Kaxon:
My biggest problem with using uniform random variables as the underpinning of the stats is that the distribution clearly isn't uniform.

Your word uniform I would interpret as non-segmented. By uniform I do not mean voter uniformity (because after all it is voter non-uniformity we are trying to model), but variable distribution. Segmented uniform random variables would be the more complex model; it would still use uniform r.v.s.

CATS is an obvious example

It's so funny how CATS is like the universal non-linearity example when he has yet to demonstrate non-linearity :) It'll probably happen in this contest though, but a lot of funky things should happen here.

while we know there are voters for whom CATS has a strength of 1.0

Indeed, segmentation.

I was thinking a big breakthrough would be if we could figure out the actual distribution of strengths for a character, but I don't know how we could do that with the data provided.

You'd be talking about a strength matrix (sort of like a covariance matrix) - but that is I think n^2/2 variables, whereas we'd only have I think 2n data points. So yeah, more or less impossible, unless you used a lot of guesstimation.

As far as actually predicting anything that's going to happen this year goes... my thoughts haven't gotten that far yet. =)

The utility for predictions here is sketchy, because of the non-linearity once you generalize past 2 characters: any deviations in actual strength from estimated strength will be magnified. If you overestimate a character, the difference between their actual and estimated performances will be greater than with only 2 characters.


King Morgoth:
I don't think Cloud can take almost 55% in a poll like this. Or that Ryu would dive all the way under 15%. It just feels awkward

I've already guaranteed plenty of awkardness in this contest! :) Now I just think that I have some better idea about what direction that awkardness might take.


Haste2:
Link got 43% on Cloud, Sephiroth, and Snake... so, Cloud (who's close to Link) getting 55% on Mega Man, Ryu, and Terra seems reasonable to me.

Indeed. I'd expect Cloud to be in that neighborhood somewhere against those opponents.

I posted some simulated results for the Battle Royale in the Stats topic. Underration of Snake seems to be huge (until it was 3-way), and overestimation of Sephiroth was also an issue. Also note that I was using 1-year old date there, except for Snake (most people would have probably used a higher value for Snake in the simulation that I did).
---
www.gamefaqscontests.com
www.gamefaqscontests.com/gallery