[EM] Being unfair at the very first step: issues for Jobst Heitzig

[Craig Carey]

At 2004-11-28 22:18  Sunday, Diana Galletly wrote:
>Sometimes I wonder why I bother replying to Craig;

Your message was only 2 lines long  and it contained a big error.
These are the facts:
(1) you referred persons at Cambridge, to an article of Mr Shulze
  that wrongly claimed that his method got a pass under the test of
   monotonicity instead of fail.

(2) These people at Cambridge university might soak up the incorrect
  idea. I guess not, but it did that you could return and trash and
  retract your previous view.

Should people at Cambridge be misled (e.g. Ross Anderson) ?.

There may be some balance of forces: i.e. by how much does your
wrong clue mislead ?, is it too small ?, and what is the force and
intellectual tenor of the corrective comment ?. The whole paper is
a sort-of solid mistake all throughout. It was right in getting
monotonicity highly regarded.

Also I say that Mr Schulze should discard all tests and method
checking ideals/rules that are not multiwinner.

Here is the text again:

_____________________________________________________________________________
| From: D.A. Galletly (dag1000 at eng.cam.ac.uk)
| Subject: Re: Cambridge election results
| View: Complete Thread (83 articles)
| Original Format
| Newsgroups: cam.misc
| Date: 2003-05-03 07:22:26 PST
|
|
| In article <slrnbb7cro.uf.ben-public-nospam at bunthorne.i.decadentplace.org.uk>,
| Ben Hutchings  <ben at decadentplace.org.uk> wrote:
| >Condorcet rules for selection for a single post.  But you probably
| >already knew that.
|
| Condorcet plus Cloneproof Schwartz Sequential Dropping, yes.  As I said
| in my speech.
| --
| +           Diana Galletly <dag1000 at eng.cam.ac.uk>          +
| +        http://www.chiark.greenend.org.uk/~galletly/     +

_____________________________________________________________________________


Some error was also made in a speech that you gave.



>four errors in the first two paragraphs that he wrote concerning me,
>I decided to persevere a little further in attempting to decipher his
>message.  I think that this is the crucial bit:
>
>Craig Carey wrote:
>
>> Also, suppose that the ballot papers are all these papers:
>>
>>     a0 * (A)  +
>>     ab * (AB) +
>>     ac * (AC) +
>>     and other papers.
>>
>> Then increasing the "ab" Real number will tend to cause B to lose
>> since appearing with a positive weight in the Heitzig-ian "A over B"
>> total.
>
>And as far as I can see it's total rubbish, because increasing *any*
>of a0, ab, ac etc. is going to cause an increase in the "A over B" total.
>

Which will tend to harm candidate B and hence the entire method could
not possibly get through a fairness checkup. We have finished here, and
Mr G-A's method is failed. Perhaps you can produce an argument that
is more persuasive than repeating my true claim.

Maybe readers have got their imaghination set wrong.
Let them consider this.

Suppose that Mr Green-Arytage was sitting at a desk and working for us
and instead of judging his own method, he was tasked with forming
fast opinions on whether randomly constructed methods were monotonic.

If there is an "ad < ac" term in the B-wins equation and the equation
is known to be properly simplified then in less than 5 seconds Mr G-A
has failed the method. He has the easy task of failing 99 wrong methods
because coefficients are on the wrong side or wrongly weighted.

As soon as own pet method comes along. he can in less than 10 seconds
that it too is unfair, and the topic would be one of coverups, shifting
rules and improper purposes.


>> So it is just shown that all Condorcet variants will be failed by the
>> rule of monotonicity.
>
>Where is it shown?
>  In your paragraph I quote above?  I don't think so.


Why is it the method that you might produce (maybe not Mr G-A's) can get
past a monotonicity rule, BUT 100% of the randomly designed bad methods
that I provide would be failed.

What is your secret ?;. Writing to me through a mailing list whilst
ignoring the whole topic of getting inevitably failed during the
design step ?.


E.g. suppose you offer this:

    (A-Wins) = (3a < b+c+d) AND ... OR ...

Obviously that wil be failed by monotonicity but it might be passed if
the "less than" was replaced with a "greater than".

You tell us what makes you believe that getting a pass under the rule
results from making no attempt to achieve that. 

If we visit while you are designing, we can say that even though you
only have ONLY 5 numbers that you can tweak, you nevertheless have
to shift perhaps HUNDREDS of polytope faces outwards, so that they then
disappear when the simplifier is run over the logic expression saying
when candidate A wins.

I sense that I am only writing something that you should already know.



>I will concede that Condorcet fails participation, which I consider to
>be almost as serious as failing monotonicity, in that statements of the
>form "any later choices you make cannot harm earlier ones" should not
>be made, and voters tend to like such assurances.


I don't know where participation is defined. But the best plan is to
fully ignore the idea.

It would have to be rejected anyway since it is not a multiwinner.
People don't know how to extend the rule to the multiwinner case.
In general we must not say that the given paper gets what it is added
and when number of winners is not 1.




...
>It's a shame that you
>seem to need to insult the people on this mailing list, e.g.
>
>> I see that DW TV periodically has articles on searching for new
>> spouses. Since not getting principles on how to be fair to individuals,
>> maybe Jobst could free up some knowledge on how to pick a partner.
>
>This kind of attack really doesn't increase the credibility of you, or
>your posts, you know ...
>

You produced that theory in an earlier e-mail. You can add details to your
irrelevant topic of me being irrelevant, in your next e-mail, if you
feel like it.

---

I have yet to devise an algorithm that can expand trees of alternating
ANDs and ORs and use the odometer feature to run faster. Has anybody
got some ideas on how to write a fast equation expanding algorithm.

(Exists)[(A or B or C)(D or E)] = Exists[AD or AE or BD or BE or CD or CE],

(Exists)(Max(a1,a2)<x<Min (b1,b2)) = (a1<b1)...etc.

Maybe Mr Heitzig can supply a needed algorithm.

Ms Galletly didn't agree that a retraction was needed.

I find this really intersting: Ms Galletly is ploying through with the
cutting edge being personal beliefs or opinions. If conflicting
personal beliefs are held then it could be even more interesting to
get to see the balance of pressures without the curve-free logic that
perfectly pervades out topic here.





---                                                        Craig Carey