[EM] Re: ATLO, defection, method/criterion model

[Kevin Venzke]

Mike,

 --- MIKE OSSIPOFF <nkklrp at hotmail.com> a écrit : 
> I've just written an almost complete reply to this posting, only to have it 
> get deleted, presumably because of some key (not the delete key) or key 
> combination that I pressed accidentally near the bottom of the keyboard. So 
> I'll start over. I guess it's a good idea to copy what one is writing every 
> 5 minutes, or after every new paragraph, so avoid losing the entire e-mail.

I'm sorry that happened.

> You said:
> 
> Where do you get "the better"? What I'm saying is: Suppose B is a lesser 
> evil,
> and hopes to persuade C voters to vote in such a way that the election won't
> be thrown to A when B voters truncate. In IRV this means the C voters can't
> vote for C. In WV+ATLO this means that the C voters can't use ATLO. I think
> the C voters will find the latter proposal easier to swallow.
> 
> I reply:
> 
> Oh, ok, I misunderstood you, because I thought that you were saying that it 
> was somehow an advantage when more drastic defensive strategy is needed. Of 
> course you didn't mean that.

But, I still wonder what your opinion is here. With WV+ATLO the C>B faction
is given the ability to threaten that if the (sincere) B>C faction doesn't vote
for C, the election will be thrown to A. But at the same time it seems to me
that the B>C faction can threaten that if the C>B faction uses ATLO, the election
will be thrown to A. Is there a clear winner?

I thought about this also in connection with defensive truncation in WV. I
think the main reason the defenders have the advantage in that scenario, is
that it's natural and expectable to not rank your candidate's primary opposition,
while it doesn't seem as natural to expressly rank third candidates over your
candidate's primary opposition.

> It's a rule that treats rank methods as special and favors them. As I said, 
> it amounts to saying "Nonrank methods fail this criterion".

If RI "favors" methods, it would be all methods that do not use approval 
cutoffs, limited slots, color voting, automatic compression options, or ratings.
It doesn't make e.g. Plurality seem any worse than it is.

It doesn't amount to saying "nonrank methods fail this criterion." That will
vary based on the criterion and based on the physical ballot and just how much
of a ranking it will permit.

As an example, suppose a given IRV variant only allows three choices to be
listed. Using RI, this variant then fails Majority (for solid coalitions),
but it still satisfies LNHarm and LNHelp.

As another example, using RI, a method can satisfy votes-only Minimal Defense
without allowing equal-ranking for non-last options. Using Steve Eppley's
"any preference order must be admissible" rule, this wouldn't be good enough.

> You'd said:
> 
> >This is why I don't feel compelled to include Steve Eppley's requirement in
> >votes-only Minimal Defense that all preference orders must be accepted
> >
> I replied:
> >
> >No, you just say the same thing in a different wording. I'll get to that 
> >soon, below.
> 
> You say:
> 
> And that's basically what I said, except that you cut that part off of your
> quote for some reason.
> 
> I reply:
> 
> Ok, then it's agreed that the rankings interpretation is just another 
> wording of "Nonrank methods fail this criterion."

Only if by "this criterion" you mean "votes-only Minimal Defense," since that
was the subject here.

> You say:
> 
> [referring to my statement that RI is just a rewording of "Nonrank methods 
> don't pass"]
> 
> That isn't true. If it were true, then it wouldn't be possible to write many 
> RI
> criteria that Plurality passes.
> 
> I reply:
> 
> How many have you written? Post their definitions.

Any of Woodall's criteria work properly as RI criteria. Actually, any criterion
written by someone who only considers ranked methods should work fine under RI.

> You continue:
> 
> Yet Plurality satisfies the Plurality criterion,
> Clone-Loser, Later-no-harm, Later-no-help, Participation, Mono-raise...
> 
> I reply:
> 
> Have you written RI versions of those criteria? The one of those that I've 
> heard of is Participation. How would you write RI Participation?

I just use Woodall's definitions for these. Participation:

"If further ballots are added that are all solidly committed to the same set
X of candidates, then the probability that the elected candidate is in X
should not be reduced."

> No, the fact is that there's no need for RI in criteria that Plurality is 
> expected to reasonably pass. RI is only needed to keep Plurality from 
> passing criteria that you don't want Plurailty to pass. RI is nothing but a 
> rewording of "Nonrank methods don't pass".

You are more obsessed with Plurality than I am. I actually do consider
Plurality to be a ranked method, just one that does not allow equal ranking
at the top.

The bigger problem is Approval, since there are several possible interpretations
of Approval as a ranked ballot method.

You also have the problems created by methods that don't allow equal ranking,
don't allow truncation, or don't allow a full ranking.

I'm not at all sure that the term "nonrank method" is well-defined.

> For instance, Participation is passed by Plurality, and it also makes no 
> mention of preference. Participation says:
> 
> Adding to the count one or more ballots that vote X over Y should never 
> change the winner from X to Y.
> 
> [end of Participation definition]

Does Random Ballot satisfy this criterion?

> You say:
> 
> I agree that either with Steve's complete rule, or using RI, Approval and
> Plurality both fail votes-only Minimal Defense.
> 
> I reply:
> 
> My point wasn't that Pluralitly fails it. My point was _why_ Plurality fails 
> it.

And, of course, it fails it ultimately because it doesn't collect the right
information. But even methods that use ranked ballots can suffer from this problem.

> I'd said:
> 
> >What would be your URF votes-only version of WDSC?
> 
> You reply:
> 
> Well, I'm sure you won't let me define your criteria for you
> 
> I reply:
> 
> Nice try. If you wrote an RI votes-only criterion that gives the same 
> answers as WDSC, that wouldn't be my criterion. Can you or can you not 
> write one?
>
> You continue:
> 
> , and also I
> don't use any version of WDSC, so I don't know.
> 
> I reply:
> 
> You wouldn't like methods that fail WDSC. Anyway, I didn't ask you to write 
> it because you use WDSC. I asked you to write it to show how (or if) your 
> criteria system can deal with the task of matching WDSC's answers.
> 
> WDSC is a good test for your criteria system, and for your claim that RI is 
> more than "Nonranked methods don't pass". That's because Plurality fails 
> WDSC, and Approval passes WDSC. That means that "Nonrank methods don't pass" 
> won't do here. And that suggests that RI won't do either.
> 
> And that shows a deficiency, a lack of applicability, of your RI system of 
> criteria.
> 
> So can you or can you not write an RI votes-only criterion that matches 
> WDSC's answers?

You ask me if I can write an RI criterion that "matches WDSC's answers." I
assume you mean *all* of WDSC's answers. The answer is no, but I don't consider
that important, since I don't think it's worth the trouble to ensure that Morse 
code voting gets a special distinction.

If you mean: Can I write an RI WDSC criterion that is passed by almost
every method that allows equal-ranking, and failed by Plurality and IRV,
then I believe the answer is yes. It would go like this:

"If more than half of the voters rank X above Y, X below no one, and Y above
no one, then Y must not be elected."

If RI has a problem here, it is that Approval is a limited slot method, and
so if you interpret that a ballot could disapprove or approve all candidates
despite the intended ranking, Approval doesn't satisfy this.

> I'd said:
> 
> >Notice that it gives special status and mention to a particular balloting 
> >mode.
> 
> That's because it can encompass most other modes.
> 
> I reply:
> 
> It doesn't encompass CR.

I agree. But, note that that doesn't mean RI criteria can't be applied to CR
(the method).

> You continue:
> 
> It would be straight-forward
> to adapt RI to also accept cardinal ratings.
> 
> I reply:
> 
> Is that why you don't do so? Even though your RI system is only half defined 
> if it doesn't apply to CR?

Is *what* why I don't do so? I don't know what you mean when you say that the
RI system "doesn't apply" to CR. The reason I don't adapt RI to use cardinal 
ratings is because there are not many criteria written that involve ratings.

> You continue:
> 
> Then "RRI" criteria could be written
> which would then be applicable to all methods.
> 
> I reply:
> 
> It could, but you don't write it.

"Criteria" is plural, so I guess by "it" you mean "RRI." I have no desire to
write criteria that involve ratings.

> Kevin, your RI system isn't defined. All you've given to us are some vague 
> sketches. I have nothing against poetry, but you've got to state definitions 
> that are precise and unambiguous, and as concise as possible. You could use 
> my defintiions as a model. You've given us nothing approaching the 
> completeness and precision of my criterion definitions, and their supporting 
> definitions. Until you do, your RI criteia aren't defined.
> 
> That means that when you speak of Minimal Defense, you don't know what you 
> mean. That's right, when you don't write precise unambiguous definitions for 
> it, you don't know if you know what you mean.

Mike, I have no idea why you say that the RI system "isn't defined."

This part of the conversation started when you said:

> >Notice that it gives special status and mention to a particular balloting 
> >mode.

This only makes a difference if you want to use criteria that do not make
sense in the given balloting mode.

Kevin Venzke



	

	
		
__________________________________________________________________
Découvrez le nouveau Yahoo! Mail : 250 Mo d'espace de stockage pour vos mails ! 
Créez votre Yahoo! Mail sur http://fr.mail.yahoo.com/