[EM] possible improved IRV method

[Allen Pulsifer]

Hello List Members,

I have a thought on a possibly improved IRV method that I would like to run
by the list.

The goal of this method is to eliminate some of the bizarre behavior that
IRV can exhibit, particularly the property that lowering a candidate's rank
on some ballots can lead to that candidate winning the election.  It is also
hoped this method would pass the Condorcet criteria, and possibly other
criteria that traditional IRV does not pass.

I call the method "Sequential Advancement".  It is based on IRV, but with a
different method of determining which candidate to eliminate.  Under this
method, the first step would be to determine if there is an IRV winner:

Test for Winner:
For each ballot, one point is awarded to the highest-ranked candidate.  If a
ballot has two or more candidates tied for the highest rank, each of these
candidates receives an equal fraction of a point.  If a ballot has no
candidate ranked, no point is awarded.  After counting all ballots, if the
point total for any candidate is greater than 50% of the total points, that
candidate wins the election.

Sequential Advancement:
If there is no winner, one candidate is eliminated, as follows:  The
candidate with the largest point total computed above automatically advances
to the next round.  The point totals are then cleared, and the ballots are
recounted, awarding one point to the highest-ranked remaining candidate on
each ballot and fractional points in the case of equal highest rank.  Again,
the candidate with the highest point total advances to the next round.  This
process of clearing the point totals, counting the ballots and advancing one
candidate continues until all candidates but one have advanced to the next
round.  At that point the last unadvanced candidate is eliminated.

Iteration:
After eliminating one candidate, the ballots are tested again to see if
there is as a winner, this time considering only the highest-ranked
candidate who had not been eliminated.  If not, the process of eliminating
one candidate continues until there is a winner.

As an alternative, the IRV test for a winner could be ignored, and instead
the elimination process used until only one candidate remains.  That
candidate is declared the winner.  It is not clear if there are any
situations in which this would result in a different outcome.

This method of elimination would seem to pass the Condorcet criteria,
because in order to be eliminated, a candidate would have to lose a
pair-wise contest against all remaining candidates.  It might also meet the
monotonic criteria and be less-susceptible to strategic voting.

Has anyone ever seen a proposal or analysis of a method similar to this?

Thank you,

Allen Pulsifer
pulsifer3 /at/ comcast /dot/ net