|
This lead me into a rabbit hole to understand how one should assert
the shape of the distribution of some empirical data w.r.t. to the
"true" expected distribution which uses the Pearson coefficient
test. Unfortunately for very large number of samples deviations sum up
quickly to a point where the computed p-value is basically 0, thus
rejecting the null hypothesis.
While this is theoretically true, there are a lot of ways to introduce
biases that will lead to rejection, one of them being the use of
modulus to select numbers from the generator's output!
|
