B0689
Title: Handling nonignorable nonresponse by using semiparametric fractional imputation for complex survey data
Authors: Sixia Chen - University of Oklahoma (United States) [presenting]
Abstract: Nonignorable nonresponse happens frequently in biomedical studies, including tobacco cessation, health disparities, and cancer research. In practice, most studies assumed missing at random in statistical analysis. However, this assumption might lead to biased results when the assumption is not randomly missing. Fully parametric approaches are vulnerable to parametric model assumptions. Fully nonparametric approaches are inefficient and might suffer from the curse of dimensionality. A novel semiparametric fractional imputation approach is proposed with a parametric model for the response mechanism and a semi-parametric model for the outcome regression model. Specifically, the strength of the empirical likelihood method is borrowed to construct fractional weights. The proposed method is further extended for incorporating multiple outcome regression and/or nonresponse models. The proposed methods can be used for handling complex survey data. The Monte Carlo simulation study shows the benefits of the proposed methods compared to some existing methods. The proposed methods are further evaluated by some real data applications.
