Tuesday, December 4, 2012

Regression & Causation

Recently I read, with interest, a thought-provoking paper by Bryan Chen & Judea Pearl. The paper is titled, "Regression and causation: A critical examination of econometrics textbooks".

Here's the abstract:
"This report surveys six influential econometric textbooks in terms of their mathematical treatment of causal concepts. It highlights conceptual and notational differences among the authors and points to areas where they deviate significantly from modern standards of causal analysis. We find that econometric textbooks vary from complete denial to partial acceptance of the causal content of econometric equations and, uniformly, fail to provide coherent mathematical notation that distinguishes causal from statistical concepts. This survey also provides a panoramic view of the state of causal thinking in econometric education which, to the best of our knowledge, has not been surveyed before."

Whatever your favourite econometrics textbook might be, once you've read this paper you'll look at that book through a different lens - I promise you!

© 2012, David E. Giles


  1. Very interesting, especially because one of my more early econometrics book is in there! (Wooldridge, albiet not the current edition). Thanks for the share.

  2. I've looked at that paper a couple of times now, and I'm still trying to get their point. On one hand, my gut reaction is "why are a couple of computer science people so interested in econometrics." On the other hand, I gather that Pearl is a big name in treatment effects and causal modeling. But it seems to me that Pearl seems to think that is all econometrics is about and comes away disappointed because none of the six econometrics texts reviewed takes the approach that he would. So?

    1. Fair enough comment.It's always interesting, though, to see a different perspective on our discipline.

    2. I think if you want to understand the substance of Pearl's concerns and contributions, which I don't entirely agree with but still think are important, this is definitely not the paper to read. He has some good published survey papers and you could also try his book. Relying on a rather poor working paper with a student is not really a fair assessment.

  3. I did a lot of work with causal networks (aka Bayesian networks, probablistic networks, belief networks) back in the 1990s as inference engines for software agents and as management decision tools. To incorporate Bayesian networks into an undergraduate econometrics course would be difficult and probably counterproductive. Bayesian networks and statistical inference seem to me to have different objectives. When I teach applied econometrics (using Stock and Watson as the text), I use the concept of "explains the variation in Y" instead of "causes Y".