[EM] real world 9-winner election using RRV

Jameson Quinn jameson.quinn at gmail.com
Sat Jun 25 15:08:04 PDT 2011


2011/6/25 Kristofer Munsterhjelm <km_elmet at lavabit.com>

> Jameson Quinn wrote:
>
>  Wait.... is that a global randomization, used across all votes? If it
>> is... or in fact, even if it isn't... I suggest you do what Warren
>> suggested, and run it several times, with different random seeds, to see if
>> the results are reasonably stable.
>>
>
> The way my program works, it deals with candidate numbers instead of names.
> The standard way to map numbers to names is to call the first named
> candidate 0, the second candidate 1, and so on. What I did was populate an
> array idx with 0...n, randomly permute it, then the first named candidate is
> idx[0], the second named candidate is idx[1] and so on.
>
> That's a global randomization because the program just sees numbers. If
> it's being unfair based on the numbers, it won't be unfair based on the
> candidate names since the mapping has been randomized.
>
> I could try reseeding many times, but I'd have to find a way of presenting
> the outcome. Say, for instance, that we have a multiwinner situation where
> six candidates are to be elected out of ten. The outcome is so that A and B
> are always in the result, but then the other four are randomly picked from
> [C-J] with every combination equally likely. How would I output that result?
> Printing them all would take a lot of space (8 choose 4 = 70). Should I just
> pick the outcome that happens most often, and break randomly if there's a
> tie?
>
> More interesting would be statistics on how many results you got, how
common the most-common one was, and similar. (How common the
second-most-common one, the median one...)

Actually better would be to randomly break the ties on each ballot before
feeding it into your algorithm. I suspect that this would be more stable
than your global randomization, because there are more potential
permutations and thus a greater chance that they cancel each other out.
20-100 runs should be enough to start to get a feel for how stable the
results are. (Of course, I don't know how easy that change would be in your
program... but if you wanted, I could make and send you a file where I do
those permutations 100 times.)

JQ
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.electorama.com/pipermail/election-methods-electorama.com/attachments/20110625/f0f322e0/attachment-0004.htm>


More information about the Election-Methods mailing list