Quote:
Originally Posted by seped
[You must be logged in to view images. Log in or Register.]
Indeed, but the downside of getting a bad approximation is nothing other then getting something unhelpful like "Item drops 4% (+-4%)" which is still useful to know that the wiki doesn't have enough data to say anything more then "this item probably has a drop rate below 8%"
|
It seems to me that the downside of getting a bad approximation is that the percentages don't tell you anything. That +-4% is coming from the assumption that your binomial distribution is acting pretty much like a normal distribution; it's saying, if we had a normal distribution with the same mean and standard deviation of the binomial one, we'd have a +-4% confidence interval. But if our 'n' is too low and our 'p' is too close to 0 or 1, then the binomial distribution does not act like a normal distribution, so those percentages are worthless. We can brute-force calculate an actual confidence interval by looking at the binomial distribution itself (this would involve taking a lot of big combinations and exponents) and we wouldn't get +-4%.