[EM] Why the concept of "sincere" votes in Range is flawed.

[Abd ul-Rahman Lomax]

At 06:04 PM 12/6/2008, Juho Laatu wrote:

> > The problem here, Juho, is that "sincere opinion"
> > is not the basis for voting,..
>
>What is the alternative basis?

The basis is choice. I can drop a marble in this bucket, or that 
bucket. Maybe I can drop a marble in each one I choose of many 
buckets. Maybe I can drop ten marbles like that, with rules constraining me.

These are not opinions. The basis for voting is choice, I choose this 
action and not that one, and presumably I do so because of expected effect.


> > Approval
>
> > It gets tricky only in the
> > situation that there are three frontrunners.
>
>You could say "more than two" here.

Yes, I could. However, what I'd consider likely strategy tends toward 
bullet voting, which indeed is connected with *sincere opinion* as to 
preference. But the basis action is a choice, not a sentiment or 
sincere opinion. You cannot derive my opinions from my vote, 
necessarily. You can derive preferences for sets from Approval votes, 
and that is an opinion.

Okay, already, sincere opinion is *a* basis for voting, but rational 
expectations, the likely *effect* of the vote, is a necessary 
element. That too, is an opinion, an expectation by the voter.

tell, me, what is the difference between a sincere opinion and an ordinary one?

> > "Sincere expression" here means, "expressing
> > some preferences sincerely, and expressing no preferences
> > insincerely." It means no preference reversal.
>
>Definition "sincere expression" = "no
>preference reversal" is not very intuitive
>to me.

When I speak to you, I tell you only a small part of what I might 
possibly express. My expression is never complete. We say that it is 
sincere if none of what I express is dishonest, we do not require 
completeness. We *may* require that no *major* or *important* fact 
remain undisclosed, where concealing this is tantamount to 
dishonesty, but we don't extent that to all possibly relevant facts.

Range and Approval votes may be presumed to all be sincere 
expressions of setwise preferences. Preference reversal violates 
this, clearly.

>I didn't refer to automatic normalization here
>but only to voter using both min and max values
>since the society thinks this is the way one
>should (typically) vote.

That's normalization, actually. There is a lost performative here, 
"who" thinks this is the way one should vote, and a lost condition as well.

One should vote that way if one wants to maximize the effect of one's 
vote. There is no requirement that one do so. It's a choice. It could 
be reasonable in some cases.

>In Approval the cutoff placement related
>request/recommendation/advice could come via
>the media or via the election organizes. The
>point was that there exists a commonly known
>friendly advice (from whatever direction) to
>guide the voter to cast an "optimized" vote.

Sure. There can exist many such sources, giving different advice. 
It's up to the voter to determine what advice to trust, if any.


> > I wouldn't start with "sincere opinion,"
> > it's far too difficult. I'd start with the
> > frontrunners, with the set of candidates who might win.
>
>There is one fundamental difference between
>these two cases. The voter either takes or
>does not take into account the opinions of
>other voters and how the voting method works.
>Both are possible scenarios. In the scale that
>I presented this is case 1 vs. cases 3, 4, 5
>and 6.
>
> > Usually there are just two frontrunners
>
>True in two-party countries. In multiparty
>countries having more than two potential
>winners (=frontrunner?) is very normal (I'm
>used to about 3.25).

How often is it that one cannot predict with confidence that the 
winner is one of two?

With Chirac/Jospin/Le Pen et all, it was pretty much a foregone 
conclusion that the winner was going to be Chirac or Jospin. Le Pen 
got everyone excited, but he had a snowballs' chance in hell of 
winning the runoff. 80% Chirac. Yes, there were three frontrunners in 
first preference, but that's not the whole story. To vote 
strategically in top two, one does need to consider more than first 
preference. Same with IRV when there are many candidates in a 
multiparty system. It's not used for that, in fact, for good reason. 
It has the same problem as top two runoff, with *maybe* a chance to 
do somewhat better.

> > some authors
> > "improved" the definition of strategic voting to
> > include, it's a little unclear to me, however one sets
> > an approval cutoff. It's strategic in the sense that one
> > does it better if one knows and uses expected outcome for
> > the election.
>
>Yes. Not a bad definition. Here "method and
>environment information based optimization
>of the vote" = "strategy".

The problem comes when it is assumed that strategy equals bad voting, 
or being affected by strategy equals bad method. The question is the 
nature of the effect, and the kind of strategy.

>...
>I didn't assume that votes are "supposed to be"
>sincere. I only assumed that we can tell what
>the difference between "sincere" (unmodified)
>and "exaggeration" is.

If we know absolute utilities, we can. But there isn't much of a way 
to tell from votes alone. Some voting patterns may indicate a 
likelihood that votes have been modified from sincere, i.e., votes of

100, 99, 1, 0

in Range creates an impression of exaggeration of preference strength 
in the 99/1 pair, with a correlated minimization in the 100/99 and 
1/0 pairs. Exaggeration is always accompanied, in Range, by minimization.

>Usually elections follow the one-man-one-vote
>principle. But one could of course also measure
>the level of adrenaline of each voter to
>determine the the weight of each ballot :-).

Or skin resistance or heart rate or other measures. It could be done.

But there is a simple one: voter turnout. Votes for voters who turn 
out are weighted by a factor of 1. The preferences of voters who do 
not turn out are weighted by 0.

These weights are correlated with preference strength over the entire 
candidate set. The adjust election results toward broader social 
utility maximization. And one conclusion is that low turnout in 
runoff elections is to be expected because of the lower preference 
strength involved. Some voters are equally displeased by both 
candidates. These may turn out to vote if there was Condorcet failure 
with sufficient preference strength. Otherwise, not. Others may be 
equally pleased and likewise not turn out. So those with strong 
preferences are the ones who vote, thus shifting results toward 
Range. A Range winner is likely to win a runoff against a Condorcet 
winner, that's my assertion.

>I have always thought "strategy" to refer to
>any kind of modification of one's vote (or
>nomination or whatever) to achieve better
>results.

Modification from what? Try to find some standard definitions. Not easy.

>If the sum of the Approval votes equals the
>"absolute utilities" then I'd say that the
>vote follows the "sincere opinion"
>(="absolute utilities"). Anything beyond that
>could be based on strategic considerations.

You would mean relative utilities, not absolute.

>If the ballots are seen as separate
>independent votes (that makes counting of
>the sum irrelevant) then I'd expect all the
>ballots to be identical, and cutoff location
>not to be based on strategic considerations.

I may sweep my approval cutoff while voting the 100 votes, starting 
out, perhaps, with tight preference, bullet voting, then lowering the 
cutoff. How far I go, that's up to me. It is not necessarily 
strategic, which, here, must mean in consideration of the election 
probabilities. And it is not defective or selfish or bad to consider 
those probabilities.
>
> > That's right. The problem with "sincere" is
> > that it is a totally loaded term.
>
>Yes, many terms are, unless we use some pure
>abstractions like "alpha" and "beta" (well,
>also here "alpha" seems to represent the
>primary choice and "beta" the secondary
>choice). In scientific discussions one should
>try to separate theoretical (unloaded) use of
>real life terms from loaded (regular English
>language real life) use of them.

It is poor practice to use loaded terms, because it invites abuse. It 
may be necessary, but that's another matter.