I'm often confronted with performing longitudinal regression with two repeated terms: year and eye (right and left representing different, but correlated, values at each year). With PROC MIXED of SAS I've been able to incorporate both class variables in the repeated statement in an unstructured variance model: "repeated timecat eye/ type=un@un". Can this — or something similar — be done with the JMP Pro Mixed Model platform? And, if so, how? It does not seem to let me enter two terms.
In JMP Pro, select Help > Books > Fitting Linear Models. When the PDF book opens in your default reader, select Edit > Find and enter "repeated measures." The search should find the first item in the table of contents. This item is a link that takes you to the right section of the Mixed Models chapter.
Only one repeated variable is allowed for the Unstructured covariance structure in the Mixed Model Personality of Fit Model in JMP Pro.
To use an unstructured covariance structure in the mixed platform you need a single variable that identifies the individual observations from each unit. As you found, with models containing more than one regressor there isn't a single variable identifying these observations, rather it is the combination of the levels of the two (or more) variables that identify the repeated observations.
To handle this, you need to generate a new variable that you won't use in the model specification, but will use in the "Repeated" section when designating an unstructured covariance structure. In your case, I believe this can be made with a concatenation of Year and Eye. In JMP 11, you could use the in-line transform columns to generate this concatenation while in Fit Model. Otherwise, you can make a new variable, and in the formula define the concatenation. If you haven't made these concatenations before let me know and I would be happy to write out some steps.
I hope this helps!
You were kind to answer my question in 2014 regarding JMP Pro 11. In now using JMP Pro 13, I see that the Repeated tab allows 2 Subject variables, for example when using an Exchangeable covariance structure. Instead of concatenating EYE and YEAR as the Repeated variable, as you recommended before, could I now use both ID and EYE (which has two levels — right eye and left eye — nested within ID) as Subject variables and YEAR as the Repeated variable? I've tried both models and their outputs are similar, but not identical. Which would be the more appropriate model for me to use for repeated measures longitudinal regression where I'm following two eyes of most patients over time? I understand from jiancao (Staff) that an alternative approach would be to define random effects, but I'd still like to understand these other options. Thanks in advance.