Direct and Indirect Effects-An Information Theoretic Perspective.

Area of Science:

• Causal inference
• Information theory
• Complex systems analysis

Background:

• Information theoretic (IT) approaches are popular for quantifying causal influences in fields like neuroscience and climate science.
• Existing IT causal measures are model-agnostic and capture non-linear interactions but differ from statistical notions by comparing effect distributions and using random variables for causes.
• A key limitation is the inability to measure causal effects for specific cause values.

Purpose of the Study:

• To introduce novel IT measures for quantifying direct, indirect, and total causal effects.
• To develop IT measures that enable causal effect assessment with respect to specific cause values.
• To provide an identifiability result for the proposed causal measures.

Main Methods:

• Development of new information theoretic measures for causal effect quantification.
• Formulation of measures for direct, indirect, and total causal effects.
• Demonstration of identifiability for the proposed measures.

Main Results:

• The proposed IT measures allow quantification of causal effects defined with respect to specific cause values.
• These measures retain the flexibility and general applicability of existing IT techniques.
• Successful application in estimating the causal effect of El Niño-Southern Oscillation on North American Pacific Northwest temperature anomalies.

Conclusions:

• The novel IT measures offer a more precise way to quantify causal relationships, especially when specific causal values are of interest.
• This approach enhances the utility of IT in fields requiring granular causal analysis.
• The method is validated through a climate science application, demonstrating its practical relevance.