[EM] Re: About random election methods

[Andrew Myers]

Eric Gorr wrote:
> Andrew Myers wrote:
> > A lot of Condorcet election methods use randomness to elect
> > a winner, but in a way that I think voters will find unsatisfactory.
> > They simply produce a winner as part of a complex algorithm that
> > uses randomness at various points. MAM is an example of such
> > an algorithm.
> 
> Actually, MAM does not use randomness at various points.
> 
> It only uses randomness at the beginning when it determines the Random 
> Voter Hierarchy (RVH).
> ...
> Unless I am mistaken, the use of randomness is required to pass several 
> desirable criteria of an election method. Passage of these criteria 
> helps minimize the potential for strategic manipulation of the election.
> 
> This is one of the reasons why MAM is currently my favorite Condorcet 
> variant.

Yes, it's when those ties are needed to be broken that I think of as
the "various points". I've implemented MAM as part of CIVS, in fact,
and it's also nearly my favorite Condorcet variant.

Don't get me wrong -- I agree that randomness is needed. I think that it would
make the election method easier to sell if the making of a random choice were
factored out as a separate final phase. If you want to use MAM for real
elections, it seems you have to start out the whole process by the public
generation of the RVH, presumably implemented by a public drawing of random
ballots. If you don't do this, paranoid voters (of which there seem to be a lot
lately) may be worried that you have rerun the election until you got the
result you wanted.

If you could make the random choice the final step, then you mostly wouldn't
have to make such choices at all, because most elections would result in one
candidate with prob=1. The whole process would look a lot simpler and more
transparent to the voters.

-- Andrew