Proc Mixed with random statement

[SylviaK]

Hi,

I am having an issue debating whether to include an interaction treatment in my random statement.  I ran an animal trial 3 times (due to lack of space for housing more animals).  So I had 3 repetitions of animals.  There were 3 treatments within each repetition and 5 animals per treatment.  So 15 total animals that we measure weight (and other things) weekly per repetition.  What I didn't expect was that each repetition is significantly different from each other.  So I included repetition as a random statement (along with repeated week/subject=animal).  Since the rep is different and the treatments are affected by rep, should I also include repetition*treatment within the random statement?  If I run rep*treatment within the model statement as a fixed effect, it is not significant. 

Any insight would be much appreciated....I'm very unsure as to what to do.  Thanks in advance to everyone for your time!!

Sylvia

[SteveDenham]

I would almost certainly include it as a random effect, if for no other reason than to get the denominator degrees of freedom more nearly correct.  Set up what Stroup refers to as a skeleton ANOVA table and check for degrees of freedom and sources of variability.  With a repeated measures design and a small dataset, I would also apply ddfm=kenwardrogers.

See W.W. Stroup (2013) Generalized Linear Mixed Models. CRC Press

Steve Denham

[SylviaK]

Thanks so much for your help!