[EM] Rejecting fairness without a spec of reasoning

[Craig Carey]

At 2004-01-07 17:40 -0800 Wednesday, matt at tidalwave.net wrote:
...
>The algorithm would need two passes.  In the first, it would attempt to
>coalesce individual grains in such a way as to minimize variance.  This
>would probably be best done via some sort of genetic algorithm or
>simulated annealing, to try out various options to find out which is
>minimal.
>
>http://www.npac.syr.edu/REU/reu94/ramoldov/proposal/section3_2.html
>

Simulated annealing and genetic algorithms can be too slow and
too dumb for the problem of testing methods (e.g. STV).

STV is checked with right tests that ensure that the method is
fair. Some toxic brew of ideas that are as good as lies some to be the
best possible for pairwise comparing variants. In fact with those,
mere claims that the method passed and omissions on how the testing
regime was fixed, seems to meet and pass the moral integrity standards
that part of the culture of the EM List (dedicated as it is to people
quitting after a few months, with the exception of a few whom it might
be best to not write on).

Simulated annealing uses random numbers so it is dumb.

Checks on fairness involve a computation of the slope of Boolean
surface and that can be found with predicting and bisecting. Once
the position and slow is available then a subroutine in the
algorithm can implement a non-linear optimizer.

Simulated annealing is using random numbers to get around multiple
minima. In testing of STV, multiple minima are mainly absent and the
convergence can be unacceptably bad because the optimizer has a
problem it makes false assumptions about.

The software is needed for fairness testing of future variants of
STV. Unfortunately this Election Methods List is a chat zone on
neverlastingly unfair preferential voting methods. With the culture
of the land hosting the mailing list, the country with the good
leaders, apparently subscribers here can be subjected to (back to back)
years of absolutely no reasoning for that stupid yet seemingly
unreformable problem.

Remarkably it is much worse than saying that the Election Methods
List is mindful of a need to reject the truth of the importance of
the correctness of fairness, but also when these changes are made
their are stil out:

(1) Replace fairness with the set of rules constraining slopes of the
  regions, e.g. the region where candidate B loses. Some some of the
  rules testing preferential voting methods are unfair

(2) Discard the reformed STV idea of perfectly constant rules that
  do not drift freely in a way that nobody would want to memorize.
  I.e. have all the (possibly unfair) differential rules from (1),
  drift inexplicably. Actually it is starting to become obvious that
  once we please Mr Shulze and follow his perfectly unspoken reasoning
  then the acceptability of the method to political parties interested
  in themselves drops away too (in addition to the plausibility to
  public of having an explosion of all-wrong-rules). E.g. monotonicity
  is rejected and then in a 3 candidate race, a person can win with
  only 1% first preference support (Fluffy the Dog).
  We got no explanation on why Fluffy the Dog problems exists.

  The nearest we got to Monotonicity was the October 2003 paper of
  Mr Schulze that censored excessively out the first stage of the
  algorithm. It was full of hoax claims about passes and yet the
  method was not defined since the words "strictly prefer" were not
  defined. The use of that idea would make a method be unfair though
  if Mr Schulze was far more competent than we know him to be, it
  might be possible to reduce the unfairness. However Mr Ossipoff's
  last 2 message drop out wrong tests doubtless so undefined as to
  be perfectly wrong and readers hand around for any retraction that
  also is no more believable since there is no algebra or admission
  that the hoax test was not usable.

(3) Now restrict the number of papers down to 3 or 4 and expect plots
  (unless passing without (3) being used).
  A plot in a triangle can show the orientation of a perpendicular
  vector rapidly. This seems to be a list no voter would have a right
  to ever argue "since that person got that, I must therefore be
  entitled to this, but I didn't get it, so the method is finding the
  wrong winners". That latter type of statement is in the style of a
  fairness argument, and it corresponds to using a polytope bundle
  of rays to expand the B-must-win region. That polytope is simply a
  dual of the other, and the other is of more interest when the
  fairness rules are defined.

  Has anybody seen a plot of a triangle here at this Election Methods
  List since 1998 ?, where the concern was on the slope (not location)
  of lines.

(4) The next steps include one of privately harassing the dim EM-er
  and attempting to get and check the fair grounds and correct logical
  sequential followable reasoning for the failure to make any relevant
  comments when the topic is fairness. Actually by use of (1) and (2)
  we have a much larger collection of rules they are maybe in the style 
  of fairness while not actually protecting fairness.

(5) In fact I might even make it even easier for the Election Methods
  List to pass the easy test: Let's see if any of their brightest
  can make some wooly useless digressions on this topic:

   'that importance is to be allocated to perfectly precise rules that
    have only a regard for infinitesimal comparisons that uphold a right
    of the participants to get what they want when having very little
    influence'

(6) They won't like the word perfectly precise. Now that the Germans of
  Passau University produced a useless QE solver (REDLOG) (useless in this
  topic except (and too factual to test about 100% of Mike Ossipoff's
  ideas), it certainly would be an error of fact to say that wrong
  pairwise comparing methods that carefully materialize the ideals of
  unfairness that is so secret and unremovable to the EM List method designer,
  is there because of some dislike of how hard it would be to actually
  solve the implications of fairness rules. But anyway, the difficulty
  might have been erased by (3). It is not obvious since fairness drags
  in implicitness and Ossipoff and the others seem to have their axioms
  and rules shift and drift according to what is on their mind at the
  moment they write their message for the Election Methods List.

(7) Now I go for the big fish: members of the Election Methods List that
  post up verbiage on unacceptably unpolitical wrong algorithms, can
  free themselves up to merely identify the existence of the topic of this
  message. This fully undoes the expectation that rules testing any type
  *never seen before* preferential voting algorithm be perfectly precise
  in the same way that (All x)[a<F(x)] is precise.

(8) To be rough, there seems to be no issue on details on fairness on 
  which Americans can be stuck for a comment. But of course most of
  the Americans of the Election Methods quit. And a strange selection
  of persons who know nothing at all about the principles and technical
  underpinnings of the design of preferential votings seems what the
  mailing list has ended up with.

---


I saw Forest Simmons' last message and he seemed to have dropped out of
the world of things that are true. No defence of fairness, but even the
minutest defence of fairness would create a noisy complete dissonance
with the theme of this e-mail mesage.

Anyway, these are just some early observations on American research into
justice in the selection of winners (I personally use the word leaders
councillors and mayors, but Americans are bit subject to having their
algorithms perfectly rejected (and who cares if their is any or cigar
ash traces) so I'd use the word "winners".



>Matt responds:
>
>There are algebraic process for obtaining an initial feasible solution
> for a given integer program model.

Adding an unneeded constraint that Integers surely can make things worse.

...
>Ernest Prabhakar wrote:
>
>The second phase - after roughly-equal grains (districts) have been
>created - would be to adjust the boundaries to both minimize
>circumference and decrease variance.   This could probably use a more
...
>Matt responds:
>
>Integer programming is a deterministic optimization method.  Constraints
>to ensure that the districts are contiguous, have no interior holes, and
>collectively cover the entire map are required.


So long as the election officials and/or legal review system runs well
enough (or whatever) then the anti-gerrymandering rule could be designed
to properly create long narrow electorates for long narrow mountain
ranges (e.g. in Chile or South America).


---

A single set of axioms: disputes maybe eventually but not in 2004 for the
topic is all too simple (except later Later-No-Harm upsets that a little
apparently).

A topic for someone who wants to publish a paper is to use the
Later No Harm rule and P2 and whatever and solve the two 3 candidate
problems. It could be a major result and the first since the Kenneth May
2 candidate result of the 1950s.





Craig Carey