowledge of first-place polls, it still seems a bit far-fetched to me that enough people are going to go out on a limb and FBLE in the hope that enough other people (as in, like, 10% of all voters) have (1) FBLE'd at all, (2) have FBLE'd the same way, and (3) the situation was just-so that it actually came through and made a difference. And maybe it is more likely than I think, but the site did not present compelling evidence of that.

Avarin wrote:Having gone through all of the links you gave, and any of the links on those pages that were interesting, I agree that range voting with 3 levels holds some promise, but it still seems to be vulnerable to voter exaggeration (that is, tactical voters will have more sway than honest voters), which IMO sounds like a big problem.

It's hard to "game" such a system, because even being completely strategic only changes any given vote by one point; more importantly, in any case where it might be attempted, it has a potential cost - ranking someone who you would normally place at "like a little" at "like a lot" means that your favorite candidate has no advantage (bad) whereas putting someone who would normally be at "like a little" at "don't like" could mean that the "sensible" choice loses and you end up with John McCain (bad).

Also, the anti-IRV arguments were not that convincing, as they relied upon somewhat contrived circumstances with small numbers of voters. It seems to me that:

Uh, the anti-IRV page was full of real-world examples where IRV failed to solve the problem of two-party domination. Have you seen this graph?

Granted, single-winner elections tend to be a lame way to elect a legislative body. Also:

Maybe you object that this example sounds rare and contrived and unlikely to occur in practice. (If so, you are wrong. And in fact a betrayal-pays scenario like this recently happened in Peru.) But that does not matter much. What matters in the eyes of these 8 voters is: Which is more likely?

Read the Peru example well.

it still seems a bit far-fetched to me that enough people are going to go out on a limb and FBLE in the hope that enough other people (as in, like, 10% of all voters) have (1) FBLE'd at all, (2) have FBLE'd the same way, and (3) the situation was just-so that it actually came through and made a difference.

People are bad at statistics and tend to vote the same way in an election by millions the same as they would in an election by hundreds. In the entire history of the United States, there has only been one election where your vote could have been decisive even if you had a thousand votes -- and if you didn't live in Florida in 2000, there has never been such a situation. That's why the "small" examples scale better than it seems like they should: they're indicative of human psychology.

By your logic, people would be expected to vote for their favorite in a plurality election, on the basis that they can't predict the votes of others. This would be nice if it were true, but it clearly isn't, because as a general rule you can predict the votes of others to some extent. Everyone was pretty sure Obummer would win in 2008 a month before the election was even held.

Incidentally, I think real runoff elections aren't a bad choice either, especially if used as an adjunct to three-level/range/approval voting. Even in Peru they worked well enough that there were more than two candidates with a chance of winning.

http://en.wikipedia.org/wiki/Tactical_v ... ar_systems

scythe wrote:It's hard to "game" such a system, because even being completely strategic only changes any given vote by one point; more importantly, in any case where it might be attempted, it has a potential cost.

That last part is silly, because all such (reasonable) systems have a potential cost for using your vote to try to game the system. Nonetheless, now that I give it more thought, I think you're right about the strength of the 3-level range voting system. It really does seem to mitigate all the important problems I can think of, though I'd still like to see some more models of it.

Also, I think I should be clearer about my stance on IRV - I don't think it's a great system, but it seems to be better than a simple plurality system. So that's the stance I am defending.

scythe wrote:Have you seen this graph?

Yeah, I don't understand what that graph is saying. The most obvious question is how can you have 2.5 or 1.7 parties, and what does "effective # of parties by seats" mean? But, more importantly, the graph is just a snapshot. It shows the nations represented by one election they've had. That doesn't show Duverger's law being true (or not being true) for anything necessarily. I'd be happy to look at more information regarding that graph, but I couldn't find anything helpful (clicking on it didn't help much).

scythe wrote:Read the Peru example well.

I did.
(1) "Caveat: Greg Dennis points out (correctly) that we do not really know IRV would not have elected Flores in Peru, because the 21% of the voters for candidates 4-20 might conceivably have voted differently about the top 3 than the other 79% of the voters – and differently-enough that they would have caused Flores to be the IRV winner." I agree. 20% is a lot.
(2) "Would this affect voters' strategies?
Yes! In this kind of scenario, if the Humala voters knew what was going to happen (and preferred Flores as the lesser evil over Garcia) then they would have been better off dishonestly ranking Flores top and betraying their true favorite Humala. That betrayal (in the runoff or IRV systems) would then cause Flores to win over the hated Garcia."
As I mentioned earlier:
1. Relies upon voters having foreknowledge of a complicated situation (not just who has the most #1's) - and the author notes this as well.
2. Relies upon voters acting in a group to game the system. (Not easily dismissed, since it involves the same "potential cost" you mentioned for range voting)
3. Relies upon those voters acting in group to game the system to significantly view Flores as the lesser evil instead of Garcia as the lesser evil.
(3) "How could it be that the candidate pairwise-preferred over all others could lose – indeed lose in the first IRV voting round?!?" It can happen through the magic of specious logic. Pairwise-preferred doesn't necessarily mean anything in terms of pluralities. If the poll had been done using a plurality method instead of a pairwise comparison, it probably would have shown something similar to the election results.

scythe wrote:People are bad at statistics and tend to vote the same way in an election by millions the same as they would in an election by hundreds.

Actually, I just meant that more voters = more variability. You are correct in pointing out that the amount of possible variability is very limited in an election with only 2 or 3 effective candidates (negating my argument). I admit I keep thinking of elections with 4 or 5 effective candidates.

scythe wrote:By your logic, people would be expected to vote for their favorite in a plurality election, on the basis that they can't predict the votes of others.

Right, but, as you know, they can predict the votes of others because of polls, and I am accounting for that. In my previous post, I said, "Even with knowledge of first-place polls, it still seems a bit far-fetched to me that enough people are going to go out on a limb and FBLE blah blah blah."

Sorry for getting so tied up in details. Hopefully I haven't lost sight of the original topic, and my overall ideas are clearer to you. Also, I feel I should mention that on this article, I am completely lost by the end of the fourth paragraph. :(

The rangevoting.org guy is a college professor who is out of tune with the fact that most people can't grok advanced math very easily. I haven't read everything on there, and it'd take me a long time to figure it out if I did. There are a number of rather clear points he makes that I like to reference, though.

I did.
(1) "Caveat: Greg Dennis points out (correctly) that we do not really know IRV would not have elected Flores in Peru, because the 21% of the voters for candidates 4-20 might conceivably have voted differently about the top 3 than the other 79% of the voters – and differently-enough that they would have caused Flores to be the IRV winner." I agree. 20% is a lot.
(2) "Would this affect voters' strategies?
Yes! In this kind of scenario, if the Humala voters knew what was going to happen (and preferred Flores as the lesser evil over Garcia) then they would have been better off dishonestly ranking Flores top and betraying their true favorite Humala. That betrayal (in the runoff or IRV systems) would then cause Flores to win over the hated Garcia."
As I mentioned earlier:
1. Relies upon voters having foreknowledge of a complicated situation (not just who has the most #1's) - and the author notes this as well.
2. Relies upon voters acting in a group to game the system. (Not easily dismissed, since it involves the same "potential cost" you mentioned for range voting)
3. Relies upon those voters acting in group to game the system to significantly view Flores as the lesser evil instead of Garcia as the lesser evil.
(3) "How could it be that the candidate pairwise-preferred over all others could lose – indeed lose in the first IRV voting round?!?" It can happen through the magic of specious logic. Pairwise-preferred doesn't necessarily mean anything in terms of pluralities. If the poll had been done using a plurality method instead of a pairwise comparison, it probably would have shown something similar to the election results.

Fair enough. I wasn't saying that IRV was worse than plurality; such an assertion would be foolish. I'd gotten a little too used to reading sites like fairvote.org that have a disturbingly blind love for IRV, and maybe a little resentful that some people seem to think it's the only alternative to what we have now.

Anyway, the big problem I have with it and also with normal range voting and many Condorcet methods is this: it provides too many choices, which confuse people. In IRV, there are going to be plenty of situations where people don't really care about whether one guy is ranked higher than another, especially if there are more than four candidates, and so they choose arbitrarily. Ideally, any voting system would severely cut down on "arbitrary" voting, because it's kind of crazy to imagine someone getting elected on such a basis. You see this in NUMA all the time -- I wouldn't be surprised to find that if someone went over their NUMA rating history, it wouldn't take too long for them to say "I rated map X a 4 and map Y a 3? What the hell was I thinking?" -- mostly because while it might sound embarrassing to say we can't decide between six levels, it's often not far from the truth. Range voting makes this even worse. I think that the possibility of the vote being influenced by spurious decisions is pretty much unacceptable in any proposed voting system.

The Schulze method, at the very least, is better than IRV and far simpler from the voter's perspective since it eliminates spurious decisions. Schulze STV is perhaps better than normal STV for a legislature.