Prev: Running SAS in batch under Windows XP.
Next: Bootstrapping a dendrogram?
From: Nilesh Timilsina on 28 Jan 2010 09:48
Hi Sas Experts

I have data on tree growth collected within plots. Plots are random effect
in my model. I am assuming that the tree within same plots are more
correlated. I want to take this into account while fitting my model. To do
this i want to specify different error correlation structure as given by
Littel et al. 2006. Following is my code:

%inc
"C:\Documents and Settings\nilesht\My Documents\Research\Final Analysis
Data\nlinmix.sas";
%nlinmix(data=test1,
procopt=method=reml,
model=
%str(
m=b0+b2*tbaac+b4*qmdac+u1;
predv = a+(b*totht)*exp(d*totht)+m*lcurdbh;
),
parms=
%str(a=-5.961 b=0.8570 b0=3.82 d=-0.045 b2=0.00057 b4=-0.0148),
stmts=
%str(
class plots tree;
model pseudo_htgr3 = d_a d_b d_b0 d_d d_b2 d_b4 / noint notest
solution cl;
random d_u1 / subject=plots type=un solution cl;
repeated/ subject= plots type = sp(sph) (x1_utm y1_utm);
),
expand=zero
)
run;
In my repeated statement should I specify plot as my subject or tree as my
subject? I don't have repeated measurements for a tree. I am assuming trees
within the same plot have higher correlation than trees in separate plots.
Also the model with the code given above fails to run properly. If i use
TOEP (3) as covariance structure it runs properly, but spatial covariance
structures such as sp(pow), sp(exp) and sp(sph) fails to run.

Thanking you
Nilesh Timilsina
 | 
Pages: 1
Prev: Running SAS in batch under Windows XP.
Next: Bootstrapping a dendrogram?