0
$\begingroup$

I know p-values behave uniformly. Now as p(np) is fixed and n goes to infinity, binomial converges to normal(Poisson). Now suppose I take random binomial samplings and fir normal(Poisson) to it, for say n = 1000. Will my p-value still be uniformly distributed or as binomial converges to normal(Poisson), p-values mostly will be in 0.8-1?

$\endgroup$
0
$\begingroup$

I think what you are referring to is sampling a binomial random variable $X$ and then looking at the distribution of $F(X)$ where $F$ is the binomial CDF. This is actually not uniform; it is a sort of discrete approximation of a uniform variable. In the case where normal approximation becomes valid, it converges to a uniform variable. In the case where Poisson approximation becomes valid, it does not converge to a uniform variable, but rather to the distribution of $F(P)$ where $P$ has the limiting Poisson distribution and $F$ is the limiting Poisson CDF. Again this isn't uniform.

$\endgroup$
  • $\begingroup$ Actually what I did was, I took some random 100 random samples from binomial n=1000,p=0.1. Then fit Normal to it, calculate the chi square value and hence the p-value. Ran this 100 times and look at the histogram of 100 p-values. It was a bit uniform looking. So, I was a bit confused as that it should converge to Normal, so p-value should be >0.8 as the fit should be good. But why was it so uniform? $\endgroup$ – Avinash Bhawnani Mar 31 at 17:41
  • $\begingroup$ @AvinashBhawnani Can you explain exactly what you mean by calculating the p-value? I think it is essentially what I am saying here, except that you're plotting $F(X)$ where $X$ are your binomial samples and $F$ is your fitted normal CDF. This will basically look exactly the same as what I said on a histogram with that many data points. $\endgroup$ – Ian Mar 31 at 18:30
  • $\begingroup$ By p-value I meant, the area of the right side in the chi square distribution graph, with the value of chi square found by test. $\endgroup$ – Avinash Bhawnani Mar 31 at 18:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.