06-19-2016 10:23 PM
PROC GLM Estimate statement: if group has 8 levels (groups 1-8), do I need to include all in effect-values even though I am testing say, 4 groups only?
e.g. estimate group 4 vs group 2 among groups 2, 3, 4, 5?
I plan to do only for comparisons (hence Bonferroni 0.05/4 rather than 0.05/8)
06-19-2016 10:39 PM
I guess you only need to specify the first N level.
estimate 'A1 VS A2' A 1 -1;
even A have more than two levels .
06-28-2016 01:56 PM
This is a great case for the LSMESTIMATE statement, available in PROC MIXED or GLIMMIX. Take a look at the shared concepts documentation for this statement, and it should be readily apparent how to do the comparisons you want, in a series of comma separated statments, followed by /adjust=bon; for a Bonferroni adjustment.
It is pretty easy to port GLM code to MIXED or GLIMMIX.