Wednesday, February 2, 2011

Alex Tabarrok on Instrumental Variables

Tabarrok shares some interesting thoughts about instrumental variables. He cites a tragedy of the commons type problem in these studies, from Morck and Yeung:

"A Tragedy of the Commons has led to an overuse of instrumental variables and a depletion of the actual stock of valid instruments for all econometricians. Each time an instrumental variable is shown to work in one study, that result automatically generates a latent variable problem in every other study that has used or will use the same instrumental variable, or another correlated with it, in a similar context. We see no solution to this. Useful instrumental variables are, we fear, going the way of the Atlantic cod."

All I can say is - sounds fantastic. A lot of instrumental variable analysis is just academic masturbation. Or perhaps exhibitionism is a better term. It's just a way of showing off, and yet nobody seems to be as impressed with the show as the exhibitionist himself. The motivation behind instrumental variables is absolutely essential, of course. But I've always found regression discontinuities, abrupt policy changes, and even "natural experiments" far more convincing (although these have their own problems).

My thoughts on this are not completely exogenous. I picked up my suspicions from Prof. David Jaeger, my undergraduate labor economics and econometrics professor. Aside from a good education in general, it's probably the single concrete insight/prejudice that I've taken with me from college. That's not to say none of my other professors ever said anything worthwhile - just that this is the one thing I've consciously carried with me. Always be suspicious of IV and discount the truth-value of the findings heavily.

1 comment:

  1. IV regressions certainly have their problems, but I'm don't particularly agree with Morck and Yeung's "tragedy of the commons" analogy. All that matters for the instrument is that is a) exogenous to the error term, and b) is relevant for explaining variation in the (endogenous) explanatory variable. These two requirements hold whether the instrument was used in a previous study or not.

    Further, once you have obtained a valid instrument for a particular estimation, why worry unnecessarily about replacing it with something new? We don't need to reinvent the wheel with every new study that comes out.

    My thoughts on this are not completely exogenous.
    Da dum. Tish.
    :)

    ReplyDelete

All anonymous comments will be deleted. Consistent pseudonyms are fine.