Longitudinal Survey Weight Calibration with Estimated Totals

• Activity Number: 671
• Type: Topic Contributed
• Date/Time: Thursday, August 8, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): Siyu Qing*+ and Michael D. Larsen
• Companies: George Washington University and The George Washington University
• Keywords: Panel survey ; Calibration estimation ; Variance estimation ; Sample survey ; Cross-sectional study ; Post stratification

Abstract

In longitudinal surveys, sampled individuals appear in multiple survey years and can be linked across time. In the NSF's Survey of Doctorate Recipients (SDR), 60% of respondents appear on 3 or more surveys from 1993-2006. In some of these surveys, such as the SDR, survey weights are created on a cross-sectional basis. Longitudinal weights enabling estimation using data from multiple survey waves together do not exist. This paper applies calibration estimation for construction of such a longitudinal weight with the help of auxiliary variables in each survey year. Population totals of auxiliary variables are an influential factor in the calibration procedure but are unavailable. Instead, cross-sectional estimated totals are used. Due to this replacement, variance could be seriously underestimated when using the traditional calibration variance estimator. A new variance estimator is derived that can mitigate the negative bias. Performance of the estimator is studied using simulation and application to SDR survey data. Choices of multivariate calibration targets are compared. Other methods such as jackknife and bootstrap for variance estimation also are discussed.