[EM] Realism of Tideman vs Schulze numerical testing

[Craig Carey]

To "Norman Petry" <npetry at accesscomm.ca>. I don't feel like sending
this to the Election Methods List (the topic is a seeming dud: which
of Tideman and Schulze is best, and there are over 40 subscribers
in that list and I don't want to waste their time on such an
unsound or uncertain topic).


I'd have a bit of an interest in having those repaired Condorcet
methods compared according to some measure of how truncation
resistant they are. I am not particularly interested in you doing
anything at all there since I prefer methods that perfectly pass
a truncation resistance test. I have not heard of any attempt to
create a repaired Condorcet style method that is perfect under
that test.

That is roughly how I would decide whether Tideman is better or
worse than Schulze. They can't be compared with STV though, since
STV is perfectly compliant under that test. The absence of
principle at the EM list is quite marked. I guess that the chances
are high that both Tideman and Schulze can trivially be show to
fail those 2 tests. Can you find examples?. I started out this
paragraph by saying you could quantify their failures, but to
have them fail BOTH properties when one or other is achievable,
raises doubts as to what exactly the humans designing them were
up to. Maybe they also fail my P2 property also (winner set
unaltered on adding papers (+k:A, -1:AB, -1:AC, -1:AD, ...)).
Who doubts the designers ignored principles. Downstream that leads
to the methods being invisible when the world of the desirable is
scanned for good methods. Both one needs to use the right vision
and get to see the fixed Condorcet methods at the same time. The
methods that sweep up wrongness into the Condorcet area, rather
say, try to keep it small in overall intensity. What is written
of here is unclear.

I guess all testing could be stopped and some list like the EM list
could be started where discussions on the method could continue in
the understanding that no resolution on which is better would occur.

These methods still do not pass the test of finding the right
number of winners in 2 candidate elections, if I understand?. Maybe
a decision algorithm to pick the best of Tideman and Schulze, should
be to pick the method that finds the right number of winners in
elections with 0,2,3,4,... winners. It is like a topic that can only
be considered in primmer school: methods that return no results.
I don't really know if the Tideman and Schulze methods are of such a
low quality that they fail to solve 2 winner elections with 2 or
more candidates. Maybe youe Schulz and Tideman code correctly solves
the 2 winner 2 candidate election but the method designers would
assert the code is buggy. I suppose the designers should be held to
be right.

I proved that the repaired Condorcet methods can't be both truncation
resistant and monotonic because of an argument using just the
papers (AB,A,B,C). I.e. the Condorcet winner is undesirable.
You wrote, make it an axioms and I proved that it couldn't be because
it would contradict with P1 (which is monotoncity + truncation
resistance). I have trouble upholding interest in the aging jellyfish
principles guarded with impassable privacy, that pervades the RL
Election Methods and its dogged aspirations to decay gracefully like
some communist party, hardly appearing to search for simple fact.

I will be (am) coding methods and testing subroutines in Ada 95:
ftp://ftp.cs.nyu.edu/pub/gnat/

There is Meek STV code here that I will soon translate to Ada 95:
http://www.bcs.org.uk/election/meek/meekm.htm