Assessing the Impact of Simplified Design Assumptions When Analyzing Data from Public-Use Complex Surveys

• Activity Number: 422
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Van Parsons*+ and Jennifer D Parker
• Companies: National Center for Health Statistics and National Center for Health Statistics
• Keywords: BRR ; linearization ; poststratification ; degrees-of-freedom

Abstract

Large national population-based surveys are often based on multi-stage cluster sampling. When using design-based estimation methods on such surveys, knowledge of the complete hierarchical sampling structure is required to correctly assess the sampling distributions of estimators. However, for public-use data releases, the survey design information for analyses is substantially simplified. For example, the NHIS is described as a stratified set of independently sampled first-stage clusters with the final adjusted survey weights to be used as fixed sampling weights. While such simplified design structures can be justified under hypothetical sampling conditions, those sampling conditions rarely hold in practice. To study the impact of simplified survey analytic structures, a finite pseudo population of 340,000 households has been created using nine years of NHIS. This population captures many of the clustering features present in the U.S. population. For this paper, sampling methods and estimation methods consistent with those of the NHIS are studied on this pseudo population. Evaluations are presented under true and simplified assumptions.