[EM] compromise proposal number one: a most (but perhaps too) simple version

[James Green-Armytage]

>
>Here's a first, very simple and perhaps too simple version of a method
>which distinguishes between more and less important binary preferences:

Dear Jobst,
	This proposal reminds me a bit of a version of weighted pairwise which I
proposed on June 19th, where people assign a value to the individual gaps
rather than rating all the candidates on a single scale.
http://lists.electorama.com/pipermail/election-methods-electorama.com/2004-June/013348.html
	The proposals are not identical, but they would be likely to have similar
properties. They differ in that the June 19th version imposes a kind of
second-order transitivity, and gives us a small scale rather than a binary
expression of strength. But the effect would probably be similar. 
	I have generally considered the primary version of weighted pairwise to
be more interesting that the June 19th version, although I'm not totally
down on the latter, and I'm not totally down on your proposal here.
>
>Each voter answers pairwise comparisons with the possible answers "A>>B"
>(strong preference), "A>B" (weak preference), "A=B"
>(equivalence/indifference), "A<B", "A<<B", and "A|B" (abstention).
>	The defeats are determined by taking both weak and strong preferences
>into account, their strengths however are defined via strong preferences
>only.
>	Then some pairwise method like Tideman, river, or beatpath is applied.
>
>So, do you think it is absolutely necessary to consider more than two
>"degrees" of preference?

	Hmm, absolutely necessary? I don't know about absolutely necessary, but I
would prefer it. I think that I would have preference gaps of varied
strengths when voting on most decently-sized sets of candidates, and I'd
like to have an opportunity to express them as such. For example, Dean >
Kerry, Kerry >> Liberman, Kerry >>> Bush.
	Still, I will go on thinking about your proposal here. At least I think
that it is potentially an improvement over our "ordinarily ordinal"
versions of pairwise methods, and it should be able to blunt some of the
more flagrant strategy abuses.
	I'm also quite interested in any further thoughts that you may have about
the other method which you proposed recently (the one I termed
"second-order pairwise"). I consider that one to be more interesting than
this one so far, although again, it is hard to evaluate fully because of
its complexity and unusually high level of abstraction... I'll probably
keep looking at it, though...

my best,
James Green-Armytage