Bayesian inference for causal mediation effects using principal stratification with dichotomous mediators and...

Area of Science:

• Social and health sciences
• Causal inference
• Bayesian statistics

Background:

• Investigating causal relationships between risk factors and outcomes is central to health and social sciences.
• Understanding direct and mediated effects is key to decomposing complex causal pathways.
• Previous methods face challenges with dichotomous mediators and outcomes.

Purpose of the Study:

• To develop a Bayesian approach for estimating direct and mediated effects.
• To address the challenges of non-identifiable parameters in causal inference with dichotomous variables.
• To apply the method to understand mediating effects of adult poverty on mortality risk.

Main Methods:

• Utilized the potential outcome framework for causal inference.
• Defined principal strata based on mediator counterfactuals.
• Developed a Bayesian approach to estimate effects using posterior distributions.
• Performed sensitivity analyses with varying prior distributions.

Main Results:

• The Bayesian approach allows estimation of direct and mediated effects even with non-identifiable parameters.
• Nonparametric bounds for causal effects were calculated under randomized treatment assignment.
• Sensitivity analyses demonstrated the robustness of the findings to prior assumptions.
• The method was applied to analyze the mediating role of adult poverty.

Conclusions:

• The proposed Bayesian method provides a flexible framework for estimating direct and mediated effects.
• This approach enhances causal inference in situations with dichotomous mediators and outcomes.
• The findings offer valuable insights for public health and socioeconomic research, exemplified by the poverty and mortality study.