[EM] Approval strategy from rankings

[Forest Simmons]

On Mon, 5 Jan 2004, Bart Ingles wrote:

> Bart Ingles wrote:
> > The main reason is that, while we have no information about the voters'
> > utilities for each candidate, the voters themselves surely would.
> >
> MIKE OSSIPOFF wrote:
> > They don't. That's the assumption. All I said was that, if a voter doesn't
> > have opinions about rating the candidate, but only has a ranking of the
> > candidates, then that voter should vote for the best half of the candidates.
>
> [BI]
> That was your assumption, and given that interpretation, I don't dispute
> your results.  But that wasn't the only possible interpretation, given
> the data.  I thought it more reasonable to assume that the voters had
> utilities (& therefore approval cutoffs), but that they weren't shown in

If I'm not mistaken, this discussion started with "Testing 1,2,3," a
proposed way of comparing various methods.

Since rankings can be easily inferred from ratings, but not vice-versa, it
seems best to use ratings as the starting place in these comparisons.

A side benefit is that the "social utilities" of the methods can be
estimated, not just the "Condorcet efficiency."


Even better, but much more ambitious, would be to start with distributions
of candidates and voters in issue space, and derive the ratings and
rankings from that information.

Richard Moore has made some forays into that approach.

Forest