This study focuses on testing the superiority of one Poisson distribution against another with a smaller mean value
for a small to moderate data set. Differing with the majority of the existing literature, we consider a non-superiority null
hypothesis and hence deal with a broader null parameter space, which introduces more difficulties for both theoretical
justification and practical computation of a p-value. Using a test that considers only the boundary of equality and ignores
the null space of inferiority risks an inflation of its type I error rate. For a strict control of the error rate, we consider the
confidence-set p-value. As the two proposed test statistics satisfy the convexity condition strictly, the calculations for the
confidence-set p-values are greatly simplified to involve at most a supremum search over a closed interval of a single
argument. On the other hand, we propose the estimated exact p-value by using the RMLEs under the null space. In the
numerical study, both the proposed p-values perform adequately.