2017 update to the SPI voting algorithm for Board elections [and 1 more messages]

Barak A. Pearlmutter bap at debian.org
Mon Feb 27 22:30:51 UTC 2017

I am very happy to see SPI adopt a voting system designed for
proportional representation.

However I do feel obligated to correct a simple incorrect statement,
namely that STV, a proportional representation system built upon IRV,
is in some sense best-of-breed. STV in fact exhibits a variety of very
serious pathologies, not just in theory but in actual practice. Even
IRV itself exhibits some serious problems, including most troublingly
non-monotonicity---meaning that under some not-unusual circumstances
changing a ballot to rank a particular candidate *higher* can cause
the candidate to go from winning to losing. STV necessarily inherits
these pathologies.

Although poorly formatted, the following case study of an actual
election, the 2009 Burlington Vermont Mayoral election, shows that the
IRV system actually used exhibited about the worst imaginable
pathology: of the three major candidates, it elected the one who lost
head-to-head to each of the other two according to the ballots cast.

Details: http://rangevoting.org/Burlington.html

Burlington subsequently changed its election system away from IRV.

Similar pathologies apparently happened in the IRV 2006 Peru
presidential election and the IRV 1970 Chile presidential election.

Using the Debian leader election data 2001-2005, in one of those five
elections IRV would have given a different winner than the Condorcet
method actually used, even though in all cases the Condorcet winner
beat all other candidates head-to-head. See
http://rangevoting.org/Debian2003.html for details. In another one of
the elections there were a variety of IRV pathologies actually
exhibited. The most interesting of these to me is that there were two
ballots each of which had the property that, had IRV been used,
removing that ballot, which prefers A to B, would have changed the
winner from B to A.

I don't have a dog in this fight. I'm just interested in the math and
in Bayesian methods. These have convinced me that Reweighted Range
Voting (RRV, http://rangevoting.org/RRV.html) and Asset Voting are the
best proportional representation voting systems currently known. And
that the problems with STV are not merely academic but occur
frequently in real elections. Asset Voting does not seem practical for
SPI---although it would be pretty fun. STV (and even IRV) do not come
off very well in any of the careful neutral analyses I've seen.


More information about the Spi-general mailing list