09-20-2016 01:04 AM - edited 09-20-2016 11:21 AM
I have 4 independent groups of individuals that I would like to compare in terms of their mental health outcomes (a continuous depression score). I am running a linear reg, with proc surveryreg. I am controlling for a slew of covariates, and adding in a weight statement, since this is survey data. My question has to do with the dummy coding. Is it enough to just reference code using the class statement, or do I need to also include an estimate statement? I will be comparing each of my groups to group 4. What I'm essentially interested in, is getting the parameter estimates and SE for each group against the reference group.
model mentalhealth=Groups /solution;
estimate 'group 1 vs 4' groups 1 0 0 -1;
estimate 'groups 2 vs 4' groups 0 1 0 -1;
09-20-2016 10:59 AM - edited 09-20-2016 11:01 AM
In my opinion, I think you will find the results of the estimate statements to be more meaningful than just looking at the solution vector.
However, you have a problem, in that the variable 'groups' doesn't appear in your MODEL statement. In the current MODEL statement, the only effect variable is 'cluster', which I don't think refers to group. You will need to correct one or the other to get any results.
09-20-2016 01:55 PM
The ESTIMATE statement will enable you to get the difference between levels, given the other covariates. Much easier would be to use an LSMEANS statement, such as: