View Single Post
  #45  
Old 05-30-2011, 12:01 PM
xodidumdum xodidumdum is offline
Truffle Shuffler
 
Join Date: May 2011
Posts: 5
Default

Bullbound, I understand this, and that's why these measurement were taken over 15000+ data points, and not eyeballing it. And when the testing was done, it was a weapon with pretty low range (like a 180-220) type weapon. Even with that limited range, it still required a tremendous amount of input to get results that were above 97% via f-test modelling.

Quote:
Originally Posted by Charista View Post
Which doesn't lead to any conclusion because that effect is the same regardless of class.

I'm not denying your numbers: I'm denying that your numbers force changes on one class vs. another.

Bb, since some weapons have tight ranges in damage, you can eliminate that effect in testing by having two opponents with tight damage ranges. That removes the effect of the Random Number Generator. Takes some work, but it can be done. Since damage scales to level, it's harder for me to find a 1150-1160 weapon than a L30 finding a 100-100, but that tiny 1% deviation can be ignored most of the time.
Charista, that is why we had so many data points. We did not have a weapon that was THAT tight. It was within 15% or so top to bottom on damage, but still required a tremendous amount of testing (about 800 battles total against the same opponent) to get output that was statistically significant and provable.

Last edited by xodidumdum; 05-30-2011 at 12:04 PM.
Reply With Quote