Please could someone help me with the following?

I wish to compare two proportions and would normally use a pooled estimate theta = (n1p1+n2p2)/n1+n2
and the normal approximation z-test where p1-p2 / sqrt(theta(1-theta)(1/n1 + 1/n2)) is normal(0,1).

However I have low sample sizes where n1<50 and n2<20.

What test should I be using - I have unknown population size so am unsure as to whether I should use the hypergeometric or fishers exact??

Please respond to [log in to unmask]

Thanks in advance