Proc Glimmix random statement

[Gubardella]

Hi all,

I have an RCBD experiment with a two-way factorial treatment design, which are fully randomized within each block. I also don't have subsampling.

My question is: In the random statement of the proc Glimmix, should I include only block, or block block*factor(a) and block*factor(b)?

Thank you in advance for the support,

Gustavo

[SteveDenham · Accepted solution]

I would start with the more specific random statement, and reduce as needed.  In other words:

RANDOM intercept A B/subject=block;

would be the first I would try.  If there is sufficient data and variability to estimate all of these components, then you are done.  If however, you get the message that the G matrix isn't positive definite, or there are zero values for the block*A or block*B variance components, you can reduce the statement and remove the zero-valued variance component terms.

Steve Denham

[Gubardella]

Thank you for the help, Steve.