GLIMMIX Repeated measures multiple means comparison

[AgReseach7]

Just want to make sure I'm "protected" when anallyzing all possible comparisons.

Study design (lamb feeding trial):

Effects of using ground juniper and feed urea in supplements fed to ewe lambs on growth, blood serum, xxx were evaluated. In a randomized design study, individually-penned lambs were fed xxx and of 1 of 8 supplements (fed separately from hay; 6 lambs/treatment) in a 4 × 2 factorial arrangement: 4 concentrations of juniper (15, 30, 45, or 60% of DM) and 2 levels of urea (1 or 3%). Lamb growth was evaluated on d 0, 5, 12, 19, 26, 33, and 40. Blood serum evaluated on d 6 to 8, 20 to 22, xxx.

A table will be produced that evaluates all possible comparisons.

Questions:

1. Is the following correct? Using Tukey adjustment?

2. Can I use "adjust = tukey" with PROC Glimmix (needd for some of my blood serum data)?

3. If JUN*UREA*Day is not significant, do I just drop it from the model & re-run it or just leave it in & just use the LSMeans of the full model? Probably doesn't matter, correct?

Note: all lambs have a different animal ID number.

PROC MIXED;

CLASS ID JUN UREA DAY;

MODEL supplementIntake = JUN UREA   JUN*UREA   JUN*UREA*DAY;

REPEATED DAY/SUBJECT=ID    TYPE = TOEPH (or other);

LSMEANS  JUN  UREA   JUN*UREA   JUN*UREA*DAY/PDIFF ADJUST=TUKEY;

RUN;QUIT;

[ballardw]

This may be more a comment on variable names. You have

Model supplementIntake =

Are you trying to model the intake? I would think that intake is more of something you are controlling and would be more of a fixed effect variable and that Growth or Blood serum measurement would be on the response side.

[AgReseach7]

Supp intake is the depend. variable. Each supplement is different, thus we analyze intake of that supplement, along with growth, serum, etc.

[SteveDenham · Accepted solution]

I would suggest adjust=simulate as opposed to Tukey as that method more accurately controls type 1 error.  

As far as the model statement, i suggest:

MODEL supplementIntake = JUN UREA   JUN*UREA  DAY JUN*DAY UREA*DAY JUN*UREA*DAY/solution;

and for the lsmeans statement, similarly incorporating DAY:

LSMEANS JUN UREA   JUN*UREA  DAY JUN*DAY UREA*DAY JUN*UREA*DAY/DIFF ADJUST=SIMULATE ADJDFE=ROW;

I would recommend not dropping any terms from the model due to "nonsignificance".

Steve Denham