|
Abstract:
|
When social scientists use population-based data for secondary analyses they sometimes ignore design specifics and mistreat the sample as a simple random sample. Many commonly used population-based studies, such as the Current Population Survey (CPS), American Community Survey, and National Health and Nutrition Examination Survey Three provide technical guideline on appropriate statistical methodologies, e.g., incorporating weights and specific variance estimation approaches. The purpose of this presentation is to illustrate the importance of using these guidelines when analyzing complex surveys. First, we discuss importance of using a finite population correction. Next, we demonstrate discrepancies in the estimates for smoking behaviors associated with disregarding suggested variance estimation methods. Specifically, we discuss applications of variance estimation via repeated replications with supplied and generated weights, and a number of Fay correction coefficients. Differences in the estimates are demonstrated using the CPS data.
|
