Categorical Data Analysis Using The SAS® System, 2nd Edition

3.2. Mantel-Haenszel Test

For the data in Table 3.1, interest lies in determining whether there is a difference in the favorable rates between Test and Placebo. Patients in both centers were randomized into two treatment groups, which induces independent hypergeometric distributions for the within-center frequencies under the hypothesis that treatments have equal effects for all patients. Thus, the distribution for the two tables is the product of these two hypergeometric distributions. You can induce the hypergeometric distribution via conditional distribution arguments when you have postrandomization stratification or when you have independent binomial distributions from simple random sampling.

Consider the following table as representative ...