[EM] Trying to out-do... a result! pt 2

Kevin Venzke stepjak at yahoo.fr
Mon Feb 21 10:11:28 PST 2011


Hi Kristofer,

--- En date de : Lun 21.2.11, Kristofer Munsterhjelm <km-elmet at broadpark.no> a écrit :
> > Now, if only we could use something more
> clone-friendly than first
> > preferences... The tricky thing is that most other
> metrics (such as
> > worst pairwise loss) are already vulnerable to
> burial.
> 
> It would be a very strange system, but perhaps IRV. Define
> the ordering so that if X is eliminated before Y, Y is
> ranked ahead of X. IRV itself is cloneproof, which Plurality
> isn't, so the result should at least be more
> clone-friendly.

That seems possible. As clones are eliminated they would boost their
fellow clones in the order.

> I'm not sure, but I have the feeling that the Plurality
> variant would not be monotone. The IRV variant wouldn't be,
> either.

I think that if you don't do the Condorcet or Smith filter at the 
beginning, the Plurality version is monotone.

Superficially this method reminds me of QR, which isn't monotone because
changing your first preference order will affect which candidates you
are compared to. But in this method (which I'll call KH, C//KH, Smith//KH,
and Smith,KH for various versions) you are always being compared to A.
So if you're going to win the election, it is to your advantage to face
A as soon as possible.

Maybe there is a doubt about whether A could want fewer first preferences?
I don't think that is likely... If A drops out of first place then B
becomes the new FPW. For this to be profitable A must have a majority over
B. But in that case there's nothing to be gained by losing first prefs
since a majority over B is decisive whether A is one above or somewhere
below B.

Incidentally, I was surprised to realize that KH and C//KH are almost
the same, DNA-wise... I wrote a method plotter that operates on the DNA
and am thinking about making a little atlas (in ASCII).

Thanks for your thoughts.

Kevin


      



More information about the Election-Methods mailing list