[EM] a similar method to weighted pairwise

[James Green-Armytage]

Brian wrote:
>My first reaction is that this won't actually break ties. A majority
>would 
>like two choices, vote them up, the tie would be between them and they 
>wound up approved on almost all of the ballots. The "almost" in that 
>leaves the tie breaking up to a scarce few who happen to draw the line 
>between those two choices, OR widespread approval-gaming trying to guess 
>what the condorcet tie might be and how to approval strategize for that 
>set.

	I'm not sure whether you're talking about pairwise ties or majority rule
cycles. The answer to that question should enable me to respond to this
paragraph. I've tried to get it from the context, but without success.
	If you're talking about pairwise ties, then this method isn't
specifically an attempt to resolve them. It's an attempt to resolve
majority rule cycles. You can use whatever rule you like for dealing with
pairwise ties in conjunction with this method.
	If you're talking about majority rule cycles, then I think that this
method should be quite decisive in resolving them. Let's assume that one
of these two candidates beats the other in pairwise comparison. Let's say
that Dean beats Kerry, as in my example, but most voters who approve Dean
also approve Kerry. Actually, that's good, important information for my
method, exactly what my method is trying to find out. If only a few Dean
voters don't approve Kerry, then it specifically suggests that the
Dean-->Kerry beat isn't very important to people. Hence it is probably the
defeat that will be dropped if there is a majority rule cycle.
	Anyway, the approval cutoff version isn't my main proposal, the ratings
version is. The approval cutoff version is just an afterthought.
>
>Also, the Ranked+Rated Condorcet plus Rated Tie Breaker could be 
>simplified by just casting a rated ballot. Since you made the rule that 
>the rankings must coincide with the relative ratings the rankings can be 
>non-ambiguously derived from the ratings (including equal rating mapping 
>to equal ranking). 

	Actually, that isn't quite the rule which I specified in my proposal,
although I thought about it. My rule says that, for two candidates A and
B, you *can* give them the same rating and still rank A over B. Actually
most of the voters in my example did that, for example, Dean > Kerry >
Bush voters who gave 100 to both Dean and Kerry but still specified a
preference for Dean.
	On the other hand, if you give A and B equal rankings, then you *have* to
give them equal ratings as well. I didn't specify that exactly, but that's
what I meant to say. I should amend my proposal to make that clear. Also,
you can't give A a higher rating but B a higher ranking. That's just not
cool.

>I generally like it, I think I'll code it up and add it 
>to the Election Calculator.

	Thank you. Please let me know if you have any clarification questions as
to how the method should work.

my best,
James Green-Armytage