User talk:Homunq/If after all there is a vote
![]() | This page was nominated for deletion on 12 September 2013. The result of the discussion was userfy. |
For now
[edit]For now, I just read upto directly before "Relationship between MJ and CMJ (highly technical)", I will continue later; but I have some small comments:
- In the introduction to the section User:Homunq/WP_voting_systems#Mechanics, with CMJ B is the winner, not A, also "median" in the computation of the score of B should be 3, not 1.
- In User:Homunq/WP_voting_systems#Desirable_characteristics it is stated, that point 2 is connected to LNH. I don't see this connection.
- In the text there is sometimes and exclamation mark before some words, e.g. "!Voters", "!vote". Does this has a special meaning?
Best wishes, --Arno Nymus (talk) 15:54, 20 September 2012 (UTC)
- Thanks for stopping by.
- 1: oops, fixed.
- 2: LNHe can be taken as a guarantee that burial strategy will not work; LNHa as a guarantee that truncation won't work. Both of these will, all else equal, reduce voters' tendency to use strategy. (Yes, I consider certain kinds of LNHa failure, such as MJ's failure, to be a minor factor here; but there's still arguably a connection, however weak.)
- 3: WP:VOTE. I'll try to explain this in the text.
- Homunq (talk) 16:07, 20 September 2012 (UTC)
tiebreaker visually
[edit]I'd like to do something like this, but it needs work:
—scratch, moved to page— Homunq (talk) 15:20, 24 September 2012 (UTC)
Understandable?
[edit]Comments on the above? Homunq (talk) 20:08, 21 September 2012 (UTC)
- I have to admit I'm quite confused with the results calculation. -- Eraserhead1 <talk> 21:10, 22 September 2012 (UTC)
- Yeah, it needs work to make it clear graphically what's happening. The equations are just a way to confuse yourself, you need a graphical intuitive understanding... Homunq (talk) 04:37, 23 September 2012 (UTC)
Winners
[edit]I think you need another criteria - which is that the voting system produces a guaranteed winner - e.g. for Condorcet there isn't always a winner. -- Eraserhead1 <talk> 21:11, 22 September 2012 (UTC)
- That's part of number 4. As for "Condorcet" failing it, that's why I refer to Condorcet systems in plural; that is, the various different Condorcet completion methods which do resolve. Homunq (talk) 04:35, 23 September 2012 (UTC)
- Fair enough. My mistake. -- Eraserhead1 <talk> 15:31, 23 September 2012 (UTC)
Strong Oppose
[edit]Any system which gives people the option of increasing the effect of their vote by prefixing it with strong is wide open to gaming. The logical tactic in such systems is always to express a strong vote for or against whenever you vote. Where you have multiple candidates/options then you can rank them in order of preference and there are several systems such as STV that cover that including some systems that work well. But allowing people to have three first choices and two fifth choices is not going to give as good a system as getting people to rank candidates 1-5. ϢereSpielChequers 00:09, 25 September 2012 (UTC)
- The above commenter has utterly failed to understand this proposal. Or perhaps I've failed to explain it. Either way, the truth is: changing a vote on a winning proposal from "oppose" to "strong oppose" will NOT affect the results in any way; and the only way changing from "support" to "strong support" could possibly matter is if the principal difference between the top two options is the amount of strong support they receive, in which case is is entirely appropriate to adjudicate based on that difference.
- As if that weren't enough, this commenter also failed to note that a voter who simply forgot the word "strong" in all cases would have it inferred in the counting.
- Essentially, this is an argument in favor of desirable characteristic #2. It therefore would cut clearly in favor of MJ over Range and Condorcet over Borda or IRV. It also weakly favors MJ over Condorcet over Approval. Homunq (talk) 15:23, 25 September 2012 (UTC)
- Perhaps it is simply that the proposal is over complex. But why in the section mechanics do you start "Voters would be encouraged to class the options into no more than 5 groups: those they "strongly support", those they "support", those they find "neutral", those they "oppose", and those they "strongly oppose"" if you don't intend to give any extra weight to those "strong" votes? Either this needs to come out as !voting strong has no effect, or it does have an effect and my criticism stands. By the way you've dropped the "weak" option which we currently have. Sometimes people use this as a valid option just one side or the other of neutral. ϢereSpielChequers 16:23, 25 September 2012 (UTC)
- As to complexity: I understand that this is a problem. I can only respond that the proposal considers literally every plausible alternative, and finds each of them inferior in one or more specific ways. (OK, I didn't specifically address Condorcet-like non-Condorcet systems, but everything I said about Condorcet applies to these as well.)
- As to strategic exaggeration: The only plausible¹ way that would make a difference is if the election came down to two generally-supported options that simply had different levels of support. In that case, over half of the people are supporting each option, so some people must be supporting both. The distinctions those "both are good" voters make between support and strong support are in general honest, not exaggerated; and in fact are in my opinion the best way to decide the matter; so in this case I think Majority Judgment is doing the right thing.
- ¹I include the weaselly "plausible" because there are two other ways it could theoretically matter, both of which I find completely implausible. One is that if two options are exactly tied in SS, S, and N, then the difference between their SO and O scores could matter. Such a precise tie on three variables at once is wildly improbable. The other is that the two best options have a median below neutral, and so the difference between their O and SO scores matters. But this is implausible because I believe that any closing administrator would rather find "no consensus" than choose a winning option whose median vote is some form of opposition.
- (Note: you may have noticed that I've essentially argued that the difference between O (opposition) and SO (strong opposition) will never matter. You could then ask why allow the difference if it never matters. The reason is that people will vote more honestly if they can do so and still express all the distinctions which matter to them. Allowing the O/SO distinction will therefore help us understand the issues and could help motivate a creative, consensus-seeking proposal.)
- As to the "weak" option: it would be easy to include it; it would take no changes to the basic voting system. However, it would make counting things a bit harder; especially when !voters don't use the right magic words (which some never will). On balance, I'd guess it's not worth it, but I could be convinced otherwise.
- So, does all that make sense? Any other questions? And if not, do you still "strongly oppose" this idea? Homunq (talk) 17:06, 25 September 2012 (UTC)
- Also, regarding weak support: I suspect that most people use only 2 options out of "strong", "weak", and flat. In that case, it would be easy to translate whatever 2 options are used into the 2 that are available. Homunq (talk) 19:02, 25 September 2012 (UTC)
- Perhaps it is simply that the proposal is over complex. But why in the section mechanics do you start "Voters would be encouraged to class the options into no more than 5 groups: those they "strongly support", those they "support", those they find "neutral", those they "oppose", and those they "strongly oppose"" if you don't intend to give any extra weight to those "strong" votes? Either this needs to come out as !voting strong has no effect, or it does have an effect and my criticism stands. By the way you've dropped the "weak" option which we currently have. Sometimes people use this as a valid option just one side or the other of neutral. ϢereSpielChequers 16:23, 25 September 2012 (UTC)
RFC: should this progress towards eventually become a guideline? How?
[edit]This is an essay on voting on Wikipedia. Though the overwhelming majority of cases are already covered adequately by WP:NOTVOTE, there is a small minority of cases which still need an explicit vote-counting system, and this essay proposes such a system. It appears to have some support as a practical system. Should I be trying to push it towards guideline status, and if so, what would that entail? Homunq (࿓) 14:41, 2 October 2012 (UTC)
Critique and response
[edit]I'm commenting on the original post advocating Majority Judgment (MJ) voting for wikepedia.
For that purpose, I'm pasting it into this comment space.
First of all, the author of the post mis-defines the criterion Later-No-Help (LNHe). LNHe isn't about "burial strategy". It just means that, while marking you ballot, when you've voted for some candidates, there is no need to vote for additional candidates in order to fully help those candidates for whom you've already voted.
But LNHe is a valuable criterion, and MJ indeed meets that criterion--as do the Approval voting system and the Score voting system (also sometimes referred to as "Range").
The author lists some "criteria", but seems to want to avoid recognized criteria.
The author presents Approval and Score unfavorably, claiming that MJ improves on them. As an advocate of Approval and Score, I therefore would like to reply.
The best way to reply is to quote text, and comment on it:
The author says:
1.Easy for voters Majority judgment, a "scored" system which uses verbal grades by preference, passes with flying colors, as the example above shows.
[endquote]
That can likewise be said for Approval and Score.
Author says:
2.It should not encourage any obvious strategy by which a partisan minority could easily try to sway the result
[endquote]
That reveals the author's belief in a fallacy. "Swaying the result" is the whole point of voting. Gibbard & Satterthaite demonstrated that all voting systems have strategy. That most definitely includes MJ, in which strategy is just as necessary as in Approval or Score.
This "don't let strategy succeed" implies a belief that there's some more noble and ethical way to vote, called "sincere voting". And what does the author mean by "sincere" voting? He's referring to giving the candidates ratings proportional to the voter's perdeption of the candidates' utilities for that voter. But why should the voter want to do that? And what makes it "sincere"? Of course it would be sincere if I asked you to vote in that way, and you promised that you would. Then, any other way of voting would be dishonest. But such an implied agreement is an entirely unjustified assumption. So, when the author says MJ is better because it doesn't let departure from utlity-proportional rating make a difference, he's expecting you to share his belief that there is something especially valid, right, ethical or honest about utility-proportional ratings. There isn't.
Author says:
No voter or minority subset of voters who agree on the rating an option deserves can do anything strategically to bring the median closer to that rating.
[endquote]
Wrong. Whatever rating they all give to that candidate, they thereby move his median closer to that rating.
Author says:
Specifically, it would NOT help a minority of voters to either exaggerate a vote (by adding "strong")
[endquote]
Neither Approval, nor Score, nor MJ have an option consisting of "adding 'strong' ".
I don't know what the author means by that, but later in his post he touts MJ's thwarting the Score strategy of extreme rating. In Score, your best strategy is to rate all of the candidates at the extremes. Top-rate the candidates you'd approve in an Approval election, and bottom rate the others.
And which ones should you approve in an Approval election? The ones that are better than your statistical expectation for the election.
That's obviously your best strategy in MJ as well. Increasing the win-probability of a candidate who is better than your expectation will obviously raise your statistial rating. So, maximally helping every above-expectation candidate maximizes your statistical expectation. That means that it's your optimal strategy.
And how do you maximally help a candidate? How do you increase his/her win probability as much as you can? Top rate him/her. That's true in Score, and it's true in MJ.
That isn't devious. That isn't dishonest. It's called "voting in your best interest". Trying to rate the candidates in proportion to your estimate or perception of their utilities to you? Why?? It's not as if you've promised to. It's not as if the rules require it.
Often there will be options that are unacceptable, but which could win. I call that a u/a election (for unacceptable/acceptable). In a u/a election, your the "above expectation" strategy amounts to approving every acceptable candidate, and none of the unacceptable candidates.
Aside from that, sometimes you don't have a good estimate of expectation, but you still have a feel for which options to approve. Certainly the ones that you like. Maybe, guided by subjective intuition, something you feel you might need as compromise, if your favorite(s) might not win.
In any case, though, if you'd approve it in Approval, then you should top-rate it in Score or MJ.
But if you don't know whether you should top or bottom rate a candidate,of course then you could give him/her some inbetween rating, or flip a coin in Approval.
The only other exception to that extreme-voting strategy is when there's a "chiken dilemma". For brevity, I shouldn't define it here. Basically the A preferrers and the B preferrers all prefer both A and B to C. They all despise C. They add up to a majority, so by combining their approvals, they oould easily defeat C. So the A voters approve B. But the B voters expect that, and, in order to ensure that B wins instead of A, they refuse to approval A. They give A a bottom rating.
In that situation, a good strategy is for the A voters to give to B some inbetween rating. ...enough to help B beat C only if B's faction is larger than C's faction. (Making B the rightful winner among {A,B}). It's guesswork, of course, but the B voters' guesses aren't any better than yours, and so if it's known that you're voting in that way, there is some deterrence of defection.
I call those inbetween ratings "Strategic Fractional Ratings (SFR).
I mention that, because it's the reason to give inbetween ratings in Score and MJ.
So what's the difference, between Score and MJ in that regard? The difference is that SFR is much more complicated and difficult in MJ, and requires probabilistic voting (using some chance means for deciding what to do).
But here's an obvious reason why MJ is more poorly suited to SFR:
In the chicken dilemma, the time when an A voter wants to help B win is when B would outpoll A anyway, and is the rightful one to be helpted to beat C. But, in MJ, when B has a large median, B's median is more likely to be above your inbetween rating of B, with the result that your rating will bring B's median downward. And if B is quite low, then your inbetween rating is more likely to raise B's median. In other words, MJ's inbetween ratings act oppositely to what is desired for SFR.
MJ requires elaborate tiebreaking bylaws. MJ doesn't have the simple and natural obviousnes of Approval and Score.
To summarize: The point of voting is to influence the outcome. To do so in a way that maximizes your expectation is called optimal strategy. That isn't dishonest, devious or unethical. It's simply right voting.
In Score and MJ, optimal voting consists of rating at extremes. Top rate those whom you'd approve in Approval, and bottom-rate the others. The ones you top-rate should be the ones who are better-than-expectation. Or the acceptable ones. Or the ones you like, and maybe some that you think you might need.
But when the chicken dilemma gives you reason to inbetween-rate, SFR, MJ makes that more complicated and difficult.
That's the strategic difference between MJ, Score and Approval.
Author says:
(Strategy is still a possibility, but it can only work by making an option get a median higher or lower than you think it deserves
[endquote]
...And what median does an option "deserve"? That notion is fallacious. Maybe you'd like an option to win, maybe you want it to lose. Maybe you'll support it as a compromise that you need in order to beat something worse. But there are only two results for an option: Winning or not winning. The voter doesen't care that an option get a certain final score. That isn't what the election is about.
Author says:
[edit] Inferior alternatives for this use case [edit] Approval voting
This does well on 4, 5, and 6, but is barely acceptable on 1, 2, and 3.
Nonsense. Author's "criteria" are based on strategic fallacies.
MJ, Approval and Score meet the same broad strategy criteria. The strategic difference is MJ's more difficult SFR.
There are a few "embarrassment criteria" that MJ fails and which Approval and Score pass. But that only matters if the methods are proposed for official public elections. I don't claim that the embarrassment criteria are important for an organization voting on options.
Author says:
[edit] Range voting / Score voting
Score voting and MJ are actually quite similar on the whole; as similar as the mean and the median. Thus, even though there is not a huge difference on the desirable characteristics, what difference there is is directly comparable. This allows us to see that MJ is detectably better on 1 and 3, and considerably better on 2.
Specifically, in Score voting, strategic exaggeration is always a viable strategy for any voter whose honest vote isn't already fully exaggerated; in MJ, it almost never is.
[endquote]
This is a product of the author's fallacious strategic beliefs that I mentioned earlier. Rating at the extremes is the optimal strategy. It isn't dishonest or devious. Author calls it "exaggeration", because Author thinks that (for some reason) voters should give ratings in proportion to the options perceived utilities for them.ll Fallacy.
I don't believe in opposing the proposals of other election reform advocates. But this author's post incorrectly claims that his Majority Judgment is better than Approval and Score. — Preceding unsigned comment added by 98.77.169.209 (talk) 06:01, 5 October 2012 (UTC)
- The above commenter clearly knows what they are talking about in terms of electoral systems (and indeed, I believe I recognize their Mode of Operations from the election methods discussion listserve). I therefore thank them for their comments here. However, I believe that they have systematically misread this essay, so they end up correctly refuting points I did not make, while not addressing the points I did make. This is certainly understandable, as the article as written often favors brevity over precision of language (since, to a non-expert, it is already TLDR.)
- Point by point:
- 1. mis-defines the criterion Later-No-Help (LNHe). LNHe isn't about "burial strategy". It just means that, while marking you ballot, when you've voted for some candidates, there is no need to vote for additional candidates in order to fully help those candidates for whom you've already voted. Correct on textual definition of LNHe. But this definition implies the need for burial strategy. If a secondary vote for Z is necessary to help X, but I prefer X>Y>Z, my most-strategic ballot may be X>Z>Y, which is the definition of burial.
- 2. lists some "criteria", but seems to want to avoid recognized criteria. I have no problem discussing mathematical criteria, but non-experts need to understand why those criteria matter. That's why I've put the 6 desiderata in non-mathematical terms.
- 3. The author presents Approval and Score unfavorably, claiming that MJ improves on them. While I agree that Score and Approval are good, you need only look at the "Strong Oppose" subsection above to see why some wiki editors will find those systems unsuitable.
- 4. That can likewise be said for Approval and Score. Agreed, as the essay says.
- 5. That reveals the author's belief in a fallacy. "Swaying the result" is the whole point of voting. Gibbard & Satterthaite demonstrated that all voting systems have strategy. That most definitely includes MJ, in which strategy is just as necessary as in Approval or Score. You are correct that Gibbard's proof applies to MJ (not Satterthwaite's... the theorems are only equivalent for ordinal/"comparative" systems, but that's a detail) . However, it only proves that there must be some theoretical cases where there is no single "dominant" vote. It doesn't say that a system must be susceptible to insincere (ie, ordinally-dishonest) strategy; nor does it show that the cases of strategic susceptibility must be common in real-world elections. In fact, as Endriss 2007 showed, Approval (and thus its extensions Score and MJ) is not susceptible to insincere strategy for vNM-rational voters. And as the arguments below show, MJ is even less susceptible to strategy than the other two.
- 6. This "don't let strategy succeed" implies a belief that there's some more noble and ethical way to vote, called "sincere voting". Commenter is putting words in my mouth. It's not about nobility or ethics. It's just that if some voters use one algorithm (call it "sincere", "naive", whatever) and others use another ("strategic", "sophisticated", whatever), an ideal voting system would tend to give those two groups voting power proportional only to their size. Yes, G-S shows that this ideal is impossible to obtain, but it can be approached, as in MJ (or fled, as in Score).
- 7. Wrong. Whatever rating they all give to that candidate, they thereby move his median closer to that rating. But they do so by precisely the same amount whether or not the exaggerate their vote to an extreme grade. This is the key point that the commenter has missed.
- 8. Neither Approval, nor Score, nor MJ have an option consisting of "adding 'strong' ". Commenter has apparently not read the examples here, which use the ratings "Strong Support", "Support", "Neutral", "Oppose", "Strong Oppose". If you want to translate that into numbers for use with Score, then "adding 'strong' " means changing a 1 to a 0 or a 3 to a 4.
- 9. That's obviously your best strategy in MJ as well. Increasing the win-probability of a candidate who is better than your expectation will obviously raise your statistial rating. So, maximally helping every above-expectation candidate maximizes your statistical expectation. That means that it's your optimal strategy. This is false. You are already maximally helping any candidate whom you grade above their median; exaggeration/top-rating is not required.
- 10. That isn't devious. That isn't dishonest. Again, it's not a moral judgment; see 6 above.
- 11. The only other exception to that extreme-voting strategy is when there's a "chiken dilemma"... Commenter gets into an extremely technical point. I don't agree with his arguments, but neither of us has a mathematically-bulletproof case; it depends on voter behavior (and I'm currently running a study of that on Amazon Mechanical Turk.) But even if he is correct, he is only proving that Score and Approval are not worse than MJ in this regard. I'd claim that MJ is better. And I'd also claim that on the contentious questions that move to a vote on wikipedia, this whole line of argument is unlikely to be relevant.
- 12. The voter doesen't care that an option get a certain final score. That isn't what the election is about. You are right; many voters will think that way, and for those voters this particular argument in favor of MJ will not hold. But for any voters who do have some notion of an absolute grade that an option deserves, this argument does hold. In any case, there are other arguments.
- 13. MJ, Approval and Score meet the same broad strategy criteria. But desiderata 1 and 3 aren't "strategy criteria", and you're wrong as shown above on desideratum 2.
- 14. This is a product of the author's fallacious strategic beliefs that I mentioned earlier. Yes, we've covered this earlier.
Critique and Response, continued
[edit]: 1.
I'd said:
mis-defines the criterion Later-No-Help(LNHe). LNHe isn't about "burial strategy". It just means that, while marking you ballot, when you've voted for some candidates, there is no need to vote for additional candidates in order to fully help those candidates for whom you've already voted.
Author says:
Correct on textual definition of LNHe. But this definition implies the need for burial strategy. If a secondary vote for Z is necessary to help X, but I prefer X>Y>Z, my most-strategic ballot may be X>Z>Y, which is the definition of burial.
[endquote]
The point is that you don't need to vote for additional candidates in order to fully help the ones for whom you've already voted. Period. In Author's example, not only should you not have to bury by ranking Z over Y, but, to fully help X, you shouldn't even have to vote for Y or Z at all. Though a need to add "...Z>Y" would indeed be a way to fail LNHe, LNHe would be equally violated if you needed to vote X>Y>Z, instead of just X, to fully help X. As I said, it isn't about burial.
2. lists some "criteria", but seems to want to avoid recognized criteria.
Author says:
I have no problem discussing mathematical criteria, but non-experts need to understand why those criteria matter. That's why I've put the 6 desiderata in non-mathematical terms.
[endquote]
But your criteria are too vague.
3. The author presents Approval and Score unfavorably, claiming that MJ improves on them.
While I agree that Score and Approval are good, you need only look at the "Strong Oppose" subsection above to see why some wiki editors will find those systems unsuitable.
[endquote]
Then they might prefer Approval to Score. Approval doesn't have a range of score ratings. You approve something or you don't.
Score differs from Approval, in offering inbetween ratings, useful for SFR (strategic fractional rating), in the event of a chicken dilemma situation. Optimal voting calls for not using inbetween ratings unless there's a chicken dilemma.
In Approval, you can probabilistically simulate inbetween ratings, by flipping a coin, or drawing a numbered piece of paper from a bag.
5. That reveals the author's belief in a fallacy. "Swaying the result" is the whole point of voting. Gibbard & Satterthaite demonstrated that all voting systems have strategy. That most definitely includes MJ, in which strategy is just as necessary as in Approval or Score.
Author says:
You are correct that Gibbard's proof applies to MJ (not Satterthwaite's... the theorems are only equivalent for ordinal/"comparative" systems, but that's a detail)
[endquote]
Fine. I've always heard that the Gibbard-Satterthwaite theorem applies to all non-probabilistic, non-dictatorial, voting systems. But whatever...
The point is that all voting systems have strategy. Some way of voting will maximize your expectation. But there's no reason to believe (unless by coincidence) that that will be (in a rank method) a sincere preference ordering, or (in a ratings method) ratings that are proportional to the options' utilities for you. Contrary to popular belief, in a ratings method, there is nothing special or, in some way, more right, about ratings proportional to the options' utilities for you. And that doesn't happen to be optimal. As I said, there's no particular reason to expect it to.
Author says:
And as the arguments below show, MJ is even less susceptible to strategy than the other two.
[endquote]
"Susceptible to strategy" implies that there is offensive strategy to be susceptible to. Approval and Score don't have offensive strategy, except for defection in the chicken dilemma. And the difference between Score and MJ is that MJ makes it more difficult to deal with that, as I've said.
The "stategy" that Author is saying that Score is "susceptible to" is extreme ranking. Extreme ranking, in Score, is the optimal, right way to vote. It's meaningless to speak of Score's "susceptibility" to that. MJ is "susceptible" to it as well, in the sense that it's also the optimal way to vote in MJ.
What Author means is that, in MJ, an inbetween rating will move an option's median as far as will an extreme rating. It will move it in an unknown direction. Why that's better, Author never quite explains.
6. This "don't let strategy succeed" implies a belief that there's some more noble and ethical way to vote, called "sincere voting".
Author says:
Commenter is putting words in my mouth. It's not about nobility or ethics. It's just that if some voters use one algorithm (call it "sincere", "naive", whatever) and others use another ("strategic", "sophisticated", whatever), an ideal voting system would tend to give those two groups voting power proportional only to their size.
[endquote]
What? :-) If, in an ideal voting system, it doesn't matter how you vote, then, in an ideal voting system, there should be no need for input from voters.
Alright, so Author isn't saying that ratings proportional to the options' utilities for you are more noble or ethical. But, for him, they must be more...something, because he wants to reduce the amount by which an extreme vote will move an option's median, in comparison to how far a "sincere" vote will move an option's median. So, in MJ, any rating will move an option's final score by the same amount (but if the rating isn't extreme, it moves that median in an unknown direction).
I said that, even in MJ, extreme rating is optimal. Author disagreed, saying that inbetween ratings move an option's median just as far. But the difference is that, with the extreme rating, you know in which direction you're moving that option's median :-)
Author says:
Yes, G-S shows that this ideal is impossible to obtain, but it can be approached, as in MJ (or fled, as in Score)
[endquote]
Both methods have the same optimal strategy when there isn't a chicken dilemma. The difference is that, when there is a chicken dilemma, MJ makes it much more complicated to deal with.
This ideal that MJ approaches, and which Score flees--Does Approval flee it too? Presumably Author is referring to some ideal of his, but he isn't being very forthcoming with us about exactly what his ideal is. ...unless he really means that ideally it shouldn't make any difference how we vote. That isn't what you meant? Forgive me; I'm just trying to guess.
The first thing that Author needs to do is be very specific about what his ideal is. Only then can it be possible to discuss how well various methods approach his ideal.
I'm guessing that, in the belief that there is someting special about ratings proportional to the options' utilities for you, Author doesn't want an extreme rating to affect an option's final score more than utility-proportional ratings can. Am I guessing right?
But MJ doesn't change optimal strategy. It's still the same as in Score. What MJ does is make chicken dilemma strategy more complicated. At the forum, I told Author how anti-defection strategy would work in MJ. It was complicated. I invited Author to tell of a different anti-defection strateg for MJ. He didn't have an answer.
Author says:
. 7. Wrong. Whatever rating they all give to that candidate, they thereby move his median closer to that rating. But they do so by precisely the same amount whether or not the exaggerate their vote to an extreme grade. This is the key point that the commenter has missed.
[endquote]
Yes, but if it's an extreme rating, then they know in what direction they're moving his rating :-) Some would say that's desirable.
That's why extreme rating is optimal in MJ, just as in Score.
And note that Author says, "whether or not they exaggerate their vote to an extreme grade". What would their vote be if they didn't "exaggerate" it? Author wants to say that an extreme rating is an "exaggeration" of something. Then what is it an exaggeration of? In other words, what would their vote be if they didn't "exaggerate" it.
I suspect that we already know Author's answer to that: What their vote would be, if they didn't exaggerate it, would be ratings proportional to the options' utilities for the voter. In order for you to be able to exaggerate something, it must have some true, unexaggerated value. Author wants to say that utility-proportional ratings are a voter's true, unexaggerated ratings. Why? What makes utility-proportional ratings more true in some sense, more right? Yes, utility-proportional ratings are certainly more unstrategic, in the sense of not having anything to do with optimality or expectation-maximization. But, then so would random ratings too. What is special about utility-proportional ratings? That fallacy that they're special, or somehow more true, is the fallacy at the center of Author's other fallacies.
Yes, extreme ratings are strategy. They're optimal strategy. What's wrong with that? Author has got to stop and ask himself these questions.
9. That's obviously your best strategy in MJ as well. Increasing the win-probability of a candidate who is better than your expectation will obviously raise your statistial expectation. So, maximally helping every above-expectation candidate maximizes your statistical expectation. That means that it's your optimal strategy.
Author says:
This is false. You are already maximally helping any candidate whom you grade above their median
[endquote]
...but you don't know where their median is. The only way to be sure that you're moving their median in the direction that you want to move it, is to rate them at an extreme.
Say, with available ratings from 0 to 10, if I rated the candidates proportional to their utility to me, I'd give X a 7 and Y a 5. X is a compromise whom I probably need in order to beat Y. So should you rate them 7 and 5, repectively, in MJ? What do you think? If their medians are both above 7, and they're the top 2 contenders, then those utility-proportional ratings will lower their ratings equally (as nearly as can be guessed). Is that really what you want to do?
If you want to fully help X, then you rate him at max. And if X is above expectation for you, and Y is below expectation for you, then you best help your expectation by reliably raising X, while reliably lowering Y. You do that by top-rating X and bottom rating Y.
Author says:
- exaggeration/top-rating is not required.
[endquote]
...unless you want to know which way you're moving the candidates' final scores :-)
And, as for "exaggeration", what is this thing that you think is being exaggerated by extreme rating? And what makes you think that it shouldn't be "exaggerated"?
11. The only other exception to that extreme-voting strategy is when there's a "chiken dilemma"... Commenter gets into an extremely technical point. I don't agree with his arguments, but neither of us has a mathematically-bulletproof case; it depends on voter behavior (and I'm currently running a study of that on Amazon Mechanical Turk.) But even if he is correct, he is only proving that Score and Approval are not worse than MJ in this regard.
[endquote]
Chicken dilemma involves emotional considerations, such as deviousness, distrust, and dislike for the defector's candidate, and the perceived culpability of the other faction, etc. It isn't really a mathematical subject. In Score (or probabilistially in Approval), give the likely defector's candidate just enough to (as you subjectively judge it) to help that candidate win only if s/he is supported by a larger faction than your candidate is. Guesswork, but the likely defector doesn't have a more accurate guess, and will hesitate to defect if he knows you're doing that. If your candidate polls high, and his lower, then he knows you might not be giving his low-scoring candidate enough to win, and he should help your stronger candidate. Both mutually-distrustful factions should give eachother's candidate such strategic fractional ratings (SFR).
So: Is Score, or MJ, better in that regard? As I've said, MJ's SFR is much more complicated and difficult than that of Score and Approval. I described it to Author, at the forum, and I invited him to tell of an alternative, easier, SFR for MJ. Author had no answer. — Preceding unsigned comment added by 184.32.143.151 (talk) 17:06, 5 October 2012 (UTC)
- I feel that at this point we're just restating our positions. The crux of the argument is the attitude towards "strategy". Commenter seems to feel that since there is no single, a priori definition of "honest" in a cardinal/evaluative system like MJ, Score, or Approval, then any talk of strategy must be misguided moralizing. I believe that aside from all moralizing, the fact is that different voters will use different strategic algorithms, and that the chosen system, in the situations it's used for on Wikipedia, should be as insensitive as possible to those variations, while remaining as sensitive as possible to the preferences/utilities/hunches/whatever that voters use as inputs to those strategic algorithms. That is the spirit of desideratum 2. If you disagree with me on desideratum 2, then score voting is arguably comparably good to MJ/CMJ; perhaps a shade worse on 1 and 3, and better on 4, 5, and 6. This essay is still useful in its analysis of systems other than Score (including Approval, which was dismissed for reason 1 primarily, not reason 2). But as I pointed out earlier, the "strong oppose" section above is evidence that at least some people agree with me (or even surpass me) in their desire to meet desideratum 2.
- Commenter: If you want to continue this argument, you have my email, and I have yours. Homunq (࿓) 17:27, 5 October 2012 (UTC)
Conclusion of Critique and Reponse (Michael Ossipoff)
[edit]Author says:
I feel that at this point we're just restating our positions.
[endquote]
Quite. We're repeating instead of answering. I asked you several simple direct questions.
I asked you what it is that is being "exaggerated" when we extreme-rate, and why it shouldn't be exaggerated.
I asked you to be very speicific about this "ideal" that you have referred to.
Instead of answering, you prefer to just repeat.
Author says:
The crux of the argument is the attitude towards "strategy".
[endquote]
The crux of the argument is about what you mean. You evidently are unwilling to answer my questions about what you mean.
If there is a "right" way vote, it's the way of voting that maximizes your expectation. Simple as that. That's called "optimal strategy".
If you want to claim otherwise, then you need to be very specific about what you mean.
Author says:
Commenter seems to feel that since there is no single, a priori definition of "honest" in a cardinal/evaluative system like MJ, Score, or Approval, then any talk of strategy must be misguided moralizing.
[endquote]
"Vague" is the word that I'd use. You are unable to answer my question about what it is that extreme rating exaggerates, and why it shouldn't be exaggerated.
I pointd out that, by helping an above expectation option, you raise your expectation. Therefore, by maximally helping all of your above-expectation options, you maximize your expectation. You do that in MJ by reliably raising them, and reliably lowering the below-expecation candidates. You do that by top-rating the above-expectation candidates and bottom-rating the below-expectation candidates.
Voting to maximize your expectation is the obvious right way to vote. If you want to claim that another way to vote is better, then you need to specify that way of voting, and explain why it's better. If you want to say that another way of voting is more "right" (whatever that means), then you need to specify that way of voting, and tell us in what way it's more right.
In other words, you need to start sharing with us what you mean. The questions that I asked you were intended to encourage you in that direction.
Author says:
I believe that aside from all moralizing, the fact is that different voters will use different strategic algorithms, and that the chosen system, in the situations it's used for on Wikipedia, should be as insensitive as possible to those variations
[endquote]
In other words, Author wants the voting system to be unresponsive, in the sense of making your expectation, as nearly as possible, independent of how you vote. If that isn't what Author means, then perhaps he'd like to tell us what the really means.
Yes, we're just repeating. But that's because Author would rather repeat his assertions than answer my questions about what he means.
He's incorrect to say that I, too, am just repeating assertions. I'm asking questions, to try to get at what Author means.
Author says:
, while remaining as sensitive as possible to the preferences/utilities/hunches/whatever that voters use as inputs to those strategic algorithms.
[endquote]
That's too vague. ...sensitive to exactly what, in particular? And what would it mean to be sensitive to it? If it should be sensitive to the "utilities" that Author refers to, what would it mean to be sensitive to them? And who decides these "should"s?
Does he mean sensitive to the ordinal prefeences expressed on the ballot? Certainly, encouraging people to just express all of their ordinal preferences, making that their optimal strategy, is an ideal. I propose a rank-count that comes closest to that ideal, though that ideal can't really be achieved. That proposed rank-count at least gets rid of the worst strategy-needss that most distort the expression of preferences in a rank balloting. But this isn't the place to introduce it. (Unless wikipedia would be interested in the best, most sincerity-optimal, rank-count for its voting). The method I refer to is called "Symmmetrial ICT", and can be found in the electowiki, under that name. "ICT" stands for Improved-Condorcet-Top". Symmetrical ICT is the result of contributions from 3 people at the election-methods mailing list. Pseudocode for a Symmetrical ICT count is available upon request. Make your request at the election-methods mailing list. I assume that it's permissible to name that mailing list here, because Author already mentioned that that is where he knows me from.
But it might not be permissible to give a URL here, which is why I don't give the URL where that pseudocode can be found.
To answer generally, regarding Author's "desideratum 2", yes it's desirable to have a voting system that is sensitive to voters' expressed preferences, in the sense that optimal strategy doesn't require drastic violation of your preferences. The rating systems, such as Score, Approval, and MJ, all have, as optimal strategy, rating all candidates at extremes, thereby leaving many preferences unexpressed. Sorry, but that's just how it is. If you don't like that, then use Symmetrical ICT instead of a rating system.
But Approval, Score and MJ will never give strategic need to vote someone over your favorite, or to vote for additional candidates in order to fully help candidates for whom you're already voting for. Few methods have those desirable properties. The ideal that I spoke of is unattainable, though Symmetrical ICT approaches it more closely.
Michael Ossipoff
(I don't know if I'm allowed (or required) to give my e-mail address in this signature, but, as Author stated earlier, I can be reached by messages to the election-methods mailing list. Look for it in google) — Preceding unsigned comment added by 184.32.143.151 (talk) 19:22, 5 October 2012 (UTC)
- Say there are two voters, P and Q. When you ask them informally, both tell you exactly the same thing: "I really like option X. Option Y is pretty good, but not as good. Option Z is pretty bad, but not the worst I've ever heard". But then when it comes to voting, P thinks "I don't want to have to think about voting, so I'll just say 'strong support' for X, 'support' for Y, and 'oppose' for Z", while Q thinks, "I want my vote to have as much power as possible, and I doubt that X can win, so I'll say 'strong support' for X and Y, and 'strong oppose' for Z".
- My position is that ideally both of these votes should have equal impact on the result. This isn't about judging whether P and Q is morally or intellectually superior. It's just about giving them as equal a franchise as possible in as many cases as possible. I realize that Gibbard's theorem means it's impossible for their votes to have equal power in all cases. But in MJ, their votes will have equal power in most cases; I'd argue that, in the kind of election this will be used for on wikipedia, almost always. In Score, their votes will never have equal pull.
- This is the third time I've tried to explain this point, which I believe is central to our disagreement here. (I've also pointed out each time that that ϢereSpielChequers above apparently agrees with me that this feature is desirable.) Your failure to understand me the first two times may be partly my fault. However, statements like "You evidently are unwilling to answer my questions about what you mean." fail to assume good faith on my part, and are not conducive to productive discussion. Homunq (࿓) 23:57, 5 October 2012 (UTC)
Ok, fair enough. Utility-proportional rating is _easier_. I won't deny that, and I won't deny that it's better to be able optimize your outcome by something as easy as utility-proportional rating, instead of dealing with strategy. No argument there.
So the idea that is that that easy rating will fully count for you, by expressing, and having counted, as you intend, your pairwise preferences. I'd like that. It would be nice if utility-proportional rating were optimal strategy. But is it true?
No. Though it would be nice if it were so, utility-proportional rating is not optimal strategy in MJ. Extreme rating, exactly as in Score, is optimal strategy in MJ, when there's no chicken dilemma. But at least then you can rate at extremes for optimality, even in MJ. But when there's a chicken dilemma, things get much worse in MJ. More about that later in this post.
Maybe you'll say it's "often true". That doesn't tell us much. "Usually true" or "Most often true"? If that means more than 50% of the time, then you couldn't say that in a journal, unless you can prove it. So what are you giving us, a "maybe usually"?
Thanks for the example. Let's look at it.
Say there's no chicken dilemma:
Say the method is 0-4 MJ. (5 rating levels, as in your example, but with the more convenient numerical names).
Suppose that, before your vote is counted last, Y and Z are tied at a median of 3.5. You're voter P, in your example. You've given the following ratings, in 0-4 MJ:
X: 4 Y: 3 Z: 1
Your ballot pulls Y and Z down by an exactly equal amount. Your preference for Y over Z isn't counted at all.
If you'd voted in the manner of voter Q, then you'd have helped Y against Z, and given Y a higher median than Z.
Was pulling Y's score down, treating Y the same as Z, really what you wanted?
You boast that an MJ ballot give full pairwise votes, without extreme rating. Yes, but in an unknown direction.
I've mentioned that before. It would be nice to know which direction you're moving a candidate's median.
That's the difference between Score and MJ:
With Score, your inbetween ratings have a measured effect, directly and straightforwardly chosen by you.
I've sometimes likened Approval to a simple, reliable, solid handtool. Same for Score (almost as much).
You advertise MJ as a labor-saving, automatic, strategy-carefree, automatic machine. But, like many cheap machines, MJ can, and often will, malfunction, acting opposite to your intent, disregarding your intent. As in your example.
Maybe you're selling MJ as a poor-man's cheap discount pairwise-count method. Note what I said above, about cheap machines.
Wikipedia is better off with a good, reliable, simple, solid handtool, than with a cheap machine that's going to often go haywire.
When there's a chicken dilemma:
As I said, that's where Score's inbetween ratings are useful. As I said above, those inbetween ratings have a measured effect, directly and straightforwardly chosen by you. That's useful for the stratetic fractional ratings (SFR) that can deter defection, in the chicken dilemma.
As I said, I've shown you what MJ's SFR looks like. Because Score's simple SFR won't work in MJ, then MJ's SFR is elaborate and complicarted, and is a probabilistic strategy, requiring some chance device for deciding one's vote.
That means that the chicken dilemma is less well dealt-with in MJ than in Score. That means that the chicken dilemma is worse in MJ than in Score.
Michael Ossipoff — Preceding unsigned comment added by 98.77.167.211 (talk) 18:06, 6 October 2012 (UTC)
What you're saying sounds like this:
"Buy this used car from me. Its advantage is that its steering direction doesn't depend so much on the steering-wheel position. So, with this car, you don't have to worry about where you hold the steering-wheel! Here, take it for a test-drive along the cliffs!"
No thanks.
Embarrassment criteria:
I wasn't going to mention these, but they're relevant, because, MJ's contradictory results show that there's something very questionable about MJ's justice and fairness.
Participation:
You, and someone who votes he same as you do, are the last to vote. Before you two vote, your favorite (favorite of you both) is winning. You vote a ballot that has your favorite alone at top, voted over everyone else. The result? You've thereby made your favorite lose. Ridiculous? Of course, but MJ will do that.
Consistency:
For an MJ election, we do 3 counts of the ballots: We of course do an overall count, the one that decides the winner. But we also divide the electorate into 2 parts, and we report the results when counting each of those two subsets of the electorate.
When we count the ballots of those two subsets, we find that, in each subset's separate count, by MJ, the same option wins. But, when we do the overall count, of all the ballots, a different option wins.
The supporters of the option that won in both subsets could say, "Our option one in each of the subsets of the electorate. There's something wrong with a count rule that chooses something else when counting the overall electorate's ballots."
Yes, there is. It's ridiculous, isn't it. But MJ will do that.
With both of those embarrassment criterion failures, if MJ does choosing justly or fairly in one instance, then it must be choosing unjustly or unfairly in the other instance.
Those embarrassment criteria say a lot about how much sense a voting system's results make. And what they say about MJ isn't good.
Ok, yes, someone could point out that I like ICT, even though it, being a Condorcet version, fails Participation and Consistency. Yes, but at least ICT gives us something for that: It automatically gets rid of the chicken dilemma, and it meets the Condorcet Criterion.
What does MJ give us in return? Unresponsiveness and unreliability? A long definition, and an elaborate set of necessary tiebreaking bylaws?
No thanks.
And could it be that MJ's unreliability and unresponsiveness could have something to do with the inconsistent and nonsensical results described above, in MJ's failures of Participation and Consistency?
Michael Ossipoff \ — Preceding unsigned comment added by 69.247.226.68 (talk) 19:10, 6 October 2012 (UTC)
- You correctly point out that in my example, if Y and Z have medians of 3.5, then P's vote will be unstrategic. But in a situation divisive enough to come to an explicit vote on Wikipedia, it is extremely unlikely that two options will both have a median that high without being very similar. Yet P obviously doesn't see Y and Z as similar.
- Furthermore, your example is completely implausible because WIKIPEDIA DOESN'T HAVE SECRET BALLOTS.
- Similarly, while you are correct to say MJ can theoretically fail various "embarassment criteria", it is unlikely that it would do so in real life. If distributions of grades for each candidate are well-behaved (for instance, having similarly-sized single peaks is sufficient, but not necessary, to be "well-behaved" for this purpose), then the system will NOT fail either of the criteria you mention. Homunq (࿓) 03:26, 8 October 2012 (UTC)
- ps. If you must continue this conversation, please try to be more concise.
Majority-Judgement, continued
[edit]Author said:
WIKIPEDIA DOESN'T HAVE SECRET BALLOTS.
[endquote]
If the ongoing election-results-so-far are kept published, up to date, then of course you can know where the median is when you vote your rating. But that gives great advantage to late voters, over early voters. Your early vote might turn out to holding back a compromise that you need, pulling its median down.
Or maybe you can change your vote as often as you want to. Fine, it's been shown that, under those ongoing repeated voting conditions, any voting system will find the Condorcet winner, the candidate who would be elected by well informed strategy (informed by previous voting results so far).
Author said:
Similarly, while you are correct to say MJ can theoretically fail various "embarassment criteria", it is unlikely that it would do so in real life. If distributions of grades for each candidate are well-behaved (for instance, having similarly-sized single peaks is sufficient, but not necessary, to be "well-behaved" for this purpose), then the system will NOT fail either of the criteria you mention.
[endquote]
I'm not saying that it will happen always or often. But the fact that it can happen tells us that there is something wrong with MJ. It tells us that MJ can contradict itself. Electorate-half A chooses Alternative X. Electorate-half B chooses Alternative X. The entire electorate, consisting of Electorate halves A and B chooses Alternative Y. ??! You vote Favorite over everyone else, and thereby make favorite lose? Something is wrong with a method that can do that. It isn't a question of how often it can happen. It's a question of wheteher MJ would do that (It would), and, if so, what that says about Majority-Judgment's judgment.
One more thing: No wonder that person objected to "Strong Support": That emotionally-evaluatve naming of the ratings is the problem.
Those names give an implied instruction for voters to indicate their emotional impressions of the alternatives. They imply that that is a rule, what the election is about.
Which is it? An emotionally evaluative impressions poll, or a vote to make a choice among alternatives? You've got to be clear with voters, regarding which it is.
My suggestion? Well, numbers would be fine--preferably fracions of full support. But I'd suggest that the highest rating be called "Support or Approve". The lowest rating could be called "Non-support or Disapprove". The inbetween continuum could be labeled "Fractional support for when uncertain, or to discourage defection", or something like that. The inbetween ratings should be numbered in decimal fractions of full support.
Easy to use in Score...
...but I wouldn't envy you the job of explaining how strategic fractional ratings (SFR) would work in MJ.
And, as I said, utility-proportional, emotion-reporting ratings are not optimal in MJ. Extreme ratings are optimal in MJ. There is no "right way to vote", other than optimal voting. Voting to get the best result or the best statistical expectation. The emotion-reporting names you suggest for the ratings instruct people to vote sub-optimally.
Michael Ossipoff
— Preceding unsigned comment added by 74.233.213.227 (talk) 20:02, 8 October 2012 (UTC)
One-paragraph CMJ explanation
[edit]- Here's a paragraph simply explaining CMJ that I wrote elsewhere:
- Another system called Continuous Majority Judgment would be the next step. Voters grade each candidate A-F and each grade is tallied for each candidate. Then you start out by counting just the A grades. If no candidate gets over 50%, you add in the B grades (as if it were approval voting, and all voters lowered their standards a bit). If there's still no majority, add in the C grades, and so on. If one candidate reaches a majority before any others — in other words, if their median is clearly higher — then they win. If two or more candidates pass 50% together — because they have the same median — then you effectively take an average of the voters nearest the median by adding the following simple adjustment to their median: for each candidate, start with the number of voters who voted them above their median, subtract the number of voters who voted them below their median, and divide by twice the number of voters who voted them at their median. This gives each candidate a "remainder" between -½ and +½ which, when added to their median (the grade at which they pass 50%), will show a clear winner.
- And here's why it's worth it:
- This system is a bit more complicated than Score or Approval to explain. But it's actually simpler for voters than either. Unlike in Approval voting, you can give middIe ratings to middling candidates, without agonizing over whether to support them as strongly as your favorite or oppose them as strongly as your worst enemy. Compared to Score, if everyone votes honestly, the two systems will almost always give the same results. But with Continuous Majority Judgment, there's less of a need to watch the polls and calculate your voting strategy, because in most cases exaggeration won't affect the results. That's because, unlike Score, if all the X voters exaggerate (give X an A and Y an F), while all the Y voters vote moderately (give Y a B and X a D), the majority still wins.
Homunq (࿓) 13:43, 11 October 2012 (UTC)
Author said;
Here's a paragraph simply explaining CMJ that I wrote elsewhere: Another system called Continuous Majority Judgment would be the next step. Voters grade each candidate A-F
[endquote]
Naming the ratings as "A-F", the evaluative grade system, amounts to an instruction to voters to give emotionally-evaluative ratings, estimated utility-proportional ratings. As I said before, that is an instruction to voters, to vote sub-optimally.
Yes, I already said that. Author is repeating his statements without answering comments about them.
Then follows a a description of an elaborate procecure called CMJ.
Anyone can describe a voting system with elaborate rules. But it's pointless to do so unless you also show why that method has the advantages that you claim for it. Just specifying the elaborate count rules won't do.
If there's some reason why the CMJ count rule achieves the advantages claimed below, then Author might want to say how he justifies such a claim.
[Elaborate CMJ count rule not copied in this reply]
Author says:
And here's why it's worth it: This system is a bit more complicated than Score or Approval to explain.
[endquote]
Maybe just a bit :-)
Author continues:
But it's actually simpler for voters than either. Unlike in Approval voting, you can give middIe ratings to middling candidates, without agonizing over whether to support them as strongly as your favorite or oppose them as strongly as your worst enemy.
[endquote]
We've been over this. I've already answered this claim. Author should reply to my answers before he repeats his claims.
Yes, as I've already agreed, in MJ you give to a candidate a max-magnitude a boost when rating him/her inbetween, just as when rating him/her at an extreme. But, though that boost is max-magnitude, you don't know what its _direction_ will be. Author thinks that that is somehow desirable. I've asked him what's desirable about that, but there hasn't been an answer.
Actually, strictly speaking, not only is the direction of the boost unknown, but its magnitude is unknown too. It depends on the voter-density in the vicinity of the candidate's median. That's another way in which MJ differs from Approval in regards to uncertain, unknown, results of a rating.
I have no objection to all ratings giving (approximately--See above) a max-magnitude boost, as in Approval. But if someone wants to say that it's better when the direction of the boost is unknown, then they need to justify that odd claim.
Author says:
Compared to Score, if everyone votes honestly
[endquote]
Author doesn't say what he means by "honestly", but let's guess. I suggest that he means emotionally-evaluative/expressive rating. Rating that attempts to be proportional to estimated utilities of the candidates. Author hasn't answered my question about why he thinks that's the honest way to vote.
Certainly emotionally-evaluative/expressive voting is more honest if the voter has promised to vote in that way, or if the balloting rules require it. And Author's suggested names for the ratings--such as the evaluative grades A-F; or Strongly Approve, Approve, Indifferent, Disapprove, Strongly Disapprove--seem to be an attempt to give emotionally-evaluative/expressive rating the status of a rules requirement.
As I said before, it's necessary to be clear with voters about whether they're voting in an emotional expression poll, intended to in some way react to emotional ratings, utility-proportional ratings; or whether they're voting in an election whose purpose is to choose an option or alternative.
The only thing that makes emotionally-evaluative ratings more honest would be if the ratings were named in such as way that participation in the voting implied a promise to vote emotionally-evaluatively.
As I said, when the purpose of a vote is to make a social choice, then the only right way to vote is the way that is best in the voter's interest, the way that best tends to optimize the voter's outcome.
Author says;
, the two systems [Score and MJ] will almost always give the same results.
[endquote]
That quantitative claim needs a proof, or at least some sort of a demonstration of validity.
The only way your ratings will help Y against Z, is if you're rating Y above his previous median, and you're rating Z below his previous median. If you're rating both of them below their previous median, then you're pulling them both down equally (as nearly as can be guessed, without knowing the voter-density in the vicinities of their medians).
Author didn't think that (in his example) both Y and Z would have high medians (3.5). But aren't contenders for the win likely to have relatively high medians? And if the two candidates are close enough for your vote to change the win, is it surprising that both of their medians are high?
But they needn't both be equally high. Maybe Y leads Z, by an amount that is a little less than twice the amount by which the Z faction can pull a candidate's median. You, by rating both of them below their medians, pull them both down equally (as nearly as can be predicted). The Z-preferrers pull Z up, by top-rating him, and pull Y down to a level below that of Z.
If Author wants to claim that such an event is unlikely, then he needs to support that claim.
What Author is that, when inbetween-rating in MJ, maybe your ballot will help a higher rated candidate against a lower-rated one. ...or maybe not.
Thanks, but no thanks.
The good news about MJ is that you can use it as Approval. You can extreme-rate candidates, and thereby know which way you're pulling their final score. (Though you still won't know exactly how much, because that depends on the voter-density in the vicinity of their medians). But, when used as Approval, MJ can be almost as good as Approval (but not quite as responsive or reliable, due to the unknown voter-densities in the candidates' median's vicinities on the ratings scale).
In Score, you can give a purposely chosen and measured amount of help to a candidate, in the form of a fractional-ratng. Reasons for inbetween ratings in Score are: 1) Maybe you just aren't sure whether or not the candidate is one that qualifies for an approval, by Approval's strategy; or 2) Maybe there's a chicken dillema, and you're giving strategic fractilnal rating (SFR), as a defection-deterrent.
MJ doesn't allow you to give that pukposely chosen and measured fractional rating. MJ only allows max-boost ratings. But you have the choice of inbetween rating, to make the direction of that boost unknown :-)
Author says:
But with Continuous Majority Judgment, there's less of a need to watch the polls and calculate your voting strategy, because in most cases exaggeration won't affect the results.
[endquote]
Translation: You aren't allowed to give fractional support.
..."affect the results"? In MJ, your inbetween ratings of two candidate often won't affect the results at all, when you're pulling them both equally in the same direction, as can and will happen in MJ.
Author says:
That's because, unlike Score, if all the X voters exaggerate (give X an A and Y an F)...
[endquote]
I asked Author exactly what it is that he thinks is being exaggerated, and shouldn't be exaggerated, by giving an extreme rating. No answer has been forthcoming from Author.
Author says:
, while all the Y voters vote moderately (give Y a B and X a D), the majority still wins.
[endquote]
Unless their previous medians are above B, in which case the Y voters are pulling both candidates down equally (as nearly as can be guessed).
What Author is saying is that maybe (He claims, without support, "more often than not") your inbetween rating will count just as would an extreme rating. Then what is the purpose of the inbetween rating? What he means is that you aren't allowed to give fractional support. He wants to call that an advantage.
Michael Ossipoff — Preceding unsigned comment added by 65.8.53.69 (talk) 16:27, 12 October 2012 (UTC)
I wasn't talking to you, which is why I made this a new section.