PROC GLIMMIX help needed!

[tka726]

I’m running a glimmix model and wanted to make sure I am modelling the data correctly.  Glimmix is a relatively new procedure for me.

Each subject was followed for 4 days, and they were randomized to one of two trainers. Subjects were with each trainer at least once, but could have been with one trainer 3 times and the other trainer once (like subject 3). At the end of each day, a test was given.

I’m trying to see if the scores differed depending on trainer, over the 4 days. Also interested in the differences in scores for each day.

This is a simplified snapshot of my actual data.

PTNO   DAY    TRAINER    SCORE

1             1             1             50

1             2             2             60

1             3             1             55

1             4             2             45

2             1             1             70

2             2             1             72

2             3             2             90

2             4             2             85

3             1             1             67

3             2             1             79

3             3             1             80

3             4             2             85

...

Is this the correct model?

PROC GLIMMIX DATA=ads;

  CLASS ptno trainer day;

  MODEL score = trainer day  trainer*day  ;

  random int / subject=ptno;

   RANDOM day / subject=ptno*trainer type=un;

Run;

[jiltao · Accepted solution]

I would probably add the RESIDUAL option to your second RANDOM statement --

 RANDOM day / subject=ptno*trainer type=un residual;

Thanks,

Jill