3 weeks ago
I'm using the REPEATED statment in PROC MIXED to specify a spatial covariance structure for the R matrix in a mixed model. For example ...
repeated / type=sp(sphga) (x y) subject=intercept local;
I'm using an anisotropic variogram model, and I specifically want the parameter estimates for the ranges in the directions of the major and minor axes of the correlation ellipse.
In past versions of PROC MIXED, I think you could get those by using the 'Range' parameter estimate, e.g., SP(SPHGA) in the above example, along with the 'Scale' parameter estimate. Specifically, I think the range parameter was the range in the direction of the major axis, and the scale parameter was the factor by which this 'major range' was 'shrunk' to obtain the 'minor range' (the range in the direction of the minor axis of the correlation ellipse).
Instead, I'm being told by SAS Tech support that it is not possible to obtain these range estimates (in the direction of the major and minor axes of the correlation ellipse) from the parameter estimates available from PROC MIXED. This seems quite unusual, i.e., key parameter estimates for a mixed model fit by PROC MIXED cannot be obtained from PROC MIXED outputs.
Has anyone else encountered this situation and, if so, do you know of any 'workarounds' to obtain these parameter estimates?