1-sided CMH

[darb]

Looking for general information about 1-sided CMH tests. Also if anyone has any code with regards to this test that would be helpful.

can use variables strata*treatment*outcome.

[SAS_Rob]

Each of the three tests that Proc FREQ reports has a very specific alternate hypothesis that is inherently two-sided.  While I suppose the Row Means score differ and the Correlation statistic could have a one-sided alternative, I have never seen these defined anywhere.  In any regard, there is no way in SAS to get them.

[Rick_SAS]

Although a one-sided CMH test isn't built-in to PROC FREQ, you can 

1. Program the test yourself if you know the distribution of the test statistic

2. Use Monte Carlo simulation to estimate the one-sided p-value, provided that you can sample from the null hypothesis.

For an example of simulating a test for a contingency table, see

"Monte Carlo simulation for contingency tables in SAS."

The article links to other articles that describe how to simulate tables that are subject to various assumptions, such as fixed N or fixed marginal frequencies.