non-est LS means on proc mixed with main effects and interactions model

[hikmatazizat]

I tried to run a proc mixed on SAS ODA. before doing that I classified the variables into several categories. Then, I added interactions in the proc mixed model; however the LS means is always non-est. Does anyone know how to solve it?

[PaigeMiller]

Non-est indicates that some of the different cells that make up the interaction are empty (no data) and so LS Means interactions cannot be computed.

Example showing the number of data points in a cell:

                     DOG           CAT

MALE              7                  6

FEMALE         6                   0

Because FEMALE/CAT has zero observations, LSMEANS cannot estimate an interaction, and it shows this as Non-Est.

[SteveDenham · Accepted solution]

I'd like to modify @PaigeMiller 's response.  It is indeed due to empty cells, but generally it is the main effects that turn out to be nonestimable.  Interactions at the highest level should still be estimable.  To get around this, the LSMESTIMATE statement is a great tool.  Fit a "one-way" model such as:

model response = var_1*var_2*var_3...*var_n;
lsmeans var_1*var_2*var_3...*var_n / e;

From the coefficients produced by the /e option in the lsmeans statement, you can pull out the terms that need to be used to estimate main effects, and by judicious use of the JOINT option, you can get F tests.

SteveDenham

[PaigeMiller]

 It is indeed due to empty cells, but generally it is the main effects that turn out to be nonestimable. 

Thanks for the clarification, @SteveDenham