Skip to content
Prev 5715 / 20628 Next

Repeated measures mixed model with AR(1) correlation structure in nlme vs SAS Proc Mixed

This makes a really nice comparison, thank you.

  It makes me think that there is some difference that we haven't yet
figured out in the way that lme and PROC MIXED are defining the model,
although nothing springs to mind, although I can't see what it would be.

  As far as I can tell, both are fitting a fixed effect of Response ~
Day where Day is a categorical predictor; both are treating Subject as a
random effect (affecting the intercept only); both are assuming AR(1)
autocorrelation in the residuals, applying only within-subject; both are
using REML ... ?

  I agree with your points about taking account of the gap between
time=3 and time=6 -- but I think the main point here is the narrower one
of "what are SAS and R doing differently for this particular (even if
sub-optimal) model"?

  What is the meaning of "Columns in X 5; Columns in Z 0" in the SAS
output;  I would have thought X (fixed effect design matrix) would have
4 columns, Z (random effect design matrix) would have 10 ... ?
On 11-03-27 03:13 PM, Jim Baldwin wrote: