[EM] RV comments

[Abd ul-Rahman Lomax]

At 01:49 AM 7/21/2007, Michael Ossipoff wrote:
>Lomax continues:
>
>Election criteria, generally, are pass-fail
>measures. A method either satisfies the Condorcet Criterion or it
>does not, at least if the Criterion is sufficiently well-defined.
>
>I reply:
>
>Just what we need--an arrogant newcomer to tell us what a criterion is.

Ah, the Ossipoff we have known so well for so 
long! He's been relatively well-behaved lately, 
but challenge him deeply and the covers come off.

I'm not "telling" Ossipoff what a criterion is. I 
write for a general audience, not for him. If I 
want to address him, I do so directly. I'm making 
a specific point, and what I wrote about criteria 
is an important foundation for that point.

>Lomax continues:
>
>But SU, properly considered, can *measure* the *relative* success of
>election methods.
>
>I reply:
>
>Of course, if that’s what you value and want to 
>measure. That your measure of method-merit is a 
>matter of degree instead of yes/no doesn’t make 
>it better, as you seem to want to imply.

It makes it more informative. And I see no other 
competitor. Election "criteria" are a problematic 
method of comparing election methods, for there 
is no particular basis for choosing which ones matter.

Measuring election success by election criteria 
is something that has been attempted for a long 
time, and what experts have come to is a basic 
conclusion that you can't satisfy all of them. 
So, then what criteria are important and what ones do we allow to fail?

>Lomax says:
>
>The Approval measure maximizes
>"approval" of the candidate. While I haven't see this from Approval
>advocates, I would define this as the voter would rather elect the
>candidate than see the office go vacant.
>
>I reply;
>
>I have to admit that Lomax has a creative, 
>original, new, and unrecognizable definition :^)

Ossipoff is utterly ignorant of deliberative 
process, he has shown this in the past. He knows 
a great deal about election methods, which are a 
detail in deliberative process, a shortcut used 
when full deliberation is considered impractical. 
The shortcut is a serious compromise, generally, 
and introduces all kinds of complications such as 
Condorcet cycles, Favorite Betrayal, and on and 
on. What seems to be new to Ossipoff is the 
concept that something is off or missing when we 
elect a candidate that a majority of the 
electorate would reject if asked the question, 
"Shall So-and-So be elected to the office?" And 
nearly every election method (Range included) 
could do this if there is no ratification process.

Now, if it were impossible to have a ratification 
process, all this would be moot. But it is not 
impossible. Ossipoff knows enough about delegable 
proxy to know that it could convert elections 
into deliberative process. I assume he also knows 
about another invention of Warren Smith, Asset 
Voting, which is a form of delegable proxy; 
indeed, it is pretty much what I described years 
ago as delegable proxy with a secret ballot front 
end. But this is beyond my intended scope, which 
is here about strategy in Range Voting. All this 
came up because Ossipoff has been using the term 
"Approved" in a way which implies that it's an 
absolute, as if you can simply determine that the 
number of voters who have "approved" of a 
candidate has been maximized, when, in fact, all 
you have maximized is the number of voters who 
have *voted* for the candidate. And the difference is important.

Because Approval of a candidate, as commonly used 
by those describing Approval Voting, is precisely 
equivalent to a Vote for the candidate, it is far 
simpler and less obfuscating to simply term it 
that, instead of bringing in the cachet of the 
word Approval, which certainly sounds stronger 
than simply having Voted for a candidate. I'll 
vote for a candidate as the lesser of two evils, 
but am I therefore "approving" of him?

>   Refusal to give someone an approval vote says 
> “I don’t want that candidate in office.”

Yes. That's correct. But the reverse doesn't mean 
that "I want that candidate in office." Rather, 
it means, "I prefer that candidate to those I did not vote for."

>  The voter isn’t asked to choose between other 
> alternatives such as “no one in office”.

That's right. And this is the shortcoming of 
election methods. You would ordinarily not do 
this in small-scale democracy. But Ossipoff is 
typical of a class of election-methods students 
who are interested in the complex high-level 
abstractions of election methods and their more 
abstruse aspects, more than the nitty-gritty of 
democratic decision-making; and it's clear from 
prior experience with Ossipoff that he has no 
experience in the latter. His idea of democracy 
seems to be putting up polls on his own 
initiative, without first vetting the 
*questions*, a basic principle of democracy. He'd 
never get away with this in an organization that 
was actually functioning with democratic process.

>  If you approve Smith and not Jones, then 
> you’re obviously saying, “I don’t want Jones in 
> office, but I accept Smith’s offer to govern.”

In a sense. But that does not necessarily involve 
"approval." To use the word outside its ordinary 
meaning is deceptive in effect. I think we 
started to use the term Approval to distinguish 
the action from Vote for, because the latter had 
some lingering impression that it was exclusive, 
whereas Approval carries an implication that more 
than one is possible. But the fact is that it is 
possible to vote for more than one candidate, and 
generally has been, except where specific rules forbid it.

We too commonly forget that choosing candidates 
for election to office is not the only use of 
election methods; they are generally of use for 
any multiple choice. But we have wisely avoided 
using public elections for complex, 
multiple-choice questions, because we know too 
well that the various election phenomena would 
bite us badly. Rather, we reserve these 
questions, generally, for representative bodies 
where the scale has been reduced (sometimes only 
barely) to one which makes deliberative process practical.

However, an exception has asserted itself, 
because there was no way to prevent it. If two 
conflicting referenda present themselves at the 
same election (or more than two), election rules 
generally provide that, if both pass, the one with the most Yes votes prevails.

This is Approval Voting, so the claim that 
Approval is not used in public elections, so 
commonly made even by supporters of Approval Voting, is false.

And in this case it is clear that if no 
referendum gets a majority "approval" -- the term 
is totally appropriate here -- no referendum is passed.

Approval Voting, as it has been proposed, is like 
a series of referenda of the form, "Candidate X 
shall be elected to the office." But the 
difference is that it is not required, except in 
some places, that each of these referenda receive 
a majority Yes vote; for that reason, we cannot 
equate a vote for a candidate with "approval."

Rather it merely means, "If we must elect someone 
from this particular candidate set, and given 
that I expect my write-in vote will be moot, then 
I choose this one, or this set of candidates."

>  Smith instead of Jones, if it’s necessary to 
> say it more explicitly. One could turn in a 
> ballot with no approvals, saying that none of 
> the candidates are acceptable, implying that 
> there should be another election with better candidates.

Yes, one could. However, with most election rules 
the effect would be the same as not voting at 
all. That is, it has no effect in the direction 
Ossipoff implies would be the intention of the 
voter, so it is problematic to claim that it means that.

There is an election between three candidates, 
and I prefer A>B>C. All three candidates are, in 
my understanding, dangerous lunatics, but the 
electorate, in my opinion, has been duped. I have 
what I believe to be solid knowledge that if A is 
elected, Doomsday will come in 200 days, if B is 
elected, it will come in 100 days, and if C is 
elected, it will come as soon as he's handed the 
football. (This explains, perhaps, my utilities 
as described in the example we have been working on.)

If I vote Approval style for A and B, or vote, in 
Range 2, 2 for A, 1 for B, and 0 for C, does this 
mean that I "approve" of A? or B?

Only in a relative sense.

>There could be a number of uses for simulations 
>with Approval, to compare its strategies, or to 
>compare Approval to other methods. I don’t know 
>how Lomax would generate utilities, but 
>typically one randomly positions voters and 
>candidates in issue space of 1 or more 
>dimensions, usually normally distributed in each 
>dimension. Then the merit of a candidate for a 
>voter is measured by the distance between 
>candidate and voter--either city block or 
>Pythagorean distance. I prefer city-block, for 
>reasons that I’ve described here an on the 
>Approval list. Briefly, it’s because I claim 
>that a candidate’s various issue distances from you are meaningfully additive.

Issue-space is one method for calculating 
utilities. Various methods are used in Warren's 
simulations, and variations on the utility 
distributions are part of the simulator's inputs.

However, I'm taking another approach, looking at 
the entire range of election possibilities from 
the point of view of a single voter with zero 
knowledge of the votes of other voters, so that 
the voter rationally may assume that those votes are evenly distributed.

There is another possible assumption which could 
be used, which is that the votes of others would 
be biased toward the votes of the voter (compared 
to random). This would be, in general, more 
realistic, but it drastically complicates matters.

Likewise, it could be assumed that candidates 
exist on a spectrum, so that, typically, far more 
often than not, voter opinion about one pairwise 
election will predict voter opinion about others. 
For example, if the voter prefers Nader to Bush, 
we might consider it quite rare that the voter would prefer Bush to Nader.

But, again, this complicates simulations greatly. 
Issue space simulation, though, should generally produce effects like this.

However, in the current argument, we don't need 
to go to those lengths. Readers will see when I write up the study.

>Whether a voter will approve a candidate of 
>course depends on those distances, and on what 
>strategy that voter is using. It may therefore 
>depend also on such things as vote totals in the previous election.

Sure.

>Lomax continues:

>However, Approval is, in fact, a Range method.
>
>I reply:
>
>Dueling definitions? Rangers use “Range” to 
>encompass all Cardinal Ratings methods. But, 
>especially since the Rangers are pretty much the 
>main users of the “Range” term, it seems useful 
>to let CR cover the broad class of 
>points-assignment methods, and let “Range” 
>denote the CR methods with more than two rating levels.

It's artificial. By making the definition 
complicated, sure, you can do that, but Range and 
Approval behave quite similarly; arguments 
against one are generally arguments against the 
other. The difference is simply the resolution, 
it is mostly a difference of degree rather than 
of kind, though there are some particular effects 
that occur from restricting the ratings to the bare minimum.

>Lomax says:
>
>We can expect that,
>generally, the SU maximizer will also maximize Approval
>
>I reply:
>
>Certainly not. A ridiculous statement.

Generally was an essential part of this.

>  Often all the methods used will give the same 
> winner. For instance, Nader won all of EM’s 
> presidential polls, by Approval, 0-100 RV, 
> Condorcet, and Bucklin. Probably by almost 
> every proposed rank count that we didn’t try, too.

Yes.

Approval is a kind of SU measure, reduced to 
binary. If we maximize SU, *usually* we will 
maximize the number of voters who rated a 
candidate above some "Approval cutoff." Not 
always, because of the issue of preference strength.

To maintain his position, Ossipoff must maintain 
that preference strength is irrelevant, because 
Approval provides no means for expressing any variation in preference strength.

>But often different methods will give different 
>results. AV and RV look at different standards, 
>count different things, and there’s no reason to 
>expect them to give the same result, though, as 
>I said, different methods sometimes do.

*This* is ridiculous. Both methods determine the 
position of a candidate on a scale. Range (>1) 
provides are more accurate position than Approval, that's all.

Elsewhere, there seems to be an assumption that 
midRange is an Approval cutoff; when I stated 
that a voter prefers A>B>C, with equal preference 
strengths in the A/B and B/C pairwise elections, 
Ossipoff claimed that this placed B at the "Approval cutoff."

It's not unreasonable to assume that; however, 
that is *not* where I would place an actual 
approval cutoff, depending on election 
probabilities, I could vote for candidates below 
midrange (sincere Range ratings, normalized) and 
not vote for candidates above midrange.

In any case, if midrange is the approval cutoff, 
voting for a candidate in Approval is an 
indication that the rating of the candidate is in 
the top half of the range, otherwise it is in the 
bottom half of the range. In Range, the range may 
be divided into more than two sections, and the 
vote indicates which section the candidate is in, 
unless the vote is distorted by strategic considerations.

But they are the same method; for Range N, the 
range is divided into N+1 sections. And Range 1 
is Approval. Approval is fully described by Range 
rules, unless special rules are added, such as N is greater than 1.

>Lomax continues:
>
>, though bare
>Approval, being binary in expression, loses data and thus cannot be
>accurate with respect to *relative approval
>
>I reply:
>
>So, Approval loses accuracy in measuring RV 
>scores. Amazing. Could that be because AV isn’t 
>RV, and we aren’t interested in RV scores? If we 
>wanted to accurately measure RV scores, we’d join the Rangers.

No. Approval loses accuracy in reporting voter 
preferences. I'll agree; Ossipoff is not 
interested in accurate reporting of voter preferences. It irritates him.

>Lomax says:
>
>Let's see what Ossipoff does:
>
>>  Say, for the moment, we disregard the fact 
>> that the SU claims depend on sincere voting, 
>> and that sincere voting is nearly always suboptimal in RV.
>
>Ossipoff continually makes this claim. It's false. "Suboptimal" is
>the trick. It is suboptimal, true, from the point of view of the
>individual voter maximizing his or her own personal utility
>
>I reply:
>
>Yes, that’s usually what means by “suboptimal.

Ossipoff interrupts my sentence midstream, which 
is an acknowledgement that *in some situations* 
the vote is suboptimal, and I specified something 
which he has routinely omitted. In those 
situations, the vote can be suboptimal for the 
individual voter, but if the vote is sincere, it 
is never suboptimal for the overall satisfaction of society.

Because these two optimizations can sometimes 
conflict *to a degree*, I made it clear that I 
was agreeing with respect to one, not the other.

>Lomax continues:
>
>, *in some
>scenarios.* In others, it is clearly optimal to vote "sincerely."
>
>I reply:
>
>Sure, sometimes your best utility-maximizing 
>strategy coincides with a frank expression of how you feel. So what?

So what? It was just stated that voting sincerely 
was *never* optimal, and Ossipoff agreed with 
that. Is he retracting that agreement? It looks 
like he isn't, but is he denying that he ever 
agreed? That would be interesting, and also easy to check!

>  For instance, in AV or RV, I’d bottom-rate 
> Hillary, Obama, and Edwards, because they don’t 
> deserve any better. My sincere opinion expressed on the ballot.

Ossipoff is here confusing sincere opinion with 
sincere rating, he is confusing votes with what 
candidates "deserve." And his personal politics, 
no matter how abominable, is not relevant here; 
he has often been known, when in difficulty in 
argument, to attempt to divert them into political flame wars.

Contrary to what Ossipoff wrote earlier, I'm not 
a newcomer. And even if I were, he's left quite a 
record that anyone can read....

>[irrelevant political rant deleted]
>
>
>Lomax continues:
>
>There are a series of problems. First of all, "sincere" vote requires
>some serious work to define. It is far from clear what a "sincere"
>Range vote is.
>
>I reply:
>
>
but that’s your problem as a Ranger, not mine. 
>But let me tell Lomax what he means by a sincere 
>RV vote: Your voted ratings are proportional to 
>the candidates’ value for you (as nearly as you can estimate that value).

No. I've written a great deal on the meaning of 
sincere with Range Votes, I won't repeat it here. 
There are algorithms for coming up with sincere 
ratings, and they don't involve proportions for 
*candidates,* they involve comparative 
*preference strengths*. This is quite not proportional to value.

Suppose we have an election with economic value. 
And I can estimate the financial return to me of 
the election of each candidate, making it easy to 
compare them and to derive my preferences and 
preference strengths. In this case, though, I can 
proceed to ratings from values, and I don't use 
proportions. There is no multiplication or 
division involved, except in the final conversion 
to a specific Range scale. What is then being 
"proportioned" is the overall scale, so the 
numbers in it are proporotional to the orginal normalized utilities.

Now, perhaps this is what Ossipoff meant and he 
simply expressed it imprecisely. Let me restate 
it. My voted rating *steps* are proportional to 
the preference strengths between the candidates 
involved. So if my A>B preference is twice as 
strong as my B>C preference, then the difference 
in rating between A and B will be twice the 
difference in rating between B and C.

In the economic model, the values would simply be 
associated with the candidate, the range of 
values would be converted to the range of the 
election (this is normalized to maximize the 
effect of my vote), and the necessary multiplier 
and offset used to adjust each value to a rating.

And if every voter votes this way, it maximizes 
the economic return to the society. Note that if 
it is a zero-sum game, there can be no winner, it 
must tie. But economic values are usually *not* 
zero-sum, zero-sum economic games are artificial and generally oppressive.

>Lomax says:
>
>We can defined a "clearly insincere" vote as being one
>where the range ratings reverse preference.
>
>I reply:
>
>Don’t be silly.

My definition is standard. The problem is that 
the word "sincere" does not state *how* sincere a 
vote is. An Approval vote is a Range vote on the 
scale of 0 to 1. By restricting the values to 
two, we cleverly avoid any consideration of *how* 
sincere it is; indeed, sincerity is totally 
unclear with Approval. Once we have more possible 
votes, though, the question is then, if a voter, 
in translating their internal preferences and 
strengths to votes, uses a scale that is larger 
than the Range election scale, thus "truncating" 
the votes, or some of them, such that more than 
one candidate sits at the extremes, is this "insincere?"

I claim that, no, it is not. It is simply not 
fully expressive. Is the lack of expression 
deceptive in some way? If I rate two candidates 
equally, does it mean that I have no preference 
between them? Clearly that's not the case; 
certainly Approval votes don't mean that!

There is no fixed, "sincere" translation from 
internal utilities to Range Votes. We can specify 
an algorithm and call that sincere, but then we 
are simply defining it specially. Absent an 
agreed-upon definition, voting methods debaters 
swing arguments on hidden and implied meanings, without ever nailing them down.

How many words have been wasted in arguments 
about what "prefer" means? Quite a few, I'll 
assert. I've pointed out, to the derision of 
some, that, as stated, Approval satisfies the 
Majority Criterion; and in the end, there are 
some very solid experts that came down with the 
position that, yes, there were some problems with 
the definitions. And, with one reasonable 
definition, requiring that "prefer" mean to "act 
to prefer," Approval satisfies the Majority Criterion.

And then various experts scrambled to come up 
with better definitions; and, often, these 
definitions still did not address the problem.

>  The opposite of a sincere RV vote is a 
> strategic vote that differs from a sincere vote 
> in order to maximize expectation, or accomplish some other strategic goal.

Attempting to define sincere by its opposite is 
quite problematic. And, here, the definition is circular.

True, I attempted to approach the definition of 
sincere by defining a "clearly insincere" behavior: reversing preference.

Consider this. There are four candidates: Gore, 
Nader, and the Antichrist. No similarity to actual elections is implied!

My internal utilities are 100, 80, and 0, let's 
say. However, I don't think there is any 
reasonable possibility that the Antichrist is 
going to win the election. So I truncate. Is my vote of 100,0 insincere?

Now, reduce the candidate set to just Gore and 
Nader. I vote 100 and 0, just to irritate Ossipoff, but also sincerely. :-)

Wait a minute, I think. I really should write in 
my true favorite. So I do, and rate her at 100. 
Does this make my first vote for Gore insincere?

I don't think so. I think that truncation is not 
insincere, in the ordinary meaning, and it is 
quite dangerous, in my opinion, to use ordinary 
words, especially loaded ones like "sincere" in 
discussing election methods in public forums.

Further, I may simple vote Approval style because 
it is, for me, in that election, easier. It's 
easier to put the candidates in two classes 
instead of rating them in more detail, perhaps. 
Is this insincere? No. But it is not "fully sincere, complete, and accurate."

>Lomax continues:
>
>What Ossipoff is talking about is voting "Approval style," in Range.
>But this isn't "insincere,"
>
>I reply:
>
>Lomax’s dueling definitions game again.

Trying to be clear with Ossipoff around is 
definitely like dueling. He's opposed to it.

Anyway, it's far too easy to go on and on with 
Ossipoff, and a huge waste of time. He's 
complained frequently about the length of my 
posts, though he's no slouch at post length. 
However, in addition, he fills up mailing lists 
with many posts as well. If someone doesn't care 
to read what I write, they can easily skip it, I 
don't mind. Many posts, on the other hand, can be 
a little more irritating. Still, I am no fan of 
censorship; where writers are to be censured or 
censored, I'd prefer, greatly, to leave it to democratic process.

>  We all know that when we speak of a sincere RV 
> vote, we mean it as I defined it above. “An RV 
> vote that is other than sincere” means a 
> strategic vote that differs from a sincere 
> vote. As I said, sometimes frank expression and 
> strategic optimality coincide, in which case 
> none of us would call the ballot insincere.



>Lomax continues:
>
>[With preferences A>B>C]
>
>Where do I rate B? Well, if the B utility is midway between A and C,
>we can define a "sincere" rating of B as 50%.
>
>I reply;
>
>That’s consistent with everyone’s definition of a sincere RV rating.
>
>Lomax continues:
>
>
>But what if the *real* pairwise election is between A and B? And C is
>actually irrelevant, C has no chance of being elected. Well, then I'd
>rate B min also.
>
>
>And in the reverse case, i.e., the real election is
>between B and C, then I'd rate B max.
>
>
>I reply:
>
>You’d be a strategic voter, voting Approval 
>strategy. So why do you want intermediate ratings?

Because I was beginning by describing certain 
special cases. Those cases are not the general 
case, and many situations call for intermediate 
votes to be optimal. Further, there is the matter 
of overall social utility. I also happen to 
desire *that* as well, and I do not prefer to 
give myself a small advantage that costs society, overall, more than I gain.

And I trust that this preference is widely enough 
shared that what goes around comes around. In 
another election, I'll be the one getting a large gain at small cost.

(Where Range and Approval differ, generally, we 
have a small preference of the majority vs. a 
large preference of a minority. In my opinion, 
the majority should consent to such a result, and 
Approval does not sufficiently express that 
consent, because of strategic forcing. Consent 
should not be coerced, and election context can 
coerce it, if precautions are not taken.)

(response to be continued)