B0508
Title: Clustering and structural robustness in causal diagrams
Authors: Santtu Tikka - University of Jyvaskyla (Finland) [presenting]
Abstract: Clustering of vertices is considered in directed acyclic graphs that represent structural causal models. Clustering can often clarify the visual representation of the causal model, but arbitrary clustering might break important causal connections. We define a specific type of cluster, called transit cluster, and show that under the corresponding clustering, the graph retains important properties related to causal effect identifiability if specific assumptions hold. We further show that a subset of transit clusters, called transit components, can be found efficiently, and that any transit cluster can be represented as a union of such components. We also consider the inverse problem, where one begins with a clustered graph and looks for larger graphs from which the original graph may have been obtained as a result of clustering.
