12-01-2016 11:06 AM - edited 12-01-2016 11:17 AM
I am analyzing disease data on individual trees. The dependent variable is a disease ranking (1-4; 1 being best, 4 being worst). The a priori model is complex and includes tree size at planting (size), breeding generation (generation), genetic family nested within generation (family; different families exist within each generation), and year (yr). Year is a repeated measure where data were taken at 0.5 year intervals for 5.5 years for each tree. We had a resolvable incomplete block design, so replication (rep) and block(rep) are random effects.
I am not sure how to incorporate the repeated measures into the model considering I have other random effects. I know how to do with this with Proc Mixed, but my data distribution is multinomial, requiring I use Glimmix. Should I have two random statements?
I also would like to make sure the default link function is what is most appropriate for the data.
I would like to make multiple comparisons and interpret differences among levels of each treatments.
Here is what I have so far:
proc glimmix data=blight14 method=laplace; ; class size generation family rep block yr tree; model maxrank= size|generation|family(generation)|yr /dist=multi cl oddsratio (diff=all label); random yr rep block(rep) /subject=tree type=ar(1);
Thanks for any help!
12-08-2016 07:09 PM
yes, try adding an intercept keyword effect to the first random statement and moving the rep and block(rep) into a second random statement with subject=tree and type=vc, fa0(1), fa0(2) or chol options.
Also adding ddfm=kr2 to the model statement options might help.
I'm not sure if maxrank needs to be specified as a class statement - does it work as expected without?