[EM] Improvement to Ranked Robin method

[Kevin Venzke]

Hi Richard,

You wrote to Forest:
> Regarding your desire to protect against clone failures, those are just
> one special kind of "irrelevant alternative," as in IIA (independence of
> irrelevant alternatives).  Personally I'm more concerned about reducing
> all kinds of irrelevant alternatives.  In this case I'm concerned that
> attempts to fully eliminate clone independence failures might have the
> side effect of increasing the rates of other kinds of IIA failures.

This is an interesting idea. It's a little hard for me to conceive of why this
would be the case. But clone independence is a difficult standard to meet, and
based on randomly generated scenarios one would likely conclude that "clones,"
as such, are pretty rare. (This sets aside of course the possibility that clones
could be deliberately created as a nomination strategy.)

There are no doubt other approaches we can take to minimize IIA failures.

I see some correlation, via simulations, between compromise incentive and the
potential for "loser withdrawal" to change the outcome. I guess this is not so
surprising since they're conceptually related: A compromising voter likely wants
to pretend that a (expected) losing candidate didn't enter the race at all, in
hopes of getting a better outcome. It's like an "irrelevant alternative" drops
out, but only from some ballots, and only if a certain kind of change can result
from it.

Kevin