Using causal diagrams to guide analysis in missing data problems.

Area of Science:

• Causal inference
• Missing data theory
• Epidemiology

Background:

• Estimating causal effects from incomplete datasets necessitates untestable assumptions about missing data mechanisms.
• Rubin's classification (MCAR, MAR, MNAR) addresses missingness but requires careful handling for causal effect estimation.

Purpose of the Study:

• To demonstrate how causal diagrams can clarify assumptions for causal effect estimation with incomplete data.
• To extend existing causal inference criteria, such as Pearl's back-door criterion, to scenarios with missing data.

Main Methods:

• Utilizing causal diagrams to represent and analyze assumptions related to missing data mechanisms.
• Formally extending the back-door criterion for application in incomplete data settings.

Main Results:

• Causal diagrams provide a clear framework for understanding missing data assumptions.
• The extended back-door criterion offers a method for unbiased causal effect estimation from incomplete data.

Conclusions:

• Causal diagrams are valuable tools for navigating the complexities of missing data in causal inference.
• The proposed extension facilitates more robust estimation of causal effects in observational studies with missing data.