On the identi…cation of causal e¤ects

This paper concerns the assessment of the eects of actions or poli-cies from a combination of: (i) nonexperimental data, and (ii) causal assumptions. The assumptions are encoded in the form of a directed acyclic graph, also called \causal graph", in which some variables are presumed to be unobserved. The paper establishes new criteria for de-ciding whether the assumptions encoded in the graph are sucient for assessing the strength of causal eects and, if the answer is positive, computational procedures are provided for expressing causal eects in terms of the underlying joint distribution.