[EM] some questions about utility

[Abd ul-Rahman Lomax]

At 08:02 AM 12/20/2005, Stephen Turner wrote:
>Is anyone there willing to write an Electowiki page
>(or point us to an existing place) about the relationship
>between voting and utility/welfare?
>There seems to be nothing on the usual sites.

Odd, is it not, that the fundamental *purpose* of voting has largely 
been neglected. Without an understanding of the purpose, 
determination of the optimum method is an exercise in argument based 
on hidden assumptions.

The concepts of spread utility ("greatest good for the greatest 
number") vs. majority utility (preference of the majority) vs. 
plurality utility (preference of the largest faction) underlie the 
preferences for, roughly, Range methods vs. Ranked methods vs. simple 
Plurality. While benefit to an oligarchy does, in fact, underlie 
existing methods in use in some places, rarely do we see anyone 
advocate it here.

>It would be useful to have a summary of
>the evidence (for and against):
>- the relevance of utility
>- its measurability
>- relationship between utility and actual vote cast.

In confined situations, utility can be easy to measure. It's a 
general concept, and I think it would be defined, essentially, as 
"measure by which a course of action or an outcome can be compared 
with others," so its relevance is a tautology. Arguments against 
utility that I've seen here have mostly been based on

(1) I have a better standard than your "utility." Which boils down to 
a disagreement over what metrics to use, really, rather than an 
argument against utility itself.

(2) Utility can't be measured. Let me put it this way: an election 
method is a device for measuring utility by using distributed 
analysis by the electors. So this argument is *really* a cynical 
argument against the effectiveness of elections. (If a better method 
of measuring utility is developed, would we use it? Isn't that an 
interesting question? I'd suggest that we would predict the utility 
of the "better method" by testing it with a vote. If the majority 
prefers the new method, it would replace elections. There is, in 
fact, such a method, at least under some conditions.)

It seems obvious to me that voters almost always cast votes according 
to their perception of the benefit either to themselves or to society 
of various outcomes. In other words, every voter is a utility predictor.

It's been argued here recently that the best election method 
*entirely* measures individual utility, that attempts to use social 
utility are doomed to failure; this appears to be based on a 
philosophy, practically a religious tenet, that fostering the pursuit 
of individual benefit always benefits society. While the success of 
free enterprise is evidence for this, so to the failures of free 
enterprise is evidence against it.

The pursuit of ndividual good *usually* produces social good. But not 
always. And the exceptions can be killers.

>Is it reasonable to suppose that someone who wishes to
>vote sincerely, in a method whose ballot design and
>counting rules permit it, would just list their
>utilities?

This is the goal of Range Voting, actually; the ballot is designed to 
permit the expression. But the question remains how to analyze the 
ballots to generate decisions. It's been presumed that Range ballots 
would be analyzed by determining the mean expressed utility, but 
there is, in fact, another understudied alternative: median rating; 
it's been overlooked because the standard definition of median 
combined with quantized data with low granularity results in, it's 
predicted, common election ties. There is a rather simple solution 
for that problem, which is to spread the ratings such that every 
rating is unique. More about that later.

>It was recently claimed here that we should seek to
>maximise total utility (sum over all voters).  Is
>there any reason why
>this is superior to any of the following:
>- maximise median utility

First of all, "superior" according to what metric? In the world of 
election methods, no method is presumed intrinsically superior, 
except for the extent to which it realizes assumed or stated goals. 
Total utility, which is estimated through what must now be called, 
more precisely, Mean Range or Mean Ratings or Mean Utility, is one 
metric, and Median Range is another. Median utility is a form of 
majority rule. Recent discussions on the Range Voting list have 
brought this out for me.

Mean utility is only majority rule if the majority accepts it. Median 
utility *is* majority rule.

Mean utility, *if* the ballot ratings are sincere, maximizes social 
good. Median utility maximizes good for a majority. So the 
superiority of one method over another depends on the questions asked 
above. It's quite a good thing that these questions are being asked!

>- utility of at least x for at least y % of voters
>(where
>   we could seek to maximise x for some fixed y,
>   or vice versa, or something else)

This is really two separate variations. Oligarchy maximizes utility 
for a limited group, and is typically justified by a belief (often a 
religious belief) that there is a group better qualified to make 
decisions that will benefit all. If the majority consents to this, it 
is, in fact, democratic. But, in my view, democracy is maximized when 
consent is maximized, and mere majority consent is precarious and dangerous.

Majority consent is like making decisions with a divided mind. In an 
emergency, that's what you may have to do. But someone who makes 
decisions in such a state, absent emergency, we would often recognize 
as dysfunctional, or even insane. A sane person continues 
investigation and consideration until the mind is much more united. 
And, I'd submit, so should society.

>- maximise the minimum utility (perhaps in
>   a small committee/tribunal).

It is not necessarily limited to that. This is called "consensus," 
and many groups are trying it. Yes, it is generally considered 
impractical for large groups, but there are organizational structures 
and techniques which can make (and, in a few circumstances, have 
succeeded in making) broad consensus possible. Usually, as the group 
becomes very large, it does become necessary to, more often than not, 
make decisions without the consent of a small number, but consensus 
organizations, even large ones, can be surprisingly effective in 
generating broad agreement. That we would often think this impossible 
is because our standard institutions don't place high value on 
consensus, and so don't put the necessary effort into finding it. 
Those organizations which *do* value consensus, if they implement it 
properly and sanely, find that the organizational unity is greatly 
strengthened. There is a major organization, founded in the 1940s, 
which succeeded where many others had failed before, to the extent 
that it rapidly, within a relatively few years, had meetings in 
almost every small town in the United States, and which is common 
around the world as well. It succeeded because of, I think, an 
analysis by the theoretician-founder, a stockbroker by trade, of what 
had caused previous attempts to fail, and he developed organizational 
principles which, among other things, were designed to maximize 
consensus. It worked.

The organization is Alcoholics Anonymous, and anyone who has 
participated in AA to the extent that they know how the structure 
works (you would not necessarily get this simply by attending a few 
meetings), I find, understands this immediately. So far, however, 
this understanding has not been translated into broader applications, 
though it *has* been translated into a host of similar programs for 
dealing with various psychological issues.


>Also, utility is generally used to define expectation,
>in terms
>of the expected value of the act of voting as compared
>to
>abstention.  The rational voter would seek to maximise
>their expectation.  What would be the correct thing to
>do if abstention were not possible?  Thanks,

Abstention can be a rational choice, maximizing expectation, so the 
constraint against abstention is contrary to the goal of allowing 
voters to maximize expectation. The rational thing for a voter to do 
who would otherwise abstain and if abstention is not possible is to 
express level preference, unless the voter *does* have preferences. 
If it is a Range method, being used for an election, vote the same 
rating for every candidate. If you think the system sucks, so to 
speak, vote 0 for every candidate. If you think it is a beautiful 
system, but you merely realize that you don't know bunk and thus 
don't want to muddy the waters with your ignorant opinion, you can 
vote the max rating for every candidate. More likely, however, you 
will have some preferences and you can express those; in a Range 
system you can vote mild preference by voting, say, in the midrange, 
such that the difference between the minimum and maximum preference 
is not great.

There is a controversy among Range advocates as to how to treat 
abstentions, which, in this context, means how to treat and analyze 
the ratings of a candidate in the presence of some voters who have 
abstained from rating that candidate. Without going into details, 
there are basically two choices: count abstentions as the minimum 
rating, or do not include abstentions in the result.

Now, I said I would write more about median Range. I'll do that in a 
separate post.