[EM] A distance based method

[Kevin Venzke]

Hi Forest,

--- En date de : Sam 16.7.11, fsimmons at pcc.edu <fsimmons at pcc.edu> a écrit :
> Kevin,
> 
> Do we agree that working from closest pairs outward solves
> the Plurality problem (at the expense of 
> compromise and less monotonicity)?

Not sure. My sim often doesn't find any Plurality failures even when 
they are possible, because the behavior of the voters may never produce
them. (And my sim doesn't identify monotonicity failures at all.)

I never found Plurality failures with "furthest in" and only did one
test with "closest out." Looking back, unfortunately I used rather odd
parameters for the test. If the test can be trusted, it said that
"closest out" had higher compromise, much less compression, much more
truncation, slightly higher burial, and zero (i.e. slightly less)
pushover. Sincere Condorcet efficiency was slightly better. Utility
maximizer efficiency was a bit worse.

I don't think I would trust this at all, except for the basic point 
that in general the methods will look different. But you did guess,
above, that compromise would be higher, and I did see that, so maybe
there's something there.

I'm interested to think about your more recent posts but don't have
time at the moment.

I did stumble upon a seemingly very nice "Single Contest" method. I
tested maybe ten versions. I will explain it later, if I sell myself on
it... 

Specifically:
If I move all the method's strategy to approval cutoff placement, so 
that none of the strategy counts as an insincere type, is that actually 
as good of a thing as it looks on paper? If not, then maybe I'm not
measuring what I want to be...

I cut out Schwartz and it helped a lot. Schwartz is kind of against the
principle I'm after anyway: It inherently has this possibility of 
creating a defeat against one candidate in order to take advantage of 
the status of a completely different defeat.

Kevin Venzke