Data-driven causal analysis of observational biological time series

Area of Science:

• Ecology
• Complex Systems Analysis
• Statistical Inference

Background:

• Complex systems often preclude manipulative experiments, necessitating alternative methods for causal inference.
• Observational time series analysis offers approaches to infer causal relationships across various scientific fields.
• Existing causal discovery methods can be prone to misinterpretation and controversy.

Purpose of the Study:

• To provide an accessible and critical review of three statistical causal discovery approaches.
• To evaluate the applicability and limitations of pairwise correlation, Granger causality, and state space reconstruction in ecological contexts.
• To clarify the assumptions and potential pitfalls associated with these causal inference methods.

Main Methods:

• Review and critical analysis of pairwise correlation, Granger causality, and state space reconstruction.
• Application of methods to ecological process examples.
• Development of novel visualizations for key concepts.
• Identification of method pathologies and assumption-free limitations.

Main Results:

• Each method (pairwise correlation, Granger causality, state space reconstruction) has specific testing targets and implications for causal statements.
• Novel visualizations enhance understanding of complex causal discovery concepts.
• Existing 'model-free' methods are not assumption-free, revealing hidden biases.
• Specific pathologies and limitations of each method are identified in ecological examples.

Conclusions:

• A clear understanding of assumptions is crucial for the appropriate application of causal discovery methods.
• This synthesis aims to improve communication and application of causal inference techniques across disciplines.
• Explicitly stating assumptions strengthens the validity and interpretability of causal discovery findings.