An Alternative Raking Approach to Reduce Design Effects

• Activity Number: 541
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): Nadarajasundaram Ganesh* and Vicki Pineau and Kennon Copeland and Zhen Zhao and Philip J. Smith and Meena Khare and David Yankey
• Companies: NORC at the University of Chicago and NORC at the University of Chicago and NORC at the University of Chicago and CDC and CDC and National Center for Health Statistics and CDC
• Keywords: calibration ; design effect ; noncoverage ; nonresponse ; raking ; weighting

Abstract

Nonresponse and raking adjustments can adversely impact the coefficient of variation of survey weights and standard errors for survey estimates. In this study, we investigate a raking approach that attempts to mitigate the impact on standard errors when weights are raked. Raking involves minimizing the distance between raked and pre-raked weights while also requiring the sum of the weights agree with population totals. In addition to this requirement, our modified approach also requires that the distance between the raked weights and mean of the raked weights within each stratum be minimized. Using 2012 and 2013 National Immunization Survey data, we compare our approach with the traditional raking approach in terms of standard error and mean squared error. Preliminary results indicate that at the national-level, the proposed method yields similar estimates as the official NIS estimates (absolute difference less than 0.26 percentage points), but decreases the standard error by 0.03 percentage points. In comparison, the unweighted estimate decreases the standard error by 0.20 percentage points, but has a difference of up to 2.15 percentage points compared to the official estimate.