Ok, suppose there are two candidates and three voters, and the voting method is Range Voting
using the scale 0-100. All three voters are completely sincere.
Voters 1 and 2 both prefer candidate A to candidate B, but not by much and they are not very impressed by either. They
vote A27, B25. Voter 3, on the other hand, is infatuated with B and thinks that A is by comparison terrible, so votes
A2, B98. This is how W.D Smith measures "regret" in his simulations.
Except that Smith apparently claims that the voter utilities used in calculating Bayesian regret should be any real numbers (although that is not what I was able to glean from his source code.) It makes no sense to me, but the exchange is on the yahoo group for approval voting, a couple of months back-- maybe someone can show me where I misinterpreted his messages.
I wonder what a reasonable distribution of voter utilities would be if the allowable utilities can be any real number? A linear distribution would be kind of hard to do, especially if you don't inadvertently use your compiler's floating point limits as minimum and maximum bounds.
Bart ---- Election-methods mailing list - see http://electorama.com/em for list info
