[EM] Lomax reply, 3/17/12

MIKE OSSIPOFF nkklrp at hotmail.com
Sat Mar 17 12:45:50 PDT 2012


You wrote:

I'd still want to be able to rank my approvals....


Maybe, but realistically, it would be better to just start by proposing Approval,
because, for one thing, there are so many ways to count rankings that there could
always be the question "Why this rank-count instead of one of the innumerable other
ones?", and, for another thing, pretty much all good rank-counts aren't suitable for
handcount. ABucklin without the delay is handcountable, if you're willing to do without
the MMC compliance made possible by the delay that I spoke of.
You wrote:

 If this is the "chicken dilemma," it's been made up. 


Every example is "made up". The chicken dilemma, however, is not made up.
It's real. It will happen.

What this 
situation means (if we interpret it realistically) is that C is 
likely to win. Period. 


Not if the situation is handled well, in Approval. And especially not with AOC,

You wrote:

People are far more alike than you might realize. If A voters betray, 
B voters also betray, they betray equally, more or less. So C wins.


We've discussed a number of ways to avoid that result in Approval and RV.
I'll list at least some of the solutions that we discussed (not counting the
use of a defection-resistant method such as AOC, MMT or GMAT):

1. I suggested that the A voters, given the appearance that they likely are more
numerous than the B voters could say "We're more numerous, so it makes sense for you
to help us win, rather than for us to help you win. So we'll just approve for A, and expect
you to approve A too, in addition to B."

2. Principled refusal. "B is an unacceptable compromise. On principle, we refuse to approve
B. If you really consider A likewise unacceptable on principle, you can refuse to approve A.
But if not, and you refuse to approve A, then C will win because we acted on principle, and 
because you acted on... what?".

3. Consequences in subsequent elections. The problem is automatically solved because
everyone knows that if the B voters defect and the A voters co-operate, then the A voters
won't help the B voters next time. It could start a never-ending spiral of mutual
defection. But, better, it could just result in one subsequent defection by A voters. They
should give the B voters another chance after that.

That's the "tit-for-tat" strategy for co-operation/defection situations. Copy the 
behavior of the other player in the previous round. Tit-for-tat is considered the
best strategy for that situation.

4. It won't be a pure 0-information co-operation/defection game. There will be discussion,
promises, declarations, regarding how the players will play, how the factions will vote.
The A voters will actually have a very good idea what the B voters are going to do. Agreements
can be made before the election by factional or party organizations. There will be a good sense
of whether memberships will go along with those agreements. Conversations, interviews, call-in
shows, interviews, polls, etc. will provide information to the A voters on what the B voters
are going to do.

5. Forest Simpson suggested that, in RV, based on an estimate of the percentages of C voters and
{A,B} voters, the A voters could declare that they're going to give to B, just enough points
to enable B to beat C if B is favorite to more people than A is. They'll say, "We're going to
give to B that many points. If your faction is bigger than B will win. If your faction isn't bigger,
C will win, unless you've done as we're doing: Give to A the points that we're giving to B. In that
way, someone from {A.B} will win either way."

Of course, in Approval that could be done probabilistically: The A voters could each approve B with
a probability such that it's as if each had given to B a certain fraction of an approval.

So the problem can be dealt with even in ordinary Approval. 

>That's the co-operation/defection problem, or the chicken dilemma.

 A false dilemma, that assumes people are playing a game different 
from what they actually play, and that society is as neatly divisible 
into factions like this. Most people won't sweat this at all!


You don't like the word "faction". We could substitute "group of people whose favorite is
the same candidate". "Faction" is just a briefer way of saying that.

I'm not saying that the chicken dilemma will always obtain. But it's certain that it
sometimes will. It doesn't now? A lot of things will be different in Approval, from how
they are in Plurality. It's widely agreed that there can, and sometimes will, be a chicken
dilemma in Approval. When Approval's opponents cite it as a problem, tell them about the
five solutions listed above.

 >If you're an A voter, you'd be glad to hear that you can give a 
>conditional approval to B, an
>approval that is conditional upon reciprocity.

This is doing something with the election process, making it a goal 
in itself..... I'm not thrilled. I'd want to see how the method 
performs in simulations.

But it can be difficult to model strategy. 



You continued:

There is a cost here, the 
cost in canvassing complexity.


Of course AOC, MMT and GMAT are more complicated than ordinary Approval. And
AOC, and probably MMT and GMAT, are not feasibly handcountable. One thing for sure
is that Approval is by far the easiest handcount other than Plurality. 
Because of the need for explanation-simplicity, proposal simplicity, and easiest
handcount, I suggest that ordinary Approval should be proposed first. AOC, MMT or
GMAT could be proposed later, if there's perception of a C/D problem, and if
computer-counts could be guaranteed to be fraud-free.

You wrote:

I'm not convinced I'd approve it.


But you won't need to worry about that for a long time. All of this discussion
about later proposals needn't be considered a problem now. Now, it's just a question
of how to do the first proposal. Now the discussion should be about how to introduce
Approval to people, and answer the objections that some will try to make.

>So, what AOC does isn't complicated to tell. People would understand 
>why they'd like it.


I'm still not convinced I really understand it. I could probably 
explain it, though, i.e., how the counts are modified. What I don't 
get is why this is really necessary.


"Necessary" isn't the word. "Helpful" is the word. It would be nice if the A voters
in the Approval bad-example (ABE) could conditionally approve B.

It's obviously devaluing 
information from the voter


It's letting the voter give more information.

, based on some assumption that... what? 
That voters have not been properly reciprocal? 


...only if the voter chooses to impose that condition.

But that would seem to 
assume that the A>B and B>A preference strengths are the same.


I don't notice how it assumes that. The chicken dilemma assumes that
both A and B voters greatly prefer both A and B to C. If that isn't so,
then it isn't really the chicken dilemma. Sometimes it will be, sometimes
it won't be. I said that usually there won't be a C/D problem. Sometimes
there will.

You said:

I think this algorithm could damage overall social utility. In fact, 
with sincere votes, it's obvious that it *will.*


If, as in the ABE, A is the sincere CW, electing hir won't be bad for SU.

If the B voters are sincerely indifferent between A and C, then A isn't 
sincere CW, and the A voters have no reason to expect the B voters to approve
A, and they probably should give to B an approval that is not conditional.

The question would be whether it balances out the damage from 
strategic voting (which, because the votes are not "maximally 
sincere," does damage S.U.) I'm pretty strongly suspecting, no, it 
causes further damage by removing a strategic voting effect that may not exist.


If the C/D problem doesn't exist, then you needn't make any of your approvals

>By the way, though Bucklin was used with a handcount, ER-Bucklin, 
>with the MMC-preserving delay that I spoke
>of, is incomparably more computation-intensive than ordinary 
>Bucklin, and therefore, almost surely unsuited to
>a handcount. And, without that delay, you lose MMC compliance.

Not sure what you mean. ER-Bucklin can be hand-counted, and was (it 
was often ER in lower ranks than first). 


I myself have often handcounted ER-Bucklin, without the delay.

Your "delay" may well 
introduce problems. I don't know what you mean, in fact.

>Suppose that, at your 3rd rank position, you've ranked 5 candidates. 
>Say that in round N, they get votes from your
>ballot. The delay provision that I speak of (and which is in the 
>electowiki definition of ER-Bucklin) says that
>your votes to your 4th ranked candidates won't be given any sooner 
>than they would be if you'd ranked your 5
>rank-3 candidates in separate consecutive rank positions. In other 
>words, in this example, your 4th ranked
>candidates don't get their votes from you until round N+5.

Gosh, people can make things complicated. Just effing count the 
votes! How in the world did ER-Bucklin become so complex? I, naively, 
assumed that it was *Bucklin* with Equal Ranking allowed. Who tacked 
all this absolutely hopeless crap onto it?


The delay preserves compliance with the Mutual Majority Criterion (MMC).

I don't know who first suggested it. Maybe Chris Benham? I don't know.

What criterion compliance do you think is lost due to the delay?

 >If you'd ranked those candidates in consecutive rank positions, then 
>one of them would get your vote in round N.
>The 2nd would get a vote in round N+1....and the 5th would get your 
>vote in round N+4. So only in round N+5
>would your ballot then give to your next candidate.

I could probably actually understand this if I suspected it were worthwhile!


Whether you consider it worthwhile depends on whether you consider MMC worthwhile.
And that depends on whether you value majority rule. To each their own, of course.

You wrote:

This is utterly damaging to social utility, as I see it. 


That statement needs support.

How does it damage SU, when we protect mutual majorities?

I see 
Bucklin as practically using a Range ballot, with an analytical 
method that slides down the approval cutoff until there is a 
majority. If voters vote sincerely, it's obvious that messing with 
the counting messes with the basic principle.


That depends on what you think the basic principle is. If you agree that
MMC is an important goal of Bucklin, then the delay prevents that goal to
be messed with.

You wrote:

(social utility optimization can violate the majority criterion).


Sure, if you have _sincere_ RV ballots, then RV will hopefully give
better SU than majoritarian methods. But we aren't comparing Bucklin
to sincerely-voted RV. In fact there will be no sincerely-voted RV.

And, absent that, hurting majority rule will lower, not raise, SU.

You wrote:

But, you should know, I dislike overcoming a majority preference 
without the voters being explicity asked if it's okay! 


Then don't take away majority rule by denying the delay that
preserves MMC compliance.

>As I said, that preserves Mutual-Majority-Criterion compliance, but 
>it greatly increases the labor of a handcount,
>almost surely making handcount infeasible.

And it also makes the voters dizzy when they try to understand the 
effect of their vote....


That's one reason why I suggest that the first proposal should be
ordinary Approval. Save the more complicated methods for later. 
Don't propose methods that can't feasibly be handcounted.

This completes my reply to this posting of yours.

Mike Ossipoff

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20120317/37aadb77/attachment-0003.htm>

More information about the Election-Methods mailing list