[EM] Dave on approval, ranked ballots

Andrew Myers andru at cs.cornell.edu
Wed Jul 27 16:45:57 PDT 2005


On Wed, Jul 27, 2005 at 09:31:33PM +0200, Jobst Heitzig wrote:
> Can anybody cite a study showing cycles would be rare in "real"
> elections with many candidates and truely ranked ballots (not 90% bullet
> votes because of lazy voters)? This claim comes up again and again and
> it seems to me that there is no evidence for this. At least my
> simulations showed that when there is a set of closely tied candidates
> and individual preferences are at least mildly independent then cycles
> occur more and *more* frequently as the number of these candidates grows...

This isn't a formal study, but experience with CIVS suggests cycles are
usually not a problem. See, for example, the following reasonably large
public elections chosen pretty much at random from past history:

http://www5.cs.cornell.edu/cgi-bin/andru/civs/results?id=E_7f2829a7cfa6c857
http://www5.cs.cornell.edu/cgi-bin/andru/civs/results?id=E_2ecaeb874ed56bac
http://www5.cs.cornell.edu/cgi-bin/andru/civs/results?id=E_c27be1d900679a9c
http://www5.cs.cornell.edu/cgi-bin/andru/civs/results?id=E_a579b65f910f1560
http://www5.cs.cornell.edu/cgi-bin/andru/civs/results?id=E_6fdd6b4b65251159
http://www5.cs.cornell.edu/cgi-bin/andru/civs/results?id=E_8524af2f699df817

Of these, only the last contains any cycles of note. For the most part,
the ranked ballots are pretty good at achieving a complete ordering of
the alternatives that does not depend on the completion method chosen.

-- Andrew



More information about the Election-Methods mailing list