Heterotic said:
As I am a new player some of my questions/comments might be silly. However, I am quite interested in the game mechanics, so I will ask them anyway. I was really surprised to find out that a linear dependence fits well (i would have expected a $x/(a + x)$ type. The linear dependence $y=a+bx$ being confirmed by wiz, what remains to be done is:
1. Find an explanation for the plateaus observed which
There are no plateaus observed of any significance.
What is seen is analagous to flipping a coin twice, and saying "two heads! This coin must always land heads!"
If we have evidence for plateaus from those numbers, then any coin that lands head twice will always land heads.
1a. can be a result of rounding (e.g. $y=c(a+[bx])$)
1b. can be a result of correlations in the samples (I might have missed it, but were all your results coming from the same mob, or the same mob type changing the mob a few times?)
So neither of these are required. The curve could be nice and simple and sloapy (44% - 1%/5 cha) with no rounding no coorelations and seeing the results displayed would not be unexpected.
Rule of thumb:
If you are measuring a percent chance using X samples (at least 100*), your results will be 19 times out of 20, within:
+ or - 100%/sqrt(X)
percent.
* before 100 samples, this simple relationship won't always hold.
So:
100 samples, 35 resists: 35% +/- 10% resist rate
400 samples, 92 resists: 23% +/- 5% resist rate
1111 samples, 300 resists: 27% +/- 3% resist rate
2500 samples, 700 resists: 28% +/- 2% resist rate
4444 samples, 1200 resists: 27% +/- 1.5% resist rate
10000 samples, 2600 resists, 26% +/- 1% resist rate
etc.
Remember the +/- value will hold 19 times out of 20.
