Instrumental Variable

While I think that GPPI's quant track could probably be condensed into two semesters, I find it gratifying when I read about statistical techniques that I would not otherwise understand without having had quant 3. That was certainly the case when I found the following passage in Paul Collier's new book, The Bottom Billion. In a discussion about what makes a country prone to civil war, he mentions research he has done with a colleague looking into whether economic growth is correlated with civil-war-risk. Without using the term "instrumental variable," he indicates that it is one of the techniques they used in their model:

On this point, one might ask whether we have the causality backward-- might it be the case instead that it is the anticipation of civil war that causes decline? After all, when a civil war looks to be in the cards, investors flee, and the economy declines. It looks like decline causes war, but actually it's the anticipation of war that causes decline. This objection can be dealt with by looking at a factor that affects growth, but has no direct connection to civil war, and seeing whether the subsequent effects make civil war more or less likely. In low-income countries rainfall shocks (too much or too little rain) affect economic growth, but they do not directly affect the risk of civil war-- that is, prospective rebels do not say, "it is raining, let's call off the rebellion."

Evidently this work has been very controversial. He takes great pains to stress that the model can not and should not be used for predictive purposes, but rather to understand the risks that face countries with long-term stagnant economies.