[Election-Methods] [EM] RV comments

[Abd ul-Rahman Lomax]

At 01:51 AM 7/29/2007, Juho wrote:
>On Jul 22, 2007, at 6:58 , Abd ul-Rahman Lomax wrote:
>
> > At 08:17 AM 7/21/2007, Juho wrote:
> >> I also think that Range is a good method in non-contentions polls and
> >> elections. But in the statement above a competitive election was of
> >> course the assumption.
> >
> > Why?
> >
> > *Assumptions should be stated.* Election methods are used in all
> > kinds of situations. If there is an ideal election method for use
> > where it would be expected that people will vote sincerely,
> > shouldn't we know it?
>
>Yes, it is good to state the assumptions. I however do assume
>election scenarios to be competitive unless other assumptions are
>explicitly stated. Especially with Range it makes a big difference to
>me if one plans to use it in a non-competitive or competitive set-up.

Let me point to something a little deeper. "Non-competitive" means 
what? I'd take it as meaning that the electorate wants to cooperate, 
to find the best solution for all, and it is assumed that it is not a 
zero-sum game. It is often not necessary that *anyone* "lose."

In "competitive" elections, there are winners and losers. While the 
game may not be zero-sum, it tends toward it, so voters are 
polarized. Which group will get its way, which group will lose and be 
disappointed?

By limiting ourselves to "competitive elections," we are limiting 
ourselves, actually, to dysfunctional societies. We need to know that.

And there is a conclusion we can make. If we care about improving the 
function of society, we should worry that an election method that 
works beautifully in a dysfunctional society might actually inhibit a 
return to function. If the election method encourages polarization 
and competition, it may prevent the society from healing.

So, I'd suggest, we should incline toward election methods that will 
make good choices in a cooperative environment, and that will not 
make bad ones in a competitive environment. It's possible that the 
ideal election method is not so good in the competitive environment, 
but if it produces better results when people cooperate toward the 
common welfare, it is superior to a method maximized toward function 
under conditions of competition.

However, it looks like we have methods that work best in a 
cooperative environment -- and Majority Criterion and Condorcet 
Criterion methods can spectacularly fail in this -- and that still 
perform as well, approximately, as the best "competitive" methods. 
Range is a candidate for this, and Range with runoffs under certain 
specified conditions is, arguably, even better.

In a cooperative environment, where we may assume a much higher 
percentage of sincere votes, the Range winner is optimal (though 
there is even then, because of the normalization problem, better 
result from Range+2), and the runoff rules might even eventually be 
discarded, perhaps to be triggered again if signs appear of serious 
competition.

In a competitive environment, the runoff rules guarantee what I 
consider the bare minimum of democracy, which is majority consent.

So let's be careful!

> > And *then* we could look at what happens when some people don't
> > vote sincerely? It's clear that there is harm done, under some
> > circumstances, but how much harm and who is harmed the most? Is it
> > the sincere voters? Or is it those who did not provide accurate
> > information to the voting method, so it can't possibly optimize
> > their satisfaction?
>
>Unfortunately with Range my understanding is that in a situation
>where we have several "parties"/"groupings", some of which vote
>strategically and some not, Range is too rewarding to the strategic
>groupings.

Note that "rewarding strategic groupings" is not the same as "harming 
the public." Elections are not a zero-sum game.

"Too" rewarding is a quantitative judgement. What is "too rewarding"?

This argument has been presented many times against Range, and I have 
never seen an analysis of what "too rewarding" is. Obviously, 
something is missing, there are assumptions being made that, for 
example, there should be no reward for strategic voting.

Yet every method, to some degree, rewards strategic voting, and the 
reward can be large. What is "too rewarding?"

Until that question is asked and answered, this argument is 
essentially meaningless, an opinion without foundation, and quite 
likely based on a moralistic judgement about strategic voting. But 
strategic voting in Range is unlike strategic voting in ranked 
methods. The latter require preference reversals, which are 
tantamount to lies. Strategic voting in Range merely requires 
truncation, which is not lying.

By being familiar with "strategic voting" in ranked methods, and 
disapproving of it because of the essential deception (which is an 
error, a vote is not testimony, it's an action), this disapproval 
gets transferred to Range by using the same terminology. Unfairly, in fact.

Voters express preferences over some scale. They normalize this scale 
to their own internal, non-normalized scale, we can presume. For 
simplicity, we assume a linear transformation. They take a piece of 
their internal scale and lay it over the range of votes allowed.

If they place the preferrred candidate at the top, and the least 
preferred at the bottom, we call this a "normalized sincere vote." A 
"fully sincere" vote is actually normalized also, but probably only 
at one end, but voters could transfer their *entire* internal scale 
to the election range, and probably would vote most candidates, or 
all, somewhere in the middle.

*It is up to them*.

In any case, it is no more or less "sincere" in the ordinary meaning 
*however* they match their internal scale to the range of the 
election, provided they don't *distort*. And some kinds of distortion 
would also be, in the ordinary meaning, "sincere." (this would mean 
stretching or compressing part of the internal scale while laying it 
down.) Only reversal of preference is actually "insincere."

And if they place their favorite, for example, above the scale of the 
range election (or, thinking of it the other way, they lay the range 
scale over their internal scale, placing the max rating on it below 
the favorite, it then becomes possible for them to max rate others, 
"sincerely."

It is a very serious error to term Approval-style voting in Range as 
"insincere." There is no basis for it other than a convention, and 
such conventions are dangerous, where they create special 
terminology, accepted in a specialized field, with implications quite 
different from general usage. We get to use big words in special 
fields, and we can even coin terms.

So Warren Smith is right. There are different kinds of strategy, and 
Range never encourages reversal, therefore it never encourages what 
is, in general usage, "insincere voting." And what I've mentioned 
elsewhere about missing candidates, write-ins where the voter 
indicates true preference, is possible in Range without harm, or with 
harm so small as to be negligible, it depends on the details. (In 
Range 2, for example, if the voter voted 2 for a write-in, and then 
wanted to maintain preference and so voted 1 max for other 
candidates, it would be harmful to this voter's participation. But 
the voter in Range 2 would simply vote 2 for the running candidate 
and 2 for the write-in. It's pretty obvious from that vote that the 
voter prefers the write-in! The voter would not bother if there was 
actual equality! And I would treat it this way in preference analysis.)

> > I don't find the answers to these questions obvious. Apparently
> > some do.
>
>I think the main rules for Range are quite straight forward. There
>are some special cases that raise interesting second thoughts, but as
>a main rule I'd say that if some "bigger than marginal" group of
>voters is strategic, then Range tends to become "Approval with option
>to cast weakened votes".

Which is not harmful. Indeed, my studies show, so far -- all the 
details have not been nailed down -- that converting a Range 2 
election to Approval harms *all* voters, that is, it reduces their 
expected outcome.

Range *is* "Approval with option to cast weakened votes," it does not 
become it. And it turns out that those who cast weakened votes, even 
if only a few (so few, indeed, that one voter can accomplish it), 
*help* not only the strategic voter, but also the sincere ones.

By making the election into actual Approval, quite as we would expect 
from the simulations, society overall is harmed. So, consider this as 
a proposal:

there is some benefit to strategic voters under Approval, under some 
scenarios, however, by denying them this benefit by making the 
election Approval, we are preventing them from gaining this benefit, 
*but nobody is gaining.*

So we are essentially *punishing" strategic voting by forcing 
everyone else to vote strategically. I.e., you *will* vote the 
extremes, because we are not going to give you any other option, 
because if we do, somebody might benefit and that would be BAD.

Please explain it to me, why we should consider strategic votes as 
something to prevent. They are a medicine that voters use when they 
are sick, when society is sick.

I have above argued that Approval style voting is not insincere. It 
merely represents a different overlay position and "stretch" between 
the internal utilities and the Range Votes than one which fixes the 
endpoints at the max an min candidate positions. We vote that way 
when we do not trust the rest of the electorate to vote sincerely. 
This is a *sickeness*.

(Either ours, or the electorate's or both.)

We know that if everyone votes sincerely, that society overall does 
better. It's true, absolutely true, that our own utility may not be 
maximized. But the method is such that our utility will *almost* be 
maximized. And it is entirely possible, even likely if we assume that 
the other votes are sincere, or at least usually sincere, that the 
benefit to others really does outweigh my own relatively small loss.

And routinely, we make decisions that are based on this.

And it comes back to us. If every election maximizes social benefit, 
overall, then while I might lose a little in some elections, I'm 
quite likely to gain *more* in other elections. Over a large series 
of elections, I gain more than I lose.

Essentially, trying to maximize my personal gain in a Range election 
by voting Approval style is short-sighted. If everyone does it, 
everyone loses, on average.

These objections to approval are themselves short-sighted, for 
essential elements are missing from the argument. How much is "too 
much reward." Quantify it and show how it can be expected to happen 
in real elections, and it might be worth looking at.

Otherwise it is simply shallow, knee-jerk thinking. Quite common, so 
Juho will find plenty of people to agree with him.

>In my terms you are interested also in elections that non-
>competitive. That's good. Let's just make it clear when we talk about
>competitive and when about non-competitive or less competitive cases
>=> Assumptions to be stated. (Term "we want" is just passive "one
>wants".)

The default is no assumption, so the logic and conclusions should 
reflect the general case. All elections. Then we, for example, 
specify "large," to mean that effects that appear when the number of 
voters is sufficiently small are neglected. It's not clear what it 
means to specify "competitive."

How does this affect conformance with election criteria, such as the 
standard ones or the Range or Approval criteria?

It affects, as far as I can see, the incidence of what is being 
called, a bit deceptively, strategic voting.

But that is not clearly defined, because, for reasons I've explained 
elsewhere, the internal normalized utilities for candidates might be 
950, but the voter votes 900 because 0 is really bad, 5 is so bad 
that the voter does not want to participate in the election of the 5. 
This isn't strategic, it is "sincere." And we can't tell the 
difference between this and a strategic vote.

Votes are actions, and they have effects. Election methods give 
voters certain power, and generally, when you have power, flexibility 
in exercising it is to be preferred. Range N is like giving every 
voter N votes to cast in an Approval election, it is like creating, 
from the voter, N voters, who are going to act as a cooperative 
community, since they are really one person with one common interest. 
And communities can divide labor or actions.

I say let each community decide how to vote. If they all act 
together, sure, they have more power, but that power has been 
artificially unified. Fine distinctions are lost, and the result is 
loss of intelligence. It's obvious from general principles.

A sane society would never prefer ranked methods to rating methods, 
once they understood the latter. This doesn't mean that they would 
abandon majority rule, the conflict is illusory and I greatly prefer 
that the method make majority consent explicit. No candidate should 
be elected without majority consent, ever.

If it is an emergency, and the office *must* be filled, then the 
majority can affirm that, quickly, thus choosing the outcome of some 
method quickly.

Many arguments about election methods are really arguments about 
democracy. Much more than we realize, many people are against 
democracy, they do not trust it. Old ideas die *very* hard.

> > Approval is a constricted Range method.
> >
> > Lost in all this is the fact that the general consensus among Range
> > advocates is that Approval is an excellent first step. It's cheap,
> > it's simple, it's easy to understand. And it *is* Range, just the
> > maximally constricted version.
>
>First step to what?

Election reform.

>  I that would be e.g. the U.S. election reform,
>then the second step might be difficult since it may require that the
>competitive elections first evolve into more non-competitive (to make
>Range really useful). I'm not very optimistic, I think the U.S.
>elections are currently quite competitive.

But the first step is to Approval, which works well in a competitive 
environment, there are no intermediate ratings, the point of the argument.

And then the next steps do make a choice, do we go toward ranked 
methods or toward sum of ratings methods? Lost in the discussion is 
sometimes the fact that Approval is also a ranked method, with two 
ranks. Given this, it is Condorcet-compliant. (However, the Condorcet 
Criterion is usually interpreted as applying to unexpressed 
preferences that it is presumed the voter would express if the method 
allowed it.)

And there could be a lot of debate over that. However, the Approval 
first step does set one precendent that would probably persevere. If 
we were to go to IRV, equal ranking would be allowed. Thus you'd have 
another option to reversing rank, making it possible to vote 
sincerely in IRV more often without harm. If I'm correct, this would 
improve IRV, cutting way back on the center squeeze effect. It also 
improves Condorcet methods, in general, to have equal ranking 
allowed, and results, probably, in fewer spoiled ballots.

General principle: allow the voters to use the ballot they are 
provided. Restricting what they can do, generally, harms democracy. 
Forcing them to make choices they don't want to make can be predicted 
to be harmful.

>I have mentioned this before. One could go in the direction of making
>Range Condorcet compliant (use it as a Condorcet completion method).
>Or alternatively one can be happy with the fact that in non-
>competitive elections/polls Range can pick better candidates (sum of
>utilities point of view) than Condorcet would pick.

*Even if the method is Condorcet* using a Range ballot to collect the 
information is far more useful, and it allows the study of election 
performance. Conversely, if we have a Condorcet ballot with as many 
ranks as there are candidates, and there are N candidates, the 
election becomes analyzable as Range N if the voters use it that way, 
collapsing some ranks to equal vote and leaving others empty.

And this may actually be optimal Condorcet strategy, I'm not sure.

The real question, to which Juho seems to be assuming an answer 
without presenting evidence, is that in *competitive* elections Range 
does not choose a better winner.

Range *is* Approval, that is, Approval behavior is allowed in Range, 
and contributes to the performance of the method. It is true that 
voting Approval style harms the overall utility maximization, but 
changing the election from Range to Approval simply *guarantees* that loss!

Range with almost all strategic voting is obviously not worse than 
Approval. It looks like, actually, it is still better, it only takes 
a single voter voting intermediate ratings (or three if we assume 
that these voters normalize) to improve the utility expectations over 
Approval. Isn't that an interesting discovery?