Sophisticated sampling designs can save money, provide more accurate answers, and answer questions that simple random sampling (or non-random data) cannot. Yet too many investigators are guilty of a multitude of sins. These include not collecting data optimally, not analyzing data according to the underlying features of the sampling designs or (horrors!) all of the above. Survey data are the lifeblood of marketing analyses, healthcare studies, environmental and sociological research, and many other areas. It is essential that data be collected and analyzed in the best possible ways. This talk will focus on key aspects of sampling theory, such as cluster sampling, weighted samples, stratified sampling, sampling errors, and non-response errors. This will be combined with a discussion of the SAS® PROCs which help one design surveys, and then analyze those surveys: PROC SURVEYSELECT, PROC SURVEYMEANS, and PROC SURVEYREG, along with the %SMSUB macro and some V9 enhancements. Attendees will also learn the hazards of using PROC MEANS when PROC SURVEYMEANS was warranted, which, despite the title, do not include bursting into flame. Whether or not such punishment is warranted will also be discussed, with examples.
Paper:
conference:
