[EM] strategy problem

[Forest Simmons]

Here's another possible solution:

All voters fill out ranked preference ballots.  In addition, each strategy
minded voter with the interest and patience fills out a supplemental
ballot, ranking pairs of candidates thusly:

{A,B}>{A,C}>{B,C}

[In the California gubernatorial race there would be lots of truncation.]

Use a respectable Condorcet method to pick a finalist pair from among
those so ranked.

Then use the standard ranked preference ballots to decide the election
between the two members of the finalist pair.

There could be some insincere rankings of the pairs on the supplemental
ballots, since even the most respectable Condorcet method is somewhat
vulnerable to manipulation, but no voter would have any incentive to rank
the candidates insincerely on their candidate ranking ballots, since these
ballots serve only to choose between the two finalists, and in no way
influence who those finalists will be.

The winner would be sincerely preferred over at least one other candidate.

Can any standard method guarantee this much?

Forest


On Mon, 18 Aug 2003, James Green-Armytage wrote:

>
>
> 	Could it be possible to design a version of Condorcet that is more
> strategy-proof than beatpath or ranked pairs? I admit that it doesn't look
> easy. I note in passing that the Raynaud system (repeatedly eliminate the
> candidate with the greatest defeat) does better in my first example than
> sequential dropping / maximin-based systems, although it doesn't help in
> all such examples, and it offers some fairly nasty strategic possibilities
> in other situations, as well as sometimes producing results that are just
> generally counter-intuitive. It may or may not pass Monroe's NIA criteria.
> 	Other than Raynaud, could there be some kind of system with little extra
> boxes for voters to check, or something like that? A multiple-round system
> of some sort? It's an interesting problem.
>
>
> James
>
> ----
> Election-methods mailing list - see http://electorama.com/em for list info
>