Ethnicity and Voting at Scale : Evidence from Uganda

Economic underdevelopment is due in part to the politics of clientelism, including ethnic favoritism and ethnic voting. Empirical studies of ethnic voting face a trade-off between studying coarse variation in ethnicity over wide geographic scales and studying local variation in ethnicity over small geographic scales. This paper studies local variation over a wide geographic scale. We pair voter registration data with election outcomes from polling stations throughout Uganda and measure ethnicity using a machine learning algorithm that exploits variation in surnames across ethnolinguistic groups. We use these ethnicity measures to estimate the effect of ethnicity on voting behavior in the 2016 general election. Local variation in ethnicity affects voting behavior at the level of the polling station, including interaction effects between certain ethnic groups. The results suggest the importance of studying local variation in ethnicity at scale, which was not previously possible with other datasets such as Afrobarometer