The most important aspect of the pre colonial economy was that

*We thank 4 referees and the Editor for many insightful comments and useful suggestions. We thank seminar participants at Dartmouth, Tufts, Oxford, Vienna, Brown, Harvard, Stanford, UC-Berkeley, UC-Davis, NYU, AUEB, the CEPR Development Economics Workshop, the World Bank, the IMF, the NBER Political Economy Meetings, the NBER Summer Institute Meetings in Economic Growth and Income Distribution and the Macro-economy for valuable comments. We also benefited from discussions with Yannis Ioannides, Rafael La Porta, Antonio Ciccone, Rob Johnson, Raphael Frank, Jim Feyrer, Ross Levine, Avner Greif, Jeremiah Dittmar, David Weil, Sandip Sukhtankar, Quamrul Ashraf, Oded Galor, Ed Kutsoati, Pauline Grosjean, Enrico Perotti, Pedro Dal Bó, Nathan Nunn, Raquel Fernandez, Jim Robinson, and Enrico Spolaore. We are particularly thankful to Andy Zeitlin, Melissa Dell, Andei Shleifer, Nico Voigtländer, and Daron Acemoglu for detailed comments and useful suggestions. We also thank Nathan Nunn for providing the digitized version of Murdock’s Tribal Map of Africa. This paper draws on material from our previous paper titled “Divide and Rule or the Rule of the Divided? Evidence from Africa”. A Supplementary Appendix with additional sensitivity checks is available online at: http://www.dartmouth.edu/~elias/ and http://sites.google.com/site/steliosecon/. All errors are our sole responsibility.

1Logan (2011) shows that ethnic leaders are equally important as the local and central governments is assigning property rights. Respondents tend to rely more on local chiefs and ethnic institutional structures for the resolution of disputes as compared to national and local government. Ethnic-specific political actors and institutions play also some role in the provision of education and health.

2Mamdani (1996), nevertheless, differs in his assessment on the beneficial contemporary role of hierarchical pre-colonial structures arguing that the legacy of indirect rule in Africa through traditional chiefs was a basis for post-independence poor institutional and economic performance.

3 In 34 instances an ethnic homeland from Murdock’s Map is assigned to more than one groups in the Ethnographic Atlas; in these cases we assigned to the ethnic homeland the median value of the ethnic institutions index.

4After intersecting Murdock’s ethnolinguistic map with the 2000 Digital Chart of the World we drop ethnic partitions of less than 100 km2, as such tiny partitions are most likely due to the lack of precision in the underlying mapping.

5 In the previous draft of the paper we added one to the luminosity data before taking the logarithm finding similar results.

6Conley’s method requires a cutoff distance beyond which the spatial correlation is assumed to be zero; we experimented with values between 100km and 3000km. We report errors with a cutoff of 2000km that delivers the largest in magnitude standard errors.

7Land suitability for agriculture, which reflects climatic and soil conditions, enters most models with a positive and significant estimate. The malaria stability index enters with a statistically negative estimate. The coefficient on land area under water is positive and in many specifications significant. Elevation enters with a negative estimate which is significant in some models. The petroleum dummy enters always with a positive and significant coefficient. The diamond dummy enters in most specifications with a negative estimate.

8We lose three observations when we condition on the rule of law or GDP, because we lack data on Western Sahara. The results are unaffected if we assign the Western Saharan ethnic homelands to Morocco.

9When we add country fixed effects we lose one observation. This is because in Swaziland we have only one group, the Swazi.

10The Hausman-type test that compares the coefficient on the jurisdictional hierarchy index of the cross-sectional to the within-country model, suggests that one cannot reject the null hypothesis of coefficient equality.

11Since population density may be both a cause and an effect of ethnic institutions, the specifications where we also control for population density should be cautiously interpreted. Following Angrist and Pishcke’s (2008) recommendation we also used lagged (at independence) population density. In these models (not reported) the estimates on the ethnic institutions measures are larger (and always significant at the 95% level).

12Fortes and Evans-Pritchard (1940) argue that “the political systems fall into two main categories. One group consists of those societies which have centralized authority, administrative machinery, and judicial institutions-in short, a government-and in which cleavages of wealth, privilege, and status correspond to the distribution of power and authority. This group comprises the Zulu, the Ngwato, the Bemba, the Banyankole, and the Kede. The other group consists of those societies which lack centralized authority, administrative machinery, and judicial institutions-in short which lack government-and in which there are no sharp divisions of rank, status, or wealth. This group comprises the Logoli, the Tallensi, and the Nuer.” Other African scholars make a trichotomous distinction between stateless societies, large chiefdoms, and centralized states.

13Since we have just two ethnic groups where the jurisdictional hierarchy index equals four, we assign these ethnicities into the groups where the jurisdictional hierarchy index equals 3.

14We are grateful to an anonymous referee for proposing this test.

15 In line with these arguments in our sample the correlation of class stratification and the jurisdictional hierarchy index is 0.63.

16Note that not all pixels have the same surface area since pixels by the coast, lakes, and ethnic boundaries are smaller.

17The results are similar using the Gennaioli and Rainer (2006, 2007) binary index of political centralization (see Appendix Table 6).

18We are thankful to Jim Robinson for providing us with this reference.

19For example, the Dagomba in Ghana, a centralized group (the jurisdictional hierarchy index equals 3) is adjacent to two non-centralized groups in Ghana, the Basari and the Konkomba. In such cases we include both pairs. The median (average) distance between the centroids of neighboring ethnicities is 179km (215km).

20See, for example, Dell (2010), Bubb (2012), and Michalopoulos and Papaioannou (2012), among others.

Page 2