[EM] D2MAC can be much more efficient than Range Voting (corrected)

[Scott Ritchie]

On Thu, 2007-03-08 at 09:50 +0100, Jobst Heitzig wrote:
> Dear Scott,
> 
> you wrote:
> > That seems really lame.  Such a method would certainly fail to elect A
> > even if there was unanimous consent!
> What do you mean by that? Unanimous consent about A like in
> 
> 100%-x: A 100, C 90, B 0
> x:      B 100, C 90, A 0
> 
> with very small x?
> 
> In that case, only somewhat more than x of the A-voters need to bullet-vote
> for A to have almost 100% winning probability. E.g.:
> 
> 100%-3x: favourite A, also approved C
> 2x:      favourite A
> x:       favourite B, also approved C
> 
> Resulting winning probabilities:
> 
> A: (100%-3x)*(100%-x) + 2x   ~ 100%-2x
> B: x*3x                      ~ 0
> C: (100%-3x)*x + x*(100%-3x) ~ 2x
> 
> So in this case it elects almost certainly the favourite of the broad 
> majority
> and only with a very small probability the compromise option.

This isn't how I read the description of the method.  It seemed quite
implied that candidates marked "favorite" are NOT also "approved".

Which, obviously, is strange.

But let us not forget that even random candidate can get better results
than any method some of the time.

Thanks,
Scott Ritchie