I am trying to find what permutation size $B$ I should be using for conducting a permutation test. Currently I have two samples which are normal, same mean but different variance. The sample sizes are the same at $N_1 = 30$ and $N_2 = 30$. Obviously it is not feasible to find and evaluate each permutation as $60 \choose 30$ = 1.182646e+17. Hence, I am trying to find the upper bound on what I should make $B$. I noticed that there was already some literature and people have asked questions:

Required number of permutations for a permutation-based p-value

However, it seems that the upper bound doesn't depend on my sample size but rather the precision. Does anyone know why and how I may be able to obtain a reasonable bound for my sample size of 60? Thanks!

  • 1
    $\begingroup$ You're trying to calculate a proportion via what is effectively simulation (random sampling). Why would the sample sizes matter with respect to what that proportion is? $\endgroup$
    – Glen_b
    Sep 24, 2015 at 2:58

1 Answer 1


The intuition is for example, I have a sample size=10, if I permute it 500 times, it should generate a distribution more approximate the true distribution. Accordingly, the estimated p value is nearer the true p value. On the contrary, if the sample size is 100, 500 times permutation may cause a larger variance of the distribution of p value. I think this is the underlying logic of the question. But I have no answer rn.


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.