Redundant variables and Granger causality.

Area of Science:

• Neuroscience
• Complex Systems Analysis
• Information Theory

Background:

• Multivariate Granger causality is a key tool for analyzing causal relationships in time-series data.
• Standard Granger causality analysis can be confounded by the presence of redundant variables, leading to underestimation of true causal influences.
• Understanding information flow is crucial for deciphering complex physiological processes, such as those underlying neurological disorders.

Purpose of the Study:

• To address the underestimation of causality in multivariate analyses caused by redundant variables.
• To quantitatively define and identify redundancy and synergy within causal frameworks.
• To develop novel methods for grouping redundant variables to improve causal inference.

Main Methods:

• Utilized the un-normalized version of the Granger causality index.
• Developed quantitative measures for redundancy and synergy.
• Proposed two distinct approaches for grouping redundant variables: target-based maximization and whole-set partitioning.
• Applied the methodology to neurological data from a migraine experiment.

Main Results:

• Demonstrated that standard Granger causality underestimates causalities in the presence of redundant variables.
• Successfully quantified notions of redundancy and synergy.
• The proposed grouping methods effectively managed redundant variables.
• Revealed significant changes in the informational patterns of the migraine brain following repetitive transcranial magnetic stimulation.

Conclusions:

• The un-normalized Granger causality index offers a robust way to handle redundant variables in complex systems.
• The developed methods enhance the accuracy of causal inference in multivariate time-series analysis.
• This approach provides deeper insights into the neurophysiological basis of migraine and the effects of brain stimulation.