Learning Simple Causal Structures1

Abstract

Humans use knowledge of causation to derive dependencies among events of interest. The converse task, that of inferring causal relationships from patterns of dependencies, is far less understood. This article established conditions under which the directionality of some dependencies is uniquely dictated by probabilistic information—an essential prerequisite for attributing a causal interpretation to these dependencies. An efficient algorithm is developed that, given data generated by an undisclosed simple causal schema, recovers the structure of that schema, as well as the directionality of all links that are uniquely orientable. A simple schema is represented by a directed acyclic graph (dag) where every pair of nodes with a common direct child have no common ancestor nor is one an ancestor of the other. Trees, singly connected dags, and directed bi‐partite graphs are examples of simple dags. Conditions ensuring the correctness of this recovery algorithm are provided.