Causation Entropy Identifies Indirect Influences, Dominance of Neighbors and Anticipatory Couplings

TL;DR

This paper introduces causation entropy, a new measure that improves the accuracy of inferring true causal relationships in complex, nonlinear, and time-dependent networks by overcoming limitations of transfer entropy.

Contribution

The paper develops causation entropy, a novel causality measure that reliably identifies direct influences in nonlinear and dynamic networks, surpassing transfer entropy.

Findings

Causation entropy outperforms transfer entropy in detecting true causal links.

Transfer entropy often misidentifies indirect influences and anticipatory couplings.

Causation entropy provides more accurate network inference in complex systems.

Abstract

Inference of causality is central in nonlinear time series analysis and science in general. A popular approach to infer causality between two processes is to measure the information flow between them in terms of transfer entropy. Using dynamics of coupled oscillator networks, we show that although transfer entropy can successfully detect information flow in two processes, it often results in erroneous identification of network connections under the presence of indirect interactions, dominance of neighbors, or anticipatory couplings. Such effects are found to be profound for time-dependent networks. To overcome these limitations, we develop a measure called causation entropy and show that its application can lead to reliable identification of true couplings.