Questionable Inference on the Power of Pre-Colonial Institutions in Africa

In their paper "Pre-Colonial Ethnic Institutions and Contemporary African Development" [Econometrica 81(1): 113-152], Stelios Michalopoulos and Elias Papaioannou claim that they document a strong relationship between pre-colonial political centralization and regional development, by combining Murdock's ethnographic atlas (1967) with light density at night measures at the local level. We argue that their estimates do not properly take into account population effects. Among lowly populated areas, luminosity is dominated by noise, so that with linear specifications the coefficient of population density is biased downwards. We reveal that the identification of the effect of ethnic centralization very much relies on these areas. We implement a variety of models where the effect of population density is non-linear, and/or where the bounded or truncated nature of luminosity is taken into account. We conclude that the impact of ethnic-level political centralization on development is all contained in its long-term correlation with population density. We also abstract from the luminosity-population nexus by analyzing survey data for 33 countries. We show that individual-level outcomes like access to utilities, education, asset ownership etc. are not correlated with ethnic-level political centralization.