In the outlying areas of two or more color fuzz the mixed colors are effectively getting tied all the time and a random winner among the tied winners chosen, thus random fuzz.
These areas usually occur when the center of the population is more than one standard deviation away from all of the candidates. Given a bunch of bad choices, it's easy for two of them to be relatively close to each other and always on the same side of the mean candidate distance for all the voters. Hopefully those qualitative descriptions of what I think is going on make sense. As for utility function, I tried 1/(dx+dy) instead of 1/sqrt(dx^2 + dy^2) and the result just looked weird to me. (dx+dy) just isn't normal distance in space as I'm used to it. I'll probably soon try linear falloff (1-r) instead of exponential (1/r). On Dec 22, 2006, at 10:18 PM, Warren Smith wrote: > The (now with random tiebreaking) Bolson pictures pretty interesting. > Approval with mean-as-threshold (at least with Bolson's utility > function) > is doing some pretty weird stuff! But "approval with poll" looks > very well behaved, at least in these examples (although I do not think > it'll be that nice always). > > So, I'm worried about approval with mean-base-thresholding. Seems > to me > there are grounds to worry approval voting can get us into trouble > if voters behave that way (and ditto range voting, therefore) and > ought to try to undertsand what is going on there... > > wds > ---- > election-methods mailing list - see http://electorama.com/em for > list info ---- election-methods mailing list - see http://electorama.com/em for list info
