User talk:Yamagawa/Drop Rates

What advantage does this method have over just averaging all drops? --  Random Time  15:53, October 17, 2010 (UTC)


 * You get a margin of error at the end...
 * This lets you know if that 5% difference in drops between X and Y is meaningful, or possibly just a variation in the data.
 * At the moment, the steps are s only about 1/3rd complete... I'm working my way back to that final bit of calcualtion...Yamagawa 17:46, October 17, 2010 (UTC)


 * To briefly elaborate... which drops more ectos? Smites or Ataxes?  I've not revisited the numbers recently, but when I had looked, the two were within their error of margin, while the raw figures said smites, statistics called it a draw (read: More detail needed). Yamagawa 22:53, October 17, 2010 (UTC)

Statistics
For my intuition, your bucketing can't give more precision, and you also can't get better error margins than computation based on statistics would already give you, i.e. you should be able to get error margins from probabilities and total sample sizes alone; and the margins get much better when you consider a total than when you consider it broken up in individual buckets. A trivial example would be "no drops in a sample of three" which tells you more than getting three times no drop in a sample of one, which told you nothing each time.

Re: your table:

For a 95% confidence interval:

If I have 29% drop rate, then not making a drop has p=0.71; not making a drop 10 times is 0.7110=. That's a lot lower than 5%. For 50@.93 we get, and 3000@.99879 makes. Correct maximum drop probabilities@95% are %@10, %@50 and %@3000. Please check your computations.

--80.228.215.218 07:41, October 27, 2010 (UTC)