Mediation analysis with the mediator and outcome missing not at random.

Area of Science:

• Biostatistics
• Causal Inference
• Statistical Modeling

Background:

• Mediation analysis is crucial for understanding causal pathways.
• Missing data in mediators and outcomes poses a significant challenge.
• Missing not at random data prevents identification of effects without assumptions.

Purpose of the Study:

• To investigate the identifiability of direct and indirect effects under missing not at random mechanisms.
• To develop and evaluate statistical methods for mediation analysis with missing data.
• To apply these methods to real-world data, such as the National Job Corps Study.

Main Methods:

• Developing interpretable mechanisms for missing not at random data in mediators and outcomes.
• Conducting simulation studies to assess statistical inference performance.
• Illustrating the proposed methods using data from the National Job Corps Study.

Main Results:

• Identifiability of direct and indirect effects is achievable under specific interpretable mechanisms for missing not at random data.
• The proposed statistical inference methods demonstrate robust performance in simulations.
• The methods are successfully applied to a practical dataset, showing their utility.

Conclusions:

• The study provides a framework for robust mediation analysis with missing not at random data.
• The developed methods enhance causal inference capabilities in the presence of missing data.
• This research offers valuable tools for researchers dealing with incomplete datasets in mediation studies.