🚀 go-pugleaf

I am trying to fit a model to data where I have fixed effects, a covariate,
and several measures on the sampling units.  The dependent variable is
number of predator insects (after a normalizing transformation).  The
sampling units are trees, and the different measurements on each tree
correspond to counts at 6 different (non-random) positions on the tree.  The
fixed effects are species of tree and geographical location of the tree.  Five
trees of each species were sampled at each geographical location.
The covariate, which is actually the most interesting effect, is the
number of host insects at each position on each tree.

I tried to test for the significance of the different effects using PROC
GLM of SAS:

PROC GLM;
CLASS LOCATION SPECIES TREE POSITION;
MODEL PREDATOR = LOCATION SPECIES TREE(LOCATION*SPECIES) POSITION
                 SPECIES*POSITION HOST / SS3;
TEST H=LOCATION SPECIES  E=TREE(LOCATION*SPECIES);

(The omitted interaction effects were judged by the investigator to be
 theoretically impossible.)

The problem when running this program - and other similar programs when I
have had a comparable problem - is that SAS gives me zero as the Type 3 SS
for the HOST (covariate) effect.  The way I understand it is that after
adjusting for the "individuality" of the tree with TREE(LOCATION*SPECIES)
there is no variance left for the covariate effect.  Does anyone have a
suggestion how to test a model with a specific TREE effect and a covariate
together?  As I say, I have run into this problem before and I need a
solution.

Thanks in advance for any advice.

Hillary Voet, Statistician
Faculty of Agriculture
Hebrew University of Jerusalem.

(I wouldn't dare try a map! )