[EM] Classifying 3-cand scenarios. LNHarm methods again. RBJ

[Kevin Venzke]

Hi Robert,

A quick response.

--- En date de : Dim 18.4.10, robert bristow-johnson <rbj at audioimagination.com> a écrit :
> my metric of goodness for an election method is not
> minimizing Bayesean regret but is in minimizing mean voter
> disappointment in the election result.

It's hard for me to imagine much difference between these two goals
on average.

>  that's why we
> don't give the office to the loser (the minority candidate)
> in a simple 2-candidate FPP race - more voters would be
> disappointed with that result than would be disappointed by
> awarding the office to the person with the greater number of
> votes.

You can use "regret" or utility to explain this as well though. With
the knowledge you will have it will be impossible to say that it would
be better to elect the FPP loser.

>  and then the question is, with 3 or more
> candidates, how do we minimize disappointment among the
> electorate?  i think the answer is to consider what all
> the possible 2-candidate permutations are (the
> "contingencies") and to award the office to the only
> consistent majority winner (if such exists).  i think
> that is obvious, because the alternative is to possibly
> award the office to a candidate when there exists another
> candidate that the majority of the electorate agree is a
> better candidate.  to pick anyone other than the CW is
> do disappoint the majority of the electorate.

I like an issue space analysis better than a disappointment analysis,
but we are probably not going to come to very different results.

> > What worries me is the possibility that every time we
> succeed in
> > implementing an election method which can handle any
> number of
> > candidates that we throw at it, we will mostly see
> scenarios with
> > one or two strong candidates and a half-dozen losers
> that never
> > coalesced into anything, so that we mostly will not be
> able to tell
> > the difference in effect from just using FPP.
> 
> i dunno about France, but is that the case in Italy? 
> or Israel?  i thought there were a bunch of countries
> with a half dozen contending parties or more.  it looks
> to me that even the UK has three significant parties.

It's trivial to encourage the existence of a large number of parties 
but it isn't as easy to encourage the existence of more than two options
that the median voter might find interesting.

When the median voter only has two real choices, it seems to me that
the axis along which voters make a decision has been decided before the
election, which I find disappointing.

But as I said originally, if it's not possible to bring in a second
axis, I'm still enthusiastic about methods that can pick the median
voter's candidate when there are a few strong candidates on a single
axis.

Kevin Venzke