Single-Stage Generalized Raking Weight Adjustment in the Current Population Survey

• Activity Number: 105
• Type: Invited
• Date/Time: Monday, August 5, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Eric Victor Slud*+ and Reid Rottach and Christopher Grieves
• Companies: US Census Bureau and US Census Bureau and US Census Bureau
• Keywords: nonresponse ; calibration ; loss function ; weight trimming ; quadratic programming ; weight tracking

Abstract

This research concerns the adaptation to the Current Population Survey (CPS) of single-stage weight adjustment techniques developed in a 2010 Census research report by Slud and Thibaudeau. Those techniques involved weight optimization with respect to a loss function in the spirit of Deville and Sarndal (1992, JASA), subject to population-control constraints, with additive penalty terms for discrepancies between weight-adjusted survey totals and corresponding known or base-weighted estimated totals for certain survey attributes, and with an additional nonlinear penalty term designed to force weights not to be too different from design weights scaled to the population total. The novel elements of the current research include: defining several appropriate quadratic penalty terms corresponding to the current multistage CPS nonresponse adjustment; developing a methodology to define penalty multipliers by tracking properties of the current CPS weights across weighting stages; enforcing weight compression by a penalty term in place of the current CPS approach based on cell collapsing; and implementing the method on CPS data for detailed comparison with the weights as currently adjusted in CPS.