Date:         Thu, 22 May 2003 21:53:29 -0800
Reply-To:     Dan Jeffers <ojodelmar@HOTMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Dan Jeffers <ojodelmar@HOTMAIL.COM>
Subject:      COV Structure in proc MIXED
Content-Type: text/plain; format=flowed

Dear SAS-L subscribers:

I am trying to fit a linear mixed affects model using proc MIXED to a
spacially correlated set of data. I am trying to fit the correlation among
measurements into a spatial powers, sp(pow), covariance structure with the
distance between my sampling points as the spacial reference variable, so
that the repeated code statement looks like this:

proc MIXED data=soil;
     class PLOT;
     model N=PLOT elevation temp/s cl;
     repeated / subject=PLOT type=sp(pow)(dist) r rcorr;
run;

where dist is a continuous measure to 2 decimal places.  In the study,
nitrogen content is the response measured repeatedly from samples taken from
each plot at different distances with soil temperature and elevation as
covariates.

I fit the model and REML converges on estimates without errors, however in
the covariance parameter estimates the variance partioned to the repeated
factor is a very small number.  Additionally, when I request the R matrix
and the correlation of the R matrix in the output, I do not get a full
matrix of values, only the diagonal value of the R matrix and the diagonal
of 1s in the r corr matrix. I have tried other covariance structures and get
similar output, the model fit statistics (AICc, BIC, etc.) give nearly
identical values for the fit of each of the structure forms to the data.

Samples are not equal among plots, and distances are also variable, samples
were not taken at the same distances along each transect at each plot.

Any thoughts on what may be happening are welocomed.

Thank you,

Dan Jeffers

_________________________________________________________________
Add photos to your e-mail with MSN 8. Get 2 months FREE*.
http://join.msn.com/?page=features/featuredemail