[EM] RV comments

[Abd ul-Rahman Lomax]

[continuation of response]
At 01:49 AM 7/21/2007, Michael Ossipoff wrote:

>Lomax continues:
>
>And if the real election is
>between A and C, I might as well rate B sincerely, i.e., 50%, it's
>not going to hurt me.
>
>I reply:
>
>Wrong. If A is much more likely to outpoll C than vice-versa, then 
>you should bottom-rate B. Admittedly, if it's almost surely between 
>A and C, it matters little what you do with B.

Because? B isn't going to win. Note that Ossipoff first disagrees, 
with a definitive Wrong, then backtracks, and agrees ("admits") what 
I just said.

"If the real election is between A and C," means that there is no 
reasonable probablity that B is going to win. So there is no harm to 
me in rating B sincerely.

The question, though, is whether or not it harms me in the 
zero-knowledge case, or where the election does not reduce to a 
simple two-candidate race in reality, for example with substantial, 
nonzero probabilities for the election of each candidate.

Ossipoff should know that strategically optimal votes depend on both 
utilities or expected satisfaction, and election probabilities. But 
the case of interest here is the zero-knowledge case.

Does it hurt the voter to vote sincerely in the zero-knowledge case?

And we can be more general than that: is the optimal vote always, as 
claimed, Approval style?

Clearly it is not. Indeed, with sufficient Range resolution, I 
believe I could show that it is *never* optimal. I think Warren's 
simulations show that, there is a page on it. The optimal vote, with 
some knowledge of probabilities, is neither the "sincere" vote nor is 
it the Approval Vote. And, geez, I'd think this would be intuitively obvious!

But, then again, I don't think like most people.

>Lomax continues:
>
>
>But what if the voter does not know who the two frontrunners are?
>What if the probability of election is equal for all three
>candidates? Well, in this case, the optimal vote is clear: rate B
>sincerely, i.e., at 50%.
>
>I reply:
>
>If there's no predictive information, then,  the optimal strategy, 
>the 0-info optimal strategy, is Above-Mean: Top-rate all the 
>candidates whose merit is above the mean. Bottom-rate the rest.

This is zero-info *Approval* strategy. It is not optimal zero-info 
Range strategy, when N>1.

Claim it, Ossipoff? Prove it!

>You say that B is halfway between the other two candidates. So it's 
>undecided whether s/he should be top or bottom-rated. Sure, then a 
>middle rating would be appropriate in RV.

Bingo!

>  In Approval you could flip a coin. Of course, the net result of 
> lots of voters flipping a coin for B is as if they'd all given B a 
> middle rating.

Yes. No disagreement there.

>So that's one instance where a middle rating has value: It saves you 
>the trouble of flipping a coin in that one particular undecided situation.

It's more than that. It improves expected outcome over voting Approval style.

>  So CR3 (CR with 3 rating levels)

I call that Range 2, because of all the times that I was finding that 
the number assigned to the method was one more than the number I 
needed to work with or express. Thus Approval is Range 1, i.e., the 
basic or first Range method, and the number refers to the preference 
gaps or preference strengths. There is only one preference strength 
in Range 1. It's basic, stripped-down to the minimum Range. Then 
Range 2 has an intermediate rating.

I'm suggesting it as standard terminology; the standard has never 
been agreed upon.

>  has a kind of justification not possessed by the CR versions with 
> more rating levels. I'd almost like to define RV as CR with more 
> than 3 rating levels, but I hesitate to ask people to accept new 
> definitions of existing terms.

Indeed. It's Range. Just like Approval.

And, politically, this is important. Start with Approval, because it 
is trivially simple and easy to understand. But it is limited in 
expression. So add 1 level. Or add one rank and do pairwise analysis 
on it. Take your pick, we don't need to decide now. Approval stands 
at the base of both paths, it's the first step up the mountain.

But the improvement that we get with Range 2 over Range 1 continues 
with additional rating levels, though with declining return. I have 
no idea, yet, where the optimum is, though I think Warren may have 
some data on that.

It is helpful, however, if there are as many rating levels as 
candidates, and preferably some more, and it becomes important if you 
are going to do pairwise analysis on the Range votes, as in proposals 
for triggering a runoff to allow the majority to ratify the kind of 
situation where the majority is making a small sacrifice (from 
personal maximum) to benefit a minority, which is what Range can do 
when the numbers are right.

And sometimes the numbers are wrong, because of strategic voting, so 
it doesn't hurt to check by asking. That's what a runoff can do.

>Lomax continues:
>
>So the kind of voting that is optimal depends on the relative
>probabilities of election, as estimated by the voter.
>
>I reply:
>
>All we needed was a newcomer to point out that amazing discovery to us. :^)

(1) I'm not a newcomer by any but the most bizarre definition. 
Perhaps Ossipoff is becoming senile, but I'm, I think, much older 
than he. If not, shame on him.

(2) This is in direct contradiction to what both Ossipoff and another 
writer quoted by Benham stated, so we needed *somebody* to point it out!

Ossipoff has many times referred to Range voters who don't vote 
Approval style as "suckers."

Voting Approval style, in *some* scenarios, for *some* candidates, 
makes sense. But I call it truncation. If your favorite is not a 
frontrunner, you are going to truncate at the frontrunner on the top 
end. I'm claiming that this is, in fact, a sincere vote. But the test 
of sincerity, really, is what you do with intermediate candidates, 
not your favorite, not so bad that you don't care if they don't win 
over the worst. In the election I described, where, after much 
badgering on my part, Ossipoff finally granted that it might be 
"okay" to vote sincerely for the middle candidate, he was still 
claiming that to vote approval style *for all candidates* was optimal.

Which is total, obvious, error.

>Yes, probability estimates influence Approval strategy, and, 
>consequently, RV strategy.

Yes. My point, in fact. Now, we can remove the probabilities from the 
question by making the situation zero-knowledge. What is the optimal vote then?

I have not studied the general case, I think Warren has, through 
simulations. In the particular case I study, which is actually 
*fairly* general, that is, I can assert that it is reasonable for 
voters to act, in common situations, as if this were the present 
situation, voting sincerely vs. Approval style appears to offer a 
small improvement in expected outcome. I'll give the numbers in a 
separate post.


>Lomax continues:
>
>Sometimes it is
>what we might call "sincere," but it is *never* insincere. It is
>merely, in some circumstances "magnified," or "truncated."
>
>I replyZ:
>
>If, as he admits, it's only sometimes sincere, then it must 
>sometimes be not sincere. That's insincere, by most people's 
>definition of "insincere". Again we have Lomax's new and 
>unrecognizable definitions.

General rule: Ossipoff hates precision.

Is "Not insincere," sincere? Only an Aristotelian would think so. Is 
a tree sincere? Well, it's not insincere!

Votes are votes, they are not sentiments; to apply the term sincere 
to them, we must have a clear definition. There is no clear 
definition of sincere as applies to Range votes. That is, I may have 
a set of utilities, and there is more than one way that I can vote 
them, more than one way, at least in some ways, sincere.

For example, if I normalize, is it "sincere"? If so, there is no 
sincere rating that is independent of the candidate set. Ratings are 
then a function not of how the voter feels about the candidate, but 
about preference strengths, which always involves pairs of candidates.

Yet, clearly, if I *don't* normalize, if I merely give absolute 
utilities, utilities that don't vary with candidate set, I'm far from 
being sincere, I've voting, in fact, in the only way that would truly 
maximize utility. Normalization loses utility information, causing 
pure Range to become suboptimal.

So, quite clearly, there is more than one set of ratings that is 
"sincere." Likewise, what if I leave the absolute ratings intact, but 
truncate them because my internal scale is "larger" than the range 
scale. Depending on the candidate set, there might or might not be 
more than one candidate at the extremes.

I could give examples, but I'll not do so at this time. If Ossipoff 
actually cared more about truth than convincing himself that he's won 
arguments, he'd figure it out for himself.

>I was referring to your opinion that it's more important to maximize 
>the sum of everyone's happiness than minimize the number of people 
>who call the outcome undeserving of approval, unacceptable in a 
>meaningful sense.

It's an inaccurate description of my opinion, for starters. Have I 
ever stated this alleged opinion?

And Approval, the method Ossipoff is promoting, doesn't do what he 
claims. At least not reliably.

>Range, dealing
>with strong preference, is Condorcet compliant. So what in the world
>is Ossipoff talking about?
>
>I reply:
>
>What in the world does that pair of sentences mean? Range does not 
>comply with the Condorcet Criterion, if that's what he means.

Actually, he's correct. I confused the Condorcet Criterion with the 
Majority Criterion. Which is the point that was being made, I just 
referred to it, obviously, with the wrong name. *Never* rely on 
Ossipoff to read through an error and respond to the intented meaning.

Range complies with the majority criterion, where the preference is 
expressed strongly, given the proper definition of strong.


>Lomax quotes me:
>
>>  There's a good case for saying that opinion is wrong. Do we 
>> really want to make the outcome unacceptable to more people, as 
>> long as, by so doing, we increase the benefit for someone already 
>> well-benefited more than we reduce it for those to whom we make it 
>> unacceptable?
>
>Lomax replies:
>
>Range maximizes, as well as we can tell, "the number of people who
>find the outcome acceptable."
>
>I reply:
>
>That ridiculous statement again. If you ask people where they draw 
>the "acceptable" line, then there's no reason to expect the most 
>accepted candidate to be the RV winner.

It's quite true that exceptions exist. Sometimes I fail to qualify 
statement, apparently, it looks like I did so fail here. What I 
intended to write -- and have written in many places, with proper 
qualification, that Range will generally select the Approval winner 
and vice-versa. It takes some special configurations of votes (the 
strong preference of a minority vs the weak preference of a majority, 
in particular, together with certain choices being made as to 
approval cutoff, to cause this coincidence to fail. And, in 
particular, I would build into the system detection of those 
conditions, and resort to further process when they exist. I still 
think that, unless there were anomalies in the voting, perhaps 
strategic voting that backfired, the Range winner *is* the best 
winner, and, if that's true, I expect that usually the Range winner 
will prevail in a direct contest with a pairwise winner (one that 
beats the Range winner, which might be in "approval,").


>Lomax continues:
>
>Ossipoff is slipping in his argument by
>avoiding defining what "acceptable" means.
>
>I reply:
>
>As I said, I'm not interested in your philosophical definition of 
>"acceptable". I use a simple operational definition: Someone 
>indicates that something is acceptable to them if they accept it. 
>It's reasonable to say that a voter accepts a candidate if s/he 
>votes Yes on that candidate's offer to govern in hir behalf.

Then I just say that the voter "votes for" the candidate, I don't 
attempt to add unwarranted meaning to it by saying that the voter 
"approves" the candidate, unless some means has been provided for the 
voter to do this explicitly.



>Lomax continues:
>
>And voters in Range are totally free to vote Approval style; if they
>do not, they are clearly granting permission to elect someone whom
>they rate at more than minimum.
>
>I reply:
>
>Fine. And they're sacrificing their own expectation maximization for 
>the greater good.

Ossipoff is claiming this, but he's already acknowledged that in some 
situations, these intermediate votes are strategically optimal *and* sincere.

And I'm coming to the position, given the lack of any solid evidence 
otherwise, and with evidence appearing from my study, that this is, 
in fact, the general case.

*Given* certain kinds of transformations of absolute ratings to Range 
Votes, that *could* result in truncation. As noted elsewhere, 
election probabilities shift the optimal Range Vote, and, as well, 
normalization is a transformation that loses some utility information 
that could be used to better optimize social benefit, but which 
always benefits the voter. This is the transformation of utilities 
from the realistic candidate set to the Range of 0 to 100%. Given the 
position of candidates in the internal utilities, some or most or 
even all candidates might be pushed to the extremes, thus voting 
Approval style. But that style is only appropriate for some voters.

Here's the simple algorithm, given absolute internal utilities as 
estimated by the voter for the entire candidate set, including 
write-ins. Consider which candidates are within a reasonable range of 
being elected. This will usually, but not always, rule out write-ins, 
but also, in general, some of the candidates on the ballot. Of these 
candidates, rate the best at 100% and the worst at 0%. Then rate all 
other candidates as indicated by a transformation of those now-fixed 
votes. So a candidate midway between the best and worst, in absolute 
utilities, would be rated at 50%.

I'm now claiming that this is not only sincere, in some meanings, it 
is strategically optimal.

In the two-party environment, where third parties have no realistic 
hope of winning, it may often reduce to Approval style voting, hence 
the common opinion that Range, optimally voted, is Approval. But it 
does not follow that other ratings than the extremes are useless, 
which is the claim that was being made, and it does not mean that to 
use these votes is suboptimal for every voter. It depends on the 
election context and the voter's preferences, and, I predict, if we 
get Approval in elections, we will also start to get Range in some 
places, and we will start to see, sometimes, third parties getting 
closer to success. And those intermediate votes will see increased use.

The reason why it reduces to Approval so often under current 
conditions is that the two-party candidates are usually, relatively, 
centrist. Still, for example, a Democrat may wish to express support 
for, say, Nader. But not indicate that he prefers Nader to Gore, 
because he doesn't. So he's use an intermediate vote for Nader. The 
reverse preference, though, he'd probably vote 100% for both Nader 
and Gore, unless either he thought Nader could win or it was Range 
with sufficient resolution that the notch down didn't seriously harm 
the Gore vote.

>  But doe we really need to legislate that into our voting statutes? 
> Shall we recommend that people do that? Or shall we just not tell 
> them that they're suckers if they do? Are you honest when offering 
> RV to people?

Ossipoff hasn't noticed, though it's been put in his face many times, 
that I'm advocating Approval, for now. I'm *not* advocating Range for 
public elections. I think it's premature. And what I think will 
happen next is that we will get Range 2, what Ossipoff calls CR3. One 
more rating.

So what is he talking about?

He's doubly wrong. Nobody is recommending that people vote like 
suckers (I think Range voters should vote how they think best, and 
I'd encourage them to fully understand the system.)


>Lomax continues:
>
>However, we do know that if voters vote sincerely, the *overall*
>outcome is improved.
>
>I reply:
>
>No, you believe that to be so.

I know it. Let Ossipoff show that it's not true, I've given in too 
many places, too many times, clear examples showing this. For 
reference, if anyone wants to look it up, the Pizza election.

>  In your arrogance, you believe that what you believe is something 
> that we all agree with you on.

No. I'm quite aware that there are idiots out there.

(Seriously, I'm also aware, *in addition*, that I'm wrong sometimes. 
Or that I'm write and intelligent people simply having come to that 
position yet. Both occur.)

>  When everyone votes RV sincerely, some are sacrificing their 
> expectation to increase the summed satisfaction of the electorate.

Hogwash. *That can occur,* but it is not the norm. It's unusual, 
relatively rare in the simulations, whether voters vote sincerely or 
strategically.

>  Compared to Approval, a greater number of those are thereby 
> getting an outcome that they'd vote "No" on, if asked to vote 
> up/down on the candidates.

Not if the implementation is the way I'd have it!

*I would ask them, if the votes indicate such a situation!*

Ossipoff is blowing smoke, and I'm wasting my time. Read the rest of 
Ossipoff's post, if you came across this one first. If he actually 
believes what he's writing, he's a sociopath, and thinks that only 
sociopaths aren't suckers.

I did not read the remainder, though, in detail, I only scanned it to 
come up with my decision not to go on with it, so if there is 
anything there that is important, please, anyone, bring it up.

I used to have a policy not to reply to Ossipoff.... maybe it's time 
to go there again.