[EM] orphaned voting method

[Alex Small]

James Green-Armytage said:
> 1. Ranked Ballots.
> 2. Count top choice vote totals.
> 3. Hold a pairwise comparison between the two candidates with the lowest
> top choice vote total.
> 4. Eliminate the loser of this pairwise competition.
> 5. Continue until no more candidates can be eliminated. (One candidate
> remaining, or a set of tied candidates.)
>
> 	This method is Condorcet efficient, in that it will always select a
> Condorcet winner, and never select a Condorcet loser.

Not only that, it will always select its winner from the inner-most
unbeaten set.  Say that the two candidates with the fewest top-choice
votes are A and B.  Say A is in the inner-most unbeaten set and B isn't. 
Then A beats B pairwise, so A is retained and B is removed.

Now, one could modify this method further and transfer votes from removed
candidates so that step 3 is modified to:

3. Hold a pairwise comparison between the two candidates who are the top
choices (among those candidates remaining) of the fewest voters.


This doesn't matter in the case of 3 candidates, nor does it matter in the
case of 4+ candidates when there is a Condorcet winner.  But when there
are 4+ candidates and a cycle (which might not include all of the
candidates) then it matters.

In the interests of simplicity (especially in vote tabulation) the
transfer could be eliminated.  But if the goal were to strike a compromise
with IRV supporters (e.g. if we ever found ourselves in a dialogue with an
IRV advocacy group that was willing to consider IRV-flavored Condorcet
methods) the transfers might be worth including.



Alex