In my understanding, CMH for X*Y*Z is similiar to (??) log-linear model in form of _response_ = x y z x*y x*z, isn't it? both are 'conditional' on X.
Jingju
That is what I try to say. CMH tests correlation of z and y when strafying x.but your full loglin model is not so.I tkink you are comparing different things. I suggest you try my model or ask your prof.
more, what do you mean '相差很远‘? no harm if you can list the test results.
JingJu