10-17-2014 04:06 PM
I am trying to do repeated measure MANOVA using proc glm. The code that I wrote as following:
proc glm data=a;
model X1 X2 X3 X4
Y1 Y2 Y3 Y4
Z1 Z2 Z3 = treatment / nouni solution;
I have 3 measurements, X and Y were measured 4 times, Z was only measured 3 times. "treatment" has 2 levels: 0 vs 1.
The data has some missing value due to loss of follow up. I did multiple imputation using Proc MI with MCMC. I ended up having 5 complete data sets.
(1) Can I do repeated measure MANOVA using Proc GLM with unbalanced outcomes (in this case, Z was only measured 3 times while X and Y measured 4 times)?
(2) If Proc glm allows unbalanced outcomes, how can I specify "repeated" statement?
(3) How do I combine the inference drawn from 5 complete data sets? Do I just focus on the coefficient of "treatment" and see if p-value in the combined result < 0.05?
Thank you very much.
10-17-2014 04:26 PM
It will be much better if you use PROC MIXED for your repeated measures, based on the information you supplied. Here is an old article, but it very nicely explains the transition from GLM to MIXED, and why GLM is not flexible enough for many applications. You have to change to a stacked data format.
Once you get the hang of this, you would be ready to look at doubly repeated measures:
10-19-2014 04:07 PM
Thank you for the paper. I read the third paper and I think it's very helpful. I guess my question is if I have three measurements: X, Y ,Z. X was measured 4 times within individual, Y was measured 4 times within individual, Z was measured 3 times within individual. Can I still use random mixed model as this paper suggested? Thank you.