查看文章

Multiple imputation (MI) has become one of leading approaches in dealing with missing data in survey research. However, existing software packages and procedures typically do not incorporate complex sample design features in the imputation process. Researcher has demonstrated that implementation of MI based on simple random sampling (SRS) assumption can cause severe bias in estimation and hence invalid inferences, especially when the design features are highly related to survey variables of interest (Reiter et al. 2006). Recent work to accommodate complex sample designs in imputation has focused on model-based methods which directly model the complex design features in the formulation of the imputation model. In this paper, we propose a semi-parametric procedure as an alternative approach to incorporate complex sample designs in MI. Specifically, we divide the imputation process into two stages: the complex feature of the survey design (including weights and clusters) is fully accounted for at the first stage, which is accomplished by applying a nonparametric method to generate a series of synthetic datasets; we then perform conventional parametric MI for missing data at the second stage using readily available imputation software designed for an SRS sample. A new combining rule for the point and variance estimates is derived to make valid inferences based on the two-stage procedure. Using health survey data from the Behavior Risk Factor Surveillance System, we evaluate the proposed method with a simulation study and compare it with the model-based method with respect to complete data analysis. Results show …