Single Versus Replicate Survey Weights: Toward Guidance for Choosing an Optimal Approach

• Activity Number: 669
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Jamie Ridenhour*+ and Andy Peytchev
• Companies: RTI International and RTI International
• Keywords: Taylor linearized weights ; replicate weights

Abstract

Creation of replicate survey weights affects estimation methods and can lead to very different variance estimates compared to the arguably more common computation of a single survey weight variable and use of Taylor Series Linearization. The choice seems to be based on familiarity with one over the other or made at the institution level, rather than based on evidence or criteria that could lead to the selection of the optimum method given a particular study. Replicate weights can overestimate the true sampling variances while the Taylor linearized weights can underestimate the variance reduction from using know population control totals. This study aims to tease out conditions that make either weighting and estimation approach more optimal, particularly with respect to the variables used in weighting and their relationship to the survey variables.