05-23-2013 05:40 PM
I'm a doctoral student in Epidemiology. For some projects, I need to analyze data from National databases, such as, BRFSS, NHANES, etc. and I generally use SAS 9.2 or 9.3 for analysis. One question that I have is: how do we get the weighted rates, or, proportions? Because, most of the surveys (even if it is a small survey) uses a stratified/ cluster/ multistage sampling procedure; some groups are oversampled and they don't use a simple random survey.
In many publications that have used these kind of data, I find that SUDAAN has been used (sometimes SAS-callable SUDAAN). I am wondering whether it is possible to adjust for survey design just by some specific commands in SAS; or, do we always need something else, like SUDAAN. I have heard about the proc survey procedures, but haven't ever used them. Do they account for the survey design?
I would appreciate any suggestions on this.
05-23-2013 05:45 PM
Depending on the analysis there's a range of procedures that are called SURVEY* that deal with survey data.
For example, logistic regression with survey data can be done using, PROC SURVEYLOGISTIC
05-23-2013 06:36 PM
You may need to add other variables containing the STRATA variables for "domain" analysis as the SAS survey procs don't generally seem to allow the use of Strata variables as Domains. The last time I compared SUDAAN and SAS survey procedure output on BRFSS data confidence limits for means were usually the same at the .001% and often to more decimal places, but there are differences in the variability calculations.