5.26.2008

Range Voting

Anyone have a reason this shouldn't be how we vote?

9 comments:

Bob Richard said...

(1) Range voting (RV) does not necessarily elect a candidate who is the first choice of a majority of voters, when such a candidate exists. Even if this isn't important to you personally, it makes it very difficult to get the method adopted for public elections.

(2) RV encourages strategic voting. In particular, it fails the most important criterion related to strategic voting, called "later no harm". This (failing the criterion) means that giving any support at all to your second or third choice can hurt your first choice.

(3) Arguments in favor of RV are based on a controversial philosophical premise: the purpose of public decisions is to maximize the sum total of welfare, which is measured by adding up the separate welfares of all individuals. Thus, if total welfare is maximized when 10% of the people enjoy 90% of it, that's the best solution for society. This philosophical view is called utilitarianism. If you reject it, you might still think RV is a useful voting method in some circumstances, but you would have to invent new arguments for it.

People instinctively do something like RV in small, face-to-face groups when a majority goes along with something that a minority (perhaps even one person) feels very strongly about because they recognize that the intensity of the minority's feelings on the subject is far greater than their own. This works when people know and trust each other. There is very little reason to think it can be used successfully in public elections.

BROKEN LADDER said...

1) Range Voting allows a majority to win if they choose to all give a top score to their favorite candidate, and minimum score to the others. If they don't, that is an expression of their desire to compromise. Thus Range Voting achieves substantially better Bayesian regret than a system in which a Condorcet ("beats-all") winner always wins.

This is especially ironic since Instant Runoff Voting, which Bob supports, can elect X instead of Y, even if Y is preferred to X by a head-to-head majority, and has more first-place votes than X.

2) Range Voting is much more resistant to the harmful effects of strategy than IRV (which Bob ironically supports, even though it is highly susceptible to strategic behavior).

That Range Voting "fails" the later-no-harm criterion is a benefit, since it causes a certain amount of revealed preference. Say you'd give 3 candidates sincere scores like X=10, Y=9, Z=0 - but I'd give them X=10, Y=1, Z=0. If we both vote strategically, you may give both X and Y a 10, though I would be more likely to give Y a 0. That's because if you cause Y to win instead of X, it only hurt you a little, whereas if Z beats Y because you did not increase Y's score, it hurts you a lot. For me, it's the opposite. With rank-order voting methods, our votes might look the same, even though our opinions differ greatly.

3) Bob repeats that he rejects the premise that social utility is just the sum of individual voter utilities - but he fails to offer any alternative social utility function. He supports IRV, even though he cannot show that there is some utility function that IRV maximizes.

The reality is that this "controversial" philosophical premise is inescapable, because there's no coherent alternative one. To prove that, I challenge him as I have in the past, to even try to come up with one.

Bob Richard is a testament to the emptiness and deceptiveness of the typical anti-RV and/or pro-IRV rhetoric.

Speaker for the Dead said...

That's interesting. I was asking it as an honest question; I just read about it, and what I read seemed very positive, so I was wondering if there was anything negative to say about it. And it seems that you are an advocate of instant runoff voting.

I definitely agree IRV is better than approval voting.

1. That's a good point, but when that happens, more people probably support the elected official (maybe not as a first choice, but to a certain extent).

2. I'm not sure that's the most important criterion. IRV doesn't let you weight your relative support for candidates, however. (Do you have to rank every candidate?) So it doesn't distinguish between the ballot of a person who strongly prefers his first choice to his second choice and the ballot of someone who barely has a preference.

3. You don't have to accept the entire moral and ethical philosophy of utilitarianism to support range voting. Because everyone's ballot could be weighted equally (in the sense that each individual could give their first choice candidate the same maximum number of points), I imagine it would be difficult to maximize welfare merely by satisfying an arbitrary tenth of the population.

BROKEN LADDER said...

I definitely agree IRV is better than approval voting.

Nope. Approval Voting is far superior to IRV.
http://rangevoting.org/UniqBest.html

It is also precinct-countable, unlike IRV, and reduces the number of spoiled ballots, whereas IRV increases them by a factor of 7.
http://rangevoting.org/SPRates.html

As for utilitarianism, Bob Richard is free to try to propose an alternative social utility function. But I know he can't/won't. The additive utility function is correct, and since Bob can't refute that, he just says the utilitarianism is "controversial". Doubting something without any scientific support for your doubt does not make it controversial, any more than evolution is controversial.

bishop said...

Range voting forces me to give a non-zero value to candidates I might equally dislike. I have to rank the Nazi above the Communist or vice versa.

Also: I would probably rank the Democratic candidate last if I'm a Republican, just to make sure that they get less mathematical support. In other words, even if I'd prefer Barack Obama to David Duke, it would be in my best interests not to vote in that order.

Speaker for the Dead said...

Well, since range voting is cardinal, not ordinal, you can (I think) give two candidates the same rating. And if you can give them both the lowest rating, you are, in effect, giving them a zero.

(I would like range voting with scores between -10 and 10.)

The whole point is that you're not ranking candidates from first to last.

And Bishop, I really like the Yale Free Press.

BROKEN LADDER said...

Bishop said:
Range voting forces me to give a non-zero value to candidates I might equally dislike. I have to rank the Nazi above the Communist or vice versa.

No, you're free to give them both a zero if you want. But there's a reason you shouldn't. Say the Nazi and the communist end up being the clear front-runners, and you found the communist to be just a little less terrible - giving a higher score to the communist helps him beat the Nazi and give you a better result. In fact, if you were pretty sure they'd end up being the front-runner, you may even want to give the communist a maximum score and the Nazi a minimum score as a wise strategy.

Also: I would probably rank the Democratic candidate last if I'm a Republican, just to make sure that they get less mathematical support.

Yes, that's precisely the kind of strategy I just described. Range Voting has remarkably good properties with regard to strategic voting. Here are some additional links on that subject:
http://rangevoting.org/RVstrat1.htmlhttp://rangevoting.org/RVstrat2.htmlhttp://rangevoting.org/RVstrat3.htmlhttp://rangevoting.org/RVstrat4.htmlhttp://rangevoting.org/RVstrat5.htmlhttp://rangevoting.org/RVstrat6.html
http://rangevoting.org/Honesty.html
http://rangevoting.org/HonStrat.html
http://rangevoting.org/PleasantSurprise.html

In other words, even if I'd prefer Barack Obama to David Duke, it would be in my best interests not to vote in that order.

It is very rarely in your best interest to vote out-of-order with Range Voting. If you preferred Obama to Duke, but gave Obama a 0 then you generally might as well give Duke a 0 as well.

BROKEN LADDER said...

I would like range voting with scores between -10 and 10.

Starting at 0 is better.

The whole point is that you're not ranking candidates from first to last.

But if you prefer X>Y, yet give Y a higher score than X, that is a clear case of mis-stated preferences, and it's something we'd like to minimize. Fortunately, this problem is rare, and is small in comparison to related mis-ordering problems present in ordinal voting methods. From RangeVoting.org/Honest.html:

Honesty Theorem. In a 3-candidate range election, nobody can ever gain strategic advantage by dishonestly range-voting as though Alice>Bob, if their honest opinion is Bob>Alice.
This is not a state of perfection. But it is still good and useful, and it is better than

Contrasting Theorem (follows from Gibbard). In a 3-candidate election conducted using any method whatever whose votes are (unlike in range voting) rank-orderings of the candidates, there is always some election scenario in which a voter can obtain a better election result (from his point of view) by dishonestly voting as though Alice>Bob, even if his honest opinion is Bob>Alice.

So range voting encourages honesty better than every rank-ordering-based voting system, including Borda, IRV, and Condorcet methods.

homunq said...

Job 34:4 seems to support range voting, certainly more than it does IRV.