Equivalence results in chi-square tests

This paper shows that two versions of the chi-square test statistics are equivalent, whether it is constructed from the entire cells in a partition of the support or one less cells. This equivalence holds not only asymptotically but also for each sample realization.