Graphical models, potential outcomes and causal inference: Comment on Ramsey, Spirtes and Glymour