[EM] Unmanipulable majority and Condorcet

[Kristofer Munsterhjelm]

On 05/01/2020 22.58, Forest Simmons wrote:
> 
> 
> On Fri, Jan 3, 2020 at 1:02 PM
> <election-methods-request at lists.electorama.com
> <mailto:election-methods-request at lists.electorama.com>> wrote:
> 
>     Send Election-Methods mailing list submissions to
>             election-methods at lists.electorama.com
>     <mailto:election-methods at lists.electorama.com>
> 
>     To subscribe or unsubscribe via the World Wide Web, visit
>            
>     http://lists.electorama.com/listinfo.cgi/election-methods-electorama.com
> 
>     or, via email, send a message with subject or body 'help' to
>             election-methods-request at lists.electorama.com
>     <mailto:election-methods-request at lists.electorama.com>
> 
>     Kristofer,
> 
> 
> So it seems like UM and Condorcet are almost but not quite compatible
> ... tantalizingly close but "no banana" as they say.
> 
> It reminds me of how tntalizingly close we can get to a method that
> satisfies both the CC and the FBC:
> 
> Consider the method Majority Defeat Disqualification Top (symmetric
> completion). This method satisfies the CC but not the FBC.
> 
>  Now change it so that the symmetric completion does not apply to equal
> top. After this change it becomes MDDT(symmetric completion below top)
> and no longer satisfies the CC but does satisfy the FBC.
> 
> Any Insights?

I can't really analyze non-Condorcet methods with my pairwise approach,
so I can't say much about relaxations that give up (some) CC. I also
don't know if there's some magical nonlinear method that would pass UM
and Condorcet (although I seriously doubt there is).

Do you know of any ways to show that a set of partial differential
inequalities plus boundary conditions has no solution? (Maybe some kind
of trigonometry trick?)

But I can say what my hunches are based on what I've done so far.

I think that Condorcet and UM are most likely incompatible. I also think
that the hard part about Condorcet-complying methods is stitching the
various win regions together (in a transitive manner) so there's no
violation when one passes from one region to another. Compared to this,
making the interior of any given region well-behaved is easy.

More specifically, it's easy to have any two of "transitive,
well-behaved along the boundary, well-behaved on the interior", and it
seems to be easy to make the interior well-behaved with respect to a lot
of criteria at once if that's one of the two you pick, but passing
through a boundary requires a much harder edge condition.

Consider e.g. Smith,Minmax (only complete ballots allowed: no equal rank
or truncation). This would pass mono-add-top in the four-candidate case
if it weren't for counterexamples where adding an A-top ballot gets you
from a region with three Smith set members to four.

I also seem to recall that someone (might be Kevin Venzke) did a
simulation of the frequency of violations of FBC under impartial
culture, and found that Schulze (somehow uniquely among the Condorcet
methods he tested) had a very low rate of incidence. So this might also
suggest that the volume of trajectories given by favorite betrayal and
going from "A wins" through the boundary to "ABCA cycle" ending up at a
failure can be made very small. But my approach doesn't let me construct
methods that minimize the failure rate (over some vote distribution),
just to determine whether there exist methods that have failure rate
zero. And currently only for three candidates with strict rankings :-)