[EM] Kevin reply, FBC wording

[Kevin Venzke]

Mike,

 --- MIKE OSSIPOFF <nkklrp at hotmail.com> a écrit : > 
> The person who's making an exception for Pluralty & Approval is the person 
> who claims that it's necessary to pretend that their rules are different 
> from what they are.
> 
> You can call Plurality's & Approval's nonranked rules "an exception", but it 
> isn't an exception that I make. It's just the actual rules of those methods. 
> To pretend that their rules are different from what they are, that would be 
> "making an exception" for those methods.

It is a useful approach to suppose that if ranked ballots provide the necessary
information to resolve the method, then it's a ranked method.  There is no need
for an exception.

Other than calling this approach ridiculous, do you have a reason to say that
this approach isn't useful?  Your argument seems to be that criteria should be
based on sincere preferences, because otherwise FPP passes Condorcet.

> You continued:
> 
> The methods that pose problems are those that require some input from the
> voter other than a ranking.  Consider Approval, which arguably requires
> a cutoff in addition to the ranking.
> 
> I reply:
> 
> No, if you're going to pretend that Plurality allows ranking all the 
> candidates, you can pretend the same thing about Approval. Approval simply 
> differs by allowing a voter to equal-rank. Sure, and let voters equal rank 
> at all rank positions--why not. Or rather _say_ that Approval allows that.

What are you disagreeing with above?

I do "pretend" that Approval allows ranking of all the candidates.  But I
would not call Approval a "ranked method" because ranked ballots don't
provide the information necessary to find the Approval winner.

I would call "vote-for-zero, random candidate" a ranked method, because 1)
ranked ballots are sufficient to find the method's winner, and 2) it is a useful 
approach if I want to compare this method to other methods.

> You continued:
> 
> Is it possible to say whether Approval meets Clone-Winner?
> 
> I reply:
> 
> Yes.
> 
> When ICC is defined in terms of sincere preference and stipulates sincere 
> voting, like my CC, then Approval doesn't pass ICC. But Approval passes the 
> ICC that Markus posted. I don't know if Approval would fail Markus's ICC  if 
> we used the ridiculous fiction that Approval allows the ranking of all the 
> candidates. Maybe. I haven't checked.

Of course, clone criteria can be defined in any way we find useful.  My point
is that a clone criterion designed for ranked methods, operating on cast
votes only, can't be applied to Approval in an obvious way.  But it certainly
can be applied to FPP.

> You continued:
> 
> Unless we have specific rules about how the cutoff may move in response
> to the introduction of clones, I suppose we have to assume that Approval
> fails this criterion.
> 
> I reply:
> 
> I once asked what ICC assumes about how people vote. I asked, does ICC 
> assume sincere voting, or that voting is undominated strategy, or that 
> voting maximizes utility expectation by some possible set of utilities and 
> probability estimates. Or does it just require that no ballot is changed 
> when a clone is deleted from the election? It turned out that the assumption 
> of sincere voting comes closest to the way that ICC is intended.

It's not clear to me how to use this assumption to say that Approval passes
or fails Clone Independence.

An alternative approach would be to append to criteria explanations of how
the criterion works with approval cutoffs, or CR levels.  For instance, if
a ballot in Approval reads "A>B," and A is cloned to get A, C, and D, is
it possible that this ballot could become "AC>BD," or do those clones have
to stay in the same slot as A?  Or can they move, so long as the adjacent
slot is empty?

> You continued:
> 
> For a more abstract example, suppose the method requires the voter to
> label each candidate "red," "green," or "blue."  Unless we have rules about 
> how and why labels may adjust in coordination with the rankings, we have to 
> assume that the method fails every criterion.
> 
> I reply:
> 
> Are you sure? A method meets the Majority Criterion if a candidate wins if 
> more than half of the voters vote hir over all the other candidates.

Yes.  I didn't mean to imply that this method even accepts rankings, although
I'm not sure if you thought I did.

Unless we have some helpful rules, for instance, "Voters always mark only their 
favorite as green" (c.f. "Voters never approve their least favorite"), then even 
if the voters can submit a ranking, we don't know whether the assigned colors will 
result in the election of the majority favorite.

This, like Approval, is intended to be an example of a method that I do not
consider "ranked," even if complete rankings can be submitted (perhaps in the
margin).  The reason I don't consider it "ranked" is because rankings alone are
insufficient to resolve the method.

I agree with your thoughts on the color method and criteria compliance.

> Those things are true regardless of what color people vote hir. Colors 
> aren't part of the requirement or premise of MC. The failure-example-writer 
> can configure color votes anyway s/he wants to, since they aren't mentioned 
> in MC's premise.
> 
> But MC's requirement doesn't mention colors either. Of course the method's 
> count rule might be strongly affected by color votes.

For a more realistic example, consider the Plurality criterion with Approval
cutoffs.  The Plurality criterion says that if X has more first-place votes
than Y has non-last votes, then Y must not win.  Unless we specify that an
approval cutoff will never be placed ahead of the favorite candidate, we can't
even say that Approval passes this criterion.


You asked about how FBC could be reworded to avoid reference to sincere
preferences.  I will take a stab at it, but no doubt someone will be able
to refine it:

Strong FBC: Lowering some candidate X from sole first place to some other
position (including equal first place) on a set of identical ballot(s) should
not improve the result of the election from the perspective of the original
ballot.

Weak FBC should be the same thing, except X starts in equal first place, and 
must be moved to some position lower than that.

These criteria are arguably desirable regardless of whether there is sincerity
behind X's placement at the top.

Actually, if the sincere preferences are the basis of judging whether the result
has improved or not, that Weak FBC definition won't work right.  Suppose you
sincerely feel A>B>C>D>E>F but vote A=B=C=D=E>F and E is elected.  Then suppose
E is downranked to obtain A=B=C=D>E>F, and this results in D being elected.
Judging by votes cast, this has not helped or harmed, but by sincere preferences
this has helped.


Kevin Venzke
stepjak at yahoo.fr



	

	
		
Yahoo! Mail : votre e-mail personnel et gratuit qui vous suit partout ! 
Créez votre Yahoo! Mail sur http://fr.benefits.yahoo.com/