Re: [EM] Majority-Judgement. Condorcet.
On 02/04/2012 01:07 AM, Kevin Venzke wrote:
Personally I don't understand why one would want to spend time on a
method that you have to defend by saying "it might work anyway," even
if as built the incentives are wrong.
I don't know if you're replying to me, but it seems to me that any
deterministic rated method that satisfies two simple criteria would have
an Approval vote as its maximally strategic vote, and voting
strategically can't strictly be worse (for each individual voter) than
not doing so.
The criteria are as follows:
- Per-candidate monotonicity: If you increase (decrease) the rating of
candidate X, holding all other candidate scores constant, that will
never reduce (increase) X's chance of winning.
- Per-candidate independence: Raising or lowering X's score while
holding other scores constant never makes the winner turn from A to B
(for some pair of other candidates {A, B}).
Range passes these. So does MJ.
By per-candidate monotonicity, increasing X's score to max will not hurt
X. By per-candidate independence, doing so won't make a compromise
candidate you might like lose, either.
So one may want to spend time on these methods because they're within a
certain class of methods (those that pass the criteria above), even if
that means that one can't help rewarding Approval-style ballots at least
as much as honest ones. (MJ's big feature is then that in a way, it
rewards Approval-style ballots as little as possible over honest ones.
It does the best it can, given the constraints.)
The name is so bad. Imagine you hear that on the news and are trying
to figure out what it means. "Majority" doesn't tell you that much
(IRV already does majorities and they didn't even need to put it in
the name) and "judgment" refers to what? The voting. They're calling
it "judgment" though. Puke. So dramatic and it doesn't even say
anything.
"Majority" says that, unlike other rated methods, it passes Majority.
"Judgement" means that you're supposed to compare the candidates to an
external grade standard (judge how good they are) instead of comparing
them to each other.
Election-Methods mailing list - see http://electorama.com/em for list info
Re: [EM] Majority-Judgement. Condorcet.
2012/2/3 Ted Stern > On 03 Feb 2012 16:07:59 -0800, Kevin Venzke wrote: > > > > Personally I don't understand why one would want to spend time on a > > method that you have to defend by saying "it might work anyway," > > even if as built the incentives are wrong. > > > > I like the idea of being able to test things, so I may be biased here. > > > > It's taking a shot in the dark. How fantastic must this method be, > > for that to seem like a good idea? It's hard to believe one couldn't > > go back and work out something that more reliably does whatever you > > were going for. > > > > Also, if MJ is a serious proposal it should be called "median > > rating" and use the Bucklin tiebreaker. You'd have a name that means > > something and a tiebreaker that isn't a pain to solve. At the top > > rating (the one we all agree might matter) the rules aren't even > > different. > > Can anyone explain how Majority Judgment differs in practice from > Bucklin with equal ratings allowed? AKA Fallback Approval? Or > one of the many versions of Majority Choice Approval (another vague > name, IMO)? > In practice? Not at all. Except there's a book about it. > > > The name is so bad. Imagine you hear that on the news and are trying > > to figure out what it means. "Majority" doesn't tell you that much > > (IRV already does majorities and they didn't even need to put it in > > the name) and "judgment" refers to what? The voting. They're calling > > it "judgment" though. Puke. So dramatic and it doesn't even say > > anything. > > > > The tie-breaker is the same thing really. It sounds neat and fair to > > pull out median votes one by one, but in practice that isn't the > > methodology, you really should use math. Try coding MJ and then see > > how much code you could delete, how much less thought it would've > > taken you, if you just wanted the Bucklin tiebreaker instead. > > > > And you can delete even more code if it is just ER-Bucklin. > Yup. But most voters don't worry about lines of code. JQ > > Ted > -- > araucaria dot araucana at gmail dot com > > > Election-Methods mailing list - see http://electorama.com/em for list info > Election-Methods mailing list - see http://electorama.com/em for list info
Re: [EM] Majority-Judgement. Condorcet.
2012/2/3 Kevin Venzke > Personally I don't understand why one would want to spend time on a method > that you have to defend by saying > "it might work anyway," even if as built the incentives are wrong. > At worst, it's approval. Not so bad. There's good reason to think that it will lead to more honesty than approval or range. People complain about approval, they really want to distinguish their favorite. So this isn't so much "it might work anyway" as "it probably has a bonus". > I like the idea of being able to test things, so I may be biased here. > OK, I'll net you know when I've tested real human behavior on Amazon Mechanical Turk. > > It's taking a shot in the dark. How fantastic must this method be, for > that to seem like a good idea? It's hard to > believe one couldn't go back and work out something that more reliably > does whatever you were going for. > I really don't see it as that ugly. > > Also, if MJ is a serious proposal it should be called "median rating" and > use the Bucklin tiebreaker. You'd have > a name that means something and a tiebreaker that isn't a pain to solve.At > the top rating (the one we all agree > might matter) the rules aren't even different. > > The name is so bad. Imagine you hear that on the news and are trying to > figure out what it means. "Majority" > doesn't tell you that much (IRV already does majorities and they didn't > even need to put it in the name) and > "judgment" refers to what? The voting. They're calling it "judgment" > though. Puke. So dramatic and it doesn't > even say anything. > > The tie-breaker is the same thing really. It sounds neat and fair to pull > out median votes one by one, but in > practice that isn't the methodology, you really should use math. Try > coding MJ and then see how much code you > could delete, how much less thought it would've taken you, if you just > wanted the Bucklin tiebreaker instead. > I didn't name it or choose the tiebreaker. I wouldn't have named it that or chosen that tiebreaker. I am sure, however, that these issues are trivial, compared to the fact that there's a book and academic articles promoting MJ. Any median system will elect the same winner in any real election. Jameson > > Kevin > > > > > Election-Methods mailing list - see http://electorama.com/em for list info > > Election-Methods mailing list - see http://electorama.com/em for list info
Re: [EM] Majority-Judgement. Condorcet.
On 03 Feb 2012 16:07:59 -0800, Kevin Venzke wrote: > > Personally I don't understand why one would want to spend time on a > method that you have to defend by saying "it might work anyway," > even if as built the incentives are wrong. > > I like the idea of being able to test things, so I may be biased here. > > It's taking a shot in the dark. How fantastic must this method be, > for that to seem like a good idea? It's hard to believe one couldn't > go back and work out something that more reliably does whatever you > were going for. > > Also, if MJ is a serious proposal it should be called "median > rating" and use the Bucklin tiebreaker. You'd have a name that means > something and a tiebreaker that isn't a pain to solve. At the top > rating (the one we all agree might matter) the rules aren't even > different. Can anyone explain how Majority Judgment differs in practice from Bucklin with equal ratings allowed? AKA Fallback Approval? Or one of the many versions of Majority Choice Approval (another vague name, IMO)? > The name is so bad. Imagine you hear that on the news and are trying > to figure out what it means. "Majority" doesn't tell you that much > (IRV already does majorities and they didn't even need to put it in > the name) and "judgment" refers to what? The voting. They're calling > it "judgment" though. Puke. So dramatic and it doesn't even say > anything. > > The tie-breaker is the same thing really. It sounds neat and fair to > pull out median votes one by one, but in practice that isn't the > methodology, you really should use math. Try coding MJ and then see > how much code you could delete, how much less thought it would've > taken you, if you just wanted the Bucklin tiebreaker instead. > And you can delete even more code if it is just ER-Bucklin. Ted -- araucaria dot araucana at gmail dot com Election-Methods mailing list - see http://electorama.com/em for list info
Re: [EM] Majority-Judgement. Condorcet.
Personally I don't understand why one would want to spend time on a method that you have to defend by saying "it might work anyway," even if as built the incentives are wrong. I like the idea of being able to test things, so I may be biased here. It's taking a shot in the dark. How fantastic must this method be, for that to seem like a good idea? It's hard to believe one couldn't go back and work out something that more reliably does whatever you were going for. Also, if MJ is a serious proposal it should be called "median rating" and use the Bucklin tiebreaker. You'd have a name that means something and a tiebreaker that isn't a pain to solve.At the top rating (the one we all agree might matter) the rules aren't even different. The name is so bad. Imagine you hear that on the news and are trying to figure out what it means. "Majority" doesn't tell you that much (IRV already does majorities and they didn't even need to put it in the name) and "judgment" refers to what? The voting. They're calling it "judgment" though. Puke. So dramatic and it doesn't even say anything. The tie-breaker is the same thing really. It sounds neat and fair to pull out median votes one by one, but in practice that isn't the methodology, you really should use math. Try coding MJ and then see how much code you could delete, how much less thought it would've taken you, if you just wanted the Bucklin tiebreaker instead. Kevin Election-Methods mailing list - see http://electorama.com/em for list info
Re: [EM] Majority-Judgement. Condorcet.
On 02/02/2012 09:40 PM, MIKE OSSIPOFF wrote: MJ: Thanks for all the answers about MJ strategy. But I'd told how easily a strategic faction can take advantage of and beat a sincere-voting faction. And, if the contest is close, then even a small difference in sincerity could decide which faction's candidate wins. And that amounts to a co-operatioin/defection problem too. No amount of speculation or discussion of MJ's other strategy issues or mystique will make that go away. Thanks, Kristofer, for confirming my conjecture: MJ strategy is like RV strategy. Don't forget the rest that I said, though. Maximal MJ strategy is like RV strategy. A Homo Economicus would vote Approval-style in a one-shot election. He might do so even when he thinks there are going to be elections after this one -- or he may not (as Jameson Quinn says). In any event, I think that MJ provides enough protection, were the majority's default sentiment to vote honestly, that they would not feel the need to vote strategically simply to head off a worse outcome. Furthermore, if there is some amount of strategy going on, the voters don't have to go all the way to a maximal ballot to defend their outcome. All they need to do, in order to support X over Y, is to vote X and Y on the other side of their respective medians. As the fraction of strategists goes to unity, the exaggeration needed goes to the maximum possible. Thus, if the voters aren't rational economic men (and turnout is evidence in itself that they aren't), the voters default to honesty (i.e. don't reason like Warren did), and the parties can't push large swathes of the people to vote strategically, then you shouldn't need to vote Approval style in an MJ election. In other words, in an MJ election, there's a certain room for honesty. If you get past this headroom, then you should vote somewhat strategically. If you get further past it, you should vote more strategically still. But as long as you're in the first area, then you have to get a lot of your friends to vote strategically too to make a difference. So the whole thing hinges on whether people will vote strategically even though they may not benefit over voting honestly. It's like a multiplayer prisoner's dilemma where, if some fraction f all defect, they get a greater payoff and the rest gets a sucker's, but if a fraction less than f tattles, they get the same payoff as if they hadn't. Would ordinary people defect? I don't think so, if f were large enough, because people are decent (i.e. not Homo Economicuses). Election-Methods mailing list - see http://electorama.com/em for list info
Re: [EM] Majority-Judgement. Condorcet.
2012/2/2 MIKE OSSIPOFF > > MJ: > > Thanks for all the answers about MJ strategy. > > But I'd told how easily a strategic faction can take advantage of and beat > a sincere-voting faction. > Not to my satisfaction. > > And, if the contest is close, then even a small difference in sincerity > could decide which faction's candidate wins. > > And that amounts to a co-operatioin/defection problem too. > > No amount of speculation or discussion of MJ's other strategy issues or > mystique will make that go away. > > Thanks, Kristofer, for confirming my conjecture: MJ strategy is like RV > strategy. > > This is for sure: In a u/a election, MJ's strategy is the same as that of > RV: Max-rate the acceptables and > min-rate the unacceptables. > This is not true. If sending a message about the relative value within either group is worth more than a thousand times less than winning the election, the rational strategy is to use the top two and the bottom two ratings. > > I conjecture that, in a non-u/a, 0-info election, MJ's strategy is > likewise identical to that of RV: Max-rate the > above-mean candidates and min-rate the below-mean candidates. > As above. > > I further conjecture that all MJ strategy, u/a and non-u/a, 0-info and not > 0-info, is the same as that of RV: > As above. > Max-rate all of the candidates whose merit, for you, is above your > expectation for the election. Min-rate all of > the candidates whose merit, for you, is below your expectation for the > election. > > Of course many Approval strategies have been discussed at EM, but they all > are instances, special cases, of the > above-stated better-than-expectation strategy. > > Condorcet: > > Condorcet is almost fine if you don't care about FBC or the > co-operation/defection problem. > > I've already told why I consider FBC and CD to be important. There's no > need to repeat that now. > > I've been saying that, when advocating a criterion, one should say why one > considers it important. I've amply done > that, regarding FBC and CD. > > So, if you advocate Condorcet, it's because we agree to disagree about FBC > and CD. > > Above, I said "almost fine". That's because, even aside from FBC and CD, > Condorcet doesn't work as well as I'd > formerly believed. My claims about Condorcet's powerful thwarting and > deterence of burial strategy were all based on > 3-candidate examples. > > Just as we all have been in denial about the co-operation/defection > problem, maybe I and some others hadn't wanted > to look at what can happen when there are more than 3 candidates in > Condorcet. > > Condorcet's thwarting and deterence of burial, it seems to me, doesn't > work nearly as well when there are more > than 3 candidates. That means that SFC doesn't mean as much. SFC was > central to my advocacy of Condorcet. > > So, for all of those reasons, I don't consider Condorcet, in any of its > versions, to be a good proposal for single-winner reform. > > In addition to Approval's FBC compliance, and the fact that CD compliance > can easily be added via the conditional > options, Approval is the answer to the question "There are so many voting > system proposals--Why should we choose > yours?". > > Approval's answer is: > > "Because Approval is the minimal, obvious, natural, easily-implemented, > improvement-change from Plurality. Plurality done right. Approval > transparently eliminates Plurality's worst problem, without adding one." > > And then, with the conditionality-options added, perhaps in a later > proposal, the method will have un-equalled > freedom from strategy problems, unmatched mild strategy. > > Could it be that the simplest and most obvious and natural is the also > best, in terms of getting rid of strategy problems? > > Yes. > > Mike Ossipoff > > > > > > Election-Methods mailing list - see http://electorama.com/em for list info > > Election-Methods mailing list - see http://electorama.com/em for list info
[EM] Majority-Judgement. Condorcet.
MJ: Thanks for all the answers about MJ strategy. But I'd told how easily a strategic faction can take advantage of and beat a sincere-voting faction. And, if the contest is close, then even a small difference in sincerity could decide which faction's candidate wins. And that amounts to a co-operatioin/defection problem too. No amount of speculation or discussion of MJ's other strategy issues or mystique will make that go away. Thanks, Kristofer, for confirming my conjecture: MJ strategy is like RV strategy. This is for sure: In a u/a election, MJ's strategy is the same as that of RV: Max-rate the acceptables and min-rate the unacceptables. I conjecture that, in a non-u/a, 0-info election, MJ's strategy is likewise identical to that of RV: Max-rate the above-mean candidates and min-rate the below-mean candidates. I further conjecture that all MJ strategy, u/a and non-u/a, 0-info and not 0-info, is the same as that of RV: Max-rate all of the candidates whose merit, for you, is above your expectation for the election. Min-rate all of the candidates whose merit, for you, is below your expectation for the election. Of course many Approval strategies have been discussed at EM, but they all are instances, special cases, of the above-stated better-than-expectation strategy. Condorcet: Condorcet is almost fine if you don't care about FBC or the co-operation/defection problem. I've already told why I consider FBC and CD to be important. There's no need to repeat that now. I've been saying that, when advocating a criterion, one should say why one considers it important. I've amply done that, regarding FBC and CD. So, if you advocate Condorcet, it's because we agree to disagree about FBC and CD. Above, I said "almost fine". That's because, even aside from FBC and CD, Condorcet doesn't work as well as I'd formerly believed. My claims about Condorcet's powerful thwarting and deterence of burial strategy were all based on 3-candidate examples. Just as we all have been in denial about the co-operation/defection problem, maybe I and some others hadn't wanted to look at what can happen when there are more than 3 candidates in Condorcet. Condorcet's thwarting and deterence of burial, it seems to me, doesn't work nearly as well when there are more than 3 candidates. That means that SFC doesn't mean as much. SFC was central to my advocacy of Condorcet. So, for all of those reasons, I don't consider Condorcet, in any of its versions, to be a good proposal for single-winner reform. In addition to Approval's FBC compliance, and the fact that CD compliance can easily be added via the conditional options, Approval is the answer to the question "There are so many voting system proposals--Why should we choose yours?". Approval's answer is: "Because Approval is the minimal, obvious, natural, easily-implemented, improvement-change from Plurality. Plurality done right. Approval transparently eliminates Plurality's worst problem, without adding one." And then, with the conditionality-options added, perhaps in a later proposal, the method will have un-equalled freedom from strategy problems, unmatched mild strategy. Could it be that the simplest and most obvious and natural is the also best, in terms of getting rid of strategy problems? Yes. Mike Ossipoff Election-Methods mailing list - see http://electorama.com/em for list info
