The principles that rule this blog

Principles that will govern my thoughts as I express them here (from my opening statement):


  • Freedom of the individual should be as total as possible, limited only by the fact that nobody should be free to cause physical injury to another, or to deprive another person of his freedoms.
  • Government is necessary primarily to provide those services that private enterprise won't, or won't at a price that people can afford.
  • No person has a right to have his own beliefs on religious, moral, political, or other controversial issues imposed on others who do not share those beliefs.

I believe that Abraham Lincoln expressed it very well:

“The legitimate object of government is to do for a community of people whatever they need to have done, but cannot do, at all, or cannot
so well do, for themselves — in their separate, individual capacities.”


Comments will be invited, and I will attempt to reply to any comments that are offered in a serious and non-abusive manner. However, I will not tolerate abusive or profane language (my reasoning is that this is my blog, and so I can control it; I wouldn't interfere with your using such language on your own!)

If anyone finds an opinion that I express to be contrary to my principles, they are welcome to point this out. I hope that I can make a rational case for my comments. Because, in fact, one label I'll happily accept is rationalist.

Wednesday, March 12, 2008

Calculations of Bayesian regret

Broken Ladder referred to the "millions" of simulated elections in Smith's calculations. But any type of calculation of this kind is going to depend on the algorithm that generates the data. And a different algorithm might generate a different set of data.

I do not mean to denigrate Smith's experiment. It performed a worthwhile function. But it is still an estimate of the validity of these different electoral systems, not a mathematical demonstration, and has to be considered in that category. Even so, I think the experiment really does show how much better range voting is than any of the alternative systems. But Smith himself would not (I imagine) claim those numbers would have perfect validity in all cases.

8 comments:

BROKEN LADDER said...

You're terribly misunderstanding the point. It is absolutely possible that in a particular election, any one method could just happen to pick a more representative leader than another, supposedly better, method would. There are times when even plurality would generate a better result than Range Voting.

The point is, what are the likely worst-case and best-case scenarios, as well as the average performance of the election method over a huge number of random elections.

You can't really complain that Smith's simulations may not have been sufficiently realistic. In them, 5 different "knobs" (things like what fraction of the voters chose to vote stategically vs. honestly) were tuned to a whopping 720 different combinations -- and for each combination, hundreds of thousands of simulations were averaged together to get the results. Different utility generators were used, some based even on real election data.

If Range Voting had outperformed the alternatives in some models, but not in others, then we would have a difficult task of trying to determine which of those models was the most like real life, so as to make a realistic assessment of the results, knowing which models to pay the most attention to.

But as luck would have it, Range Voting beat the other common methods (e.g. Condorcet, Borda, IRV, plurality, approval) in all 720 models.

So your criticism is just one more in a line of naive inexperienced novices to come along and repeat exactly the same misguided criticism over again.

I can't tel you how tired I get of having to explain this to people because they won't take the time to read the site more thoroughly and learn all this. Somehow people just think they can read the gist of it and have a deep expert understanding of this subject. It's baffling.

Greg said...

Broken Ladder says "Range voting beat the other common methods in all 720 models." That shows you how incomplete the set of models were. Consider the "bullet vote" model where voters give their first choice the highest score and all other choices the lowest score. Under that strategy, Range devolves to plurality and therefore could not have "beat the other common methods".

I don't know whether voters will use that strategy in practice, but neither do BrokenLadder or Smith. Nobody knows which strategic model voters will follow in real political elections. Smith's experiments explore a tiny subspace of potential strategic models. Under some, Range will beat the other methods; under others (like the bullet vote model) it won't.

This all presumes, by the way, that Smith's idea of maximizing global utility (minimizing Bayesian regret) is the metric by which we want to compare voting systems. It just happens to contravene centuries of thinking that majoritarian voting systems are preferable.

Range voting advocates could also better address the issue of fairness. They admit that RV will respond readily to strategic voting -- they just claim its susceptibility to strategy is irrelevant because it will still produce a lower Bayesian regret than other common methods, even when strategy is involved. While it may still produce a lower BR, the result will nevertheless be biased towards the preferences of the dishonest voters. Do we want that in a voting system?

If many will inevitably use an Approval Voting strategy, why not just recommend Approval so that everyone is on equal footing? I am do not recommend Approval Voting, but it does seem more logically consistent than general Range.

Overall, Range Voting seems to lack serious vetting and understanding of how it will work in practice. Its advocates seem to rely heavily on a couple of un-peer-reviewed papers. I would to see the same critical eye that has been directed at IRV, directed at Range as well. When that has happened in the recent serious scholarship, such as the case of Nagel's paper and Tideman's book, Range has been deemed worse than IRV.

Opinionator said...

Broken Ladder wrote:

"Range Voting beat the other common methods (e.g. Condorcet, Borda, IRV, plurality, approval) in all 720 models."

I think that this makes sense, and I have said I favor range voting over all the other methods; I thought you had been defending approval voting as "almost as good." But the question is, can we assume that the distribution of different types of election situations was correctly forecast in Smith's simulations? I'm not so sure. In some cases (where there are clearly some good candidates and some bad ones), approval voting will do well. In others, where there are a lot of "in-between" ones, it will be much worse, because one could misjudge the candidates with a high probability of winning and end up hurting someone you like or helping someone you don't. There similarly are problems with Borda, where if my rankings are very nonlinear I can hurt a good choice or help a bad one by my ranking. Only range voting allows me to honestly rate the candidates in all types of situations. And no simulation system is guaranteed to generate the situations in the same proportion that they will occur in real life.

Greg wrote:

"Range Voting seems to lack serious vetting and understanding of how it will work in practice."

This is probably true. It needs more trial in real-life situations. But the only way to give it that "serious vetting and understanding of how it will work in practice" would be to implement it somewhere!

BROKEN LADDER said...

Greg,

Most of your statements are uninformed nonsense coming from inexperience on this subject.

Consider the "bullet vote" model where voters give their first choice the highest score and all other choices the lowest score. Under that strategy, Range devolves to plurality and therefore could not have "beat the other common methods".

Bullet voting isn't "strategic"; in fact it's worse than sincere Range Voting on average.
http://rangevoting.org/RVstrat3.html

Then there's the question, "Will people naively do it anyway?" Well, obviously not -- at least in the situations where it counts the most. Consider the Nader voters in 2000; estimates showed about 90% of those voters who preferred Nader actually voted for Gore, because they didn't want to waste their vote power. So, forced to bullet vote (forced to use plurality voting) those voters cast a vote for Gore. If they could suddenly use Range Voting they'd still have the same incentive to vote for Gore, but then they'd also have every reason to vote for Nader too. Sure there are some dumb people out there who would not, but they obviously would be rare. So bullet voting does not seem to be a problem for Range Voting, based on reasonable notions of human psychology.

But beyond that, there have been Range Voting "poll" experiments which suggest that many/most voters will indeed use intermediate values, and even the ones who only use "0's" and "10's" will tend to give those 10's to more than one candidate -- so empirical evidence also says bullet voting is rare.

I did one small poll myself -- in Beaumont, TX -- http://rangevoting.org/Beaumont.html

I don't know whether voters will use that strategy in practice, but neither do BrokenLadder or Smith.

Ooops...you were wrong, because you didn't take just a few minutes to look at the research that backs up our claim, and shows overwhelmingly that bullet voting will be rare and non-problematic. It might help to also think about an analogy that William Poundstone made in his pro-Range-Voting book Gaming the Vote. The ability to rate more than one candidate is like having airbags in your car. If you go to the polls and just cast a "10" for Gore or Bush, it doesn't really matter that you didn't rate Nader. The fact is, if you felt confident in just voting for one candidate, you probably knew that candidate was one of the front-runners anyway. If that candidate was a long-shot, you'd also want to give high marks to other more viable candidates you liked. For instance, if I was voting in the Democratic primary back when there were more candidates, and I gave a 10 to Obama and rated no others, that wouldn't have really been a problem. But say my favorite was Kucinich. Well, after giving him a 10, I'd think "Okay? Do I really not want to also help Obama beat the vile and disgusting Hillary Clinton? Would it not help me to also distinguish among my favorite front-runners so I don't waste my vote?"

So, a little thought goes a long way toward clearing up your confusion.

Nobody knows which strategic model voters will follow in real political elections.

That's both false and misleading>. It's false in the sense that we've studied voter behavior enough to have identified common strategies that voters use, that we can then model. It's misleading in that the simulations also incorporate various amounts of randomness that simulate ignorance and/or the use of completely incomprehensible strategies. So these things are, at least in very reasonably approximate ways, accounted for.

Ultimately, the fact that you can so vastly tweak so many parameters, and Range Voting always comes up tops (out of the common "non-exotic" voting methods), is powerful evidence that it would continue to dominate despite any tweaks you could make to increase the realism of the simulations.

Everything about your criticisms is based on ignorance, and has been said -- and refuted -- before.

Smith's experiments explore a tiny subspace of potential strategic models.

No, not tiny. You are just asserting things without having any clue what you are talking about. I've been intensively studying this issue for almost two years, and I know a lot about how these simulations work -- whereas you have demonstrated that you know virtually nothing about how they work, and that you will juts make out-of-hand assertions with zero evidence.

The fact is, there is a pretty limited set of strategies that voters will use, given ratings or rankings, which cover 95% or more of the scenarios. For the rest, the randomness inherent to the models, including the ignorance factors, account for the vast swath of inconceivable and incomprehensible "strategies" that voters may employ.

But of course, you didn't know that, because you decided to start asserting expertise on this subject before having the slightest clue how these simulations work.

Under some, Range will beat the other methods; under others (like the bullet vote model) it won't.

But massive evidence says the bullet voting model is unrealistic, so...you'd better come up with something else pal. You don't know what you're talking about and you're flailing.

This all presumes, by the way, that Smith's idea of maximizing global utility (minimizing Bayesian regret) is the metric by which we want to compare voting systems. It just happens to contravene centuries of thinking that majoritarian voting systems are preferable.

That thinking is wrong, and that has been mathematically proven via reductio ad absurdum. In effect, a rule that says "X is better than Y if preferred by a majority to Y" can easily contradict itself, showing that it cannot be a valid axiom.
http://groups.google.com/group/socorg/web/majority-myth

If you propose any alternative social utility function than summation of individual utilities, then present it. I myself, as well as others, have rigorously tried to do that, and it is apparently impossible. Maximizing social utility is forced.
http://rangevoting.org/UtilFoundns.html

It goes beyond that. Maximizing social utility is desirable to all voters as "selfish" individuals, because it means the greatest expected satisfaction with election results. Think of it this way. Say that in an election that you didn't care that much about, you were in the majority, and the minority generally cared a lot more about the issue than the majority. Then in another election, the tables were turned. Now before you go on raving about the importance of "majority rule" (which, as I noted has been mathematically disproved as the correct goal) think about how much happier society would be overall, if your majority could have conceded the election that they didn't care so much about in order to over-rule the majority in the election you did care a lot about. You would WANT that, because you and most other people would be more happy on the whole.

Range voting advocates could also better address the issue of fairness. They admit that RV will respond readily to strategic voting -- they just claim its susceptibility to strategy is irrelevant because it will still produce a lower Bayesian regret than other common methods, even when strategy is involved.

And that all other voting methods also are (more) susceptible to strategic voting! A non-trivial point that you shouldn't carelessly leave out.

While it may still produce a lower BR, the result will nevertheless be biased towards the preferences of the dishonest voters. Do we want that in a voting system?

Strategic voting is possible in ALL deterministic voting methods. Stop, and read this, and actually understand before continuing to respond. For instance, in a ranked voting election where I prefer Green>Dem>GOP, I will generally be better off to say Dem>Green>GOP in almost all ranked voting methods.

The ONLY way you can make a voting method free from strategy is to add randomness to it. For instance, say we order all the candidates, and then two candidates are randomly selected and pitted head-to-head based on our rankings. That theoretical method is 100% free from strategy. But it totally sucks, because it can easily pick the majority winner out of the two least liked candidates.

Also it is profoundly errant to say that strategic voting equates to unfairness. All election outcomes make some people happy and others unhappy -- and that's not "fair". It doesn't matter whether the people who are less happy happen to be honest voters, or happen to be people with the last name Smith. It's not as if one disparity is more "fair" than another because of the criteria on which the disparity is based.

No voting method can ever be "fair" in the simplistic sense we're used to thinking of. Bush wins and lots of conservatives cheer. I cry. The ideal is to pick the candidate who makes the most people the most happy.

If many will inevitably use an Approval Voting strategy, why not just recommend Approval so that everyone is on equal footing?

Because that will lower the average voter happiness. And it will not make everyone on equal footing. Some people will be happy with election outcomes, and others will be unhappy. There's no equality in voting.

Moreover, you forget that if you change from Range Voting to Approval Voting to "help" a naively honest Range Voting user cast a more strategic vote, then

1) You do the same for all the naively sincere Range Voting users whose feelings differ from his, which means you strengthen his enemies as much as you strengthen him and his allies!

2) Those naive voters may easily cast sub-optimal Approval Voting ballots, which are worse for them than sincere Range Voting ballots:
http://rangevoting.org/RVstrat6.html

Thus Range Voting not only makes the whole of society better off, on average, but it also may be better than Approval Voting even for sincere voters.
http://rangevoting.org/ShExpRes.html <= experiment I suggested to prove this

I am do not recommend Approval Voting, but it does seem more logically consistent than general Range.

You're wrong. And the reason you're wrong is that you form opinions without first getting familiar with the subject, so that you can make informed conclusions.

Overall, Range Voting seems to lack serious vetting and understanding of how it will work in practice.

Again, you don't have a clue what you're talking about. We've studied and considered extensively how it will behave in practice:
http://rangevoting.org/Honesty.html
http://rangevoting.org/HonStrat.html
http://rangevoting.org/OrsayTable.html
http://rangevoting.org/HaikuIcon.html
http://rangevoting.org/Beaumont.html
http://rangevoting.org/RVstrat1.html
http://rangevoting.org/RVstrat2.html
http://rangevoting.org/RVstrat3.html
http://rangevoting.org/RVstrat4.html
http://rangevoting.org/RVstrat5.html
http://rangevoting.org/RVstrat6.html

Okay? So we've studied this, and we know what we're talking about, and you are making assertions out of total ignorance.

Its advocates seem to rely heavily on a couple of un-peer-reviewed papers. I would to see the same critical eye that has been directed at IRV, directed at Range as well.

The material has been reviewed extensively by those on the forefront of the election methods community, many of whom have advanced math and science degrees. And in fact you may review it yourself and make informed and intelligent criticisms.

When that has happened in the recent serious scholarship, such as the case of Nagel's paper and Tideman's book, Range has been deemed worse than IRV.

Tideman's assessment was severely flawed, as Warren Smith has pointed out to him.
http://rangevoting.org/TidemanRev.html

A simple sanity check is that Tideman's "strategy resistance" measures say that plurality voting is better than approval voting, despite the fact that this contradicts virtually every election methods expert out there -- approval voting was designed to be strategy-resistant.

A lesson you are going to have to quickly learn is that you do not know nearly as much as you think you know, and the sources of much of your thinking are flawed and out-dated.

Opinionator said...

I think that Broken Ladder does not understand one thing: the concept of "sampling error." He writes:

"You can't really complain that Smith's simulations may not have been sufficiently realistic. In them, 5 different "knobs" (things like what fraction of the voters chose to vote stategically vs. honestly) were tuned to a whopping 720 different combinations -- and for each combination, hundreds of thousands of simulations were averaged together to get the results. Different utility generators were used, some based even on real election data."

The point is that no number of samples is sure to guarantee that the sample will be representative of what would occur in real life. Even if there were "720 different combinations," did each of thoce combinations occur as often as it would occur in a real situation? Suppose you go to Connecticut Avenue and P St. in Washington, D. C. and poll the passers-by on a subject like gay marriage. Do you think that the poll will be representative? Obviously not, if you know the area. It's a location where there are a lot of gay bars, a gay bookstore, and other businesses oriented toward a gay clientele. The portion of the passers-by who would be gay is probably much more than a correct sample of USA population would require. Ever hear of the Literary Digest poll prior to the 1936 election?

Now be sure that you understand what I am claiming. I do not say that Smith's results are wrong, or unreliable. I am only saying that we cannot tell how reliable they are. We do not have any real knowledge as to what sort of distribution these parameters would take in a real-life situation. Smith could only guess. His guesses may be right or wrong. We just don't know!

But I think the general conclusion is justified that range voting better represents the voters' desire than any other system. I say this, not because I think Smith's demonstrations proved it (though I think they did give good evidence for tis proposition) but because I think that it allows a voter, more accurately than any other system, to express what he wants to see in a winning candidate. And for this reason I'm for the idea.

Greg said...

Clay brings up the case of a Nader voter in the 2000 election. This is a good example of the problem with Range Voting. Consider the dilemma faced by a Nader voter: "Should I rate Gore a 10 on the ballot?" If a Nader support votes honestly, they would rate Gore less than a 10, but if they desperately want to avoid a Bush presidency, they may rate him a 10 despite their true preferences. In general, voters will be under a lot of pressure to give a 10 to the front-runner they like the best.

This is why Range Voting doesn't actually eliminate the spoiler problem. The mere presence of Nader in the race would, in practice, cause more people to rate Gore lower.

Thus, the fallacy of the Range advocates logic is exposed. They assume that the rating one gives to a candidate on an RV ballot is reflective of some fixed measurements of utility a voter derives from the candidates, regardless of who the other candidates may be. But of course, the rating one gives to a candidate will in practice always be relative to the other candidates running.

In practice, Range Voting will not satisfy IIA, because the existence of a candidate in the race will naturally cause the other candidates to be given a lower rating. In short, Nader would still be called a spoiler under Range Voting.

In contrast to what Clay wrote, I did read the "research" claiming that voters would vote honestly under Range Voting. I just think an exit poll of no consequence to them gives any indication of how they will respond when faced with the strategic tension under a real RV election.

I'm all for Smith, Clay, and others experimenting with Range Voting at a small scale first, in some student organizations or private associations, so that they have real, empirical evidence as to how well it works, rather than merely computer simulations. I would be opposed, however, to foisting such an untested method on the public for any election of significance. I don't expect them to have much success convincing anyone to implement RV, given its serious problems,

BROKEN LADDER said...

I think that Broken Ladder does not understand one thing: the concept of "sampling error." He writes:

"You can't really complain that Smith's simulations may not have been sufficiently realistic. In them, 5 different "knobs" (things like what fraction of the voters chose to vote stategically vs. honestly) were tuned to a whopping 720 different combinations -- and for each combination, hundreds of thousands of simulations were averaged together to get the results. Different utility generators were used, some based even on real election data."

no number of samples is sure to guarantee that the sample will be representative of what would occur in real life. Even if there were "720 different combinations," did each of thoce combinations occur as often as it would occur in a real situation?

This is a pretty unfocused criticism. If you mean to generally say that the simulations are not "perfect", not exactly like real life, then of course you're correct. But consider that one of the knob settings is "percentage of voters who are strategic (as opposed to sincere)". If you gradually vary that percentage from 0 to 100, you eventually have to come to a value that is approximately reflective of real life averages. That is inevitable. If you vary the ignorance levels from "voters know everything there is to know about the candidates" to extreme ignorance bordering on randomness (choices are as reflective of what a voter wants as they would be if he drew a name out of a hat, because that voter is so misinformed), you inevitably come across a value that is approximately reflective of the real ignorance levels in real voters. Approximate does not mean perfect, but the point is that the results were so strongly in favor of range voting that they amount to a very very powerful case.

As I pointed out before, if the voting methods had performed quite differently depending on the model employed, then we would have difficulty determining which models mimicked real life. In fact, that is actually just the case with various other methods. Borda does better than Condorcet if many voters are honest. But if voters are more stategic, Condorcet does better because it is not quite as harmed by strategic voting. In terms of finding out which of those methods is generally better, we'd have to do thorough real-world studies to find out which conditions are most like real life. But the strength of range voting is shown by the fact that it surpasses these alternatives no matter what model is assumed.

Suppose you go to Connecticut Avenue and P St. in Washington, D. C. and poll the passers-by on a subject like gay marriage. Do you think that the poll will be representative? Obviously not, if you know the area. It's a location where there are a lot of gay bars, a gay bookstore, and other businesses oriented toward a gay clientele. The portion of the passers-by who would be gay is probably much more than a correct sample of USA population would require. Ever hear of the Literary Digest poll prior to the 1936 election?

The simulations used the entire simulated electorate not just certain factions within it -- so this "criticism" is totally bunk. Try again.

Now be sure that you understand what I am claiming. I do not say that Smith's results are wrong, or unreliable. I am only saying that we cannot tell how reliable they are. We do not have any real knowledge as to what sort of distribution these parameters would take in a real-life situation. Smith could only guess. His guesses may be right or wrong. We just don't know!

Yes, Smith could only guess. But the point is, no matter what guesses were used -- whether we guessed that all voters would be honest, or all voters strategic, or some intermediate mixture -- range voting came out on top.

But I think the general conclusion is justified that range voting better represents the voters' desire than any other system.

Actually no, there are better systems, but the are infeasible in their complexity for general public elections. In one system, we successively eliminate the lowest-rated candidate, and then re-normalize every ballot. It is more resistant to strategic voting, but mathematically monstrous.

Range voting is nice both in its quality and simplicity.

I say this, not because I think Smith's demonstrations proved it (though I think they did give good evidence for tis proposition) but because I think that it allows a voter, more accurately than any other system, to express what he wants to see in a winning candidate. And for this reason I'm for the idea.

But it's important not to focus on how expressive the ballot allows you to be; instead you should focus on how well the election outcome expresses what you wanted. Case in point, IRV and approval voting. Some IRV supporters naively feel that IRV is more representative, because it lets them say e.g. Nader is better than Gore is better than Bush; whereas with approval voting they might have approved both Nader and Gore, without being able to say that they preferred Nader to Gore. They hate the way that makes them feel, but that is stupid. It is stupid because with IRV, much of the information you put down is ignored -- completely discarded -- because IRV only looks at the top level of preference, and will never see deeper preferences that could have picked a better winner. Hence IRV incentivizes a strategy where the aforementioned voter would want to insincerely vote Gore>Nader>Bush, and so Nader would ultimately get fewer votes, even though those naive IRV advocates would have thought it was better because it better expressed their preferences.

However I do agree that it is good if a voting method can both provide representative results and give the voter a feeling that he was able to express himself -- and range voting does that.

BROKEN LADDER said...

Clay brings up the case of a Nader voter in the 2000 election. This is a good example of the problem with Range Voting. Consider the dilemma faced by a Nader voter: "Should I rate Gore a 10 on the ballot?" If a Nader support votes honestly, they would rate Gore less than a 10, but if they desperately want to avoid a Bush presidency, they may rate him a 10 despite their true preferences. In general, voters will be under a lot of pressure to give a 10 to the front-runner they like the best.

That's not a problem at all. Studies say about 90% of Nader supporters in 2000 voted for someone else, which was probably Gore for the most part. All of those voters were concerned with getting the best possible outcome with the election, and so with range voting they would still want to give their favorite front-runner a 10, as well as give their sincere favorite a 10. Range voting wouldn't substantially change their support for Gore, but it would allow them to also show their support for Nader, and therefore Nader would have done massively better under range voting.

For those who cared more about expressing their true preferences, and actually voting for Nader, they'd want to do something more like

Nader = 10
Gore = 5
Bush = 0

So compared with plurality voting, they'd go from giving no support to Gore to giving enough support to tilt the balance toward the candidate that pretty clearly was more well supported by the electorate at large.

So what you call a problem simply is not a problem. It's a benefit. And here's something really important to notice about those strategic voters with range voting. Even if they want to help their favorite front-runner (e.g. help Gore beat Bush), they never have an incentive to betray their sincere favorite. That is, no matter what level of support they give to Nader's opponents, they'll never have a reason not to give Nader a 10. That means range voting passes the "favorite betrayal criterion". Very few voting methods pass this. The favorite betrayal incentive is why Instant Runoff Voting is horrible for third parties -- because it gives voters who prefer Third>Major party 1>Majory party 2, an incentive to vote Major party 1>Third>Major party 2.

That is a kind of strategic voting problem that is far worse than anything range voting faces.
http://rangevoting.org/TarrIrv.html

This is why Range Voting doesn't actually eliminate the spoiler problem. The mere presence of Nader in the race would, in practice, cause more people to rate Gore lower.

Range voting eliminates the spoiler problem in the sense that, if you take all the ballots that have been cast, and order the candidates by their scores, the order of any two candidates stays the same no matter which of their opponents you remove or add. No rank-order voting method can say that. For instance, with Instant Runoff Voting, candidate X can be the winner, yet if you remove an "irrelevant" candidate Y, then candidate Z can suddenly become the winner -- even though all the ballots have already been cast and no voters' preferences for any of the candidates are actually changing.

Because voters will re-scale their ballots in real life, range voting cannot perfectly fix the spoiler problem -- and nor can any other voting method. Range voting does much better than the alternatives in this regard however. Another benefit of range voting.

Thus, the fallacy of the Range advocates logic is exposed. They assume that the rating one gives to a candidate on an RV ballot is reflective of some fixed measurements of utility a voter derives from the candidates, regardless of who the other candidates may be.

No we do not, as I just explained. You are misinformed about what range voting advocates actually assume, Q.E.D.

But of course, the rating one gives to a candidate will in practice always be relative to the other candidates running.

If you had spent even a few minutes looking at the range voting web site -- http://rangevoting.org/ -- you would know that range voting proponents are well aware of this.

In short, Nader would still be called a spoiler under Range Voting.

Nope. The vast majority of experts agree that Nader would not have been a spoiler under most any alternative voting method, from IRV to Borda to range voting. There are plenty of realistic scenarios where the introduction of an irrelevant alternative would change the election outcome, but they are fewer and less damaging with range voting than with alternatives like IRV.

Again, no voting method can be "spoiler-proof" in practice, so yours is not a criticism of range voting. In fact it simply points to a benefit of range voting -- that it is more spoiler resistant than alternatives.

In contrast to what Clay wrote, I did read the "research" claiming that voters would vote honestly under Range Voting. I just think an exit poll of no consequence to them gives any indication of how they will respond when faced with the strategic tension under a real RV election.

Wrong. Immense historical analysis of such polling shows that people will still be strategic and lie when talking to pollsters. Perhaps not as much as as with elections of consequence, but definitely to a substantial degree. One example of this is the Bradley effect.

Considering that range voting also makes sincerity far more safe than plurality (it is "safe" for voters to rate no-hope candidates sincerely), and that substantial numbers of plurality voters are prepared to vote sincerely rather than strategically (97,488 Nader voters in 2000 from Florida alone), there is ample evidence that range voting will inspire a respectable degree of sincere voting. And sincere it already operates so well under strategic voting, that means it can only be statistically better for every voter who does choose to be sincere.

I'm all for Smith, Clay, and others experimenting with Range Voting at a small scale first, in some student organizations or private associations, so that they have real, empirical evidence as to how well it works, rather than merely computer simulations.

The computer simulations are better than small scale real world use of range voting, since there's no way to read voters' minds and find out "how well it works" in real life. People who make this argument continue to misunderstand that in computer simulations, range voting excels even if we make the voters more strategically oriented than real humans could ever be expected to be. So range voting is ready for prime time, right now.

I mean, consider that right now we are using plurality voting, and have been using it all across the world for hundreds of years -- and there is absolutely no conceivable way that range voting could be worse than plurality. So it's irrational to urge caution about range voting. Heck, range voting even experimentally results in fewer spoiled ballots than our current method, so in some sense it's easier for voters.

Also, range voting has had hundreds of years of use in places like ancient Venice and Sparta.

So please let's take these tired and misled arguments out into a field and shoot and bury them. Experts on this subject know that your concerns are unjustified.

I would be opposed, however, to foisting such an untested method on the public for any election of significance.

Why? You haven't given any good reasons not to use it. You've just demonstrated that you're very ignorant on this subject.

I don't expect them to have much success convincing anyone to implement RV, given its serious problems

What serious problems? Most everything you think of as a problem for range voting is actually a benefit, or is based on a grievous misunderstanding on your part. It is clear that you have only a rudimentary understanding of this very complex subject. If you're going to say range voting has serious problems, but ignore that plurality voting, instant runoff voting, and every other method under the sun has vastly larger problems, then it's clear you are an ideologue and have no interest in an enlightened fact-based debate.