Social Science Statistics Blog

How about Angrist, Graddy, and Imbens? Disclaimer: I haven't read the paper.

Google is right, most of the work on this topic comes from the "Heckman family". The topic is discussed in Heckman Lalonde & Smith (1999, Handbook of Labour Economics). Here are some other articles that may be relevant:

http://papers.nber.org/papers/w6426
http://ideas.repec.org/p/iza/izadps/dp758.html
http://ideas.repec.org/p/qed/wpaper/1074.html

Would you consider non-deterministic computational models leading to an equilibrium condition an example of "potential outcomes arising naturally from the model"?

Thanks for the links above (btw, if you post enough links in a comment, our spam filter picks it up and holds if for approval - sorry about that Structural Model). Tom, I'm not sure whether the computational models that you describe would meet the somewhat ambiguous criterion that I proposed. That is in part because I'm not sure what a successful attempt to embed the potential outcomes framework in an equilibrium model would look like. For example, in the Angrist, Graddy, and Imbens article linked above, the treatment as I understand it is (plausibly) exogenous variation in the availability of fresh fish due to weather conditions etc., which the authors use to estimate demand equations. A cool idea, don't get me wrong, but in some ways external to the model. I wonder, if one takes both the potential outcomes framework and an equilibrium approach to modeling behavior, whether it is basically impossible to estimate (and possibly to define) causal effects of human actions if we assume that those actions are taken by rational maximizers, since Y(0) is only observed off the equilibrium path. If this is the case (and I'm just throwing it out there), it would be kind of ironic given the criticism of manipulationist views of causality as being too human-centered.