[EM] Copeland//Plurality --- can it beat IRV?

[Richard Lung]

The most rational method of counting preference votes is by keep value, or the ratio of the quota to a candidates votes. Exactly the same count can be conducted for the exclusion of candidates as for their election, simply by reversing the preference vote into an unpreference vote.
The exclusion count of unpreferences is inverted for an auxiliary election count. The two counts are averaged, by geometric mean, for the final keep values to the candidates.

Tha advantages of this binomial stv is that it uses all the preference information; does so with rational accuracy; is a simple repetition of a rational count for exclusion, as well as election. 
Unlike IRV or the alternative vote, it is not stuck with an approximate ordinal Last Past The Post count, no better than the ordinal First Past The Post, which it is trying to replace.

Richard Lung.




On 22 Jan 2022, at 11:10 pm, Kristofer Munsterhjelm <km_elmet at t-online.de> wrote:

> On 22.01.2022 21:33, Daniel Carrera wrote:
> Hi guys,
> 
> As you know, like many of you I am dismayed that so many election reform
> advocates are promoting IRV, apparently thinking that it is the only or
> the best alternative to FPTP. So once again I'm trying to think of
> methods that might appeal to an existing IRV advocate, to see if I can
> get them to ditch IRV and pick something that is actually good. In the
> past you've seen me ask about BTR-IRV, Benham, Smith[//,,]IRV, etc. So
> let me present another idea:
> 
> What about Copeland//Plurality?
> 
> 1) The method is incredibly easy:  "Among the candidates that win the
> most head-to-head matches, pick the one with the most first votes."
> (yes, I'm setting it so the score for ties is zero)
> 
> 2) I'm pretty sure that the method is Smith efficient and monotone.

It probably fails monotonicity, because it's X//Y and it's not Borda.
The thing that makes Copeland//Borda monotone is that the pairwise
matrix doesn't change when you remove candidates. But the number of
first preferences for remaining candidates *does* change when candidates
are eliminated.

I think the scenario would be something like this: Suppose we have an
A>B>C>D>A cycle. Some voters uprank A which turn the D>A leg into A>D,
so that D is booted from the Copeland set. D-first voters tend to vote
for B second, and A has the most first preferences with B in second
place. After D is booted out, the revealed first preferences for B make
B win instead of A; so raising A made A lose.

I would also guess that it would be pretty susceptible to strategy. It
isn't cloneproof either because neither Copeland nor Plurality is.

> 3) For an IRV advocate it might have better intuitive appeal than other
> alternatives because it has that Plurality component that some of them
> want. I've read IRV advocates say that IRV > Condorcet because the
> winning candidate should have strong 1st-place support. I can't imagine
> any reason why that would possibly be true, but it means that an
> X//Plurality method might appeal to them.

I always find the core support argument kind of weird. If core support
is the end-all be-all, then Plurality's your man. But clearly it isn't.
On the other hand, IRV can elect a candidate who has only two first
preferences, so the worst case (lack of) reliance on core support can be
pretty bad -- and IRV can fail to elect the CW even when the CW has more
initial first preferences than the IRV winner. So it doesn't really hold up.

-km
----
Election-Methods mailing list - see https://electorama.com/em for list info