[EM] Chicken Dilemma--To whom is it a problem? KV to MO

[Kevin Venzke]

Hi Mike,


----- Mail original -----
> De : Michael Ossipoff <email9648742 at gmail.com>
> 
> Just a few more words about a previous topic. Maybe a few things I've
> already said, but with additional things, and different emphasis:
> 
> When I first proposed SDSC, I did so because it's something that the
> best wv Condorcet methods offer, and i wanted to name that desirable
> property, so that I could say which methods have it.
> 
> Beatpath and Ranked-Pairs meet SDSC. Kevin, your 2nd example shows
> that IRV, Benham and Woodall don't meet SDSC.
> 
> It's a desirabe property, but it can't be called a _necessary_
> property. A failure of SDSC isn't an _outrage_.  A failure of the
> harder-to-fail, easier-to-meet Mutual Majority Criteion is an outrage.
> Here is a majority of the voters, all supporting eachother's
> candidates...and they lose?
> 

First of all, I don't think that MMC is easier to meet than SDSC. 
Almost every method based on the "implicit approval" concept satisfies
SDSC, while MMC requires the mechanisms of the method to take care 
with regard to clone-proofing the majority favorites.

I would agree that an MMC violation would be a greater outrage than 
an SDSC violation. But I think MMC's premise would apply exceedingly 
rarely. That is why I think MMC is not a very useful guarantee.


> As for why SDSC-failure isn't an outrage, look at candidate B's
> win-qualifications, in your 2nd example:
> 
> The B voters want B to be elected by a majority coalition that they
> don't support. They have no right to expect anything from that
> non-existent majority coalition.
> 
> B isn't CW. There is no voted CW. If the B voters are sincere in their
> truncation, there's no sincere CW either.
> 

Sure, by MMC or Condorcet logic there is no basis. It's purely about 
seeing the majority specifically. Why do they matter: Since via 
insincerity the majority can achieve any outcome it could want, when 
you violate SDSC you can be confident that a majority didn't get 
what it would have wanted, and that it could have secured it via some 
other way of voting.

> Maybe SDSC can be advocated for the rights of the A voters to support
> B to defeat C.
> 
> Sure, and that's SDSC's justification. Probably, ideally, even a
> 1-sided "coalition", like the one in your 2nd example, should have the
> power to defeat C. When I introduced SDSC, it was because I considered
> the defeat of the despised to be the important thing. That's why you
> ike SDSC too.
> 

More precisely I do not want a majority to complain that the result
is spoiled, in a situation where you can look at the ballots and see 
that the winner is the one guy with the only majority-strength loss.

Also, when the method can't see past unviable "noise" candidates 
there is pressure not to nominate or vote for them. I tend to think 
that without SDSC, the prerequisite situations for CD and MMC won't 
even happen, because it's too dangerous to those would-be coalitions.

It's definitely my beliefs about nomination disincentive that will
keep us in disagreement on this issue.

> 
> I don't know if MMC,CD, CC and SDSC are compatible. Probably not,
> because I don't know of a method that meets them all.
> 

Interesting question there. This method would certainly fail 
Plurality as you note. My immediate question would be whether 
CD is sufficiently distant from LNHarm that it could be 
compatible with Condorcet (which is not compatible with LNHarm) 
or SDSC (which I strongly suspect to be incompatible with LNHarm 
with 4+ candidates).

You may have noticed that CD is LNHarm-friendly, tending to agree
with those methods in what is required.

> I don't know of a method that meets MMC, CD and FBC either. That
> combination, too, is probably incopatible.
> 

MMPO (break ties with FPP) would be the closest I think. If 
they are incompatible, the likely reason is that the clone-proofing 
needed for MMC is too "path-tracey" for FBC. MMC+FBC is not
impossible, as in ER-Bucklin, but LNHarm is completely lost there.

Kevin Venzke