Written for smalltalkeconomics.com.
Since the very beginnings of market economies, there have been two general views of how things should be done; how the governments should behave. Laissez-faire vs. regulation, right vs. left, liberalism vs. social state, republicans vs. democrats, keep-your-money-and-use-it-however-you-want vs. give-us-your-money-and-we-will-redistribute-it. One might think that by now, we should have known for a long time which system works better. We can just compare two nations that each use one of the systems, right? However, comparing two nations with different systems shows to be a very difficult task due to great heterogeneity in virtually all other characteristics that affect the outcome. In the words of the theory of treatment econometrics, constructing a counterfactual is incredibly difficult. In the words of a normal person, we don’t know what economies would look like had they adopted a different system.
Alexander Lee and Kenneth A. Schultz of Stanford have used a unique natural experimental setting in Cameroon in their 2011 paper to shed more light on this agelong discussion. Cameroon was first colonized by the German empire in 1884. After the defeat of Germany in World War I, it became a League of Nations mandate territory and was artificially divided between the French and British colonial powers until Cameroon gained independence in the beginning of the 1960’s. During the more than 40 years of colonial regime, the French and the British took a very different approach as to which institutions to put in place to ensure effective economic development of their part of Cameroon. While the French flooded their part (East) of Cameroon with investment into infrastructure and made the public sector in general very powerful, the British (West Cameroon) preferred letting the economy work largely on its own.
This situation made for a very promising experimental setting, but large regional differences remained an issue. That is why the authors focused on pairs of rural towns that are close to each other geographically, but divided by the French-British border. This method is called local average treatment effect and it is an integral part of regression discontinuity design. To see its fundamental idea on a simpler example, suppose you want to estimate the effect of studying at the best law school on post-university wage as compared to studying at a worse (less challenging) law school. Taking a simple difference in wage between the two groups leads to a huge selection bias, as more skilled people usually get to the best school, because they score better on the entrance exam. Using the local average treatment effect in this example means to compare people who scored close to the cut-off level but below, and those who scored close but above. These two groups are roughly the same (since entrance exams are only rough approximations of your true skills), but one of them got the treatment (i.e. the best law school) and the other did not (i.e. they went to a worse law school). Similarly, in Cameroon, two villages that are close to the French-British border are roughly the same (in geographical and other characteristics), but they received different treatment (i.e. the French vs. the British colonial legacy).
And the result? As shown in the figure below, the British part appears to have higher levels of economic dynamism, evidenced by greater household wealth, and better functioning local government institutions, evidenced by its higher level of public goods provision. These findings are consistent with the hypothesis that the mix of institutions and practices associated with British colonial rule generated superior outcomes. This does not imply, of course, that British-colonized areas always perform better or that West Cameroon is an elysia of wealth and strong institutions. But for the rural areas in this particular case, it seems that the British way was the better one to go with.
Reference: Lee, A., & Schultz, K. A. (2012). Comparing British and French Colonial Legacies: A Discontinuity Analysis of Cameroon. Quarterly Journal of Political Science, 7(4), 365-410. Available here.