No abstract

The practice of identifying structure of couplings between elements of complex systems from experimental recordings of their oscillations (time series) using the Wiener – Granger causality method has revealed a number of problems that prevent one from obtaining reliable results. In particular, the presence of observational noise can lead to the “effect of spurious coupling”, i.e. to inference of mutual coupling between two elements that are actually coupled in a unidirectional way. A quantitative analysis of this phenomenon is carried out and recommendations allowing one to reduce its probability are presented. It is shown that the effect typically takes place only for large noise levels comparable to the level of observed oscillations. However, we have also singled out less typical situations where the effect occurs at much weaker noise.