Discovering cause-effect relationships in spatial systems with a known direction based on observational data

Konrad P Mielke, Tom Claassen, J Huijbregts, Aafke M Schipper, Tom M Heskes
Proceedings of the 10th International Conference on Probabilistic Graphical Models, PMLR 138:305-316, 2020.

Abstract

Many real-world studies and experiments are characterized by an underlying spatial structure that induces dependencies between observations. Most existing causal discovery methods, however, rely on the IID assumption, meaning that they are ill-equipped to handle, let alone exploit this additional information. In this work, we take a typical example from the field of ecology with an underlying directional flow structure in which samples are collected from rivers and show how to adapt the well-known Fast Causal Inference (FCI) algorithm (Spirtes et al., 2000) to learn cause-effect relationships in such a system efficiently. We first evaluated our adaptation in a simulation study against the original FCI algorithm and found significantly increased performance regardless of the sample size. In a subsequent application to real-world river data from the US state of Ohio, we identified important likely causes of biodiversity measured in the form of the Index of Biotic Integrity (IBI) metric.

Cite this Paper

BibTeX
@InProceedings{pmlr-v138-mielke20a, title = {Discovering cause-effect relationships in spatial systems with a known direction based on observational data}, author = {Mielke, Konrad P and Claassen, Tom and Huijbregts, Mark A J and Schipper, Aafke M and Heskes, Tom M}, booktitle = {Proceedings of the 10th International Conference on Probabilistic Graphical Models}, pages = {305--316}, year = {2020}, editor = {Jaeger, Manfred and Nielsen, Thomas Dyhre}, volume = {138}, series = {Proceedings of Machine Learning Research}, month = {23--25 Sep}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v138/mielke20a/mielke20a.pdf}, url = {https://proceedings.mlr.press/v138/mielke20a.html}, abstract = {Many real-world studies and experiments are characterized by an underlying spatial structure that induces dependencies between observations. Most existing causal discovery methods, however, rely on the IID assumption, meaning that they are ill-equipped to handle, let alone exploit this additional information. In this work, we take a typical example from the field of ecology with an underlying directional flow structure in which samples are collected from rivers and show how to adapt the well-known Fast Causal Inference (FCI) algorithm (Spirtes et al., 2000) to learn cause-effect relationships in such a system efficiently. We first evaluated our adaptation in a simulation study against the original FCI algorithm and found significantly increased performance regardless of the sample size. In a subsequent application to real-world river data from the US state of Ohio, we identified important likely causes of biodiversity measured in the form of the Index of Biotic Integrity (IBI) metric.} }
Endnote
%0 Conference Paper %T Discovering cause-effect relationships in spatial systems with a known direction based on observational data %A Konrad P Mielke %A Tom Claassen %A J Huijbregts %A Aafke M Schipper %A Tom M Heskes %B Proceedings of the 10th International Conference on Probabilistic Graphical Models %C Proceedings of Machine Learning Research %D 2020 %E Manfred Jaeger %E Thomas Dyhre Nielsen %F pmlr-v138-mielke20a %I PMLR %P 305--316 %U https://proceedings.mlr.press/v138/mielke20a.html %V 138 %X Many real-world studies and experiments are characterized by an underlying spatial structure that induces dependencies between observations. Most existing causal discovery methods, however, rely on the IID assumption, meaning that they are ill-equipped to handle, let alone exploit this additional information. In this work, we take a typical example from the field of ecology with an underlying directional flow structure in which samples are collected from rivers and show how to adapt the well-known Fast Causal Inference (FCI) algorithm (Spirtes et al., 2000) to learn cause-effect relationships in such a system efficiently. We first evaluated our adaptation in a simulation study against the original FCI algorithm and found significantly increased performance regardless of the sample size. In a subsequent application to real-world river data from the US state of Ohio, we identified important likely causes of biodiversity measured in the form of the Index of Biotic Integrity (IBI) metric.
APA
Mielke, K.P., Claassen, T., Huijbregts, J., Schipper, A.M. & Heskes, T.M.. (2020). Discovering cause-effect relationships in spatial systems with a known direction based on observational data. Proceedings of the 10th International Conference on Probabilistic Graphical Models, in Proceedings of Machine Learning Research 138:305-316 Available from https://proceedings.mlr.press/v138/mielke20a.html.

Related Material