[EM] the 'who' and the 'what' - trying again

[Michael Allan]

Dave Ketchum wrote:
> I see the Condorcet phantom as going thru the same motions as a real 
> primary or general election, but letting the real elections do the 
> nominating and pay no official attention to results of the phantom.

I see a phantom that will nevertheless have real effects (ghost in the
machine).  You see a test bed or proving grounds for an election
method (machine in the ghost).  Same ghost, different machines.

(But I've interrupted your discussions.)

> That the phantom votes would be shiftable because current counts should be 
> displayed during the voting/polling period does not make consensus exciting 
> to me.
>
> I mentioned cycles because their resolution formulas are a hot topic and a 
> variety of examples could help thinking.

Maybe decision rings could help.  The resolution is slow (depends on
vote shifting), but maybe someone can improve that.  (I needed a slow
and thoughtful process to solve a real world problem, external to the
counting mechanism.)  Just to illustrate, here's a "Condorcet
resolution" by a decision ring:

  0.  A clear Condorcet winner (null case).

  http://zelea.com/project/votorola/d/_/decision-0-stable.png

No need for a resolution with that result.  All 58 voters are in
agreement.

  1.  A Condorcet cycle.

  http://zelea.com/project/votorola/d/_/decision-1-vacuum.png

Call that a "Condorcet cycle" because it's (as you say) a "near tie".
Say the tie includes all those receiving 5+ votes apiece (but ignore
the fiver on the bottom, pretend she's a four).

Two problems with above i) it's not apparent to the voters that
there's a cycle (tie), and ii) if we make it clear and turn up the
decision heat ("hurry up, we're picking the winner now") they may
behave chaotically.  They may pile up on the winner or something, so
the end result is overly sensitive to initial vote shifts.

  2.  A decision ring.

  http://zelea.com/project/votorola/d/_/decision-2-ring.png

So rather than resolve by addition (pile up), we'll resolve by
subtraction.  The tied candidates form into a "decision ring" by
voting for each other.  For your purposes, the system automatically
aligns their votes in this manner.  The resulting whirlpool of assent
(flow volume 52) is equally shared by all members of the ring.  This
formalizes their tie, and signals that it's time for a final decision.

  3.  The first vote shift.

  http://zelea.com/project/votorola/d/_/decision-3-out.png

Votes then shift to resolve the tie.  In this case, the first move was
made by a ring member.  She shifts her vote to another member, and
thus exits the ring.  She's signalling that she's no longer a
contender.

This is only a resolution *framework*.  It leaves much unspecified
(not needed for my own problem).  How do you force the timing?  Maybe
you eject the weakest members of the ring at regular intervals.

What is the resolution criterion?  It's not ring membership (that is
only a signal).  Raph Frank has suggested counting only the votes that
are received from outside of the ring.  In other words, factor out the
cyclic flow that *equalizes* the ring members, and leave them with
what *distinguishes* them.  That's your resolution count.

(All of this applies only to cascade voting.  There are other methods,
and I'm afraid I interrupted your discussion of them.  Please
resume...)

-- 
Michael Allan

Toronto, 647-436-4521
http://zelea.com/