Title: An alternative sensitivity analysis approach for missing not at random data
Authors: Chiu-Hsieh Hsu - University of Arizona (United States)
Chengcheng Hu - University of Arizona (United States) [presenting]
Yulei He - CDC (United States)
Abstract: Missing mechanism is unverifiable. Often researchers perform sensitivity analysis to evaluate the impact of various missing mechanisms. All the existing sensitivity analysis approaches for missing not at random (MNAR) data require fully specifying the relationship between the missing value and the missing probability or simply use the delta adjustment approach. The relationship is specified using a selection model, a pattern-mixture model, or a shared parameter model. We propose an alternative sensitivity analysis approach for MNAR using a nonparametric multiple imputation approach, which only requires specifying the correlation between the missing value and the missing probability. The correlation is a standardized measure and can be directly used to indicate the magnitude of MNAR. We perform simulation studies to compare the proposed sensitivity analysis approach and the delta adjustment procedure. Numerical results indicate that the proposed approach performs well and can be used as an alternative approach for MNAR. The proposed sensitivity analysis approach is demonstrated on hemoglobin A1c data of open heart surgery patients, which are subject to missing especially for non-diabetic patients.