The CMH option in the TABLES statement of PROC FREQ produces three different versions of the Cochran-Mantel-Haenszel statistic. All three test the null hypothesis of no association, but the tests have different assumptions regarding the scale of the row and column variables, and they test for different alternative hypotheses. The table below lists the assumptions underlying each test and the corresponding alternative hypothesis.

Test	Assumption	Alternative Hypothesis
General Association	Does not require an ordinal scale for either the row or column variable	For at least one stratum, there is some kind of association between X and Y
Row Mean Scores Differ	The column variable lies on an ordinal (or interval) scale	For at least one stratum, the mean scores of the R rows are unequal
Correlation Statistic	Both variables lie on an ordinal scale	There is a linear association between rows and columns in at least one stratum.