CMStatistics 2018: Start Registration
View Submission - CMStatistics
B1649
Title: A causal modeling framework in search of a graphical representation Authors:  Joris Mooij - University of Amsterdam (Netherlands) [presenting]
Tineke Blom - University of Amsterdam (Netherlands)
Abstract: Structural Causal Models (SCMs) provide a causal modeling framework that is used in many fields such as economy, the social sciences, and biology. It offers appealing features, e.g., a graphical representation that simultaneously expresses conditional independence properties and causal properties of the model, which lies at the basis of many of the theoretical and algorithmic results in the area. We show that SCMs are not flexible enough to give a complete causal representation of equilibrium states of dynamical systems in general. We propose a generalization of the notion of SCM, that we call Causal Constraints Model (CCM), and prove that CCMs are flexible enough to capture the essential causal semantics of dynamical systems at equilibrium. As an illustration, we consider a simple and ubiquitous chemical reaction. The price one pays for the improved generality and flexibility of CCMs over SCMs is that no appropriate graphical representation of CCMs that simultaneously expresses conditional independence properties and causal properties is known. We challenge the graphical modeling community to invent such a representation.