replying to Chun-Wei Chang et al. Nature Communications https://doi.org/10.1038/s41467-022-30359-8 (2022)
The preceding Matters Arising1 on our Article2 states that (1) empirical mode decomposition (EMD) method incorrectly distinguish causation from correlation for the system of two independent variables driven by a shared external forcing (aka Moran effect3); and (2) we used convergent cross mapping (CCM) in a manner that was not intended by the original paper4 and that leads to incorrect causal relationship interpreted by the CCM method. While we think the second comment has certain merits, the first point needs to be addressed in detail.
First, Chang et al.1 criticized that causal decomposition fails to correctly identify causal relationships in adult-recruitment model to simulate Moran effect in which two independent variables are driven by a shared external force. Here we reach a different conclusion, which we now elaborate. It’s obvious that N1 and N2 in the adult-recruitment model are correlated because of the common environmental variable in the differential equation. However, with appropriate mathematical deduction (see supplementary information), both the values of N1 and N2 are found to be coupled with the past values of its counterpart, which accords temporal precedence principle of cause and effect. Furthermore, we found the causal strength between N1 and N2 was mainly driven by the ratio of φ1 and φ2, which represents the magnitude of environmental forcing in the adult-recruitment model. Fig. 1a shows that a change in relative causal strength by causal decomposition is observed in various settings of φ1 and φ2. Interestingly, a consistent finding of convergence of cross mapping between N1 and N2 is also observed with the CCM method (Fig. 1b), suggesting a bidirectional coupling exists between N1 and N2, which contradicts the conclusion by Chang et al.1 that CCM would indicate no causation in this model.
It is worthy to note that the Moran effect model may be a case of instantaneous causality5. The example illustrated by Chang et al.1 indicates that CCM accounts for instantaneous causality as the states of the variables compared are at the same time point and therefore shows no causality, whereas the result of the causal decomposition on the other hand indicated lag-causality instead of instantaneous causality. The presence of instantaneous causality in the nonlinear system such as Moran effect models requires further study6. Additionally, both CCM and causal decomposition analysis are bivariate causality analyses, common drivers and other indirect causal effects may be better identified using multivariate causality approaches7,8.
Second, we appreciate Chang’s et al.1 clarification of the importance of the convergence of cross mapping in the interpretation of the method. This criterion gives clue to the existence of causation but do not indicate the difference in coupling strengths such as top-down or bottom-up control in the predator and prey relationship. For clarity, here we quote the original statement for the CCM results of Didinium and Paramecium by Sugihara et al.4: “The results in Didinium and Paramecium suggest bidirectional coupling, which accords with what is known. Moreover, the higher level of skill in cross mapping Didinium from the Paramecium time series than the reverse suggests that top-down control by the predator, Didinium, is stronger than bottom-up control by the prey, Paramecium.” Therefore, it seems to us that the interpretations of CCM are two folds: (1) the existence of directional coupling is determined based on the convergence or improvement of cross-mapping skills4,9, and (2) the strength of coupling in each direction (e.g., top-down or bottom-up control in the case of predator and prey) is determined by difference in the level of cross-mapping skill (e.g., correlation)4.
Based on these definitions, we did have objective interpretation of CCM results in Fig. 5 of our Article2 that all ecology data showed bi-directional coupling but only Didinium and lynx have a clear pattern of top-down control, whereas CCM failed to show differential control in Lotka Volterra model and the convergence of cross-mapping skill is ambiguous in wolf-moose relationship. In contrast, the causal decomposition consistently identifies top-down control of predator over prey across four types of ecology data. As already mentioned in our paper, the relative causal strength is more important than absolute causal strength in that a meaningful causality is only observed when differential coupling strengths exist. Having that said, the interpretation of CCM by McCracken et al.10 using difference of cross-mapping skill is consistent with Sugihara et al.4. But we do concur that the convergence criteria of cross-mapping skill needs to be rigorously evaluated for the presence of causation9.
For additional comments by Chang et al.1, we concur that CCM does not rely on predictability but information recovering as the criterion. However, we would like to highlight that the ability of information recovering in state space is determined by the choice of embedding dimension, which is related to the length of time lag in time series model. Furthermore, there indeed have been methods to improve CCM with trend removing techniques, and we have discussed its pros and cons in our published Peer Review File11.
Chang et al.1 also commented that causal decomposition does not meet the expectation that if the predator dies off exponentially in the absence of the prey, then the prey will grow exponentially in the absence of the predator; rather, the remaining components continue to cycle after subtracting causal IMFs. We would like to point out that the causal decomposition method was not intended to mathematically solve the differential equation of predator and prey model. Specifically, although we would expect the predator and prey will have respectively exponential decay or growth in the absence of the other pair, it’s apparent that removing a causal-related IMF in a time series will not induce exponential change as predicted by the predator and prey model because the data has been recorded in a manner where both predator and prey are present. Furthermore, because EMD is designed to separate oscillations in different temporal scales12, the method itself does not necessarily violate the mathematical intuition of non-separability given in the nonlinear differential equations. For example, we have shown that EMD method is able to delineate phase and amplitude coupling from nonlinear oscillations generated by the multiplicative process13.
Importantly, we concur that real-world data is blended with stochastic and deterministic mechanisms. Our paper never intended to disparage Granger or CCM method. We have objectively shown the merit of use of covariation principle of causality, and illustrate advantages and limitations of causal decomposition compared to other methods. Nevertheless, more data and validation are absolutely needed to evaluate the applicability of existing causality methods in various modeled and real-world data.
Reporting Summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
References
Chang, C. W., Munch, S. B. & Hsieh, C. H. Comments on identifying causal relationships in nonlinear dynamical systems via empirical mode decomposition. Nat. Commun. https://doi.org/10.1038/s41467-022-30359-8 (2022).
Yang, A. C., Peng, C. K. & Huang, N. E. Causal decomposition in the mutual causation system. Nat. Commun. 9, 3378 (2018).
Moran, P. A. P. The statistical analysis of the Canadian lynx cycle. II. Synchronization Meteorol. Aust. J. Zool. 1, 291–298 (1953).
Sugihara, G. et al. Detecting causality in complex ecosystems. Science 338, 496–500 (2012).
Koutlis, C., Kimiskidis, V. K. & Kugiumtzis, D. Identification of hidden sources by estimating instantaneous causality in high-dimensional biomedical time series. Int. J. Neur. Syst. 29, 1850051 (2019).
Lütkepohl, H. New introduction to multiple time series analysis. (Springer-Verlag GmbH, 2005).
Runge, J. et al. Identifying causal gateways and mediators in complex spatio-temporal systems. Nat. Commun. 6, 8502 (2015).
Manshour, P., Balasis, G., Consolini, G., Papadimitriou, C. & Paluš, M. Causality and information transfer between the solar wind and the magnetosphere–ionosphere system. Entropy 23, 390 (2021).
Chang, C. W., Ushio, M. & Hsieh, C. H. Empirical dynamic modeling for beginners. Ecol. Res. 32, 785–796 (2017).
McCracken, J. M. & Weigel, R. S. Convergent cross-mapping and pairwise asymmetric inference. Phys. Rev. E 90, 062903 (2014).
Yang, A. C., Peng, C. K. & Huang, N. E. Peer review file of causal decomposition in the mutual causation system. https://static-content.springer.com/esm/art%3A10.1038%2Fs41467-018-05845-7/MediaObjects/41467_2018_5845_MOESM1_ESM.pdf (2018).
Huang, N. E. et al. The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis. Proc. R. Soc. Lond. Ser. A-Math. Phys. Eng. Sci. 454, 903–995 (1998).
Huang, N. E. et al. On Holo-Hilbert spectral analysis: a full informational spectral representation for nonlinear and non-stationary data. Philos. Trans. Ser. A Math. Phys. Eng. Sci. 374, 20150206 (2016).
Acknowledgements
This study was supported by the Ministry of Science and Technology with grants to Dr. Albert C. Yang [MOST-110-2628-B-A49A-509; MOST-110-2321-B-A49A-502; MOST-110-2634-F-A49 -005]. Dr. Albert C. Yang was also supported by Mt. Jade Young Scholarship Award from the Ministry of Education, Taiwan, as well as the Brain Research Center, National Yang Ming Chiao Tung University and the Ministry of Education (Aim for the Top University Plan), Taipei, Taiwan.
Author information
Authors and Affiliations
Contributions
A.C.Y. analyzed the data, performed computation and simulation, and drafted the manuscript. N.E.H. and C.-K.P. contributed to the critical comment and revisions of 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
Yang, A.C., Peng, CK. & Huang, N.E. Reply To: Comments on identifying causal relationships in nonlinear dynamical systems via empirical mode decomposition. Nat Commun 13, 2859 (2022). https://doi.org/10.1038/s41467-022-30360-1
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41467-022-30360-1
This article is cited by
-
Detection of intermuscular coordination based on the causality of empirical mode decomposition
Medical & Biological Engineering & Computing (2023)
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.