Depicting deterministic variables within directed acyclic graphs: an aid for identifying and interpreting causal...

Area of Science:

• Causal Inference
• Data Analysis Methodology
• Statistical Modeling

Background:

• Deterministic variables are functionally dependent on parent variables, common in derived variables and compositional data.
• Accurate causal effect interpretation is challenging with these variables.
• Existing methods may not fully address the complexities introduced by deterministic relationships.

Purpose of the Study:

• To introduce a method for depicting deterministic variables within directed acyclic graphs (DAGs).
• To enhance the identification and interpretation of causal effects involving derived and compositional data.
• To provide a framework for analyzing data with functionally determined variables.

Main Methods:

• Proposing a 2-step approach for variable consideration within DAGs.
• Visualizing deterministic variables and their parent-child relationships.
• Applying DAGs to scenarios with derived variables and compositional data.

Main Results:

• DAGs facilitate the identification and avoidance of tautological associations.
• Improved understanding of conditioning on 'whole' variables in compositional data.
• Enhanced scrutiny of consistency and exchangeability assumptions for derived variables.

Conclusions:

• Depicting deterministic variables in DAGs aids in planning and interpreting analyses.
• This graphical approach clarifies causal pathways involving derived and compositional data.
• DAGs offer a valuable tool for researchers dealing with functionally determined variables.