Well, my results are still only about 90% accurate, but when off they're always close...
You can use a million different models, and get a few different data points that are wrong. Nothing is perfect, but I think the conclusions are quite clear:
For the blocking bar, something along the lines of...
0.2*STR + 0.42*BLK + 0.1*AGI + 0.1*CONF + 0.07*VIS + 3
or
0.21*STR + 0.46*BLK + 0.1*AGI + 0.1*CONF + 0.07*VIS
work really well. The latter just spreads that 3 across strength and blocking..
Thus it takes 15 strength to move the bar, 6-7 blocking to move the bar, 30 agility, 30 conf, over 40 vision.
When I look at the overall bars, I'm seeing no major differences across positions. What I'm finding is something like...
Strength 0.171448042
Blocking 0.142170238
Agility 0.184115072
Vision 0.090454058
Confidence 0.180340944
Tackling 0.057557981
Stamina 0.180880099
I really don't like tackling in there. But whatever, the point to make about the overall bar is that it's very even. Strength, blocking, agility, confidence and stamina all move it well. Vision isn't as important.
Speed actually showed up if I took tackling out..
Strength 0.189509129
Blocking 0.150404004
Agility 0.166137928
Vision 0.058986864
Confidence 0.161627095
Stamina 0.163147963
Speed 0.095809368
Once again, str/blk/agi/conf/sta are about the same, with vision far behind.
I'm not satisfied with the overall results.
What's bugging me most is that the formula isn't exact, nor is it changing whatsoever as I add more data. If it is using the 8-10s in the stuff I'm not inputting, this is all constant, thus would be absorbed into the "+3" at the end, which didn't solve anything.
The other possibility is that height and/or weight come into play, or that the linear model is simple incorrect.
However the relationships that come up I think are, for the most part, correct.
If you have any other model propositions, let me know.
A few ideas would be: actual skill point value (including caps...), including height/weight/level.