[EM] Defensive strategy for Condorcet methods JQ

[Kevin Venzke]

Hi Jameson,

--- En date de : Mar 14.6.11, Jameson Quinn <jameson.quinn at gmail.com> a écrit :
[start quote]
Of course, this causes favorite betrayal strategy, because you may 
care more about giving a penalty than about helping your honest 
favorite. And this strategy is "obvious" enough that I think people 
would overuse it, even when there was an honest CW (for instance, 
Nader voters in a Nader/Gore/Bush scenario).

One way to avoid such "overfitting" (solving one problem but 
causing another) is to have a runoff between the winners of two 
different methods, if they differ. For instance, minimax and FPC. 
Of course, that throws simplicity entirely out the window.
[end quote]

I am torn on the runoffs. Experimentally they seem to be very effective
at selecting the sincere CW winner, no doubt because they can go to the
voters twice, with the second time having sincerity assured, so the
voters can correct errors of the first round. They can avert disasters.

But, again experimentally, they seem to have some of the worst 
insincerity in the first round. While this might be acceptable on
balance, it makes me think that it is a waste to use methods of much
complexity there.

Kevin Venzke