Confounding effects of indirect connections on causality estimation.

Area of Science:

• Neuroscience
• Complex Systems Analysis
• Information Theory

Background:

• Effective connectivity research often overlooks indirect influences between system components.
• Bivariate measures of causality can be misled by unobserved common causes or mediating pathways.
• Robust causal inference requires accounting for the broader network environment.

Purpose of the Study:

• To introduce and evaluate partial transfer entropy for assessing effective connectivity.
• To quantify indirect coupling effects within complex networks.
• To investigate the impact of indirect connections and model misspecification on causal link detection.

Main Methods:

• Development of a multivariate transfer entropy-based Granger causality measure.
• Quantification of indirect coupling mediated by the network environment.
• Comparison of partial transfer entropy with bivariate transfer entropy for causal inference.

Main Results:

• Partial transfer entropy demonstrates higher sensitivity in identifying robust causal relations than bivariate methods.
• Variations in indirect coupling significantly confound the detection of true causal links.
• Model misspecification poses challenges to connectivity analysis, even with information-theoretic approaches.

Conclusions:

• Partial transfer entropy offers a more reliable method for inferring causal relationships in complex systems.
• Understanding and accounting for indirect effects are crucial for accurate effective connectivity analysis.
• Robustness of causal inference is sensitive to both network structure and model assumptions.