[EM] RE : Re: A few concluding points about SFC, CC, method choice, etc.

[Abd ul-Rahman Lomax]

At 10:52 AM 2/19/2007, Kevin Venzke wrote:
>--- Abd ul-Rahman Lomax <abd at lomaxdesign.com> a écrit :
> > Election criteria sometimes presume omniscience. For example, the
> > Majority Criterion is based upon voter preferences that may not be
> > expressed, or even expressable, in the votes. "Prefer," as it was
> > clearly interpreted here, refers to a mental state of the voter.....
>
>Well, I wouldn't define MF that way. But I can go with this. We can
>say that MF says that if there is a majority favorite on sincere
>preferences, and voting is sincere, the MF wins.

If you use actual votes rather than unexpressed 
but sincere preferences, then Approval satisfies 
Majority Favorite. But when I pointed that out 
here, I was told, quite clearly and with nobody 
chiming in with support for my position, that the 
Majority Criterion -- which I think is the same 
as MF -- was about sincere preferences, not about actual votes.

And even though the Majority, under Approval, has 
a means of expressing "strict preference," which 
is to bullet vote. Because they might elect to 
not do this, for whatever reason, it is alleged that Approval fails the MC.

If we *don't* allow the concept of "sincere" 
preference, but only expressed preference, then 
Approval fails MC in a manner that it obviously 
*should* fail it, that is, the failure is purely 
technical, since it would be failing to elect the 
preference of majority because a *larger* majority preferred another.

>[...]  And we know, without doubt, that the
> > Expected Utility Criterion and the Majority Criterion are not
> > mutually compatible.
>
>Presumably the EUC would also have to require sincerity in the votes.
>Yes, I agree that EUC and MF aren't compatible given this.

Right. Now, EUC is an actual measure of election 
success, whereas MF is what I'd call a secondary 
measure. It is a criterion which *seems* rational 
if one neglects preference strength, and many 
would neglect it on first thought. In the pizza 
examples I've given, the pizza election is an 
unqualified success if everyone is happy with the 
outcome, and it is a partial success if as many 
as possible are happy. "Unqualified success" is 
often *only* possible if the MF is violated.

But the *real* criterion behind MF, the totally 
legitimate one, is majority rule. That is, the 
majority has the right of decision. As 
parliamentarians know, this right is best and 
most clearly expressed when it is made through a 
Yes/No decision that has been the subject of full 
process in the determination of what question is 
to be asked; at every step along the way, the 
majority has the right of decision.

The problem with election methods in general is 
that they attempt to short-circuit this process. 
So, for me, the question of election methods 
reduces to the question to how to *best* 
short-circuit it, to obtain a result that is the 
mostly likely result that would be obtained 
through standard deliberative process, but 
without the time and effort involved.

Part of the deliberative process is a 
consideration of the impact of the decision on 
minorities. If a majority simply steam-rollers a 
decision through based on its undenied majority 
power, it can make *very* bad decisions, 
decisions that polarize society and make friends 
into enemies. Decisions that cause civil wars or 
insurgencies. Decisions that make societies 
dysfunctional in many ways, as people 
increasingly consider government to be "them" rather than "us."

MF in an election method is that steamroller. 
It's fast, it's easy to understand. And it can 
flatten far too many people. *Usually* MF will 
pick the best winner. That's why democracies 
using it have been as successful as they have. 
But if you look at the rough edges, the places 
where democracies *aren't* working well, you 
might see what I've seen: MF is part of the 
problem. It only has to make a bad decision 
occasionally to have this effect, for these 
decisions accumulate over the years.

Range or Approval won't suddenly cause elections 
to violate MF, the vast majority of Range and 
Approval elections are likely to satisfy it. My 
point is that there Range and Approval *don't* 
satisfy it, they do so to find a better winner.

So MF is in indirect measure of election success, 
not a direct one. And most election criteria are 
like this, except this SUC which was have not 
precisely defined, but which we can understand 
clearly in at least some situations, enough to 
know that MF should be sometimes violated.


>I do think it is interesting to consider whether a method fails
>"gracefully." The worst result under Condorcet methods is probably
>worse than the worst result under IRV. You have to ask also how much
>to worry about this relative to other concerns.

It's a real concern. If we want to talk about 
"worst" case, IRV can fail spectacularly because 
of its peculiar method, Warren has posted an 
example that shows a seriously terrible outcome. 
Obviously, not very likely, in the extremity 
which he shows, but .... you said "worst."

Range and Approval are criticized for failing MF 
when, in fact, they do so to choose a more 
broadly acceptable winner. That is hardly a bad 
outcome..... IRV can choose, fairly easily, a 
winner who is strongly *disliked* by a majority, 
if I'm correct. (Warren's example is extreme in 
this, but there are much more likely outcomes; it 
is basically the well-known center squeeze effect.)

> > For me, the only question about SU is the best way to measure it. The
> > Condorcet Criterion has *nothing* to do with measuring SU. It's just
> > a guess at a beneficial method, an intuititively satisfying one that
> > turns out to miss something extremely important.
>
>In simulations it seems to be quite a good guess, assuming sincerity.

Sure. Particularly when you have many candidates, 
Condorcet converges on Range in some ways. What 
happens, I think, is that preference strength 
becomes more uniform. Range can handle preference 
strength in a three-candidate election, Condorcet 
can't, generally. But with a lot of candidates 
spread across the spectrum, Condorcet effectively 
*does* consider preference strength, so it is not surprising that it does well.

Condorcet is the king of *preference* election 
criteria. To do better, one must start 
considering preference strength, which is what Range does.

>Also, I can substitute "EUC" for "Condorcet" in that paragraph and
>really have the same criticism.

No. EUC is *about outcome* and Condorcet is about process.

(More accurately, of course, EUC is about 
"expected* outcome, expected outcomes are summed 
in Range, across the electorate, providing a 
measure of total satisfaction. Condorcet doesn't 
attempt to do this, rather it does something else 
that we intuit will do it. We intuit that the 
winner of every pairwise election will satisfy 
the most people. But, unfortunately, this can be 
far from true. Range, instead, asks people *how 
much* they will be satisfied by each possible 
outcome. It is thus approaching, as closely as 
possible, the goal of elections. The real 
question, to me, is not whether or not Range is 
superior with sincere votes, it obviously is, but 
what happens when voters, for whatever reason, 
are *not* sincere. It is quite a different 
question than "what is the best election method," 
for the answer to that latter question may depend 
greatly upon context. If we have a collection of 
people who value social unity, we wouldn't even 
consider Condorcet if we know about Range. People 
who are polarized, who only think about their 
faction winning, this is more difficult. Warren's 
work appears to indicate that Range is still as 
good or better than Condorcet methods. It would 
certainly be useful if others would confirm this 
result, or show that it is defective.)

> > > > It is obvious that MF has a utility. It is unlikely to choose a truly
> > > > bad candidate.
> >
> > It can and has, so I don't know about "unlikely." Think Ruanda. (Now,
> > I don't know the full details of the election involved, but it was
> > won by the leader of the majority Hutu tribe. I'm guessing that he
> > was a majority victor.)
>
>Well, you're replying to yourself here. I didn't say this.

Right. But the question does stand. Is MF 
unlikely to choose a truly bad candidate.

*Normally*, no. But under conditions of 
polarization, critical conditions, it can. That's 
the point. MF works well *most* of the time. Just 
not when something better is needed!

>I just want to know that you're positing sincerity when you say that
>methods which specifically optimize utility fail MF.

Yes. Of course. That is, all voters can vote 
sincerely, utility can be optimized *as they 
define utility*, and the method will fail MF, and might do so fairly often.

In fact, it's possible, I haven't looked at this 
in detail, that *nearly always* MF is not the 
socially optimum result, when the election is 
about something important and controversial.

MF is brilliant when the decision is very small 
and uncomplicated. It is very good when the 
decision is Yes/No *and* there has been thorough 
and fair debate, when voters are informed.

>But when you posit sincerity it isn't clear that this is an interesting
>point. It is hard for me to take this as an argument that MF is not
>desirable for real elections.

I'm saying that a good election method should 
work with sincere voters!!! It should *also* work 
with some segment of voters which aren't sincere, 
and it should not truly *punish* sincere voters.

What this means for me is that a good election 
method *must* fail MF, at least in the 
simple-minded definition of MF that can be used 
for one-step elections. If the majority has an 
opportunity to change its preference based on 
accurate information about the state of the whole 
electorate, that's another matter. If that 
opportunity is provided, MF can be satisfied 
*and* SU can be maximized, and I give MF priority *under these conditions.*

Why? Precisely because of the measurement 
problem. If the measure of SU is defective, for 
some reason, who is to determine that? 
Essentially, I've been led inexorably to the 
conclusion that the majority has the right of 
decision, but that it better make sure it knows 
what it is doing, because the majority 
unnecessarily imposing its choice on a minority 
can be quite destructive of social unity and 
welfare. You win the election and then your 
family is wiped out in the civil war that results. Some victory!

So a good method will, among other things, 
function to inform the electorate. Range really 
provides the necessary information, alone among 
all the methods. The question of how to analyze that information remains.

And the majority has the right to say, "this 
process didn't work, I smell a rat." No."

>[...]
> > Yes. However, that is post-facto information. How would I know that I
> > will seriously disappoint half of society? *The poll is the
> > information.* I have argued again and again that single-step election
> > methods are doomed to be inferior to more complex, essentially
> > deliberative, process.
>
>Well, I would argue that public elections actually are multi-step and
>complex. Since in public elections there should already have been
>polls.
>
>I believe in this scenario I really would be likely to know that almost
>half of the voters will be seriously disappointed by A.

Sure. And what we have works as well as it does 
because this works, *to a degree.* It's context, 
and, obviously, context can determine how an 
election method functions. What is good for one 
context might not be so for another.

The multistep process we have in the U.S. for 
many elections, i.e., party primaries followed by 
a general election, unfortunately, tends to 
produce polarization. Of course, the whole U.S. 
legal system is based on polarization, on 
contests. Attorneys duke it out before a judge or 
jury. Problem is, the truth can get lost in the 
process, the results can depend on the skill of 
the attorneys or the funds available to each 
side. If one outcome is profitable for one party, 
and the other outcome has no single big winner, 
merely general public benefit, public benefit can get short shrift.

Make no mistake about it, election methods are 
the tip of the iceberg. The real issue is how 
society is organized, how collections of human 
beings make decisions about how to communicate and cooperate.

> > Here is a way in which a Range method satisfies the Majority
> > Criterion: A Range poll is taken and the winner is presented for a
> > second vote to the electorate, "Shall the Range Winner be elected?"
> >
> > If the majority votes Yes, we have, with the overall method,
> > satisfied the Majority Criterion.
>
>What happens if the majority votes No? If it's "elect the range winner
>or have a new election" I wouldn't say that satisfies MF.

Not quite. In the case mentioned, the electorate 
has explicitly rejected a candidate. It doesn't 
matter why. That candidate should be ineligible. 
The exact ensuring process doesn't have to be 
stated for our purposes here. It could be to 
present the next winner in sequence, or to 
present the Condorcet winner (not so good, in my 
opinion), or, best, a new Range election, same 
rules, but without the previous Range winner. 
When the majority says No, it should mean No.

And this method satisfies the *goal* of the 
Majority Criterion because the Majority has 
explicitly elected the winner by voting Yes. The 
Majority, in that case, prefers that winner over 
the alternative (i.e., someone else). If the 
method failed because of strategic voting, the 
majority can simply toss that result out.

Unless the results were very close, or, unless 
there was serious distortion for some reason, 
such as some kind of last-minute campaigning that 
was drastically unfair, I think that it would be 
quite unusual for the Range winner to be dumped. But possible.

> > We could also do the same thing by presenting the Range Winner
> > together with the best winner by any other method, in which case, the
> > question would be a runoff between two candidates.
>
>I think Range winner vs. MF runoff would at least not fail MF in a
>meaningful way.

Sure. In this case, the majority would have the 
explicit opportunity to choose its favorite, and 
would only choose the Range winner if it decides 
that this will be better for society.

Which is *true* preference, informed preference.

The question is whether or not preserving the MF 
in this way is worth the cost. It's not clear to 
me. If people are worried about Range failing due 
to strategic voting, the runoff, where the MF and 
Range winners differ -- which won't happen very 
often, I'd predict -- then the MF/Range runoff would make sense.

Range fanatics -- we have a few -- might disagree 
with me. They would claim that the Range winner 
is the best, period. But they depend too much, 
I'd say, on perfection of the method, and don't 
trust the majority sufficiently.

Besides, the *real* problem is how the majority 
becomes an informed majority. We don't have to 
inform everyone, indeed if only a relatively 
small number of voters -- well short of a 
majority -- become well-informed, it would be 
enough. How can groups of people become informed, 
which reduces to, how can truth be found amidst 
the chaff and deliberate deception that is raised 
up, how can we find information that we can 
trust? Even more important, how can we find 
*analysis* that we can trust, because information 
is not enough if it is buried in the mountain of 
facts (and lies) that come before us every day?

I probably wouldn't raise the question here if I 
didn't think I had an answer, but what I find 
really odd is that fundamental questions like 
this receive so little attention....