[EM] utility agreement - I wish...

[Warren Smith]

>robla: Warren, we don't agree.  I said there is NO systematic, fair way of
measuring utility.  I didn't say it's hard, I said it's impossible.
Ergo, for purposes of studying electoral systems, it might as well not
exist.

Using Bayesian regret on numeric utilities is begging the question.  By
stating utility as a numeric range, you're using Range-style metric.
It's not entirely surprising that Range Voting does well measured in its
own terms.

> However, Heitzig has repeatedly and clearly stated that it 
> "does not exist."
> I have repeatedly stated that it does exist, it is just hard to measure
> and hard to get people to tell it to you honestly.

Could you cite an example you're referring to?

----------------

WDS reply: - sigh - seems I spoke too soon.

Examples, you ask? OK, quote #1:

> But what we definitely
> know is a false claim is that there "is no such thing."
Do we know that?
Interesting. I don't know it since as you know I state the contrary.

end quote #1.

Quote #2:
Utilities are just fantasy.

Quote #3:
> The only reason we can claim A>B is Util(A)>Util(B).
No, there is no such thing as Util(A) or Util(B).

Quote #4:
...since it doesn't exist...

----------------------

Now let me address these statements by you (robla):
>I said there is NO systematic, fair way of
>measuring utility.  I didn't say it's hard, I said it's impossible.
>Ergo, for purposes of studying electoral systems, it might as well not
>exist.

--All three of these sentences by Robla are false.
One systematic fair way to measure utility of some societal policy is,
you come back in 100 years and you see what that country's GNP,
life expectancies, and literacy rate (say) are.  If they are high,
then that policy had pretty good utility.  You need a lot of control 
and experiment countries,
but this appraoch can work and has often been used with apparent success.

Second, even if something is completely unmeasurable, that does not
at all imply ("ergo") it "does not exist for purposes of studying XXX."
For example, the temperature in the inner core of Alpha Centauri is
completely unmeasurable.  Does that mean it does not exist for the
purpose of studying astrophysics?

Third, I have already demonstrated, in work you do not appear to have understood
(or perhaps not read) that you can use "bayesian regret" to compare voting systems,
purely by using the fact that utility exists, and without ever having to measure any utility of
anything to any human.  This is not just me being weird.  The same idea
was also thought of and used by S.Merrill, R.Bordley, and R.Weber.
So I urge you to inform them all that they are insanely using a concept that
might as well not exist for the purposes of their field.

It is perfectly ok if you have not read or understood my work, or theirs.  I am trying
to at least get you up to the 20th century, i.e. to appreciate 
that utility exists and has... uh...
utility... so that you can at least hope to begin to do that.
If you simply start by asseting utility does not exist (or might as well not exist)
then you cannot go any further and have reached a dead end somewhere around 1850 or so.

Bayesian regret is not the only use of utility in voting theory.  Characterizing voting 
strategy is another.  Again, if  you simply start by asseting utility does not exist (or 
might as well not exist)
then you cannot go any further and have reached a dead end somewhere around 1850 or so.

It is very odd to behold the inhabitant of the EM list, I must say.
In some ways they seem well in advnace of the "official" political science
community in their investigations (at least, at the high points).
In other ways they seem not to have reached even the year 1900.  It sometimes
feels to me like beholding a remarkably intelligent set of dinosaurs.

------

Now let me address these other statements by you (robla):

>Using Bayesian regret on numeric utilities is begging the question.  By
>stating utility as a numeric range, you're using Range-style metric.
>It's not entirely surprising that Range Voting does well measured in its
>own terms.

--I am not sure what this is supposed to mean, but I have a guess.

First of all, utility is a number. It is not "unfair" to state utility
as a number, because it *is* a number.

(Perhaps you would criticize analysing a voting system with the aid of numbers, because 
that voting system involves using numbers, hence that analysis was "unfair"?  Ludicrous.)

Second, you may be suffering from the delusion, which I have heard before from
Bart Ingles and quickly refuted,
that my studies were somehow cooked by making utilities be
random numbers in the interval [0,1] and then making range votes
also be scores in the range [0,1], and hence "designing
the experiment to get a good result".

That is a delusion for several reasons. First of all, many of
my experiments did NOT have utilities restricted to [0,1] - for example
normal random variates were used (unrestricted).  Only a subset
of my experiments used the range [0,1].  
Second, even in those experiments, my range votes, even with "honest" range voting,
were NOT the same as the utility values, because of rescaling by the voters.
That rescaling causes problems like the "free the slaves vote" gets the wrong
answer, utility-wise.  (Range voting is not immune to every problem, unfortunately.)
Third, if with the first two delusions corrected you think it is somehow
"unfair" that I measure bayesian regret with the aid of utilities, you
are wrong.  Bayesian regret is defined in terms of utilities.  There is a
unique right way to define it.
If I were measuring it in any other way, *that* would be unfair.

Finally, I obviously could not have been "cooking" the experiment
because at the time I was not a range
advocate.  I had no idea how the experiment would turn out.  I became a range advocate
after I saw the results of the experiment, which were vastly clearer and more
in favor of one system (range) than I ever would have expected a priori.
(And after I saw the results of a second experiment, this time with real humans, 
which made clear the "nursery effect" which made it clear that range voting, and only 
range voting, could hope to garner unified thrid-party support.)

If I had been wise enough to predict how the results were going to come out, and
that I would later become a range advocate, then  I would have designed
the experiment to use DIFFERENT ranges for the votes and utilities - i.e. I
would have made an artificial weird range like [-77, +82] or kludged
it to keep everythign unbounded - to prevent
the automatic assumption by "knee-jerk, read only 1 paragraph in the paper, find a way
to try to interpret it as a `cook', quit reading the rest of the paper,
and then claim Smith cooks his data" pre-biased people like Bart Ingles, that
I was some kind of fraud.  I however at the time was not smart enough to
anticipate this sort of reaction.  I wish I had been.

Now, you may still say: "range voting is based on numbers and somehow this utility
stuff unfairly favors it."   I would sort of agree with that, but I would
interpret it in a more positive way, i.e.: "BECAUSE range voting is close
to what the perfect voting system woudl be if only we could count
on people to tell us undistorted honest utility values, THEREFORE
it is not too surprising that it works well."

Eh?  It is kind of like somebody saying "it is unfair to have to run
a race against sprinter Michael Johnson - the setup makes it likely I will
lose" but somebody else says "this suggests that Micahel Johnson is
a faster runner than you are."  They are two different somewhat-true
views of the same thing, but I suggest to you that the productive view,
is the latter!  Only a narrow-minder pre-biased person would take the former view, and
I think you are bigger than that.

wds