Are Instrumental Variables useful for policy making?

A few weeks ago, Bernard Hoekman, a World Bank “expert”, presented a paper on trade in services and growth here at IHEID. He started with a critique of instrumentals variables (IVs) in the recent literature on institutions and growth, quoting a paper by Dixit. I thought that he had no idea what he was talking about, being a World Bank employee, and as this literature is much praised in the academic world. His critic was that instruments for institutions or openness, such as settler mortality, origin of the legal system, Protestantism, or geographic ones such as ruggedness, climate or distance to the equator, were not helpful for policy making, being things you cannot affect. That is definitely not the point of an IV I thought. He must be all wrong.

But then yesterday, Verdier was making the same comment in the PhD Micro Seminar, which lead to a unsatisfying class discussion. This must be a World Bank paradigm I thought. One important guy says something, everybody repeats it, without understanding, it gets distorted, and it spreads out of the World Bank into the entire economics world, as a new, but false, paradigm.

To test my hypothesis, I looked at the Dixit paper, called “Evaluating Recipes for Development Success”, published in the World Bank Research Observer, in 2007. His critic was that these IVs, even though they identify causality, do not tell you how to affect your instrumented variable. In the literature, the policy recommendation is to create good institutions, as they are the key to growth. However, IVs do not tell you how to change institutions. This is the point made by Dixit. But that is not due to IVs at all. If you want to affect something more precise than institutions, such as bureaucrats’ salaries, put that variable in your regression!

The IVs certainly don’t tell you to affect geography or settler mortality, nor that a country’s future is completely pre-determined. (see Jared Diamond for a geographic deterministic approach). “Of course, this is not the interpretation the researchers intend; they intend many of their history and even geography variables to have only indirect effects on economic outcomes through some other proximate determinant of success or to be mere econometric instruments used for identifying the direction of causation”, Dixt writes. Indeed, IVs are used only to extract the exogenous component of your explaining variables, so that you can econometrically evaluate the causation effect.

So the problem with this literature is that it doesn’t tell you how to enhance institutions or openness, it just tells you this is what you have to do. This critique, however, has nothing to do with IVs, it has to do with your choice of explaining variables. Dixit wasn’t clear and he was misinterpreted.