Seems "Condorcet compliant" is not a warranty of value all by itself.
Looking at an easily findable example I see:
19 voters
6 rank A top - more than any other.
A wins, per BTR-IRV - seems reasonable
Add in 6 more voters ranking A top - total of 12 now
C wins, per BTR-IRV. HUH!!! Look close and C is truly CW
Game is to discard candidates with the least votes on the ballots,
comparing a pair at a time. This will end at the CW if there is one.
Order is based on how many top ranked each, which can be a different
order - and the difference affecting who is last when there is no CW.
If we went for such as this we could expect the same kind of
complaints as we write about IRV.
As to what I suggest, based on the N*N matrix, and, for shortening
discussion, assuming no equal comparisons:
For N-1 races, discard loser in each pair. We thus find at
least a cycle member.
For N-1 races, see if this cycle member is loser to another. If
so we have two members of a cycle and can proceed to complete it; if
not we have the CW.
Dave Ketchum
On Aug 14, 2010, at 9:41 PM, robert bristow-johnson wrote:
Re: [EM] it's been pretty quiet around here...
On Aug 14, 2010, at 6:45 PM, Dave Ketchum wrote:
On Aug 14, 2010, at 2:18 PM, robert bristow-johnson wrote:
the other method, BTR-IRV (which i had never thought of before
before Jameson mentioned it and Kristofer first explained to me
last May), is a Condorcet-compliant IRV method. i wonder how well
or poorly it would work if no CW exists. i am intrigued by this
method since it could still be sold to the IRV crowd (as an IRV
method) and not suffer the manifold consequences that occur when
IRV elects someone else than the CW. does "BTR" stand for "bottom
two runoff"? and who first suggested this method? is it
published anywhere? Jameson first mentioned it here, AFAIK. the
advantage of this method is that is really is no more complicated
to explain than IRV, and it *does* resolve directly to a winner
whether a CW exists or not. i am curious in how, say with a Smith
Set of 3, this method would differ from RP or Schulze.
For Condorcet you have the N*N matrix and precinct summability but,
unlike IRV, you better do nothing that involves going back to look
at any ballots.
i guess you're right. i was just intrigued about this variant of
IRV that is Condorcet compliant. but the actual method should be
precinct summable so that leaves BTR-IRV out.
--
r b-j [email protected]
----
Election-Methods mailing list - see http://electorama.com/em for list info
