[EM] acceptable Chicken Proof methods (was chain climbing)
fsimmons at pcc.edu
Wed May 28 20:25:53 PDT 2014
Any form of pairwise sorting of the approval order (bubble sort, sink sort,
approval margins sort, etc.) is highly chicken resistant as long as the
approval cutoff is explicit as opposed to implicit:
For example, the ballot set
27 A >>B
will elect C under any of those methods, since in the approval order
no adjacent pair is out of order pairwise.
However, in a chicken situation if the A supporters like B so much that
they cannot bring themselves to vote B below the approval cutoff, the B
voters can take advantage of this and get B elected.
So it is not quite as chicken proof as IRV (say), but I think that it is
adequately chicken proof: it adapts to either of the possible sincere
scenarios without requiring any order reversals or collapses ... just a
lowering of B below the approval cutoff in the case of the true preferences
27 A >B
24 B>A (or even B>>A)
On Wed, May 28, 2014 at 2:12 AM, Kristofer Munsterhjelm <
km_elmet at t-online.de> wrote:
> (belated reply)
> On 04/29/2014 12:53 AM, Forest Simmons wrote:
>> you suggested ratings or grading, including MJ as possibilities.
>> But these cannot (at least in their present forms) be chicken proof.
> I suppose this is related to the general proof that monotone score rules
> where every candidate's score is calculated by that candidate's ratings
> alone can't reward non-Approval strategy more than Approval strategy.
> In retrospect, it then doesn't seem that surprising that rules whose
> strategy is Approval-like would also behave like Approval with respect to
> CD (which is a strategy criterion).
> In part, my answer of "grades and/or ratings" to "where can we go from
> Approval?" was intended to be a more general one. If we need more
> information than just rankings and the binary options given by Approval,
> then grades or ratings may provide it. But that does no good, in the CD
> context, if they're poisoned anyway, and I do understand that.
> But could we use ratings or grades to augment a method so that it can
> distinguish settings where the CD-ish ballot is sincere from ones where it
> is not? Well, if so, it can't be like MJ, Range, Approval etc (if there is
> a connection between the proof mentioned above and CD failure). But how
> about something like cardinal pairwise? Could that be adapted to increase
> the resistance?
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Election-Methods