[EM] Gilmour: Approval

Fri Dec 6 23:53:40 PST 2002

Bart wrote (29 Nov)
>Or the answer could be to adopt approval voting, especially if the only
>practical alternative is IRV.

Mr. Gilmour replied:

But there are some other serious problems with Approval.

I reply:

When you're that unspecific, you're not very convincing. If you believed
that you could show a genuine problem for Approval, a problem-claim
that hasn't already been answered here, you'd have actually named
a problem.

Bart had said:

>If you acknowledge that voter rankings
>will be utilized in such a haphazard way that you would prefer to keep
>information about subsets of the vote secret, wouldn't it be better to
>avoid collecting information you can't use reliably?

Gilmour replied:

I do not acknowledge that voter rankings will be utilised "in such a 
haphazard
way"

I reply:

...unless those rankings are counted by IRV

Gilmour continued:

, unless "haphazard" means something very different your side of the pond.

I reply:

Pretty much the same, I'd expect. But many words have radically different 
meanings in IRV-land.

There is nothing haphazard about the situation I described. Publishing
"results"
precinct by precinct is just totally irrelevant when all that matters is the
city-wide totals. It is not a question of keeping them secret. Rather the
question is why on earth would you want to publish such irrelevant 
information?

I reply:

It's something that's regularly done here. People are intersted in which
nationwide or statewide candidate would have won if only their own
County's votes were counted. You see, they're curious about what kind
of people they're sharing  a county with.

Gilmour continued:

I never recommend collecting such information. Of course, in a 
non-preferential
voting system, it is possible to count the votes locally at each precinct 
and
remit only the totals to the central "counting" station. That would be more
difficult with a preferential voting system

I reply:

Not at all. It's easy with Condorcet: Each precinct county could count
the pairwise votes of the ballots, and send them to a central national
location. But I agree with you that that couldn't be done in IRV, where
each of the 100,000,000 rankings would have to be stored and repeatedly
referred to during the count. Of course even if Condorcet rankings were
sent to a central national location for all of the counting job, looking
at the rankings would be a 1-time thing. IRV requires much more memory
usage and computation-time in a big national election. Also much more
opportunity for tampering. See the "Summability Criterion" discussion
at the technical evaluation page at http://www.electionmethods.org

Bart had said:

>In return,
>approval ballots contain information not present in ranked ballots,
>namely an indication of the voters' strength of preference.

Gilmour replied:

I don't buy that. In Approval each voter just sorts the candidates into two
sets - acceptable and not acceptable.

I reply:

You've got to study Approval strategy more before you make statements
like that. Yes, acceptable/unacceptable is the optimal strategy if
you believe that the election has completely unacceptable candidates who
are winnable. Personally I believe that to always be so in our public
political elections, but many or most would disagree with me on that.

Probably the most popular strategy will be to vote for whichver of the
expected frontrunners you like the best, and for everyone whom you
like better. That's the obvious extension of typical Plurality strategy
to Approval.

That could be refined by also voting for every candidate whose utility
to you is greater than PxUx+PyUy, where Px is the probability that if
only one of the 2 expected frontunners, X & Y, is in a tie or near-tie
for 1st place, it will be X. Ux is X's utility for you.

If you don't have any information or impression about the relative 
magnitudes of Px & Py, then they're equal as far as you're concerned,
and they're each 1/2, so that PxUx+PyUy becomes the mean of Ux & Uy.

This refinement to the popular "best-frontrunner" strategy is derived
at http://www.barnsdle.demon.co.uk/vote/sing.html  in one of the
Approval Strategy articles there.

Alternatively, one could just vote for each candidate whom one consideres so 
good that one would rather have him/her in office instead
of holding the election. Both are good strategies.

Gilmour continued:

That seems to me to be LESS information
than on a typical ranked ballot.

I reply:

...but did Bart say that the Approval ballot contains more information
than a ranking, or merely that it contains information that the ranking
doesn't contain?

Gilmour continued:

If you really want information about "strength
of preference" you will have to introduce some system that allows each voter 
to
weight his or her preferences as they wish.

I reply:

Incorrect. In a good rank method, one can just rank sincerely. Freedom
from need to do otherwise is a goal of rank-methods. With Approval,
where you must give a candidate 0 or 1 point, the matter of whether or
not you give a point to a candidate depends partly on how much you like
the candidate, in absolute terms rather than ordinally.

For instance, in the extreme case of 0-information, you vote for all
the candidates whose utility for you is above the mean. As someone else
has pointed out, that means that Approval, then, is maximizing the
number of people who consider the winner above-mean.

And the
acceptable/unacceptable strategy that you mentioned also is about
non-ordinal rating. If people are using that strategy, Approval maximizes 
the number of people who consider the winner acceptable.

So in these instances the Approval ballot is based
on information that doesn't affect ballots in the better rank methods.

If you're merely voting for whichever of the 2 expected frontrunners that 
you prefer to the other, then of course your ballot doesn't use
any cardinal rating information.

Bart had said:

>In computer models conducted by Merrill and others, approval voting
>produced results more in line with Condorcet's method than did IRV,
>especially when there are many candidates.

Gilmour replied:

Maybe, but that does not remove the serious defect in Approval. One person, 
one
vote is violated.

I reply:

Are you aware that that objection was answered during the last few
days on EM? On EM, we customarily reply to messages that claim to refute
something that was said, before we go on repeating the refuted
statement.

1) 1-person-1-vote is intended to mean that each person gets one ballot,
   with the same voting opportunities, and counted by the same rules,
   rules that are invariable with respect to the name of the voter.
   Approval meets that criterion.

2) In Approval or Plurality, your ballot affects the outcome if
   you've voted for X and notfor Y,  and, when all the ballots but yours
   are counted, either X & Y are equal, or Y is one ahead of X.

   So your ballot has effect because you've voted between X & Y.
   In Approval, if you vote for 1 candidate, and I vote for all but 1,
   we're both voting between the same number of pairs of candidates.

   We're both giving a point assignment to all the candidates: You're
   giving "high" to one and "low" to the rest, and I'm giving "low" to
   one and "low" to the rest.

3) It's been recently shown here that Plurality allows voting power to
   vary by a much larger factor than Approval does.

4) Approval doesn't let anyone give more than 1 vote to a candidate.

5) In Approval, any voter can cancel-out any other voter.

6) If, in Approval, more people voted Smith over Jones than vice-versa,
   of what relevance is it if some of those voting Smith over Jones
   also voted for John Doe? That doesn't change the fact that they
   voted preference for Smith over Jones, and that Smith is obviously
   preferred to Jones in terms of voted preferences.

What do you mean by 1-person-1-vote?

Mike Ossipoff

_________________________________________________________________
STOP MORE SPAM with the new MSN 8 and get 2 months FREE* 
http://join.msn.com/?page=features/junkmail

----
For more information about this list (subscribe, unsubscribe, FAQ, etc), 
please see http://www.eskimo.com/~robla/em