09-01-2013 09:24 PM
I am comparing dichotomous health-related variables (ie do you have diabetes? asthma? health insurance? etc) among three separate surveys. Two of the surveys are national and sponsored by the CDC: NHANES and NSCH.
The third survey is conducted by a local hospital system here in Texas.
I have run PROC SURVEYFREQ on all of my variables to obtain the proportions and std error.
Now that I have them, I need to compare between them in order to determine if they are statistically different.
I'm considering using PROC MULTTEST...but if I do that, then I would either
a) have to input the p-values (and how would I generate those?)
b) ask for the Procedure to run a test and the one that fits best i think would be Test Fisher Exact.
Then I can ask for the Bootstrap and Bonferroni corrections on the mult comparisons.
So, am I on the right track? or is there a completely different way for me to think about this??
Thank you very much in advance,
09-03-2013 09:11 AM
I don't think you can get Fisher's Exact Test out of PROC SURVEYFREQ. However, for large surveys such as NHANES, using a Wald chi-square seems appropriate, as would a Rao-Scott. If you do that, then adding an ODS statement should get the unadjusted p values into a dataset. You may have to do a little bit of editing on that dataset, but you can input the p-values using the INPVALUES= option in the PROC MULTTEST statement.