[EM] MMPO objections (hopefully better posted)

[C.Benham]

Mike,

The  MinMax Pairwise Opposition (MMPO)  "bad example" we are talking about:

x: A
1: A=C
1: B=C
x: B

x  = any number greater than 1.   MMPO elects C.


On 9/19/2016 4:15 AM, Michael Ossipoff wrote:
>
> Why should 2 voters have the power to elect someone bottom-rated by 
> nearly everyone?
>
> How about because everyone is bottom-rated by at least half of the voters.
>
> ...& because it isn't a positional method.
>
> Those 2 voters didn't do it on their own. They had a lot of help from 
> everyone else.
>
> ...because the A voters & the B voters prefer C to each other's 
> candidate.
>

C: There's no evidence on the ballots for that assertion.

> Surely the importance of a bad-example depends on its plausibility.

C: Not when it's that bad.  And not even when it's merely very bad in 
such a simple example.  It is more understandable
and perhaps forgiveable for an algorithm to become "confused" in a 
complicated example (with say, lots of candidates
and cycles within cycles).

> Would you give up the best combination of the best strategy properties 
> because of a funny, but not outrageous result, one that doesn't wrong 
> anyone, in a thoroughly implausible example?

C: I don't agree with most of the premises in that question. Other 
methods meet FBC and CD. What's so good about Later-no-Harm with a 
random-fill incentive?

The result is completely outrageous and absurd.

The correct result is an A=B tie.  All but 2 of the voters were wronged, 
because their favourites should have a 50%  probability of winning.

Chris Benham




On 9/19/2016 4:15 AM, Michael Ossipoff wrote:
>
> A few more comments:
>
> Why should 2 voters have the power to elect someone bottom-rated by 
> nearly everyone?
>
> How about because everyone is bottom-rated by at least half of the voters.
>
> ...& because it isn't a positional method.
>
> Those 2 voters didn't do it on their own. They had a lot of help from 
> everyone else.
>
> ...because the A voters & the B voters prefer C to eachother's candidate.
>
> Given that, C's win isn't so surprising or outrageous.
>
> Anyway, the example has no plausibility, at all.
>
> Surely the importance of a bad-example depends on its plausibility.
>
> Yes, MMPO doesn't strictly always elect the CW, and I don't like that. 
> It's a distinct disadvantage. We expect better from a pairwise-count 
> method.
>
> But with sincere voting, & with no indifference, the CWs (sincere CW) 
> always wins.
>
> For the CW to lose, it's necessary for one of hir pairwise comparisons 
> to have high turnout, & be relatively nearly tied.   ...& for someone 
> else's pairwise comparisons to all be very low turnout & hir defeats 
> nearly tied.
>
> Would you give up the best combination of the best strategy properties 
> because of a funny, but not outrageous result, one that doesn't wrong 
> anyone, in a thoroughly implausible example?
>
> Michael Ossipoff
>
> On Sep 17, 2016 1:51 PM, "Michael Ossipoff" <email9648742 at gmail.com 
> <mailto:email9648742 at gmail.com>> wrote:
>
>     ---------- Forwarded message ----------
>     From: "Michael Ossipoff" <email9648742 at gmail.com
>     <mailto:email9648742 at gmail.com>>
>     Date: Sep 17, 2016 12:52 PM
>     Subject: MMPO objections
>     To: <t at gmail.com <mailto:t at gmail.com>>
>     Cc:
>
>     Though NEO, so far, to me at least, seems to show promise, it
>     hasn't been thoroughly checked out enough to be a proposal.
>
>     But it's different with MMPO. We've heard people's best arguments
>     against MMPO, & it can be said to have  already been well-discussed.
>
>     No rank method's result will always look right. All will sometimes
>     do something ridiculous.
>
>     A method optimized for 1 purpose or standard can't do well by
>     other standards.
>
>     MMPO achieves what it achieves by looking only at pairwise
>     unpreferredness.
>
>     It isn't a positional method, & so you can find an example in
>     which it does terribly, positionally.
>
>     In Kevin's bad-example, it chooses someone twice as bottom-voted
>     as the other candidates, & nearly not top-voted at all.
>
>     It certainly isn't a positional method
>
>     MMPO isn't a pairwise-defeats method. So you can find an example
>     where it does terribly by pairwise defeats.
>
>     In Kevin's example, it elects the Condorcet loser, who pairwise
>     loses to the others by 1000 to 1, if X = 1000.
>
>     It certainly isn't a pairwise defeats method.
>
>     We've been looking at pairwise defeats methods for so long that we
>     tend, maybe subconsciously, to evaluate by pairwise defeats standards.
>
>     A "beats-diagram" shows
>     an "=" sign between A & B. They have no defeat, but C has one.
>
>     But look under that "=" sign. Half the voters bottom-vote A, & the
>     other half bottom-end vote B.
>
>     Say two groups both despise eachother. Does that mutual despising
>     cancel out, making both groups un-despised?
>
>     But that's the fallacy that the beats-diagram & its "=" sign
>     allows you to believe.
>
>     If the A voters voted among themselves, between B & C, they'd
>     choose C.
>
>     If the B voters voted among themselves, between A & C, they'd
>     choose C.
>
>     C is the compromise preferred by the A voters, & by the C voters,
>     to eachother's candidates.
>
>     Yes, it's natural to reject a low-favoriteness compromise. Rob
>     Richie would be proud.
>
>     Of course this bad-example makes that compromise as little
>     top-voted as possible.
>
>     I've told, here, why the bad-example isn't as bad as you think.
>
>     It doesn't look good by standards other than the one by which it
>     achieves the elusive goal of MAM-like strategy, without
>     chicken-dilemma.
>
>     Distinguish between a harmless election of a low favoriteness
>     compromise, a compromise outcome that looks bad to an outside
>     observer vs an actual practical problem, one that will routinely
>     make strategy problems for voters, and give tangibly (not just
>     aesthetically) bad results.
>
>     When proposing better voting to a community of jurisdiction, of
>     whatever size, offer them a list of methods, telling the
>     objections to each, & their answers.   ...& telling the advantages
>     of each.
>
>     It would be irresponsible to leave out one with an impressive,
>     unique, powerful combination of strategy advantages.
>
>     Let the community, jts voters &/or the initiative proposal
>     committee choose for themselves. It isn't necessary to make
>     decisions for them.
>
>     Michael Ossipoff
>
>
>
> ----
> Election-Methods mailing list - see http://electorama.com/em for list info
>
>
> No virus found in this message.
> Checked by AVG - www.avg.com <http://www.avg.com>
> Version: 2016.0.7797 / Virus Database: 4656/13038 - Release Date: 09/18/16
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20160921/2a2fb490/attachment.htm>