Dynamical Causal Model

A dynamical causal model (DCM) is a framework for inferring causal structure in systems where variables interact through feedback loops and evolve over time. Developed initially in neuroimaging by Karl Friston and colleagues, DCM extends the structural causal model tradition by replacing static directed edges with differential equations that describe how the state of each variable changes in response to the states of others.

Unlike a DAG, which encodes causal assumptions as a static graph, a DCM encodes them as a dynamical system: a set of coupled ordinary differential equations with parameters that can be estimated from observed time series. The causal question is not "which variables are connected?" but "which connections carry dynamical influence that shapes the system's trajectory?"

DCM has been most widely applied in neuroscience, where it is used to infer effective connectivity — the causal influence that one neural population exerts over another — from fMRI and EEG data. But the framework generalizes to any domain where feedback dominates and time-series data are available: ecology, economics, climate science, and system dynamics.

The limitation of DCM is computational: estimating parameters for a system of coupled differential equations is far harder than identifying a DAG. The trade-off is that DCM can model phenomena — oscillation, synchronization, regime shifts — that DAGs cannot even represent. The deeper question is whether DCM represents a genuine extension of causal inference or a different paradigm entirely.