Abstract

Real-world systems are often modelled by sets of equations with exogenousrandom variables. What can we say about the probabilistic and causal aspects ofvariables that appear in these equations without explicitly solving for them?We prove that, under a solvability assumption, we can construct a Markovordering graph that implies conditional independences and a causal orderinggraph that encodes the effects of soft and perfect interventions by making useof Simon's causal ordering algorithm. Our results shed new light on discussionsin causal discovery about the justification of using graphs to simultaneouslyrepresent conditional independences and causal relations in models withfeedback.

 

Quick Read (beta)