Comparison of Alternative Variance Estimators for Raking in the Presence of Nonresponse

• Activity Number: 169 - SPEED:Improving Survey Data Quality with Multiple Data Sources, Administrative Data, and Nonresponse Bias Control
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): Daifeng Han* and Richard Valliant
• Companies: Westat and University of Maryland and University of Michigan
• Keywords: raking; nonresponse adjustment; variance estimation; linearization; replication

Abstract

Raking is a widely used calibration method in government surveys. In the context of using single-step calibration weighting to reduce nonresponse bias, D’Arrigo and Skinner (2010) evaluated the properties of the GREG estimator, raking ratio estimator, and maximum likelihood raking estimator as well as the performance of several linearization variance estimators. The alternative forms of linearization variance estimators were defined via the choices of (1) the weights applied to the residuals from the regression model; and (2) the weights used in the regression model to estimate regression coefficients and residuals. In this paper, we examine several alternative variance estimators for raking in the presence of nonresponse, including the linearization estimators defined in D’Arrigo and Skinner (2010) and a jackknife replication variance estimator. Our simulation results show that when raking is model-biased, none of the linearization variance estimators under evaluation is unbiased. In contrast, the jackknife replication method performs well in variance estimation, although the confidence interval may still be centered in the wrong place if the point estimate is inaccurate.