[EM] Correlated Instant Borda Runoff, without Borda

[Ken Kuhlman]

It seems to me the choice between most or least correlated candidates
for the tournament seeding should be based on whether you consider
candidate nomination or voter burial strategy of graver concern.  
Using most correlated should help pass Eppley's Independence of
Simular Alternatives (ISA) criterion, and using least correlated
should minimize burial incentive.

I personally am more concerned about nomination strategy, but I
understand that you, Forest, are more concerned about FBC.  I think
there's room for us to disagree on this.

As far as monotonicity goes, your concern is well-founded.  It would
be interesting to look into this further to find an example, see how
extensive the potential problem is, and if it could be resolved with
the addition of a repechage stage.


As far as names go, I've been calling this method "Correlated Instant
Condercet Runoff (CICR),"  but I believe Chris Benham came up with the
variant first, and Dan Bishop wrote it up publicly for the first time,
so if either of them have a prefered name, I'll defer to them.

-Ken

Ref:
http://alumnus.caltech.edu/~seppley/Independence%20from%20Similar%20Alternatives.htm


On 12/23/05, Simmons, Forest <simmonfo at up.edu> wrote:
>  Great ideas in a much neglected area!
>
> A couple of comments:
>
> 1.  It seems to me that it is better to start by eliminating the pairwise loser of the least (as opposed to most) correlated pair of candidates.  This reduces burial incentive.
>
> 2.  These kinds of methods tend to lack monotonicity because increasing support for a winner can change the correlations in such a way that the winner faces an unfavorable pairwise contest that didn't materialized before.
>
> 3.  To overcome the monotonicity problem, the correlation data could be obtained separately from the rankings.  However, this tends to open up opportunities for manipulations of the correlations, since they are not tied to the rankings.
>
> Taking into account (1), (2) and (3) I've come up with the following idea, which I call "Narrowing In:"
>
> (A)  Have the candidates fill out extensive questionnaires with a wide variety of questions related to a wide variety of issues.
>
> (B)  Publish their responses, as well as the correlations between the candidates based on their responses.
>
> (C)  Have the voters rank the candidates.
>
> (D)  While there remain two or more candidates, eliminate the pairwise loser of the least correlated pair.
>
> Remarks:
>
> Note that if issue space turns out to be essentially one dimensional, the method starts eliminating candidates from the outside, narrowing in on the Condorcet winner.
>
> Because the candidates' responses to the questionnaire are published before the vote, they have no (unusual) incentive to lie about their position on the issues.
>
> This method is monotone.  It has little incentive for favorite betrayal since Favorite and Compromise tend to be highly correlated, so the decision between them tends to come late in the game, if at all.
>
>  In fact, the only time there could be a favorite betrayal incentive is if Compromise and Favorite formed the least correlated pair, while there still remained at least one other candidate to be eliminated.
>
> Even then there would be no betrayal incentive if the pairwise winner of the pair had has great a chance against the other remaining candidate(s) as the pairwise loser of the pair.
>
> What do you think?
>
> Forest
>
> ----
> election-methods mailing list - see http://electorama.com/em for list info
>
>
>