[EM] Re: IMHO, IRV superior to Approval

[Kevin Venzke]

Chris,

 --- Chris Benham <chrisbenham at bigpond.com> a écrit : > Kevin,
> Where did you get that idea?  It happens all the time in Australia. The 
> Federal  government here is based on two conservative
> parties in coalition, which are permanent allies. In some seats they 
> each field a candidate, and in those seats they tightly "exchange
> preferences".

Ok, I stand corrected, although clones who agree beforehand to tightly exchange
preferences don't strike me as likely to be distinct alternatives.

> >This is a tired scenario, but consider this while thinking of "half-clones":
> >40 A
> >35 C>B
> >25 B
> >
> >I call {BC} a half-clone set.  In Approval the C>B voters can vote CB and still get
> >B to win, and this doesn't involve any insincerity.  In IRV those voters should put
> >C below B in order to avoid electing A.  Not only does IRV encourage insincere
> >voting here, but the fact that it does so will be a strong incentive for C not to
> >enter the race at all.
> >
> >So the biggest reason I prefer Approval to IRV is IRV's nomination disincentives.
> >
> I find it a bit implausible that the C voters all give their second 
> -preferences to B, with no recipricocity  at all. 
> If  A, B, and C
> were the endorsed candidates of rival establised political parties  for 
> a seat in Parliament in  Australia ,

I do not care if C is an endorsed candidate of an established political party;
I want to suppose that C is an independent candidate who does not have a vote-trading
arrangement with a major party.  My argument is that in this case, even people
who like C best, under IRV, can only support C by endangering the "more realistic"
choice, B.  And this is a reason why C will be reluctant to run in the first place.

If every serious candidate is going to belong to one of two partisan camps, with
vote-swapping agreements with others in their camp, then I don't care what election 
method is used.  Primaries+Plurality is adequate.

> >> I  like methods that have no zero- information strategy, 
> >> and  doesn't 
> >> reward  indecisive voters by giving them extra clout.  If  voter  A
> >> ranks candidate X last,  and  because of  prejudice, ignorance and 
> >> stupidity has no ranking of the other candidates; and voter B ranks
> >> candidate Y last but also strictly ranks all the other candidates; then 
> >> both candidates X and Y should have the same probabilty of being
> >> elected.
> >
> >I still think "probability of being elected" is undefined here.  Usually that
> >term is used when there is a tie...
> >
> By "probabilty of being elected", I meant exactly that. To make it 
> clearer, instead of referring to two individual voters, I could
> instead have said that there are two equal-sized factions of voters.

Do you have an example, for any election method, where these assumptions lead to 
X and Y having unequal probability of being elected?

> >> "No Zero-Information Strategy"  also implies that Later-no-harm and 
> >>  Later-no-help should be in balance. That is, that the chances of
> >> harming or helping an already ranked candidate by ranking another below, 
> >> should be the same.
> >
> >1. Can you tell me a method besides Approval which fails "no zero-info strategy"?
> >Because I take it you really mean that zero-info strategy should be equivalent to
> >a sincere vote.
> >2. Can you tell me a method which fails "no zero-info strategy" where LNHarm and
> >LNHelp are not in balance?
> >
> Yes and Yes. A method that simply asks the voters to rank the 
> candidates, but  gives voters who do anything else (based on their
> ratings) a better expectation, fails "No Zero-Info Strategy".  I gather 
> from posts in the archives, that Condorcet (Winning Votes) fails
> because if voters have a sufficiently large gap in their ratings, say 
> between second and third place, they do better to insincerely rank
> the candidates above the gap in equal-first place. Also, it is true, or 
> has been strongly conjectured, that voters who do not have a sincere
> full ranking do better to strictly rank all the candidates below 
> first-place .
> Bucklin, Weighted  Median Approval, the Woodall methods Quota-Limited 
> Trickle-Down  and Descending Acquiescing Coalitions all
> fail Later-no-harm , but meet Later-no-help.  That means that truncation 
> is a big implicit  approval-cutoff, and if voters have a big gap
> in their ratings they would do better to insincerely not rank the 
> candidates below it.
> On the other hand, the Woodall  set-intersection method Descending Solid 
> Coalitions meets Later-no-harm but fails Later-no-help.
> That means that the voter definitely does better to "random-fill", ie if 
> the voter does not have a full ranking, s/he should strictly rank
> all the bottom candidates at random.

I see.  This turns out to be a rather difficult standard, I think.  Only
IRV, FPP, and Borda (without 0-pt truncation) strike me as meeting this.
But if you forbid equal ranking, I believe Condorcet and DAC/DSC meet the
standard.  Bucklin too, if no truncation.

I'm not sure how to apply these standards to Approval.  One would have to define
what the approval ballot assumes the voters are doing (just as with a ranked ballot
we assume that voters are giving a preference ordering), because only then could
we say what it means for voters to "do anything else" besides follow the
implicit instructions.

Suppose we assume that anyone filling out an approval ballot is voting for
the candidates they would rank sincerely in first place.  Then obviously it
is possible to "do anything else" to get a better result: They could approve
more candidates who they like almost as much.

But suppose we assume that anyone filling out an approval ballot is marking every
candidate who is better than the voter's expectation.  Then it's not possible for 
a voter to "do anything else" to get better expectation.

> >> And of course voters who succeed in ignoring this circus and instead just concentrate on the 
> >> policies and qualities of the candidates, will potentially be greatly disadvantaged (much
> more
> >> than a "naive", sincere IRV voter).
> >
> >Here I completely disagree.  A gut-sentiment Approval voter will probably not be far off
> >from his optimal way of voting.  He screws up only if he approves or disapproves both of
> >the strongest candidates.  In IRV he can screw up by not guessing who needed his traveling
> >vote when.
> >
> I don't see why the it is particularly unlikely that the 
> "zero-information"  (poll-ignoring) voter wouldn't distinguish between 
> the two
> leading candidates (not approving either or approving both).

It may not be unlikely.  I'm disagreeing with the idea that, because IRV accepts
a full ranking from the voter, IRV is more likely to give a zero-info voter a fair
shake than Approval is.

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/

Dialoguez en direct avec vos amis grâce à Yahoo! Messenger !Téléchargez Yahoo! Messenger sur http://fr.messenger.yahoo.com