[EM] full rankings, voter desire for

[Kevin Venzke]

Rob,

--- Rob Lanphier <robla at robla.net> a écrit :
> > As Mike said, MMPO satisfies all three of these (and Sincere Favorite).
> > But it fails SDSC (and minimal defense) and Plurality.
> [...]
> >  Failing Plurality is
> > probably not acceptable in a public election, since it makes the winner
> > very hard to justify (i.e. you'd have to explain what positive incentives
> > the method offers, to balance the counter-intuitive winner).
> 
> No argument here.   I'm assuming, though, that Plurality isn't mutually
> exclusive of any of the other three (SFC, LNH, and FBC). 

Right. Schulze(wv) satisfies Plurality and SFC. IRV satisfies Plurality and
LNHarm. Approval (and most of the FBC methods) satisfies Plurality and FBC.

> As a thought exercise for purposes of this conversation, and not really
> as a serious proposal, I'd like to propose "Plurality-patched MMPO".
> The procedure would be as follows:
> 
> 1.  Eliminate all candidates whose selection would violate the Plurality
> criterion
> 2.  Determine the MMPO winner from the remaining candidates.
> 
> I'm going to play around with this myself, and try to understand its
> properties and differences to plain MMPO.  If something immediately
> obvious that's bad about this strikes you, let me know.

Well, it just breaks Later-no-harm. Here's the obvious example:

48 A
26 B
26 C>B

MMPO returns a BC tie (another questionable thing about MMPO). Plurality-
filtered MMPO elects B. But when the B votes are changed to B>C, we are back
to a BC tie, so that LNHarm is violated.

I view LNHarm a lot like FBC: Failing just a little bit isn't much better than
failing by a mile, since the main point is to assure voters that certain kinds
of incentives don't exist.

> A simpler variant of this would be "Majority ranked MMPO":
> 1.  Eliminate all candidates who aren't ranked on a majority of ballots
> 2.  Determine the MMPO winner from the remaining candidates.
> 
> I imagine that this filter causes a LNHarm failure, but I think it also
> points to a slightly weaker variant of LNHarm that may be more useful
> than pure LNHarm.

This is very similar to MAMPO, which is on Electowiki. The definition is:
1. If fewer than one candidate is ranked on a majority of ballots, that candidate
ranked on the most ballots is elected.
2. Disqualify all candidates who are not ranked by a majority.
3. Elect the remaining candidate whose MMPO score (with opposition counted from
*all* candidates) is lowest.

This satisfies FBC, SDSC, SFC, and Plurality. But in my opinion, it doesn't
come very close to satisfying LNHarm.

The methods that come closest to satisfying LNHarm while still satisfying SFC and
SDSC are the CDTT+LNHarm combination methods: CDTT,MMPO, CDTT,FPP, CDTT,IRV, 
CDTT,DSC. These methods still fail Plurality pretty badly, but not as badly
as MMPO, I don't think. They don't fail LNHarm very often. (For LNHarm failures, 
you need four candidates and a three-candidate majority-strength cycle.) At
least in CDTT,MMPO, FBC failures should be very rare (MMPO itself satisfies FBC,
and CDTT creates FBC failures basically as rarely as LNHarm failures).

I don't know of a way to weaken LNHarm which would still result in a guarantee
that voters could "take to the bank."

Kevin Venzke



	

	
		
___________________________________________________________________________ 
Appel audio GRATUIT partout dans le monde avec le nouveau Yahoo! Messenger 
Téléchargez cette version sur http://fr.messenger.yahoo.com