[EM] Call for Ideas on Automatic Approval Cutoff Finding

[Forest Simmons]

On Sat, 13 Sep 2003, [iso-8859-1] Kevin Venzke wrote:

> What your Max Power method does in the three-candidate, weak centrist
> scenario is a perfect example of this: The low-utility centrist will be embraced
> as soon as a major faction "learns" they can't win.
>

Thanks for this comment.  I'm going to work on modifying the Max Power
collapsing rule to take into account CR differences in order to address
this weakness.

For those just tuning in, Max Power is an example of an automatic approval
cutoff method that continually adjusts the approval cutoff based on
previous results, similar to Repeated Approval Balloting except that
required collapses at each step ensure that the method cannot cycle
forever.

Note that Generalized Bucklin also has required collapses at each step.
In Generalized Bucklin the collapses are always the top two remaining
levels, and the new cutoff is always between the new top two levels.

In Max Power neither the collapses nor the cutoffs are determined so
rigidly: instead the Max Power rules try to approximate what the voter
would do if he were there to choose which pair of adjacent categories to
collapse on his ballot and where to place the new approval cutoff.

Up 'til now it hasn't made any difference in Max Approval if the ballots
were CR ballots or ordinal ballots allowing equal rankings.

But as Kevin points out, unless we use CR values somewhere in the method,
we cannot expect to weed out low utility CW's.

Forest