Comparing Alternative Weight Adjustment Methods

• Activity Number: 671
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Kimberly Henry*+ and Richard Valliant
• Companies: IRS/SOI and University of Maryland
• Keywords: model-assisted estimation ; model-based estimation ; weight trimming ; p-splines ; weight smoothing

Abstract

Several design-based, model-based, and model-assisted methods have been developed to adjust survey weights for nonresponse or coverage errors, to reduce variances through the use of auxiliary data or by restricting the range of the weights themselves. Some methods directly change the weights, like calibration weighting and design-based ad hoc weight trimming methods. Other methods implicitly adjust the weights, like robust superpopulation modeling approaches. The generalized design-based method models the weights as a function of the survey response variables and using the smoothed weights predicted from the model to estimate finite population totals. This paper provides empirical examples of how several adjustment methods change a given sample's weights and the resulting impact on estimates.