- Elections are not enough to generate accountability and legitimacy for governments (something readers of the New York Times Letter to the Editor Page have known for over a decade). But it's even worse than that- in very poor countries, holding elections appears to lead to worse policies and governance outcomes, and increase the risk of civil war.
- About 12% of foreign aid ends up financing militaries, regardless of what the aid money is actually going towards. Governments of beneficiary countries simply spend less on whatever issues donors are involved in, and more on their militaries.
- Most interesting is Collier's "feasibility hypothesis" of civil war. The idea is that the best way to understand civil war is to look at whether rebellion (as opposed to peaceful opposition) is a feasible means for the opposition to contest the ruling government's authority. He makes the case by showing that you can actually predict whether or not a country will have a civil war fairly well just by looking at the circumstances that make it feasible (access to resources that rebels can exploit, mountainous terrain that is more difficult to police, etc.) This is a bolder statement than it sounds, because it implies that preventing civil war is less a matter of understanding or fixing the underlying grievances, and more a matter of making it harder to pull off (more on this later).
Wars, Guns, and Votes is a fine illustration of how the methodology of economics- rational agent models and econometric analyses of data- can yield important insights when applied to other fields, such as political science. The book also illustrates the critical importance of contextual knowledge in this kind of thing. Collier's expertise in African politics is evident, and it is only in bringing this expertise to bear that the results of his economic analysis are convincing. Some have criticized Collier for "data mining," and undoubtedly some of the details of his findings will not bear further empirical scrutiny. However, what seperates Collier's work from other more egregious data mining exercises is that Collier's empirical findings are supported by both coherent theories and an extensive knowledge of context, and as a result his stories tend to tend ring true.