[EM] IRVers guarantee that maybe IRV won't fail.

[MIKE OSSIPOFF]

Guarantees that contain the word "maybe" don't sound very reassuring.

James uses the old, familiar IRVer argument that says, "Ok, IRV can fail in 
a worse way than Approval, so, instead of judging by how badly the methods 
can fail, let's change the subject to the probability of failure, how often 
it will occur."

First, I'd like to clarify for James what a criterion is: It's a 
precisely-defined yes/no test. Always or nearly always, criteria are about a 
guarantee that a method has. Usually there's something that a method is 
guaranteed to never to.

So, when James starts talking about probabilities, likelihood or frequency 
of failures, he's departing from the subject of criteria. "Ok, but how 
_often_?". IRVers like to argue that IRV guarantees that IRV won't very 
often make you regret that you didn't bury your favorite to help a 
lesser-evil.

A bit vague? Yes.

Without a quantitative measurement of the probability or frequency of the 
failure, and a quantitataive measure of it's disutility when it happens, 
compared to the disutility of Approval's equal rating, the "But how often?" 
argument doesn't mean much.

It can probably best be put in terms of a person's inclination to gamble. 
Approval gives a solid guarantee: For the first time, no one would ever need 
to ever bury their  favorite. And, no need, ever,  for a majority to reverse 
a preference in order to defeat a greater evil.

When you choose IRV, your are choosing to place your bet in a gamble. Maybe 
you'll get lucky and IRV won't make you regret that you didn't bury your 
favorite to help a lesser-evil. Or maybe it will.
If you're a gambler, maybe you'll like IRV.

But can we somehow put these failures and their likelihood on a quantitative 
basis?

That's what Merrill's spatial simulations accomplish. Merrill compared 
various methods, including IRV & Approval, in terms of social utility (SU). 
If there's any way to quantitatively evaluate a method that sometimes might 
do better, but sometimes does worse, SU is the best way to examine 
quantitatively a method's overall merit, over time.

In Merrill's study, IRV scored consistently significantly worse than 
Approval.

Another thing: James objection to Approval's balloting amounts to a 
rules-criterion. If a rules-criterion is the best that you can do to 
criticize a method, then you don't really have much of a criticism of the 
method. So what about results criteria?

Approvl fails SDSC, which requires that a majority be able to defeat a 
greater evil without voting a less-liked candidate equal to or over a 
more-liked one. But IRV fails SDSC too.

Well, as I said before, the IRVers have ICC & MMC. Of course a clone set 
preferred to everyone else by a majority is also a mutual majorilty set.

But, as I've been saying, every MMC example that makes IRV look better also 
puts under the IRVer's nose an example in which IRV fails WDSC & FBC.

But IRV will also often fail WDSC & FBC when there isn't a mutual majority 
or a clone-set.

That gives James an answer to his "how often" question.

And it agrees with IRV's much worse SU scores, as compared to Approval.

ERIRV(whole) is a whole different subject. We're talking about which is 
better, Approval or IRV. Ordinary IRV of the kind that the CVD IRVers 
propose. The IRV promoters have never accepted an IRV mitigation coimpromise 
like ERIRV(whole), or even ERIRV(fractional). If we're going to use a rank 
method, there are much better ones--Condorcet wv. So the IRV mitigation 
compromises are good only as compromises with IRVers. And since the IRVers 
don't accept IRV mitigation compromises, they're not really of any use.

Mike Ossipoff

_________________________________________________________________
Getting married? Find great tips, tools and the latest trends at MSN Life 
Events. http://lifeevents.msn.com/category.aspx?cid=married