[EM] Hey guys, look at this...

[Kristofer Munsterhjelm]

On 2/22/23 04:39, Forest Simmons wrote:
> Why have the STAR folks stubbornly dug in their heels with a clone 
> dependent version when one simple change would make it both Condorcet 
> efficient and clone-independent?

And for those who like a theoretical challenge, there's still the design 
question of making a method that takes von-Neumann-Morgenstern utilities 
to their logical conclusion while remaining cloneproof and dealing with 
Range's strategy problems.

In a three-candidate election, we don't know if a particular voter's 
9/10 is the same as another voter's 9/10. But we can infer lottery 
probabilities. So each voter's ballot can be normalized to (1, a, 0) if 
we use l_infinity normalization.

So we could do just a normalized Range, but this has undesirable 
strategic consequences - the usual Burr dilemma, minmaxing, etc. With 
different normalization types (e.g. l_2 cumulative voting) that's 
mitigated but at the cost of it no longer being cloneproof (vote-splitting).

I thought that perhaps something with the logic of a Condorcet method 
would work better, with, instead of pairwise contests, there would be 
pairwise triples: the minimum necessary to capture strength of 
preferences. But I just couldn't make it cloneproof.

-km