04-30-2014 07:48 AM
Firstly, I am not an experienced SAS user (using v9.3).
I am using PROC PLS to derive RRR patterns in dietary data. I am using a national survey. The survey has a weight variable which can be applied to make it more nationally representative. Is there a way to account for this weighting during the PLS procedure?
Thank you for any advice you can offer.
04-30-2014 08:08 AM
By "weight variable", I am assuming you mean that each observation accounts for a certain proportion of the national population, instead of each observation accounts for 1/n of the national population if there are n observations.
In that case, you would have SAS replicate the observations such that now the proportions are correct, and then run PLS. This probably isn't perfect, since PLS works by explaining the variance of your predictors and responses, but its better than nothing.
04-30-2014 11:16 AM
You may want to check on the sampling methodology for the national survey to see if it is a complex sample. Procedures that would assume a simple random sample often under estimate variability in responses from complex samples.