06-12-2015 02:34 AM
i need small clarification for proc mixed, if any one knows then please let me know.
what is the difference b/w
Repeated variable/ subject=subjid type=UN;
Repeated variable/ type=UN;
1) if i mention subject=subjid then what it's means?
2) if not what?
Note: In My proc mixed model, subject is considered as Random effect. so that random statment will effect the repeated statment?
please clarify my doubts.
06-16-2015 01:02 PM
For an unstructured R matrix (repeated statement) the two statements are very different. Without specifying the subject=, the covariances (off-diagonal) values are zeroes. When you specify subject=subjid, the covariances between timepoints are estimated. This is a good thing.
However, specifying the subject as a random effect in combination with type=un (or cs or csh) will almost certainly lead to either convergence problems or statements in the output that the G matrix is not positive definite.
If, instead you model the repeated effect with a structure that specifically involves correlations, such as AR, ARH, SP(POW), ARMA or ANTE, inclusion of the random effect of subject will almost always result in a more stable and computationally precise estimate of standard errors.
07-23-2015 02:20 AM
Thanks for your valiable reply.
so in this case if i used "Repeated variable/subject=subjid type=UN"; is it necessary to mention random statment with "Subject"?
in Repeated statment, "Subject=subjid" is random effect or fixed effect?
07-23-2015 08:05 AM
Q1: No, and in fact adding a separate RANDOM statement with the subject alone will result in a non-positive definite message regarding the G matrix, as all the variability will have been explained in the R matrix.
Q2: In the repeated statement, subject=subjid isn't considered a random or fixed effect, but is a subject classifier that denotes what the repeated measure is observed upon. I suppose in that sense, it is fixed, as it only addresses the subjects sampled in the trial, and they are assumed to be independent of one another.