Comment on "Comparison of Six Methods for the Detection of Causality in a Bivariate Time Series"
Abstract
Krakovska et. al. suggest that convergent cross mapping (CCM) performance was among the `least successful' of the methods tested to detect causality due to a preponderance of false-positive results when detecting unidirectional causality. This conclusion stems from the method's detection of bidirectional causation when the description of the known state space dynamics is defined with a unidirectional link between coupled subsystems. The authors admit that this is not contrary to the theory of CCM in that it only claims to produce high cross mapping skill when the link exists, not low skill when no link exists. Though this point is true, this observation misses a critical note on the impact of generalized synchronization in the theory of CCM which motivated the development of extended CCM in subsequent work. Furthermore, because the issue of misidentified bidirectionality has been cited as a concern for the application of CCM in several recent publications and is used to justify caution against the use of CCM on real-world data, the issue warrants further discussion.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 15, 2020
- Accession Number
- AD1127390
Entities
People
- Robert R. S. Martin
Organizations
- Air Force Research Laboratory