Causal inference with noisy data: Bias analysis and estimation approaches to simultaneously addressing missingness...

Area of Science:

• Statistics
• Biostatistics
• Epidemiology

Background:

• Causal inference methods often fail with noisy data.
• Binary outcomes can suffer from both missingness and misclassification.
• Accurate estimation of average treatment effects (ATE) is crucial.

Purpose of the Study:

• To develop valid statistical methods for estimating ATE with binary outcomes affected by missingness and misclassification.
• To investigate the impact of ignoring missingness and misclassification on ATE estimation.
• To propose robust methods that account for potential model misspecification.

Main Methods:

• Examination of asymptotic biases caused by missingness and misclassification.
• Development of weighted estimation methods for simultaneous correction.
• Proposal of a doubly robust correction method for enhanced protection against model misspecification.

Main Results:

• Established intrinsic connections between missingness and misclassification effects on ATE estimation.
• Developed weighted and doubly robust estimators that provide valid corrections.
• Simulation studies demonstrated the performance of the proposed methods.

Conclusions:

• The proposed methods effectively address challenges in causal inference with binary outcomes affected by missingness and misclassification.
• Doubly robust methods offer protection against model misspecification.
• The methods are applicable to real-world data, as shown in a smoking cessation study.