Various exact tests for showing a difference between two treatments or the noninferiority (therapeutic equivalence) based on the difference of two binomial proportions are compared. It is found that a frequently used test has to be applied with great caution due to its numerical instability. Furthermore, a test based on the score statistic can be recommended as a good compromise between a simple and powerful procedure. Finally, a likelihood ratio based exact test is introduced, which slightly outperforms all other tests from the literature with respect to power. The issue of sample size determination is briefly addressed. All methods are illustrated with help of an example where two antihelmintic agents are compared. Key words: likelihood ratio test, shifted null hypotheses, unconditional exact p-values, therapeutic equivalence, non-inferiority trials, Fisher's exact test
G. Skipka, Axel Munk, G. Freitag