[EM] Introduction (cont.)

[Forest Simmons]

On Fri, 3 Aug 2001, Douglas Greene wrote in part:

> I currently tend to favor AV, for all the reasons you're familiar
> with.  Besides the ones I've seen mentioned, it's also possible in
> some cases to implement it with no changes in statute.
> 
> But I was surprised to see that no one had brought up range voting.  I
> was first tipped off to this system through a post from (of all
> people) the CVD's Rob Richie.
> 
> Warren D. Smith of the NEC Research Institute modeled 120 or so voting
> systems with both sincere and strategic voters.  He found that range
> voting produced the least Bayesian regret as compared to all the other
> systems.  His work is available at:
> 

This is not surprising in the sincere case.

It's always nice to hear that the strategic case of Range voting optimizes
some desirable quantity, since the strategic case of Range voting is
equivalent to Approval. 

So the strategic case result raises some interesting questions.

Was strategy calculated from "perfect information"?  If so, what kind of
"perfect information"?

How was optimum strategy calculated? Was it something equivalent to
Cranor's method?

Did the author (Smith) notice that all of the range ballots in the
strategic case assigned points only at the extremes of the range? 

When Smith announced that range voting produced the least "Bayesian
regret" compared to other systems, did he include Approval in the "other"
category? Or did he notice that coincidentally Approval scored identically
to Range voting?

Forest