On the statistical performance of Granger-causal connectivity estimators

Area of Science:

• Neuroscience
• Statistics
• Signal Processing

Background:

• Assessing linear connectivity is crucial for understanding complex systems.
• Previous evaluations of statistical detection performance exist.
• Multivariate Granger causality tests are widely used but require robust performance evaluation.

Purpose of the Study:

• To extend the statistical detection performance evaluation of linear connectivity.
• To investigate the behavior of various Granger causality measures under different data lengths.
• To identify anomalies in conditional multivariate Granger causality.

Main Methods:

• Conducted Monte Carlo simulations on three established toy models.
• Evaluated a classic time domain multivariate Granger causality test.
• Assessed information partial directed coherence and information directed transfer function.
• Included conditional multivariate Granger causality in the analysis.

Main Results:

• Performance varied across different data record lengths.
• Information partial directed coherence and information directed transfer function showed expected behaviors.
• Conditional multivariate Granger causality exhibited anomalous behavior, affecting detection accuracy.
• The study provides a comprehensive performance evaluation of these connectivity measures.

Conclusions:

• The findings highlight the importance of rigorous performance evaluation for causality detection methods.
• Anomalous behavior in conditional multivariate Granger causality necessitates caution in its interpretation.
• This research contributes to a better understanding of linear connectivity assessment in time series data.