Help with estimate statement in proc glimmix resulting in non-est !

[sbampah]

I have count outcomes collected longitudinally at baseline and two post baseline time points: Time: 0,1,2 and Treatment group of two levels: A and B

proc glimmix data=work.Regg method=quad;

   class Study_ID Trt Time;

   model Y= Trt Time Trt*Time /solution link=log s dist=nb;

   random int / subject=Study_ID;

run;

Negative binomial mixed effects model:

log(Mui_ofY)=Beta_0+Beta_1Trt_A + Beta_2Time1 +Beta_3Time2+ Beta_4Trt_A*Time1 + Beta_5Trt_A*Time2 + Random_int

Note: Refs for Trt is B, and Time is Time 0

I am interested in estimates/contrast for reduction in mean of Y at time 1 relative to baseline for only Treatment A:

which is Beta_2 + Beta_4 and used the estimate statement below but does not work for me.

   estimate "Trt A  Time 1 vs Time 0"  intercept 0

                                Trt     0  0

                                Time     1 0  0 

                                Trt*Time 0 1 0 0 0 0;

This results in non-est, can someone help please ?

[PaigeMiller]

Typically, non-est for interactions happens because the interaction cannot be estimated. This is almost always due to cells in the design where there are zero observations.

For example, in this simple data set below, you cannot estimate the X1*X2 interaction because there is an empty cell where x1=0 and x2=0

   x1    x2   y
    0     1   5
    1     0   8
    1     1   4

Also, there is almost never a need to compare Time 1 versus Time 0 via an ESTIMATE statement. The LSMEANS statement will do the comparison for you with much less effort on your part.

[sbampah]

Thank you for the insight it is very helpful, however, how do we specify this test with lsmeans ?


Steve B. Ampah,

[PaigeMiller · Accepted solution]

There are examples at the link I gave.

Maybe something like

lsmeans time/lines;

[sbampah]

Thanks very much, while it worked I had to add a little more code to get what I wanted.
lsmeans Time*Trt/ slice=Trt lines;
The above code gives me what I want.
Thank you.