Generalized Bayesian Inference for Causal Effects Using the Covariate Balancing Procedure.

Area of Science:

• Biostatistics
• Epidemiology
• Causal Inference

Background:

• Propensity scores are crucial for estimating causal effects in observational studies.
• The inverse probability weighting (IPW) estimator is widely used but sensitive to propensity score model misspecification.
• Existing robust methods require complex parameter considerations.

Purpose of the Study:

• To propose a novel Bayesian estimating procedure for propensity scores.
• To address limitations of existing robust propensity score methods.
• To enable more reliable causal effect estimation in observational studies.

Main Methods:

• Developed a Bayesian procedure for propensity score estimation.
• Leveraged the general Bayesian paradigm applicable to loss functions.
• Avoided full likelihood considerations, requiring standard causal inference assumptions.

Main Results:

• The proposed Bayesian method achieved equal or superior performance in simulation experiments compared to previous methods.
• Demonstrated robustness against propensity score model misspecification.
• Successfully applied to real-world data, including the Whitehall dataset.

Conclusions:

• The novel Bayesian propensity score estimation procedure offers a robust alternative.
• This method enhances the reliability of causal effect estimates from observational data.
• The approach is flexible and requires minimal additional assumptions.