Over at Boston Review, a forum on Paul Collier's work on poverty, economic development and military intervention, with Collier, Stephen Krasner, Mike McGovern, Nancy Birdsall, Edward Miguel, and William Easterly. William Easterly:
I have been troubled by Paul Collier’s research and policy advocacy for some time. In this essay he goes even further in directions I argued were dangerous in his previous work. Collier wants to de facto recolonize the “bottom billion,” and he justifies his position with research that is based on one logical fallacy, one mistaken assumption, and a multitude of fatally flawed statistical exercises.
The logical fallacy leads to the conclusion that the poorest countries systematically fall behind everybody else in economic growth. Of course they do! Collier selected countries that were on the bottom at the end of a specific period, so naturally they would be more likely to have had among the worst growth rates in the world over the preceding period. This ex post selection bias makes the test of poor-country divergence invalid. The correct test would be to see who is poor at the beginning of the period and then see if they have worse growth than richer countries in the following years. When the test is run this way, there is no evidence that poor countries grow more slowly than richer countries.
Collier’s bottom billion is in fact a constantly changing mishmash of initially poor countries with average growth combined with other countries that were initially richer countries that had sharply negative growth. (Côte d’Ivoire and Zimbabwe are classic examples of the latter.)
Easterly has mounted an aggressive critique, I will take a moment to defend myself. Eight years ago my colleague Anke Hoeffler and I proposed that three economic characteristics—low income, slow growth, and dependence upon natural resources—all made conflict more likely. Each of these propositions has since become considerably more robust. The balance of the statistical evidence suggests that the propositions are correct.
As to method, my colleagues and I adopt the statistical approach of “general-to-specific,” in which insignificant variables are systematically and progressively deleted by the rule of stepwise deletion: this is not data mining. If it were, our results would not have been accepted by professionally refereed economics journals. Easterly’s twists on my remarks concerning new results—where a doubling of the sample and other improvements led to a minor refinement in our previous results on the effect of ethnic diversity—typify his bias.
Also see here for Easterly's rejoinder.