At 08:32 PM 9/3/2007, Warren Smith wrote: >3. For voting methods whose ballots are NOT just rank-orderings, >(e.g. approval and range voting) there is no such thing as "truncation." >But it still is possible to cast a plurality-style range or approval vote.
Actually, I use the word "truncation" with Range Voting to refer to Approval-style votes, i.e., max or min votes for more than one candidate when the voter has, in fact, some preference between them. Hybrid systems like Range Plus Pairwise Analysis would encourage the expression of such preference, but they would remain distorted in the Range Vote itself; however, the Range ballot becomes analyzable as a ranked ballot and has, presumably, true preferences, not reversed preferences. Why this would work is more complex a question than I care to answer here. But the "distortion" I referred to is effectively truncation, just as truncation in ranked ballots refers to equal-rating bottom. It's just that in Range and Approval, you can equal rate top just as easily. Pure truncation at the top would be, as an example, absolute utilities, normalized to the full election set would give, for A, B, C, 100, 90, 0, but, instead, the voter normalizes to the frontrunners, B and C. So the votes become 100, 100, 0. A voter might vote -- many of us say we would vote that way, 100, 99, 0, tolerating the very small loss of voting power in favor of expressing rank correctly. With pairwise analysis, this would increase the likelihood that a runoff would be between the favorite and the preferred frontrunner, a happy result for the voter, who can then express exclusive preference for the favorite without harm. Now, some Range supporters object that doing pairwise analysis could not fail to increase Regret, since the Range winner is -- supposedly -- the regret minimizer. However, there are two reasons to do this. First is an S.U. reason. When there is a pairwise winner who beats the Range winner, we have, necessarily, the weak preference of a plurality faction vs the strong preference of a weaker faction. As we know, the Range winner is not necessarily the true regret minimizer, because of normalization and strategic voting. I think that a symptom of the range winner not being the regret minimizer would be this discrepancy between the Range and ranked results. Not a proof, a symptom. And so testing the case becomes advisable. How do we test it? I've before only suggested on way, but there is actually another that is even more to the point, and which brings with it another benefit. But, first, the first way. Hold an actual runoff between two candidates, the Range winner, and, probably, the pairwise winner who beats the Range winner, and if there is more than one of these -- very, very unlikely in real elections -- then the one with the highest Range total. Why wouldn't this just reproduce the original ranked result? Well, it's a nuisance to go and vote. If, in fact, the preference is weak, the supporters of the pairwise winner have less motivation to vote; they get a decent outcome with no more effort. On the other hand, the supporters of the Range winner are, relatively, more motivated to vote. Further, it is easier to convince the supporters of the ranked winner to shift to the other side, whereas the reverse is more difficult. If the Range winner was truly the regret minimizer, he or she is very likely to prevail in the runoff... unless it is all very close, in which case it's a toss-up, as it was originally, really. But the other method is one that I've generally rejected before even saying it, and I'm starting to think that was a mistake. This is to do exactly what is routinely done in direct democracies: ratify the election result, no election result is complete until it has been officially ratified by whoever is charged with that task, and in direct democracies, that is the assembly of citizens itself, normally. It's put up for a Yes/No vote on a motion to accept the result. This is quite enough! There is an obstacle to voting No on the ratification, which is that the election will then have failed, and the group will presumably then need to start over again. It tests preference strength, but it also tests the entire process, including issues of election fraud, everything. If the process is flawed *for whatever reason*, the electorate can simply reject it by majority vote. In fact it takes majority plus one to accept it.... and some organizations actually have supermajority rules for accepting certain offices.... In any case, both these methods make Range satisfy the Majority Criterion, and the second makes it satisfy the Condorcet Criterion, overall. Satisfying the Majority Criterion is politically very important. And, in fact, we can have the majority criterion and minimize regret *at the same time*. The trick, of course, is the second poll. >Will range and approval voters, therefore, just plurality-vote? >No. We know that does not happen because it doesn't happen. Well, there is some evidence in the other direction, and it's quite likely that Range and Approval would be different as settled, routine methods than in either non-election polls or actual elections when first implemented. The contrary evidence is the history of Bucklin. But, of course, the significance of that history has been twisted by FairVote. From the FairVote information on it, "only" 13% of voters added a second rank. But in a two-party system, that's huge! Why should a supporter of a major party add another candidate in second rank? They further report that in a series of elections, a majority winner was never produced by the addition of votes from the lower ranks. Again, this is misleading. This presumes that the goal of an election method is to produce a majority winner, the IRV people fall for this. IRV does not produce "majority winners," except by a trick, it stops considering discarded, exhausted ballots in order to declare that a candidate "won by a majority." Which is nonsense, and if all ranks were exhausted on all ballots, IRV will end up, almost certainly -- though I haven't seen the actual rules, with a plurality winner in spite of it all. The question, and the proper goal of election methods, is to choose the best candidate given the condition of the electorate. It's entirely possible that there is no majority winner at all. However, the real question is the response of the method to those few voters who *do* add second ratings or more. Counting these extra votes is not at all an onerous burden, certainly not compared to a real runoff! If most voters don't use the extra votes, if only 13% of voters add another rank, then counting those votes is 13% of the work involved in counting the whole election, and those votes will improve the results. These are independents and swing voters and third-party supporters. Far less than 13% would have turned Florida 2000. So, the truth: there will be a lot of bullet voting. The *full* truth: it's not a problem. As I wrote on BlueNC today, people instinctively realize that there is something harmful about bullet voting. It seems narrow and selfish. We know that it will, on average -- in Range -- increase the regret. *However*, the real question is "Does it increase regret enough to *forbid* voting 'sincerely'?" That is the paradox behind the FairVote argument. Range, even with lots of bullet voting, improves election results, as shown by the simulations, and it's not mysterious, though there is one interesting open question that I think is worth exploration. FairVote is hoodwinking people by making a reasonable-sounding argument that actually, followed in depth, would lead to a conclusion opposite to the one FairVote is trying to manufacture. FairVote believes that Approval does not have a prayer, politically. Why? Well, they know that there are certain knee-jerk arguments against Approval, most notably the one-person one-vote myth. Ossipoff points out how often he's run into that, talking with people, encountering vehement opposition on the basis that allowing people to vote for more than one violates this basic rule. FairVote has complete contempt, not only for experts on election methods, but for the people as well. They do not believe that people can be educated on this. I think differently, I think that people simply have not considered the matter sufficiently, and that there are ways around the block. The 1P1V error is an easy one to make. But it's also easy, if you can get a minute's attention from someone, to counter it. I think the fact that Approval is already used in public elections in the U.S., broadly (though not commonly, i.e., few elections actually take place where this multiple voting is allowed), with Ballot Questions, once realized, will help to turn the tide. It's already an established democratic practice. I know of at least three separate lines of argument that easily and quickly show that Approval does not violate 1P1V, and then there is the additional argument that even if it did, it's fair, everyone has the same power. >Here are some range voting and/or approval voting election data > >http://www.rangevoting.org/OrsayTable.html >http://www.rangevoting.org/FrenchStudy.html >paper #82 at http://math.temple.edu/~wds/homepage/works.html > >in all these real-world studies, >plurality style voting was quite rare, well in the minority. Warren, while the data there is quite interesting, these were polls, not elections. We don't know from them how voters will behave in actual elections, and I would expect some movement toward bullet voting. But it's beside the point. Many voters may bullet vote and it is just fine! Consider this: in a two-party system, we expect that, by definition, most people are supporters of one or the other of the two major parties. These voters have little incentive to do more than bullet-vote, and it would be quite rare that if they add another vote, it would actually change the outcome. So why should they bother to add second rank votes? And why should we care that they do not? Now, there are two groups of people for whom the situation is different: Third party supporters, who may now express full support for their favorite and full or partial support for their preferred frontrunner. Major party supporters who want to move their party toward the position of a third party, so they add vote strength to that party. Allowing this freedom to the voters will, I'm certain, either have no effect on outcome, or will improve it. The Bucklin history, we should really investigate; I have no access to the necessary documents. I highly suspect that Bucklin was dropped because it was working, not because of the supposedly poor use of second rank votes, which makes no sense. FairVote has presented, I'm sure, a distorted view of this history to advance their own cause. FairVote has also cited the dropping of Approval Voting by the IEEE. That history definitely shows that it was dropped for political reasons. The board adopted Approval to head off a spoiler candidate, and then dropped it once the job was done. Why didn't they leave it in place? Well, supposedly few members were using the ability to vote for more than one. But, if so, it wasn't costing anything. No extra votes, no extra counting. No, almost certainly, once the immediate purpose had been served, the board realized that they had, by allowing Approval, opened the door for dissident groups to start to build strength without spoiling elections. That's the last thing they wanted. So they dropped it. My question is, what in the world is a board of a membership organization doing manipulating the election method that is the means for the members to exercise proper control over board composition, without the approval of the members? And even if they submitted a question to the members, the way in which that question was worded, and the information going out to the members about it, would very likely be biased. Was there a debate? If so, where is it, I'd really like to see it! ---- Election-Methods mailing list - see http://electorama.com/em for list info
