[EM] Will to Compromise

[Jobst Heitzig]

Dear Raph,

you answered to Greg:
> > 1) Using Bayesian utility, randomness is worse than FPTP.
> >
> > This is a pretty powerful indict, depending on how often the method
> > has to resort to random ballot.
> 
> Hmm, I am not sure how true that is.  The randomness in those
> simulations is picking a random candidate.
> 
> Random ballot should be superior to random candidate.

That clarifies the claim a bit. Of course, picking each option with equal probabilty is crap since it doesn't use any preference information at all. That's obvious. Random Ballot is very different from that. The simulations I will present this weekend show that Random Ballot usually performs quite well from the social utility perspective.

> It isn't entirely.  There randomness creates an incentive to approve
> compromise candidates.  This means that it isn't like pure approval.
> A 55% bloc that refuses to compromise and thus wins the approval
> stage, will likely end up causing a compromise failure.  That is
> completely different to an approval election where a 55% bloc can
> guarantee a win.

Absolutely!

> I think that finding an acceptable compromise is an important point.
> The specific method is separate from the concept that you can allow
> voters to in effect trade their winning probability.

Although that might be interesting too! One could imagine a method in which voters delegate their decision power to a small number of representatives who will then negotiate and trade the winning probabilities they represent. 

> Strategy needs to be tested.  The example that was used was a 3
> candidate race, finding a compromise is harder when there is more
> candidates.

Yes, definitely. I hope someone will help with that analysis!

Yours, Jobst