WHEN ETHNIC EXCLUSION IS GOOD POLITICS:
ETHNIC EXCLUSION, ARMED CONFLICT, AND LEADERSHIP TENURE
IN SMALL-COALITION SYSTEMS
By
Hyun Jin Choi

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Political Science
2012

ABSTRACT
WHEN ETHNIC EXCLUSION IS GOOD POLITICS: ETHNIC EXCLUSION, ARMED
CONFLICT, AND LEADERSHIP TENURE IN SMALL-COALITION SYSTEMS
By
Hyun Jin Choi
Why do some leaders deliberately foster ethnic hatred and exclusion even though such a policy
increases the risk of ethnic conflict?

Contrary to common belief, I find that ethnic exclusion is

good politics (but not good policy) for non-democratic leaders with small winning coalitions,
despite its positive impact on the risk of ethnic conflict.

To explain this mechanism, I modify

the selectorate theory of Bueno de Mesquita, et al. (2003) by explicitly accounting for the role of
ethnic ties in the formation of the incumbent’s coalition.
theory for explaining ethnic exclusion and leader survival.

Four hypotheses are deduced from my
H1 maintains that, in small-coalition

systems, leaders who employ ethnic exclusion are more likely to survive longer in office than
those who do not employ it.

H2 predicts that, if small-coalition leaders do not pursue an

exclusive ethnic policy, they are more likely to be removed from office in an irregular manner.
H3 suggests that small-coalition leaders are less likely to lose power during civil war if they
employ ethnic exclusion.

Lastly, if ethnic exclusion really is good politics for small-coalition

leaders, H4 predicts, there should be higher levels of ethnic exclusion in small-coalition systems
than in large-coalition systems.
These hypotheses are empirically tested and supported by Cox’s proportional hazard
regressions using data on the tenures of 982 leaders from 1946 to 2004. My results show that in
small-coalition systems: (1) the hazard of deposition for leaders who implement a strong
exclusion policy is about 80% lower than that of leaders who do not promote ethnic exclusion; (2)
the risk of irregular turnover among leaders who employ ethnic exclusion is only about 1.3% of

the risk for those who do not employ such a policy; and (3) the risk of irregular removal from
office virtually disappears even in times of civil war if a leader employs a strong exclusion
policy.

Case studies of Iraq, Burundi, and Rwanda further corroborate causal claims made by

the exclusion theory.

In all three cases, major ethnic groups had been excluded from

participation in the incumbent’s coalition until a small group of ethnic elites monopolizes key
positions in the army and government.

The cases of Burundi and Rwanda further demonstrate

how small-coalition systems face large-scale violence when inclusive ethnic policy is
implemented after years of exclusive rule.

Overall, my findings lead to an “unwelcome”

conclusion: ethnic exclusion is good politics in small-coalition systems even if it could increase
the risk of ethnic conflict.

This conclusion suggests that the crucial element in the prevention

and resolution of ethnic conflict is the development of policies that can address leaders’
incentives for ethnic exclusion.

Copyright by
Hyun Jin Choi
2012

To My Great Mom Chu Insook and Lovely Wife Jungeun

v

ACKNOWLEDGEMENTS
This dissertation would not have been possible without the guidance and the help of several
individuals who in one way or another contributed and extended their valuable assistance in the
preparation and completion of this study.

First and foremost, my utmost gratitude to Michael

Colaresi, chair of my dissertation committee, whose sincerity and encouragement I will never
forget.

Dr. Colaresi has been my inspiration as I hurdle all the obstacles in the completion of

this dissertation.

Ravi Bhavnani, for his unselfish and unfailing support as my adviser.

Dr.

Bhavnani’s input was crucial to putting me on the current research path.
Dissertation committee members Cristina Bodea, Jakana Thomas and Christina DeJong, for
numerous helpful comments on my work; Dr. Clionadh Raleigh, for the invaluable feedback she has
provided on my dissertation; Dr. Dan Miodownik, for sharing his expertise in computational
modeling and visual analysis; Dr. Steve Kautz, Director of Graduate Studies, for his financial support
and steadfast encouragement to complete this study; and Dr. Im Kangsoo, Park Chunho and Han
Sungmin for their moral support and friendship.

Last but not the least, my family and the one

above all of us, the omnipresent God, who loves me and gives me the strength despite my
weakness and failings.

Thank you so much my Lord!

East Lansing, December 2012
Hyun Jin Choi

vi

TABLE OF CONTENTS
LIST OF TABLES ....................................................................................................................... ix
LIST OF FIGURES .......................................................................................................................x
CHAPTER 1: INTRODUCTION ..................................................................................................1
1.1 The Puzzle ........................................................................................................................1
1.2 Literature Review .............................................................................................................4
1.3 My Contributions .............................................................................................................8
CHAPTER 2: A SELECTORATE THEORY OF ETHNIC EXCLUSION ................................11
2.1 Summary of Selectorate Theory .................................................................................... 11
2.2 Concepts and Assumptions ............................................................................................13
2.2.1 Coalition Identity ..................................................................................................13
2.2.2 Leader’s Choice of Coalition Identity ...................................................................16
2.2.3 Selectorate’s Choice ..............................................................................................17
2.2.4 Coalition Size ........................................................................................................18
2.3 The Exclusion Model .....................................................................................................19
2.3.1 Ethnic Exclusion and Leadership Tenure .............................................................25
2.3.2 Coalition Size and the Likelihood of Ethnic Exclusion ........................................27
2.4 Chapter Conclusion ........................................................................................................28
CHAPTER 3: A COMPUTATIONAL MODEL OF ETHNIC EXCLUSION ...........................30
3.1 The Model ......................................................................................................................31
3.1.1 Incumbent Attributes ............................................................................................32
3.1.2 Challenger Attributes ............................................................................................33
3.1.3 Selector Attributes ................................................................................................33
3.1.4 Global Parameters .................................................................................................34
3.2 Sequence of Play ............................................................................................................36
3.2.1 Incumbent’s Policy Choice ...................................................................................38
3.2.2 Challenger’s Policy Choice ...................................................................................38
3.2.3 Selectors’ Choice ..................................................................................................39
3.2.4 External Shocks ....................................................................................................41
3.2.5 Evolutionary Algorithm ........................................................................................42
3.3 Experiments and Analysis ..............................................................................................44
3.3.1 Evolution of Ethnic Exclusion ..............................................................................44
3.3.2 Leadership Tenure ................................................................................................48
3.3.3 Political Mistakes ..................................................................................................50
3.4 Chapter Conclusion ........................................................................................................54
CHAPTER 4: EMPIRICAL ASSESSMENT I: STATISTICAL ANALYSIS ............................56
4.1 Ethnic Exclusion, Coalition Size, and Leader Survival .................................................56
4.1.1 Research Design I .................................................................................................57
4.1.2 Nonparametric Analysis ........................................................................................63

vii

4.1.3 Regression Analysis I: Leader Survival ................................................................66
4.2 Coalition Size and Ethnic Exclusion ..............................................................................73
4.2.1 Research Design II ................................................................................................74
4.2.2 Regression Analysis II: Ethnic Exclusion in the State ..........................................77
4.2.3 Regression Analysis III: Ethnic Exclusion of Minorities .....................................81
4.3 Chapter Conclusion ........................................................................................................83
CHAPTER 5: EMPIRICAL ASSESSMENT II: ILLUSTRATIVE EXAMPLES ......................85
5.1 Iraq (1958–2003) ...........................................................................................................86
5.2 Burundi (1962–1993) .....................................................................................................92
5.3 Rwanda (1962–1994) .....................................................................................................97
5.4 Chapter Conclusion ......................................................................................................101
CHAPTER 6: CONCLUSION ..................................................................................................103
6.1 Summation of the Argument ........................................................................................103
6.2 Implications and Conclusion ........................................................................................106
6.2.1 Theoretical Implications .....................................................................................107
6.2.2 Policy Implications ............................................................................................. 110
BIBLIOGRAPHY ........................................................................................................................ 114

viii

LIST OF TABLES
Table 3.1: Model Parameters .......................................................................................................35
Table 3.2: Input Assumptions for Model Runs ............................................................................43
Table 4.1: Ethnic Exclusion, Coalition Size, and Leader Survival ..............................................67
Table 4.2: The Determinants of Ethnic Exclusion (at the Country Level) ..................................76
Table 4.3: The Determinants of Ethnic Exclusion (at the Group Level) .....................................80

ix

LIST OF FIGURES
Figure 2.1: Changing Cost of Ethnic and Non-ethnic Identities ..................................................14
Figure 2.2: Supply and Demand for Coalition Membership ........................................................22
Figure 3.1: Screenshots of the Simulation Process ......................................................................36
Figure 3.2: Evolution of Ethnic Exclusion Type 1, by w ............................................................45
Figure 3.3: Evolution of Ethnic Exclusion Type 2, by w ............................................................46
Figure 3.4: Survival of Leaders by Ethnic Exclusion ..................................................................49
Figure 3.5: Number of Supporters in Large-W Systems (w = 0.5) ..............................................51
Figure 3.6: Political Liberalization (at t = 28) in Small-W Systems (w = 0.1) ............................53
Figure 4.1: Ethnopolitical Configuration of the State ..................................................................59
Figure 4.2: Frequency Distribution of Exclusion Scores .............................................................61
Figure 4.3: Kaplan-Meier Survival Estimates, by W ...................................................................64
Figure 4.4: Kaplan-Meier Survival Estimates, by W and Exclusion ............................................65
Figure 4.5: Ethnic Exclusion, Civil War, and Predicted Hazard Rates .......................................71
Figure 4.6: Predicted Hazard Rates during Civil War .................................................................72
Figure 4.7: The Effect of W on the Level of Ethnic Exclusion ...................................................77
Figure 4.8: The Probability of High-level Exclusion (Exclusion≥3) by W ..................................78
Figure 4.9: The Effect of W on the Risk of Ethnic Exclusion .....................................................82

x

CHAPTER 1
INTRODUCTION

1.1 The Puzzle
When British rule ended in 1961, Sierra Leone’s domestic politics was largely dominated by the
Mende people, then the largest ethnic group comprising 36 percent of the population. The ruling
Sierra Leone People’s Party (SLPP) led by Sir Milton Margai was widely perceived as
promoting Mende interests with many of the important positions in civil service filled with
Mendes (Allen, 1968, p. 309; Horowitz, 2000, p. 474). However, the election of Siaka Stevens,
an ethnic Limba, as prime minister effected a drastic change in the composition of the governing
coalition. After assuming office in 1968, Stevens began to eliminate his ethnic rivals from
political power. First he purged Mende officers in the army and eliminated the SLPP from
electoral competition; then he removed even the Temne―the second largest ethnic group that
had allied with him during his rise to power―from key positions in the government, the party,
and the army and filled these positions with members of the Limba community. By 1971,
Stevens ended up ruling with the active support of Limbas that comprise less than 10 percent of
the population. Although a group of soldiers and civilians, mostly Mende and Temne, plotted a
coup d’état in 1974 (Cox, 1976, pp. 227-228), Stevens had survived in power for 18 years until
he peacefully retired from office in November 1985.
Similar tendencies are observed in Michel Micombero’s Burundi (1966-1976), Idi
Amin’s Uganda (1972-1979), Hafez al-Assad’s Syria (1971-2000), and Saddam Hussein’s Iraq
(1979-2003). In each case, a leader continued to exclude his ethnic rivals from state power until
only one or a small number of ethnic minorities came to dominate a much larger population.

1

Members of excluded ethnic groups attempted coups or rebellions against the state, but they
were easily thwarted by dedicated soldiers who belong to the same ethnic group as the leader.
Despite abounding ethnic grievances and internal military threats to their regimes, these leaders
had survived relatively long in office.

1

We might naturally expect that leaders who promote

integration and unity among ethnic groups and thus face lower risks of ethnic rebellion are more
likely to survive longer in office than those who face violent oppositions within their countries.
These examples, however, suggest that the opposite could in fact be true. Why is this so? Why
do some leaders deliberately foster ethnic hatred and exclusion even to the point of excluding the
majority of the population? Why do they continue to do so even though such a policy may
increase the risk of ethnic conflict? Also, when and under what conditions will ethnic exclusion
(or inclusion) be good politics for leaders, and when will it turn self-defeating?
In this dissertation, I develop a systemic model to explain the puzzling relationship
between ethnic exclusion and leader survival. I consider the relevance of well-known selectorate
theory (Bueno de Mesquita, Smith, Siverson, & Morrow, 2003) in exploring the underlying logic
of this relationship. The selectorate theory posits that (1) political survival is a primary concern
for all leaders, and (2) leaders attempt to attain this goal by allocating various mixes of private
and public goods to members of their winning coalition. Employing these basic concepts, I
develop a theory of ethnic exclusion (hereafter called the exclusion theory). The principal
characteristic of the exclusion theory is its emphasis on the role of ethnic ties as a key
determinant of membership in the leader’s coalition.
My analysis yields four major findings. When the winning coalition is small: (1) ethnic
groups outside of the leader’s coalition are more likely to be subjected to formal political
1

Idi Amin survived only 7 years in office, but he was overthrown not by domestic opposition
but by invasion of a foreign power.
2

2

exclusion ; (2) leaders who promote ethnic exclusion are more likely to survive longer in office
than those who do not promote such a policy; (3) leaders are more likely to be removed from
office in an irregular manner when they fail to promote ethnic exclusion; and (4) political
benefits from ethnic exclusion are large enough to offset the risk of being involved in domestic
armed conflicts. Overall, my findings suggest that ethnic exclusion is good politics for smallcoalition leaders even if it could increase the risk of civil war. The argument does not purport to
suggest that ethnic exclusion is also good politics for the country and bringing peace. In fact, the
practice of good governance without discrimination of any kind has been an important part in
reducing both the onset and recurrence of conflict (Hegre & Nygard, 2012). However, the
exclusion model does specify the conditions under which the promotion of such “good
governance” can be less effective or even harmful―not just for the leader, but for the citizenry
as a whole.
This dissertation is divided into four main chapters. In the first chapter, I construct a
theory that explains how institutions for selecting leaders create different incentives for leaders
to employ ethnic exclusion. Here I summarize the selectorate theory and then extend it by
explicitly accounting for the role of ethnic ties in the formation of the leader’s coalition before
deducing hypotheses about ethnic exclusion and leader survival. The second chapter formally
describes the exclusion model with an agent-based computational model (ABM). I use ABM to
study the origin and evolution of ethnic exclusion in small-coalition systems, and to understand
how random external shocks, such as major political and economic crises, and “mistaken” ethnic
policy choices made by leaders affect the length of leadership tenure. In chapters 3 and 4, the

2

I define ethnic exclusion as “intentional and targeted” exclusion of elites from particular ethnic
categories from state power including key positions in the ruling party, the army, and the central
government.
3

hypotheses generated in the first two chapters are put to empirical tests, using data on the tenures
of 982 leaders from 1946 to 2004 and four illustrated examples that are chosen to illuminate the
workings of the exclusion theory. In conclusion, I summarize the key findings of the preceding
chapters and discuss the policy implications of the study.
The rest of this introductory chapter will review the literature on ethnic exclusion and
conflict and describe how my dissertation contributes to existing debates on these topics.

1.2 Literature Review
Ethnic exclusion and inequality play a central role in the contemporary civil war literature. In
his research on ethnic minorities, Gurr (1993, 2000) argued that a group’s collective
disadvantages vis-à-vis other social groups are indirect sources of rebellion that act through
grievances and group mobilization. Adopting the concept of horizontal inequalities―defined as
“inequalities in economic, social or political dimensions or cultural status between culturally
defined groups”―Stewart (2008, p. 3) argued that both economically underprivileged and
politically excluded groups are likely to be more easily mobilized for participation in rebellion.
Similarly, based on geo-referenced survey data on welfare and socioeconomic inequalities in 22
Sub-Saharan African countries, Østby, Nordås, and Rød (2009) found that regional inequalities
in terms of education and household assets are positively associated with civil war onset.
More recently, Cederman and his colleagues (Cederman, Wimmer, & Min, 2010)
distinguished three categories of ethnic groups based on the new Ethnic Power Relations (EPR)
data set: (1) those who hold full control of government without sharing power with other groups;
(2) those who hold dominant control in a power-sharing government; and (3) those who are
excluded from the executive power. They found that excluded groups (category 3) are about

4

three times more likely to challenge the state than those groups included in the executive branch
(categories 1 and 2), and this effect is more pronounced when excluded groups have recently
experienced a “downgrade” in their power status. By extending the empirical scope of research
on horizontal inequalities and violent conflict, Cederman, Weidmann, and Gleditsch (2011) also
found that ethnic groups that are wealthy or poor are more likely to experience civil war than
those groups with average wealth in highly unequal societies.
In line with the above studies, a number of scholars proposed the removal of ethnic
exclusion and inequality as a way to prevent future civil wars. After estimating an empirical
model of civil war incidence in 161 countries between 1960 and 1999, Elbadawi and Sambanis
(2000) claim that civil wars in ethnically diverse societies are due to the failure of inter-group
cooperation. Based on this finding, they conclude that

Taking the view that civil wars are the extreme case of non-cooperation among social
groups, this paper has argued that, under the right conditions, Africa’s ethnic diversity
would actually enhance development efforts by promoting positive inter-group
interactions. A pre-condition for this is the ‘appropriate’ political framework which
focuses on participation, inclusion and consensus-building among social and especially
ethnic groups (Elbadawi & Sambanis, 2000, pp. 265-266).

To contain conflict in the future, Gurr (2000) also calls for democratic governance
directed at reducing discrimination, recognizing cultural pluralism, and promoting power sharing
and political inclusion for minority groups. Noting that there is a positive association between
the political exclusion of powerful ethnic groups and the onset of ethno-nationalist conflicts,

5

Vogt (2007, p. 3) suggests that policymakers need to “ensure that all relevant ethnic groups are
included within the coalition of power and that a (more or less) equal distribution of the state’s
benefits and costs among the whole population is provided.” In a similar vein, Hegre and
Nygard (2012) find evidence that countries characterized by low quality governance have a
higher risk of conflict recurrence. Thus, they suggest that any reform that improves governance
in terms of less corruption, better bureaucracies, and less repression and political exclusion may
reduce the risk of conflict.
While the scholarship discussed above identifies the broad condition—political exclusion
(or inclusion)—under which we can expect ethnic conflict (or ethnic peace), other scholars have
attempted to explain why some ethnic groups became subject to political exclusion in the first
place. Several studies focused on the relation between ethnic exclusion and competition for
political goods. Bates (1983) argues that ethnicity is a useful means of attaining the benefits of
modernity. According to Bates, ethnicity serves as the basis of political coalitions in Africa by
enabling group members to secure goods and services from the modern sector and by excluding
others from these benefits. In other words, ethnicity provides “a form of minimum winning
coalition (Riker, 1962), large enough to secure benefits in the competition for spoils but also
small enough to maximize the per capita value of these benefits (Bates, 1983, p. 165).”
Fearon (1999) also suggests that ethnic exclusion is more likely to occur in pork-based
politics—the rationale being that when pork is at issue, winners of the election have strong
incentives to take the maximum share of the spoils by limiting the size of their coalition.
Ethnicity serves this purpose, he argues, because it is hard to modify and convenient for
excluding losers. Similarly, Caselli and Coleman (2006) formally demonstrate that political
coalitions can be formed along ethnic lines in ethnically heterogeneous societies. Once the

6

winning coalition is formed, they note, members of the losing coalition will attempt to infiltrate
the winning one in order to participate in the allocation of political goods. For this reason,
members of the wining coalition have a strong incentive to use ethnic identity—race in
particular—as a “visual marker” to exclude non-members from the winning coalition.
Somewhat relatedly, Chandra (2004) suggests that ethnic political parties are more likely to
succeed in “patronage democracies.” Based on the study of ethnic politics in India, she shows
that voters tend to support a party that represents the interests of their own ethnic group to the
exclusion of others when this party has a favorable history of allocating jobs and services to coethnics and has a reasonable chance of winning.

3

Taken together, the studies discussed above show that (1) ethnic exclusion increases the
likelihood of civil conflict; (2) ethnic inclusion can decreases the likelihood of civil conflict; and
(3) ethnic groups are more likely to be subject to political exclusion in pork-ridden political
systems. However, these studies remain silent on the motivations and constraints that influence
leaders’ choices regarding ethnic exclusion. If policies of ethnic exclusion are simply a means to
benefit members of specific ethnic groups, why do leaders care about the welfare of her coethnics instead of her own wellbeing (or personal political success), let alone the welfare of the
broader public? Moreover, why do some leaders deliberately foster ethnic exclusion even if such
a policy increases the risk of domestic armed opposition, which may put her tenure in danger?
This dissertation attempts to fill this gap in understanding the strategic incentives faced by
leaders who choose to employ ethnic exclusion rather than integration.

3

Taking a more historical view of ethnic exclusion, Wimmer (2002, p. 5) shows that ethnic
conflicts and racism are integral parts of political modernization because “modern institutions of
inclusion (citizenship, democracy, welfare) are systemically tied to ethnic and national forms of
exclusion.”
7

1.3 My Contributions
Recently, an attempt to explain the strategic logic behind ethnic exclusion has been made by
Roessler (2011). Based on original data on the ethnicity of coup leaders and insurgents in subSaharan Africa, he argues that in Africa a policy of ethnic exclusion reduces the risk of coup
d’état at the cost of increasing the risk of rebellion. Roessler (2011, p. 302) describes this
relationship in the following way:

[E]thnic exclusion and subsequent group rebellion are the outcome of strategic
interactions between elites incorporated in the central government, especially
coconspirators, who collaborated to seize the state by force but find it difficult to sustain
cooperation due to mutual fears that they may be ousted from power in the future. To
resolve this commitment problem, African rulers tend to exclude their coconspirators
from the central government, though at the cost of increasing the risk of civil war with
their former allies.

In line with Roessler (2011), I analyze the relative risk of coups and rebellions associated with
ethnic exclusion. Yet, in examining how domestic institutions shape the politics of exclusion,
my analysis departs from previous scholarship in four important ways.
First, I consider leaders and domestic institutions as the centerpiece for understanding the
logic of ethnic exclusion. By modifying the selectorate theory (Bueno de Mesquita et al., 2003)
in a way that accounts for the role of ethnicity more directly, I examine how selection institutions
influence the probability that leaders will remain in power and show that different institutions
create different incentives for leaders to adopt ethnic exclusion as their survival strategy. This

8

approach not only subsumes existing explanations based on pork-based politics, but goes beyond
them to derive novel implications for the relationship between ethnic exclusion and leadership
tenure―that is, ethnic exclusion can be good politics for leaders despite its positive impact on
the risk of armed conflict.
Second, while the literature on ethnic conflict focuses on the impact of ethnic exclusion
and inequality on conflict onset, the literature on ethnic politics tends to focus on the underlying
causes of discrimination and ethnic salience.

This dissertation brings together these two

literatures, which rarely speak to each other, by analyzing both ethnic politics and conflict within
a coherent theoretical framework. It proposes that, in systems with small winning coalitions,
promoting ethnic exclusion can enhance the probability that leaders will survive in office despite
the increased risk of civil war, while failure to do so can put them at risk of being deposed in an
irregular manner―such as coup, revolt, and assassination―despite the reduced risk of civil war.
The exclusion model thus highlights more nuanced notions of ethnic exclusion and the
interconnections between domestic institutions and ethnic conflict in a manner not easily
addressed by previous conflict literature.
Third, I use ABM to specify micro-level mechanisms that link ethnic exclusion to
leadership tenure. The flexibility provided by ABM is well suited for analyzing systems of
“boundedly rational agents” who are able to make mistakes, learn and adapt their behavior based
on past experience (Miller & Page, 2007, pp. 81-83). Hence, ABM allows me to demonstrate the
evolution of ethnic exclusion in systems with different sizes of winning coalition, and to
understand how random external shocks and mistaken ethnic policies interact with domestic
institutions in determining the fate of leaders.

9

Fourth, the explanation I provide calls into question some of the policy recommendations
that are designed to eliminate ethnic-based discrimination and oppression. This dissertation
suggests that a fundamental problem of small-coalition systems is that there are no easy ways to
simultaneously promote good governance and leaders’ survival. In fact, reforms to improve
governance could return the country to violence and instability if policymakers ignore the
political incentives of leaders and thus fail to understand the risks associated with these reforms.
This does not mean to rationalize the continuation of discrimination, but advise caution and
preventive measures before attempting to abolish ethnic exclusion in small-coalition systems.

10

CHAPTER 2
A SELECTORATE THEORY OF ETHNIC EXCLUSION

This chapter provides the theoretical foundations of the arguments that I will make in this
dissertation. I start with a summary of the selectorate theory (Bueno de Mesquita et al., 2003),
from which I borrow a number of key concepts and assumptions. Building on the selectorate
theory, I develop a model that shows why ethnic exclusion can be good politics in smallcoalition systems. This chapter concludes with the derivation of four testable hypotheses on the
likelihood of ethnic exclusion and leader survival.

2.1 Summary of Selectorate Theory
In The Logic of Political Survival, Bueno de Mesquita and his colleagues (2003) assume that all
leaders have political survival as their primary goal. In order to stay in power, leaders must
maintain the support of their winning coalition by providing a mix of private and public goods.
Public goods, such as national defense and highway systems, are those which cannot be withheld
from anyone without withholding them for everyone. Hence, public goods benefit all members
of the society if they are provided. On the other hand, private goods, such as jobs in state-owned
companies and luxuries for the ruling class, are both rival in consumption and excludable.
Hence, leaders can restrict access to private goods to members of their winning coalition. The
smaller the winning coalition, they suggest, the larger the share of private goods received by
each coalition member and hence the greater the value of private goods to retain the support of
their coalitions. Therefore, leaders in small-coalition systems are more likely to rely on private
goods, rather than public goods, to reward their key supporters. When the winning coalition is

11

large, on the other hand, leaders have to dilute each member’s share of private goods due to
budget constraints. This decreases the value of private goods and makes public goods a more
effective way to reward coalition members. Therefore, in large-coalition systems, leaders tend to
shift the mix toward public goods.
In addition, Bueno de Mesquita et al. (2003) suggest that leaders with small winning
coalitions are more likely to stay longer in office than their large-coalition counterparts due to
the following reasons. First, the provision of private goods is central to small-coalition systems;
hence small-coalition leaders can attract support by promising their supporters private benefits.
Second, in small-coalition systems, members of the incumbent’s coalition know that they have a
low probability of being included in the winning coalition if the new leader comes to power.
This makes the incumbent’s supporters loyal to their present leader for fear of losing access to
future private benefits. In contrast, supporters in large-coalition systems have a low risk of
exclusion in future coalitions because new leaders will continue to provide them with high levels
of public goods. This makes defection to the challenger less risky in large-coalition systems.
Third, when an individual decides whether to support the challenger or not, she must
consider the probability of being excluded from the challenger’s new winning coalition after the
challenger comes to power. Such a probability increases as the size of the winning coalition
decreases. So, in small-coalition systems, the challenger cannot credibly convince her potential
supporters that they will receive private goods in the future, and therefore finds it harder to
garner support. This commitment problem faced by the challenger provides a huge advantage
for the incumbent leader in small-coalition systems. For all these reasons, leaders with small
winning coalitions tend to be long-lived in office even if they promote bad public policy. In their

12

words, “good policy is bad politics and bad policy is good politics for small-coalition leaders
(Bueno de Mesquita et al., 2003, p. 325).”

2.2 Concepts and Assumptions
In this dissertation, I extend the selectorate theory by explicitly accounting for the role of ethnic
ties in the formation of the leader’s coalition. To do so, I borrow ideas from Chandra (2006),
Caselli and Coleman (2006) and Fearon (1999) to make four assumptions that I use in building
the exclusion theory.

In this section, I provide detailed explanations of assumptions and

definitions of related concepts.

2.2.1 Coalition Identity
I assume that every leader chooses to form his coalition on the basis of specific identity.
Identities can be broadly divided into two different types: ethnic identities and non-ethnic
identities. In line with Chandra (2006), I define ethnic identity as a subtype of identity categories
whose membership is determined based on descent-based attributes. Descent-based attributes
can be acquired genetically (e.g., skin color, eye color, and height), through culture and family
history (e.g., language, accent, religion, last name, and parents’ place of birth), or through
government action (e.g., ethnicity printed on identity card) (Chandra, 2006, p. 400). On the other
hand, non-descent-based attributes include those acquired through personal history (e.g., age,
education, and income), or through one’s voluntary choices (e.g., political ideology, party
membership). Of course, this is not a hard and fast distinction, and in reality there is an area of
overlap between the two types of identity. For example, a Labor Party membership in North
Korea is not really a voluntary choice, but one of prestige inherited from parents. There is also

13

Ethnic Identities
Non-ethnic Identities

Less Costly

More Costly
The Cost of Changing Identity (C)

Figure 2.1: Changing Cost of Ethnic and Non-ethnic Identities

evidence that even political ideology is genetically transmitted by biological parents (Alford,
Funk, & Hibbing, 2005). Nevertheless, for the sake of conceptual clarity, I will distinguish
between ethnic and non-ethnic identities in the rest of this chapter.
Another key property that distinguishes ethnic identity from non-ethnic one is the cost of
changing identity (hereafter denoted as C). As a general rule, ethnic identities have higher C
than non-ethnic identities (Chandra, 2006, pp. 414-416; Fearon, 1999, pp. 16-17). Simply put, it
is more difficult to change one’s ethnic identity than non-ethnic identity. For example, if a
liberal ideology is used to determine who can be included in the leader’s coalition, a conservative
who wishes to become a coalition member can easily change his identity by simply declaring
himself a liberal or by supporting a liberal political party. However, it would be extremely costly,
if not impossible, to change one’s skin color if the coalition is formed among individuals with
particular racial characteristics.

This argument is illustrated in Figure 2.1, which shows a

hypothetical dimension in which all types of identities can be ordered according to their value of
C. Ethnic identities are generally located on the right side of the line, and non-ethnic identities
on the left.

14

It is important to note that ethnic identities, as defined here, are not necessarily equal to
nominal ethnic categories―such as Shia Muslim, Sunni Muslim, Hutu, Tutsi, and Korean―that
are commonly used in calculating the ethno-linguistic fractionalization (ELF) index (Roeder,
2001).

4

For instance, the ELF value of South Korea is 0 because Korea is considered by many

as a homogenous society consisting of a single ethnic group and single language. However,
ethno-regionalism has been a key factor in South Korean politics. Voters have been clearly
divided between the southeastern (Yongnam) and southwestern (Honam) provinces of South
Korea in almost every national election since 1971. Between 1968 and 1999, father’s place of
birth (Bonjok) had been printed on the national identity card, and it had been widely believed that
Honam people were discriminated by the Yongnam government, and vice versa. Hence, based
on the above definition of ethnicity, I regard Honam and Yongnam as ethnic identities because
one must be Honam (or Yongnam) if his father was Honam (or Yongnam).
A similar example is found in Rwanda. It is often considered that Rwanda is composed
of two major ethnic groups: Hutu and Tutsi. However, between 1962 and 1994, there had been
significant regional divisions within the Hutu population, in which northerners and southerners
were opposed to one another to the point where one group massacred members of the other
group. In this case, I consider both northern Hutu and southern Hutu as ethnic identities (or subethnic identities (Scarritt & Mozaffar, 1999) within the Hutu ethnic group) because they are
based on descent-based attributes―the place of birth and ancestral origin. Generally, sub-ethnic
identities are associated with higher values of C (thus are more difficult to change) than ethnic
identities at the higher level. It should be more difficult to change both the place of birth and the
origin of ancestors than to change only one.
4

The ELF index is interpreted as the probability that two randomly selected individuals within a
country will not belong to the same ethnic group.
15

Theoretically, the role of “coalition identity” in my model can be compared to the role of
“affinity” in the selectorate theory (Bueno de Mesquita et al., 2003, pp. 60-65), in which “the
incumbent chooses a coalition of those selectors from whom she expects the highest affinity” (p.
62).

The selectorate theory regards affinity as “small idiosyncratic tastes” that play only

secondary roles in the decision making process of leaders and members of the selectorate (Bueno
de Mesquita et al., 2003, p. 61). They write, “For the purpose of the selectorate theory, affinities
need not be large. Indeed in the mathematical model, affinities are only used to break ties when
all other considerations are identical” (Bueno de Mesquita et al., 2003, p. 61). My concept of
coalition identity can be understood as an expansion of affinity, with an emphasis on the nature
of different types of affinity. By explicitly considering the changing costs associated with
different kinds of coalition identity―with ethnic identities generally associated with higher
changing costs―I am able to show that coalition identity not only plays a significant role in the
decisions of leaders and selectors, but is also pivotal in determining which leaders survive in
office. I defer the explanation of this relationship until I present the basic framework of the
theory in Section 2.3.

2.2.2 Leader’s Choice of Coalition Identity
Second, I assume that the leader chooses a specific coalition identity that she thinks best ensures
her political survival. There are a multitude of historical examples where leaders choose their
own ethnic identity to form their winning coalitions. Saddam Hussein created his coalition on
the basis of “Sunni Arab north-west,” which corresponds to his own ethno-regional identity
(Tripp, 2007, p. 219). Before Idi Amin came to power in 1971, members of the Lango and
Acholi ethnic groups dominated the government and military in Uganda. However, Amin was a

16

Kakwa, and once in power he filled his coalition with Kakwa officers to the exclusion of Langis
and Acholis (Horowitz, 2000, pp. 487-492). Similarly, Juvénal Habyarimana, a Hutu, came from
the northwest province of Rwanda, and his coalition was largely dominated by the Hutu northwest (Straus, 2006, p. 23). In Zambia, leaders utilized linguistic identities in national elections
and tribal identities in local elections (Posner, 2004). Of course, not every leader uses ethnicity
as their coalition identity.

For instance, political coalitions in advanced industrialized

democracies tend to form along non-ethnic lines such as political ideology (Wagner & Kritzinger,
2012), party identity (Richardson, 1991), or socio-economic status (Simmons, 1967). Regardless
of whether they are ethnic or non-ethnic, my model assumes that every coalition identity is a
deliberate choice made by the leader.

2.2.3 Selectorate’s Choice
Third, I assume that selectors who are outside the leader’s coalition are not able to enter the
coalition unless they change their identity to the one chosen by the leader. For example, if the
Tutsi minority comprises the winning coalition in Rwanda, members of the Hutu majority must
change their ethnic identity to Tutsi before trying to enter the coalition.

For this reason,

becoming a coalition member entails a cost to those who are initially excluded from the
incumbent’s coalition. This distinguishes my approach from Bueno de Mesquita et al. (2003), in
which selectors choose between the incumbent and the challenger by comparing the benefits they
expect to receive from both sides (Bueno de Mesquita et al., 2003, pp. 85-86). In my model,
selectors assess not only the benefits, but also the costs of joining rival coalitions. Especially,
selectors attempt to enter the coalition if and only if the cost of changing their identity is less than

the expected benefit of public and private goods received by each coalition member (Caselli &

17

Coleman, 2006, p. 1; Fearon, 1999, pp. 16-17). Therefore, it is not surprising in my model that a
selector supports the challenger even if the incumbent provides a larger amount of benefits.

2.2.4 Coalition Size
Lastly, I treat the size of the winning coalition (W) as conceptually different from the “actual”
number of supporters in the leader’s coalition. Bueno de Mesquita et al. (2003, p. 51) defines the
winning coalition as a “subset of the selectorate of sufficient size such that the subset’s support
endows the leadership with political power over the remainder of the selectorate as well as over
the disenfranchised members of the society.” In other words, the size of the winning coalition is
the minimum number of supporters in the selectorate that a leader needs to stay in power (Bueno
de Mesquita et al., 2003, p. 90; Riker, 1962), and it can be understood as one of the institutional
characteristics of country. For instance, in many liberal democracies with simple majority voting,
W is typically about half of the selectorate; while in autocratic systems with rigged elections, W
can be much less than one-tenth of the selectorate.
However, I emphasize that the actual number of supporters in the leader’s coalition is not
necessarily equal to W. Simply put, it can be larger or smaller than W. I call the coalition
“downsized” if the actual size of the leader’s coalition is small in large-W systems. On the other
hand, it is “oversized” if the actual coalition size is large in small-W systems. Domestic politics
has seen many instances of this type of mistake, where leaders have failed to form a minimum
winning coalition (Riker, 1962). Consider, for instance, Abd al-Karim Qasim who ruled Iraqi
between 1958 and 1963. During his years in office, Iraq was a small-W system where political
power resided exclusively in the army (Andrews & Ra'anan, 1969, p. 73). However, Qasim had
an integrative vision for an Iraq, and thus tried to represent all Iraqis by including civilians with

18

diverse ethnic and ideological backgrounds in his government (Tripp, 2007, pp. 146-147).

5

In

the next section, I offer reasons why failure to pursue an exclusive ethnic policy produces an
oversized coalition in small-W systems, which in turn contributes to violent leadership turnover.

2.3 The Exclusion Model
In the previous section, I introduced the key concepts and assumptions that are needed to extend
the selectorate theory.

Here, I put these components together to explain how selection

institutions encourage certain ethnic policies and punish leaders who fail to respond to the
encouragement. In large-W systems, the demand for membership in the leader’s coalition is
generally low due to the low amount of private goods allocated to each coalition member. The
tragedy of large-coalition leaders lies in the fact that in order to survive, they must be engaged in
intense competition over who include more supporters in their coalition, despite the low demand
for its membership. Thus, it is rational for them to decrease the cost of changing identity, so that
more people are allowed to join their coalition. This can be done by forming a coalition based
on non-ethnic identities such as ideological beliefs. It is also straightforward to see why largecoalition leaders do not want to exclude anyone from joining their coalition based on ethnicity.
Doing so increases the hazard of losing office by allowing the challenger to include these
excluded people in his coalition without needing to attract them away from the incumbent’s
coalition. Therefore, forming an exclusive coalition based on ethnic identity is political suicide
in large-W systems.
In small-W systems, the demand for coalition membership is generally high due to the
high amount of private goods provided to each coalition member. The tragedy of small-coalition

5

A more detailed history of Iraq in this period will be discussed in Chapter 5, Section 5.1.
19

leaders is that in order to survive, they must exclude a large number of selectors from their
coalition, even though these people want to become their coalition partners. Since survival in
office depends upon the provision of private benefits to key supporters, leaders in small-W
systems have a strong incentive to limit the size (or prevent the expansion) of their coalition by
preventing the entry of outsiders. Forming a coalition based on ethnic identity serves this
purpose well. If the coalition is formed by the leader’s ethnic group, members of ethnic outgroups are not able to enter the coalition unless they change their ethnic identity. This strategy is
most effective when the leader chooses an ethnic identity associated with high C. That is, when
the ethnic identity of the leader’s coalition is highly distinguishable from all other types, it
reduces the transaction costs of enforcing coalition membership because it helps identify who is
real and who is disguising his identity, making the coalition less subject to infiltration by ethnic
others (Caselli & Coleman, 2006).
The theory also explains why small-coalition leaders do not want to pursue an inclusive
ethnic policy by forming their coalition based on non-ethnic identity. Doing so increases the risk
of losing office by producing an oversized coalition. When the leader’s coalition is organized
along non-ethnic lines in small-W systems, the size of the coalition will rapidly expand as large
numbers of selectors will try to enter the coalition in order to participate in the distribution of
private goods. Now, with additional members, the incumbent is forced to reduce the rewards she
provides to each coalition member, thereby diminishing both the welfare and loyalty of existing
coalition members. Thus, to protect their privilege, existing members of the leader’s coalition
may attempt to replace their current leader when their privilege is eroding. In addition, if the
challenger has not reduced the private benefits promised to his coalition, members of the
incumbent’s coalition―both existing and new―are more likely to defect to the challenger when

20

they expect to receive a larger amount of benefits from the challenger than from the current
leader, a situation highly unlikely in small-W systems without an oversized coalition. For these
reasons, forming an inclusive coalition based on non-ethnic identity increases the risk of losing
office in small-W systems.
Other than preventing the expansion of the coalition, “ethnicity” provides additional
benefits to small-coalition leaders. First, ethnic coalition is favored in small-W systems because
the leader finds it more effective to distribute benefits on the basis of bloc identity. Rather than
distributing private goods to every individual in the winning coalition, it is more cost effective to
buy support among a few important individuals who can deliver a bloc of votes through patronclient relationships (Bueno de Mesquita et al., 2003, pp. 63-64). Ethnicity can provide an
attractive basis for a bloc of votes. Members of the ethnic group are likely to support the same
candidate as their group leader because they can easily know who will be the beneficiaries of
private goods once a candidate endorsed by their leader comes to power (Chandra, 2004).
Secondly, coalition formation based on ethnicity increase the loyalty of coalition
members to the incumbent leader. Bueno de Mesquita et al. (2003, pp. 65-68) suggests that the
loyalty norm is generated by the risk of exclusion from a successful challenger's future winning
coalition. When political coalitions are organized along ethnic lines, access to private benefits is
also structured along ethnic lines. Suppose that there are two rival ethnic groups, A and B, in a
fictitious small-W system where the incumbent coalition represents group A and the challenger’s
coalition represents group B. In these circumstances, members of group A would fear that, if the
incumbent is deposed and the challenger enters office, the probability would be high that
members of group B would exclude them from access to private goods. Such a high risk of
exclusion from the challenger’s coalition drives group A’s loyalty to the incumbent leader.

21

A
D

P
P1

P2
S
0

W1

W2

W

B
D

D’

P
P1

P3

P2
S
0

W1

W3

W2

W

Figure 2.2: Supply and Demand for Coalition Membership. The y-axis represents the amount of
private goods per coalition member (P). The x-axis represents the size of the incumbent’s
coalition (W). For interpretation of the references to color in this and all other figures, the reader
is referred to the electronic version of this dissertation.
22

Figure 2.2 (cont’d)

C

D’

P

D

P1

P3

P2
S
0

W1

W3

W2

W

Excluded groups (E)

The risk of exclusion may not only be related to their access to private goods but also to their
physical survival in an ethnically divided society. For example, when Michel Micombero, a
Tutsi, came to power in Burundi, he acted ruthlessly to eliminate Hutus who held high positions
in the military and government under the previous regime. Similarly, Juvénal Habyarimana
executed key figures, mostly southern Hutus, from the previous regime after he seized power in
Rwanda. All of these are theoretical reasons for predicting that leaders are likely to form an
ethnically exclusive coalition in small-W systems.
The causal mechanism of the exclusion model is graphically depicted in Figure 2.2. The
supply curve, labeled S in Figure 2.2A, shows how the size of the leader’s coalition changes as
the amount of private goods received by each coalition member changes. The supply curve is

23

downward sloping, as suggested by the selectorate theory; the larger the coalition, the less
leaders are able to provide private goods to their coalition members. On the other hand, the
demand curve, labeled D, shows how the size of the coalition demanded by citizens depends on
the amount of private goods provided to each coalition member. The demand curve is upward
sloping. That is, holding other things equal, more citizens will want to join the coalition as the
level of private benefits goes up.
Let’s suppose a leader in a small-W system (W = W1), who initially provides a large
amount of private goods to coalition members—say, P1 in Figure 2.2B. This leader has an
incentive to form a minimum winning coalition of size W1, while citizens want a larger coalition
of size W2—a situation in which the coalition demanded exceeds the coalition supplied. What
would happen if the leader’s coalition is based on an identity with C = 0? The imbalance
between the supply and demand for coalition membership would create downward pressure for
the amount of private goods (per coalition member) as citizens outside the coalition tried to enter
the coalition and the leader did not block their entry. Eventually, the leader’s coalition would
expand until its size reaches W3, resulting in an oversized coalition. This outcome makes both
the leader and the existing coalition members worse off, as the leader is less likely to survive in
office with an oversized coalition and existing coalition members receive a smaller amount of
private goods, P3.
Alternatively, as shown in Figure 2.2C, the same leader can maintain her coalition size at
W1 by excluding a group of people E (or W2 − W1) from entry into her coalition. The leader can
do this by requiring that any individual who is a coalition member to have a particular set of
ethnic identities such as skin color and language. By doing so, she can not only prevent entry by
new members but continue to provide a large amount of private benefits to key supporters. This

24

is exactly how many dictators practice ethnic exclusion to maintain their small coalition.
Graphically, such an exclusive policy increases the cost of changing identity for those outside the
coalition, thereby shifting the demand curve from D to D’. As a result, the system clears at a
higher amount of private goods P1 and a smaller coalition size W1. This outcome makes both
the leader and coalition members better off, as the leader is more likely to remain in office with a
minimum-sized coalition and coalition members continue to enjoy a large amount of private
goods.

2.3.1 Ethnic Exclusion and Leadership Tenure
Having specified my causal mechanisms, I now move on to look at the observable implications
of the exclusion model. The first set of hypotheses concerns the relationship between ethnic
exclusion and leadership tenure. Although we might expect that leaders who promote ethnic
integration deserve a long tenure in office, my theory suggests that the opposite may be true in
small coalition systems. When W is small, leaders who pursue an exclusionary ethnic policy
would have longer terms in office because it prevents oversized coalition. When small-coalition
leaders do not promote ethnic exclusion, however, the size of their coalition will expand (as
shown by the right-pointing arrow in Figure 2.2B). This will not only reduce the rewards
provided to coalition members but also weakens their loyalty to the incumbent. Given that
political competition in small-W systems is based on the ability to provide private goods, such an
inclusive policy makes defection to the challenger more likely. Therefore, I hypothesize that

H1: Leaders who promote ethnic exclusion are more likely to stay longer in office than
those who do not promote ethnic exclusion in small-W systems.

25

In addition, I expect that leaders are more likely to be replaced in an irregular manner if
they do not employ ethnic exclusion in small-W systems. As indicated by the downwardpointing arrow in Figure 2.2B, the expansion of coalition in small-W systems decreases the
welfare of existing members in the leader’s coalition. Thus, in order to return their welfare to the
previous higher level, dissatisfied members of the leader’s coalition may attempt to displace the
incumbent in an irregular manner. They can carry out Coup d’état, revolt, or assassination as a
means to achieve this end (Bueno de Mesquita et al., 2003, p. 397). Moreover, this process can
be especially violent when inclusive ethnic policy is preceded by a long period of exclusive rule.
The longer the ruling elite profits from the small-coalition system, the less likely they relinquish
their privilege without a fight. This makes them more determined to use violence as a means to
assert power when their power and privilege are deeply threatened. Through a coup or other
means of irregular turnover, they eventually want to install a new leader who can advance their
interests in a smaller coalition. From this argument, I derive the following hypothesis:

H2: Leaders who do not promote ethnic exclusion are more likely to be removed from
office in an irregular manner than those who promote ethnic exclusion in small-W
systems.

Both H1 and H2 suggest that ethnic exclusion is good politics for leaders in small-W
systems because it enhances their prospects for remaining in office. However, according to
recent literature on civil wars (Cederman et al., 2010; Wimmer, Cederman, & Min, 2009), ethnic
exclusion also increases the likelihood of civil war by sharpening the grievances of excluded

26

ethnic groups. This finding raises a critical question: is ethnic exclusion good politics for smallcoalition leaders even if it increases the probability of facing armed conflict? Based on the
exclusion theory, I argue that ethnic exclusion is good politics even under the worst case scenario,
where the leader is involved in civil war, due to the loyalty of the military to the incumbent
leader. In small-coalition systems, the key positions in the military are often filled by members
of the leader’s ethnic group. Bashar al-Assad has filled almost all key positions in the Syrian
army and security services with members of the Alawite minority. Under Saddam Hussein,
members of the Sunni north-west held a near-monopoly on key posts in the Republican Guard
and regular armed forces. The commanders of the Sierra Leone’s army were all Limba under the
Stevens regime. Because these individuals in the military are so closely tied with the incumbent
in terms of ethnicity, their ethnic identity signals potential disloyalty to the challenger’s future
coalition, which makes them face high risk of purge if the current leader is overthrown and the
rival group comes to power (McLauchlin, 2010). For this reason, the loyalty of the military is
especially high when the small-coalition leader employs an “ethnic preference policy” in the
armed forces (McLauchlin, 2010). Since victory is essential for their access to private goods,
they also fight hard in the face of rebellion, making the incumbent more likely to win the conflict.
I summarize this reasoning in my third hypothesis:

H3: During civil war, leaders who promote ethnic exclusion are less likely to lose power
than those who do not promote ethnic exclusion in small-W systems.

2.3.2 Coalition Size and the Likelihood of Ethnic Exclusion

27

The next set of hypotheses proposes a negative relationship between the size of the winning
coalition and the likelihood of ethnic exclusion. If the causal mechanisms operate as the model
suggests―that is, if ethnic exclusion is good politics in small-W systems―there should be a
statistically significant tendency that leaders pursue higher levels of ethnic exclusion in small-W
systems. Another prediction that naturally follows from this tendency is that ethnic groups
outside of the leader’s coalition have a higher risk of being subjected to political exclusion in
small-coalition systems. I submit these predictions to separate tests by dividing H4 into two subhypotheses:

H4a: Leaders are more likely to engage in higher levels of ethnic exclusion in small-W
systems than in large-W systems.
H4b: Ethnic groups that are not in the incumbent’s coalition are more likely to be
subjected to formal political exclusion in small-W systems than in large-W systems.

2.4 Chapter Conclusion
In this chapter, I have outlined a theory of how the size of the winning coalition creates different
incentives for ethnic exclusion. While large-coalition leaders are encouraged to employ an
inclusive strategy towards out-group members, small-coalition leaders are incentivized to pursue
an ethnically exclusive strategy. Theoretically, I extend the selectorate theory of Bueno de
Mesquita et al. (2003) by accounting for the role of ethnic ties in the leader’s coalition. When
the winning coalition is large, leaders choose to form an inclusive coalition based on non-ethnic
identity; otherwise they will face undersized coalition with a higher risk of losing office. When

28

the winning coalition is small, leaders want to form an exclusive coalition based on ethnic
identity; otherwise they will produce oversize coalition with a greater risk of removal from office
in an irregular manner. Therefore, leaders who employ ethnic exclusion are less likely to lose
office in small-W systems than those who do not promote such a policy. I argue that this
relationship holds even in times of civil war caused by exclusionary ethnic policies. The overall
implication of the exclusion theory is that ethnic exclusion is good politics in small-W systems
even if it brings about civil war.
The exclusion theory left me with four testable hypotheses regarding how the size of the
winning coalition is related to ethnic exclusion and leader survival. Before I subject these
hypotheses to empirical tests in Chapters 4-5, I will formally describe the exclusion theory in the
form of agent-based computational model in Chapter 3. Although the exclusion model outlined
in this chapter fully focuses on the behavior of the incumbent in the absence of any disturbing
factors, the computation model will involve a strategic interaction between the incumbent and
the challenger within an artificial environment that allows mistakes, random noises, and external
shocks.

29

CHAPTER 3
A COMPUTATIONAL MODEL OF ETHNIC EXCLUSION

This chapter will build upon the theoretical framework established in the previous chapter, and
from it will present how computational modeling illuminates the role of leaders in the exclusion
of ethnic groups and how this exclusion can be an evolutionary outcome. While the theory
outlined in chapter 2 confines its focus to the incumbent leader, the computational model
introduced in this chapter accounts for both the incumbent and the challenger competing for
political office. I choose to describe my argument in the form of an agent-based model (ABM),
which provides the following advantages.
First, ABM is appropriate for modeling “changes” in the size of the leader’s coalition.
The size of the incumbent’s coalition can be broken down into three broad categories. The first
is simply an “appropriate (or minimum)” size that corresponds to the size of the minimum
winning coalition. The second is “undersized” coalition, which occurs when a leader forms a
small coalition in large-W systems. In this case, she is most likely to be punished by voters in
regular elections. The third category is “oversized” coalition, which occurs when a leader
chooses to form a large coalition in small-W systems. In this case, an oversized coalition may
bring about a violent leadership turnover that could have been avoided if the leader’s coalition
had been small. The formal mathematical approach works best for the first category. For
instance, Bueno de Mesquita et al. (2003) use their formal model to prove that the leader prefers
the “minimal winning coalition (p. 90)” since she “never benefits from adding additional
members to her coalition” (p. 123). The core of my theory, however, concerns oversized
coalition. When a leader has an oversized coalition in small-W systems, I argue, she does not

30

adopt prudent ethnic policies and chooses to base her coalition on non-ethnic identity. ABM is
well suited to modeling dynamic systems with oversized coalition. Specifically, the flexibility
provided by the ABM approach allows me to understand how boundedly rational behavior of
leaders results in either shrinkage or expansion of the leader’s coalition and how these outcomes
contribute to peaceful or violent turnover of leadership.
Second, with ABM, I can demonstrate the “evolution” of ethnic exclusion in small-W
systems. Rather than assuming perfect rationality and static equilibrium, I assume a scenario
where leaders occasionally make mistakes, learn from past mistakes or successes, and adapt their
behavior in ways that promote their survival. I use this assumption to show that ethnic exclusion
does not result from a single leader’s policy choice, but evolves from multiple generations of
leaders in small-W systems. With a variety of evolutionary algorithms available, computational
methods provide a convenient tool for studying the evolutionary process of ethnic exclusion.
Lastly, I can use ABM to investigate the robustness of the causal relationship proposed by the
exclusion theory. Especially, it can test whether the expected effects of ethnic exclusion on
leader survival are due to general causal mechanisms of the theory or to chance, mistakes, or
random external shocks. I turn next to a specification of my computational framework.

3.1 The Model
The model contains three agent types—selectors, the incumbent leader (L), and a challenger
(C)—with simple attributes and behavioral rules local to their environment.

The model

landscape consists of a 100 × 100 torus with each patch representing an individual selector
(therefore, a total of 10,000 selectors). L and C are randomly placed on one of the patches at the
beginning of each model run. The positions of the agents are not geographical locations, but

31

represent selectors’ relation to L or C―the closer the selector’s location is to L or C, the greater
the proximity between them in social identity space. For instance, if a selector is located on the
same patch with L (or C), she shares the same identity with L (or C). If her location is far from L
(or C), her identity is highly distinguished from L’s (or C’s). The attributes of the agents are as
follows.

3.1.1 Incumbent Attributes
The incumbent leader is defined by five distinct attributes: a budget b; a coalition identity e; the
size of the disenfranchised population d; a memory t; and a probability of mistake m. A budget
(b) ranges from 1 to 20,000 and denotes the total amount of financial resources available to the
leader during a given time step.

6

Ranging from 0 to 1, a coalition identity (e) denotes a key

identity that establishes the criteria for determining what constitutes a leader’s coalition. It
approaches 1 (or 0) as the leader’s winning coalition is organized along the lines of ethnic (or
non-ethnic) identity. So the higher the value of e, the higher the cost of changing identity; and
therefore the harder it becomes for selectors outside the leader’s coalition to become a coalition
member, and vice versa.
The size of the disenfranchised (d) represents the proportion of the population that is
disenfranchised by the leader, and ranges from 0 to 1. Members of the disenfranchised group
have neither a say in the selection of leaders, nor an opportunity to become a member of a
winning coalition of L or C. This model assumes that the leader disenfranchises her citizens in

6

The choice of b in this range is arbitrary at the outset of the simulation. In my model, the
initial value of b does not affect the simulation results as long as the challenger has the same
level of (prospective) budget (b*) with the incumbent. But, b can be affected by external shocks
during simulation, creating a situation in which the incumbent and the challenger have different
levels of budget size.
32

the order of who is farthest from her position in identity space. For example, if the leader
decides to disenfranchise only one citizen, this person should be chosen among the selectors who
have the longest distance to L on the model landscape. Taken together, e and d determine the
degree of ethnic exclusion employed by the leader—the former representing exclusion from the
leader’s coalition (hereafter called exclusion type 1), and the latter, exclusion from the polity
(exclusion type 2). In addition, leaders are defined by a memory of previous leaders’ ethnic
policies and lengths of tenure during the last t generations. Lastly, I include a parameter to
reflect the possibility that a leader mistakenly fails to imitate successful ethnic policies adopted
by her predecessors with probability m.

3.1.2 Challenger Attributes
The challenger is characterized by two attributes: its prospective budget b* and coalition identity
e*. A prospective budget (b*) denotes the total amount of budget in the hands of the challenger
once he comes to power, with a range between 1 and 20,000. A coalition identity (e*) is a key
identity that constitutes a challenger’s winning coalition and has the same properties as e—that is,
the higher the value of e*, the higher the difficulty in changing one’s identity. Since the
challenger has not assumed office yet, she, unlike the incumbent leader, has no legal power to
disenfranchise any citizens.

3.1.3 Selector Attributes
Each selector is endowed with an ideological preference r. The random variable r follows a
normal distribution with mean 1 and standard deviation σ r , which is a model parameter that is
fixed at the beginning of the simulation. I assume that selectors with ideological preferences

33

below 1 (the leftists) are more likely to favor a change in the status quo and thus are more
favorably inclined toward the challenger, while those with ideologies above 1 (the rightists) are
inclined toward the incumbent.

3.1.4 Global Parameters
Other than these attributes of the agents, this model has four global parameters: the size of the
minimum winning coalition w, the price of public goods p, the frequency and strength (denoted f
and s) of economic crisis and political scandal, a threshold for entering a coalition τ, and a
discount factor δ. Ranging from 0 to 1, w denotes the minimum proportion of supporters in the
selectorate that a leader needs to stay in power. In liberal democratic systems with simple
majority rule, the value of w is typically about 0.5. However, in autocracies, this number can be
much less than half. For example, the size of the minimum winning coalition in North Korea
consists of a tiny proportion of the population who are members of Kim Jung-un’s royal family
and top commanders of the military loyal to Kim. The price p ranges from 1 to 1,000 to denote
the cost of producing a unit of public goods. Next, this model assumes that a leader’s budget (b)
and selectors’ ideologies (r) are affected by two types of external shock—economic crisis and
political scandal, respectively—that occur with a probability f and strength s, both with a range
of 0-1. I assume that external shocks decrease the survival prospects of the incumbent leader
7

either by reducing his budget or by making selectors less favorable toward him. In addition to

7

Previous research suggests that a decrease in economic growth, especially during the economic
crisis, is commonly associated with an increase in the risk of removal from office (Bueno De
Mesquita & Smith, 2010; Debs & Goemans, 2010; Goemans, 2008) or governmental collapse
(Alesina, Özler, Roubini, & Swagel, 1996). A number of leaders have also been removed from
office due to political scandal, which is major corruption or abuse of power leading to severe
political consequences. These leaders include, for instance, Syngman Rhee of South Korea
(1948-1960), Richard Nixon of the United States (1969-1974), Yasuhiro Nakasone of Japan
34

Table 3.1: Model Parameters

Agent-Level Parameters
Incumbent Attributes
b
e
d
t
m

b ∈ [ 1 , 20000 ]
e∈ [ 0 , 1 ]
d∈[0,1]
t ∈ [ 0 , 20 ]
m∈[0,1]

Budget
Coalition Identity
Size of the Disenfranchised
Memory
Probability of Mistake

Challenger Attributes
b*
e*

b* ∈ [ 1 , 20000 ]
e* ∈ [ 0 , 1 ]

Budget
Coalition Identity

Selector Attributes
r

2
r ~ N [ 1 , σr ]

Ideological Preference

Global Parameters
w
p
f
s
τ
δ

w∈[0,1]
p∈ [ 1 , 1000 ]
f∈[0,1]
s∈[0,1]
τ∈[0,1]
δ∈[0,1]

Minimum Winning Coalition Size
Price of Public Goods
Frequency of External Shocks
Strength of External Shocks
Coalition Threshold
Discount Factor

these parameters, a coalition threshold τ and a discount factor δ, both of which range from 0 to 1,
are defined globally and common to all selectors. All of these global parameters are fixed at the
outset of a model run. Table 3.1 summarizes model parameters and value ranges.

(1982-1987), Chuan Leekpai of Thailand (1992-1995, 1997-2001), and Christian Wulff of
Germany (2010-2012).
35

A: Leader, Challenger, Selectors

B: Minimum Winning Coalitions
WL

Selectors
Leader

WC
Challenger

C: Oversized Incumbent Coalition

D: Incumbent Removed

Figure 3.1: Screenshots of the Simulation Process. In 3.1C and 3.1D, blue-colored patches
D, blue
represent the incumbent’s supportres, while red colored patches represent the challenger’s
s
red-colored
challenger
supporters.

3.2 Sequence of Play
At the beginning of the simulation, both the incumbent and the challenger are created and
randomly located on the two-dimensional torus composed of 10,000 patches, each representing
dimensional
an individual selector (see Figure 3.1A) Once the initial conditions are set, the simulation
3.1A).
proceeds through the following successive steps
steps:

36

1. The incumbent decides to disenfranchise a group of size d among selectors in the order of
who is farthest from his location in identity space, and then simultaneously both the
incumbent and the challenger announce their coalition identities (e and e*).
2. Both the incumbent and the challenger pick their minimum winning coalitions of size w ×
10,000―denoted WL and WC ―from selectors in the order of who is closest to their
locations in identity space (see Figure 3.1B).
3. The incumbent proposes a plan for xL public and g L private goods from a fixed budget b,
while the challenger proposes provisions of xC public and gC private goods from his
prospective budget b*.
4. All selectors (except those who are disenfranchised) simultaneously choose to support either
the incumbent or the challenger, or to remain independent.
5. The incumbent is removed and the challenger becomes the new leader if and only if the
following two conditions hold: (1) the incumbent retains less than w supporters in his
coalition; and (2) the challenger has more supporters than the incumbent (see Figure 3.1D).

8

If the challenger becomes the new incumbent leader, nature picks a new challenger and the
simulation returns to step 1. Otherwise, the incumbent remains in power and a new round
begins at step 3 where the incumbent and the challenger update their policy provisions based
on the number of supporters in their coalitions.

8

The same deposition rule is used by Bueno de Mesquita et al. (2003, p. 81)
37

Now, I move on to the specification of decision rule for each player.

9

3.2.1 Incumbent’s Policy Choice
I assume that selectors who are included in the incumbent’s coalition receive payoffs from two
different sources—the incumbent-provided public goods (xሻ and private goods (݃ሻ—and have
the following specific utility function: ܷሺ‫݃ ,ݔ‬ሻ = √‫+ ݔ‬ඥ݃. The incumbent’s strategy is, given a
budgetary constraint, to find the best possible provision of public and private goods in order to
maximize the welfare of her coalition members:

Incumbent rule: max xL, gL U ( xL , g L )

subject to the budget constraint that b = | WL | · g L + p · xL , where | WL | refers to the actual size
of the incumbent’s coalition. The value of | WL | is equal to the size of the minimum winning
coalition (w × 10,000) in the first round of simulation. If the incumbent survives the first round,
| WL | will be the actual number of supporters in the incumbent’s coalition in each subsequent
round.

3.2.2 Challenger’s Policy Choice
The challenger’s strategy, like the incumbent’s, is to find the spending policy that maximizes the
welfare of his coalition members within the limit of his prospective budget. Therefore, the
challenger’s optimal policy choice is
9

I apply insights and formulas from the selectorate model (Bueno de Mesquita et al., 2003, pp.
106-120) in specifying decision rule for each player.

38

Challenger rule: max xC , gC U ( xC , gC )

subject to the budget constraint that b* = | WC | · gC + p · xC , where | WC | is actual number of
supporters in the challenger’s coalition. The value of | WC | is updated in the same way as in the
incumbent case.

3.2.3 Selectors’ Choice
All members of the selectorate know how many public and private goods are offered by the
incumbent and the challenger. They also know whether they are chosen by the incumbent and/or
the challenger to be part of their minimum winning coalitions as well as their distance from L
and C in social identity space. Having this information, each selector calculates her utility from
joining each side’s coalition and enters the one that provides her with greater utility.
Specifically, the expected reward to a selector from joining the incumbent’s coalition in
the current and future rounds is given by Z L =

1
U ( x* , g * ) , where δ is a common discount
L L
1−δ

factor that takes the value between 0 and 1, and ( x* , g * ) are optimal policy provisions of the
L L
incumbent in each round. However, when a selector calculates the expected reward from joining
the challenger’s coalition, she must consider the probability of being excluded from the
challenger’s long-term coalition after the challenger comes to power. The selectorate theory
defines the probability that a selector is included in a successor coalition as (w / S), where S is the

size of the enfranchised population (or the selectorate) and is defined as (1 − d) in this model
(Bueno de Mesquita et al., 2003, pp. 65-68). Hence, the expected reward from joining the

39

challenger’s coalition is ZC =

1 w
w
* *
*
( U ( xC , gC ) + (1 − )U ( xC )) . The benefit from joining the
1−δ S
S

incumbent’s or challenger’s coalition is then calculated according to the following formula.

BL = ( Z L · r) / ( Z L + ZC ) and BC = ZC / ( Z L + ZC )

As such, I assume that selectors with ideological preferences (r) below 1 (or above 1)
underestimate (or overestimate) the benefit provided by the incumbent’s coalition. Both BL and
BC are normalized to lie between 0 and 1 by dividing by ( Z L + ZC ).
Entering a coalition also entails a cost to those who are not chosen by L or C as their
minimum coalition members. Those individuals have to change their identities before entering
the coalition. Therefore, for a selector who has not offered coalition membership, the cost of
joining either the incumbent’s or challenger’s coalition is given by

CL = DL · e

and

CC = DC · e*

where DL and DC indicate the normalized distance of a selector from the locations of L and C
in identity space, respectively, and take values between 0 and 1. Coalition identities (e and e*)
determine how costly it is for a selector far from the location of L (or C) to enter his coalition.
The key intuition behind these equations is straightforward. The cost of changing identity is
higher when coalitions are organized along the lines of ethnic identity. The selector’s overall
utility from joining either the incumbent’s or challenger’s coalition is then given by

40

U L = BL − CL

and

U C = BC − CC

Therefore, the selector’s simple behavioral rule is:

Selector rule: if U L − U C > τ (where U L > 0), then enter the incumbent’s coalition; if U C − U L
> τ (where U C > 0), then enter the challenger’s coalition; else remain independent.

3.2.4 External Shocks
In addition to the basic rules described above, this model also assumes that the incumbent budget
(b) and selectors’ ideologies (r) can be affected by random external shocks. Given that surprises
are inevitable in systems of humans, ABM is particularly useful in analyzing the effects of
unpredictable events on leadership tenure, which are not easily addressed by other traditional
modeling methods. In this model, I assume that two types of external shock—economic crisis
and political scandal—create chances for the incumbent to be removed from office either by
decreasing the incumbent’s resources or by shifting selectors’ ideologies to the left. When
external shocks occur, they affect b and r according to the following sets of equations:

(1) The Impact of economic crisis on b at time t
bt = (1 − ρ )b + ρ ⋅ bt −1 − vt
vt = ϕ ⋅ bt −1
(2) The impact of political scandal on r at time t
rt = (1 − ρ )r + ρ ⋅ rt −1 − ε t

ε t = ϕ ⋅ rt −1
41

where 0 ൏ ߩ ൏ 1 and ϕ is a random number greater than or equal to 0 but strictly less than s
with probability f, or 0 with probability (1 − f ) in each round of play. As such, the current
values of budget and ideologies are part of their original and previous values plus a random
shock weighted by coefficient ρ. How fast the effect of previous shocks decays depends on the
value of ρ. Since 0 < ρ < 1, all shocks will eventually dissipate over time and b and r will revert
to their original values.

3.2.5 Evolutionary Algorithm
The evolutionary algorithm used in this model is imitation-based learning of ethnic exclusion
(Alkemade, 2004; Page, 1999, p. 37). This algorithm is based on the assumption that a current
leader learns from previous leaders’ experiences in order to acquire knowledge of different types
of ethnic policy. Specifically, the learning mechanism allows a leader to formulate her ethnic
policy in the following way:

1. After assuming office, a new incumbent observes previous leaders’ ethnic policies—
coalition identity (e) and the scale of disenfranchisement (d)—as well as their lengths of
tenure during the last t generations.
2. Then, she imitates the most successful strategy―the one that brought the longest
tenure―with probability (1 − m), or simply chooses a strategy randomly with probability m.

This is a process of learning whereby successful strategies are imitated by following leaders, and,
occasionally, new strategies are introduced by random exploration or mistake. When a leader

42

Table 3.2: Input Assumptions for Model Runs
W size
Model
Variant

w

Exclusion parameters
e

Simulation with Evolutionary Algorithm
0.05
―
1
0.05
―
2
0.1
―
3
0.1
―
4
0.25
―
5
0.25
―
6
0.3
―
7
0.3
―
8
0.45
―
9
0.45
―
10
0.5
―
11
0.5
―
12

Noise parameters

d

m

f

s

2
σr

―
―
―
―
―
―
―
―
―
―
―
―

0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.2

0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.2

0.1
0.3
0.1
0.3
0.1
0.3
0.1
0.3
0.1
0.3
0.1
0.3

0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.2

N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A
N/A

0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.1
0.1

0.1
0.3
0.1
0.3
0.1
0.3
0.1
0.3
0.1
0.3
0.1
0.3
0.1
0.1
0.1

0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.2
0.1
0.1
0.1

Simulation without Evolutionary Algorithm
0.05
Random
0.25
13
0.05
Random
0.25
14
0.1
Random
0.25
15
0.1
Random
0.25
16
0.15
Random
0.25
17
0.15
Random
0.25
18
0.4
Random
0.1
19
0.4
Random
0.1
20
0.45
Random
0.1
21
0.45
Random
0.1
22
0.5
Random
0.1
23
0.5
Random
0.1
24
0.5
0.1
0.1
25
0.5
0.9
0.1
26
0.1
1.0→0.1
0.25
27

Note: In all model variants, the following parameters were held constant: incumbent budget (b =
10,000); memory (t = 10); challenger budget (b* = 10,000); challenger coalition identity (e* =
random number between 0 and 0.2); price of public goods (p = 500); coalition threshold (τ = 0.1);
and discount factor (δ = 0.9).

chooses a random strategy with probability m, it can be either harmful or beneficial to leader
survival. The process of evolution is facilitated by this “randomness,” which allows various
43

strategies to be rejected or imitated over time. In addition, I assume that the “first generation”
leader, who has no predecessors, always chooses e < 0.1 and d < 0.1. Hence, the goal of this
evolutionary algorithm is to show how ethnic-based exclusion evolves out of self-interested
interactions between three types of agent although no discrimination was present at the
beginning of simulation.

3.3 Experiments and Analysis
I run 27 distinct model variants in NetLogo (v. 4.0.5) to examine the effects of coalition identity
(e) and the scale of disenfranchisement (d)—taken together to denote the level of ethnic
exclusion—on leadership turnover in different sizes of minimum winning coalition. While
exclusion parameters (e and d) are programmed to evolve during the simulation in variants 1-12,
they are fixed at the outset of a model run in variants 13-26. For replicability of the simulation
results, Table 3.2 provides input assumptions for all model variants.

3.3.1 Evolution of Ethnic Exclusion
This section focuses on variants 1-12. I run each variant for a period of 80 generations of leaders
to examine how ethnic exclusion (type 1 and type 2) evolve over time in small-, medium- and
2
large-coalition systems under slightly noisy (m = 0.1, f = 0.1, s = 0.1, σ r = 0.1) or highly noisy
2
(m = 0.2, f = 0.2, s = 0.3, σ r = 0.2) conditions. Each variant is run 10 times with unique random

seeds to generate a total of 120 independent model runs. Figure 3.2 compares the evolution of
coalition identity (exclusion type 1) for three different sizes of minimum winning coalition. The
vertical axis shows the mean values of coalition identity (e), while the horizontal axis is the
number of generations. In small-coalition systems with 0.05 ≤ w ≤ 0.1, ethnic exclusion (type 1)

44

0

.2

.4

e

.6

.8

1

0.05 ≤ W ≤ 0.1

0

20

40

60

80

Number of generations

0

.2

.4

e

.6

.8

1

0.25 ≤ W ≤ 0.3

0

20

40

60

80

Number of generations

0

.2

.4

e

.6

.8

1

0.45 ≤ W ≤ 0.5

0

20

40

60

80

Number of generations

Figure 3.2: Evolution of Ethnic Exclusion Type 1, by w. The bandwidth for lowess curves is 0.5.

45

0

.2

.4

d

.6

.8

1

0.05 ≤ W ≤ 0.1

0

20

40

60

80

Number of generations

0

.2

.4

d

.6

.8

1

0.25 ≤ W ≤ 0.3

0

20

40

60

80

Number of generations

0

.2

.4

d

.6

.8

1

0.45 ≤ W ≤ 0.5

0

20

40

60

80

Number of generations

Figure 3.3: Evolution of Ethnic Exclusion Type 2, by w. The bandwidth for lowess curves is 0.5.

46

gradually evolves until the value of coalition identity reaches about 0.75. When small-coalition
leaders are allowed to imitate the successful strategies of their predecessors, they learn and
implement more exclusive strategies through trial and error and, eventually, establish winning
coalitions along the lines of identity with a high changing cost. This holds true even if ethnic
exclusion is absent in the first generation of leaders. On the other hand, in large-coalition
systems with 0.45 ≤ w ≤ 0.5, mean values of coalition identity hardly go above 0.2 during 80
generations, suggesting that leaders favor coalitions based on individual identities that are easy to
change. I also observe evolutionary process in systems with medium-sized winning coalitions
(0.25 ≤ w ≤ 0.3), but its trajectory is not as upward as the one in small winning coalitions.
Figure 3.3 illustrates the different evolutionary dynamics of type 2 exclusion for small-,
medium- and large-coalition systems.

The vertical axis shows the mean values of

disenfranchisement scale (d). The horizontal axis is the number of generations. Consistent with
theoretical expectations, leaders with small winning coalitions tend to exclude a larger
proportion of selectors from the polity than those with large coalitions. When the winning
coalition is small (0.05 ≤ w ≤ 0.1), leaders will end up disenfranchising about 30 percent of
selectors after round 20. On the other hand, when the coalition is large (0.45 ≤ w ≤ 0.5), only 15
percent of selectors, on average, are disenfranchised by leaders.
But, the question remains as to why the evolution of type 2 exclusion is less noticeable
than type 1 in small-coalition systems. The selectorate theory provides an answer. When d starts
off small, increases in the size of the disenfranchised enhances a leader’s survival in office by
preventing those disenfranchised from participating in the challenger’s coalition. Beyond a
certain point (0.3 in this experiment), however, further increases in d may put a leader at a higher
risk of losing office. This happens because the increases in d also increase the probability that a

47

selector is included in the successor winning coalition, or w / S. The higher the probability of
being included in a successor coalition, the less risk a selector is taking by supporting the
challenger. For this reason, excluding too many selectors from the polity weakens coalition
members’ loyalty to the incumbent leader, and therefore increases the hazard of losing office.

3.3.2 Leadership Tenure
In order to examine how the survival of leaders differs according to ethnic exclusion and
coalition size, I now run a model without the evolutionary mechanism, repeating the simulation
1000 times for each variant (from 13 to 24) to generate a total of 12,000 leaders with random
values of coalition identity (e). I focus on the effects of coalition identity (e) on leadership
tenure by holding the scale of disenfranchisement (d) either at 0.25 (for variants 13-18) or at 0.1
(for variants 19-24), which are the most plausible values for d given my previous experiments
using evolutionary algorithm.
Figure 3.4 shows the proportion of leaders who survive at a particular time in office,
estimated by the Kaplan–Meier method. In small-coalition systems with w ≤ 0.15 (Figure 3.4A),
leaders who organize coalitions along ethnic lines (e ≥ 0.7) find it easier to survive than leaders
who build their coalitions based on non-ethnic identities (e ≤ 0.4). 82 percent of leaders with e ≥
0.7 survive in office beyond the first round, while 68 percent of leaders with e ≤ 0.4 survive their
first round. By the end of round fifty, 52 percent of leaders with e ≥ 0.7 are still in office, while
only 33 percent of leaders with e ≤ 0.4 remain. The log-rank test of the difference in survival
rates between these two groups is statistically significant at the 0.01 level. In large-coalition
systems with w ≥ 0.4 (Figure 3.4B), however, coalition building along ethnic lines poses a threat
to leaders’ survival, although its effect is not as strong as in small-coalition systems. The median

48

1.00

A: Kaplan-Meier Survival Estimates, w ≤ 0.15

0.00

Proportion of leaders surviving
0.25
0.50
0.75

Ethnic coalition (e >= 0.7)
Non-ethnic coalition (e <= 0.4)

0

20

40
60
Time in office

80

100

1.00

B: Kaplan-Meier Survival Estimates, w ≥ 0.4

0.00

Proportion of leaders surviving
0.25
0.50
0.75

Ethnic coalition (e >= 0.7)
Non-ethnic coalition (e <= 0.4)

0

20

40
60
Time in office

80

Figure 3.4: Survival of Leaders by Ethnic Exclusion

49

100

survival times for leaders with ethnic and non-ethnic wining coalitions are 9 rounds and 11
rounds, respectively. The difference in survival rates is also statistically significant at the 0.01
level. Consistent with my theoretical predictions, these results suggest that ethnic exclusion is
good politics for small-coalition leaders although large-sized coalition can turn the same policy
into bad politics.

3.3.3 Political Mistakes
How does a political system punish leaders who do not implement an appropriate ethnic policy?
In order to examine the consequences of mistaken ethnic policy, I illustrate two scenarios where
a leader employs ethnic exclusion in large-W systems or fails to employ it in small-W systems.
In variant 25, a leader organizes a coalition based on non-ethnic identity (e = 0.1) in large-W
systems (w = 0.5). As shown in Figure 3.5A, four external shocks have threatened her survival
during 100 rounds of the simulation. Out of these shocks, only the second one (Shock #2) was
big enough to remove her from office. On the other hand, in variant 26, the incumbent’s
coalition is based on ethnic identity (e = 0.9) in large-W systems (w = 0.5). As shown in Figure
3.5B, this policy results in an undersized coalition, in which the number of supporters in the
leader’s coalition remains below the minimum winning coalition of 5,000 selectors. Three
external shocks have occurred during this simulation, and all of them led to the removal of
leaders. This result suggests that leaders who form an undersized coalition by promoting ethnic
exclusion are less immune to external shocks than those who do not promote it in large-W
systems. This happens because selectors who are excluded from the incumbent’s coalition
quickly switched their support to the challenger, making it possible that even a small-scale
external shock could remove the incumbent from office.

50

Number of supporters
2000
3000
4000

5000

A: Non-ethnic Coalition (e = 0.1)

Shock #2
Shock #1 Turnover #1

Shock #3

Shock #4

1000

Leader
Challenger
0

20

40
60
Time in office

80

100

Number of supporters
2000
3000
4000

5000

B: Ethnic Coalition (e = 0.9)

Shock #1
Shock #2
Turnover #1 Turnover #2

Shock #3
Turnover #3

1000

Leader
Challenger
0

20

40
60
Time in office

80

100

Figure 3.5: Number of Supporters in Large-W Systems (w = 0.5). Parameter settings are based
on variants 25 (Figure 3.5A) and 26 (Figure 3.5B).

51

Model variant 27 represents a scenario in which a small-coalition leader decides to
abolish ethnic exclusion at one point in time during her tenure. Figure 3.6 shows the results
when I decrease coalition identity (e) of the incumbent in small-W systems with w = 0.1. I hold
the incumbent’s coalition identity at 1.0 for the first 28 rounds. Then, I gradually decrease it to
0.1 between rounds 29 and 60. Figure 3.6A plots the welfare for members of the incumbent’s
coalition. It shows that the rewards received by each coalition member diminish over time after
the incumbent begins to reduce ethnic exclusion (or decrease her coalition identity) at round 28.
Figure 3.6B shows the number of supporters in the winning coalitions―both the leader’s and the
challenger’s―in each round of play. Even though the number of supporters in the leader’s
coalition tends to increase for some period of time after round 28, there is a sharp rise in the size
of the challenger’s coalition beginning in round 55.
Why is this so? When strong exclusion is present at the beginning of the simulation, a
gradual decrease in e produces an oversized coalition because selectors who were previously
excluded from the incumbent’s coalition will try to enter the coalition in order to share the
benefits of private goods. Beyond a turning point (round 53 in Figure 3.6B), however, further
decreases in e begin to decrease the number of supporters in the leader’s coalition as the amount
of private goods received by coalition members continues to decrease. Combined with the
promise of future payoffs from the challenger and weakened loyalty toward the incumbent, a
period of oversized coalition is followed by a sudden outburst of opposition to the incumbent in
round 55. As predicted by the exclusion theory, this example shows that the abolition of ethnic
exclusion in small-W systems place leaders at a higher risk of irregular removal from power.
Faced with decreasing welfare, dissatisfied members of the leader’s coalition may attempt to
enhance their welfare by displacing the incumbent in an irregular manner through coup d’état or

52

46

Welfare of coalition members
48
50
52
54
56

58

A: Welfare of Coalition Members

0

20

28
40
Time in office

60

8000

B: Number of Supporters in WL and WC

0

Number of supporters
2000
4000
6000

Leader
Challenger

0

20

28
Time in office

40

60

Figure 3.6: Political Liberalization (at t = 28) in Small-W Systems (w = 0.1). Parameter settings
are based on variant 27.

53

assassination. Bueno de Mesquita et al. (2003, p. 397) state, “[i]n a coup, they can displace the
incumbent with someone from their own ranks in the hope that the newly installed leader will
promote their interests.” This leader could also be removed by popular uprising (or an outburst
of opposition as shown in Figure 3.6B), as exemplified by the recent removal of long-time Arab
dictators, Muammar Gadhafi and Hosni Mubarak.

3.4 Chapter Conclusion
In this chapter, I developed a computational model to explore how the size of the winning
coalition influences the motivation for ethnic exclusion and leader survival. The following
results are obtained from the simulation. First, ethnic exclusion is an evolutionary outcome of
political competition. Leaders in small-W systems are likely to establish an ethnically exclusive
regime through multiple generations of trial and error. Second, leaders in small-W systems are
more likely to survive in office when they implement ethnic exclusion. Third, a large-coalition
leader who employs ethnic exclusion jeopardizes her tenure by producing an undersized coalition,
which makes the incumbent more vulnerable to external shocks. Fourth, a small-coalition leader
who renounces ethnic exclusion faces a higher risk of irregular turnover by producing an
oversized coalition, which might give rise to an outburst of opposition to the incumbent leader.
Having simulated hypothetical effects of ethnic exclusion and turnover dynamics with a
computational model, the next chapter (Chapter 4) will test whether ethnic exclusion, as
demonstrated by the model, is indeed associated with a longer tenure in small-W systems using
empirical data on leadership tenure. In Chapter 5, I will demonstrate the evolutionary process of
ethnic exclusion in three historical cases, together with an illustration of how political
liberalization in small-W systems was followed by an outburst of violence in two small African

54

countries―Burundi and Rwanda―in the early 1990s.

55

CHAPTER 4
EMPIRICAL ASSESSMENT I: STATISTICAL ANALYSIS

In Chapter 2, I argued that, in small-W systems, ethnic exclusion can enhance the probability that
leaders will remain in office despite its positive impact on the risk of armed conflict, while
failure to implement it can put them at higher risk of being deposed in a violent manner. In this
chapter, I present an empirical test of this argument. This chapter is composed of two broad
parts. In the first part, I test the determinants of leadership turnover and find that ethnic
exclusion is good politics in small-coalition systems even in times of civil war. In the second
part, I examine the determinants of ethnic exclusion and show that leaders really do implement
higher levels of ethnic exclusion in small-W systems. I now turn to an empirical assessment of
ethnic exclusion and leader survival.

4.1 Ethnic Exclusion, Coalition Size, and Leader Survival
In this section, I test how the size of the winning coalition and ethnic exclusion affect leader
survival based on new measures of selection institutions (W and S) and ethnic exclusion.
Consistent with the theoretical prediction, I find that ethnic exclusion has a significant effect on
the risk of removal from office. A strong exclusion policy decreases the risk of losing office in
small-W systems, while the same policy increases the risk of deposition in large-W systems.
Leaders without promoting ethnic exclusion are punished by violence in small-coalition systems,
while leaders with exclusion policy are punished by election in large-coalition systems. I also
find that civil war increases the risk of an irregular removal from office, but its effect is
conditioned by the level of ethnic exclusion employed by the leader.

56

4.1.1 Research Design I
I test the determinants of leadership tenure and irregular turnover (H1, H2, H3) using Cox's
semi-parametric proportional hazards models, with leader-years as the unit of analysis. The
dependent variable is the length of time leaders survive in office before he left office (in a
particular manner), and is taken from the Archigos data on leaders (Goemans, Gleditsch, &
Chiozza, 2008).

This data set provides information on the time and manner―regular or

irregular―of leaders’ entry into office as well as their departure in 188 countries from 1875 to
2004. In order to relax the proportional hazards assumption, I include an interaction term
between the natural logarithm of survival time and independent variables that fail to pass the
proportional hazards test based on the analysis of Schoenfeld residuals (Box-Steffensmeier &
Jones, 2004, pp. 131-133). In addition, I account for unobserved country-specific factors by
including country-level frailty terms that are assumed to follow a gamma distribution with mean
1 and unknown variance θ .
The first main independent variable is the size of the winning coalition (W). Previously,
a 5-point measure of W (hereafter denoted as WLPS ) was developed by Bueno de Mesquita et al.
(2003). One disadvantage of using WLPS is that it is hard to separate the effects of W from the
effects of regime type because this measure is constructed based on a subset of indicators used in
Polity IV democracy index (Monty G. Marshall & Jaggers, 2011). Given that the size of the
winning coalition is not by itself equivalent to regime type, a failure to control for regime type
can induce omitted relevant variable bias (Clarke & Stone, 2008). Recognizing this limitation of
WLPS , I develop a new measure of W (hereafter denoted as WEPR ) based on the EPR data set
(Cederman et al., 2010).

57

The EPR data set provides information about whether “the political elites that claim to
represent an ethnic group” are participating in the state’s executive branch in all countries from
1946 to 2005 (Cederman et al., 2010, p. 100). Participating groups are further divided into four
subtypes: monopoly (14%), dominant (15%), senior partner in power-sharing regime (25%), and
junior partner in power-sharing regime (46%). On the other hand, non-participating groups are
divided into regional autonomy (30%), separatist autonomy (3%), powerless (42%), and
discriminated (25%). It is important to note that groups participating in the executive branch do
not necessarily employ exclusionary ethnic policies in the EPR data set. More than 70 percent of
participating groups are involved in power-sharing arrangements with other ethnic groups. Even
among “dominant” ethnic groups, 64 percent are not engaged in intentional or targeted exclusion
of other ethnic groups. For instance, although “Whites” in the USA are coded as “dominant”
between 1966 and 2005, they did not employ any exclusion policy. Likewise, only 25 percent of
non-participating groups in the data are subjected to political discrimination. Therefore, in the
EPR data, participating in the executive branch does not mean promoting exclusive rule.
Similarly, not participating in the executive does not mean becoming the target of exclusive rule.
Having noted that, I move on to discussing how I measure W based on the EPR data. I
argue that the proportion of ethnic groups participating in the executive branch is a useful
indicator of W. In many societies, political coalitions are organized along ethnic lines and
ethnicity provides a basis for the distribution of political goods (Bates, 1983; Chandra, 2004;
Fearon, 1999; Ferree, 2004; Heger & Salehyan, 2007; Horowitz, 2000). In addition, it is the
executive branch that actually distributes most of the political goods―private benefits and public
goods―to coalition members (Bueno de Mesquita et al., 2003, pp. 163-164). For this reason, I
argue that ethnic groups participating in the executive branch―including the presidency, the

58

I

Participating in the
Executive Branch

Not Participating in the
Executive Branch

Discriminated

Size of the winning coalition (W)
Size of the selectorate (S)
Territorial boundary of the state

Figure 4.1: Ethnopolitical Configuration of the State

cabinet, the army, and key posts in the administration (Cederman et al., 2010, p. 99)―provide a
useful proxy for ethnic groups included in the winning coalition.
Thus, I use the proportion of ethnic group(s) participating in the executive branch, which
varies from 0 to 1, as an indicator of W. Also varying from 0 to 1, the indicator of the size of the
selectorate (S) is 1 minus the proportion of discriminated ethnic group(s). The way I measure W
and S is graphically displayed in Figure 4.1. The main advantage of using a new measure of W

59

is that this is not based on the Polity IV components, unlike the one used in Bueno de Mesquita
et al. (2003). This allows me to separate the effects of W from the effects of regime type by
including the Polity IV democracy score as a control variable. However, the use of WEPR is
based on the assumption that the winning coalition is organized along ethnic lines in every
country. For this reason, when I use WEPR , I restrict my analysis to countries where politically
relevant ethnic groups are present.

10

In practice, it is not easy to measure the exact size of the

winning coalition, nor do I believe that WEPR is always superior to other measures of W.
Therefore, rather than resting my analysis solely on WEPR , I will examine whether my results
are consistent when WLPS is used in my statistical model.
The second major independent variable is ethnic exclusion. I estimate the degree of
ethnic exclusion as a composite index Exclusion based on four indicators: 1) exclusion from
central government; 2) exclusion from national power; 3) exclusion from political participation;
and 4) exclusionary ideology of the ruling elite. For each of these indicators, I use the variables
MONOPOLY and DISCPOP, taken from the EPR data (Cederman et al., 2010), PARCOMP from
the POLITY IV data (Monty G. Marshall & Jaggers, 2011), and ELCELITI from the Political
Instability Task Force (PITF) data (Monty G. Marshall & Cole, 2011). Specifically, I assign
Exclusion one point if MONOPOLY is equal to 1, meaning that elite members who claim to
represent a particular ethnic group have monopoly power in central government at the exclusion
of other ethnic groups. When DISCPOP―the percentage of discriminated population―is larger

10

I follow the criteria used by Cederman et al. (2010) in classifying an ethnic group as
politically relevant. An ethnic group is considered political relevant “if at least one political
organization claims to represent it in national politics or if its members are subjected to state-led
political discrimination (Cederman et al., 2010, p. 99).”
60

2000
Frequency of Leader-Years
500
1000
1500
0

0

.25

.5
Exclusion Index

.75

1

Figure 4.2: Frequency Distribution of Exclusion Scores

than 0.01, another point is assigned to Exclusion. A DISCPOP larger than 0.01 means that
ethnic group members that account for more than 1 percent of the country's population “are
subjected to active, intentional, and targeted discrimination, with the intent of excluding them
from both regional and national power (Cederman et al., 2010, p. 101).” I assign Exclusion
another point if PARCOMP―the competitiveness of political participation―is not equal to code
4 or 5, so that participation is either repressed, suppressed, or factional competition between rival
ethnic groups, and there are no secular political groups regularly competing for political
influence at the national level (Monty G. Marshall & Jaggers, 2011, pp. 26-27). Finally, I add
one more point to the index of Exclusion when ELCELITI―the ideological character of the
ruling elites―is equal to 1, meaning that governing elites hold exclusionary ideology, which
they use to discriminate, oppress, or eliminate particular categories of people. Exclusion is then

61

a five-point ordinal scale; that is, leaders of systems with higher values are likely to have a
higher degree of ethnic exclusion. I normalize Exclusion to a number between 0 and 1 by
dividing it by 4 for the ease of comparing its effect with that of other independent variables.
Figure 4.2 is a frequency distribution of Exclusion scores. About 16 percent of leaders
(875 leader-years) have no exclusion policy of any kind (Exclusion=0), while about 5 percent
(274 leader-years) have the strongest form of ethnic exclusion (Exclusion=1). A score of 0.25
represents the modal level of exclusion to which 37 percent of leaders (2,032 leader-years)
belong. In addition, the Pearson correlation coefficients of Exclusion vs. WEPR , WLPS , and
Regime Type are -0.15, -0.33, and -0.44, respectively, suggesting that my index of ethnic
exclusion is not simply reflective of small-coalition systems or authoritarian states.
Next, I investigate the effects of civil war on leadership tenure by including a
dichotomous variable Civil War that equals 1 if a leader is engaged in civil war that reaches a
threshold of 25 battle-related deaths within a given year based on the UCDP/PRIO Armed
Conflict Dataset (Gleditsch, Wallensteen, Eriksson, Sollenberg, & Strand, 2002). I also check to
see whether my results are different if I include an alternative measure of civil war Civil War
(COW) with a higher threshold of 1,000 battle deaths per year based on the Correlates of War
(COW) data set (Singer & Small, 2006).
I add both leader-level and country-level control variables. At the leader-level, I include
the leader’s Age and the manner of Entry into office, both taken from Goemans et al. (2008). I
expect a positive coefficient on Age since old leaders are more likely to experience difficulty in
promising future benefits to coalition members than young leaders (Bueno De Mesquita & Smith,
2010, p. 943). The manner of Entry is coded as 0 if a leader came to power through a regular
process, or as 1 if through an irregular process, for example, as a result of coup d’état.

62

Especially, I am interested in examining how the manner of entry affects the hazard of losing
office in a specific manner. To control for country-level characteristics, I use four control
variables: GDP Growth, GDPPC , Population, and Regime Type. Growth, which is taken from
Debs and Goemans (2010), is measured by dividing the current year’s GDP by the previous
year’s GDP, and has been found to be a highly significant predictor of leadership tenure, with an
increase in GDP growth decreasing the risk of deposition (Bueno De Mesquita & Smith, 2010;
Debs & Goemans, 2010; Goemans, 2008). I also include the natural log of gross domestic
product per capita ( GDPPC ) and logged country Population as indicators of socioeconomic
development and demographic conditions. Both of these variables are taken from Debs and
Goemans (2010). Lastly, I rely on the Polity IV data to measure Regime Type (Monty G.
Marshall & Jaggers, 2011). The Polity score ranges from -10 indicating countries that are full
autocracies to 10 indicating countries that are full democracies. I include Regime Type to
separate the effects of W from the effects of democracy and expect its positive influence on the
hazard of losing office.

4.1.2 Nonparametric Analysis
Before performing regression analyses to test the proposed hypotheses, it appears relevant to let
the data “speak” for themselves by making no assumptions about the functional form of the
survivor function for the leaders’ data. Figure 4.3 plots the Kaplan-Meier estimates of the
survivor function for leaders from small-W and large-W systems. A system is considered to be a
large-coalition system if WEPR ≥ 0.75 and a small-coalition system if WEPR < 0.75.

As

predicted by the selectorate theory (Bueno de Mesquita et al., 2003), the curve representing the
survival function of small-coalition leaders is at each time above the one representing the

63

1.00
0.00

Proportion of leaders surviving
0.25
0.50
0.75

W > 0.75
W < 0.75

0

5

10

15

Years in office

Figure 4.3: Kaplan-Meier Survival Estimates, by W

survival function of large-coalition leaders. About 18 percent of leaders from small-coalition
systems remain in office by the end of ten years, whereas only 12 percent of leaders do so in
large-coalition systems. In addition, the log-rank test rejects the null hypothesis of the equality
of the two survival functions (chi-square=16.92, p=0.000).
Next, I split the sample into two separate groups―no exclusion and exclusion. “No
exclusion” group consists of leaders with Exclusion=0 and “exclusion” group includes leaders
with Exclusion=0.75/1.0. Figure 4.4 compares the survival curves of small-W and large-W
leaders within each group. If ethnic exclusion does not influence the leader’s survival, the
survival curves in the exclusion and no exclusion groups should display similar patterns. Figure
4.4A shows the survival curves for the exclusion group. Again, consistent with the selectorate
theory (Bueno de Mesquita et al., 2003), small-W leaders have a significantly lower risk of

64

1.00

A: Exclusion (Exclusion=0.75/1.0)

0.00

Proportion of leaders surviving
0.25
0.50
0.75

W > 0.75
W < 0.75

0

5

10

15

Years in office

1.00

B: No Exclusion (Exclusion=0)

0.00

Proportion of leaders surviving
0.25
0.50
0.75

W > 0.75
W < 0.75

0

5

10
Years in office

Figure 4.4: Kaplan-Meier Survival Estimates, by W and Exclusion

65

15

leaving office than large-W leaders in the exclusion group. By the end of their second year, 65
percent of leaders in small-W systems and 43 percent of leaders in large-W systems remain in
office. By the end of their fifteen years, 12 percent of small-coalition leaders are still in office,
while only 6 percent of large-coalition leaders remain. The difference between the two survival
curves is statistically significant at the 0.01 level (chi-square=15.03, p=0.000).
On the other hand, Figure 4.4B compares the survival of leaders in the no exclusion
group. Unlike what the selectorate theory predicts, I do not find any evidence that smallcoalition leaders survive longer than large-coalition leaders in this group. The log-rank test fails
to reject the null hypothesis of the equality of the two survival functions at the 0.05 level (chisquare=0.43, p=0.512). Overall, the results of nonparametric analysis suggest that leaders in
small-W systems have an incumbency advantage over their large-W counterparts only when they
employ ethnic exclusion. When leaders do not promote ethnic exclusion in small-W systems,
they no longer have such an advantage.

These findings are generally consistent with my

prediction that leaders who employ ethnic exclusion are more likely to survive in office than
those who do not employ it in small-W systems. In the next section, I test whether these results
are robust to the inclusion of other relevant factors by performing a regression analysis.

4.1.3 Regression Analysis I: Leader Survival
Table 4.1 presents the results of fitting six Cox regression models to the data on leaders. Model
1 is a baseline specification that replicates the findings of Bueno de Mesquita et al. (2003) and
consists of coalition size, selectorate size, and five control variables. As predicted by the
selectorate theory, the estimated hazard of losing office among large-coalition leaders (W=1) is
82.2% higher than that of small-coalition leaders (W=0). This effect is statistically significant at

66

Table 4.1: Ethnic Exclusion, Coalition Size, and Leader Survival
Cox Proportional Hazard Regression
(1)
VARIABLES
Coalition size (W)

0.600***
(0.194)

W * Exclusion
-0.467*
(0.283)

Civil War

GDP Growth
Log( GDPPC )
Log(Population)
Entry
Entry * ln(t)
Regime Type
Regime Type * ln(t)

(4)

(5)

(6)

W
(LPS)
-0.389
(0.306)
-1.592***
(0.577)
2.115***
(0.700)
-1.007**
(0.424)

S * ln(t)

Age

(3)

Civil War
(COW)

Regular
Turnover

Irregular
Turnover

-0.366
(0.284)
-1.715***
(0.342)
2.031***
(0.533)
0.739***
(0.141)
-0.305***
(0.057)
0.253**
(0.100)
0.021***
(0.004)
-2.632***
(0.484)
0.177***
(0.062)
0.015
(0.043)
0.157
(0.113)
-0.315***
(0.050)

-0.372
(0.306)
-1.498***
(0.577)
2.005***
(0.699)
-0.958**
(0.425)

-0.065
(0.396)
-0.952
(0.768)
1.929**
(0.910)
-0.896
(0.593)

-1.239**
(0.582)
-4.376***
(1.088)
4.989***
(1.379)
-2.383***
(0.730)

0.301**
(0.119)
0.014***
(0.004)
-2.486***
(0.485)
-0.019
(0.057)
0.022
(0.039)
0.400***
(0.102)

0.039
(0.129)
0.008*
(0.005)
-2.458***
(0.636)
-0.021
(0.068)
0.073
(0.048)
0.619***
(0.147)

1.025***
(0.179)
0.020**
(0.008)
-3.291***
(0.788)
-0.341**
(0.138)
-0.179**
(0.090)
-0.066
(0.169)

0.050***
(0.009)
0.025***
(0.003)

0.129***
(0.012)
0.033***
(0.004)

-0.111***
(0.019)
0.036***
(0.009)

-4830.06
0.165
5,276
982
850

-3116.06
0.170
5,276
982
552

-1137.78
0.838
5,276
982
214

Baseline

Exclusion

Selectorate size (S)

(2)

0.019***
(0.003)
-2.059***
(0.422)
-0.000
(0.049)
-0.022
(0.037)
0.160*
(0.084)
-0.231***
(0.038)

0.276***
(0.099)
0.014***
(0.004)
-2.583***
(0.474)
-0.009
(0.057)
0.010
(0.040)
0.385***
(0.103)
0.049***
(0.009)
0.025***
(0.003)

Log-likelihood
-6532.85
-4829.35
-4868.61
Theta
0.199
0.167
0.225
Observations
6,401
5,276
5,236
Number of subjects
1,218
982
981
Number of failures
1,090
850
855
Standard errors in parenthesis
* p < .1; ** p < .05; *** p < .01 (two-tailed tests)

67

the 0.01 level. On the other hand, the negative coefficient for S indicates that leaders in systems
with a large selectorate find it easier to survive as anticipated by the selectorate theory, although
the effect is only marginally significant (p=0.100). Therefore, the results based on new measures
of selection institutions ( WEPR and S EPR ) are consistent with the selectorate theory’s main
predictions.
The control variables also provide additional insight into the logic of leader survival. The
positive coefficient on the age variable tells us that the hazard of losing office is 1.9% higher for
each additional year of a leader’s age. For variables on economic performance, only GDP
growth significantly influences survival. For every 1% increase in Growth, the hazard of losing
office is 87.2% lower. The hazard rate for leaders who entered office in an irregular manner is
17.4% higher than that for those with no such history after one year in office. However, this
effect gradually decreases over time as the length of tenure increases. The coefficients on
GDPPC and Population are not distinguishable from zero.
Model 2 examines the effects of ethnic exclusion and civil war on leadership tenure by
including the Exclusion and Civil War variables as well as the interaction of Exclusion with
coalition size (W * Exclusion). This model also includes Regime Type as a control variable in
order to separate the effects of coalition size from the effects of democracy. The negative
coefficient estimates on the Exclusion variable indicates that the practice of ethnic exclusion
significantly decreases the risk of losing office for small-coalition leaders, lending support to H1.
In Model 2, the hazard of deposition for leaders who substantially exclude ethnic out-groups
from the political process (Exclusion=1) is 79.7% lower than that of leaders who do not employ
such a policy (Exclusion=0) in small-coalition systems (W=0). In large-coalition systems (W=1),
the effect of ethnic exclusion is the sum of the coefficients on Exclusion and its interaction with

68

W. This aggregate effect is positive and statistically significant (p=0.048), indicating that ethnic
exclusion increases the risk of deposition by about 68.6% for large-coalition leaders. These
results are consistent with the results of computational simulation in Chapter 3 (section 3.3.2). In
addition, the presence of civil war influences leader survival. In times of civil war, the hazard of
deposition increases by 31.8% for all leaders.
Model 3 repeats this specification, but replaces WEPR and S EPR with WLPS and S LPS in
order to see how my results change if I use an alternative measure of W and S. I do not find any
major differences in the effect of key independent variables: Exclusion, W * Exclusion and Civil
War remain consistent and significant even after using different measures of selection
11

institution.

In Model 4, which bases the coding of civil war on the COW data, I obtain a quite

similar coefficient and significance level for the Civil War variable. Overall, my statistical
findings are robust to alternative measures of selection institution and civil war, and suggest that
ethnic exclusion is good politics for small-coalition leaders, whereas the reverse is true for largecoalition leaders.
In Models 1-4, the event of interest was whether or not leaders lose office; however, there
are different ways in which leaders exit office. Goemans (2008) identifies two distinct ways of
exit: a regular and an irregular way. Leasers may lose office in a regular manner due to electoral
defeat, term limits, or ill health. But, they can also be removed from office in an irregular
manner as a result of “the threat or use of force” such as a coup, revolt, or assassination
(Goemans, 2008, pp. 4-5). For this reason, in Models 5-6, I perform a competing risks analysis

11

The GDPPC variable gains some of its impact and is now significant at the 0.01 level in
Model 3.
69

where I distinguish two different types of exit. In so doing, I can obtain a more refined estimate
of how ethnic exclusion and coalition size are associated with a particular way of losing office.
In Model 5, the negative coefficient estimate on the Exclusion variable indicates that the
presence of ethnic exclusion decreases the hazard of “regular” turnover for small-coalition
leaders (W=0). However, this effect is statistically insignificant. On the other hand, the effect of
ethnic exclusion on the hazard of “irregular” turnover is large and significant in small-coalition
systems, providing support for H2. In Model 6, I find that the risk of irregular turnover among
small-coalition leaders who employ ethnic exclusion (Exclusion=1) is only about 1.3% of the
risk for those who do not employ such a policy (Exclusion=0). These findings suggest that while
ethnic exclusion is good politics in small-coalition systems, its mechanism is driven by the
reduction in the hazard of irregular turnover.
In large-coalition systems (W=1), the policy of ethnic exclusion increases the hazard of
deposition both in a regular and an irregular manner, although its effect is statistically significant
only for the hazard of regular removal. In Model 5, the estimated hazard of regular turnover
among large-coalition leaders with exclusion policy (Exclusion=1) is more than 2.5 times that of
those for whom Exclusion=0. As such, unlike in small-coalition systems, ethnic exclusion is bad
politics for large-coalition leaders and its mechanism is driven by the increase in the hazard of
regular turnover.

The control variables provide additional insights.

Larger GDPPC and

Population decrease the hazard of irregular removal from office, while an irregular entry to
office (Entry=1) increases the hazard of regular removal. In addition, the impact of civil war is
much higher for the risk of irregular turnover but has less of a relationship to the risk of regular
turnover. That is, the outbreak of civil war does not significantly increases the risk of regular
removal from power but strongly increases the risk of irregular removal from power. These

70

.3
.25
.2
.15
.1

Smoothed hazard function

.05

3

6

9

12

15

Years in Office
Exclusion=0 Civil War=0
Exclusion=0 Civil War=1

Exclusion=1 Civil War=0
Exclusion=1 Civil War=1

Figure 4.5: Ethnic Exclusion, Civil War, and Predicted Hazard Rates, smoothed by a Gaussian
kernel density estimator. I hold all other variables at their means.

results show that each mode of exit may be influenced by different mechanisms and the two
approaches—the pooled and competing risks models—can result in different conclusions
regarding the effects of individual variables on leadership tenure.
Coupled with these numerical results, Figure 4.5 plots the predicted hazard rates
generated from Model 2 for W = 0.1. This figure provides compelling evidence that a variable
measuring ethnic exclusion has a powerful effect on the hazard of losing office in small-coalition
systems. Holding all other variables at their mean values, small-coalition leaders who employ
ethnic exclusion (Exclusion=1) in times of peace (Civil War=0) face the lowest hazard of losing
office. This leader has a lower risk of deposition than small-coalition leaders who do not employ
the same policy (Exclusion=0) in times of peace (Civil War=0). On the other hand, leaders with

71

A: Overall Hazard of Losing Office

0

Smoothed hazard function
.1
.2
.3

.4

Civil War
Civil War and Exclusion

3

6

9
Years in office

12

15

.4

B: Hazard of Losing Office in an Irregular Manner

0

Smoothed hazard function
.1
.2
.3

Civil War
Civil War and Exclusion

3

6

9
Years in Office

12

15

Figure 4.6: Predicted Hazard Rates during Civil War, smoothed by a Gaussian kernel density
estimator. I hold all other variables at their means. Shaded areas represent 95% confidence
intervals.

72

small winning coalitions face the highest risk of deposition if they do not employ ethnic
exclusion (Exclusion=0) in times of war (Civil War=1). However, the risk of deposition would
drop significantly if this leader adopted ethnic exclusion (Exclusion=1) even during civil war
(Civil War=1).

Overall, in small-coalition systems, both ethnic “inclusion” and civil war

increase the hazard of losing office, although the effect of the former is much larger than that of
the latter—that is, ethnic inclusion is riskier than civil war for small-coalition leaders. These
results suggest that ethnic exclusion is good politics for small-coalition leaders even if it could
increase the risk of civil war.
As a way to focus on the relationship between ethnic exclusion and civil war in small-W
systems (W=0.1), I plot the difference that these two variables have on the predicted hazard rates,
holding other variables at their mean values. Figure 4.6 displays the hazard rates associated with
different values of the Exclusion variable (Exclusion=0 or Exclusion=1) in times of civil war
(Civil War=1) with 95 percent confidence intervals. In Figure 4.6A, it is notable that smallcoalition leaders without exclusion policy are about 5 times more likely to be removed from
office during civil war as compared with those who promote exclusion along ethnic lines, thus
supporting H3.

The effect of ethnic exclusion is especially pronounced for the hazard of

irregular turnover. In Figure 4.6B, small-coalition leaders with Exclusion=0 are about 40 times
more likely to be removed from office in an irregular manner during civil war than those with
Exclusion=1. Surprisingly, in small-coalition systems, the risk of irregular turnover virtually
disappears even in times of civil war if a leader has a strong exclusion policy.

4.2 Coalition Size and Ethnic Exclusion

73

In this section, I test whether leaders in the real world behave as predicted by the exclusion
model. In other words, if ethnic exclusion really is good politics for leaders in small-W systems,
I should find higher levels of ethnic exclusion in small-W systems than in large-W systems. For
the same reason, I also expect that minority ethnic groups are more likely to be subjected to
political exclusion in small-W systems.

4.2.1 Research Design II
In order to test the determinants of ethnic exclusion implemented in the state (H4a), I estimate
ordered logit regression models using country-years as the unit of analysis. The dependent
variable is a five-point index (from 0 to 4) of ethnic exclusion (Exclusion), which is obtained in
the same way as described in section 4.1.1. I account for unobserved country-specific factors by
including random intercepts at the country level. To test whether ethnic minorities face a higher
risk of violence in small-W systems (H4b), I estimate logistic regression models using groupyears as the unit of analysis. The Minorities at Risk (MAR) data set (Birnir & Gurr, 2009) is
appropriate for testing this hypothesis because it identifies a total of 283 politically-active and
mobilized minority ethnic groups in all countries from 1940 to 2003. Ethnic groups included in
the MAR data set can either suffer or benefit from “systematic discriminatory treatment vis-à-vis
other groups in a society,” and thus relevant to the causal process to be tested (Birnir & Gurr,
2009). The dichotomous dependent variable equals 1 if the POLDIS variable in the MAR data
set is coded “Exclusion/Repressive policy” where the group’s political participation is
substantially restricted by public policies in comparison to other ethnic groups (Birnir & Gurr,
2009). In order to account for unobserved country- and group-specific factors, I use three-level
logistic regression with a random intercept for each group and country. The explanatory variable

74

in both analyses is the size of the winning coalition (W), which is measured in the same way as in
section 4.1.1.
I use the following control variables. The first country-level control variable is the
amount of natural resource rents. I expect that leaders are less likely to exclude members of
ethnic out-groups when rents from natural resources allow them to pay off the larger coalition. I
choose an annual Oil Production (in barrels) as a reasonable proxy for resource rents. The data
for this variable are taken from Wimmer and Min (2006). This variable is measured for all oil
producing countries for the years 1946-2005. The next country-level control variable Past
Conflict Country, which is taken from the PITF data (Monty G. Marshall & Cole, 2011), is the
number of times a country has previously been engaged in ethnic conflict. I expect a positive
effect for this variable since we can assume that leaders of the country with a past history of
rebellion are more likely to engage in exclusionary politics against rival ethnic groups. Other
controls― GDPPC , Population, and Regime Type―are the same country-level variables as
described in section 4.1.1.

12

In order to control for group-level characteristics, we use six control variables. Group
Size is the natural log of group size as estimated by Birnir and Gurr (2009), which is expected to
have a positive effect, because large ethnic groups tend to have more salient ethnic identities
(Posner, 2004) and mobilization potential (Gurr, 1993), and thus are more likely to be the target
of political discrimination. The next group-level control variable Past Conflict Group is the
number of times an ethnic group has previously been engaged in ethnic conflict, and is taken
12

I include GDPPC as an indicator of socioeconomic development and expect a negative effect
on ethnic exclusion. I also include Population since large populations are generally associated
with a higher probability of conflict between ethnic groups (Fearon & Laitin, 2003; Weidmann,
2009).
75

Table 4.2: The Determinants of Ethnic Exclusion (at the Country Level)
Ordered Logit
VARIABLES

Logit
Model 3
LPS’s W

Model 1

Model 2

Model4

-3.747***
(0.320)
-0.123***
(0.013)
0.977***
(0.101)
-0.564***
(0.111)
-0.830***
(0.140)
-19.916***
(1.416)
-15.123***
(1.392)
-11.876***
(1.383)
-8.532***
(1.374)

-3.224***
(0.346)
-0.118***
(0.014)
0.741***
(0.110)
-0.560***
(0.121)
0.108
(0.151)
-0.314***
(0.011)
-11.557***
(1.540)
-5.793***
(1.518)
-1.981
(1.514)
2.162
(1.519)

-18.492***
(1.414)
-13.496***
(1.390)
-10.116***
(1.380)
-6.581***
(1.373)

21.020

20.519

18.974

127.83

-3430.26
4,409
126

-735.99
4,409
126

Fixed part:
Coalition size (W)
Log(Oil Production)
Past Conflict Country
Log( GDPPC )
Log(Population)
Regime Type
Constant 1
Constant 2
Constant 3
Constant 4

-3.978***
(0.203)
-0.090***
(0.013)
0.710***
(0.104)
-0.515***
(0.111)
-0.767***
(0.141)

-4.865***
(0.706)
-0.234***
(0.036)
1.269***
(0.186)
-0.796***
(0.240)
-0.759***
(0.285)
-0.202***
(0.017)
6.562**
(3.132)

Random part:
[Country]

Log likelihood
-3569.37
-3050.28
Observations
4,409
4,409
Number of countries
126
126
Standard errors in parenthesis
* p < .1; ** p < .05; *** p < .01 (two-tailed tests)

from Cederman et al. (2010). I expect a positive effect for this variable since we can assume that
ethnic groups with a past history of rebellion are more likely to be excluded from state power.
Finally, I test how patterns of settlement are associated with the probability of ethnic exclusion
by classifying ethnic groups into four broad settlement types―Concentrated in one region,

76

Majority in one region, Minority in one region, and widely Dispersed―as coded by Birnir and
Gurr (2009).

13

Previous studies suggest that the spatial distribution of ethnic groups has a

significant effect on conflict onset (Melander, 2009; Toft, 2003; Weidmann, 2009) and violence
against civilians during ethnic war (Bhavnani & Choi, 2012). I test whether settlement patterns

0

Probability of ethnic exclusion
.1
.2
.3
.4

are also related to the risk of ethnic exclusion.

0

.2

.4

.6

.8

1

W
Pr(E=0)

Pr(E=1)

Pr(E=3)

Pr(E=4)

Figure 4.7: The Effect of W on the Level of Ethnic Exclusion. I hold all other variables at their
means.

4.2.2 Regression Analysis II: Ethnic Exclusion in the State
Table 4.2 shows the results of the regression analysis. The results in Model 1 provide strong
support for H4a. The coefficient for W is negative and highly significant at the 0.01 level,

13

I use Dispersed as a reference category in all models.
77

Probability of ethnic exclusion
.12
.17
.22
.07
0

.2

.4

.6

.8

1

W

Figure 4.8: The Probability of High-level Exclusion (Exclusion≥3) by W. I hold all other
variables at their means. Shaded areas represent 95% confidence intervals.

indicating that the probability of the highest level of ethnic exclusion (Exclusion=4) in large-W
systems (W=1) is only about 30% of that employed in small-W systems (W=0), holding all other
variables constant at their means. In Model 2, I do not detect any major difference in the effects
of W even after controlling for the effects of democracy. In Model 3, I replace WEPR with
WLPS in order to test whether my models are sensitive to different ways of measuring W. The
use of WLPS makes no substantive difference for my results. The coefficient for W is still
negative and significant in Model 3 and its effect is slightly stronger than those in Models 1 and
2. Figure 4.7 provides a graphic depiction of how the size of the winning coalition influences the
level of exclusion policy implemented in the state. The plot, which is based on Model 2, shows
that small-coalition systems are more likely to have higher levels of ethnic exclusion than large-

78

coalition systems, lending support to H4a. Consider the probability of the highest level of
exclusion (Exclusion=4), which is indicated by black solid line. At W=0, the probability is about
0.1. As W increases, the predicted probability decreases to 0.06 at W=0.5 and 0.025 at W=1. On
the other hand, the probability of no exclusion (Exclusion=0), which is indicated by red dash-dot
line, begins at 0.075 at W=0 and ends at 0.18 at W=1. In Model 4, I dichotomize my dependent
variable (Exclusion) into low- and high-level of exclusion. The dependent variable is now
defined as 1 if Exclusion ≥ 3, and 0 otherwise. Figure 4.8 shows the relationship between W and
the probability of “high” exclusion based on the logistic regression in Model 4. Consistent with

H4a, a change in W from 0 to 1 decreases the probability of high-level exclusion by 15%,
holding all other variables at their means.
Regarding the control variables, the coefficient for Oil Production remains negative and
statistically significant across all four regressions, as predicted, and suggests that leaders tend to
employ a weaker form of ethnic exclusion in resource-rich countries. This finding is somewhat
counterintuitive given that resource abundance has been commonly associated with negative
political outcomes such as corruption (Leite & Weidmann, 1999), dictatorship (Jensen &
Wantchekon, 2004), and civil war (Collier & Hoeffler, 2004; Ross, 2004). In addition, GDPPC
is significant across all four models, decreasing the level of ethnic exclusion. The coefficient for
Past Conflict Country also has the expected positive sign, indicating that countries that have
previously been engaged in ethnic conflict are more likely to have higher levels of ethnic
exclusion. Less robust is my control for the country’s population. Population is negative and
significant in Models 1 and 3, but loses significance and changes its sign when I control for the
effects of regime type in Model 2.

79

Table 4.3: The Determinants of Ethnic Exclusion (at the Group Level)
Random-intercept Logistic Regression
VARIABLES

Model 4
LPS’s W

Model 1

Model 2

Model 3

-3.199***
(0.296)

-2.826***
(0.307)

-0.065***
(0.009)
-1.351***
(0.258)
0.364***
(0.069)
-1.459
(0.942)
-1.754*
(1.057)
-2.120*
(1.179)
-0.274***
(0.084)
0.395
(0.267)
17.057***
(1.990)

-0.056***
(0.010)
-1.329***
(0.253)
0.349***
(0.071)
-1.646*
(0.944)
-1.904*
(1.063)
-2.354**
(1.179)
-0.252***
(0.089)
0.640**
(0.265)
-0.111***
(0.008)
14.017***
(2.016)

-1.189*
(0.700)
-3.938**
(1.533)
2.516
(1.764)
-0.083***
(0.012)
-1.068***
(0.257)
0.312***
(0.076)
-1.653*
(0.950)
-1.983*
(1.071)
-2.307*
(1.215)
-0.224**
(0.097)
0.684**
(0.274)
-0.108***
(0.009)
9.154***
(2.382)

3.060
3.896

2.859
3.951

2.702
3.969

3.225
3.721

-3222.02
10,650
286

-3686.27
11,680
294

Fixed part:
Coalition size (W)
W * Difference
Difference
Log(Oil Production)
Log(Group Size)
Past Conflict Group
Concentrated
Majority
Minority
Log( GDPPC ), lagged
Log(Population)
Regime Type
Constant

-1.265***
(0.160)

-0.052***
(0.009)
- 1.181***
(0.243)
0.308***
(0.068)
-1.145
(0.906)
-1.537
(1.023)
-1.966*
(1.144)
-0.182**
(0.085)
0.130
(0.257)
14.784***
(1.923)

Random part:
[Country]
[Group]

Log likelihood
-3648.37
-3473.19
Observations
11,680
11,497
Number of groups
294
294
Standard errors in parenthesis
* p < .1; ** p < .05; *** p < .01 (two-tailed tests)

80

4.2.3 Regression Analysis III: Ethnic Exclusion of Minorities
The logistic regressions in Table 4.3 assess how the size of the winning coalition affects the
probability of political exclusion against ethnic groups. In Model 1, the negative coefficient of 3.199 on W indicates that as the size of the winning coalition gets larger, ethnic groups are less
likely to be subjected to formal political exclusion, thus supporting H4b. The control variables
provide additional insight into the logic of exclusion. As expected, the coefficient for Oil
Production is both negative and statistically significant, suggesting that ethnic groups tend to
face a lower risk of political exclusion in resource-rich countries. Both Group Size and GDPPC
have negative and statistically significant influences on the probability of ethnic exclusion. In
addition, the impact of Past Conflict Group is positive and significant, suggesting that ethnic
groups that had previously been engaged in conflict are more likely to be excluded from the
political process. The negative coefficients on the variables for spatial distribution indicate that
widely dispersed groups are most vulnerable to political exclusion, but these effects are only
marginally significant. The coefficient for Population fails to reach statistical significance at the
0.1 level.
In order to separate the effects of coalition size from the effects of regime type, I include
the Regime Type control variable in Model 2. The addition of this variable does not alter my
main results. Model 3 contains ethnic difference index (ETHDIFXX) in MAR data set and its
interaction with W. Ranging from 0 to 11, ethnic difference index measures the extent to which
a given ethnic minority group is different from the dominant group in terms of language, religion,
customs, and race (Birnir & Gurr, 2009). I name this index “Difference” and divide it by 11 so
that it varies between 0 and 1. According to my theory, the more “distinguishable” an ethnic
group is from the dominant group, the more difficult it is to disguise one’s identity to infiltrate

81

Probability of ethnic exclusion
.1
.2
.3
0
0

.2

.4

.6

.8

1

W

Figure 4.9: The Effect of W on the Risk of Ethnic Exclusion. I hold all other variables either at
their means (Group Size, GDPPC , Population, and Past Conflict Group) or at zero (Concentrated,
Majority, and Minority). Shaded areas represent 95% confidence intervals.

into the leader’s coalition. Hence, such a group is more likely to become the target of political
exclusion in small-coalition systems. Indeed, this is what I find. The large and negative
coefficient of -3.938 for the interaction of W and Difference indicates that the effect of W is
much stronger when ethnic groups have distinct characteristics that distinguish them from the
dominant group.
Model 4 is a robustness check of my primary findings. In order to test whether my model
is sensitive to the way I measure W, I replace a measure of W with WLPS , while otherwise
keeping the specification of Model 1 unchanged. The results from Model 4 demonstrate the
robustness of my main hypothesis (H4b). My results hold in predicting political exclusion of
ethnic groups with an alternative measure of coalition size. The variables of theoretical interest

82

(W and Oil Production) retain their effects and significance, as does the Group Size, Past
Conflict Group and GDPPC variables. Figure 4.9, which is generated from Model 2, displays the
predicted probability of ethnic exclusion associated with ethnic groups as a function of W,
holding all other variables either at their means or at zero. This figure clearly demonstrates the
effect of wining coalition, with higher probabilities of ethnic exclusion in smaller coalition
systems. Minority ethnic groups in small-W systems (W=0.2) are about three times more likely
to be subjected to political exclusion as compared with those in large-W systems (W=0.8).

4.3 Chapter Conclusion
Empirical tests of the proposed hypotheses using data on leaders and ethnic groups suggest that
there is indeed a powerful connection between the size of the winning coalition, ethnic exclusion,
and leader survival. Specifically, I find that in small-W systems: (1) leaders who employ ethnic
exclusion are more likely to survive in office than those who do not promote exclusionary
politics; (2) leaders are less likely to be removed in an irregular manner when they promote
ethnic exclusion; and (3) political benefits from ethnic exclusion are large enough to offset the
risk of civil wars. Therefore, in small-W systems: (4) leaders are more likely to engage in strong
exclusion policy; and (2) ethnic minorities are more likely to be subjected to formal political
exclusion.

Overall, my statistical analysis confirms this “unwelcome” conclusion: ethnic

exclusion is good politics in small coalition systems even if it could increase the risk of ethnic
conflict. On the other hand, ethnic exclusion is bad politics in large-coalition systems since
leaders who employ exclusion policy are more likely to be punished by voters in regular
elections. These results proved robust even after separating the effects of W from the effects of
democracy, and after using alternative measures of W and civil war. In the next chapter, I will

83

provide brief case histories that are chosen to illuminate and illustrate the workings of the
exclusion theory.

84

CHAPTER 5
EMPIRICAL ASSESSMENT II: ILLUSTRATIVE EXAMPLES

The exclusion model offered in Chapter 2 attempts to explain why leaders with small winning
coalitions tend to establish ethnically exclusive polity and why they do so in spite of the risk of
armed rebellion. I tried there to provide five hypotheses, which are strongly supported by both
computational and statistical evidence provided in Chapters 3 and 4. Nevertheless, whether my
theory is useful depends on how well it explains the actual behavior of small-coalition leaders.
Do they act as the exclusion model predicts? In order to answer to this question, Chapter 5
examines the ethnic policies of leaders in three small-coalition systems: Iraq from the time of the
First Republic in 1958 until the fall of Saddam Hussein in 2003; Rwanda from its independence
from Belgium in 1962 to the outbreak of genocide in 1994; and Burundi from the 1962
independence to the beginning of civil war in 1993. I choose to examine the behavior of
multiple leaders in each country’s history rather than focusing on a single leader because doing
so helps show that an exclusive polity governed by a small group of people with particular ethnic
backgrounds is not an outcome of one leader’s choice, but, as computational simulation suggests,
an evolutionary process over multiple generations of leaders.
Iraq (1958–2003) is a straightforward case that provides strong support for the exclusion
model. During this period, major ethnic rivals―the Kurds and Shia Muslims―were excluded
from important positions in the Iraqi government, and the country had become more and more a
narrowly defined regime which ended up being controlled by a small proportion of Sunni
Muslims, especially those affiliated with Saddam Hussein’s family. Surviving a defeat in the
Gulf War, economic sanctions, revolts, and assassination attempts, Hussein remained in office

85

for 24 years until he was finally deposed not by domestic challenges but by external powers in
2003. Developments after 1962 in Burundi and Rwanda display similar evolutionary trajectories
to Iraq. However, these two cases are distinguished from the Iraqi case in that one of their
leaders―Pierre Buyoya in Burundi and Juvénal Habyarimana in Rwanda―attempted to end
ethnic exclusion after a long practice of exclusive rule, thus showing how small-coalition leaders
are punished when they fail to employ ethnic exclusion. Rwanda is further distinguished from
Iraq and Burundi in that its leaders did not belong to minority ethnic groups. The exclusion
model predicts that leaders in small-coalition systems will end up ruling with the active support
of a small group of co-ethnics, in combination with the systemic exclusion of the ethnic majority.
But, the problem in post-colonial Rwanda appears to be the opposite: it was the Hutu majority
(constituting about 85 percent of the population) that had been in power until the Rwandan
Patriotic Front (RPF) victory in 1994. Hence, if my theory is to be persuasive, I must show that
1) a similar evolutionary (or narrowing) process had taken place in Rwanda even if the Hutu
majority was believed to be in control of political power; and 2) Rwandan leader(s) became to
rule with a small segment of the Hutu population to the exclusion of Tutsis and many other
Hutus. In the next section, I begin with an assessment of Iraqi ethnic policies between the First
Republic and the outbreak of the second Iraq War.

5.1 Iraq (1958–2003)
After General Abd al-Karim Qasim, a Sunni Arab, came to power in a 1958 coup d’état, Iraq
became a military dictatorship where the winning coalition was composed of a small group of
military elites who participated in a coup, including Qasim, Colonel Abd al-Salam Arif, and
twelve other young army officers who called themselves the Free Officers (Andrews & Ra'anan,

86

1969, p. 73). Sunnis played a dominant role in Qasim’s conspiracy. Of these fourteen members,
twelve were Sunnis, although Sunnis comprise about 30 percent of the Iraqi population, while
only two were Shiites, though about 60 percent of the Iraqi population is Shiite, with no Kurdish
officers (Andrews & Ra'anan, 1969, p. 71).
Even though the Iraqi government under Qasim was essentially a Sunni regime, Qasim
did not actively discriminate against the Shia and Kurd populations. During his early years in
office, Qasim had an integrative vision in which he represents all Iraqis, and diverse Muslim
communities work together for the Iraqi state (Tripp, 2007, p. 146).

14

Qasim created a three-

man Council of Sovereignty as a symbolic head of the state to represent Iraq’s three main ethnic
groups: Shia Muslims, Sunni Muslims and Kurds. His first cabinet contained a few of the Free
Officers including Arif as minister of the interior, but otherwise included civilians with diverse
ideological and ethnic backgrounds (Hurewitz, 1969, p. 155; Tripp, 2007, p. 147). In addition,
Qasim did not obey the structural imperative of small-coalition systems by spending
considerably more resources on public goods than other small-coalition leaders. He allocated oil
revenues to fund internal reform projects such as the construction of new roads, canals, schools,
and houses for poor people around Baghdad (Andrews & Ra'anan, 1969, p. 74; Tripp, 2007, p.
161). During his rule, for example, the number of students tripled and Baghdad and some other
cities were transformed into more modern and healthy places.
Although Qasim’s rule led to a number of positive changes that benefitted all members of
Iraqi society, he was unable to win over the support of members in his winning coalition. Some
of the Free Officers were disappointed by their share of profits and political role under the Qasim
regime. Moreover, members of the Iraqi officer corps were deeply divided between pan-Arab
14

This is partly explained by his family background. Qasim’s father was a Sunni Muslim while
his mother was a Kurdish descent and Shia Muslim.
87

nationalists and communists. Pan-Arab nationalists were supported by Qasim’s political rival
Arif (a Sunni Arab), and his military faction, while Qasim sought the support of the communists
(Andrews & Ra'anan, 1969, pp. 75-76). In October 1959, a group of disgruntled pan-Arab Free
Officers attempted an unsuccessful coup, in which the 22-year old Saddam Hussein participated.
This event was followed by an extensive purge of pan-Arab sympathizers in the armed forces as
well as large public demonstrations in support of Qasim. Despite Qasim’s repressive measures,
Arif’s supporters were not discouraged and started to build up their support base throughout the
country. They staged another military coup in February 1963. Fierce fighting continued around
the Ministry of Defense for two days until Qasim was forced to surrender to the rebel forces.
Qasim and his followers were executed after a show trial on February 9, and Colonel Arif
became the new president of Iraq.
The new government was also a military dictatorship with a small winning coalition
composed of military elites. However, unlike his predecessor, Arif was well aware of the
importance of patronage network in the armed forces. After removing military officers who
were suspected of opposing him, Arif consolidated his power in the military by utilizing his
personal and ethnic networks. He appointed his brother, Abd Al-Rahman Arif, to Chief of Staff,
and created the Republican Guard as an elite unit of the Iraqi army. The Commander of the
Republican Guard, Colonel Said Slaibi, was a relative of Arif, and most of the Guard members
were recruited from Sunni Muslim tribe of “Al-Jumayla” (Marashi & Salama, 2008, p. 97).
Better trained, armed and paid than the regular army, the main mission of this unit was to protect
the regime against future coup attempts (Tripp, 2007, p. 169).
Being confident about his control of the armed forces, Arif decided to start the gradual
withdrawal of the military from politics. In September 1965, he appointed a civilian, Abd al-

88

Rahman al-Bazzaz, as deputy prime minister to build a larger winning coalition participated by
civilians. Faced with the impending loss of power and privileges, several leading officers in the
military and the prime minster at that time launched a coup against Arif while he was attending
Arab summit in Morocco. The coup was quickly suppressed by the Republican Guard and, upon
his return, Arif appointed al-Bazzaz as new prime minster. As the first civilian prime minister
since the overthrow of the monarchy in 1958, al-Bazzaz held a liberal view that the leadership of
the Iraqi government could return to civilians and that the military should return to its primary
mission of national defense. A number of domestic reforms were implemented during his
premiership. A National Council of the Revolutionary Command (NCRC) was abolished and the
legislative power of NCRC was transferred to a newly created cabinet.

Al-Bazzaz also

encouraged private enterprise and constructive criticism against him, and promised to legalize
political parties and hold elections. (Tripp, 2007, p. 177). Especially, he promoted a policy of
ethnic integration with the Kurdish population. For example, his twelve-point agreement in June
1966 was the most extensive concessions to the Kurds, including an amnesty to the Kurd rebels,
constitutional recognition of the Kurdish identity, recognition of Kurdish as an official language,
and representation of Kurds in all branches of the executive (Ghareeb & Dougherty, 2004, p. lxii;
Tripp, 2007, p. 181).
Arif’s tenure came to an end when he was killed in a helicopter crash in April 1966. His
older brother, Abd Al-Rahman Arif, succeeded the presidency, and Al-Bazzaz was reappointed
as prime minister in the new cabinet. However, the officer corps remained dissatisfied with the
new government. They opposed al-Bazzaz’s proposed reduction in the defense budget (Tripp,
2007, p. 179) and inclusive policies toward the Kurds. Disappointed with their decreasing
privileges under the Arif government, a number of senior officers including the head of the

89

Republican Guard launched a bloodless coup against the government in July 1968. Arif was
exiled to Turkey while al-Bazzaz was arrested and accused of conspiracy against the state. One
of the coup leaders, General Ahmed Hassan al-Bakr, was appointed the new president of Iraq on
July 17.
Al-Bakr consolidated his power by appointing himself prime minister and his cousin,
Saddam Hussein, as deputy chairman of the Revolutionary Command Council (RCC), the second
most powerful position in Iraq.

Al-Bakr and Hussein were as quick as Arif to sense the

importance of ethnic networks in the armed forces, but their “exclusion” was far more
comprehensive than that of previous leaders. The Iraqi government led by Al-Bakr and Hussein
was a small-coalition system that favored a small segment of the Sunni minority, especially those
from their home region of Takrit as well as others from the Sunni lands of the north-west (Tripp,
2007, p. 191). The commanding positions in the army and the Republican Guard―including
Minister of Defense, Chief of Staff, and Commander of the Air Force―came to be occupied by
privileged officers from Takrit, and other members from the Sunni north-west staffed many key
positions in the state’s central administration and the security services (Marashi & Salama, 2008,
p. 144).
Throughout the 1970s, Hussein strengthened his power in the party and the state through
security services, eventually becoming de-facto leader of Iraq. On 16 July 1979, the aging AlBakr was forced to step down from his position, officially for health reasons, and within a few
hours the 42-year-old Saddam became Iraq’s new president. Hussein’s rule depended on ethnic
loyalties and exclusion. Like Al-Bakr, Hussein appointed clan members of the Sunni north-west
to key government positions, the tenure of which depends exclusively on their unconditional
loyalty to the leader (Balaghi, 2006, p. 73). The officer corps and intelligence networks were

90

also filled with members of his clan and immediate family for the purpose of preventing any
coup attempts. Consisting of only 500,000 out of a population of nearly 26 million (W=0.02)
(Tripp, 2007, p. 259), these coalition members of Saddam Hussein enjoyed privileged access to
valuable state resources and opportunities. Foods and resources were more widely distributed in
central Baghdad and Takrit regions (Balaghi, 2006, p. 98), and even the black market and
smuggled goods were used as private benefits to loyal supporters (Bueno de Mesquita et al.,
2003, p. 202). They relied upon Hussein for all of their privileges and even survival, and if
Hussein fell in a coup, they would lose everything. Charles Tripp (2007, p. 216) describes
Hussein’s coalition in the following way,

As an engine of power, accumulating resources, deploying patronage and maintaining
control over its inhabitants, it was centered on the restrictive circles of Saddam Hussein’s
associates, linked to him either through bonds of kinship and regional background or
through a history of personal trust. These men formed the inner circle of the Iraqi regime,
having been put to the test on numerous occasions during the preceding fifteen years,
when they could have sided with other clansmen, other ideological tendencies in the party
or with restless and opportunistic military officers. Instead, they had followed Saddam
Hussein. This made them the Ahl al-Thiqa [“the trusted”] in whom … Saddam Hussein
could have confidence. His cause had become theirs and they were so closely identified
with him that their political fate would be linked to his.

The exclusionary political system of Hussein led to suffering of vast numbers of those
excluded, exacerbating grievances among Shia Muslims and Kurds, who together constitute

91

about 70 percent of the Iraqi population. Discriminated in their access to private goods and
excluded from state power, these ethnic groups occasionally took up arms against the regime of
Saddam Hussein. However, they made little impact on Hussein’s leadership. For instance, in
March 1991, Shia Muslims in southern Iraq launched an armed struggle against the Sunnidominated regime. However, support for the rebellion was confined to southern cities and towns,
and the Republican Guard units crushed the rebel forces within a couple of weeks, killing some
30,000 Shiites and displacing more than 70,000 people (Balaghi, 2006, p. 90; Tripp, 2007, p.
246). Soon thereafter, the Kurdish rebels in the north initiated conflict against Saddam. In 1988,
Saddam used chemical weapons against the Kurds, killing over 100,000 Kurd civilians. This
time again, the rebellion was ruthlessly repressed by the Republican Guard, and massive
numbers of Kurds fled Iraq. In 1998, the U.S. Congress passed the Iraq Liberation Act and
authorized President Bill Clinton to spend up to $98 million to assist qualified Iraqi opposition
groups, mostly Shia and Kurdish ("Iraq Liberation Act," 1998). However, this still could not
weaken Saddam’s hold on power. There were also a number of assassination attempts against
Hussein made by disgruntled Shia groups (Tripp, 2007, p. 238), although they were ineffective
and poorly organized in comparison to military coups against previous regimes.

In sum,

combined with ethnic solidarity, the loyalty of the Republican Guard and intelligence service
controlled by Saddam’s inner circle provided little opportunity for those who planned to
overthrow the Hussein regime.

After staying in power for 24 years, Hussein was finally

overthrown not by a coup or civil war but by a full-scale invasion of the country by American
forces in 2003.

5.2 Burundi (1962–1993)

92

The two major ethnic groups in Burundi are the Hutu and the Tutsi. Although the Tutsi
constituted only about 14 percent of Burundi’s population and the Hutu nearly 85 percent, the
Tutsi maintained the privileged status under the Belgian colonial regime. The Belgian authority
replaced all Hutu chiefdoms with Tutsis, and “Tutsi was evolving to become synonymous with
power and high status … and Hutu to become synonymous with subordination and exclusion”
(Eller, 1999, p. 213). When Burundi gained independence in 1962, it became a constitutional
monarchy under the Tutsi King Mwambutsa IV.

After independence, however, the Hutu

Revolution (1959-1961) in neighboring Rwanda sparked a new phase of conflict between the
two ethnic groups in Burundi.

In Rwanda, Tutsis were subject to mass murder by Hutu

politicians after political power was transferred from the Tutsi minority to the Hutu majority.
King Mwambutsa feared that Burundian Tutsis would meet the same fate.

Therefore he

appointed Tutsis in key military and police positions, while at the same time making concessions
to the Hutu by co-opting emergent Hutu elites into the bureaucracy and by allowing them to gain
status and prestige (Weinstein & Schrire, 1976, p. 14). Tutsis perceived this situation as a threat
to their survival and to “the privileges and power that Tutsi had been able to accumulate and
establish for themselves, as a result of preferential treatment under Belgian … colonial rule”
(Weinstein & Schrire, 1976, p. 15). The first parliamentary elections were held in May 1965,
with the Hutu winning an overwhelming victory.

However, King Mwambutsa refused to

recognize the result of the election, and instead appointed a Tutsi as Prime Minster. In response,
a group of Hutu officers attempted an unsuccessful revolt against the King, which was followed
by the purge of a large number of Hutu intellectuals and politicians by Tutsi hardliners. In fear
of losing more privileges and power after the 1965 events, the Tutsi-controlled army led by
Captain Michel Micombero overthrew the monarchy in November 1966 and declared a republic

93

with Micombero as its first president.
President Micombero―a young Tutsi born in Bururi province―abolished parliament and
ruled the country through the seventeen-member National Revolutionary Council (NRC),
effectively creating a small-coalition system.

Consistent with the exclusion model, the

Micombero’s dictatorial regime was based on the political exclusion of Hutus. He eliminated
Hutus from the military and high positions in the administration, with a number of Hutu officers
being executed for treason. The Hutus responded to discrimination by launching a rebellion
against Tutsi rule in April 1972, although it was brutally suppressed by the Tutsi-dominated
army, resulting in the deaths of about 200,000 Hutus. Approximately a quarter of Hutu students
disappeared during 1972 and 1973. And, almost all Hutu officials and administrators either fled
or were killed during this period (Weinstein, 1977, p. 53).
With the Hutu under control, fissures quickly opened up inside the Tutsi elite. The most
important was a regional rivalry between Tutsis from the southern province of Bururi, one of
Burundi’s 15 provinces, and those from other provinces.

Consistent with computational

simulation in Chapter 3 (Section 3.3.1), the Burundian regime evolved to a system with a smaller
winning coalition of Bururi Tutsis who comprise well under 10 percent of Burundi’s population
(W<0.1). Although non-Bururi Tutsis benefited from power more than Hutus, they were largely
excluded from important positions―including the presidency, key ministers, and army
commanders―as well as from education opportunities (Nkurunziza & Ngaruko, 2002, p. 19).
Tutsis from central provinces were also suspected of plotting a coup against Micombero, leading
to the arrest and execution of several non-Bururi Tutsi politicians and army officers in 1971
(Weinstein, 1977, p. 53; Weinstein & Schrire, 1976, pp. 26-27). Micombero had been in office
for 10 years until he was overthrown by Deputy Chief of Staff Jean-Baptiste Bagaza, a distant

94

relative, in a bloodless coup of November 1976. It was reported that Micombero was not able to
control events during his later years in office due to chronic alcoholism and paranoia (Bartrop,
2012, p. 210; Weinstein, 1977, p. 55), and the coup was carried out to restore Bururi Tutsis’
control over the government.
The dominance of Bururi Tutsis had been more secure under President Bagaza, who
stayed in power until 1987. Its main manifestation was through the economic benefits provided
to Bururi Tutsis by the central government. Illustrating the concentration of economic benefits
in Bururi province, Ngaruko and Nkurunziza (2005, pp. 42-43) note that Bururi has received
significantly more public services―in terms of overall school enrollment, population/hospital
ratio, teachers per classroom, illiteracy rate, and per capita income―than other provinces while
paying significantly less taxes. Hutus residing in Bururi province were excluded from these
benefits except for positive external effects of public infrastructure (Ngaruko & Nkurunziza,
2005, p. 43).
The Bururi Tutsi’s hold on power, however, began to erode in the late 1980s. Under
pressure from the international donor community, President Pierre Buyoya, a Bururi Tutsi who
seized power through a bloodless coup in September 1987, decided to end ethnic exclusion that
had been in effect since the 1960s. Buyoya appointed Hutus to key political posts, legalized
multiparty politics, and agreed to hold presidential and legislative elections in June 1993. Both
elections were won by Hutu-dominated parties, and Melchior Ndadaye became Burundi’s first
Hutu president in July. The exclusion model predicts that a leader who attempts to end ethnic
exclusion in small coalition systems would be removed from office in an irregular (or violent)
manner by existing (and now less loyal) coalition members who try to recover their share of
private goods. This is similar to developments in Burundi following the 1993 elections. Bururi

95

Tutsis viewed the advent of Hutu domination as a peril to their survival and to every privilege
they had enjoyed under previous governments. This suspicion was confirmed when President
Ndadaye replaced large numbers of Tutsi officeholders with Hutus after his election (Ngaruko &
Nkurunziza, 2005, p. 45). In the words of Ngaruko and Nkurunziza (2005, p. 44),

Such changes were particularly threatening because the system had been enforced for so
many years with so much determination; there was much to lose. Indeed, Burundi’s
history shows that when the change of leadership does not threaten the interests of the
dominating group, it is peaceful. Between 1966 and 1987, for instance, three Tutsi
military presidents from Rutovu in Bururi province succeeded each other through
bloodless palace coups. This contrasts sharply with the bloodbath that resulted from the
victory of a non-Bururi Hutu in the June 1993 democratic elections.

The ruling Bururi Tutsi had little intention of giving up their power without a fight. As
predicted by the theory, the Bururi Tutsi-dominated army assassinated Ndadaye on October 1993.
This event was followed by violent reprisals against Tutsi civilians throughout the country,
resulting in the deaths of about 50,000 civilians, mostly Tutsis, in the first week following the
assassination. Buyoya came back to power through a military coup in July 1996. But, a Burundi
had fallen into a civil war between the Tutsi-dominated army and Hutu rebel groups. Between
1993 and 2005, approximately 200,000 people were killed in ethnic violence. These events are
clear examples of violent leadership turnover in the wake of ethnic inclusion preceded by a long
period of exclusive rule. For Bururi Tutsis, it was better to fight now, while they still have some
advantages, than to risk waiting until their hold on power may be significantly weak. For those

96

excluded, it was better to start a war because now, for the first time, they had a good chance to
come to power.

5.3 Rwanda (1962–1994)
The population of Rwanda consists of three ethnic groups: the Hutu (85%), the Tutsi (14%), and
the Twa (1%). In pre-colonial times, Hutus and Tutsis spoke the same language, had the same
religion, lived in the same villages, and often intermarried. However, it was under Belgian
colonial rule (1914-1962) that a divide between the two ethnic groups was solidified. The Tutsis
enjoyed a privileged status in the Belgian administration, which ruled Rwanda indirectly through
the Tutsi minority at the expense of the majority Hutu. The Belgian colonizers believed that
Tutsis were genetically superior to Hutus who were perceived as short, stocky, and wide-nosed.
On the other hand, Tutsis were perceived as tall, elegant, and thin-nosed with “superior”
Caucasian ancestry (Straus, 2006, pp. 20-21). In the 1930s, ethnic identity cards were issued to
all Rwandans, so that a Hutu could not become a Tutsi. Such a discriminatory treatment created
deep resentments and grievances among the Hutus. The result of this tension was the 1959 Hutu
Revolution that led to the overthrow of the Tutsi monarchy and to the establishment of Hutudominated government. On 1 July 1962, Rwanda achieved independence from Belgium and
Grégoire Kayibanda, a Hutu, became the first President of the Rwandan Republic.
The young republic was a single party, small-coalition system where the ruling
Parmehutu party occupied all the seats in the National Assembly. Like most other leaders in
small-coalition systems, President Kayibanda was re-elected in 1965 with 99 percent of the vote
in a rigged election where he was the sole candidate. The Kayibanda regime was characterized
by ethnic exclusion and widespread violence against the Tutsis. Under Kayibanda, the majority

97

Hutu established its firm control over all aspects of the Rwanda state: the central government, the
single party, and the military. On the other hand, the minority Tutsi was increasingly excluded
from any meaningful participation in politics. Between 1962 and 1964, there were a series of
anti-Tutsi pogroms. In February and March 1962, a group of Tutsi exiles, who fled Rwanda for
neighboring countries during the Hutu Revolution, began to attack Rwanda, killing a handful of
Hutu policemen and government soldiers.

15

In response, Kayibanda chose to target the rebel’s

base of support in the Tutsi population (Straus, 2006, pp. 186-187). During this period, 10,000
to 15,000 Tutsis were killed, and many others fled Rwanda.

Another round of anti-Tutsi

violence followed in 1973 when several Tutsi government officials were executed and large
numbers of Tutsi students were expelled from universities and secondary schools.
With the Tutsi eliminated from military contention, the Hutu began to be divided along
regional lines. The Hutu who make up 85 percent of the population was clearly an oversized
winning coalition. Since it was impossible to provide private goods to such a large winning
coalition, Kayibanda favored southern Hutus in the distribution of posts in the civilian
government (Jones, 2001, p. 25; Newbury, 1992, p. 197; Straus, 2006, p. 190), which created
resentment among northern Hutus. Despite this regional favoritism, there was no “systemic”
exclusion of northern Hutus within the regime. Though there was unrest among northern Hutus
who dominated the Rwanda army, Kayibanda did not purge northerners from the military (Enloe,
1980, p. 46; Newbury, 1992, pp. 197-198; Straus, 2006, p. 191). It was this northern component
of the officer corps that ultimately toppled Kayibanda from power. On 5 July 1973, a group of
military officers, mostly northern Hutus, led by General Juvénal Habyarimana seized power in a
coup d'état. The new authority executed key figures from the previous regime and placed

15

The Hutus called the Tutsi rebels “inyenzi” or “cockroaches” because they attacked at night.
98

Kayibanda under house arrest until he died in 1976 (Straus, 2006, p. 191).
After assuming power, President Habyarimana established a single party dictatorship
under the exclusive rule of the MRND (National Republican Movement for Democracy and
Development) party. As predicted by the theory, ethnic exclusion was reinforced under the
small-coalition system of Habyarimana. A system of ethnic quota was introduced to limit Tutsis’
access to post-primary education and employment in government sector. The winning coalition
of Habyarimana was largely composed of Northern Hutus, especially those from the
northwestern province of Rwanda (Straus, 2006, p. 23). Using discriminatory appointments to
public office, excluding not only Tutsis but also Hutus from other provinces, northwestern Hutus
have monopolized control over key positions in the government, in the army, and in state-owned
corporations (Straus, 2006, p. 23). At the elite level, the inner circle of the Habyarimana regime
was dominated by members of his clan and immediate family, a group known as akazu or “little
house” (Jones, 2001, pp. 26-27). For instance, the Presidential Guard, an elite unit in the army,
was filled with clan members loyal to Habyarimana in order to protect his regime against future
coup attempts.
That said, the evolution of ethnic exclusion in Rwanda diminished the significance of
overarching ethnic categories, such as the Hutu and Tutsi, and demanded more narrowly defined
identity. If one was a Hutu, it made a critical difference whether he was a southern Hutu or a
northwestern Hutu. On the ethnic categories in Rwanda, Bruce Jones (2001, p. 26) states,

Whereas the standard picture of the Rwandan regime depicted it as a Hutu regime, this
failed to capture the reality of pre-genocide Rwandan politics. Even though the vast
majority of those who held power in Rwanda before the genocide were Hutu, they were

99

Hutu of particular clans―especially from the Bushiru region. As in pre-colonial times,
the clan was a powerful factor in Rwandan political life. The Habyarimana regime was in
fact a clan-based northern Hutu regime that was discriminatory against Hutus from
southern Rwanda as against Tutsis.

For this reason, I argue that the “ancient hatred” model is not appropriate for explaining ethnic
politics in pre-genocidal Rwanda. Rwandan politics was not simply an ethnic contest between
the two ethnic groups, but an ethno-regional rivalry in which both southern Hutus and Tutsis
were excluded from state power by northwestern Hutus to consolidate the latter’s privilege
within the leader’s coalition. Furthermore, the ancient hatred model cannot capture the ways in
which Rwandan leaders changed the meaning and scope of ethnic categories in order to secure
their political survival.
A major change in Rwanda began with end of the cold war in the early 1990s. Under
pressure from Western aid donors, President Habyarimana announced in July 1990 his intention
to end the single party system. Immediately thereafter, Habyarimana and his party began to be
challenged by various Hutu opposition parties from southern Rwanda. Another major change
was the invasion of the country by the Rwandan Patriotic Front (RPF) in October 1990. The
RPF rebels were primarily descendants of Tutsi exiles who had fled Rwanda in the 1960s. The
size of the winning coalition expanded in April 1992 when Habyarimana agreed to form a
coalition government with opposition parties. A coalition government then reached a peace
agreement with the RPF rebels in August 1993. Known as the Arusha Accord, the agreement
aimed to establish a broad-based transitional government with significant Tutsi representation;
the RPF was given a third of the cabinet positions, 40 percent of the regular army, and 50 percent

100

of the officer corps.

16

Rwanda’s ruling elites―Habyarimana’s inner circle and its allies from the
northwest―perceived these changes as a threat to their hold on power as well as to the privileges
they had enjoyed under Habyarimana. Theoretically speaking, these changes meant a rapid
expansion of the winning coalition that will diminish their share of private benefits significantly.
The more they will lose from the new system, the more they are willing to fight for the current
system (Bueno de Mesquita et al., 2003, p. 370). Consistent with the theory’s prediction, the
ruling elites in Rwanda adopted hardline stances and chose to keep their power through irregular
means, ultimately laying the groundwork for the 1994 genocide. They created and trained a
youth militia known as interahamwe, spread anti-Tutsi propaganda from a radio station, and
developed “death lists” of moderate Hutus. When Habyarimana was killed in April 1994, the
hardliners, who once dominated key positions in the Habyarimana government, were
coordinating the genocide that claimed at least 500,000 lives.

5.4 Chapter Conclusion
The preceding three cases―Iraq, Burundi, and Rwanda―support my claim that leaders in smallcoalition systems tend to strengthen exclusionary politics against ethnic “others” over time. In
each case, major ethnic rivals―the Kurds and Shia Muslims in Iraq, Hutus in Burundi, and
Tutsis in Rwanda―were excluded from participation in central government. Then, another
round of power struggle begins among former ethnic allies. Eventually, the regime becomes
dependent on the active support of a small group of political elites who share the same ethnoregional backgrounds with the leader. Moreover, the cases of Burundi and Rwanda also show
16

The full text of the Arusha Accords (1993) can be downloaded from
http://www.incore.ulst.ac.uk/services/cds/agreements/pdf/rwan1.pdf.
101

that political survival can be threatened by an attempt to end ethnic exclusion in small-coalition
systems. The outcomes of such “mistaken” ethnic policy could be especially dangerous (not
only to the leader, but also to the country itself) when the abrogation of ethnic exclusion was
preceded by a long period of exclusive rule. However, as we will see in the next chapter, these
findings should be interpreted with great caution. Especially, they should not be interpreted as a
recommendation for ethnic exclusion as a way to ensure domestic stability and regime survival.
Overall, the behavior of leaders and their survival (or failure) in the above cases provide
strong support for my theory. Although there is plenty of room for further case studies and data
collection in future research, I believe that the results presented in this and previous chapters
(Chapters 3 and 4) are of high theoretical importance as well as of policy relevance. In the next
chapter, I will summarize the key findings of this dissertation and discuss their implications for
the peacemaking efforts of international community.

102

CHAPTER 6
CONCLUSION

This dissertation started with a simple question: why do some leaders deliberately employ ethnic
exclusion even though it increases the risk of ethnic conflict? The political exclusion of ethnic
groups and the resulting “grievances” have been recognized as one of the most important causes
of civil war (Cederman et al., 2011; Cederman et al., 2010; Gurr, 1993, 2000; Horowitz, 2000).
However, leaders’ incentives for ethnic exclusion have been largely forgotten or ignored in the
existing literature on ethnic conflicts and in many attempts to implement policies designed to
resolve ethnic grievances. The central purpose of this dissertation is to emphasize the crucial
role of leaders in explaining ethnic exclusion and violence. To this end, I introduced a theory of
ethnic exclusion by modifying the selectorate theory of Bueno de Mesquita et al. (2003).
Contrary to common belief, my theory suggests that ethnic exclusion is good politics for leaders
with small winning coalitions because it enhances their prospects for remaining in office, despite
its positive impact on the risk of ethnic conflict. In the remainder of this concluding chapter, I
will summarize the theoretical and empirical findings of the previous chapters and discuss their
broader theoretical and policy implications.

6.1 Summation of the Argument
My theory builds upon the key assumptions and arguments of the selectorate theory (Bueno de
Mesquita et al., 2003); namely, (1) all leaders have political survival as their primary objective;
(2) leaders maintain the support of their coalition by providing a mix of private and public goods
to coalition members; and (3) small-coalition leaders are expected to survive longer than largecoalition leaders due to the high loyalty to the incumbent in small-coalition systems. Yet, in
103

examining how notions of ethnicity affect leader survival, my analysis extends and differs from
the selectorate theory in four notable respects.
First, I assume that every leader chooses to form her coalition on the basis of coalition
identity that she thinks best ensures her political survival. Second, every coalition identity is
associated with a different cost of changing identity (C). As a general rule, C is higher when
coalition identity is based on ethnic attributes―such as skin color, religion and language―than
when it is based on non-ethnic ones―such as party affiliation and ideological preferences
(Chandra, 2006; Fearon, 1999). Third, individuals outside of the leader’s coalition may attempt
to enter the coalition by changing their identity. But, they do so if and only if the cost of
chaining their identity is less than the expected benefit of private goods received by each
coalition member (Caselli & Coleman, 2006; Fearon, 1999). Fourth, the size of the winning
coalition is not necessarily equal to the “actual” number of supporters in the incumbent’s
coalition. The coalition is (1) “minimum-sized” if the actual size of the incumbent’s coalition
corresponds to the size of the minimum winning coalition; (2) “undersized” if the actual coalition
size is small in large-W systems; or (3) “oversized” if the actual size is large in small-W systems.
An extension of the selectorate theory based on the above assumptions provides novel
implications for the relationship between ethnic exclusion and leader survival. In large-W
systems, the demand for coalition membership is low due to the low amount of private goods
provided to each coalition member. Since leaders in this system have to include large numbers
of supporters in their coalition, it is rational for them to form an inclusive coalition based on nonethnic identity; otherwise they will generate undersized coalition with a high risk of removal
from office. In small-W systems, the demand for coalition membership is high due to the high
amount of private goods allocated to each coalition member. Since survival in office depends

104

upon the provision of private benefits to key supporters, leaders in this system have a strong
incentive to limit the size of their coalition by forming an exclusive coalition based on ethnic
identity; otherwise they will face an oversized coalition with a greater risk of losing office in an
irregular manner.

This makes ethnic exclusion good politics for leaders in small-coalition

systems.
Four hypotheses were deduced from my theory for explaining ethnic exclusion and leader
survival. H1 maintains that, in small-coalition systems, leaders who employ ethnic exclusion are
more likely to survive longer in office than those who do not employ it. This is because ethnic
exclusion prevents oversized coalition in small-W systems. H2 maintains that, if small-coalition
leaders do not pursue an exclusive ethnic policy, they are more likely to be removed from office
in an irregular manner. In this case, dissatisfied members of the leader’s coalition may attempt
to displace the incumbent through coup d’état or other violent means in order to advance their
interests in a smaller coalition. H3 suggests that, during civil war, leaders who promote ethnic
exclusion are less likely to lose power in small-W systems because the loyalty of the military to
the incumbent is especially high in such a system. H4 predicts that, if ethnic exclusion really is
good politics in small-W systems, there should be higher levels of ethnic exclusion in small-W
systems than in large-W systems.
These hypotheses were then empirically tested and supported by Cox’s proportional
hazard regressions using data on the tenures of 982 leaders from 1946 to 2004. My results show
that in small-coalition systems: (1) the hazard of deposition for leaders who implement a strong
exclusion policy is about 80% lower than that of leaders who do not promote exclusion; (2) the
risk of irregular turnover among leaders who employ ethnic exclusion is only about 1.3% of the
risk for those who do not employ such a policy; and (3) the risk of irregular removal from office

105

virtually disappears even in times of civil war if a leader employs a strong exclusion policy.
Case studies of Iraq, Burundi, and Rwanda further corroborate causal claims made by the
exclusion theory. In all three cases, major ethnic groups had been excluded from participation in
the leader’s coalition until a small group of ethnic elites monopolizes key positions in the army
and government. The cases of Burundi and Rwanda also show how the occurrence of oversized
coalition in small-W systems leads to large-scale violence when ethnic inclusion is attempted
after years of exclusive rule. Overall, my findings lead to an “unwelcome” conclusion: ethnic
exclusion is good politics in small-coalition systems even if it could increase the risk of ethnic
conflict.

6.2 Implications and Conclusion
The theory of ethnic exclusion bridges the gap between the role of political institutions in
leadership survival and the politics of ethnicity. In common with the selectorate theory, I believe
that institutions for selecting leaders play a central role in influencing leadership survival;
namely, leaders in small-W systems have a greater incumbency advantage than those in large-W
systems. I also believe, however, that the type of coalition identity and the resulting ethnic
policy play a major role in shaping leaders’ survival and manner of exit. More importantly, I
believe that such ethnic variables interact with the size of the winning coalition; that is, when
leaders fail to promote ethnic exclusion in small-W systems, oversized coalition is likely, and
when they promote ethnic exclusion in large-W systems, undersized coalition is likely. While
the impact of selection institutions upon leadership survival is filtered through leaders’ stance on
ethnicity, I argue, leaders often jeopardize their own survival by making errors in their choice of
ethnic policy. The marriage between political institutions and ethnic politics within a coherent
theoretical framework suggests a number of theoretical and policy implications.
106

6.2.1 Theoretical Implications
First, my theory can be expanded to research on conflict onset and escalation.

In recent

statistical research on ethnic conflicts, many scholars focused on the opportunity for rebellion as
a primary determinant of conflict onset (Collier & Hoeffler, 2004; Fearon & Laitin, 2003;
Melander, 2009; Toft, 2003; Weidmann, 2009). These studies made a notable contribution to
identifying and understanding the economic or geographic conditions that make insurgency a
more feasible option for excluded ethnic groups. However, they do not explicitly explain what
motivates ethnic groups to initiate conflict in the first place. A theory of ethnic exclusion can
provide an explanation for the underlying causes of ethnic grievances. Leaders in small-W
systems utilize ethnic exclusion as a way to promote their personal political success. Doing so
generates grievances among excluded ethnic groups by increasing political and economic
inequalities between them and members of the incumbent’s coalition. The smaller the size of the
winning coalition, the higher the degree of grievances it will generate, since excluded groups are
expected to have fewer political goods and higher levels of taxation in small-W systems.
After all, motivation matters, but opportunity also matters. While the exclusion theory
can tell us when excluded ethnic groups hold high levels of grievance, it is the opportunity
mechanism that tells us how effectively these groups will mobilize manpower to initiate a
conflict. For example, previous studies find that a geographic concentration facilitates group
mobilization by alleviating collective action problems (Laitin, 2004; Toft, 2003; Weidmann,
2009). They argue that concentrated ethic groups will be more likely to mobilize against the
state when they are motivated to do so. On the other hand, geographically dispersed groups will
be less effective in coordinating their actions and thus in resolving collective action problems.

107

Therefore, I expect that excluded and highly motivated ethnic groups can engage in different
forms of violence: while concentrated groups are more likely to initiate full-fledged civil wars,
dispersed groups are more likely to carry on their struggle in the form of low-intensity violence
such as scattered riots, terrorism, and violent demonstrations.
Second, although my theory deals mostly with the incumbent leader, it can also be used
to understand the behavior of insurgent leader. Especially, when lootable resources―such as
gold, diamonds and drugs―are available to the rebel group, the insurgent leader can distribute
them as private rewards to a small group of essential supporters. In this case, the internal politics
of rebel organizations should reflect, in part, the basic logic of the exclusion theory. Consider
Foday Sankoh who was the leader of the Revolutionary United Front (RUF) of Sierra Leone
during its 11-year-long civil war between 1991 and 2002. Sankoh used access to diamond
resources as a private benefit to his loyal supporters at the expense of local populations. Sankoh
is known particularly for his brutality against his own countrymen, including loot, murder, child
conscription, rape, amputations of limbs, and forced labor in the diamond fields (Dobbins, 2005,
p. 133).

Rebel leaders of this kind resemble small-coalition leaders practicing ethnic

discrimination; in this sense, my theory may provide explanations for why some rebel leaders are
more likely to survive a long-running war when they act harshly against local populations rather
than seeking their support.
A third theoretical implication has to do with the role leaders in shaping ethnic cleavages.
In the literature on ethnic identity and violence, ethnicity is often assumed to be fixed by history
and deeply held values of an ethnic group (Isaacs, 1989; Smith, 1986; Van den Berghe, 1981).
According to this “primordial” view, ethnic violence results from antagonisms and hostilities that
are unchaining characteristics of rival ethnic groups. However, a history of ethnic conflict is rife

108

with examples where members of specific ethnic groups target their co-ethnic rivals. During the
Rwandan genocide, some 10,000-30,000 moderate Hutus were killed by their Hutu neighbors
(Prunier, 1995, p. 265). President Nimeiri broke the Addis Ababa Agreement and plunged Sudan
back into civil war in 1983 for fear of his own radical Muslims in the north rather than rebels in
the south (Fearon & Laitin, 2000, pp. 866-867). Similarly, a study of Northern Ireland conflict
by Kennedy-Pipe (1997, pp. 53, 63) shows that when moderate Catholics pursued the peace
process with Protestants and British authority, radical Catholics in the Provisional Irish
Republican Army (IRA) attempted to spoil this process by provoking British troops to shoot
ordinary citizens in the hope that they can garner support from moderate Catholics, and thus
strengthen their hold on power and social privileges. These examples suggest that the boundary
of ethnic identity is not fixed, but rather is multi-layered and changing over time.
The theory of ethnic exclusion is in line with the “instrumentalist” view that ethnicity is
socially constructed by elites for instrumental purposes (Esman, 1994, pp. 10-11). In addition,
following Chandra and Wilkinson (2008), I draw a distinction between “nominal” and “activated”
ethnic identities, where the former “are those for which we possess the attributes of membership
while [the latter] are that subset of our nominal categories in which we profess membership or
are assigned membership by others” (p. 517). Rather than simply arguing that ethnic identities
are socially constructed, my theory specifies when and under what conditions a particular ethnic
category is “activated” by political leaders. Specifically, the theory suggests that ethnicity is
activated (and thus become politically salient) when leaders in small-coalition systems form an
exclusive coalition along ethnic lines. When a small-coalition leader is a member of a large
ethnic group A, for instance, she is likely to activate a smaller ethnic identity B, which is a subset
of A, by excluding members not belonging to B from her coalition. In all three cases examined

109

in Chapter 5, leaders activated sub-ethnic cleavages―the Sunni north-west in Iraq, the Bururi
Tutsi in Burundi, and the northwestern Hutu in Rwanda―within the nominal ethnic identity.

These cleavages, not the nominal ones, correspond to ethnic groups that are actually engaged in
the politics of exclusion that is likely to generate ethnic grievances.

6.2.2 Policy Implications
Scholars and policy makers have supported the idea of “inclusive coalition” as a way to manage
conflict in ethnically divided societies. In Politics in West Africa, for instance, Lewis (1965)
proposed a coalition government in which all the major parties participate in decision-making.
In such a system, he writes, “it is necessary to get right away from the idea that somebody is to
prevail over somebody else; from politics as a zero-sum game. Words like ‘winning’ and ‘losing’
have to be banished from the political vocabulary of a plural society” (Lewis, 1965, p. 67). In
Democracy in Plural Societies, Lijphart (1977) introduced another model of inclusive coalition,
called consociational democracy.

The key characteristic of consociationalism is that “the

political leaders of all significant segments of the plural society cooperate in a grand coalition to
govern the country” (Lijphart, 1977, p. 25). Other elements of consociationalism include (1) the
mutual veto that protects minority interests; (2) proportionality in the allocation of government
jobs and political goods; and (3) segmental autonomy whereby minorities rule themselves
(Lijphart, 1977, p. 25).
More recently, the World Bank (2011) supports the idea of “inclusive-enough coalition”
as one of the basic principles for violence prevention and peaceful resolution of conflicts.
Defined as the governing coalition encompassing broader segments of society―including local
governments, private businesses, civil society movements, and opposition parties―this strategy

110

aims at restoring confidence needed to create continued momentum for political reform (World
Bank., 2011, p. xvii). With regard to the characteristics of inclusive-enough coalition, the report
(World Bank., 2011, pp. 12-13) says,

Coalitions are “inclusive-enough” when they include the parties necessary for
implementing the initial stages of confidence-building and institutional transformation.
They need not be “all-inclusive.” Inclusive-enough coalitions work in two ways: (1) at a
broad level, by building national support for change and bringing in the relevant
stakeholders, through collaboration between the government and other sectors of
society—as well as with regional neighbors, donors, or investors, and (2) at a local level,
by promoting outreach to community leaders to identify priorities and deliver programs.

Forming an inclusive coalition may be a good policy for the country. Indeed, previous
studies provide evidence that the presence of inclusive coalition leads to democratic transition
(Linder & Bachtiger, 2005), democratic consolidation (Reynolds, 1999), and conflict resolution
in Africa (Sisk & Reynolds, 1998). The exclusion theory, however, suggests that such a policy is
not necessarily good politics, at least for the leader. The promotion of inclusive coalition in
small-W systems gives rise to oversized coalition, making the leader vulnerable to a violent
turnover of power. Especially, when members of one ethnic group have enjoyed exclusive
privileges for a long period of time before this policy is implemented, they are unlikely to
relinquish their privileges without a bloody fight. This was the pattern of behavior that led to the
Rwandan genocide and Burundi civil war in the early 1990s. The above recommendations for
inclusive coalition have not taken an adequate account of this dangerous trade-off that may arise

111

in the course of policy implementation.
What can be learned from the research presented in this dissertation, and what policies
should be considered in the future? The theory of ethnic exclusion suggests that there is no easy
way to simultaneously promote ethnic inclusion and leaders’ survival in small-coalition systems.
To be sure, a successful transition from an exclusive toward an inclusive coalition will reduce the
risk of ethnic conflict. However, this transition should be accompanied by institutional reforms
to increase the size of the minimum winning coalition. My theory makes a clear distinction
between the institutional size of the winning coalition (W), and the actual size of the incumbent’s
coalition. To the extent that the incumbent’s coalition is larger than W, efforts toward an
inclusive coalition generate violent outcomes in small-coalition systems. Hence, only after
leaders have undertaken significant institutional reforms toward a large W can they safely form
an ethnically inclusive coalition to escape the vicious cycle of exclusion and violence.
Secondly, intervention by the international community through organizations such as the
United Nations and African Union can facilitate a peaceful transition to a larger and more
inclusive coalition by taking defensive measures to protect the incumbent during a transition
period.

The expansion of coalition in small-W systems decreases the welfare of existing

coalition members. My empirical findings show that, faced with the erosion of welfare, they
could choose to remove their current leader by force. International actors can reduce this risk by
providing an effective protection against any attempts at unconstitutional changes of government.
The recent approval of tougher anti-coup measures by the African Union is therefore a positive
development (Voice of America, 2010). Lastly, efforts must be made to reassure hardliners in
the incumbent’s coalition that the transition toward an inclusive coalition is not aimed at
excluding them, but rather at promoting a stable system that provides all citizens with peace and

112

prosperity.
To conclude, policy makers must be acutely aware that the formation of inclusive
coalition could become another source of violence and instability in small-W systems, especially
when such a measure is implemented after years of authoritarian rule based on ethnic exclusion.
Those who lack a clear understanding of this grim reality are likely to find that their efforts to
build an inclusive coalition will end in violence.

113

BIBLIOGRAPHY

114

BIBLIOGRAPHY

Alesina, A., Özler, S., Roubini, N., & Swagel, P. (1996). Political Instability and Economic
Growth. Journal of Economic Growth, 1(2), 189-211.
Alford, J. R., Funk, C. L., & Hibbing, J. R. (2005). Are Political Orientations Genetically
Transmitted? American Political Science Review, 99(02), 153-167.
Alkemade, F. (2004). Evolutionary Agent-Based Economics. (Ph.D.), Eindhoven University of
Technology, Eindhoven, Netherlands.
Allen, C. (1968). Sierra Leone Politics since Independence. African Affairs, 67(269), 305-329.
Andrews, W. G., & Ra'anan, U. (1969). The Politics of the Coup D'état: Five Case Studies. New
York,: Van Nostrand Reinhold Co.
Balaghi, S. (2006). Saddam Hussein : A Biography. Westport, Conn.: Greenwood Press.
Bartrop, P. R. (2012). A Biographical Encyclopedia of Contemporary Genocide. Santa Barbara,
Calif.: ABC-CLIO.
Bates, R. H. (1983). Modernization, Ethnic Competition, and the Rationality of Politics in
Contemporary Africa. In D. S. Rothchild & V. A. Olorunsola (Eds.), State versus ethnic
claims : African policy dilemmas (pp. x, 356 p.). Boulder, Colo.: Westview Press.
Bhavnani, R., & Choi, H. J. (2012). Modeling Civil Violence in Afghanistan: Ethnic Geography,
Control, and Collaboration. Complexity, 17(6), 42-51.
Birnir, J., & Gurr, T. R. (2009). Minorities at Risk Dataset. Retrieved from:
http://www.cidcm.umd.edu/mar/
Box-Steffensmeier, J. M., & Jones, B. S. (2004). Event History Modeling: A Guide for Social
Scientists. Cambridge ; New York: Cambridge University Press.
Bueno De Mesquita, B., & Smith, A. (2010). Leader Survival, Revolutions, and the Nature of
Government Finance. American Journal of Political Science, 54(4), 936-950.
Bueno de Mesquita, B., Smith, A., Siverson, R. M., & Morrow, J. D. (2003). The Logic of
Political Survival. Cambridge, Mass. ; London: MIT Press.

115

Caselli, F., & Coleman, W. J. (2006). On the Theory of Ethnic Conflict CEP discussion paper no
732 Retrieved from http://cep.lse.ac.uk/pubs/download/dp0732.pdf
Cederman, L.-E., Weidmann, N. B., & Gleditsch, K. S. (2011). Horizontal Inequalities and
Ethnonationalist Civil War: A Global Comparison. American Political Science Review,
105(03), 478-495.
Cederman, L.-E., Wimmer, A., & Min, B. (2010). Why Do Ethnic Groups Rebel? New Data and
Analysis. World Politics, 62(01), 87-119. doi: doi:10.1017/S0043887109990219
Chandra, K. (2004). Why Ethnic Parties Succeed: Patronage and Ethnic Headcounts in India.
Cambridge, UK ; New York: Cambridge University Press.
Chandra, K. (2006). What is Ethnic Identity and Does It Matter? Annual Review of Political
Science, 9(1), 397-424.
Chandra, K., & Wilkinson, S. (2008). Measuring the Effect of “Ethnicity”. Comparative
Political Studies, 41(4-5), 515-563.
Clarke, K. A., & Stone, R. W. (2008). Democracy and the Logic of Political Survival. American
Political Science Review, 102(03), 387-392.
Collier, P., & Hoeffler, A. (2004). Greed and Grievance in Civil War. Oxford Economic Papers,
56(4), 563-595.
Cox, T. S. (1976). Civil-military relations in Sierra Leone. Cambridge, Mass: Harvard
University Press.
Debs, A., & Goemans, H. E. (2010). Regime Type, the Fate of Leaders, and War. American
Political Science Review, 104(03), 430-445.
Dobbins, J. (2005). The UN's Role in Nation-Building: From the Congo to Iraq. Santa Monica,
CA: RAND Corporation.
Elbadawi, E., & Sambanis, N. (2000). Why Are There So Many Civil Wars in Africa?
Understanding and Preventing Violent Conflict. Journal of African Economies, 9(3), 244269.
Eller, J. D. (1999). From Culture to Ethnicity to Conflict: An Anthropological Perspective on
International Ethnic Conflict. Ann Arbor: University of Michigan Press.
Enloe, C. H. (1980). Police, Military, and Ethnicity: Foundations of State Power. New
Brunswick, N.J.: Transaction Books.
116

Esman, M. J. (1994). Ethnic Politics. Ithaca etc.: Cornell University Press.
Fearon, J. D. (1999). Why Ethnic Politics and "Pork" Tends To Go Together. Paper presented at
the MacArthur Foundation-sponsored conference on Ethnic Politics and Democratic
Stability, Chicago, IL.
Fearon, J. D., & Laitin, D. D. (2000). Violence and the Social Construction of Ethnic Identity.
International Organization, 54(04), 845-877.
Fearon, J. D., & Laitin, D. D. (2003). Ethnicity, Insurgency, and Civil War. American Political
Science Review, 97(01), 75-90. doi: doi:10.1017/S0003055403000534
Ferree, K. E. (2004). The Micro-foundations of Ethnic Voting: Evidence from South Africa. Cape
Town :: Institute for Democracy in South Africa.
Ghareeb, E., & Dougherty, B. (2004). Historical Dictionary of Iraq. Lanham, Md.: Scarecrow
Press.
Gleditsch, N. P., Wallensteen, P., Eriksson, M., Sollenberg, M., & Strand, H. (2002). Armed
Conflict 1946-2001: A New Dataset. Journal of Peace Research, 39(5), 615-637.
Goemans, H. E. (2008). Which Way Out? The Manner and Consequences of Losing Office.
Journal of Conflict Resolution.
Goemans, H. E., Gleditsch, K. S., & Chiozza, G. (2008). ARCHIGOS: A Data Set on Leaders
1875-2004.
Retrieved
from:
http://www.prio.no/CSCW/Research-andPublications/Publication/?oid=57558
Gurr, T. R. (1993). Why Minorities Rebel: A Global Analysis of Communal Mobilization and
Conflict since 1945. International Political Science Review, 14(2), 161-201.
Gurr, T. R. (2000). Peoples versus States: Minorities at Risk in the New Century. Washington,
D.C.: United States Institute of Peace Press.
Heger, L., & Salehyan, I. (2007). Ruthless Rulers: Coalition Size and the Severity of Civil
Conflict. International Studies Quarterly, 51(2), 385-403.
Hegre, H., & Nygard, H. M. (2012). Governance and Conflict Relapse. Paper presented at the
International Studies Association Annual Convention, San Diego, CA.
Horowitz, D. L. (2000). Ethnic groups in conflict (2nd ed.). Berkeley: University of California
Press.

117

Hurewitz, J. C. (1969). Middle East Politics: The Military Dimension. New York,: Published for
the Council on Foreign Relations.
Iraq Liberation Act, H.R.4655, 105th Congress (1997 - 1998) (1998).
Isaacs, H. R. (1989). Idols of the Tribe: Group Identity and Political Change ([Reprint] ed.).
Cambridge, Mass. etc.: Harvard University Press.
Jensen, N., & Wantchekon, L. (2004). Resource Wealth and Political Regimes in Africa.
Comparative Political Studies, 37(7), 816-841.
Jones, B. D. (2001). Peacemaking in Rwanda: The Dynamics of Failure. Boulder, CO: Lynne
Rienner Publishers.
Kennedy-Pipe, C. (1997). The Origins of the Present Troubles in Northern Ireland. London etc.:
Longman.
Laitin, D. D. (2004). Ethnic Unmixing and Civil War. Security Studies, 13(4), 350-365.
Leite, C., & Weidmann, J. (1999). Does Mother Nature Corrupt? Natural Resources, Corruption,
and Economic Growth. IMF Working Paper No. 99/85
Retrieved from SSRN:
http://ssrn.com/abstract=259928 or http://dx.doi.org/10.2139/ssrn.259928
Lewis, W. A. (1965). Politics in West Africa. London: Allen & Unwin.
Lijphart, A. (1977). Democracy in Plural Societies: A Comparative Exploration. New Haven:
Yale University Press.
Linder, W., & Bachtiger, A. (2005). What Drives Democratisation in Asia and Africa? European
Journal of Political Research, 44(6), 861-880.
Marashi, I., & Salama, S. (2008). Iraq's Armed Forces: An Analytical History. London:
Routledge.
Marshall, M. G., & Cole, B. R. (2011). Global Report 2011: Conflict, Governance, and State
Fragility. Retrieved from: http://globalpolicy.gmu.edu/political-instability-task-forcehome/pitf-phase-v-findings-through-2004/
Marshall, M. G., & Jaggers, K. (2011). Polity IV Dataset and Users' Manual: Political Regime
Characteristics
and
Transitions,
1800-2010.
Retrieved
from:
http://www.systemicpeace.org/inscr/p4manualv2010.pdf

118

McLauchlin, T. (2010). Loyalty Strategies and Military Defection in Rebellion. Comparative
Politics, 42(3), 333-350.
Melander, E. (2009). The Geography of Fear: Regional Ethnic Diversity, the Security Dilemma
and Ethnic War. European Journal of International Relations, 15(1), 95-124.
Miller, J. H., & Page, S. E. (2007). Complex Adaptive Systems: An Introduction to
Computational Models of Social Life. Princeton: Princeton University Press.
Newbury, C. (1992). Rwanda: Recent Debates over Governance and Rural Development. In G.
Hyden & M. Bratton (Eds.), Governance and Politics in Africa (pp. 193-219). Boulder,
Colorado: Lynne Rienner Publishers.
Ngaruko, F., & Nkurunziza, J. D. (2005). Civil War and Its Duration in Burundi. In P. Collier &
N. Sambanis (Eds.), Understanding Civil War: Evidence and Analysis (Vol. 1, pp. 35-61).
Washington DC: The World Bank.
Nkurunziza, J. D., & Ngaruko, F. (2002). Explaining Growth in Burundi: 1960-2000. The Center
for the Study of African Economies Working Paper Series, Paper 162, 1-68.
Østby, G., Nordås, R., & Rød, J. K. (2009). Regional Inequalities and Civil Conflict in SubSaharan Africa. International Studies Quarterly, 53(2), 301-324.
Page, S. E. (1999). Computational Models from A to Z. Complexity, 5(1), 35-41.
Posner, D. N. (2004). The Political Salience of Cultural Difference: Why Chewas and Tumbukas
Are Allies in Zambia and Adversaries in Malawi. American Political Science Review,
98(04), 529-545.
Prunier, G. r. (1995). The Rwanda Crisis: History of a Genocide. New York: Columbia
University Press.
Reynolds, A. (1999). Electoral Systems and Democratization in Southern Africa. Oxford etc.:
Oxford University Press.
Richardson, B. M. (1991). European Party Loyalties Revisited. The American Political Science
Review, 85(3), 751-775.
Riker, W. H. (1962). The Theory of Political Coalitions. New Haven,: Yale University Press.
Roeder, P. G. (2001). Ethnolinguistic Fractionalization (ELF) Indices, 1961 and 1985, from
http://weber.ucsd.edu/~proeder/elf.htm

119

Roessler, P. (2011). The Enemy Within: Personal Rule, Coups, and Civil War in Africa. World
Politics, 63(02), 300-346.
Ross, M. L. (2004). What Do We Know about Natural Resources and Civil War? Journal of
Peace Research, 41(3), 337-356.
Scarritt, J. R., & Mozaffar, S. (1999). The Specification of Ethnic Cleavages and Ethnopolitical
Groups for the Analysis of Democratic Competition in Contemporary Africa.
Nationalism and Ethnic Politics, 5(1), 82-117.
Simmons, J. W. (1967). Voting Behaviour and Socio-Economic Characteristics: The Middlesex
East Federal Election, 1965. The Canadian Journal of Economics and Political Science /
Revue canadienne d'Economique et de Science politique, 33(3), 389-400.
Singer, J. D., & Small, M. (2006). Correlates of War Project: International and Civil War Data,
1816-1992.
.
Retrieved
from:
http://dx.doi.org/10.3886/ICPSR09905.v1
Sisk, T. D., & Reynolds, A. (1998). Elections and Conflict Management in Africa. Washington,
DC: United States Institute of Peace Press.
Smith, A. D. (1986). The Ethnic Origins of Nations. Oxford: Basil Blackwell.
Stewart, F. (2008). Horizontal Inequalities and Conflict: An Introduction and Some Hypotheses.
In F. Stewart (Ed.), Horizontal Inequalities and Conflict: Understanding Group Violence
in Multiethnic Societies (pp. 364 S.). Basingstoke: Palgrave Macmillan.
Straus, S. (2006). The Order of Genocide: Race, Power, and War in Rwanda. Ithaca: Cornell
University Press.
Toft, M. D. (2003). The Geography of Ethnic Violence: Identity, Interests, and the Indivisibility
of Territory. Princeton, N.J.: Princeton University Press.
Tripp, C. (2007). A History of Iraq (3rd ed.). Cambridge: Cambridge University Press.
Van den Berghe, P. L. (1981). The Ethnic Phenomenon. New York: Elsevier.
Vogt, M. (2007). Ethnic Exclusion and Ethno-Nationalist Conflicts: How the Struggle over
Access to the State Can Escalate. (Lic phil I Univ Zèurich, 2008 - Ref Lars-Erik
Cederman), Zèurich.

120

Voice of America. (2010). AU Summit Approves Tougher Anti-Coup Measures Retrieved
October 2, 2012, from http://www.voanews.com/content/au-summit-approves-tougheranti-coup-measures-83263502/153002.html
Wagner, M., & Kritzinger, S. (2012). Ideological Dimensions and Vote Choice: Age Group
Differences in Austria. Electoral Studies, 31(2), 285-296.
Weidmann, N. B. (2009). Geography as Motivation and Opportunity. Journal of Conflict
Resolution, 53(4), 526-543.
Weinstein, W. (1977). The Burundi Coup. Africa Report, 22(01), 52-56.
Weinstein, W., & Schrire, R. A. (1976). Political Conflict and Ethnic Strategies: A Case Study of
Burundi. Syracuse, N.Y.: Maxwell School of Citizenship and Public Affairs, Syracuse
University.
Wimmer, A. (2002). Nationalist Exclusion and Ethnic Conflict: Shadows of Modernity.
Cambridge: Cambridge University Press.
Wimmer, A., Cederman, L.-E., & Min, B. (2009). Ethnic Politics and Armed Conflict: A
Configurational Analysis of a New Global Data Set. American Sociological Review,
74(2), 316-337.
Wimmer, A., & Min, B. (2006). From Empire to Nation-State: Explaining Wars in the Modern
World, 1816–2001. American Sociological Review, 71(6), 867-897.
World Bank. (2011). World Development Report 2011: Conflict, Security, and Development.
Washington, DC: World Bank.

121