For my dissertation, I have a pooled national dataset (Behavioral risk factor surveillance system) which is a weighted complex survey data. My study outcome is binary and I am trying to do weighted multilevel modeling (MLM) using this complex survey data and I am using the PROC GLIMMIX procedure available in SAS 9.4. The link is here: SAS Help Center: Example 49.18 Weighted Multilevel Model for Survey Data In my model, I am trying to analyze a subsample consisting of non-pregnant and who are Latino and Asian Americans. I know the weighted Glimmix syntax in SAS does not take the "Domain" command line and hence I could not use the "Flag" variable to subsample the data. I can and have used the "Where" or 'By' command in the syntax and it is giving me results, but I know it may be giving me biased results. I need to find an accurate way to subsample data from my main large dataset while doing PROC GLIMMIX weighted MLM with survey data. STATA has 'subpop' option in this case but I cannot use STATA because my original dataset is too large (2300000 observations) and STATA cannot take it. I have been trying to find a solution for the last 2 months but could not find a reasonable and accurate way. If you can find me a solution, I will be grateful.
... View more