On the role of the Markov condition in causal reasoning

The Markov condition describes the conditional independence relations present in a causal graph. Cartwright argues that causal inference methods have limited applicability because the Markov condition cannot always be applied to domains, and gives an example of its incorrect application. We question two aspects of this argument. One, causal inference methods do not apply the Markov condition to domains, but infer causal structures from actual independencies. Two, confused intuitions about conditional independence relationships in certain complex domains can be explained as problems of measurement and of proxy selection.