A Comparison of Statistical Methods for Standardized Estimates and Confidence Intervals with Survey Data

• Activity Number: 437
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Government Statistics Section
• Author(s): Yi Mu*+
• Companies: Centers for Disease Control and Prevention
• Keywords: Standaridized Estimates ; Confidence Intervals ; Survey ; Predicted Marginals ; Modeling

Abstract

Researchers are often interested in making inference for the population from survey. This paper presents a discussion of standardized estimates and confidence intervals using two different classes of approaches: direct standardization and modeling procedures. Direct standardization is further analyzed as simple random sampling design (DSR1) with Gamma distribution as confidence intervals and direct standardization for clustered stratified sampling design (DSR2). DRI1 is simple to implement and is useful when design features of a study is unknown. DIR2 can account for the complex design features of a study. Modeling procedures are designed based and models are developed from sampled data. The sampled weight estimators are either directly applied to population or used to calculate predicted marginals to obtain standardized estimates. Predicted marginals is an appealing method because it is a generalization of the direct standardization, shares all advantages of modeling procedures and can be estimated using some standard statistical software products. Results are demonstrated using data collected for Clostridium Difficile infection across multiple U.S. geographic locations.