Dear @sld, Thanks very much for the suggestions. Following are my responses: 2) The TABULATE procedure is new to me and very handy indeed. Thanks! And yes--it shows a "1" in all cells of the dataset. (3) Sorry for the confusion. The original dataset was "pops" but I had converted it to the "tall" format (i.e., 3 entries per individual). Looks like I made an error in the copy/paste. I corrected the dataset name in the original post. (4) Apologies for not reporting back, but yes--I ran the model you had suggested: proc mixed data=tall method=reml covtest asycov;
class var pop id;
model y=var;
random var/subject=pop type=unr;
repeated var /subject=id(pop) type=unr;
run;
quit; Unfortunately, this returned the "infinite likelihood" error mentioned above. (5) And yes, it had occurred to me that I could simply generate populations means for the three response variable, then simply estimate the variance-covariance matrix from those means (and indeed, this is common practice for this particular type of analysis). Others have shown similar results doing this versus the random effect variance component approach... My hope was to have both to be able to compare. And to flesh out the dataset a bit more: 1. there are no missing data 2. the dataset is unbalanced--the number of 'ids' per population ranges from n=10 to n=85 (mean is 39.3 individuals per populations) Finally, I should mention that I had been working off of this document which discusses multivariate analysis using PROC MIXED. Unfortunately, none of the examples therein included both a RANDOM and a REPEATED statement. Thanks again, kpboh
... View more