One problem with this model is that the RANDOM and REPEATED statements are fitting the same covariance structure in Z. The RANDOM statement does this through G and the REPEATED through R. You should get a message about the G matrix or Hessian being non-positive definite with this model. Most of the time, MIXED will set one of the variance components to 0 to make this more obvious for you. To "fix" this, try using AR(1) (or any structure other than CS) on the REPEATED statement.  Your LSMEANS statement is making comparisons among levels of your 3-way interaction at the average value of the covariate. You can try using the AT option to see if there are differences in the interaction at other, possibly more interesting, values of the covariate. ... View more

With three non-nested random effects, your model falls into a class of models (cross-classified models) that can be very time-intensive and memory-intensive to fit. Knowing a little more about the data structure and number of levels to your 3 random effects will help the community understand your problem better. If you have a large number of levels to any of these effects, sampling observations from those levels might help. You can run repeated samples to build up a distribution for your random effect variances.  ... View more

Try using just WEEK as the repeated effect (the term after the REPEATED statement) and see if that will converge. I am not sure what MIXED will do if you include SEGMENT as both part of the repeated effect and the subject effect. You can try specifying the R option to see how MIXED is correlating the errors in your data. ... View more

Posting the code used and actual output will help the community answer your question more quickly and more accurately. Comparison of likelihood stats from competing mixed models can by tricky. Changes in the model can make the comparisons can make those comparisons invalid. Seeing the code will help the community provide a better response. Thanks! ... View more

The RANDOM statement random patient(treatment); correlates all the observations from the same patient with a CS structure.  The RANDOM statement random week*patient(treatment); correlates all the observations from the same week for a patient. So that statement puts a CS structure on all the segments from the same week on the same patient.  From your description, you may not need to further correlate the errors in your model. You state that segment is of no interest to you, so you do not need an effect that includes segment. It does appear that you are measuring both eyes on a patient, with 56 observations coming from each eye. If both, do you need to correlate observations from the same eye? You could correlate the observations from the same segment across the weeks, using something like segment*eye*patient(treatment) as the subject effect on a REPEATED statement. That might look something like repeated week / subject=segment*eye*patient(treatment) type=??; where you can choose the covariance structure applied. This model will get expensive to estimate if you have lots of patients and many weeks, however.  ... View more

It is best just to avoid PROC ANOVA. It was meant as a procedure for textbook-type problems. The method behind PROC ANOVA requires balanced data and there is no way to work around that requirement. Use GLM for your modeling needs that use only fixed effects. For models with random effects and/or repeated measures, use PROC MIXED.  ... View more

Balanced/Unbalanced data refers to the counts in each cell of your design (same number of observations per treatment group). PROC ANOVA requires balanced data in the design, PROC GLM and PROC MIXED do not. The data situation you describe is slightly different. Subjects with incomplete repeated measures are not included in PROC GLM. The method of moments used in GLM requires complete data for each subject. Subjects with incomplete data are used in PROC MIXED. Maximum likelihood methods do not require that subjects have observations for all time points. MIXED does allow only one observation per time point for a subject. So, GLM and MIXED will not agree if you have incomplete data on your subjects in a repeated measures analysis.  ... View more

You can explore the Kronecker product structures for TYPE= on the REPEATED statement. That is a handy construct for two-dimension repeated measures. You can set up an unstructured covariance between the eyes of a patient and CS, AR(1), or (on occasion) UN for observations within an eye. The model you have now is used quite often as well. You could try adding covariance within an eye as an extra random effect, with random int eye / subject=ptid; That RANDOM statement correlates observations from the same patient and further correlates observations from the same eye within a patient. You could fit AR(1) as a structure on the REPEATED statement you have now with this RANDOM statement.  ... View more

If you can post the code you are using, with and without the interactions, and the output from both then that would give us a better understanding of what is happening here. ... View more

Any success with your overnight run? Were you able to try HPMIXED? ... View more

Posting your results and providing more detail about your data (replications, specifically) will help us diagnose the problem ... View more

You may need to specialize some error terms here. This example in the GLM doc will get you started. PROC MIXED is a good alternative to GLM in these situations, too. ... View more