Hello. I have longitudinal survey data supplied with weights and I am aiming to fit a generalised linear mixed model to it, the response variable being employment status at one of five successive survey interviews, and the predictors being things like qualification level, gender, method of job search, and so on. Everybody in my subsample is unemployed at first interview.
My question is as follows. I have a set of survey weights, and I suspect it really doesn't matter to the analysis how they are scaled. However, I was wondering whether the WEIGHT option in PROC GLIMMIX, which allows me to supply non-response weights, requires the total sum of the weights to be scaled to the sample size.
PS Although the survey sampling scheme is complex, a single set of weights has been provided for use in longitudinal analysis, scaled to the population.
Many thanks.
I've never done this, but the example in the SAS/STAT14.1 documentation (Example 45.18 Weighted Multilevel Model for Survey Data) does give some hints, which would probably make a lot more sense if I spent the time reading the Rabe-Hesketh & Skrondal paper..
It appears to me that the example does scale the weights, so if your research question aligns with the example, it looks like the Method 2 weights need to be precalculated in a data step and supplied to PROC GLIMMIX.
But i bet you already knew this.
Steve Denham
Thanks very much for taking the time to check that paper out Steve. I will have a read of it. I might also experiment with different scalings and see if the results change!
Thank you for this hint. I am testing this procedure and corresponding packages in R. My simulation finds that weight inflates significance. Hope scaled weight can correct such inflation. I think R packages have the same problem.
Are you ready for the spotlight? We're accepting content ideas for SAS Innovate 2025 to be held May 6-9 in Orlando, FL. The call is open until September 25. Read more here about why you should contribute and what is in it for you!
ANOVA, or Analysis Of Variance, is used to compare the averages or means of two or more populations to better understand how they differ. Watch this tutorial for more.
Find more tutorials on the SAS Users YouTube channel.