Dear Greg,
you wrote:
Nondeterminism is a delightful way of skirting the
Gibbard-Satterthwaite theorem. All parties can be coaxed into exposing
their true opinions by resorting or the threat of resorting to chance.
Actually, if I remember correctly, that theorem just said that Random Ballot
was the only completely strategy-free method (given some minor axioms such as
neutrality and anonymity), so it's not really skirting it but just taking it
seriously.
However, it seems some minor possibilities for strategizing are acceptable when
they allow us to make the method more efficient. FAWRB tries to be a compromise
in this respect.
I don't dispute that. The nondeterminsitc methods I have seen appear
to be designed to tease out a compromise because a majority cannot
throw its weight around.
Right, that's the main point.
The abilities of nondeterministic methods to generate compromises is
formidable, but since we speak of utility, I would like to point
something out.
1) Using Bayesian utility, randomness is worse than FPTP.
Two answers: i) Please cite evidence for this claim, ii) Bayesian utility is
not a good measure for social utility in my opinion. We had lengthy discussions
on this already a number of times on this list, so I won't repeat them.
Instead, I will produce evidence from simulations this weekend which shows that
no matter what measure of social utility is used, Random Ballot does not
perform much worse than optimal.
2) False compromises are damaging
What do you mean by false? If a proposed compromise fails to be desirable by
most voters over the Random Ballot lottery, it will not get much winning
probability. If it is, on the other hand, it is not a false but a good
compromise. The simulations I will report about this weekend show that usually
we can good compromises to exist which have quite large social utility.
The reduced power of a majority means that at any choice with a
greater-than-random-ballot average utility is a good compromise
Notice how lousy the Bayesian utility of random ballot is and you
begin to see my point.
See above. In simulations with well-known preference models, Random Ballot
results are not lousy at all.
Also note that the method for determining the compromise is
majoritarian (to the extent that approval is) so the intermediate
compromise procedure is a red herring that produces some nasty
side-effects. The compromise is determined to be the most-supported
at-least-above-average candidate. How does this avoid the original
criticism of majoritarian methods?
You are right in that the majority still has some special influence on the
*nomination* of the compromise. But the important difference to majoritarian
methods is that they can't make any option get more winning probability than
their share without the minority cooperating in this. So, yes, they can present
the minority with a compromise they value only slightly better than Random
Ballot. This is not perfect yet, but it guarantees the minority to get a
better-than-average result where a majoritarian method doesn't guarantee a
minority anything!
Yours, Jobst
Election-Methods mailing list - see http://electorama.com/em for list info