Simultaneous Calibration and Nonresponse Adjustment with Explicit Variance Formulas

• Activity Number: 563
• Type: Topic Contributed
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Eric Victor Slud and Yves Thibaudeau*+
• Companies: U.S. Census Bureau and U.S. Census Bureau
• Keywords: selection probabilities ; nonresponse adjustments ; design-consistency ; linearization formula

Abstract

Statistical agencies invest considerable resources in the scientific design of survey samples, one aspect of which is the availability of explicit unit and joint selection probabilities. In textbook theory, these selection probabilities enable design-consistent estimation of the variance through linearization formulae. In practice, however, the inverse selection- probability weights are subject to multiple rounds of adjustments to reflect population benchmarks and to compensate for nonresponse. After these operations the weights no longer have a clear interpretation. Then, formula-based design-consistent variance estimation methods are often not longer easily available, and replication techniques are used instead. We propose a comprehensive optimization-based method for adjusting the initial survey weights for population benchmarking and nonresponse and also accommodating weight compression or smoothing, in a single operation. The benefit is a unified theoretically supported weight adjustment with guaranteed bounds on the weights, and which maintains design-consistency under nonresponse or linear-model assumptions.