A0512
Title: Multiply robust estimation for two-part regression models with missing semi-continuous response
Authors: Chunlin Wang - Xiamen University (China) [presenting]
Qiyin Zheng - Xiamen University (China)
Abstract: Two-part regression models are widely used for analyzing semicontinuous response data consisting of a mixture of excess zeros and skewed positive continuous data. The problem of missing semicontinuous response data is often encountered in many applications, and simply ignoring it may lead to erroneous conclusions. Multiple robust estimation procedures are proposed for the two-part regression parameters to allow for multiple candidate models for both the missing mechanism and imputation. The multiple robustness properties of the proposed estimators are established in the sense that they are consistent if any one of these multiple models is correctly specified. Other methods, including inverse probability weighting, imputation, and doubly robust estimators, have also been constructed and compared in simulation studies. The simulation results show the desirable finite-sample performance of the proposed multiply robust estimators under a variety of model settings. Real psychology data with the missing semicontinuous responses is analyzed for illustration.
