[EM] The Repoman strikes again

Bart Ingles bartman at netgate.net
Tue Jan 9 00:09:36 PST 2001



LAYTON Craig wrote:
> 
> Mike wrote:
> 
> >The example dramatizes a genuine IRV problem: IRV tends to often
> >jump away from the voter median and choose an extreme. That's why
> >IRV did so poorly in Merrill's social utility simulation. Some here
> >don't like Approval, but both in Merrill's study and Norm's study,
> >Approval did better than IRV. You may not base your judgements on
> >social utility, but the fact remains that those SU studies
> >show IRV doing poorly because of its frequent jumps to extremes.
> 
> I basically accept the SU arguments.  I don't necessarily accept the idea
> that a candidate who recieves a lower SU is necessarily more "extreme", nor
> do I really accept the idea of voters voting in a neat line from left to
> right (or vice versa).  The example is contrived.

I think Merrill's study was really showing IRV's vote-splitting problem,
in which having several candidates clustered together will tend to
benefit candidates who are outside of the cluster.  The studies assumed
normal distributions of candidates, which of course places the cluster
near the center, thereby favoring the more extreme candidates to either
side.

When he reduced the standard deviation (tightened up the cluster) for
candidates but not for voters, the effect was intensified.  The
conclusion was that the candidates who moved toward the center were in
effect punished for doing so.

-Bart



> >So it isn't really correct to say that Demorep's example is meaningless
> >or that any method could be made to look just as bad. The example
> >demonstrates a genuine problem that IRV has, something that IRV
> >really will do fairly often. IRV jumps to extremes.
> 
> Say I pretend to be an IRV advocate.  I give the following example with
> Hitler, Mussolini and Roosevelt as the candidates;
> 
> H>M>R   42
> M>R>H   10
> R>M>H 48
> 
> Roosevelt wins in IRV. Mussolini wins in Condorcet.  It isn't that hard to
> think of some anti-Condorcet arguments to go with it; something about an
> unpopular and irrelevant candidate winning, when the contest should really
> be between the other two.
> &c.



More information about the Election-Methods mailing list