On causality of extreme events

Area of Science:

• Complex systems analysis
• Causality detection
• Data science

Background:

• Existing causality metrics primarily focus on time-series data.
• Detecting causality in static datasets, especially non-linear relationships, remains a challenge.
• Confounding factors can often be mistaken for true causal links.

Purpose of the Study:

• To propose a new metric for identifying causality within static datasets.
• To develop a method capable of detecting non-linear causalities.
• To differentiate true causalities from mere correlations in static data.

Main Methods:

• Analysis of extreme event occurrences between elements in static datasets.
• Application to cross-sectional and longitudinal data.
• Validation using synthetic, dynamical, chaotic, and human brain activity datasets.

Main Results:

• The proposed metric successfully detects non-linear causalities in static data.
• It effectively distinguishes genuine causal links from correlations due to confounding factors.
• Performance surpasses classical metrics when non-linearities and large datasets are present.

Conclusions:

• A robust metric for static causality detection has been established.
• This method offers a powerful tool for analyzing complex systems where time-series data is unavailable.
• The metric provides a significant advancement in understanding causal relationships in diverse data types.