03-11-2013 06:29 PM
I have calculated a standardized weight for running weighted analysis using PROC FREQ. The issue is, I obtain standard errors that are half the size of the standard errors I obtain when using all the survey parameters in PROC SURVEYFREQ. I calculate the same proportions when I use the standardized weight in PROC FREQ and the original weight in PROC SURVEYFREQ, but I have a larger standard error in PROC SURVEYFREQ than in PROC FREQ.
I am trying to figure out if there is a standard way to inflate the standard error so that it does not underestimate the variance estimates in PROC FREQ. Thank you for any guidance you can provide!
03-13-2013 08:28 AM
If you have a survey, use PROC SURVEYFREQ, not PROC FREQ. The standard errors and the variances calculated in PROC FREQ do not account for the survey design (for example, stratification, clustering).