orgtheory.net

pick & twist part 2

Sean mentioned the Economist’s story of a few weeks ago. As I hinted in an earlier post, the identification movement has not been received enthusiastically in all corners of the economics profession. I think it is useful to bear in mind what the alternatives favored by Heckman & Urzua and Deaton are; my guess is that they will sound pretty unappealing to most orgtheory readers.

Among their critics, the knock against identificationistas is that they are “mere statisticians.” The models estimated in the typical IV paper are often loosely inspired by economic theory (horror of horrors, they might even be motivated by sociological or psychological theories!); the parameters identified are not policy-relevant because they cannot be traced in a simple way to equilibrium relationships derived from a model in which agents optimize (to which I answer: “guilty as charged, your honor”).

The alternative is to build a strucural model. The idea is to “take directly to the data” a particular model, as opposed to test some or all of the implications of a model. As an interesting example, I would offer my colleague Tavneet Suri‘s analysis of hybrid corn use among Kenyan farmers.

The great advantage of the structural approach is that one recovers directly estimates of theory-relevant parameters. As a result, researchers can make statements about the welfare effects of a policy, or simulate the effects of alternative policies. This comes at a cost, for all the conclusions require that we maintain the assumption that the particular model estimated is a plausible and not-too-inaccurate approximation of the real world. Personally, I tend to view economic models as parables; there are very few that I would have the confidence to take directly to the data.

Very recently, combining randomization with structural modelling has emerged as an approach delivering the best of worlds. A recent paper by Esther Duflo, Rema Hanna, and Stephen Ryan provides a nice illustration. The call for more papers using such “mixed methods” was also one of the punchlines in Guido Imbens’ thoughtful reply to Deaton (2009) and Heckman & Urzua (2009).

Written by Pierre

August 23, 2009 at 3:50 pm

%d bloggers like this: