Approval's effect on candidates

[MIKE OSSIPOFF]

Rob L. wrote [about methods' effect on candidates]:

Does anyone know whether anyone has studied this, and has figured out any
way to quantify this?

I reply:

Yes. Weber & Myerson discuss that. I believe that Weber talks about
it in his article in the Winter '95 issue of _Journal of Economic
Perspective_. His article is entitled "Approval Voting".

Also in that article, and/or its refereces at the end of the article,
he refers to an aricle by Myerson in which Myerson proposes a model
in which an election is openly for sale. Each candidate has a certain
equal amount of money to divide among the voters after being elected.
Each candidate may promise to divide it among the voters in any way
s/he chooses. Each voter votes for the candidate who promises him/her
the most money.

Myerson found that IRV, much more than Approval, gives candidates
strategic incentive to make promises to many more voters.

That's the opposite of what CVD claims will happen with IRV &
Approval.

Rob L. continues:

However, Approval voting, it seems to me, leads to centrist weasels.
Basically, it's the "don't offend anyone" voting method, because that's
how I think candidates will win elections. It seems that when talking
about electing leaders, anyone who actually gets things done is more
susceptable to criticism than a do-nothing.

I reply:

But the opposite was found when Myerson compared how Approval and
IRV give candidates strategic incentive to try to please everyone.

As for encouraging centrists, of course Approval does that. Any
good method rewards candidates at the voter median position, the
sincere CW position. IRV's jumps away from center to extremes accounts
for its lower SU scores in simulations, and its need for more
drastic defensive strategy.

Rob L. continues:

That's why I'm worried about the support for Approval on this list from
Condorcet advocates. It seems that Approval has been deemed "almost as
good", based on mathematical properties having everything to do with voter
behavior, and nothing to do with candidate behavior.

I reply:

Approval's advantages discussed by Weber & Myerson do include
candidate behavior.

Rob L. continues:

Perhaps looking at it another way, it can be spun as purely a voter
behavior problem. Approval is almost as good at finding the center of the
spectrum if the spectrum is one dimensional, but I worry that it fails the
two dimension test.

I reply:

I expect that even with several issue dimensions, Approval would
still home in on the voter median position, the sincere CW position,
and stay there. If someone away from that position won, people who
prefer someone closer to median to someone even farther away from
it than the winner is are likely to notice that that farther candidate
has a lower vote total than their preferred candidate, and so they
neen't vote for that winner in the next election.

Approval is much better than one would expect a nonranked method
to be. IRV's benefit from rankings, and even much of the better
methods' benefit from rankings is illusory. The best rank methods,
in my opinion, bring more improvement than problems, but that isn't
true of IRV.

People are surprised to find that even with Condorcet, adding
to the count some ballots that vote Smith over Jones can change
the winner from Smith to Jones. As I said, I consider that to
be outweighed by Condorcet's advangages.

And if there are any unnacceptable but winnable candidates in the
election, even with the best rank methods, I'd rank all the
acceptable candidates in 1st place, as Blake suggested.
Condorcet(wv) doesn't give defensive strategic need to do that, but
it gives incentive under some conditions. And there _are_
unacceptable but winnable candidates in our public elections, as
I judge the candidates, and so I'd vote Condorcet as if it were
Approval.

I still like Condorcet, because I like the luxury of expressing
all my preferences, and I like GSFC & SDSC. But, as I often say,
the best 1-balloting nonprobabilistic rank methods can't improve
on Approval as much as we might expect them to. Improving on Approval
is more difficult than I expected.

For me, FBC and SARC are valuable. Approval and Condorcet both
have their exclusive criterion compliances. They're both excellent.
For me, Approval is a very close second to the best Condorcet versions.

And Approval is much easier to define, explain, propose, and
implement than any rank method. And it's much easier to decide how
to count Approval ballots than rank ballots: Add them up.

It's might take a long time to educate the public about how to
count rank ballots. In the meantime we can have Approval's considerable 
benefits.

We can talk about getting to the top, but tremendous benefit can
be gotten by getting off the bottom.

Mike Ossipoff






_________________________________________________________________
Chat with friends online, try MSN Messenger: http://messenger.msn.com