[EM] A possible solution to SNTV vote-splitting

[Kristofer Munsterhjelm]

A note before I reply, about something I didn't clarify: the first 
version of this method, which would not handle excess votes, should use 
plain SNTV counts, not Sainte-Lague. That is, award a seat to the k 
highest votes. This is equivalent when there is no excess - i.e. no 
candidate gets more than one seat - and when there is an excess, using 
Sainte-Lague means we have to find out what to do with it.

-

Raph Frank wrote:

> Ahh Ok, so the change assumes that those outside the group hold their
> votes constant.

For each "round" it does, yes, though the removal of candidates from one 
side may prompt the removal of candidates from the other in the next 
"round" - as with my example where first the excess B candidates are 
removed, and then the excess A candidates are removed.

> Also, "prefer" means according to the cumulative ballot, it is assumed
> to be an honest range ballot?

It is according to the (original, unmodified) cumulative ballot, so that 
the method passes Majority in the single-winner case.

I suppose you could have this as a parameter: if the parameter is on, it 
uses rescaled ballots - if not, it treats them as Range.

>>> CPO-STV says that if your first choice has a seat, then you can't vote
>>> for any of the other seats (subject to surplus transfers)
>> Ordinary STV doesn't go beyond first choices, either, which I think
>> amplifies its chaos.
> 
> I was reading a proposal to use an alternative method to decide the
> elimination ordering.
> 
> PR-STV retains its Droop criterion compliance no matter what order the
> candidates are eliminated, as long as elected candidates are not
> eliminated.
> 
> I think this might have been it (couldn't find an ungated one)
> 
> http://preview.tinyurl.com/y8648yt
> 
> The idea is that you have a loop where elimination ordering affects
> who is subsequently eliminated, which causes chaos, as each stage
> causes a change in later changes.
> 
> Anyway, his idea is to use borda to decide elimination ordering.

I think I've heard of that, and I actually implemented it in my 
simulator. It works, but the results suffer. I haven't checked if it's 
more monotone than ordinary STV, though.

Instead of Borda, you could use any elimination order (for instance, by 
some Condorcet method), as well. That would be an easy way to reduce the 
single-winner instance to your favorite Condorcet method: all candidates 
but your winner ends up eliminated.