Equilibrium Causal Models: Connecting Causal Models & Dynamical Systems


Feedback loops and cyclic relationships are a mainstay of dynamical systems modelling, yet they are conspicuously absent in much of causal inference, where Directed Acyclic Graphs (DAGs) abound. Cycles do indeed have a place in causal inference, yet, in order to understand them, it is necessary to map causal models and causal intuitions back onto dynamical systems. To do this we can make use of Equilibrium Causal Models (ECMs), causal abstractions of dynamical systems which a) allow researchers to reason about the long time-scale effects of interventions and b) can potentially be learned from cross-sectional data. In this short talk, we give a brief motivation of ECMs, discuss how they relate to contemporary issues in psychological research, and sketch a simulation study to assess the feasibility of learning cyclic causal structures in practice.