Partial cross mapping eliminates indirect causal influences.

Area of Science:

• Nonlinear dynamics
• Complex systems analysis
• Causal inference

Background:

• Causality detection methods often misinterpret indirect effects as direct ones due to causation transitivity.
• Existing methods struggle with non-separable dynamical systems with weak or moderate interactions.
• Distinguishing direct from indirect causation is vital for understanding complex system mechanisms.

Purpose of the Study:

• To develop a data-based, model-independent method for identifying direct causations from indirect ones.
• To address the challenge of disentangling causal effects in complex, interacting systems.
• To provide a tool for accurate causal discovery in real-world data.

Main Methods:

• Developed a partial cross-mapping technique integrating phase-space reconstruction, mutual cross mapping, and partial correlation.
• The method is data-driven and does not rely on pre-defined system models.
• Validated using data from diverse representative models and real-world systems.

Main Results:

• Successfully differentiated direct and indirect causal links in complex dynamical systems.
• The proposed method demonstrated efficacy even with non-separable variables and weak interactions.
• Empirical validation across various systems confirmed the method's robustness.

Conclusions:

• The novel partial cross-mapping method effectively identifies direct causations in complex systems.
• This approach overcomes limitations of traditional methods in analyzing interacting variables.
• Anticipated to be essential for deciphering mechanisms in diverse scientific disciplines using data.