A0312
Title: Model-assisted weighting methods for improving robustness and efficiency of relative risk estimation in the population
Authors: Lingxiao Wang - University of Virginia (United States) [presenting]
Abstract: Model-assisted calibration weighting methods are widely used to improve the robustness and efficiency of the weighted estimator in survey statistics. Calibration on auxiliary variables has been shown not only to gain efficiency in estimating finite population total/mean when the auxiliary variables are highly correlated with the outcome of interest but also to reduce bias if the calibration totals are error-free. However, for regression coefficients, such as linear association, relative risks, and hazard ratios, calibration on auxiliary variables only does not improve efficiency. Model-assisted calibration weighting methods are presented using novel auxiliary variables generated from the estimating equations for parameters of interest to obtain robust and efficient estimators. The proposed method can be applied under various data structures, such as two-phase samples and data integration. The proposed methods are applied to estimating hazard ratios of lung cancer incidence from a large-scale volunteer-based epidemiologic cohort by using the National Health Interview Survey (NHIS) as the reference.
