Three Essays on Politics in Kenya

Abstract

This dissertation examines ethnic patronage, local conflict, and election fraud in Kenya in three separate essays. Fraud, violence, and ethnicity are difficult to measure, and they often play a central role in narratives and theories about African politics. The essays in this dissertation draw on natural language processing, spatial statistics, and demography to improve measurement of these concepts and, in turn, our understanding of how they function in Kenya. The approaches developed here can be generalized to conflict, ethnicity, and fraud in other contexts. The first essay presents a method for extracting ethnic information from names. Existing methods give biased estimates by ignoring uncertainly in the mapping between names and ethnicity. I apply my improved, approximately unbiased method to data on political appointments from 1963 to 2010 in Kenya, and find that existing narratives about distributive politics do not accord with empirical patterns. The second essay examines patterns of violent ethnic targeting during Kenya's 2007-2008 post-election violence. I focus on patterns of arson, one of the key types of violence used in the Rift Valley. I find that incidence of arson is related to the presence of ethnic outsiders, and even more strongly related to measures of land quality, accessibility, and electoral competition. Using a difference-in-differences design, I show that arson caused a significant decrease in the number of Kikuyu and other immigrant ethnic groups registered to vote; no such decline is observed in indigenous ethnic groups. The third essay documents the prevalence of dead voters on Kenya's voter register prior to the contentious 2007 presidential elections, and shows how dead registered voters may have facilitated electoral fraud. Simply accounting for the number of dead voters demonstrates that turnout was greater than 100% in several opposition constituencies, and implausibly high in most of the incumbent president's home province. Ecological inference suggests that ballot-stuffing occurred in candidate strongholds, rather than competitive constituencies. These results are consistent with the opposition party's allegations of fraud.