IRV mailing party adventures

[Bart Ingles]

Martin Harper wrote:
> 
> 3 candidates, 100 voters, IRV, ~20,000 tests
> Random electorate: in around 2-3% of elections.
> Two-dimensional electorate with a gaussian distribution: less than 1%
> 
> That is, in that percentage of the elections, there will be a Condorcet winner,
> and IRV will elect a different candidate.
> 
> IRV spec: draws broken according to which of the two had the higher number of
> votes in the last round. Low series detected and eliminated all at once, which
> slightly changes the characteristics with this method of draw resolution.
> 
> 2-d electorate spec: both candidates and voters have the same independant random
> gaussian distribution in a 2-d space. voters vote for candidates according to how
> close they are, and on no other criteria.


That's pretty close to Merrill's published results.  Why not try running
the same trial, but with the distribution of candidates having a
standard deviation of about 1/2 that of the voters' S.D.

-Bart