[EM] Range-Voting article--basically right, but a few disagreements

[Brian Olson]

nice post. a couple comments.

On Jan 3, 2005, at 8:19 AM, MIKE OSSIPOFF wrote:

[snip]
> Another person told me that, though Nader is more honest than Kerry, 
> and offers better policies than Kerry, that person would vote only for 
> kerry in Approval. Due to a mistaken impression of how Approval works? 
> Maybe. Or maybe due to the completely misguided progressive anger at 
> Nader for allegedly spoiling the corrupt Democrat. This may sound 
> surprising, but many progressives confuse  stratgegy with candidate 
> merit, and will telll you that Nader is less desirable--and when asked 
> why, can only give you the alleged strategic reasons for not voting 
> Nader. But somehow they begin to perceive that as a matter of 
> candidate merit rather than strategy. "But isn't Nader egotistical" 
> (because takes votes from the corrupt media-anointed candidates)? 
> "He's whacko" (Further questioning reveals that he's whacko because he 
> runs though he isn't viable, not because of his policies).

On confusing "strategy" vs "candidate merit", I remember there being 
many polls from early 2004 saying that Kerry's most important and 
desirable quality was his "electability", a strategy issue.

[snip]
> Smith mentioned Bayesian regret. I've heard of it, but not heard it 
> defined. If I read his definition correctly, it's the SU (social 
> utility) of the SU maximizer minus the SU of the winner. Is that 
> right? One way to minimize it is to prevent high SU candidates from 
> running. Another way is to maximize SU of the winner. Smith says that 
> CR does best at that. But that would only be true under sincere 
> voting, and under the special conditions where the CW doesn't maximize 
> SU.

That's what I keep wanting to understand also. I like talking about 
maximizing social utility. What I learned in probability class about 
"Bayes rule" keeps confusing me when I "Bayesian regret". The 
probability of regret based on the probability of some candidate and 
{mumble mumble}?

A potential difference could be along the lines of, instead of 
maximizing social utility, minimize the square of voter disapproval, so 
that no one is terribly unhappy with the result. Following that goal 
would lead to different results.

[snip]
> Smith wrote:
>
> 6.BAYESIAN REGRET (FOR STATISTICS NERDS):
> Approval voting does quite badly (measured by Bayesian Regret)
> compared to many other more expressive voting systems such
> as Borda and Black, when there are 4 or more candidates and we have 
> honest voters.
>
> I reply:
>
> He means nonstrategizing voters. Strategizing isn't dishonest. Borda 
> isn't going to give us nonstrategic voting. It can't be counted on in 
> any method other than maybe Condorcet.
>
> But if voters are divided between  strategy and sincere rating, maybe 
> CR will do better than Approval by SU, if the Approval strategy is 
> based on incorrect information. I don't expect CR to do better than 
> Condorcet by SU under actual conditions.

My simulations found something similar, except under conditions of 
error (counting error, candidate lies, voter confusion). Under such 
conditions Approval degrades less rapidly than Borda. I didn't test 
Black. These where honest/non-strategic voters.

Brian Olson
http://bolson.org/