Simplifying the Analysis of Complex Survey Data Using the SAS

Large sample-based surveys often have complex sample designs, with design features including stratification, clustering, multi-stage sampling, and unequal probability of selection of observations. The calculation of the associated sampling weights often involves nonresponse adjustments and raking to external control totals. The analysis usually includes descriptive statistics such as frequencies, means, totals and their standard errors. Using standard statistical software modules such as PROC SUMMARY, PROC FREQ and PROC MEANS to analyze such data results in underestimation of variance, as these routines assume that the data is from a simple random sample and do not take into account the complex nature of the sample. Now, however, the survey analysis procedures such as PROC SURVEYMEANS and PROC SURVEYFREQ that have been added to SAS/STAT ® software can compute variances that accurately reflect complex sample design and estimation procedures. This paper compares the complexity of the variance estimation code used in earlier projects with the simplicity of the code that is possible using the survey analysis procedures.