arising from Albert C. Yang et al. Nature Communications https://doi.org/10.1038/s41467-018-05845-7 (2018)
The empirical mode decomposition (EMD) method proposed in Yang et al.1 fails to correctly identify causal relationships for a system of two independent variables driven by a shared external forcing (aka Moran effect). Using a simple, two-species Moran effect model (Fig. 1a), it is obvious that the EMD method erroneously concludes that N1 and N2 have a causal relationship (IMF 1 in Fig. 1b), although in fact they do not. This is because effects of external forcing were recorded in both N1 and N2 time series, and at least one IMF decomposed from N1 and N2 time series is associated with that external forcing. Therefore, removal of such IMF from N1 makes the remaining less coherent with N2, and vice versa. As such, EMD methods based on diminished coherence due to IMF removals fail to falsify spurious causations caused by sharing external forcing. In contrast, convergent cross mapping (CCM) correctly identifies the lack-of causal relationship between N1 and N2 (Fig. 1c). The efficacy of CCM to distinguish Moran effects depends on the strength of the shared external forcing. In extreme cases, when external forcing is too strong, N1 and N2 are synchronized (the pathological case, as noted in a previous study2). Methods to cope with such situation have also been developed3,4.
Second, Yang et al.1 argued that CCM provides incorrect causal relationships. However, Yang et al.1 used CCM in a manner that the original authors2 did not intend, producing incorrect conclusions. Specifically, following McCracken et al.5, Yang et al.1 used correlation difference (X cross-map Y – Y cross-map X) as the definition of CCM causation (Fig. 3 in ref. 1) without examining the convergence of the cross-mapping skill; this is an incorrect definition. The correct definition of causation under CCM is improvement of cross-mapping skill with increasing time series length (i.e., convergence). In addition, as in real systems, CCM causation can be bidirectional2. Yang et al.1 used an incorrect definition (i.e., correlation difference) and incorrectly concluded that CCM misidentified the lynx versus hare and Didinium versus Paramecium interactions as top-down control systems and the Lotka Volterra predator–prey model and wolf versus moose interactions as either no or confusing causation. However, in each of these examples, CCM exhibits clear convergence with increasing library size in both directions. By using the convergence definition of causation in CCM, we concluded that these prey–predator systems exhibited bidirectional causation. That is, CCM correctly identified the reciprocal nature of predator–prey interactions in all of these systems (Fig. 5 in ref. 1). As such, we suggest any description of coupling in predator–prey systems as “directional” claimed by Yang et al.1 may be misleading (Fig. 3 in ref. 1) because predators causally influence prey by consuming them and prey causally influence predators by providing them the energy needed for population growth. Moreover, the relative strength of each direction can be quantified based on the rate of convergence2,6, with proper consideration of potential lagged effects3. We also disagree with the claim that CCM incorrectly identified causal coupling in white noise. Again, this also stems from using an incorrect definition of CCM. In contrast, we find no evidence of convergence when applying CCM to paired white noise signals (Fig. 2a) and a false positive rate consistent with p = 0.05 as the level of significance (Fig. 2b, c).
Several additional misunderstandings about CCM in Yang et al.1 warrant clarification. (1) CCM does not rely on predictability as the criterion. Rather, CCM relies on information recovering2,6 that identifies whether the present state of an effect variable contains information about the present state of causal variables7 (i.e., nowcast) and thus enables CCM to identify simultaneous influences. (2) Oscillatory dynamics may confound the efficacy of CCM; however, methods to remove cycles or construct a null model that accounts for cycles (e.g., seasonality) have been developed4,8,9,10. (3) Yang et al.1 stated that “CCM is developed under the constraints of perfect deterministic system”; this is incorrect. In reality, real-world systems usually contain a deterministic skeleton convolved with stochastic processes11. In fact, modeling and empirical examples have demonstrated that CCM correctly identifies causation, even when stochastic processes are convolved with deterministic signals2,6,12.
It is noteworthy that interpretation of findings based on IMF subtraction needs to be done with caution. Yang et al.1 included a strong statement regarding interpreting their findings, stating that removals of causal-related IMFs enable us to exclusively recover intrinsic dynamics of the target series from the residual IMFs. Although this statement is correct when the system dynamic is a result of superposition of signals, it is not always correct for general dynamical systems in which causal influences cannot be easily separated from intrinsic dynamics. Using a simple example of prey–predator model, \(\frac{dx}{dt}=\alpha x-\beta xy;\frac{dy}{dt}=\delta xy-\gamma y\), we have a clear expectation that the prey will grow exponentially in the absence of the predator, or at the very least, prey cannot oscillate. However, these expectations are not realized after subtracting causal IMFs; rather, the remaining components continue to cycle—quite at odds with intuition based on the statement provided by Yang et al.1. The salient point is that interpretation of the residual IMF is not as unambiguous as the original text implies. Certainly, IMF subtraction is not equal to mathematical subtraction. However, real-world biologists applying Yang et al.’s approach1 to predator–prey systems may be confused when interpreting the results according to Yang et al.’s statement about separability in causal inference. In fact, after subtracting the effect of the predator on the prey (e.g., Fig. 1c in ref. 1), prey continue to oscillate. Based on countless chemostat experiments, prey grown in isolation reach a steady state set by the rate of nutrient input and media outflow. Thus, sustained oscillations (e.g., remaining series after accounting for predator effect) suggest the existence of other factors. Assured by their IMF analysis that these oscillations are neither driven by the predator nor by its intrinsic dynamics, biologists might conclude that there must be another variable causing the oscillation (perhaps, time-varying fluctuations in resource availability or temperature) and fruitlessly search without success. Of course, they never find one, because the oscillation is artificially introduced by performing an additive decomposition on a non-separable system. Thus, we caution potential over-interpretation of the meanings of various IMFs in EMD.
To summarize, the EMD method of Yang et al.1 clearly works for systems in which superposition is obtained, but does not provide unambiguous results for non-separable, nonlinear dynamical systems.
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Data availability
Empirical datasets analyzed in this study are open-access and available by following the same instruction addressed in Yang et al.1. R source codes used to generate synthetic datasets from the Moran-effect model and white noises are available on GitHub, https://github.com/biozoo/CommentEMD.
Code availability
Matlab and R source codes for conducting EMD causal decomposition and CCM analyses, respectively, are provided on GitHub, https://github.com/biozoo/CommentEMD.
References
Yang, A. C., Peng, C.-K. & Huang, N. E. Causal decomposition in the mutual causation system. Nat. Commun. 9, 3378, https://doi.org/10.1038/s41467-018-05845-7 (2018).
Sugihara, G. et al. Detecting causality in complex ecosystems. Science 338, 496–500, https://doi.org/10.1126/science.1227079 (2012).
Ye, H., Deyle, E. R., Gilarranz, L. J. & Sugihara, G. Distinguishing time-delayed causal interactions using convergent cross mapping. Sci. Rep. 5, 14750, https://doi.org/10.1038/srep14750 (2015).
Sugihara, G., Deyle, E. R. & Ye, H. Reply to Baskerville and Cobey: misconceptions about causation with synchrony and seasonal drivers. Proc. Natl Acad. Sci. USA 114, E2272–E2274, https://doi.org/10.1073/pnas.1700998114 (2017).
McCracken, J. M. & Weigel, R. S. Convergent cross-mapping and pairwise asymmetric inference. Phys. Rev. E 90, 062903, https://doi.org/10.1103/PhysRevE.90.062903 (2014).
BozorgMagham, A. E., Motesharrei, S., Penny, S. G. & Kalnay, E. Causality analysis: identifying the leading element in a coupled dynamical system. PLoS ONE 10, e0131226, https://doi.org/10.1371/journal.pone.0131226 (2015).
Paluš, M., Krakovská, A., Jakubík, J. & Chvosteková, M. Causality, dynamical systems and the arrow of time. Chaos: Interdiscip. J. Nonlinear Sci. 28, 075307, https://doi.org/10.1063/1.5019944 (2018).
van Nes, E. H. et al. Causal feedbacks in climate change. Nat. Clim. Change 5, 445–448, https://doi.org/10.1038/nclimate2568 (2015).
Deyle, E. R., Maher, M. C., Hernandez, R. D., Basu, S. & Sugihara, G. Global environmental drivers of influenza. Proc. Natl Acad. Sci. USA 113, 13081–13086, https://doi.org/10.1073/pnas.1607747113 (2016).
Ye, H. et al. Equation-free mechanistic ecosystem forecasting using empirical dynamic modeling. Proc. Natl Acad. Sci. USA 112, E1569–E1576, https://doi.org/10.1073/pnas.1417063112 (2015).
Sugihara, G. Nonlinear forecasting for the classification of natural time series. Philos. Trans. R. Soc. A 348, 477–495 (1994).
Ushio, M. et al. Fluctuating interaction network and time-varying stability of a natural fish community. Nature 554, 360–363, https://doi.org/10.1038/nature25504 (2018).
Author information
Authors and Affiliations
Contributions
C.-W.C., S.B.M., and C.-H.H. conceived the research idea; C.-W.C. designed the models and analyzed data; C.-W.C., S.B.M., and C.-H.H. wrote the manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Nature Communications thanks Dimitris Kugiumtzis, Juergen Kurths, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Chang, CW., Munch, S.B. & Hsieh, Ch. Comments on identifying causal relationships in nonlinear dynamical systems via empirical mode decomposition. Nat Commun 13, 2860 (2022). https://doi.org/10.1038/s41467-022-30359-8
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41467-022-30359-8
This article is cited by
-
Detection of intermuscular coordination based on the causality of empirical mode decomposition
Medical & Biological Engineering & Computing (2023)
-
Reply To: Comments on identifying causal relationships in nonlinear dynamical systems via empirical mode decomposition
Nature Communications (2022)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.