70
Institutions and Electoral Violence Maurice René Dunaiski March 24, 2015

Why do some elections turn violent? The impact of institutions on electoral violence

Embed Size (px)

Citation preview

Institutions and Electoral Violence

Maurice René Dunaiski

March 24, 2015

Paris School of International Affairs (PSIA)

Sciences Po Paris

Master in Human Rights and Humanitarian Action

Academic Year 2014/2015

Supervised by Dr Bruno Cautrès

The copyright of this Master’s thesis remains the property of its author. No part of the contentmay be reproduced, published, distributed, copied or stored for public or private use without writtenpermission of the author. All authorization requests should be sent to [email protected] [email protected].

Abstract

Politically motivated violence and large-scale human rights abuses are a common feature of electoralcontests in the developing world. But why are some countries more likely to experience electoralviolence than others? Using a panel dataset on election-related social conflict in Africa since 1990,this paper shows that institutions matter in explaining a country’s risk of electoral violence. Goingbeyond previous quantitative research on electoral violence, this paper demonstrates that institutionalfactors such as the electoral system, decentralization, the separation of powers, and state capacity allhave an important impact on the likelihood of electoral violence. My findings indicate that, all elsebeing equal, centralized states with majoritarian electoral systems, semi-parliamentary regimes andweak state capacity are most at risk of experiencing electoral violence. Importantly, this paper alsoprovides evidence suggesting that the effect of institutions on electoral violence differs between thepre- and the post-election period as well as between pro- and anti-government perpetrators.

Contents

1 Introduction 2

2 Literature Review 4

2.1 Armed conflict . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.2 Repression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.3 Electoral violence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3 Theoretical Framework 16

3.1 What are institutions? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3.2 How can institutions shape behaviour? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3.3 Institutions and electoral violence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4 Data 28

4.1 Measuring electoral violence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

4.2 Measuring institutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

4.3 Controls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

5 Analysis 35

5.1 Pre-election violence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

5.2 Post-election violence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

6 Conclusion 51

Bibliography 53

Appendix 62

1

Chapter 1

Introduction

In the immediate aftermath of the 2007 general elections in Kenya more than 1000 people were killedand roughly 300.000 people were internally displaced as the result of violence between governmentloyalists and opposition supporters (International Crisis Group, 2008). Even though the magnitude ofthe killings and the scale of the destruction were unprecedented, the phenomenon of electoral violenceis certainly not unique to the case of Kenya. Recent elections in countries as diverse as Afghanistan,Côte d’Ivoire, Egypt, Nigeria, Guatemala and Thailand have also been marred by politically motivatedviolence and grave human rights abuses. Why did these elections turn violent, when so many otherelections in recent years remained peaceful? Despite the enormous destructive potential of electoralviolence, the question of what makes some elections more violent than others has been largely ignoredin the political science literature. We know relatively little about the frequency and geographicaldistribution of electoral violence across countries. We also do not have a lot of reliable informationabout who is primarily responsible for electoral violence - the state, or opposition actors? Furthermore,we have only a very limited understanding of the situational and structural factors that determine thelikelihood of electoral violence in a given country.

Only recently, political scientists have started to examine the phenomenon of electoral violence andbegun to offer explanations as to why some elections turn violent. While most of this work has takenthe form of qualitative case studies, in the last few years scholars have produced a number of pio-neering articles and working papers that systematically examine electoral violence from a quantitativeperspective (see Straus and Taylor, 2009; Hafner-Burton et al., 2010; Daxecker, 2012; Linebarger andSalehyan, 2012; Fjelde and Höglund, 2015). This development is undoubtedly due to the recent pro-liferation of high-quality data sources on ‘low-level’ conflict events, which now allow researchers toexamine electoral violence from a quantitative, cross-national perspective (see Raleigh et al., 2010;Salehyan et al., 2012; Sundberg and Melander, 2013). Previous explanations of electoral violence havemostly focused on situational factors such as the ‘closeness’ of the electoral race, the extent of electoralfraud, or the presence of international election monitors (see Wilkinson, 2006; Collier and Vicente,2008; Daxecker, 2014; Hafner-Burton et al., 2014). However, the potential impact of country-levelinstitutions on the likelihood of electoral violence has been almost completely ignored1. This gap inthe literature on electoral violence is surprising, given that the more developed literature on conflictmanagement and ‘institutional engineering’ provides empirical evidence suggesting that institutionscan have an important impact on the likelihood of civil war onset (e.g. Brancati, 2006; Schneider andWiesehomeier, 2008; Hendrix, 2010; Bogaards, 2013), the level of state repression (e.g. Cingranelli andFilippov, 2010) and the likelihood of electoral fraud (e.g. Birch, 2007; Hicken, 2007).

This paper aims to fill the above-mentioned gap in the quantitative literature on electoral violence.It shows that institutions matter in explaining why some countries are more prone to election-relatedviolence than others, and that accordingly any explanation of electoral violence needs to take country-level institutional factors into account. I use negative binomial regression models and a panel dataset

1For two important exceptions, see Opitz et al. (2013) and Fjelde and Höglund (2015).

2

on election-related social conflict events in Africa between 1990 and 2010 to uncover the institutionaldrivers of electoral violence. I focus on electoral violence in Africa for a number of theoretical andmethodological reasons (see Chapter 4). My statistical analysis shows that a country’s electoral system,its state structure, the separation of powers, as well as its state capacity all have an important impacton the risk of electoral violence. My results suggest that, all else being equal, unitary states withmajoritarian electoral systems, semi-parliamentary regimes and weak state capacity are most at riskof experiencing electoral violence. Using interactions, I also show that a country’s ethnic compositionhas an important mediating impact on the violence-inducing effect of majoritarian electoral systems.

This paper advances the existing literature on electoral violence in several ways. To begin with, itrepresents the first serious attempt to systematically examine the impact of a whole range of insti-tutional factors on the likelihood of electoral violence. Previous studies on the relationship betweeninstitutions and electoral violence have for the most part taken a qualitative, case-study approach(see Höglund, 2009; Opitz et al., 2013), or have narrowly focused on one specific institutional fac-tor (see Hafner-Burton et al., 2014; Fjelde and Höglund, 2015). Secondly, this paper is the first tosystematically examine how the impact of institutions on electoral violence differs between the pre-and the post-election period. Previous quantitative studies on electoral violence have either failed toadequately distinguish between the two periods (see Fjelde and Höglund, 2015), or have only focusedon one of the two (see Daxecker, 2012; Borzyskowski, 2013; Daxecker, 2014; Hyde and Marinov, 2014).This is unfortunate, given the different dynamics of violence at play during the pre- and post-electionperiods. Lastly, most of the previous empirical literature has failed to distinguish between violenceperpetrated by incumbents and violence perpetrated by ‘challengers’ or other non-state actors (forexceptions see Straus and Taylor, 2009; Daxecker, 2014; Fjelde and Höglund, 2015). However, as thispaper shows, the two competing electoral camps can have divergent incentives to resort to electoralviolence depending on a number of institutional constraints.

The paper is structured as follows. To begin with, Chapter 2 develops a working definition of electoralviolence and provides a comprehensive overview of the relevant literature. The literature review alsolinks this new field of research with the more established literature on armed conflict (see Section 2.1)and state repression (see Section 2.2). In Chapter 3 I then develop the theoretical framework that willinform my empirical analysis. In particular, I develop a number of hypotheses about the relationshipbetween specific country-level institutional factors and the likelihood of electoral violence. Thereafter,Chapter 4 presents my research design and the panel dataset in more detail. In Chapter 5 I discussthe main findings from my regression analysis. Overall, I find that institutional factors are importantdeterminants of both pre- and post-election violence. Importantly, my findings also suggest that theimpact of individual institutional factors can differ between the two periods and in some cases alsodepend on the perpetrator. Lastly, Chapter 6 concludes my analysis and presents possible avenues forfuture research on the subject of electoral violence.

3

Chapter 2

Literature Review

One of the defining features of democratic regimes is that they hold regular and competitive elections(Dahl, 1973). However, it is clear that simply holding elections does not make a country democratic.Since the end of the Cold War it has become increasingly common to see regular elections takingplace in countries that lack even the most basic attributes of liberal democracies, such as freedom ofexpression or freedom of association (Diamond, 2002). In fact, elections have become so common thatonly ten countries failed to hold some form of direct national election between 2000 and 2006 (Hydeand Marinov, 2009). Scholars have acknowledged the increasing variety of regime types that holdregular elections by creating typologies that distinguish mature democracies from “hybrid”, “competitiveauthoritarian”, “electoral authoritarian” and other types of autocratic regimes (see e.g. Diamond, 2002;Levitsky and Way, 2002; Schedler, 2002; Gandhi and Lust-Okar, 2009). Elections in these less-than-democratic countries lead to what Gates et al. (2006) call “institutionally inconsistent” regimes thatexhibit traits of both democracies and autocracies. In such contexts, elections are primarily used toplacate external constituents such as foreign donors or pressure groups and their outcomes are oftenmanipulated through violence, intimidation and fraud (Linebarger and Salehyan, 2012). It is thereforeunderstandable that scholars and practitioners have started to raise questions about the potentiallynegative impacts of premature democratization efforts and the role of violence in electoral competition(see e.g. Collier, 2011; UNDP, 2011). In the following chapter I provide an overview of the academicliterature that addresses the connection between elections and violence. To begin with, I reviewthe relatively well-developed fields of research that examine the link between elections and two moreprominent forms of political violence: armed conflict and repression. Thereafter I turn to the small butburgeoning field of research that deals specifically with the phenomenon of electoral violence. I showthat the few existing quantitative studies on electoral violence have not yet adequately addressed thequestion of how institutional factors such as the electoral system, decentralization or state capacityshape a country’s predisposition to experience electoral violence.

2.1 Armed conflict

The pacifying effect of democracy has found substantial empirical support in the literature on intra-and inter-state armed conflict. A large number of scholars have shown that democratic regimes aremore peaceful in international affairs than their authoritarian counterparts (Doyle, 1986; Lake, 1992;Schultz, 1998). The impact of democratic governance on the likelihood of internal armed conflict issomewhat less clear and scholars have found both linear and inverted ‘U-shaped’ relationships (seeHegre et al., 2001; Hegre and Sambanis, 2006; Schneider and Wiesehomeier, 2008; Vreeland, 2008).Nevertheless, the negative association between democracy and war, also known as the “democraticpeace”, is one of the most robust and consistent findings in the literature on armed conflict (see e.g.Dafoe et al., 2013; Hegre, 2014). However, the picture becomes much less clear-cut when we look at therelationship between elections - only one component of full-fledged democracies - and the occurrence of

4

armed conflict. The main concern of scholars studying this relationship is that the premature holdingof elections in countries emerging from civil war and authoritarianism could trigger renewed violenceby further polarizing deeply divided communities (see Collier, 2011). This concern has resulted in alarge number of papers that investigate the impact of elections on the likelihood of armed conflict.

There is considerable debate about the advantages of early versus delayed elections in post-conflictand post-authoritarian societies. Representing the more optimistic school of thought, Cheibub et al.(2012) argue that elections in unconsolidated democracies do not lead to more conflict. In fact, theirquantitative analysis suggests that elections can have a pacifying effect on divided societies. In theirrecent paper they examine the causal relationship between multiparty elections in Africa and theinitiation of civil wars in the period 1960-2005. Their data provides no support for the competinghypothesis that elections increase the probability of a civil war onset. The optimistic view of Cheibubet al. is shared by a number of other scholars, who stress that repeated elections can contribute tosuccessful transitions from civil war and authoritarianism to stable democratic governance (see e.g.Wantchekon and Neeman, 2002; Birnir, 2007; Lindberg, 2008). In contrast, a number of scholars haveprovided empirical evidence showing that premature democratization efforts can indeed precipitatecivil war in post-conflict or post-authoritarian societies. For example, Collier has argued that electionsin “dangerous places” - i.e. countries with weak institutions and deep social cleavages - often act as atrigger of civil war (Collier et al., 2008; Collier, 2011). Similarly, several recent papers analyze cross-national datasets and claim to establish a negative causal impact of elections on civil war initiation(Hegre et al., 2001; Mansfield and Snyder, 2009; Cederman et al., 2010a; Flores and Nooruddin, 2012;Cederman et al., 2013). These papers suggest that international democracy promotion efforts can becounterproductive in countries that do not yet exhibit the institutional and societal prerequisites forpeaceful democratic competition.

Cheibub et al. criticize this more pessimistic school of thought for not adequately addressing en-dogeneity issues. They maintain that “multiparty elections in democratizing, non-institutionalizedauthoritarian regimes are observed precisely when the leader already faces a significant threat of beingremoved by force: they are held when a civil war is already a real possibility and elections are nothingbut an attempt to avoid war.” (Cheibub et al., 2012; p.8). According to Cheibub et al. it is this dy-namic, rather than electoral competition per se, that seems to be the driving factors behind the morepessimistic findings in the literature. They show that if one takes the issue of endogeneity seriously,the impact of elections on armed conflict turns out to be much more benign. In short, while scholarsgenerally agree that democracy decreases the likelihood of armed conflict, there is still no consensuson whether elections can be said to have the same effect. However, recent scholarship has tried tobridge the gap between the optimistic and pessimistic schools of thought by suggesting that the perilsand promises of elections may depend on the nature and constellation of the groups competing forpower (see Cederman et al., 2013). Such an approach will allow future research to further disaggregateand unpack the concept of electoral competition and hopefully offer valuable new insights into therelationship between elections and armed conflict.

One of the major shortcomings of the quantitative literature on the elections-conflict nexus is its re-liance on country-year time-series datasets. As a consequence of this approach, scholars cannot analyzethe violent dynamics at play during the electoral process. In other words, they cannot account fordifferences between pre- and post-election violence and they can only examine the long-term con-sequences of elections. In addition, their data usually fails to capture any violence associated withelections that does not reach the very high threshold of armed conflict2 (Linebarger and Salehyan,2012). Yet, for conflict researchers it is important to understand election-related violence short ofarmed conflict, given that in many cases such ‘low-level’ violence can serve as a “training ground forwarfare” (Höglund, 2010; p.4). In this respect, one question that future research will need to addressis whether the spread of elections to less-than-democratic countries has inadvertently fueled ‘low-level’conflicts that do not immediately reach the threshold of armed conflict (see Goldsmith, 2014).

2One of the most frequently used datasets, the UCDP/PRIO Armed Conflict Dataset, defines armed conflict as “acontested incompatibility that concerns government and/or territory where the use of armed force between two parties,of which at least one is the government of a state, results in at least 25 battle-related deaths” (see Gleditsch et al., 2002;Themner and Wallensteen, 2013).

5

2.2 Repression

The literature on state repression3 represents another well-developed field of research that examines thelink between elections and political violence. It is widely accepted that democracy and human rightsgo together. Indeed, there is ample empirical evidence showing that stable democracies are much lesslikely to violate basic human rights than any other regime type (Poe and Tate, 1994; Poe et al., 1999;Davenport and Armstrong, 2004; Bueno De Mesquita et al., 2005; Davenport, 2007; Young, 2009).In contrast, the relationship between elections and the abuse of human rights is much less clear-cut.Existing quantitative studies have largely focused on examining whether state repression increases ordecreases during election years. However, these studies have come to conflicting conclusions about theimpact of elections on state repressive behaviour. Representing the more optimistic camp, Davenport(1997) finds that the level of repression in authoritarian regimes decreases significantly during yearsin which elections are held. He argues that this is due to the fact that authoritarian regimes want tolegitimize themselves and improve their image during election years. In a follow-up study he also findsthat national elections in both democracies and autocracies tend to reduce censorship of the mediaand state restrictions placed upon individuals and groups (Davenport, 1998).

A number of scholars have since challenged Davenport’s more optimistic findings. For example,Richards (1999) finds that the impact of national elections on levels of human rights abuse is notstatistically significant when controlling for population size, economic development, armed conflictand the extent of suffrage. Similarly, Conrad and Moore (2010) argue that elections have little effecton the likelihood that states terminate the use of torture when they are faced with violent dissent orcivil war. More recent studies have also tried to further disaggregate the concept of elections. Theyfind that the effect of elections on levels of state repression depends on the type of electoral competitionand the incentives created by different electoral rules. For example, Richards and Gelleny (2007) arguethat executive elections increase the level of human rights abuse, while legislative elections tend to havethe opposite effect. In a similar vein, Cingranelli and Filippov (2010) show that low-magnitude pro-portional representation districts and candidate-centered voting rules are the two factors that can besaid to significantly decrease the level of state repression in democratic countries. In short, while thereis widespread agreement that democracies are less likely to resort to repression than any other regimetypes, there is still no consensus on whether elections have the same potential to reduce human rightsviolations. As with the literature on armed conflict, one important shortcoming of the quantitativeliterature on state repression is that it relies almost exclusively on aggregate annual measures of humanrights abuse4. The problem with such highly aggregated measures is that the recorded abuse may ormay not have been related to elections. Furthermore, the ‘country-year’ format makes it impossiblefor researchers to distinguish between pre- and post-election violence (Hafner-Burton et al., 2014).

2.3 Electoral violence

What makes electoral violence distinct from other forms of political violence such as armed conflictor state repression? What does previous academic work say about the causes and consequences ofelectoral violence? What are the knowledge gaps in this field of research? In the following sectionI provide answers to these three questions. I begin with developing a working definition of electoralviolence, before reviewing the literature on this subject and highlighting its main shortcomings.

3For the purpose of this paper I define state repression as “the actual or threatened use of physical sanctions againstan individual or organization, within the territorial jurisdiction of the state, for the purpose of imposing a cost onthe target as well as deterring specific activities and/or beliefs perceived to be challenging to government personnel,practices or institutions” (Davenport, 2007; p.2) The violation of physical integrity rights (i.e. torture, disappearance,political imprisonment and extrajudicial killing) is the category of human rights abuse that is most frequently used asthe dependent variable in the quantitative literature on state repression (Cingranelli and Filippov, 2010).

4The two most frequently used datasets on human rights violations are the Cingranelli-Richards (CIRI) Human RightsData Project (Cingranelli and Richards, 2010) and the Political Terror Scale (PTS) (Gibney et al., 2014). Both datasetsuse country-years as their unit of analysis. However, while CIRI measures human rights ‘practices’ of governments, PTSmeasures human rights ‘conditions’ in countries (Cingranelli and Filippov, 2010).

6

What is electoral violence?

In line with Höglund (2009), I suggest that it is primarily the motive and timing that set electoralviolence apart from other forms of political violence such as armed conflict, state repression or terrorism.Even though electoral violence often occurs in countries that are experiencing other forms of violentconflict, it is the motive and timing that make it an analytically distinct phenomenon and a form ofviolence with unique causes and consequences (see Höglund, 2009; p.415). The general motive behindelectoral violence is to influence the electoral process. This characterization is in line with most ofthe conceptual literature on electoral violence, which stresses the instrumental or goal-oriented natureof such violence (see e.g. Laakso, 2007; Sisk, 2008; Höglund, 2009; Collier and Vicente, 2012). Themore specific motives of actors engaging in electoral violence usually take one of the following fourforms. Firstly, actors might use violence in opposition to elections of any sorts (e.g. the Taliban inpost-2001 Afghanistan). Secondly, actors might violently oppose only specific electoral contests, butnot democracy per se. Thirdly, actors might accept electoral competition, but resort to violence asa means to influence the election results in their favour5. Lastly, actors might use violence in theaftermath of elections to overturn or defend the proclaimed results (Höglund, 2009). In order to drawa conceptual distinction between electoral violence and armed conflict, Fjelde and Höglund (2015;p.2) argue that electoral violence is “employed alongside other constitutional and non-constitutionalstrategies for retaining power”, whereas the outbreak of armed conflict “represents an exit strategyfrom the domain of regular political competition”. Unfortunately, this distinction fails to capture thefirst, second and fourth forms of electoral violence identified above, given that they can clearly alsorepresent ‘exit strategies’ from the domain of regular political competition. Hence, I suggest that asecond dimension (i.e. “timing”) needs to be taken into account if we want to conceptually separateelectoral violence from other forms of political violence.

With regard to timing, electoral violence takes place during the electoral process6 and generally fallsinto one of the following three periods: the pre-election period, the election day(s), or the post-election period. In this respect it is important to note that the motives behind the use of violenceare different in the three periods. While violence in the the pre-election phase and during the electionday(s) is generally aimed at increasing the vote share of the perpetrator or disrupting the electoralprocess altogether, violence in the post-election period tends to be aimed at challenging or defendingthe ‘official’ results (Hafner-Burton et al., 2014). This conceptual distinction between pre- and post-election violence is also highlighted by Daxecker (2014), who notes that the underlying causes ofelectoral violence may differ depending on when it occurs. She suggests that pre-election violence andviolence during the election day(s) should be conceptualized as “strategic manipulation” or a form ofelection fraud, whereas post-election violence follows a different logic and should be conceptualized as“a response to outcomes, particularly if fraud occurred” (Daxecker, 2014; p.233). With these conceptualissues in mind, we can now define electoral violence as the use or threat of violence aimed at influencingan impeding electoral contest or an announced electoral result. This definition of electoral violenceencompasses direct physical violence, harassment and the destruction of property (e.g. polling stations)by state actors as well as non-state actors, with the purpose of influencing the electoral process, eitherin the pre-election period, during polling day, or in the post-election period (for similar definitions seeStraus and Taylor, 2009; Opitz et al., 2013; Goldsmith, 2014).

Data on electoral violence

The phenomenon of electoral violence is much more frequent than is commonly thought. For example,in a first systematic attempt to map the prevalence of electoral violence in Africa, Straus and Taylor(2009) find that around 60 percent of all national elections since 1990 have been accompanied by some

5This category can also include the use of violence to change the demography of a constituency and the use of violenceto discipline would-be defectors from a winning coalition (see Straus and Taylor, 2009; p.19).

6While it is impossible to draw clear temporal boundaries around the concept of ‘electoral process’, it has beensuggested that this process typically begins with events such as voter and party registrations or campaign initiations,and that it usually ends with the inauguration of the newly elected officials (Höglund, 2009).

7

form of violent intimidation, even though only 20 percent experienced large-scale civilian casualties.Another recent study finds that roughly one fifth of all elections worldwide are marred by electoralviolence and that this phenomenon is not confined to Sub-Saharan Africa, but affects nearly all worldregions7 (Norris, 2012). Other studies have pointed out that in some countries such as Ethiopia,Kenya, Nigeria, the Philippines, India, or Thailand the electoral process is routinely accompanied bymassive, organized campaigns of violence and intimidation (see e.g. UNDP, 2011; Hafner-Burton et al.,2014; Daxecker, 2014). Yet, despite the high frequency and prevalence of electoral violence across theworld, this phenomenon has only recently started to be examined systematically. The question of whatmakes some elections more violent than others has been largely ignored in the empirical political scienceliterature. This stands in stark contrast to the numerous articles reviewed above, which have examinedthe question of whether elections can trigger armed conflicts and state repression. A number of earlystudies deal with theoretical and conceptual issues related to electoral violence (see e.g. Rapoport andWeinberg, 2000; Fisher, 2002; Höglund, 2009). In addition, several recent case studies and comparativepapers examine electoral violence from a qualitative angle8 (see e.g. Guelke, 2000; Pausewang et al.,2003; Wilkinson, 2006; Laakso, 2007; Bratton, 2008; Collier and Vicente, 2012). These papers providedetailed information about the micro-level dynamics of electoral violence and they develop valuablehypotheses about the causes and consequences of electoral violence. However, the lack of cross-nationaldata on electoral violence has largely prevented scholars from studying the issue more systematicallyand has hampered efforts to test the theories developed in the qualitative literature (Linebarger andSalehyan, 2012).

Until recently, very little data on political violence were available in cross-national format, aside fromthe commonly used ‘country-year’ datasets on armed conflict9 and state repression10. The few cross-national datasets that included forms of ‘low-level’ violence such as the Cross-National Time-SeriesArchive (Banks, 2011), typically lacked sufficient detail to disaggregate violent events by issue-typeand to attribute them to electoral contests (Linebarger and Salehyan, 2012). However, the last coupleof years have witnessed an explosion of micro-level data-gathering efforts that will allow researchers tosystematically analyze the phenomenon of electoral violence without needing to rely on the restrictive‘country-year’ format. These include the Social Conflict in Africa Database (SCAD) (Salehyan et al.,2012), the Armed Conflict Location and Event Data Project (ACLED) (Raleigh et al., 2010), and theUCDP Georeferenced Event Dataset (Sundberg and Melander, 2013)11. In addition, the relativelynew dataset on National Elections across Democracy and Autocracy (NELDA), created by Hyde andMarinov (2012b), now allows researchers to study patterns of electoral violence using ‘election rounds’instead of ‘election years’ as the unit of analysis. How has recent scholarship utilized this new wealthof data on electoral violence? In the following sub-sections I provide an overview of the burgeoningfield of research that examines the causes and consequences of electoral violence. In keeping withthe important conceptual distinction between pre- and post-election violence identified earlier, I beginwith an examination of the existing literature on the causes and consequences of pre-election violence,before turning to the related but distinct literature on post-election violence.

7The main exceptions are Western Europe and North America, even though some sporadic electoral violence can beobserved in these regions. For example, the Basque separatist organization ETA has repeatedly used violence to disruptelections in Spain (Norris, 2012; p.13).

8Two recent case studies of electoral violence in Kenya and Burundi also examine the phenomenon from a quantitativeperspective (see Dercon and Gutierrez-Romero, 2012; Colombo et al., 2014). However, due to the case study format,their empirical findings cannot be readily generalized to different contexts.

9The two most commonly used cross-national datasets on armed conflict are the Correlates of War (Singer and Small,1994) and the UCDP/PRIO Armed Conflict Dataset (Gleditsch et al., 2002).

10see Cingranelli-Richards (CIRI) Human Rights Data Project (Cingranelli and Richards, 2010) and the PoliticalTerror Scale (PTS) (Gibney et al., 2014).

11So far the geographic scope of these disaggregated, event-based datasets is limited to countries in Africa. However,ACLED researchers have started gathering data on some non-African countries (see http://www.acleddata.com/data/)and Ursula Daxecker at the University of Amsterdam has initiated an events-based dataset on electoral violence withglobal coverage for the years 1990-2013 (personal correspondence). As far as I am aware, there are also three smallerdatasets on electoral violence in Africa that are unfortunately not publicly available (see Lindberg, 2008; Straus andTaylor, 2009; Arriola and Johnson, 2012).

8

Literature on pre-election violence

The literature on the causes and consequences of pre-election violence can be roughly divided intotwo groups. The first group focuses on the incentives and motives of political actors who use violencestrategically in order to manipulate the electoral process in their favour12 (Chaturvedi, 2005; Wilkinson,2006; Collier and Vicente, 2012). The second group focuses on structural or ‘enabling’ factors thatmake some countries more susceptible to electoral violence than others (Höglund, 2009; Straus andTaylor, 2009; Linebarger and Salehyan, 2012). Only recently have scholars tried to explain pre-electionviolence by integrating the two approaches (see Hafner-Burton et al., 2014; Fjelde and Höglund, 2015).Chaturvedi (2005) uses formal models and journalistic evidence to develop an agent-centered theoryof pre-election violence. He treats pre-election violence as a form of electoral manipulation that can besubstituted with “ideological exhortation” (Chaturvedi, 2005; p.189). He argues that, in an electoralcontest between two parties, the potential for pre-election violence will decrease as the fraction ofundecided voters goes up. In other words, very ‘close’ electoral contests do not bode well for stability.He also suggests that incumbent parties will be more likely to resort to violent manipulation of theelectoral contest given their access to state resources. Unfortunately, Chaturvedi provides no systematicempirical evidence to support his hypotheses.

In his seminal study on ethnic riots and elections in India, Wilkinson (2006) also treats pre-electionviolence as a form of strategic manipulation that can be used by politicians to increase their vote sharein ‘close’ electoral contests. He provides a detailed subnational analysis of electoral violence in Indiaand argues that politicians incite or suppress ethnic violence depending on whether the support ofethnic minorities is necessary for their electoral success. Even though Wilkinson does provide somestatistical evidence to support his claims, the focus on regional elections in India means that his findingscannot be readily generalized to other countries. Collier and Vicente (2012) develop a theory of pre-election violence that is similar to the one put forward by Chaturvedi (2005) and Wilkinson (2006).They also treat electoral violence as a form of “strategic manipulation” used by politicians and politicalparties to shape the electoral process in their favour. However, in their formal models, violence canbe substituted with both vote-buying and ballot fraud as tactics to increase vote shares. They arguethat electoral violence is most likely when unpopular incumbents fear losing their grip on power as theresult of elections. Similarly, electoral violence is likely to be the dominant strategy of weak challengersthat want to increase their vote share. In contrast, strong incumbents facing localized competitionare more likely to resort to the non-violent manipulation strategies of vote-buying and ballot fraud.Again, the main shortcoming of Collier and Vicente’s paper is that they cannot systematically testtheir theory of pre-election violence. The lack of cross-national data on vote-buying and ballot fraudmeans that they can only refer to illustrative case studies to back up their theoretical arguments (seeCollier and Vicente, 2012; p.135ff).

In her conceptual analysis of electoral violence, Höglund (2009) highlights three structural factors thatcan explain why elections in some countries are more violent than in others. To begin with, she suggeststhat “patrimonialism” - i.e. a system of governance based on patron-client networks - fosters electoralviolence13. While this hypothesis sounds plausible, the relationship between “patrimonialism” and elec-toral violence does not readily lend itself to empirical investigation. Besides problems related to theoperationalization of “patrimonialism”, it is also likely that the phenomenon is endogenous to politicalviolence. Indeed, Höglund herself suggests that “politics of patronage tends to fester in countries withhigh levels of insecurity” (Höglund, 2009; p.420). Secondly, she suggests that political competitionin itself can foster electoral violence, given that it exacerbates pre-existing social cleavages. In linewith arguments previously put forward by Chaturvedi (2005) and Wilkinson (2006), she maintainsthat politicians in “close races” have strong incentives to foment pre-election violence (Höglund, 2009;p.421). Lastly, Höglund argues that the design of the electoral system will impact a country’s likeli-hood to experience electoral violence. Drawing on previous literature in the field of electoral systemdesign and conflict management (e.g. Reilly et al., 1999; Reilly, 2002b; Norris, 2004), she predicts that

12Some of this research provides robust empirical evidence showing that electoral violence can indeed shape voterturnout (and precipitate opposition boycotts) in favour of the perpetrator (Bratton, 2008; Collier and Vicente, 2008;Hafner-Burton et al., 2012).

13A similar argument is made by Arriola and Johnson in an unpublished manuscript (Arriola and Johnson, 2012).

9

single-winner voting systems are more prone to electoral violence than voting systems that allow formultiple winners. In this paper, Höglund develops a number of interesting hypotheses about the causesof electoral violence. It is particularly noteworthy that she points to ‘institutional’ factors such as acountry’s electoral system as potential determinants of pre-election violence. However, she unfortu-nately does not attempt to systematically test her hypotheses and fails to offer substantive evidenceto support her claims.

To my knowledge, Straus and Taylor (2009) are the first to analyze the phenomenon of electoralviolence from a cross-national, quantitative perspective. They code a relatively crude four-point scaleto measure the level of violence in all African elections between 1990 and 2007. While their primaryaim is to map the frequency and spread of electoral violence across Africa, Straus and Taylor alsodevelop a number of hypotheses to explain the variation in electoral violence across countries. Amongstother things, they predict that the likelihood of electoral violence is influenced by the ‘closeness’ ofthe electoral race, the country’s regime type, its growth rates and the level of ethnic polarization.Unfortunately, Straus and Taylor do not attempt to systematically test their propositions, despite theeffort they put into gathering data on electoral violence. While the bivariate correlations presentedin their paper seem to support at least some of their arguments, this rudimentary statistical methodsimply does not allow them to infer robust causal patterns. Using the newly available SCAD databaseon ‘low-level’ social conflict in Africa for the period 1990-2009, Linebarger and Salehyan (2012) examinethe conditions under which elections are more or less likely to become violent. Their events-basedmeasure of electoral violence is more sophisticated than the four-point scale used by Straus and Taylor(2009), yet it does not distinguish precisely between pre- and post-election violence. Rather, it codesviolent events that occurred before, during and after ‘election months’. Linebarger and Salehyan findthat electoral contests in Africa are strongly associated with higher levels of ‘low-level’ social conflictand election-related violence. They also hypothesize that elections held during civil war as well aselections in poor, authoritarian, post-conflict countries are more susceptible to violence. However, theydo not find statistically significant evidence that elections during civil war or in post-conflict contextsare more likely to suffer from violence. In contrast, their findings indicate that mature democraticregimes as well as higher GDP per capita greatly diminish the level of violence during election months.

Focusing on an international determinant of pre-election violence, Daxecker (2014) argues that thepresence of international election monitors results in a temporal shift in the use of pre-election violence.She uses a SCAD-based measure of electoral violence and original data on election monitoring missionsin Africa to show that the presence of reputable election monitors makes violence in the pre-electionperiod more likely. She argues that this is the result of domestic elites strategically adjusting theirmanipulative tactics to the presence of monitors by ‘moving’ the violence from the election day(s)to earlier periods in the campaign. Daxecker does not find evidence for a statistically significant ef-fect of monitors on the level of violence on election day itself, but this might be due to her researchdesign. The problem is that the level of pre-election violence is likely to alter the bargaining dynam-ics among domestic actors and hence influence the level of confrontation on election day. In otherwords, it becomes difficult to observe the direct effect of monitors on election-day violence, given thatthey influence the level of violence in the pre-election period. One of the most systematic empiricalanalyses of the causes of pre-election violence is presented in a recent paper by Hafner-Burton et al.(2014). They argue that the likelihood of pre-election violence can be predicted by a combination oftwo factors: Firstly, the incumbent’s fear of losing power as the result of an election and, secondly, the“institutionalized constraints” placed on her decision-making powers. Hafner-Burton et al. hypothesizethat the more uncertain the incumbent’s victory and the less constrained she is to use state powers inher favour, the more likely it is that a country will experience pre-election violence. They systemati-cally assess the observable implications of their argument, using variables from the NELDA dataset,which contains cross-country information on all election rounds between 1981 and 2004 (see Hyde andMarinov, 2012b). Hafner-Burton et al. find empirical support for their theoretical argument, whilecontrolling for perceptions of pre-election fraud, the incumbent’s age and tenure, economic develop-ment, population size, civil war, and the overall level of repression. Their paper presents perhaps themost thorough quantitative investigation into the causes of pre-election violence to date. However, itsuffers from a number of methodological shortcomings.

10

Most importantly, Hafner-Burton et al. do not specify the perpetrator of the electoral violence in thecoding of their main dependent variable (“Nelda33”). This is problematic because the paper’s centralargument relies on the assumption that it is the incumbent who commits acts of pre-election violence inorder to increase her chances of winning the election (see Hafner-Burton et al., 2014; p.150). However,Nelda33 codes “any significant violence relating to the elections that resulted in civilian deaths” (Hydeand Marinov, 2012a; p.16), regardless of whether it was committed by the incumbent’s forces or bysupporters of the opposition. This conceptual confusion is particularly unfortunate in light of recentstudies on electoral violence in Africa, which suggest that the issue of who is the primary perpetratorof electoral violence is far from settled. While Daxecker (2014; p.240) and Collier and Vicente (2012)suggest that non-state actors or ‘challengers’ are the primary perpetrators of electoral violence, Strausand Taylor (2009), Onapajo (2014) as well as Fjelde and Höglund (2015) find that incumbents areresponsible for an overwhelming majority of violent events in the electoral period. Secondly, Hafner-Burton et al. do not differentiate between pre- and post-election violence in the coding of their maindependent variable14 even though their argument relies on the conceptual separation of the two phases.Their main argument is that the likelihood of pre-election violence is determined primarily by theincumbents fear of losing power as well as the extent of institutional constraints placed upon her.However, in the second part of their paper, Hafner-Burton et al. also make a related (but distinct)argument about the consequences of pre-election violence. They suggest that the use of violence inthe pre-election period increases the likelihood of post-election protests and state repression. In otherwords, they argue that pre-election violence, while often successful in manipulating elections in favourof the incumbent, can at the same time ‘backfire’ by increasing the likelihood of post-election protest.These protests can in turn trigger state repression and can even precipitate the incumbent’s downfall.Nelda33 is problematic as a measurement of pre-election violence given that it also includes violenceperpetrated in the post-election period. This makes it difficult for Hafner-Burton et al. to claim thatthe former has a causal impact on the latter.

Lastly, the key ‘institutional’ explanatory variable used by Hafner-Burton et al. is not very wellsuited to establish an independent effect of institutional state features on the likelihood of pre-electionviolence. The variable “executive constraints” from the Polity IV Project is based on the executive’spractice rather than a country’s institutional or constitutional design and hence might be endogenous tothe executive’s resort to violence and repression (Marshall et al., 2013; p.24). Exactly which institutionsimpose “executive constraints” is also not at all apparent from the coding. According to the PolityIV codebook, any “accountability groups” can theoretically impose constraints on executive power.In Western democracies these are usually legislatures. However, other kinds of accountability groupsincluded in the Polity IV measure are “the ruling party in a one-party state, councils of nobles orpowerful advisors in monarchies, the military in coup-prone polities, and in many states a strong,independent judiciary” (Marshall et al., 2013; p.24). The number and variety of “accountability groups”is so large that it becomes difficult to draw any policy-relevant conclusions from the finding that theexistence of “executive constraints” reduces the likelihood of electoral violence. Their argument aboutthe consequences of pre-election violence is further developed in Hafner-Burton et al. (2012). In thisworking paper, the authors examine how the use of pre-election violence affects the chances of theincumbent to stay in power. To begin with, they hypothesize and show that the use of pre-electionviolence by the incumbent increases her chances of being re-elected by swaying voter turnout in herfavour and precipitating election boycotts of opposition parties. Secondly, they argue that the use ofpre-election violence increases the likelihood of post-election protests. These protests in turn make theannulment of elections more likely and often force the incumbent to resign. The empirical analysispresented in this working paper has similar shortcomings as the journal article reviewed above (seeHafner-Burton et al., 2014). Again, Hafner-Burton et al. use a composite measure of pre-electionviolence (“Nelda15” and “Nelda33”) that does not distinguish between pre-election violence and post-election violence. In addition, the measurement of pre-election violence also includes anti-incumbentviolence and thus muddles up their argument about the impact of government-initiated violence onincumbent tenure.

14The Nelda33 variable indicates whether there was “significant violence involving civilian deaths immediately before,during or after the election” (Hyde and Marinov, 2012a; p.16, emphasis added).

11

In a forthcoming article, Fjelde and Höglund (2015) argue that majoritarian electoral systems makeunconsolidated democracies more susceptible to electoral violence given that the ‘stakes’ in such sys-tems are much higher than in proportional representation (PR) systems. They also hypothesize thatmajoritarian electoral systems are more likely to fuel violence in countries where economic inequality ishigh and where large ethno-political groups are systematically excluded from state power. Fjelde andHöglund (2015) test and confirm these three arguments using cross-national data on electoral violencein Sub-Saharan Africa between 1990 and 2010. Their paper represents the first serious attempt in thequantitative literature to analyze the way in which an ‘institutional’ factor can influence a country’spredisposition to experience electoral violence. One of the main methodological shortcomings of Fjeldeand Höglund’s paper is that their SCAD-based measure of electoral violence does not distinguish be-tween pre- and post-election violence. In fact, they use ‘country-months’ as their unit of analysis andthey identify instances of electoral violence only by issue, thus completely disregarding the questionof timing. This is problematic given that the dynamics of violence in the pre- and post-election periodare likely to be fundamentally different (see Daxecker, 2014; Hafner-Burton et al., 2014) As mentionedearlier, the use of pre-election violence is primarily aimed at reducing the uncertainty of the electionoutcome, which can realistically be influenced by the type of electoral system in place. In contrast,post-election violence is a response to this outcome, which largely takes place outside the framework offormal electoral competition. Fjelde and Höglund (2015) ignore this important difference between pre-and post-election violence and hence cannot account for the fact that incentives to use violence maychange depending on when it occurs. For example, Hafner-Burton et al. (2014) show that the incen-tives of incumbents to use violence in the pre-election period are primarily determined by the ‘stakes’of the electoral contest15. However, they also show that in the post-election period these incentiveschange. Opposition mobilization in this period is now primarily determined by allegations of fraud inthe pre-election period, whereas the incentives of the incumbent to use force against the opposition areprimarily determined by institutionalized constraints placed upon her executive power. Thus, whileFjelde and Höglund (2015) can make a plausible argument about the impact of majoritarian systemson the likelihood of pre-election violence, they cannot at the same time assume that this ‘institutional’factor influences the likelihood of post-election violence in exactly the same way. Indeed, it seems moreprobable that post-election violence is determined by factors unrelated to the electoral system, giventhat the violent confrontation has now moved outside the domain of regular electoral competition. Bethat as it may, Fjelde and Höglund (2015) simply cannot explore the different dynamics at play duringthe pre- and post-election period due to their restrictive research design. A second shortcoming relatesto their distinction between “government violence” and “opposition violence” (see Fjelde and Höglund,2015; p.11). They use a SCAD-based measure of opposition violence that includes “all election-relatedviolent activities by non-state actors that are ‘anti-government’ or directed towards a distinct ‘other’group” (p.11, emphasis added). The problem with this measure of anti-government electoral violenceis that it includes violent events where the government is clearly not the target. Indeed, some of the“opposition violence” captured by Fjelde and Höglund’s measure is perpetrated by non-state actorsagainst opposition parties and opposition leaders16. The authors would have been well advised toexclude these events from their measure of anti-government electoral violence.

In the last few years, scholars have produced a number of pioneering articles and working papersthat systematically examine pre-election violence from a quantitative perspective. This developmentis undoubtedly due to the recent proliferation of high-quality data sources on electoral violence and‘low-level’ conflict. Unfortunately, the quantitative studies reviewed above have not adequately incor-porated the ‘institutional’ explanations of electoral violence developed in the qualitative literature (seee.g. Höglund, 2009; Opitz et al., 2013). One important exception in this regard is the forthcomingpaper by Fjelde and Höglund (2015). However, this paper only examines the impact of one particu-lar institution (i.e. majoritarian electoral systems) on electoral violence and does not address otherinstitutional factors such as decentralization, the separation of powers or state capacity. This gapin the literature is surprising, given that the more developed literature on conflict management and

15While the ‘stakes’ in Hafner-Burton’s argument are determined by two different measures of electoral ‘closeness’(“Victory Uncertain” and “Polling Unfavourable”), the ‘stakes’ in Fjelde and Höglund’s argument are determined by theelectoral system.

16Other targets of such non-state violence include foreign governments, rival tribes and private companies.

12

‘institutional engineering’ provides empirical evidence suggesting that a number of institutional factorscan have an important impact on the likelihood of civil war onset (e.g. Brancati, 2006; Schneider andWiesehomeier, 2008; Hendrix, 2010; Bogaards, 2013), the level of state repression (e.g. Cingranelli andFilippov, 2010) and the extent of electoral fraud (e.g. Birch, 2007; Hicken, 2007). Furthermore, mostof the above-mentioned quantitative papers have failed to distinguish between violence perpetratedby incumbents and violence perpetrated by ‘challengers’ or other non-state actors (for exceptions seeStraus and Taylor, 2009; Daxecker, 2014; Fjelde and Höglund, 2015). This is problematic because thetwo competing electoral camps are likely to have divergent incentives to resort to electoral violencedepending on a number of situational and structural factors (see e.g. Collier and Vicente, 2012). Inaddition, it seems inappropriate to ignore non-state actors as potential perpetrators of electoral vi-olence (cf. Hafner-Burton et al., 2014), considering that there is still no consensus in the literatureabout who is primarily responsible for electoral violence (see Straus and Taylor, 2009; Daxecker, 2014;p.240). This paper aims to fill these knowledge gaps by systematically examining the impact of awhole range of institutional factors (i.e. electoral system, district size, regime type, state structureand state capacity) on the likelihood of electoral violence. It thus goes beyond the forthcoming articleby Fjelde and Höglund (2015), which narrowly focuses on the violence-inducing effect of majoritarianelectoral systems. This paper also represents the first attempt in the literature to systematically ana-lyze how the impact of ‘institutions’ on the likelihood of electoral violence varies between the pre- andpost-election period and how pro-government and opposition actors respond differently to the sameinstitutional constraints.

Literature on post-election violence

In this sub-section I briefly review the literature on the causes of post-election violence and post-election protests. It is important to keep in mind that this literature is distinct from the fields ofresearch that treat ‘post-election violence’ as either election-related armed conflict or state repression(see Sections 1 and 2 of this Chapter). Here, I only review academic research that examines the causesof post-election violence without relying on the relatively crude country-year time-series datasets onarmed conflict and state repression. In a seminal qualitative contribution to the literature on post-election protests, Tucker (2007) uses a collective action framework to explain the “coloured revolutions”that swept through Serbia, Georgia, Ukraine and Kyrgyzstan in the early 2000s. He argues that majorelectoral fraud can solve the collective action problems faced by potential protesters in authoritariancountries. This is because it exposes the entire population to the same type of abuse and hence providesa “focal point” for action (Tucker, 2007; p.541). One major shortcoming of Tucker’s study is that hedoes not systematically test the hypothesis that electoral fraud increases the likelihood of post-electionprotests. Instead, he relies on a few case studies to illustrate his argument.

In a more recent qualitative paper, Opitz et al. (2013) argue that ‘inclusive’ election monitoring boards(EMBs) can play a decisive role in reducing the likelihood post-election violence. They suggest thatopposition representation in EMBs can prevent ‘sore loser’ protests, which may trigger a tit-for-tatescalation of post-election violence between supporters of the opposition and the incumbent. However,Opitz et al. also do not systematically test their hypothesis and instead draw their conclusions basedon a comparison of electoral processes in Zanzibar, Malawi and Ethiopia. In a recent quantitativeworking paper, Norris (2012) puts forward a similar argument as Tucker (2007). She maintains that thefailure to observe international standards of “electoral integrity” is one of the most important factorscontributing to post-election protests (and electoral violence in general). She also argues that theprobability that electoral manipulation will trigger post-election protests is particularly high in post-authoritarian regimes that have little experience with democratic competition. While her statisticalmodels seem to corroborate the first argument, they offer only mixed support for the second. One ofthe main problems with Norris’s empirical analysis is that her main explanatory variable (“electoralintegrity”) also includes incidences of electoral violence17. This is problematic because she is tryingto establish the extent to which violations of “electoral integrity” impact the likelihood of electoral

17Her measurement of “electoral integrity” includes a variable from the NELDA dataset that indicates whether therewas “evidence that the government harassed the opposition” (Norris, 2012, p.11).

13

violence. Furthermore, Norris does not use the dependent variable that her theoretical argumentseems to require and that she continuously blurs the important conceptual distinction between pre-election violence and post-election protests established in the literature (see e.g. Hafner-Burton et al.,2014; Daxecker, 2014). As Hafner-Burton et al. (2014), Norris relies on “Nelda33” as a measureof electoral violence. Unfortunately, this does not allow her to distinguish between pre- and post-election violence. More importantly however, her theoretical argument seems to require a dependentvariable that measures post-election protests and not electoral violence in general. For example, shehypothesizes “that electoral malpractices can be expected to exacerbate violent protests and coerciveregime reactions” (Norris, 2012; p.9). However, she then includes these two phenomena into herstatistical model as explanatory variables in order to predict the likelihood of electoral violence acrossall election periods (p.25). Norris should have used NELDA’s measure of post-election protest as hermain dependent variable instead of including it in her model as an explanatory variable. The findingthat post-election protests and subsequent state repression are associated with more civilian deaths -her measure of electoral violence - is in itself not very interesting and does not relate to her theoreticalarguments.

Using an ACLED-based measure of post-election violence and original data on election monitoring inAfrica, Daxecker (2012) shows that the presence of reputable international election monitors increasesthe likelihood of post-election violence if the elections in question were fraudulent. She argues thatfraudulent elections monitored by international organizations will be more susceptible to subsequentviolence because an independent third-party can reveal fraud more reliably than domestic institutions.Thus, international election monitors can serve as a trigger for violent contestation of election resultsin the aftermath of fraudulent elections. Daxecker finds support for her theoretical argument ina systematic analysis of post-election conflict events for African elections in the 1997-2009 period.However, one weakness in her analysis is that the conflict events cannot be directly attributed to theelections, given that the ACLED dataset does not include information on the issue(s) at stake duringthe violent event (see Raleigh et al., 2010). In line with Daxecker’s argument, Borzyskowski (2013)maintains that the presence of international monitors and their criticism of fraudulent elections increasethe likelihood of post-election violence. She uses data on electoral violence in Africa gathered by Strausand Taylor (2009) to measure the occurrence and intensity of post-election violence. Borzyskowskicriticizes Daxecker’s research design for assuming that fraudulent elections are always mechanicallytranslated into negative reports if international monitors are present. However, as Kelley has shownin her recent book, even reputable international election monitors are often reluctant to criticizefraudulent elections for fear of flaming post-election violence (see Kelley, 2012). Borzyskowski’s maincontribution to the literature on post-election violence is that she directly measures the impact ofnegative election monitor reports on the level of post-election violence.

Going one step further than Daxecker (2012) and Borzyskowski (2013), Hyde and Marinov (2014)show that the information provided by credible external election observers increases the likelihood andduration of post-election protests when election fraud occurred, and simultaneously discredits ‘soreloser’ protests when the elections were actually free and fair. This study builds upon previous theo-retical work by Svolik and Chernykh (2013), whose formal models suggest that (even biased) electoralcommissions, courts, and international monitors have the capacity to prevent ‘wasteful’ post-electionconflicts by making citizens’ mobilization more ‘accurate’. However, Hyde and Marinov (2014) are thefirst to find empirical evidence for the positive relationship between international election monitors and‘accurate’ post-election protests. Based on their findings, they suggest that international monitors canfacilitate “self-enforcing democracy”, by “increasing incentives for leaders to hold democratic electionsin the long term” (Hyde and Marinov, 2014; p.329). Yet they unfortunately do not test this proposi-tion systematically. It is hence up to future research to establish whether ‘accurate’ protests actuallyincrease the probability of free and fair elections in the long-run (see Daxecker and Schneider, 2014;footnote 26). As was the case with the literature on pre-election violence, the literature on post-electionviolence has so far failed to adequately examine the role of ‘institutions’ in reducing or increasing thelikelihood of violent confrontation. While Opitz et al. (2013) do provide an ‘institutionalist’ accountof post-election violence, their case study approach does not allow them to draw conclusions that arereadily generalizable. The few studies that have examined post-election violence from a quantitative

14

perspective have largely focused on the impact of external election observers and do not address thequestion of how domestic institutions can shape a country’s predisposition to experience post-electionviolence (see Daxecker, 2012; Borzyskowski, 2013; Hyde and Marinov, 2014). My paper aims to fillthis gap in the literature by systematically examining how institutional factors such as the electoralsystem, state structure and state capacity can fuel or discourage violent confrontation in the wake ofelections.

15

Chapter 3

Theoretical Framework

In the previous chapter I argued that the empirical literature on electoral violence has so far failedto adequately examine how ‘institutions’ can shape a country’s predisposition to experience election-related social conflict. But which ‘institutions’ are we talking about? And how can they influence thelevel of electoral violence in a country? In this chapter I provide answers to these two questions. Tobegin with, I define and unpack the concept of ‘institutions’. Thereafter, I show how institutions canconstrain and encourage the behaviour of political actors. Lastly, I develop a number of hypothesesabout the relationship between specific country-level institutional factors and electoral violence.

3.1 What are institutions?

Political scientists and sociologists have come up with a myriad of different ways to define and con-ceptualize ‘institutions’. For example, Hall defines institutions as “the formal rules, compliance proce-dures, and standard operating practices that structure the relationship between individuals in variousunits of the polity and economy” (Hall, 1986; p.19). This relatively broad definition of institutionsencompasses all formalized structures and conventions in society, such as marriage, private propertyor criminal courts. An even broader definition is provided by Norgaard, who defines institutions as“legal arrangements, routines, procedures, conventions, norms, and organizational forms that shapeand inform human interaction”(Norgaard, 1996; p.39). In a similar vein, Helmke and Levitsky defineinstitutions as “rules and procedures, both formal and informal, that structure social interaction byconstraining and enabling actors’ behavior” (Helmke and Levitsky, 2004; p.727). The latter two defi-nitions go beyond Hall’s conceptualization of institutions by taking into account informal institutionssuch as clientelistic networks, traditional rituals (e.g. circumcision) or queuing. In this paper, I adopta broad definition of what constitutes an ‘institution’. However, I am for the most part concerned withformal institutions and their impact on political behaviour. Helmke and Levitsky argue that formaland informal institutions can be conceptually separated from one another by looking at how theyare codified and enforced. They suggest that formal institutions are “rules and procedures that arecreated, communicated, and enforced through channels widely accepted as official ”, whereas informalinstitutions are “usually unwritten [...] and enforced outside of officially sanctioned channels” (Helmkeand Levitsky, 2004; p.727, emphasis added). When political scientists talk about formal institutionsthey often mean state institutions, such as constitutions, courts or bureaucracies (see e.g. Person andTabellini, 2004; Basedau, 2011). But as Helmke and Levitsky rightly point out, many non-state organi-zations such as trade unions, political parties or churches can also be considered as formal and codifiedinstitutions (Helmke and Levitsky, 2004). Nevertheless, the focus of my analysis is on formal stateinstitutions. In particular, I am interested in formalized and officially sanctioned rules that govern thedistribution of power in public office. For example, I examine specific institutional arrangements thataffect the way in which votes translate into parliamentary seats (i.e. electoral systems), that regulatehow decision-making power is divided between the executive and the legislative (i.e. regime types),

16

or that determine the extent to which sub-national units can influence public policy decisions at thenational level (i.e. state structure). All these formal state institutions are what Tsebelis (1990; p.104)refers to as “redistributive” institutions, which benefit one interest group in society at the expense ofanother. They can be contrasted with “efficiency” institutions, such as farmers’ cooperatives, whichhave the potential to improve the welfare of everyone involved. This means that, while all formalstate institutions aggregate conflicting interests within a population into public policy decisions, therelative importance given to specific interest groups can vary according to a country’s institutionalset-up (Person and Tabellini, 2004; p.85).

3.2 How can institutions shape behaviour?

Formal state institutions can either constrain or empower specific political actors. For example, PRelectoral systems with multi-member districts tend to create incentives for politicians to implementsocial policy that benefits broad groups of voters. In contrast, majoritarian electoral systems withsingle-member districts usually result in public policy that only benefits small geographical constituen-cies (Person and Tabellini, 2004). Another example of how formal state institutions can influence thedistribution of power within a country is related to the degree of government centralization. Whilefederal state structures tend to empower regional interest groups vis-à-vis the central authorities, uni-tary state structures can be expected to have the opposite effect on the distribution of power (see e.g.Brancati, 2006; Basedau, 2011). The crucial link between institutions and public policy outcomes isthe behaviour of political actors. Yet, while scholars generally agree that institutional arrangementsaffect outcomes, there is little agreement when it comes to the relationship between institutions (i.e.structure) and the behaviour of individuals (i.e. agency) (see Koelbe, 1995; Hall and Taylor, 1996;Hay and Wincott, 1998). In particular, we are confronted with the difficult question of whether oneof the two is logically prior to the other. Is the behaviour of individuals largely predetermined by theinstitutional context they find themselves in, or do political actors have the capacity to strategicallyshape their institutional context?

We can identify three different ‘institutionalist’ approaches to the structure-agency problem. Accordingto the rational choice approach, institutions can affect the actions of individuals, but ultimately theycannot determine the preferences of individuals. Specific institutional settings are seen as the outcomeof strategic interactions between rational actors that aim to maximize their own utility. However,once these institutions are in place, they have the potential to constrain the choices of individuals (seee.g. Calvert, 1985; Shepsle, 1989; Ostrom, 1991). In contrast, the historical institutionalist approachconceives of institutions as both a causal as well as a ‘caused’ factor. According to this perspective,“structure may indeed be created by agents; but subsequently, like Frankenstein’s monster, structuretakes on a life of its own”, with the potential to shape individuals’ preferences (Aspinwall and Schnei-der, 2000; p.10). Lastly, the sociological institutionalist approach suggests that the preferences ofindividuals are largely pre-determined by their institutional context, which is in turn dependent uponlarger, underlying structural factors such as society or culture (see e.g. Finnemore, 1996). Whereasthe agent-centered rational choice approach to institutions can be traced back to the work of MaxWeber, the more structural approaches of historical and sociological institutionalism are to a largeextent influenced by Emile Durkheim18 (Aspinwall and Schneider, 2000). Yet, despite their differentunderstandings of the relationship between structure and agency, all three institutionalist paradigmsin political science agree that institutions can have an important impact on societal outcomes.

At this point it is necessary to make the distinction between ‘institutional theories’ and ‘theories ofinstitutions’ (see Diermeier and Krehbiel, 2003). Whereas the former try to explain how the behaviourof political actors is influenced by different institutions, the latter try to explain how these institutionscome about in the first place. Explaining the choice and development of different state institutionsacross the African continent (i.e. developing a ‘theory of institutions’) is unfortunately beyond the

18It is worth noting that, despite its putative focus on individual agents and their choices, the ‘rational choice’ approachalso exhibits a form of ‘deep structuralism’, given that it strips individuals of all distinctiveness (see Hay and Wincott,1998).

17

scope of this paper19. Instead, my focus is on identifying institutional factors that can explain variationsin the level of electoral violence between African countries (i.e. developing an ‘institutional theory’of electoral violence). It is evident that state institutions can at the same time shape the behaviourof political actors as well as be shaped by them. However, in my empirical analysis, I assume thatstate institutions are relatively stable structures that are rarely changed in response to election-specificdynamics20. This assumption seems reasonable in light of recent studies, which suggest that changesto state institutions, such as the electoral system, generally only occur during or directly after episodesof exceptional political upheaval (e.g. revolutions, civil wars, or occupations), rather than in responseto electoral outcomes (see Lijphart, 1995; Benoit, 2004; Pospieszna and Schneider, 2013). For example,Benoit (2004) observes that, with the exception of France and Greece, no country in Western Europehas changed its electoral system since the end of the Second World War. The intuition underlyingthese studies is also confirmed by my panel data on state institutions in Africa. For example, out of atotal of the 47 African countries in my sample, only seven ever changed their electoral system in theperiod between 1990 and 2010 (see Chapter 4). The empirical evidence thus allows us to assume thatformal state institutions are relatively stable and long-lasting structures.

This assumption also seems reasonable in light of the aim of ‘institutional theory-building’. In thisrespect, Diermeier and Krehbiel note that “institutional theories often elicit a somewhat misguidedcriticism for assuming that institutional features cannot be altered by actors. The criticism is notempirically misguided because, often, decision-makers can and do change the structural arrangementsunder which they operate. However, the criticism is theoretically misguided inasmuch as it loses sightof the limited aim of institutional theories: structural features must be exogenous when the aim isto learn how and why contextual features affect choice processes. If the researcher wants to identifythe institutional factors that explain a particular pattern of behaviour, the institutional features sim-ply cannot be modeled simultaneously as causes and consequences of that behavior” (Diermeier andKrehbiel, 2003; p.10). Thus, there are also good methodological reasons for treating state institutionsas ‘sticky’ structures. In this paper I furthermore assume that the preferences of political actors areexogenous to the institutional context21. Hence, I adopt an ontological perspective that is perhapsclosest to that of rational choice institutionalism or the more ‘rationally-minded’ historical institution-alism. I do so because the aim of any quantitative analysis of institutions is to uncover generalizablehypotheses about how different institutional settings impact human behaviour. In order to do so, itis a “methodological necessity to hold fixed the behavioral postulate” (Diermeier and Krehbiel, 2003;p.9). Therefore, I assume that political actors are (at least to some extent) interested in gaining andretaining public office and that their behaviour is largely determined by the incentives created by theinstitutional context. Again, it is obvious that some political actors are motivated by ideology oraltruism rather than pure self-interest. However, in my empirical analysis, I adopt a more reductionistinterpretation of the motivations of political actors. With these theoretical issues in mind, we can nowturn to the task of explaining how state institutions affect the likelihood of electoral violence.

19For an analysis of constitutional choice after civil wars, see Pospieszna and Schneider (2013). For an overview of theliterature on electoral system change, see Benoit (2004).

20An instrumental variable approach would be necessary to address the possibility that electoral systems and otherinstitutional factors are endogenous to electoral violence. However, previous studies on armed conflict suggest that formalstate institutions rarely change in response to specific outbreaks of violence and that colonial legacy is a much moreimportant determinant of current institutional arrangements (see e.g. Pospieszna and Schneider, 2013). A more pressingissue for my analysis is whether countries that are particularly prone to experience electoral violence (e.g. because theyhave a history of ethnic strife) are at the same time more likely to have particular institutional arragements. I addressthis problem by controlling for each country’s colonial history, which is the most likely confounding factor in this case(see Blanton et al., 2001).

21In this paper, the term ‘political actor’ usually refers to (either opposition or incumbent) political parties, which areassumed to be unitary actors.

18

3.3 Institutions and electoral violence

State institutions shape the way in which populations are governed in many different ways. Forexample, they can have an impact on social spending priorities (Person and Tabellini, 2004), theprotection of fundamental rights (Davenport, 2007), or the nature of a country’s foreign policy (Schultz,1998; Bueno de Mesquita and Smith, 2012). However, in this paper I am only interested in how stateinstitutions can influence a country’s risk of electoral violence. Indeed, the main aim of this paper is toshow that institutions matter in determining a country’s predisposition to experience electoral violence.In the following section I therefore develop a number of hypotheses about how specific institutionalfactors can be expected to impact the likelihood of electoral violence. To begin with, I explain how acountry’s electoral system as well as its average district size can influence the level of electoral violence.Thereafter, I examine how the degree of government centralization and the separation of state powercan impact the likelihood of electoral violence. Lastly, I move beyond an analysis of formal institutionsby explaining how ‘state capacity’ can influence the level of electoral violence. ‘State capacity’ measuresthe extent to which state institutions are effective and formal rules are implemented on the ground.I believe that it is particularly important to account for the effectiveness of state institutions, giventhat in most developing countries, there still exists a large discrepancy between what formal stateinstitutions are supposed to do and what they actually can do (see e.g. Berman, 1998). Bringing ‘statecapacity’ into the equation thus allows me to provide a more complete picture of the effects of stateinstitutions on electoral violence22.

Electoral systems

Previous studies have shown that a country’s electoral system design significantly influences its risk ofexperiencing civil war23 (e.g. Saideman et al., 2002; Schneider and Wiesehomeier, 2008), the likelihoodof electoral fraud (see Birch, 2007), as well as popular confidence in the electoral process (see Birch,2008). In contrast, the relationship between electoral systems and electoral violence has to date notbeen adequately examined24. In this section, I argue that a country’s electoral system can be expectedto influence the risk of electoral violence. In particular, I hypothesize that majoritarian electoralsystems increase the risk of electoral violence compared to PR systems, given that they significantlyheighten the stakes of the electoral contest. As argued earlier, electoral violence should be conceived ofas the strategic use of violence by unscrupulous politicians and their supporters, aimed at influencingthe election results in their favour. In line with this reasoning, a number of recent studies have shownthat electoral violence can indeed shape voter turnout and precipitate opposition boycotts in favourof the perpetrators (Bratton, 2008; Collier and Vicente, 2008; Hafner-Burton et al., 2012). Thus, ifwe understand electoral violence as strategic violence, rather than as spontaneous outbursts of populardiscontent25, there are two main reasons to believe that elections held under majoritarian electoralrules will be more prone to violence than those held under PR rules.

Firstly, losing elections in majoritarian systems is much more costly than in PR systems. This isbecause in majoritarian systems, even parties with relatively large support bases can end up with littleor no representation in state institutions (Reilly, 2002b). On the flip side, the benefits that successfulpoliticians and parties can expect in terms of decision-making power and access to state resources tend

22The decision to investigate these specific institutional factors was largely based on previous work in the field of conflictmanagement and institutional design (e.g. Basedau, 2011). However, data availability also played a role. For example,Opitz et al. (2013) argue that the institutional design of Electoral Management Boards is likely to have an impact onlevels of post-election violence. Unfortunately, cross-national data on electoral management design are relatively crudeand currently not available in time-series format (see http://www.idea.int/elections/emd/).

23Importantly, there is no scholarly consensus regarding the impact of different electoral systems on the risk of civilwar (see Selway and Templeman, 2012; Bogaards, 2013).

24One important exception is the forthcoming paper by Fjelde and Höglund (2015), which I believe has a number ofmethodological shortcomings (see Chapter 2).

25It is important to note that no all instances of electoral violence are necessarily orchestrated by unscrupulous elites.However, as Fjelde and Höglund argue, “political elites often play a central role in defining the limits and the meaningsof even spontaneous and localized riots or resource conflicts during election periods (Fjelde and Höglund, 2015; p.3).

19

to be higher in majoritarian systems than in PR systems. Indeed, in majoritarian systems the benefitsof resorting to electoral violence are more likely to outweigh the costs incurred (in terms of reputation),compared to PR systems (see Birch, 2007). As a result of the ‘winner-takes-all’ logic of majoritarianelectoral rules, individual politicians thus have much more to gain from risky manipulative strategiessuch as electoral violence (Fjelde and Höglund, 2015). Furthermore, political parties in PR system havea stronger incentive to discipline individual politicians engaging in electoral malpractice, given that“parties in PR systems stand together or fall together, both in terms of reputation and overall levelsof electoral support” (Birch, 2007; p.1538). This stands in stark contrast to the candidate-centeredmajoritarian electoral system, where the benefits and costs of electoral violence can be expected topertain directly to the candidate who incites it, rather than being borne by the party as a whole (ibid.).Secondly, electoral violence is much more efficient under majoritarian rules than under PR rules. Thisis simply because the number of votes that must be altered to change the electoral outcome is typicallysmaller. In majoritarian systems, often a very small margin of votes separates winners from losers andthe overall level of support needed to win a majority can be quite low. For example, in a simple‘first-past-the-post’ electoral system, the ruling party needs only 25 percent of the national vote inorder to remain in power. As long as the ruling party wins 50 percent of the votes in 50 percent of thedistricts, it can receive no votes at all in the remaining voting districts and still end up with a majorityin parliament (see Person and Tabellini, 2004; p.86). This suggests that the number of voters thatneed to be ‘targeted’ by strategic electoral violence is generally lower in majoritarian systems than inPR systems, which in turn makes the recourse to such risky practices more likely (see Birch, 2007).Importantly, the violence-inducing effect of majoritarian voting rules can be expected to apply to thepre-election period as well as the post-election period. Given that majoritarian systems tend to produceall-out winners, the losing side has an increased incentive to resort to violence in the post-electionperiod in order to overturn the proclaimed results through ‘extra-constitutional means’. Similarly, theincumbent party has more to lose in the face of post-election protests demanding the nullification ofthe election results, given its complete control over government power and state resources. This in turnincreases the likelihood of a government crackdown in the post-election period, aimed at defending theproclaimed election results (whether they were legitimate or not).

In mature and stable democracies, the prospect of electoral defeat is usually not a sufficient motive tolead politicians to incite violence against their opponents26. However, in newly established democraciesacross the African continent, electoral defeat can result in a whole community losing access to politicalinfluence and economic resources (Berman, 1998). The reason for this is that in unconsolidated democ-racies, formal state institutions tend to stand in competition with informal institutions that also havean important impact on the nature of political competition (Helmke and Levitsky, 2004; Fjelde andHöglund, 2015). In particular, the two informal institutions of corruption (i.e. the misuse of publicoffice for private gain) and clientelism (i.e. the trading of patronage in return for political loyalty)can significantly increase the costs of electoral defeat and thereby heighten the risk of election-relatedviolence. Both practices are distinct dimensions of what has been termed ‘neo-patrimonial rule’ andcan be considered widespread or even endemic in most African countries (e.g. Van de Walle, 2003;Bratton, 2007). Importantly, the mechanisms through which these two informal institutions increasethe costs of electoral defeat operate at different levels. Corruption translates into higher personalstakes for individual politicians. In the context of neo-patrimonialism, winning elections means gain-ing increased access to state resources, which can be used to buttress one’s own political standingand perhaps the organizational capacity of one’s party (Van de Walle, 2003). The loss of access tostate resources can in turn mean the end of political careers for election losers, given that incumbentswill be at a clear financial advantage. Thus, widespread corruption exacerbates the ‘incumbency bias’that has also been observed in consolidated democracies (see Sundstroem and Waengnerud, 2014), andthereby further increases the stakes of electoral competition.

In contrast, clientelism translates into higher stakes for entire communities. According to Berman, thepractice of clientelism involves “mutual obligations of support and assistance and extends the ties ofkinship and sentiment into the wider structures of [the state]. Wealth and power [of the patron] rests on

26It is important to note that electoral violence was also quite common during the early phases of democratization inEurope and North America (see Rapoport and Weinberg, 2000; Wasserman and Jaggard, 2007).

20

the ability to mobilize and maintain a following of both kin and unrelated dependents. For the [clients]the ‘lopsided friendship’ of clientelism provides access to resources through an indigenous paternalism.”(Berman, 1998; p.325). A number of scholars have pointed out that in most African countries the stateis still conceived of as the main source of patronage, both by politicians and ordinary citizens (Bayart,1993; Berman, 1998; Van de Walle, 2003). In this context, it becomes vital for communities to havetheir patrons in power, because only then will they be able to benefit from state patronage (e.g.infrastructure projects, welfare programs, cash hand-outs, etc.). Thus, clientelism further increasesthe stakes of electoral competition, given that it binds the fate of entire communities to those ofindividual politicians. Importantly, this dynamic should apply irrespective of whether clients actuallybenefit from patronage. In fact, some studies suggest that the benefits that clients can expect fromtheir political patrons are negligible in most African contexts (Van de Walle, 2003). Rather, whatcounts is the perception amongst ordinary citizens that they will lose out if their patron loses accessto political power and state resources. This alone should suffice to heighten the stakes of electoralcontests and enable politicians to mobilize their clients against opposing candidates (see Fjelde andHöglund, 2015).

The above discussion suggests that the informal institutions associated with ‘neo-patrimonialism’ havean important influence on how political actors engage with the electoral system. Both clientelism andcorruption significantly increase the costs of electoral defeat and therefore amplify the violence-inducingeffect of majoritarian electoral rules. Importantly, we can also expect these informal institutions toinfluence the way in political actors engage with other formal ‘redistributive’ institutions, such as acountry’s state structure, the regime type or the separation of powers. Furthermore, we do not needto assume that clientelism and corruption are less prevalent in countries with PR electoral systems. Infact, the empirical evidence on the relationship between electoral systems and corruption is somewhatambiguous (see Person and Tabellini, 2004). However, what is different between majoritarian and PRsystems is that the former effectively exclude opposition actors from access to state resources to beused for patronage and rent-seeking, whereas the latter offer at least some benefits to election losers.As a result of this dynamic, we can expect countries with majoritarian electoral systems to be moreat risk of election-related violence than countries with PR systems.

Electoral systems and ethnicity

A number of scholars have noted that clientelistic networks in Africa are for the most part structuredalong ethnic lines (see e.g. Berman, 1998; Van de Walle, 2003). This means that the fates of individualpoliticians or political parties become closely intertwined with those of entire ethnic communities. Forexample, Murkommen maintains that “in Kenya, certain leaders embody the ideals of their respectivecommunities and that is why they are ‘kingpins’ where they come from. [An] attack on these individualsis construed to be an attack on the larger community” (cited in Hansen, 2012; p.25). Accordingly, anytensions between Kenya’s political leaders can spread directly to their respective ethnic groups, whosee the fate of their ‘kingpins’ as being closely connected to their own (Klopp, 2001). With reference tothe entire African continent, Van de Walle observes that “the single most important factor explainingparty loyalty is ethnicity or region, and ethnic identity provides a remarkably precise prediction ofvoting behaviour” (Van de Walle, 2003; p.305)27. In light of the close connection between politicalloyalty and ethnic identity in African politics, it is therefore important to further investigate how theabove-mentioned violence-inducing effect of majoritarian electoral systems interacts with a country’sethnic composition. The effect of different electoral systems on the likelihood of ethnic conflict istheoretically ambiguous and a contested topic in the peace research literature. While advocates ofPR electoral systems typically highlight the importance of ethnic minority representation for conflictprevention (Lijphart, 1977; Reilly et al., 1999; Reynal-Querol, 2002; Lijphart, 2004; Schneider and

27However, this does not mean that there is something immutable (or primordial) about the close relationship betweenethnicity and political loyalty in Africa. Rather, the continued salience of ethnicity in African party politics suggeststhat politicians still successfully use ethnic competition as a means to mobilize voters. In line with this reasoning, arecent cross-national study of African elections finds that individuals are much more likely to identify in ethnic termsduring electoral campaigns, compared to when elections are in the distant future (see Eifert et al., 2010).

21

Wiesehomeier, 2008), their critics argue that the institutional affirmation of group differences invitesethnic competition and increases the risk of inter-ethnic conflict (Horowitz, 1985; Lardeyret, 1991;Horowitz, 2003). The latter group of scholars thus recommends the adoption of majoritarian systemsrather than PR systems, as they are thought to encourage vote-pooling and inter-ethnic coalition-building28. However, proponents of both camps have almost exclusively relied on anecdotal evidenceto support their respective arguments and the few existing systematic, cross-national studies have sofar come up with rather inconclusive results regarding the impact of different electoral systems onethnic conflict (see Selway and Templeman, 2012; Bogaards, 2013; Pospieszna and Schneider, 2013).

In this paper I adopt a third perspective, given that I am not interested in the drivers of ethnic conflictper se, but rather in the question of how a country’s ethnic composition can mediate the effect ofelectoral rules on election-related violence. I argue that the violence-inducing effect of majoritarianelectoral systems is likely to be particularly strong in societies that are polarized along ethnic lines.Conversely, the effect is likely to be less pronounced in societies that are highly fractionalized29. Thereasoning behind these expectations is simple: In countries with majoritarian electoral systems, wheretwo equally large ethnic groups compete for power, politicians have an incentive to mobilize supportersbased on their shared ethnicity. This is because both camps stand a roughly equal chance of gainingaccess to political power and state resources, without support from members of the other ethnic group.Furthermore, electoral violence targeting the competing ethnic group is likely to be relatively efficient,given that opposition supporters are easier to identify, and inexpensive in terms of reputation, giventhat the violence only targets a distant ‘Other’ (see Brubaker and Laitin, 1998; p.441ff). Lastly, wecan expect that ethnic polarization exacerbates the effect of majoritarian rules because electoral defeatis likely to lead to the exclusion of an entire ethnic community from political influence. This providesunscrupulous politicians with a powerful narrative with which supporters can be mobilized and incitedto engage in electoral violence.

The violence-inducing effect of majoritarian systems is likely to be less pronounced in societies thatare highly fractionalized along ethnic lines. In these contexts, politicians usually need the supportof a number of ethnic groups in order to obtain a majority in parliament. The language of ethniccompetition is therefore less attractive as a tool to mobilize support amongst one’s own ethnic group,given that it might scare off members of other ethnic groups. In addition, the use of electoral violenceis bound to be less efficient than when only one opposition group needs to be targeted and more costlyin terms of reputation. All other things being equal, we can thus expect majoritarian systems to be lessviolence-inducing in societies that are ethnically fractionalized. In contrast to majoritarian systems,we can expect PR systems to be relatively accommodating to different forms of ethnic composition.This is because each ethnic group will be able to gain access to political influence and state resourcesin proportion to its size. When it comes to the likelihood of electoral violence, the level of ethnicfractionalization and polarization in countries with PR systems should therefore not matter as much asin countries with majoritarian systems. Importantly, there are a number of methodological limitationsassociated with the measures of ethnic division employed in this paper. To begin with, the staticmeasures of ethnic fractionalization and polarization do not tell us anything about the distribution ofpolitical power amongst different ethnic groups at a given point in time. For example, in a settingwith one numerically dominant ethnic group and one minority ethnic group, these measures simplycannot tell us whether the minority group is in power (as in apartheid South Africa) or whether itis without any meaningful political representation (as the Tuareg in Mali) (Cederman et al., 2010b;Basedau, 2011). Furthermore, the two measures also cannot capture the regional distribution of ethnicgroups within a country, given that they aggregate information at the country-level. However, insocieties where ethnic groups are concentrated in particular districts and constitute local majorities,

28In a recent study, Huber firmly rejects the proposition that PR systems are more likely to politicize ethnicitythan majoritarian systems. In fact, his empirical findings suggest the exact opposite. He shows that, in contrast tomajoritarian systems, PR electoral systems are much more conducive to the creation of political parties “that appeal onbases other than ethnic identity, with the result being that voters from the same group often divide their support acrossa number of parties, often non-ethnic ones” (Huber, 2012; p.1000).

29My measure of ‘ethnic fractionalization’ reflects the probability that two randomly selected people from a givencountry will belong to two distinct ethnic or ‘ethnoreligious’ groups, whereas my measure of ‘ethnic polarization’ indicateswhether a society is polarized between two large ethnic groups or not.

22

the effect of majoritarian electoral rules on the disproportionality in political representation can besignificantly reduced (Wagner and Dreef, 2014). Lastly, given that the two measures are static, theycannot capture changes in the ethnic composition of countries over time. For example, they do notaccount for large structural changes in the ethnic make-up of societies induced by phenomena suchas migration or genocide, but rather ‘freeze’ ethnic divisions at a given point in time (Basedau, 2011;p.5).

District size

The average district magnitude of a country indicates the number of representatives that can be electedto the parliamentary assembly by each constituency. On one end of the spectrum, all representativesare elected in single-member districts, as is the case in elections to the House of Representatives in theUSA. On the other end of the spectrum, all representatives are elected in a single, country-wide district,as is the case in elections to the Israeli Knesset (Person and Tabellini, 2004; p.78). While countrieswith ‘first-past-the-post’ electoral rules tend to also have single-member districts and countries withPR systems usually have multi-member districts, I nevertheless analyze these two institutional factorsseparately. This is because there is a large but not perfect correlation between these two dimensionsof the voting system across countries (see Blais and Massicotte, 1997; Person and Tabellini, 2004).In this paper I argue that a country’s district size is an important factor determining the risk ofelectoral violence. In particular, I suggest that an increase in the average district size will reducethe likelihood of electoral violence. This hypothesis is based on the fact that large districts providemore opportunities for minority groups to obtain a certain level of representation and to gain accessto state resources. Single-member districts, in contrast, only allow one representative to be electedfrom a given area and thus create a ‘winner-takes-all’ logic that is likely to increase the risk of violentconfrontation between competing camps. Again, I expect this dynamic to apply not only to violencein the pre-election period, but also to violence in the post-election period.

State structure

Whether a country has a unitary or a federal state structure is likely to be an important determinantof electoral violence. Whereas unitary states concentrate political power and fiscal resources at thenational level, federal states accord more decision-making power to sub-national and regional politicalunits. Previous studies in the field of peace research have shown that federal states are less proneto armed conflict than their unitary counterparts, given that they allow for better representation ofethnic and regional minorities (see e.g. Saideman et al., 2002; Schneider and Wiesehomeier, 2008).However, others have cast some doubt on this positive assessment of decentralization, by showingthat it can be associated with a decline in good governance (Gerring et al., 2005) and a growth ofregional parties, which in turn increases the risk of violent conflict (Brancati, 2006). In this paperI side with the more optimistic school of thought, given that I expect decentralization to reduce thestakes of national elections, which in turn should decrease the risk of electoral violence. In particular,I argue that unitary state structures increase the likelihood of election-related violence in comparisonto federal state structures, given that a larger share of the state’s resources and more political powerare at stake during national elections30.

30Since I only have data on national elections, I cannot investigate whether decentralization increases the risk ofviolence associated with regional elections.

23

Regime types

Previous studies have identified a country’s regime type as an important institutional factor influencingthe risk of violent conflict (Schneider and Wiesehomeier, 2008)31, the quality of governance (Gerringet al., 2008), and social spending priorities (Person and Tabellini, 2004). I argue that whether a countryhas a presidential, a semi-parliamentary, or a parliamentary regime also has an impact on the likelihoodof election-related violence. In purely presidential regimes, citizens directly elect the country’s chiefexecutive, i.e. the head of government. In contrast, in parliamentary regimes this power is typicallybestowed upon the national parliament, which is in turn elected by the country’s citizens (Personand Tabellini, 2004). Importantly, there are a number of ‘mixed’ regimes that combine features ofthe presidential and the parliamentary system (see e.g. Gerring et al., 2008; p.337). Following Becket al. (2001), I therefore use a third category (‘semi-parliamentary regimes’) to capture countries inthe mid-range between pure presidentialism and pure parliamentarism32. In countries with ‘semi-parliamentary’ regimes, the chief executive is also elected by the parliament. However, the head ofgovernment enjoys a higher degree of independence than in pure parliamentary systems, given thatthe legislature needs a two-third majority or a dissolution in order to remove the chief executive (seeBeck et al., 2001).

I expect countries with presidential regimes to be more at risk of electoral violence than countrieswith parliamentary or semi-parliamentary regimes. The reasoning behind this expectation is thatpresidential regimes significantly heighten the stakes of electoral contests (especially those for thechief executive) in comparison to parliamentary regimes. In line with this reasoning, Basedau (2011;p.17) suggests that “among other ‘perils of presidentialism’ [...], the winner-takes-all logic may resultin the marginalization of ethnic groups, thus fostering violent reactions by the losing group. Hence,parliamentary systems may be more suitable for avoiding ethnic conflict”. With reference to the Africancontinent, Van de Walle (2003; p.310) also observes that in presidential regimes “throughout the region,power is highly centralized around the president. He is literally above the law, controls in many casesa large proportion of state finance with little accountability, and delegates remarkably little of hisauthority on important matters. In most [African] countries, the presidency emerges as the dominantarena for decision-making”. Thus, a direct implication of this centralization of power in presidentialregimes is that access to state patronage and clientelistic benefits is also highly concentrated aroundthe presidential office. As a result, political competition in presidential systems is much more likelyto be viewed as a zero-sum game compared to parliamentary systems. Furthermore, the fact that inpresidential regimes the chief executive is relatively unrestrained by the other branches of governmentcan be said to increase the likelihood of government violence aimed at intimidating opposition actorsduring the election period (Hafner-Burton et al., 2014). In sum, we can thus expect a higher level ofelectoral violence in countries with presidential regimes.

State capacity

Lastly, I argue that ‘state capacity’ is an important factor influencing the risk of electoral violencein a given country. This argument builds on a large body of peace research literature, which hasidentified ‘state capacity’ as a crucial structural driver of conflict onset, duration and intensity (seee.g. Fearon and Laitin, 2003; DeRouen and Sobek, 2004; Lacina, 2006; Gates et al., 2006; Fjelde andDe Soysa, 2009; Hendrix, 2010; Sobek, 2010; Hanson and Sigman, 2013). Strictly speaking the term‘state capacity’ does not refer to a formal institution (such as the electoral system), but rather tothe effectiveness of a specific ‘meta-institution’ which we call the state. Like no other institution, thestate is equipped to regulate all of the administrative, legal, bureaucratic, and coercive systems thatstructure social relations within its territory (Evans et al., 1985; p.7). The state is thus considered

31In contrast, Brancati (2006) finds that the effect of regime type on the likelihood of ethnic conflict is not statisticallysignificant.

32One could also further sub-divide presidential regimes into semi-presidential and fully presidential regimes (seeGerring et al., 2008). However, for the sake of simplicity, I simply adopted the three-fold categorization of regime typesused by the World Bank’s Database of Political Institutions (Beck et al., 2001).

24

to be a so-called ‘meta-institution’, which determines the way in which sub-ordinate institutions areenforced and implemented (Zierhofer, 2005). My measure of ‘state capacity’ indicates the extent towhich state institutions are effective and formal rules are implemented on the ground. As mentionedearlier, I think that it is particularly important to account for the effectiveness of state institutions,given that in most developing countries, there still exists a large discrepancy between what formalstate institutions are supposed to do and what they actually can do (see e.g. Berman, 1998). Politicalscientists have defined and operationalized the term ‘state capacity’ in a number of different ways.

Broadly speaking, we can identify three distinct approaches to the concept of ‘state capacity’ (seeHendrix, 2010). The first approach defines state capacity simply in terms of a state’s coercive power.This conceptualization can be traced back to Max Weber’s famous definition of the state as the soleorganization that can successfully claim the monopoly of the legitimate use of force within a giventerritory (see Weber, 1958). Scholars in this school of thought tend to focus on the state’s militarycapacity and argue that stronger militaries will decrease the likelihood of civil unrest in a country(Mason et al., 1999; Fearon and Laitin, 2003; Balch-Lindsay et al., 2008; Buhaug, 2010). State capacityis thus mostly operationalized in terms of military personnel per capita or military spending per capita(see Hendrix, 2010). However, this operationalization of state capacity has several shortcomings inrelation to the study of electoral violence. Firstly, it is not at all evident that a strong military isnecessary to control a country’s domestic population. Indeed, there are a number of countries in theworld that maintain a high degree of internal order without having to rely on armed forces (e.g. CostaRica or Iceland33). More plausibly, it is the strength of the police services and the state bureaucracy,rather than the military, that are key in preventing civil unrest (see Fjelde and De Soysa, 2009; p.10-11). Hence, while a country’s military capacity surely matters in times of war, it does not necessarilyinfluence the state’s ability to suppress domestic opposition or organize peaceful elections. Secondly, alarge military budget might actually be the result of low state capacity. For example, weak states thatare unsuccessful at deterring challenges to their authority might invest heavily in military capacity,given that they can expect to fight more wars than strong states (Hendrix, 2010; p.277). Similarly, anincrease in military spending might reflect a country’s geo-strategic insecurity (e.g. hostile neighbors),rather than its capacity to control the domestic population. Lastly, large military expenditure indeveloping countries might be the result of corruption and thus stand in direct opposition to capablestate institutions (see Gupta et al., 2001). In sum, the approach to state capacity that focuses solely ona country’s coercive capacity has significant shortcomings in relation to the study of electoral violence.

The second approach conceptualizes state capacity as institutional coherence. Scholars in this school ofthought typically focus on the degree to which countries combine democratic and authoritarian regimecharacteristics (see e.g. Hegre et al., 2001; Reynal-Querol, 2002; Gates et al., 2006). They argue thatcountries with mixed regimes, that have both democratic and authoritarian traits, are least capable ofaverting social unrest, given that they combine an “inadequate capacity for repression with insufficientability to accommodate opposition through institutionalized channels” (Fjelde and De Soysa, 2009;p.6). Most scholars who subscribe to this line of reasoning use some measure of ‘democraticness’ basedon the Polity dataset to capture a country’s ‘state capacity’ (see Marshall and Jaggers, 2002). Again,I think that this approach to state capacity is flawed for a number of methodological reasons. Tobegin with, the ‘institutional coherence’ approach conflates the two concepts of regime type and statecapacity and thus makes it impossible to separately examine their respective impact on the likelihoodof conflict (Hanson and Sigman, 2013). Secondly, there is little systematic evidence which suggests that‘institutional coherence’ is a good indicator of both the repressive capacity and the ‘accommodative’capacity of states, which puts the causal argument of this approach on a shaky empirical ground(Fjelde and De Soysa, 2009; Hendrix, 2010). Lastly, any investigation into the relationship betweenstate capacity and conflict that relies on the Polity index to capture ‘institutional coherence’, is likely tobe somewhat tautological. This is because countries are coded in the middle-range of the Polity index(i.e. as ‘institutionally incoherent’) precisely when “competition between groups is intense, hostile, andfrequently violent” (Vreeland, 2008; p.402). In other words, the purported relationship between lowstate capacity and conflict may simply be definitional.

33See https://www.cia.gov/library/publications/the-world-factbook/fields/2055.html#cw [accessed 10.02.15].

25

In light of these methodological shortcomings, I believe that the third approach to state capacity isthe most appropriate in relation to the study of electoral violence. This approach focuses on theadministrative or bureaucratic capacity of the state. Besides being insulated from political pressures,a capable state bureaucracy requires “technical competence, trusted and professional state agents,monitoring and coordination mechanisms, and effective reach across the state’s territory and socialgroupings” (Hanson and Sigman, 2013; p.4). Defined in this way, the concept of state capacity thuscaptures the extent to which state institutions can effectively implement official goals (see Sikkink,1991). I believe that this conceptualization of state capacity has distinct advantages over the twocompeting approaches. Firstly, the pacifying effect of states with strong administrative capacities isless contested than the relationship between military capacity or ‘institutional coherence’ and socialconflict. Indeed, Hendrix (2010; p.277) notes that “there is broad agreement that countries with morecapable bureaucracies experience fewer civil conflicts”. Secondly, the conceptualization of state capacityas administrative capacity is indifferent towards a country’s regime (e.g. democratic, autocratic, ortheocratic) and thus enables us to disentangle the impact of different forms of government (on thelikelihood of conflict) from different degrees of government (see Huntington, 1968). Lastly, the focus onadministrative capacity also prevents the concept of state capacity from being diluted with ideologicalpreferences and normative ideas about how societies are best governed (Hanson and Sigman, 2013).

In this paper I use a measure of state capacity that is based on expert assessments of a country’squality of government (see Chapter 4, Section 2). I believe that this is the most appropriate way tooperationalize the concept of administrative capacity given the shortcomings of existing alternativemeasures. For example, Fearon and Laitin (2003) use GDP per capita as a proxy for a country’sadministrative capacity and argue that higher GDP per capita reduces the likelihood of conflict byincreasing the state’s ability to monitor its population. I believe that this approach is questionablebecause the two concepts of GDP per capita and administrative capacity do not necessarily overlap.Gabon, for example, has a relatively high GDP per capita, but excessive corruption and rent-seekingmeans that the country’s wealth has not translated into a capable state bureaucracy34. In addition,low GDP per capita may be associated with civil conflict via causal channels that have nothing todo with the state’s administrative capacity (Hendrix, 2010). Collier and Hoeffler (2004), for example,provide a competing causal explanation and argue that low GDP per capita reduces the perceivedopportunity costs associated with participation in armed rebellion. Other measures that have beenused to capture the administrative capacity of states include the ratio of tax revenue to GDP, theratio of total government revenue to GDP, as well as the ratio of actual tax revenue to expected taxrevenue given a country’s GDP per capita, mineral exports and other factors (see Arbetman-Rabinowitzand Johnson, 2007; Hendrix, 2010; Thies, 2010). However, all these relatively objective measures ofadministrative capacity suffer from the problem that revenue generation is likely to be endogenous toconflict (see Thies, 2010). Furthermore, the lack of reliable data on taxes and government revenuesfor large parts of the African continent means that these measures are inappropriate for the purposeof my analysis. In sum, we can thus say that a measure of state capacity based on expert assessmentsof a country’s quality of government presents the most suitable option35.

I argue that countries with strong and capable state bureaucracies are less at risk of experiencingelectoral violence than countries with low administrative capabilities. This is because countries witha capable state apparatus are better equipped to address the grievances of their populations throughpeaceful mechanisms, which reduces the risk of oppositional violence during the election period. Inline with this reasoning, Gurr observes that “civil violence is most likely to manifest in societies thatrely on coercion to maintain order in lieu of providing adequate patterns of value-satisfying action”(Gurr, 1970; p.317). Similarly, Sobek suggests that “citizens in strong states are more able to havetheir grievances ameliorated or, at the very least, a strong state may be able to limit any escalationof the dissent” (Sobek, 2010; p.267). According to Sobek, strong and capable state bureaucracies arethus not only able to co-opt oppositional violence by satisfying the demands of the population, but

34See http://www.theguardian.com/global-development/2013/sep/06/gabon-fake-civil-servants-receiving-salaries [ac-cessed 10.02.15].

35The biggest threat to the validity of such survey-based measures of state capacity is endogeneity (i.e. the possibilitythat recent violent events in a given country influenced the expert’s assessment of that country’s state capacity). However,in Chapter 4, Section 2 I offer a (partial) solution to this problem.

26

they also successfully deter oppositional violence by presenting a less “tempting target” for challengersto state authority (ibid., p.269). We can also expect that countries with capable and independentadministrative structures will experience less pro-government violence during the election period. Thisis because state bureaucracies (in particular police forces) that are sufficiently insulated from politicalinterference are less likely to be used by the incumbent to obstruct and intimidate opposition actors.In sum, we can expect that the risk of electoral violence, both by pro- and anti-government actors,will be lower in countries with high levels of administrative capacity.

Importantly, I expect this violence-reducing effect to be most relevant in the post-election period.There are two principal reasons to believe that strong and independent state administrations can sig-nificantly decrease the risk of violence in the post-election period. Firstly, in weak states, abrupt powerchanges associated with elections can result in a complete breakdown of government services and thuslead to a general state of insecurity and lawlessness. One of the main reasons for this dynamic is that inweak states, the fate of incumbents and state bureaucrats tends to be closely intertwined as the resultof widespread clientelism and uncompetitive recruitment practices (see Lindberg, 2008). In contrast,countries with capable administrations are generally more immune to changes and interruptions ingovernmental authority that occur in the immediate aftermath of elections. All other things beingequal, strong state bureaucracies should therefore reduce the risk of post-election violence, by contin-uing to enforce the law and providing basic services in times of rapid political power-shifts. Secondly,strong states are better equipped to enforce the proclaimed results of the elections. This is becausethe overall level of bureaucratic quality is also likely to influence the legitimacy of the electoral results,which in turn impacts the likelihood of post-election protests against the proclaimed results. In linewith this reasoning, Opitz et al. suggest that “distrust in the government to run fair elections oftencoexists with the lack of institutional capacity to conduct electoral administration in a satisfying way,particularly in an environment of general resource scarcity. Technical problems may often convergeinto political ones, and reactions often break out at the intersection between political suspicion andtechnical incapacity” (Opitz et al., 2013; p.715). Hence, competent state bureaucracies can reducethe risk of post-election violence by increasing the legitimacy of the proclaimed election results. Inparticular, independent electoral management bodies, recruited on a meritocratic basis, can be saidto decrease the likelihood of post-election protests aimed at overturning the ‘official’ election results(see e.g. Lehoucq, 2002; Reilly, 2002a; Birch, 2008; Opitz et al., 2013)36. This is because indepen-dent electoral management bodies are perceived to be less susceptible political interference, which inturn makes it more difficult to undermine their legitimacy by portraying them as partisan or biased.Overall, we can expect countries with higher levels of state capacity to be less at risk of experiencingpost-election violence.

36Importantly, Opitz et al. (2013) argue that it is the inclusiveness rather than the formal independence of electoralmanagement bodies that matters most in terms of reducing the risk of post-election violence. In contrast, Reilly (2002a)recommends formally independent electoral management bodies, staffed by non-partisan civil servants, as the mostappropriate institutional design. Unfortunately, both camps base their recommendations on individual case studies orsmall-N comparisons. To date, there is no systematic evidence available to judge which option is most effective atreducing the risk of electoral violence.

27

Chapter 4

Data

In order to systematically assess the empirical implications of my arguments, I use a cross-nationaltime-series dataset on electoral violence in Africa for the period 1990-2010. African elections presenta suitable sample to test my arguments for several reasons. Firstly, most African countries introducedregular national elections at about the same time, i.e. after the end of the Cold War in the early1990s (Shin, 1994; Daxecker, 2014). This means that most countries in my sample will have had asimilar amount of time to ‘get used to’ democratic practices and that variations in the level of electoralviolence will not be driven by the fact that some countries simply have more experience in organizingelections than others37. Secondly, my theoretical arguments imply a focus on unconsolidated regimes,given that in mature democracies (e.g. in Europe or North America) electoral defeat is generally nota sufficient motive for politicians to incite electoral violence (see Fjelde and Höglund, 2015). Thirdly,countries in Africa also exhibit significant variation in both the level of electoral violence as well astheir institutional design. Lastly, disaggregated datasets on ‘low-level’ political violence are currentlyonly available for Africa. Datasets with global reach unfortunately lack sufficient detail to attributeviolent events to specific electoral contests or rely on crude binary indicators to measure electoralviolence (see Hafner-Burton et al., 2014).

4.1 Measuring electoral violence

I use information from the National Elections Across Democracy and Autocracy Dataset (NELDAVersion 3) to identify all elections in Africa between 1990 and 2010 (Hyde and Marinov, 2012b).NELDA provides detailed information on all national-level executive and legislative elections wherevoters “directly elect the person or persons appearing on the ballot” (Hyde and Marinov, 2012a; p.1).Importantly, elections do not need to be ‘competitive’ in order to be included in the dataset. NELDArecords elections for all countries with a population above five hundred thousand inhabitants based onthe List of Independent States by Gleditsch and Ward (1999). It codes election rounds rather thanelections. Hence, a given country can have more than one election event per year. If a country’sexecutive and legislative elections took place on the same day, I collapsed them into one election roundin order to avoid double-counting violent events in the corresponding pre- and post-election periods.Based on these criteria, my dataset includes 47 countries from both Northern and Sub-Saharan Africa,with a total of 353 election rounds. Most elections (80%) have only one round and the maximumnumber of election rounds per country-year in my sample is 4, which only occurred in Gabon in 1990.

To measure the level of pre- and post-election violence per election round, I use data from the SocialConflict in Africa Database (SCAD Version 3). This highly disaggregated, events-based dataset con-tains information on violent and non-violent social disturbances such as protests, strikes and riots in

37In my regression models I nevertheless control for the number of election rounds per country between 1990 and 2010(see Section 4.3)

28

Africa since 1990 (Salehyan et al., 2012). SCAD codes the onset of conflict events based on Englishand French language newswires compiled by Lexis-Nexis38. The advantage of SCAD over other events-based conflict datasets such as ACLED (Raleigh et al., 2010) or the UCDP Geo-referenced Dataset(Sundberg and Melander, 2013) is that each conflict event is identifiable by issue-type (one of whichis “elections”) and hence directly attributable to a specific electoral contest. Furthermore, the scopeof the SCAD dataset is broad enough to capture conflict events that did not result in deaths. This isimportant given the ‘low-intensity’ nature of most electoral violence (see Lindberg, 2008; Goldsmith,2014). In addition, SCAD codes the initiator and target of each conflict event and thus enables me todistinguish between government and opposition electoral violence. My unit of analysis is the pre- andpost-election period respectively. This allows me to investigate the different violent dynamics at playbefore and after elections39. I define the pre-election period as 180 days before election day and thepost-election period as 90 days after election day40. The level of pre-election violence for each electionround is measured by the number of election-related conflict onsets during the relevant pre-electionperiod41. I apply the same procedure to the post-election period to measure the level of post-electionviolence. In contrast to the measures used by Daxecker (2012) and Fjelde and Höglund (2015), mymeasure thus identifies instances of electoral violence by both issue and timing. On the one hand, theissue-based identification allows me to exclude conflict events that are unrelated to electoral violence,such as civil wars, gang violence or terrorist attacks. On the other hand, the time-based identificationenables me to examine the dynamics of violence in the pre- and post-election period separately.

My data suggest that electoral violence is a relatively common phenomenon across Africa. Roughly onequarter of all elections between 1990 and 2010 experienced some form of pre-election violence, whileabout 13 percent of all elections experienced some form of post-election violence. Figure 1 displays thetotal number of election-related violent events per country that occurred between 1990 and 2010. Wecan see that Zimbabwe, Nigeria, Côte d’Ivoire and Egypt are outlier countries in terms of overall levelsof electoral violence. However, it is important to keep in mind that this map does not take into accountthe number of elections held in each country. Hence, it puts countries with many elections between1990 and 2010 at a ‘disadvantage’, given that they were also more ‘exposed’ to the threat of electoralviolence. Figure 2 displays the average number of violent events per election round that each countryexperienced during the time-period in question. It shows that elections in the Democratic Republicof Congo (DRC) and Kenya are as (if not more) violent than elections in the three above-mentionedcountries, if one takes the frequency of electoral contests into account.

38The focus on conflict onsets (rather than incidence) means that violent events that lasted longer than one day arenevertheless only coded once. One shortcoming of this measure is that it fails to capture differences in the intensity andduration violent events. For example, while most violent events in my dataset resulted in less than 10 deaths, some eventsled to more than 1000 deaths (e.g. Kenya in 2007). However, only 10% of all events in my dataset lasted longer than twodays. It is also important to note that any coding that relies on media reports of conflict events rather than on actualevents is susceptible to information bias, given that some countries (and elections) receive more international attentionthan others (Eck, 2012; Clark and Sikkink, 2013; Daxecker, 2014). For more information on the SCAD coding proceduressee Salehyan et al. (2012) or the SCAD Codebook, which is available at: https://www.strausscenter.org/scad.html.

39This stands in contrast to the research design of Fjelde and Höglund (2015), who use ‘country-months’ as their unitof analysis.

40This timeframe is of course somewhat arbitrary. However, a preliminary analysis of the data suggested that roughly70% of all election-related SCAD events fall within 180 days before and 90 days after election day. Previous workby Goldsmith (2014) also suggests that most electoral violence in Africa takes place within this timeframe. To avoidexcessive double-counting of events for elections with multiple rounds, the pre- and post-election periods can only extendto the previous (or next) election round. Elections with little time in between successive rounds therefore sometimeshave shorter pre- and post-election periods than the ‘standard’ election. Despite this precautionary measure, 37 conflictevents were double-counted, given that they occurred where the pre- and post-election periods of two successive electionrounds overlapped. In these few cases it was impossible to attribute the event to one of the two election rounds.

41Conflict events were coded as “election-related” if SCAD explicitly mentioned “elections” as the first, second or thirdsource of disorder or issue at stake. In addition, I manually reviewed all conflict events that SCAD coded as relating to“human rights & democracy”. In this category the SCAD event descriptions were searched for keywords (“elect”, “oppo”,“candid”, “poll”, “vot”, “campaign”, “ballot”, “race”, “referend”, “incumb”, “challen”) in order to identify additional conflictevents that were clearly related to elections. In total, my dataset is made up of 585 unique election-related conflictevents, each attributable to a specific pre- or post-election period.

29

≤ 56 - 1011 - 1516 - 2021 - 30No data

Figure 1: Total number of election-related vi-olent events per country since 1990

≤ 0.50.6 - 11.1 - 22.1 - 44.1 - 6No data

Figure 2: Average number of election-relatedviolent events per election round

My measure of electoral violence also distinguishes between Government Violence and OppositionViolence. Several scholars have highlighted how the incentives to resort to electoral violence can differbetween the two competing camps, depending on a number of situational and structural factors (seee.g. Straus and Taylor, 2009; Collier and Vicente, 2012; Fjelde and Höglund, 2015). My measureof government violence includes all election-related violent events that were initiated by governmentand pro-government actors42. My measure of opposition violence in turn includes all election-relatedviolent events perpetrated by non-state actors that specifically targeted the government or governmentsupporters. In contrast to Fjelde and Höglund (2015), I did not automatically include all ‘violent riot’SCAD events in the category of ‘opposition violence’, given that some of this violence was clearlyperpetrated by non-state actors against opposition parties and opposition candidates43. Instead, Imanually attributed all ‘violent riot’ events to either category, based on whether the primary target wasthe government or the opposition44. The categorization of events by election period as well as by targetand perpetrator allows me to investigate the determinants of four different ‘types’ of electoral violence:(1) government pre-election violence, (2) opposition pre-election violence, (3) government post-electionviolence and (4) opposition post-election violence. Table 1 displays the summary statistics of these fourdependent variables. According to my data, pro-government actors are responsible for the majority ofpre- and post-election violence across Africa between 1990 and 2010. This finding is in line with recentstudies by Straus and Taylor (2009) and Fjelde and Höglund (2015), but contrasts with evidence putforward by Collier and Vicente (2012) and Daxecker (2014), who suggest that ‘challengers’ or non-stateactors are the main perpetrators of electoral violence.

42This measure of Government Violence does not include anti-government events that were subsequently repressedby government forces. However, it does include ‘violent riot’ events that were perpetrated by non-state actors againstopposition parties and candidates.

43For example, this category includes violent riots by supporters of President Kibaki targeting opposition supportersin the aftermath of the 2007 elections in Kenya. As mentioned earlier, other targets of such non-state violence includeforeign governments, rival tribes and private companies.

44In line with Fjelde and Höglund (2015), my measure of ‘opposition violence’ does not include anti-governmentviolence waged by actors with semi-permanent or permanent militant wings in order to exclude violence related toongoing civil wars.

30

Table 1: Number of election-related violent events per electoral period

Variable Observations Mean Standard Deviation Min. Max.

Government pre-election violence 353 .227 .648 0 4

Opposition pre-election violence 353 .178 .537 0 5

Government post-election violence 353 .113 .475 0 6

Opposition post-election violence 353 .074 .282 0 2

4.2 Measuring institutions

The data for my institutional explanatory variables come from a number of different sources. Theinformation on each country’s electoral system was taken from the updated version of the Institutionsand Elections Project (IAEP Version 2) dataset (Wig et al., 2014a). Based on this information I codedan electoral system variable, which indicates whether a country’s legislative elections were organizedaccording to (1) Majoritarian45, (2) Mixed, or (3) Proportional Representation (PR) rules, with thelatter being the reference category46. It is important to keep in mind that this threefold measureof electoral systems hides important differences within each category. For example, mixed electoralsystems can differ significantly amongst one another with regard to how many legislative seats arefilled according to majoritarian or PR rules (see Reynolds et al., 2005). My data suggests that roughly50% of African election between 1990 and 2010 were held according to majoritarian electoral rules.PR systems were operational 30% of the time and mixed systems make up 20% of the overall sample.Figure 3 shows the distribution of these three different electoral systems across the African continentas of 201047.

Data on each country’s average district size was taken from the Database of Political Institutions(DPI) hosted by the World Bank (Beck et al., 2001). The District Size variable captures the weightedaverage of the number of representatives elected to a country’s lower house by each constituency size48.Following Schneider and Wiesehomeier (2008), Huber (2012) and Fjelde and Höglund (2015), I log-transformed this measure because of its high skewness and the theoretical expectation that beyond acertain threshold, an increase in district size should have no further violence-reducing effect. Strictlyspeaking, a country’s average district size simply represents another dimension of its voting system;the other two being the ballot structure and the electoral formula49 (Person and Tabellini, 2004).Nevertheless, I model the effect of a country’s electoral rules and average district size separately, giventhat there is a large but not perfect correlation between these two dimensions of the voting systemacross countries50. While Anglo-Saxon countries tend combine plurality electoral rules with single-

45This category includes electoral systems that follow ‘plurality’ voting rules, such as the British ‘first-past-the-post’system, as well as ‘majority’ voting rules, such as the French ‘two-round system’.

46The original IAEP dataset only provides information up until 2005. Havard Hegre has kindly provided me with theupdated version, which extends the temporal scope of the dataset until 2012. The IAEP dataset codes information onformal institutions in existence as of January 1 each year. If an election happened in a given year for which there wasno electoral system recorded, the system coded for the following year (t+1) was assumed to have been in place at thetime of the elections. If there was also no electoral system information in the following year, missing values were kept.

47For those countries that changed electoral systems during the period 1990-2010, the last electoral system in place isdisplayed. The countries in question are: Chad, Madagascar, Mali, Mauritania, Morocco, Sudan and Togo.

48If data on constituency size was not available, DPI coders divided the number of seats in the lower house by thenumber of constituencies. Like the IAEP dataset, DPI codes information on institutions in existence as of January 1each year. If an election happened in a given year for which there was no information on average district size recorded,the coding for the following year (t+1) was assumed to have been in place at the time of the elections. If there was alsono district size information in the following year, missing values were kept.

49In this paper I use the terms ‘electoral system’, ‘electoral rule’ and ‘electoral formula’ interchangeably to refer tothe distinction between majoritarian/plurality, mixed and proportional representation rules.

50In this paper I do not examine the interaction between electoral systems and average district size. Previous studiessuggest that large voting districts can to some extent mediate the impact of majoritarian electoral systems on policyoutcomes (see Cox, 1997; Schneider and Wiesehomeier, 2008).

31

Figure 3: Electoral systems in Africa

Majoritarian

Mixed

PR

No data

member districts and countries with PR systems usually have large (multi-member) districts, there arealso a substantial number of countries with ‘mixed’ voting systems (see Blais and Massicotte, 1997).The mean district magnitude for all countries in my sample is 6.8 parliamentary seats per constituency.

To measure the degree of government centralization, I again use the updated IAEP dataset. IAEPcodes countries as Unitary states if they have “a strong central government with few if any regionaladministrative structures, where such structures have no autonomy” (Wig et al., 2014b; p.8). Unitarystates make up roughly 80 percent of my sample and are thus clearly the most ‘popular’ type ofgovernmental structure in Africa. The reference category is Federal states, which are characterizedby “semi-autonomous regional political units or subordinate provincial governments” (ibid.). Thiscategory is relatively rare in Africa and includes countries such as Benin, Ethiopia, Gabon, Nigeria andMadagascar. Information on whether a country has a presidential, semi-parliamentary or parliamentarysystem was taken from the DPI dataset. Countries in which the legislature elects the chief executiveare coded as having Parliamentary regimes. However, if the legislature needs a two-third majorityor a dissolution in order to remove the chief executive and elect a new one, the country is coded asSemi-parliamentary. The former category makes up only 5 percent of my sample and includes countriessuch as Botswana, Ethiopia, Lesotho and Mauritius. The latter category represents roughly 10 percentand includes countries such as Congo-Brazzaville, Egypt, South Africa and Togo. All other countriesin my dataset have Presidential regimes, which is by far the most frequent regime type in my sample(see Beck et al., 2001).

To measure the degree of State Capacity in each country, I use the ICRG Quality of Governmentindex from the Quality of Government database (Teorell et al., 2013). This index is based on expertassessments of three factors that determine a country’s quality of government. These are (1) theperceived level of corruption in the political system, (2) the degree of abidance to law and order, and

32

(3) the autonomy and quality of the state bureaucracy51. The index ranges from 0 to 1, with Liberiain 1997 and the DRC in 2006 having the lowest state capacity (0.11) and Namibia in 1994 havingthe highest state capacity (0.84). The biggest threat to the validity of such expert measures of statecapacity is endogeneity to the conflict events that they are supposed to predict (Hendrix, 2010). Itherefore use a lagged version of the ICRG index, to avoid the possibility that the expert’s assessmentof a country’s quality of government for a given year is largely determined by the occurrence ofelectoral violence that year. In order to examine how electoral systems interact with ethnic divisions,I use two different measures of ethnic distribution based on data gathered by Fearon (2003). TheEthnic Fractionalization index ranges from zero (for perfectly homogeneous societies) to one (forhighly fractionalized societies)52. This measure reflects the probability that two randomly selectedpeople from a given country will belong to two distinct ethnic or ‘ethno-religious’ groups. My measureof Ethnic Polarization in turn indicates whether a society is polarized between two large ethnic groupsor not. The measure is taken from Schneider and Wiesehomeier (2008), who apply a formula originallyproposed by Reynal-Querol (2002) to Fearon’s ethnicity data. Their binary measure is preferableto the continuous one used by Reynal-Querol because it does not correlate so much with the ethnicfractionalization index. They construct an indicator of polarization that excludes countries abovethe 45-degree line of the correlation matrix between ethnic fractionalization and ethnic polarizationand also use a threshold of 0.5 of the Reynal-Querol polarization index. Their indicator thus largelycaptures highly polarized countries that are in the medium range of the ethnic fractionalization index(see Schneider and Wiesehomeier, 2008; p.191).

4.3 Controls

In my regression models I control for a number of factors that potentially impact the likelihood and levelof electoral violence. Previous empirical studies in the literature on armed conflict and state repressionsuggest that the level of economic development as well as population size significantly influence acountry’s likelihood to experience social conflict (see e.g. Fearon and Laitin, 2003; Davenport, 2007). Itherefore include measures of real GDP per capita and total population size in my regression models.Both measures are based on data from the Penn World Table (Heston et al., 2012). Furthermore, Icontrol for the level of democracy in each country given that previous research has found that socialconflict is less likely in mature democracies (e.g. Davenport and Armstrong, 2004; Bueno De Mesquitaet al., 2005; Hegre, 2014). I use the Revised Combined Polity IV index (Marshall et al., 2013), whichranges from -10 for strongly autocratic regimes to +10 for strongly democratic regimes, and alsoinclude its quadratic term to capture the potentially curvilinear relationship between regime type andviolence (see Hegre, 2014; Fjelde and Höglund, 2015). The GDP, population and Polity variables areall lagged by one year in order to address concerns of reverse causation53. Ongoing civil wars are likelyto polarize electoral contests and can thereby increase the likelihood of election-related violence (seeHafner-Burton et al., 2014). I therefore include a binary variable, which indicates whether a givencountry experienced an internal or internationalized internal armed conflict at the time of the election.The information for this variable was taken from the UCDP/PRIO Armed Conflict Dataset (Themnerand Wallensteen, 2013). I also control for a country’s level of ethnic fractionalization and polarization,given that the civil war literature has previously advocated one or both factors as important structuraldeterminants of conflict onset and intensity (see Esteban and Ray, 2008; Esteban et al., 2012). Tocontrol for temporal dependence among consecutive elections in the same country, I include a laggedversion of the dependent variable in all of my models54. Lastly, I control for the number of electionrounds per country between 1990 and 2010. This is necessary because some countries in my sample

51For more information on the ICRG Quality of Government index, see Teorell et al. (2013; p.107). Ideally, I wouldhave preferred to use the disaggregated ICRG measure of ‘bureaucratic quality’ to capture the degree of state capacityin each country. However, the original data for the ICRG Quality of Government index are provided by a commercialrisk assessment firm, the PRS Group, which protects its products behind a high paywall.

52The Ethnic Fractionalization measure is taken from the Quality of Government database (Teorell et al., 2013).53All three measures are taken from the Quality of Government database (Teorell et al., 2013).54This variable is coded as 1 if the preceding election round experienced at least one violent event and 0 if no violent

event occurred. It is thus a lagged ‘dummy version’ of my dependent variable(s).

33

have held only a couple of elections in this time period (e.g. DRC, Liberia or Rwanda), while othershave experienced more than 15 election rounds (e.g. Comoros, Egypt or Mali). This means that Ineed to account for the possibility that countries with more ‘democratic experience’ might be lesssusceptible to electoral violence, even though their ‘exposure’ to electoral violence can be said to behigher55. Additional controls are discussed in the sections on election-specific factors and colonialhistory.

55A related problem is that I am working with an unbalanced panel dataset, with some countries having more pre- andpost-election periods than others. It is likely that a country’s probability of not holding elections (i.e. to be “missing”from my dataset) also depends to some extent on the level of electoral violence in a previous election (i.e. my dependentvariable). Therefore, countries in my dataset might not be “missing completely at random”, which can potentially biasmy estimates (see Frees, 2004; p.265).

34

Chapter 5

Analysis

I use negative binomial regression models with robust standard errors clustered by country to analyzethe drivers of electoral violence. Negative binomial regression is the most suitable estimation methodin this case, given that my dependent variables are over-dispersed count measures. Negative binomialregression can be considered as a generalization of Poisson regression, which is commonly used formodeling count data (Hilbe, 2011). It has the same mean structure as Poisson regression, but includesan additional parameter to model over-dispersion. A basic characteristic of the Poisson distributionis that its mean is equal to its variance. However, the conditional variance of my outcome variablesalways exceeds the conditional mean. Hence, a basic assumption of the Poisson regression modelis violated and the negative binomial regression model is preferable56. In this chapter I discuss themain findings from my regression analysis. Overall, I find that institutional factors are importantdeterminants of both pre- and post-election violence. Importantly, my findings also suggest that theimpact of individual institutional factors can differ significantly between the two periods. To beginwith, I examine how institutions influence a country’s likelihood to experience violence in the pre-election period (Section 5.1). Thereafter, I analyze the impact of institutional factors on violence inthe post-election period (Section 5.2). In each section, I also use interactions to examine how differentforms of ethnic diversity mediate the effect of majoritarian electoral systems on electoral violence.Furthermore, I test whether my findings remain robust when taking into account election-specificfactors as well as the colonial history of each country.

5.1 Pre-election violence

Institutional drivers

The results presented in Table 2 show that, all other things being equal, majoritarian systems signif-icantly increase the number of violent events per pre-election period compared to PR systems. Thiseffect applies to both pro-government violence as well as oppositional violence. Mixed systems are alsomore susceptible to violence than PR systems, even though this effect is less significant and disappearscompletely when considering only oppositional violence. The incidence rate57 of both pro-governmentas well as oppositional violence is about three times higher in countries with majoritarian electoral

56Given that I use ‘election rounds’ rather than ‘country-months’ or ‘country-years’ as my unit of analysis, my countmeasure does not suffer from ‘excess zeroes’ (i.e. it does not code ‘zero’ electoral violence during periods without elections)and a zero-inflated model is not necessary (cf. Fjelde and Höglund, 2015). For more information on modeling countoutcomes, see Long and Freese (2006; Chapter 7) and Hilbe (2011).

57The incidence rate is the rate at which events occur within a given time-period. To obtain the effect of a one-unitincrease in a particular explanatory variable on the incidence rate, one only needs to exponentiate the negative binomialregression coefficient of the variable of interest, which yields the incidence rate ratio (IRR). While the coefficients reportedin my regression tables have an additive effect in the log(y)-scale, the IRRs have a multiplicative effect in the y-scale(see Hilbe, 2011; Chapter 2). Due to space constraints, I do not report IRRs in my regression tables.

35

Table 2: The effect of electoral systems and district size on pre-election violence(1) (2) (3) (4)

Governmentviolence

Oppositionviolence

Governmentviolence

Oppositionviolence

1. Majoritarian 1.145∗∗∗ 1.108∗∗(0.421) (0.463)

2. Mixed 0.808∗ -0.845(0.443) (0.600)

District size (log) -0.276∗ -0.596∗∗(0.148) (0.255)

Civil war -0.464 0.00567 -0.531 0.0786(0.446) (0.308) (0.522) (0.307)

Population (log, lag) 0.706∗∗∗ 0.633∗∗∗ 0.675∗∗∗ 0.516∗∗∗(0.144) (0.110) (0.144) (0.131)

GDP pc (log, lag) -0.171 0.359∗ 0.0514 0.538∗(0.248) (0.194) (0.278) (0.283)

Polity (lag) 0.0143 0.0136 0.00529 0.0120(0.029) (0.029) (0.032) (0.027)

Polity squared (lag) -0.00639 -0.0149∗∗ -0.0117 -0.0215∗∗(0.007) (0.007) (0.008) (0.009)

Ethnic fractionalization -0.321 -0.760 0.00248 0.751(0.779) (1.147) (0.827) (1.100)

Ethnic polarization 0.508 0.546 0.687 1.109∗∗(0.417) (0.385) (0.527) (0.466)

Number of elections -0.0777∗ 0.0593∗ -0.0836∗ 0.00240(0.046) (0.036) (0.045) (0.036)

Previous gov. violence 0.0272 0.208(0.295) (0.326)

Previous opp. violence 0.336 0.396(0.335) (0.349)

Constant -6.657∗∗∗ -10.75∗∗∗ -6.996∗∗ -10.40∗∗∗(2.561) (2.723) (2.844) (3.595)

ln(alpha) 0.258 -1.854 0.408 -1.451(0.415) (1.256) (0.399) (0.940)

Observations 277 277 258 258Clusters 45 45 44 44Robust standard errors clustered by country in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

36

systems than in countries with PR systems. Countries with majoritarian systems experience on aver-age 0.25 pro-government violent events per pre-election period, while countries with PR systems canonly expect 0.08 violent events per pre-election period58. The corresponding estimates for oppositionalviolent events are slightly lower, but also three times higher in majoritarian system than in PR sys-tems. Even though the overall predicted number of violent events per pre-election period might seemnegligible, it is important to keep in mind that outbreaks of electoral violence are relatively rare eventsand that the ‘baseline risk’ in a given country is quite low (see Table 1 in Chapter 4). Nevertheless,my findings suggest that a country’s risk of experiencing electoral violence is three times higher if ithas a majoritarian electoral system rather than a PR system. Given the life-and-death nature of muchelectoral violence, this certainly represents a sizable increase.

Table 2 shows that my hypothesis regarding the impact of a country’s district size on the risk ofelectoral violence also finds support in the data. The negative coefficients indicate that an increase inthe average number of representatives elected per constituency significantly reduces the risk of bothpro-government as well as oppositional violence. This is because larger districts tend to lower thebarriers for representation of minority groups and reduce the stakes of the elections by providing moreopportunities for electoral success (see Huber, 2012). A one-unit increase in the (logged) number oflegislative seats per constituency reduces the incidence rate of pro-government pre-election violence bya factor of 0.75 (or 25%). This means that moving from the 10th to the 90th percentile of the (logged)district size variable reduces the expected number of pro-government violent events per pre-electionperiod from 0.23 to 0.11. The effect is even stronger for opposition violence. In this case, a one-unitincrease in the (logged) average district size of a given country reduces the incidence rate of oppositionviolence by a factor of 0.55 (or 45%). Accordingly, the expected number of opposition violent eventsper pre-election period decreases from 0.2 to 0.04 when moving from the 10th to the 90th percentileon the district size variable. These results show that increasing a country’s average district size canbe an important tool for policy makers to lower the stakes of electoral contests and thereby reduce thelikelihood of election-related violence.

In Chapter 3 I argued that electoral systems are not the only institutional factor that can shape acountry’s predisposition to experience electoral violence. In line with this reasoning, Table 3 showsthat unitary states are significantly more likely to experience pre-election violence than federal states.Importantly, the violence-inducing impact of unitary states only applies to pro-government violence.This finding suggests that federal state structures place particular constraints on the actions of gov-ernmental forces, which do not apply in unitary states. We can argue that in federal states, the controlover police forces is more likely to be dispersed among regional (opposition-controlled) governments,rather than being firmly in the hands of the incumbent in the national capital. This constraint onexecutive power should decrease the risk of pro-government electoral violence in federal states com-pared to unitary states. Furthermore, we can argue that federal state structures diminish a country’srisk of experiencing electoral violence by reducing the stakes of national elections and providing com-peting groups with more opportunities for electoral success and representation (see Schneider andWiesehomeier, 2008). However, given that I only find that unitary states increase the likelihood ofpro-government violence (and not oppositional violence), we have to treat this conclusion with somecaution. Be that as it may, a country with a unitary state structure can be expected to have 2.5times (or 150%) more pro-government pre-election violent events than a country with a federal statestructure. This means that unitary states can expect to experience on average 0.2 violent events perpre-election period, whereas federal states can expect only 0.07 events. These findings show that gov-ernment decentralization is an important institutional determinant of electoral violence, which has sofar been completely ignored in the literature.

I do not find evidence that ‘state capacity’ matters in determining the level of pre-election violencein a given country59. This is in line with my expectation that the violence-reducing effect of capable

58All substantive terms in this paper were calculated using the margins command in Stata 12. This command allowsme to calculate the predicted event counts at each level of my explanatory variable(s), while holding all other variablesin the model(s) at their means. The binary civil war variable as well as the lagged, binary version of the dependentvariable were held at zero (i.e. the mode).

59Due to space constraints, I do not present a results table of this regression model.

37

Table 3: The effect of state structure and regime type on pre-election violence(1) (2) (3) (4)

Governmentviolence

Oppositionviolence

Governmentviolence

Oppositionviolence

Unitary 0.931∗ 0.312(0.511) (0.302)

1. Semi-parliamentary 1.026∗∗∗ 0.170(0.366) (0.626)

2. Parliamentary -0.986 -0.322(0.666) (0.700)

1. Majoritarian 1.307∗∗∗ 1.177∗∗ 1.505∗∗∗ 1.189∗∗(0.433) (0.506) (0.415) (0.514)

2. Mixed 0.834∗ -0.919 1.005∗∗ -0.829(0.476) (0.620) (0.410) (0.616)

Civil war -0.457 0.00788 -0.365 0.0480(0.388) (0.274) (0.387) (0.295)

Population (log, lag) 0.832∗∗∗ 0.684∗∗∗ 0.659∗∗∗ 0.628∗∗∗(0.185) (0.144) (0.137) (0.109)

GDP pc (log, lag) -0.171 0.376∗ -0.193 0.362∗(0.262) (0.209) (0.247) (0.188)

Polity (lag) 0.0313 0.0189 0.0392 0.0199(0.031) (0.031) (0.032) (0.033)

Polity squared (lag) -0.00875 -0.0156∗∗ -0.00991 -0.0155∗∗(0.007) (0.007) (0.007) (0.007)

Ethnic fractionalization -0.0971 -0.590 -0.340 -0.793(0.722) (1.180) (0.738) (1.145)

Ethnic polarization 0.680 0.589 0.715∗ 0.585(0.422) (0.421) (0.427) (0.426)

Number of elections -0.0610 0.0740∗ -0.110∗∗ 0.0537(0.047) (0.041) (0.048) (0.039)

Previous gov. violence -0.0985 -0.00287(0.286) (0.301)

Previous opp. violence 0.306 0.307(0.346) (0.353)

Constant -9.043∗∗∗ -11.87∗∗∗ -6.095∗∗ -10.70∗∗∗(2.782) (3.263) (2.377) (2.702)

ln(alpha) -0.00768 -2.077 0.117 -1.987(0.523) (1.633) (0.441) (1.504)

Observations 273 273 277 277Clusters 45 45 45 45Robust standard errors clustered by country in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

38

state bureaucracies should mostly apply in the post-election period. The results presented in Table 3show that a country’s ‘regime type’ has a significant impact on the likelihood of pre-election violence.Countries with semi-parliamentary regimes are 2.8 times more likely to experience pro-government pre-election violence than countries with presidential regimes60. The incidence rate of government violenceincreases by about 180 percent when moving from presidential to semi-parliamentary systems. Coun-tries with semi-parliamentary systems are also significantly more likely to experience pro-governmentviolence than countries with parliamentary systems. The incidence rate of violence is roughly 7.5 timeshigher in semi-parliamentary systems than in parliamentary systems. Purely parliamentary systemsseem to be less prone to pro-government violence than presidential systems. However, the relevantcoefficient just misses the 90 percent significance threshold, which provides only limited support forthe hypothesis that parliamentary systems have a pacifying effect in comparison to presidential sys-tems. Most importantly, my results indicate that countries with semi-parliamentary regimes are moreviolence-prone than countries that adopted one of the ‘pure’ alternatives.

In Chapter 3 I argued that countries with presidential regimes should be most at risk of electoralviolence, given the winner-takes-all logic of presidentialism. The empirical findings are thus somewhatat odds with my theoretical expectations. One explanation for this is that my findings are simplydriven by the make-up of my sample. Given that the overwhelming majority (85%) of African coun-tries adopted presidential systems of government, the reference category (i.e. presidential regimes) islikely to ‘hide’ a significant amount of variation in terms of electoral violence. Furthermore, the fewAfrican countries that have semi-parliamentary regimes (including Egypt, South Africa and Togo) allhave a long history of violent conflict or civil unrest, which might explain their choice of this specificregime type as well as their high predisposition to experience electoral violence. An alternative expla-nation for my findings suggests that semi-parliamentary regimes are most at risk of election-relatedconflict because they combine the negative attributes of both presidentialism and parliamentarism.Commenting on the conflict-potential of both regime types, Gerring et al. (2008; p.334) note that “ina separate powers system [i.e. presidential regime], conflict is endemic and continual. Each branch isassumed to represent a somewhat different constituency or the same constituency in different ways.Yet, because a higher threshold of consensus is necessary for agreement on any policy measure, itmight be said that consensus is mandated by a separate powers constitution. By contrast, powerin a parliamentary system is temporarily monopolized by a single party or coalition. Other groupsmay voice their opposition, but they have no formal mechanism by which they might affect policyoutcomes. Consequential conflict is thus episodic, occurring during elections but not in between.”By combining the conflict-potential of both presidentialism as well as parliamentarism, countries withsemi-parliamentary regimes might therefore be most at risk of election-related violence.

Looking at the control variables in Tables 2 and 3, my results largely confirm the findings of previousempirical studies on electoral violence (see Daxecker, 2014; Hafner-Burton et al., 2014; Fjelde andHöglund, 2015). Countries with larger populations are significantly more at risk of experiencing bothforms of pre-election violence than countries with smaller populations. This finding is robust across allmodels. Furthermore, strongly authoritarian and firmly democratic countries are much less likely tosuffer from oppositional pre-election violence than ‘transitional’ countries, which combine authoritar-ian and democratic regime characteristics. The coefficient of the squared Polity index is negative andsignificant across all models of opposition pre-election violence. This finding suggests that stronglyauthoritarian regimes tend to have sufficient determination and repressive capacity to prevent oppo-sitional non-state actors from mobilizing during the election campaign. Firmly democratic regimesin turn provide oppositional actors with sufficient space to voice their concerns in a peaceful mannerand hence also tend to be less susceptible to electoral violence. However, ‘transitional’ regimes arepartly open to opposition mobilization but lack experience and effective means to solve conflict peace-fully and are therefore at an increased risk of violence (see Hegre, 2014). I find no evidence that acountry’s ‘democraticness’ influences the level of governmental pre-election violence (cf. Section 1.2.on post-election violence). Higher levels of economic development, measured in real GDP per capita,tend to make a country more susceptible to opposition pre-election violence. This finding is consis-tent across all four models of opposition violence and suggests that opposition actors need access to

60I did not find evidence that regime type has an impact on the level of oppositional pre-election violence.

39

a certain amount of resources before they can successfully mobilize against their governments duringelections. A country’s level of economic development only influences the extent of anti-governmentmobilization and does not seem to impact the level of governmental pre-election violence (cf. Section1.2. on post-election violence). The number of election rounds per country is a significant predictorof electoral violence across most of my models. Interestingly, while countries with more ‘democraticexperience’ tend to suffer less governmental violence, they also tend to experience more oppositionalviolence. This finding suggests that there is indeed a ‘learning effect’ attached to every successiveelection round, which makes governments better at implementing peaceful elections. At the same timehowever, a larger number of elections also provides oppositional actors with more opportunities toorganize and mobilize against the government. I do not find evidence that ongoing civil wars makean outbreak of electoral violence more likely. Furthermore, the occurrence of electoral violence in theprevious election round is not a significant predictor of violence in the subsequent election round.

Interactions

In order to examine how the violence-inducing effect of majoritarian electoral systems is mediated bydifferent forms of ethnic diversity, I run a number of interactions. Table 4 reports the interactions oftwo measures of ethnic composition, Ethnic Fractionalization and Ethnic Polarization, with a binaryelectoral system variable, which equals 1 if a country has a majoritarian system and 0 if it has a non-majoritarian system61. In Chapter 3 I hypothesized that the violence-generating impact of electoralsystems will depend on the specific nature of ethnic divisions within society. I argued that majoritariansystems are particularly likely to lead to violence in societies that are made up of a few large ethnicgroups rather than many small ones. The results in Model 1 provide some support for this hypothesis.The negative coefficient of the interaction term indicates that in majoritarian systems, an increasein ethnic fractionalization is associated with a decrease in the level of pro-government pre-electionviolence. This suggests that majoritarian systems are less violence-inducing in societies where manysmall ethnic groups vie for political influence than in societies where a few large groups compete forpower. This makes intuitive sense given that in highly fractionalized contexts politicians usually needthe support of a number of different ethnic groups in order to obtain a majority in parliament. Incitingethnic competition is thus a less attractive tool to mobilize support amongst ones own ethnic group,given that it might lead to the loss of support amongst other ethnic groups.

Given that interaction terms with continuous predictors make the coefficients in the model difficult tointerpret, a graphical representation of the interaction term is the most suitable option (see Bramboret al., 2006). Figure 4 displays the predicted number of pro-government violent events per pre-electionperiod in majoritarian and non-majoritarian systems for different levels of ethnic fractionalization. Theoverlapping confidence intervals around the two lines suggest that there is no statistically significantdifference between the two types of electoral systems in terms of how ethnic fractionalization mediatestheir impact on electoral violence. However, the different slopes of the two lines indicate that theviolence-inducing effect of majoritarian systems is mediated by the level of ethnic fractionalization(i.e. it is more pronounced in ethnically homogenous societies), whereas the effect of PR systemson electoral violence does not seem to depend on a country’s level of ethnic fractionalization. Theresults for opposition pre-election violence in Model 2 of Table 4 are roughly the same, only that thecoefficient of the interaction term just misses the 90 percent significance threshold. Models 3 and4 examine how a country’s level of ethnic polarization mediates the impact of electoral systems onelectoral violence. My theoretical argument implies that the violence-inducing impact of majoritariansystems is particularly pronounced in societies that are highly polarized (i.e. where my binary measureof ethnic polarization equals 1). The positive and significant coefficient of the interaction term in Model3 supports this proposition. Countries with majoritarian electoral systems are much more likely toexperience pro-government violence when they are highly polarized along ethnic lines than when theyare not. In fact, highly polarized countries with majoritarian systems are roughly 3.5 times more likelyto experience pro-government violence than countries with majoritarian systems that are not ethnically

61In order to make the interaction models easier to interpret, mixed and PR electoral systems were collapsed into onecategory.

40

Table 4: Majoritarian systems, ethnic divisions and pre-election violence(1) (2) (3) (4)

Governmentviolence

Oppositionviolence

Governmentviolence

Oppositionviolence

Majoritarian 2.556∗∗∗ 3.277∗∗ 0.371 1.180∗∗∗(0.992) (1.404) (0.329) (0.403)

Ethnic fractionalization 0.282 0.350(0.848) (0.850)

Majoritarian × Ethnic frac. -2.589∗ -2.965(1.380) (1.889)

Ethnic polarization -0.0112 0.636(0.540) (0.543)

Majoritarian × Ethnic pol. 1.243∗ 0.416(0.683) (0.906)

Civil war -0.312 0.238 -0.356 0.0886(0.473) (0.360) (0.470) (0.300)

Population (log, lag) 0.720∗∗∗ 0.597∗∗∗ 0.748∗∗∗ 0.639∗∗∗(0.131) (0.089) (0.128) (0.097)

GDP pc (log, lag) -0.0235 0.468∗∗∗ -0.0552 0.407∗∗(0.199) (0.174) (0.193) (0.173)

Polity (lag) 0.00992 0.0254 0.0243 0.0289(0.028) (0.028) (0.032) (0.031)

Polity squared (lag) -0.00849 -0.0181∗∗∗ -0.0109 -0.0181∗∗(0.007) (0.006) (0.008) (0.008)

Number of elections -0.0502 0.0450 -0.0518 0.0512(0.047) (0.037) (0.044) (0.035)

Previous gov. violence 0.170 0.172(0.325) (0.330)

Previous opp. violence 0.419 0.376(0.344) (0.342)

Constant -7.854∗∗∗ -11.71∗∗∗ -7.649∗∗∗ -11.79∗∗∗(2.131) (2.042) (1.917) (1.902)

ln(alpha) 0.265 -1.167 0.226 -1.461(0.408) (0.940) (0.407) (0.993)

Observations 277 277 277 277Clusters 45 45 45 45Robust standard errors clustered by country in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

41

Figure 4: Predicted number of pro-government pre-election violent events by electoral system fordifferent levels of ethnic fractionalization

−1

01

23

Pre

dict

ed N

umbe

r O

f Eve

nts

(with

90%

CI)

0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1Ethnic Fractionalization

Non−majoritarian Majoritarian

polarized. While the average predicted number of violent events per pre-election period is 0.97 in theformer scenario, it is only 0.28 in the latter. In countries without majoritarian electoral systems, ethnicpolarization does not seem to have the same violence-inducing effect as in countries with majoritariansystems. This makes intuitive sense, given that non-majoritarian systems can usually provide largecompeting ethnic groups with sufficient opportunities for political influence and representation inorder to avoid violent confrontation. I do not find evidence that the same interaction effect applies tooppositional pre-election violence.

Overall, my interactions suggest that a country’s ethnic composition does indeed influence the way inwhich majoritarian systems increase the risk of electoral violence. In particular, the interactions withboth measures of ethnic composition indicate that majoritarian electoral systems in highly polarizedsocieties are the worst possible combination. While PR systems seem to be able to accommodate differ-ent forms of ethnic diversity, majoritarian systems are much more sensitive to different forms of ethnicdivision and are particularly prone to electoral violence when two large groups compete for power.This is because an opposition party with a relatively large electoral constituency based on sharedethnicity represents a much more credible threat to the ruling party than a highly factionalized oppo-sition, which in turn heightens the stakes of electoral competition. The violence-inducing interactionbetween majoritarianism and ethnic polarization is best exemplified by the case of Zimbabwe, whichcombines a British-style ‘first-past-the-post’ electoral system with high ethnic polarization. Whilethe ruling ZANU-PF party is largely supported by members of the Shona people, the second-largestethnic group, the Ndebele, overwhelmingly votes for the oppositional MDC. As a result, elections inZimbabwe are not only ‘high-stakes’ contests between the ruling party and the opposition, but alsocontests between the two largest ethnic groups. For example, the Minority Rights Group notes thatin the run-up to the 2005 elections, the ruling ZANU-PF party “turned the ongoing economic crisisto its advantage by withholding food aid from disproportionately Ndebele MDC supporters, while

42

directing increased distribution to disproportionately Shona ZANU-PF supporters” (Minority RightsGroup, 2008). That same year, the ruling ZANU-PF party also initiated ‘Operation Murambatsvina’,which forcibly displaced thousands of slum-dwellers - overwhelmingly MDC supporters - in an attemptto reshape the urban electoral landscape in its favour (ibid.). In light of these observations, it is notsurprising that, according to my data, Zimbabwe has experienced more election-related violence thanany other country in Africa.

Election-specific factors and colonial history

We have seen that electoral violence is most likely in centralized states that have majoritarian electoralsystems, semi-parliamentary regimes and weak state capacity. However, in this section I try to con-textualize these general findings and to ensure that they are not driven by election-specific dynamics.I control for two additional factors that can be expected to have a more immediate effect on electoralcompetition than the structural drivers in my main models. The results tables of these additionaltests can be found in the Appendix. Firstly, I include a binary variable in my main regression models,which indicates whether the election in question was ‘competitive’. The measure is taken from theNELDA dataset and equals 1 if “opposition was allowed, more than one party was legal, and more thanone candidate competed” (see Hyde and Marinov, 2012b; p.192). The theoretical expectation is that‘competitive’ elections are on average more prone to electoral violence than ‘uncompetitive’ elections,given that in the latter scenario there is no uncertainty regarding the outcome of the election. Sec-ondly, I control for the economic climate during the election period based on the expectation that lessfavorable economic conditions will increase the likelihood of violence by allowing manipulative elitesto channel socio-economic grievances into violent behaviour. Furthermore, it can be assumed thatfavorable economic conditions will make it easier for the ruling elite to ‘buy off’ potential electoralchallengers, which will in turn reduce the likelihood of election-related violence (see e.g. Collier andVicente, 2012). I measure the economic climate during each election period by using the World Bank’sannual GDP growth rates lagged by one year (WB, 2013). All my main findings regarding the impactof ‘institutions’ on the level of electoral violence are robust to the inclusion of these election-specificfactors (see Appendix, Tables 8 & 9). The economic climate during elections is itself a significantpredictor of electoral violence. In other words, positive GDP growth rates in the year preceding theelections significantly reduce the likelihood of election-related violence. However, whether an electionis ‘competitive’ or not does not seem to matter in terms of increasing a country’s risk of electoralviolence.

In addition to the above-mentioned election-specific factors, I also control for the colonial historyof the countries in my panel. Previous studies have shown that former African colonies have largelyadopted the electoral systems used by their colonial masters (see e.g. Reilly et al., 1999; Bogaards, 2013;Pospieszna and Schneider, 2013). Furthermore, the type of rule to which a country was subjected toduring the colonial era can be said to influence its susceptibility to violent conflict in the post-colonialperiod. For example, Blanton et al. (2001) show that the distinctive colonial styles of the Britishand French empires created different systems of ethnic stratification, which left former British coloniesexposed to a relatively higher risk of post-colonial ethnic strife. This suggests that both the typeof electoral system as well as the level of electoral violence can be to some extent explained by acountry’s colonial legacy. In order to control for this possibility, I include a categorical Colonial Originvariable in the two models that examine the impact of electoral systems on pre-election violence.The variable distinguishes between former British and French colonies in Africa, with all countries notcolonized by these two empires serving as the reference category62. The main finding that majoritarianelectoral systems are more prone to electoral violence than PR and mixed systems remains robust whencontrolling for colonial legacy (see Appendix, Table 10). The results also indicate that a country’scolonial origin is not necessarily a good predictor of current levels of electoral violence. However, Ido find that former British colonies are roughly 3.5 times more likely to experience oppositional pre-election violence than countries in the reference category. To some extent this confirms Blanton et al.

62The information on each country’s colonial origin comes from the Quality of Government Dataset (Teorell et al.,2013).

43

(2001), who argue that the British colonial style of ‘divide-and-rule’ fostered ethnic competition, whichin turn increased the risk of conflict in the post-colonial era.

5.2 Post-election violence

Institutional drivers

Most of the institutional factors that I identified as important structural drivers of pre-election violencealso have an impact on the likelihood of post-election violence. However, there are some crucialdifferences between the two periods, which I highlight in this section. Table 5 shows that majoritarianelectoral systems significantly increase a country’s risk of experiencing violence in the post-electionperiod. The direction of the effect is thus the same as in the pre-election period. The incidencerate of pro-government post-election violence is about 5 times higher in countries with majoritarianelectoral systems than in countries with PR systems. In the case of oppositional post-election violence,countries with majoritarian electoral systems have an incidence rate that is 10 times higher than incountries with PR systems. The violence-inducing effect of majoritarian electoral systems is thus muchmore pronounced in the post-election period than in the pre-election period. Given that the effect isparticularly strong for oppositional post-election violence, it seems reasonable to suggest that theresults are mainly driven by ‘sore loser’ protests against the incumbent (see Hyde and Marinov, 2014).The ‘winner-takes-all’ logic of majoritarian systems significantly increases the chance of such ‘soreloser’ protests in the aftermath of elections because, in contrast to PR systems, majoritarian formulaswill not allow opposition actors to share in the ‘spoils’ of public office. An attempt by opposition actorsto alter the declared results of the elections through ‘extra-constitutional means’ (i.e. protests, riots,etc.) is thus much more likely in majoritarian systems than in PR systems. Table 5 also shows thatthe violence-reducing effect of an increase in a country’s average district size is roughly the same inthe post-election period compared to the pre-election period. All other things being equal, a one-unitincrease in the (logged) district size variable reduces the number of pro-government violent events inthe post-election period by around 50 percent. This means that moving from the 10th to the 90thpercentile of the (logged) district size variable reduces the expected number of pro-government violentevents per post-election period from 0.09 to 0.01. At first sight, the average predicted number of violentevents per post-election period might seem negligible. Yet again, it is important to keep in mind thatoutbreaks of post-election violence are relatively rare events and that the ‘baseline risk’ in a givencountry is quite low. An increase in the average district size seems to also have a violence-reducingeffect on oppositional post-election violence. However, the relevant coefficient in Table 5 just missesthe 90 percent significance threshold.

I do not find evidence that a country’s state structure (i.e. unitary vs federal structure) has a significanteffect on post-election violence63. However, ‘state capacity’ does seem to be an important institutionalfactor affecting a country’s susceptibility to post-election violence (see Table 6). The higher a countryranks on the ICRG Quality of Government index, the less likely it is to experience violence in theaftermath of elections. This effect applies to both pro-government as well as oppositional violence.Moving from the 10th to the 90th percentile of the Quality of Government index (scaled 0-1) reducesthe expected number of oppositional violent events per post-election period from 0.1 to 0.03. Theestimates are roughly the same when looking only at pro-government violence. My findings suggestthat a competent and credible state machinery makes it much more likely that both electoral campswill accept the election results once they are proclaimed by the relevant authorities. This is in line withmy argument that a capable state administration can significantly increase the perceived legitimacy ofthe proclaimed election results and thereby reduce the risk of post-election violence aimed at overturn-ing the results. Furthermore, my findings suggest that strong and independent state bureaucraciesreduce the risk of insecurity and disorder in the post-election period due to the fact that they arerelatively independent from the incumbent’s patronage network and therefore less likely to break downas the result of abrupt power transfers. Whether a country has a presidential, semi-parliamentary or

63Due to space constraints, I do not present a results table of this regression model.

44

Table 5: The effect of electoral systems and district size on post-election violence(1) (2) (3) (4)

Governmentviolence

Oppositionviolence

Governmentviolence

Oppositionviolence

1. Majoritarian 1.587∗∗ 2.332∗(0.619) (1.213)

2. Mixed 0.834 1.311(0.695) (1.047)

District size (log) -0.673∗∗ -0.513(0.290) (0.337)

Civil war -1.035 -1.092 -1.091 -1.590(0.838) (0.750) (0.977) (0.994)

Population (log, lag) 0.320∗∗ 0.376∗∗∗ 0.252∗ 0.299∗(0.143) (0.136) (0.138) (0.178)

GDP pc (log, lag) -0.886∗∗∗ 0.0803 -0.609∗∗ 0.498(0.316) (0.510) (0.270) (0.631)

Polity (lag) 0.141∗∗ 0.0348 0.125∗∗ 0.0436(0.060) (0.056) (0.063) (0.057)

Polity squared (lag) -0.0427∗∗∗ -0.0141 -0.0494∗∗∗ -0.0299∗∗(0.016) (0.014) (0.017) (0.015)

Ethnic fractionalization -3.020∗∗∗ -0.784 -3.007∗∗∗ 0.0506(0.770) (1.642) (0.871) (1.849)

Ethnic polarization -0.0537 -0.452 0.147 -0.0115(0.487) (0.627) (0.495) (0.882)

Number of elections 0.0171 0.00469 -0.00987 0.00254(0.074) (0.054) (0.067) (0.058)

Previous gov. violence 0.687 0.602(0.739) (0.792)

Previous opp. violence -0.176 -0.000259(0.630) (0.706)

Constant 2.552 -7.282∗ 3.203 -7.716(2.526) (4.391) (2.520) (6.372)

ln(alpha) 0.361 -14.95∗∗∗ 0.574 -1.230(0.665) (3.644) (0.698) (3.899)

Observations 277 277 258 258Clusters 45 45 44 44Robust standard errors clustered by country in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

45

Table 6: The effect of state capacity and regime type on post-election violence(1) (2) (3) (4)

Governmentviolence

Oppositionviolence

Governmentviolence

Oppositionviolence

State capacity (lag) -4.189∗∗ -4.455∗∗(1.711) (2.203)

1. Semi-parliamentary 1.106∗ -0.147(0.612) (0.704)

2. Parliamentary -0.513 -13.69∗∗∗(0.990) (0.739)

1. Majoritarian 1.881∗∗∗ 2.487∗∗ 1.837∗∗∗ 2.476∗(0.697) (1.004) (0.648) (1.267)

2. Mixed 1.585∗ 1.548∗ 0.963 1.361(0.892) (0.920) (0.740) (1.057)

Civil war -2.814∗∗ -0.903 -0.937 -0.821(1.110) (0.746) (0.717) (0.654)

Population (log, lag) 0.406 0.250∗ 0.355∗∗ 0.371∗∗∗(0.276) (0.135) (0.158) (0.138)

GDP pc (log, lag) -1.161∗∗∗ 0.213 -0.884∗∗∗ 0.145(0.442) (0.342) (0.320) (0.512)

Polity (lag) 0.0514 0.00830 0.181∗∗∗ 0.0494(0.058) (0.049) (0.047) (0.062)

Polity squared (lag) -0.0331∗∗ -0.00435 -0.0476∗∗∗ -0.0136(0.015) (0.014) (0.014) (0.014)

Ethnic fractionalization -3.678∗∗∗ -1.344 -3.134∗∗∗ -1.071(0.861) (1.175) (0.770) (1.617)

Ethnic polarization -1.155∗ -0.342 0.201 -0.363(0.682) (0.729) (0.511) (0.628)

Number of elections -0.117 -0.0558 -0.0126 -0.00389(0.079) (0.063) (0.082) (0.055)

Previous gov. violence 0.703 0.658(0.803) (0.739)

Previous opp. violence -0.423 -0.236(0.561) (0.621)

Constant 6.536 -4.583 2.243 -7.507∗(4.258) (3.989) (2.641) (4.365)

ln(alpha) 0.362 -357.5 0.313 -16.94∗∗∗(0.572) (.) (0.711) (0.468)

Observations 218 218 277 277Clusters 35 35 45 45Robust standard errors clustered by country in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

46

parliamentary regime also has an impact on the level of violence in the post-election period. In linewith my findings on pre-election violence, the results presented in Table 6 indicate that countries withsemi-parliamentary regimes are also significantly more prone to pro-government post-election violencethan those with purely presidential regimes. In fact, the incidence rate of violence in the former cat-egory is three times higher than in the latter. I do not find evidence that the same dynamic appliesto oppositional post-election violence. However, purely parliamentary regimes seem to be significantlyless prone to oppositional violence than their presidential counterparts. This provides some supportfor the argument that presidential regimes increase the stakes of electoral contest (especially those forthe chief executive) by concentrating political power and state resources in one single office.

Tables 5 and 6 show that the impact of my control variables on post-election violence is similar totheir impact on pre-election violence. Countries with larger populations are significantly more at riskof experiencing both pro-government as well as oppositional post-election violence than countries withsmaller populations64. As is the case for pre-election violence, ongoing civil wars and occurrences ofelectoral violence in the previous election round are not significant predictors of post-election violence.The impact of a country’s ‘democraticness’ on levels of violence is also similar in the post- and pre-election periods. Again, my results suggest that countries around the mid-range of the Polity index aremost susceptible to electoral violence. Interestingly, it is government post-election violence that is muchless likely in strongly authoritarian and firmly democratic countries than in ‘transitional’ countries. Thecoefficient of the squared Polity index is negative and significant across all models of government post-election violence, but loses significance in most models of opposition violence. This finding suggeststhat strongly authoritarian and firmly democratic states do not need to rely on excessive force inorder to uphold and enforce the proclaimed results of an election. In contrast, ‘transitional’ countriestypically lack the legitimacy of firmly democratic states and they also generally do not have the samecapacity as strongly authoritarian states to deter ‘challengers’. As a consequence, the electoral resultsin such ‘transitional’ countries are much more likely to be challenged ‘on the streets’, which in turnincreases the likelihood of violent repression by government forces. The results in Tables 5 and 6also indicate that higher levels of economic development tend to make a country less susceptible togovernmental post-election violence. This suggests that richer states are more effective at enforcingthe proclaimed results of an election by non-violent means, given that they have more resources andsophisticated crowd-control measures at their disposal.

Interactions

In the post-election period, the violence-generating effect of majoritarian systems is also mediated bya country’s ethnic composition. Thus, electoral systems and ethnic diversity interact similarly in bothelection periods. Again, I find support for the hypothesis that majoritarian systems are particularlylikely to lead to violence in societies that are made up of a few large ethnic groups rather than manysmall ones. However, in contrast to the findings of pre-election violence, the interaction effect seemsto apply mostly to oppositional violence. We can assume that this finding is driven by ‘sore loser’protests of opposition groups, which have more incentives (than the incumbent) to resort to violencein the post-election period, given that they will try to challenge or overturn the ‘official’ election results(see Hyde and Marinov, 2014). The negative coefficient of the interaction term in question indicatesthat in majoritarian systems, an increase in ethnic fractionalization is associated with a decrease inthe level of oppositional post-election violence. Figure 5 displays the predicted number of oppositionviolent events per post-election period in majoritarian and non-majoritarian systems for different levelsof ethnic fractionalization. The dynamics are identical to those observed in the pre-election period.Again, the overlapping confidence intervals around the two lines suggest that there is no statisticallysignificant difference between the two types of electoral systems in terms of how the level of ethnicfractionalization mediates their impact on electoral violence. Nevertheless, the different slopes of thetwo lines indicate that the violence-inducing effect of majoritarian systems is strongly influenced by thelevel of ethnic fractionalization (i.e. it is more pronounced in ethnically homogenous societies), whereas

64This finding is robust across all models of post-election violence. The only exception is Model 1 in Table 6, wherethe relevant coefficient just misses the 90 percent significance threshold.

47

Table 7: Majoritarian systems, ethnic divisions and post-election violence(1) (2) (3) (4)

Governmentviolence

Oppositionviolence

Governmentviolence

Oppositionviolence

Majoritarian 0.523 5.161∗∗∗ 0.870 0.959(1.310) (1.822) (0.595) (0.585)

Ethnic fractionalization -3.683∗∗ 2.115(1.543) (1.860)

Majoritarian × Ethnic frac. 1.174 -5.160∗∗(2.020) (2.550)

Ethnic polarization 0.438 -15.86∗∗∗(0.702) (0.621)

Majoritarian × Ethnic pol. 0.521 16.38∗∗∗(0.832) (1.114)

Civil war -1.035 -0.908 -1.034 -0.920(0.834) (0.790) (0.771) (0.754)

Population (log, lag) 0.328∗∗ 0.661∗∗∗ 0.437∗∗ 0.485∗∗∗(0.154) (0.160) (0.208) (0.154)

GDP pc (log, lag) -0.878∗∗∗ 0.423 -0.657∗∗ 0.178(0.287) (0.378) (0.322) (0.415)

Polity (lag) 0.127∗∗ 0.0344 0.0835 0.0453(0.056) (0.058) (0.062) (0.057)

Polity squared (lag) -0.0408∗∗∗ -0.0183 -0.0342∗∗ -0.0168(0.016) (0.015) (0.016) (0.015)

Number of elections 0.0258 0.0910 0.0257 0.0424(0.069) (0.060) (0.078) (0.051)

Previous gov. violence 0.734 0.776(0.726) (0.722)

Previous opp. violence -0.452 -0.257(0.614) (0.602)

Constant 2.993 -14.27∗∗∗ -2.019 -8.666∗∗∗(2.805) (4.306) (3.295) (2.944)

ln(alpha) 0.388 -21.90 0.349 -24.87(0.655) (.) (0.648) (.)

Observations 277 277 277 277Clusters 45 45 45 45Robust standard errors clustered by country in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

48

Figure 5: Predicted number of opposition post-election violent events by electoral system for differentlevels of ethnic fractionalization

−2

−1

01

23

4P

redi

cted

Num

ber

Of E

vent

s (w

ith 9

0% C

I)

0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1Ethnic Fractionalization

Non−majoritarian Majoritarian

the effect of PR systems on electoral violence is in no way dependent on a country’s level of ethnicfractionalization. The interaction with my second measure of ethnic composition provides additionalevidence to support my hypothesis. Countries with majoritarian systems that are at the same timealso highly polarized along ethnic lines (i.e. where my binary measure of ethnic polarization equals 1)are much more at risk of experiencing post-election violence than those that are not polarized. Whileethnically polarized countries with majoritarian systems can expect roughly 0.22 violent events perpost-election period, the corresponding estimate is only 0.13 for countries with majoritarian systemsthat are not polarized. In contrast to my findings on pre-election violence, the interaction effect onlyapplies to post-election violence that is perpetrated by opposition actors. As mentioned earlier, thisresult is likely to be driven by the phenomenon of ‘sore loser’ protests. In sum, we can say that thecombination of majoritarian electoral rules with ethnically polarized societies does not bode well forstability; neither in the pre-election period, nor in the post-election period.

Election-specific factors and colonial history

As with my analysis of the pre-election period, I examine whether my main findings on the institutionaldrivers of post-election violence are independent of election-specific factors. I include three election-specific control variables in all relevant regression models in order to test whether the impact ofinstitutions on post-election violence remains the same when situational factors are taken into account.To begin with, I control for allegations of electoral fraud. Previous studies on post-election protests haveshown that allegations of vote-rigging and other fraudulent electoral tactics can increase the likelihoodof protests in the aftermath of elections (see Tucker, 2007; Hafner-Burton et al., 2014). Electoralfraud is coded as a binary variable, which indicates whether there were “significant concerns that theelections will not be free and fair”, based on reports by domestic or international observers (Hyde

49

and Marinov, 2012a; p.8). I also include a binary control variable, which indicates whether electionresults that did not favour the incumbent were canceled. In these cases, the likelihood of post-electionprotests by the opposition can be expected to be significantly higher. Lastly, I include a third binaryvariable, which indicates whether the incumbent’s party lost the election in question. In these cases,there is a heightened risk of post-election violence perpetrated by pro-government forces, given thatunscrupulous incumbents will try to hold on to political power at any cost (see Hafner-Burton et al.,2014). All my main findings on the impact of ‘institutions’ on the likelihood of post-election violenceremain unchanged when controlling for the above-mentioned election-specific factors (see Appendix,Tables 11 & 12). The only result that is sensitive to election-specific factors is the violence-reducingeffect of state capacity. The direction of the effect remains the same as in the main models, but thep-value drops below the 90 percent significance threshold. I do not find evidence that allegations ofelectoral fraud have an impact on the likelihood of post-election violence. However, I find support forthe argument that the risk of oppositional post-election violence increases if election results that did notfavour the incumbent were subsequently canceled. My results also show that the risk of pro-governmentpost-election violence increases significantly if the incumbent’s party lost. As with my analysis of thepre-election period, I also control for each country’s colonial origin in the two models that examine theimpact of electoral systems on post-election violence65. The main finding that majoritarian electoralsystems are more prone to sore-loser protests (i.e. oppositional post-election violence) than PR systemsremains robust when controlling for colonial legacy. However, the relevant coefficient just misses the90 percent significance threshold when looking only at pro-government post-election violence.

65Due to space constraints, I do not present the results tables of these regression models.

50

Chapter 6

Conclusion

This paper represents the first attempt in the literature to systematically examine the impact of a wholerange of country-level institutional factors on the risk of electoral violence. My analysis has shownthat institutions matter in explaining why some countries are more prone to electoral violence thanothers. The statistical evidence presented in this paper suggests that highly centralized states, withmajoritarian electoral rules, semi-parliamentary regimes and weak state capacity are most susceptibleto election-related violence. Furthermore, my findings indicate that the violence-inducing effect ofmajoritarian electoral rules is particularly pronounced in societies that are highly polarized alongethnic lines. Importantly, my analysis also shows that the impact of institutions is not always uniformacross the two electoral periods and can vary depending on who perpetrates the violence. The factthat institutions have an important impact on the risk of electoral violence provides some hope for theso-called ‘institutional engineers’, who believe that societal conflicts can be managed or even avertedby smart constitutional designs and innovative institutional reforms (see Reilly, 2002b; Basedau, 2011).

This study represents a first step in terms of explaining the relationship between country-level insti-tutional factors and the risk of electoral violence. In this respect, it is important to keep in mindthat my analysis has a number of shortcomings. To begin with, my research design did not allowme to investigate the possibility that certain state institutions are endogenous to electoral violence.An instrumental variable approach would be necessary to address this issue appropriately (see e.g.Thies, 2010). Furthermore, my paper only examined the impact of a very narrow set of institutionalfactors on electoral violence, based on data availability and previous research priorities in the fieldsof conflict management and ‘institutional engineering’ (e.g. Basedau, 2011; Kurtenbach and Mehler,2013). However, it is likely that a number of other institutional factors that were not considered inthis study also play an important role in determining a country’s predisposition to experience electoralviolence. For example, Opitz et al. (2013) argue that the institutional design of Electoral ManagementBoards is likely to have a significant impact on the risk of post-election protests. Yet they cannotprovide systematic evidence to support their argument, given that data on electoral management de-sign are currently not available in a format that is suitable for statistical analysis. Future research onthe relationship between institutions and electoral violence would thus benefit greatly from increaseddata-gathering efforts in this area.

Another important shortcoming of my study relates to the interaction effect between electoral systemsand ethnicity. We need to keep in mind that the measures of ethnic fractionalization and polarizationused in this paper are static and aggregated at the country-level. Therefore, they do not tell usanything about the relative political power of different ethnic groups at a given point in time or theirgeographical distribution in the country. In the future, it would be interesting to employ data thatcapture changes in ethnic power relations at the group-level, rather than relying on static macro-levelindices of a country’s demographic (see e.g. Cederman et al., 2010b). My research design also didnot allow me to directly measure the salience of ethnicity during electoral contests (see Huber, 2012).Rather, I assumed that ethnicity automatically becomes a salient issue when the electoral system

51

provides politicians with incentives to mobilize constituencies based on their shared ethnicity. Futureresearch in this area would benefit from disaggregated data on electoral violence that allows researchersto distinguish ethnically motivated violence from other forms of violence.

Lastly, due to the fact that I only analyze data from the African continent, my findings are notnecessarily applicable to other contexts. However, even though my theoretical arguments about thecauses of electoral violence do not apply to mature democracies in Europe and North America, itwould certainly have been interesting to include electoral contests in Asia, Latin America and theMiddle East in the analysis. Like most of Africa, these regions are characterized by variations of‘neo-patrimonial rule’ that increase the stakes of electoral contests and undermine the functioning offormally democratic institutions (see Helmke and Levitsky, 2004). In addition, Africa is by no meansthe only continent where electoral violence erupts on a regular basis (see e.g. Norris, 2012). Futureresearch on electoral violence should therefore aim to expand the geographical scope of this study toinclude other parts of the developing world.

The systematic study of electoral violence is still very much in its infancy. In order to improve ourunderstanding of why some elections turn violent while others remain peaceful, we need to gathermore fine-grained and cross-nationally comparable data on election-related social conflict. The task offurther investigating the phenomenon of electoral violence should however not only be undertaken forthe sake of academic curiosity, but it should also be motivated by the desire to end the gross violationsof human rights that go hand in hand with electoral violence.

52

Bibliography

Arbetman-Rabinowitz, M. and K. Johnson: 2007, ‘Relative political capacity: empirical and theoreticalunderpinnings’. Sentia Group, Claremont, California.

Arriola, L. and C. Johnson: 2012, ‘Election Violence in Democratizing States’. Unpublishedmanuscript.

Aspinwall, M. D. and G. Schneider: 2000, ‘Same menu, seperate tables: The institutionalist turn inpolitical science and the study of European integration’. European Journal of Political Research38(1), 1–36.

Balch-Lindsay, D., A. J. Enterline, and K. A. Joyce: 2008, ‘Third-party intervention and the civil warprocess’. Journal of Peace Research 45(3), 345–363.

Banks, A. S.: 2011, ‘Cross-National Time-Series Data Archive, 1815-2011’. Retrieved on 20.10.14.

Basedau, M.: 2011, ‘Managing ethnic conflict: The menu of institutional engineering’. GIGA WorkingPaper No. 171.

Bayart, J.-F.: 1993, The state in Africa: the politics of the belly. London: Longman.

Beck, T., G. Clarke, A. Groff, P. Keefer, and P. Walsh: 2001, ‘New tools in comparative politicaleconomy: The Database of Political Institutions’. The World Bank Economic Review 15(1), 165–176.

Benoit, K.: 2004, ‘Models of electoral system change’. Electoral studies 23(3), 363–389.

Berman, B. J.: 1998, ‘Ethnicity, patronage and the African state: the politics of uncivil nationalism’.African Affairs 97(388), 305–341.

Birch, S.: 2007, ‘Electoral systems and electoral misconduct’. Comparative Political Studies 40(12),1533–1556.

Birch, S.: 2008, ‘Electoral institutions and popular confidence in electoral processes: A cross-nationalanalysis’. Electoral Studies 27(2), 305–320.

Birnir, J. K.: 2007, Ethnicity and electoral politics. Cambridge University Press.

Blais, A. and L. Massicotte: 1997, ‘Electoral formulas: a macroscopic perspective’. European Journalof Political Research 32(1), 107–129.

Blanton, R., T. D. Mason, and B. Athow: 2001, ‘Colonial style and post-colonial ethnic conflict inAfrica’. Journal of Peace Research 38(4), 473–491.

Bogaards, M.: 2013, ‘The Choice for Proportional Representation: Electoral System Design in PeaceAgreements’. Civil Wars 15(sup1), 71–87.

Borzyskowski, I. V.: 2013, ‘Sore Losers? International Condemnation and Domestic Incentives forPost-Election Violence’.

53

Brambor, T., W. R. Clark, and M. Golder: 2006, ‘Understanding interaction models: Improvingempirical analyses’. Political analysis 14(1), 63–82.

Brancati, D.: 2006, ‘Decentralization: Fueling the fire or dampening the flames of ethnic conflict andsecessionism?’. International Organization 60(03), 651–685.

Bratton, M.: 2007, ‘Formal versus informal institutions in Africa’. Journal of Democracy 18(3),96–110.

Bratton, M.: 2008, ‘Vote buying and violence in Nigerian election campaigns’. Electoral Studies 27(4),621–632.

Brubaker, R. and D. D. Laitin: 1998, ‘Ethnic and nationalist violence’. Annual Review of Sociologypp. 423–452.

Bueno De Mesquita, B., F. M. Cherif, G. W. Downs, and A. Smith: 2005, ‘Thinking inside the box:A closer look at democracy and human rights’. International Studies Quarterly 49(3), 439–458.

Bueno de Mesquita, B. and A. Smith: 2012, ‘Domestic explanations of international relation’. AnnualReview of Political Science 15, 161–181.

Buhaug, H.: 2010, ‘Dude, where’s my conflict? LSG, relative strength, and the location of civil war’.Conflict Management and Peace Science.

Calvert, R. L.: 1985, ‘The value of biased information: A rational choice model of political advice’.The Journal of Politics 47(02), 530–555.

Cederman, L. E., K. S. Gleditsch, and S. Hug: 2013, ‘Elections and ethnic civil war’. ComparativePolitical Studies 46(3), 387–417.

Cederman, L. E., S. Hug, and L. F. Krebs: 2010a, ‘Democratization and civil war: Empirical evidence’.Journal of Peace Research 47(4), 377–394.

Cederman, L.-E., A. Wimmer, and B. Min: 2010b, ‘Why do ethnic groups rebel? New data andanalysis’. World Politics 62(01), 87–119.

Chaturvedi, A.: 2005, ‘Rigging elections with violence’. Public Choice 125(1-2), 189–202.

Cheibub, J. A., J. Hays, and B. Savun: 2012, ‘Elections and Civil War in Africa’. UnpublishedManuscript.

Cingranelli, D. and M. Filippov: 2010, ‘Electoral Rules and Incentives to Protect Human Rights’. TheJournal of Politics 72(01), 1–15.

Cingranelli, D. L. and D. L. Richards: 2010, ‘The Cingranelli and Richards (CIRI) human rights dataproject’. Human Rights Quarterly 32(2), 401–424.

Clark, A. M. and K. Sikkink: 2013, ‘Information effects and human rights data: Is the good newsabout increased human rights information bad news for human rights measures?’. Human RightsQuarterly 35(3), 539–568.

Collier, P.: 2011, Wars, guns and votes: Democracy in dangerous places. Random House.

Collier, P. and A. Hoeffler: 2004, ‘Greed and grievance in civil war’. Oxford economic papers 56(4),563–595.

Collier, P., A. Hoeffler, and M. Söderbom: 2008, ‘Post-conflict risks’. Journal of Peace Research 45(4),461–478.

Collier, P. and P. Vicente: 2012, ‘Violence, bribery, and fraud: the political economy of elections inSub-Saharan Africa’. Public Choice 153(1-2), 117–147.

54

Collier, P. and P. C. Vicente: 2008, ‘Votes and Violence: Experimental Evidence from a Nigerian Elec-tion.’. In: Economics Series Working Papers WPS/2008-16. Department of Economics (Universityof Oxford).

Colombo, A., O. D’Aoust, and O. Sterck: 2014, ‘From Rebellion to Electoral Violence: Evidence fromBurundi’. CSAE Working Paper WPS/2014-20.

Conrad, C. R. and W. H. Moore: 2010, ‘What stops the torture?’. American Journal of PoliticalScience 54(2), 459–476.

Cox, G. W.: 1997, Making votes count: strategic coordination in the world’s electoral systems. Cam-bridge Univ Press.

Dafoe, A., J. R. Oneal, and B. Russett: 2013, ‘The democratic peace: Weighing the evidence andcautious inference’. International Studies Quarterly 57(1), 201–214.

Dahl, R. A.: 1973, Polyarchy: participation and opposition. Yale University Press.

Davenport, C.: 1997, ‘From ballots to bullets: an empirical assessment of how national electionsinfluence state uses of political repression’. Electoral Studies 16(4), 517–540.

Davenport, C.: 1998, ‘Liberalizing Event or Lethal Episode?: An Empirical Assessment of How Na-tional Elections Affect the Suppression of Political and Civil Liberties’. Social Science Quarterly79(2), 321–340.

Davenport, C.: 2007, ‘State repression and political order’. Annu. Rev. Polit. Sci. 10, 1–23.

Davenport, C. and D. A. Armstrong: 2004, ‘Democracy and the violation of human rights: A statisticalanalysis from 1976 to 1996’. American Journal of Political Science 48(3), 538–554.

Daxecker, U.: 2012, ‘The cost of exposing cheating International election monitoring, fraud, and post-election violence in Africa’. Journal of Peace Research 49(4), 503–516.

Daxecker, U.: 2014, ‘All quiet on election day? International election observation and incentives forpre-election violence in African elections’. Electoral Studies 34, 232–243.

Daxecker, U. and G. Schneider: 2014, ‘Electoral Observers: The Implications of Multiple Monitors forElectoral Integrity’. In: P. Norris, R. Frank, and F. Martinez (eds.): Avancing Electoral Integrity.Oxford University Press.

Dercon, S. and R. Gutierrez-Romero: 2012, ‘Triggers and characteristics of the 2007 Kenyan electoralviolence’. World Development 40(4), 731–744.

DeRouen, K. and D. Sobek: 2004, ‘The dynamics of civil war duration and outcome’. Journal of PeaceResearch 41(3), 303–320.

Diamond, L. J.: 2002, ‘Thinking about hybrid regimes’. Journal of democracy 13(2), 21–35.

Diermeier, D. and K. Krehbiel: 2003, ‘Institutionalism as a Methodology’. Journal of theoreticalpolitics 15(2), 123–144.

Doyle, M. W.: 1986, ‘Liberalism and world politics’. American Political Science Review 80(4), 1151–1169.

Eck, K.: 2012, ‘In data we trust? A comparison of UCDP GED and ACLED conflict events datasets’.Cooperation and Conflict 47(1), 124–141.

Eifert, B., E. Miguel, and D. N. Posner: 2010, ‘Political competition and ethnic identification in Africa’.American Journal of Political Science 54(2), 494–510.

Esteban, J., L. Mayoral, and D. Ray: 2012, ‘Ethnicity and conflict: An empirical study’. The AmericanEconomic Review 102(4), 1310–1342.

55

Esteban, J. and D. Ray: 2008, ‘Polarization, fractionalization and conflict’. Journal of Peace Research45(2), 163–182.

Evans, P. B., D. Rueschemeyer, and T. Skocpol: 1985, Bringing the state back in. Cambridge UniversityPress.

Fearon, J. D.: 2003, ‘Ethnic and cultural diversity by country’. Journal of Economic Growth 8(2),195–222.

Fearon, J. D. and D. D. Laitin: 2003, ‘Ethnicity, insurgency, and civil war’. American political sciencereview 97(01), 75–90.

Finnemore, M.: 1996, ‘Norms, culture, and world politics: insights from sociology’s institutionalism’.International organization 50(02), 325–347.

Fisher, J.: 2002, ‘Electoral conflict and violence: A strategy for study and prevention’. IFES WhitePaper.

Fjelde, H. and I. De Soysa: 2009, ‘Coercion, Co-optation, or Cooperation? State Capacity and theRisk of Civil War, 1961-2004’. Conflict Management and Peace Science 26(1), 5–25.

Fjelde, H. and K. Höglund: 2015, ‘Electoral Institutions and Electoral Violence in Sub-Saharan Africa’.Forthcoming in British Journal of Political Science (FirstView Advance Online Publication).

Flores, T. E. and I. Nooruddin: 2012, ‘The effect of elections on postconflict peace and reconstruction’.The Journal of Politics 74(02), 558–570.

Frees, E. W.: 2004, Longitudinal and panel data: analysis and applications in the social sciences.Cambridge University Press.

Gandhi, J. and E. Lust-Okar: 2009, ‘Elections under authoritarianism’. Annual Review of PoliticalScience 12, 403–422.

Gates, S., H. Hegre, M. P. Jones, and H. Strand: 2006, ‘Institutional inconsistency and politicalinstability: Polity duration, 1800-2000’. American Journal of Political Science 50(4), 893–908.

Gerring, J., C. Moreno, and S. C. Thacker: 2005, ‘Are unitary systems better than federal systems?’.Unpublished manuscript.

Gerring, J., S. C. Thacker, and C. Moreno: 2008, ‘Are parliamentary systems better?’. ComparativePolitical Studies.

Gibney, M., L. Cornett, R. Wood, and P. Haschke: 2014, ‘Political Terror Scale 1976-2012’. Retrievedon 20.10.14.

Gleditsch, K. S. and M. D. Ward: 1999, ‘A revised list of independent states since the Congress ofVienna’. International Interactions 25(4), 393–413.

Gleditsch, N. P., P. Wallensteen, M. Eriksson, M. Sollenberg, and H. Strand: 2002, ‘Armed conflict1946-2001: A new dataset’. Journal of peace research 39(5), 615–637.

Goldsmith, A.: 2014, ‘Electoral Violence in Africa Revisited’. Terrorism and Political Violence (Oc-tober), 1–20.

Guelke, A.: 2000, ‘Violence and electoral polarization in divided societies: Three cases in comparativeperspective’. Terrorism and Political Violence 12(3-4), 78–105.

Gupta, S., L. De Mello, and R. Sharan: 2001, ‘Corruption and military spending’. European Journalof Political Economy 17(4), 749–777.

Gurr, T. R.: 1970, Why men rebel. Princeton University Press.

56

Hafner-Burton, E., S. Hyde, and R. Jablonski: 2010, ‘Terrorizing Freedom: When Governments UseRepression to Manipulate Elections’.

Hafner-Burton, E., S. D. Hyde, and R. S. Jablonski: 2012, ‘Surviving Elections: Election Violence andLeader Tenure’. APSA 2012 (January).

Hafner-Burton, E. M., S. D. Hyde, and R. S. Jablonski: 2014, ‘When Do Governments Resort toElection Violence?’. British Journal of Political Science 44(1), 149–179.

Hall, P. A.: 1986, ‘Governing the economy: The politics of state intervention in Britain and France’.

Hall, P. A. and R. C. Taylor: 1996, ‘Political science and the three new institutionalisms*’. Politicalstudies 44(5), 936–957.

Hansen, T. O.: 2012, ‘Transitional Justice in Kenya? An Assessment of the Accountability Process inLight of Domestic Politics and Security Concerns’. California Western International Law Journal42, 1–35.

Hanson, J. and R. Sigman: 2013, ‘Leviathan’s Latent Dimensions: Measuring State Capacity forComparative Political Research’. In: APSA 2011 Annual Meeting Paper.

Hay, C. and D. Wincott: 1998, ‘Structure, agency and historical institutionalism’. Political studies46(5), 951–957.

Hegre, H.: 2014, ‘Democracy and armed conflict’. Journal of Peace Research 51(2), 159–172.

Hegre, H., T. Ellingsen, S. Gates, and N. P. Gleditsch: 2001, ‘Toward a democratic civil peace?Democracy, political change, and civil war, 1816-1992’. American Political Science Review 95(1),33–48.

Hegre, H. and N. Sambanis: 2006, ‘Sensitivity analysis of empirical results on civil war onset’. Journalof conflict resolution 50(4), 508–535.

Helmke, G. and S. Levitsky: 2004, ‘Informal institutions and comparative politics: A research agenda’.Perspectives on politics 2(04), 725–740.

Hendrix, C. S.: 2010, ‘Measuring state capacity: Theoretical and empirical implications for the studyof civil conflict’. Journal of Peace Research 47(3), 273–285.

Heston, A., R. Summers, and B. Aten: 2012, ‘Penn World Table Version 7.1’. Center for InternationalComparisons of Production, Income and Prices at the University of Pennsylvania.

Hicken, A. D.: 2007, ‘How do rules and institutions encourage vote buying?’. In: F. C. Schaffer (ed.):Elections for sale: the causes and consequences of vote buying. Boulder, Colorado: Lynne RiennerPublishers., pp. 47–60.

Hilbe, J. M.: 2011, Negative binomial regression. Cambridge University Press.

Höglund, K.: 2009, ‘Electoral violence in conflict-ridden societies: concepts, causes, and consequences’.Terrorism and Political Violence 21(3), 412–427.

Höglund, K.: 2010, ‘Leaders, Violence and Political Power: Exploring Linkages between PoliticalLeadership and Electoral Violence’. In: ISA Annual Convention, New Orleans. pp. 1–19.

Horowitz, D. L.: 1985, Ethnic groups in conflict. Univ of California Press.

Horowitz, D. L.: 2003, ‘Electoral systems: A primer for decision makers’. Journal of Democracy 14(4),115–127.

Huber, J. D.: 2012, ‘Measuring ethnic voting: Do proportional electoral laws politicize ethnicity?’.American Journal of Political Science 56(4), 986–1001.

57

Huntington, S. P.: 1968, Political order in changing societies. Yale University Press.

Hyde, S. and N. Marinov: 2009, ‘National Elections Across Democracy and Autocracy: Putting the"Competitive" into Competitive Authoritarianism’. Unpublished Manuscript.

Hyde, S. and N. Marinov: 2012a, ‘Codebook for the National Elections across Democracy and Autoc-racy (NELDA) Dataset, Version 3’. Retrieved on 20.10.14.

Hyde, S. and N. Marinov: 2012b, ‘Which elections can be lost?’. Political Analysis 20(2), 191–210.

Hyde, S. and N. Marinov: 2014, ‘Information and self-enforcing democracy: The role of internationalelection observation’. International Organization 68(02), 329–359.

ICG: 2008, ‘Kenya in crisis’. Africa Report Nr.137 of the International Crisis Group.

Kelley, J. G.: 2012, Monitoring democracy: When international election observation works, and whyit often fails. Princeton University Press.

Klopp, J. M.: 2001, ‘Ethnic Clashes and Winning Elections: The Case of Kenya’s Electoral Despotism’.Canadian Journal of African Studies 35(3), 473–517.

Koelbe, T.: 1995, ‘Review Article: The New Institutionalism in Political Science and Sociology’.Comparative Politics 27(2), 231–243.

Kurtenbach, S. and A. Mehler: 2013, ‘Introduction: Institutions for Sustainable Peace? Determinantsand Effects of Institutional Choices in Divided Societies’. Civil Wars 15(sup1), 1–6.

Laakso, L.: 2007, ‘Insights into Electoral Violence in Africa’. In: M. Basedau, G. Erdmann, and A.Mehler (eds.): Votes, money and violence: political parties and elections in Sub-Saharan Africa.

Lacina, B.: 2006, ‘Explaining the severity of civil wars’. Journal of Conflict Resolution 50(2), 276–289.

Lake, D. A.: 1992, ‘Powerful pacifists: democratic states and war.’. American Political Science Review86(01), 24–37.

Lardeyret, G.: 1991, ‘The problem with PR’. Journal of Democracy 2(3), 30–35.

Lehoucq, F.: 2002, ‘Can parties police themselves? Electoral governance and democratization’. Inter-national Political Science Review 23(1), 29–46.

Levitsky, S. and L. Way: 2002, ‘The rise of competitive authoritarianism’. Journal of democracy 13(2),51–65.

Lijphart, A.: 1977, Democracy in plural societies: A comparative exploration. Yale University Press.

Lijphart, A.: 1995, Electoral Systems and Party Systems. Oxford University Press, USA.

Lijphart, A.: 2004, ‘Constitutional design for divided societies’. Journal of democracy 15(2), 96–109.

Lindberg, S. I.: 2008, Democracy and elections in Africa. JHU Press.

Linebarger, C. and I. Salehyan: 2012, ‘Elections and Social Conflict in Africa, 1990-2009’. In: 2012Annual Convention of The International Studies Association.

Long, J. S. and J. Freese: 2006, Regression models for categorical dependent variables using Stata.Stata press.

Mansfield, E. D. and J. Snyder: 2009, ‘Pathways to war in democratic transitions’. InternationalOrganization 63(02), 381–390.

Marshall, M. G., T. R. Gurr, and K. Jaggers: 2013, ‘Polity IV Project Dataset Users Manual’. Retrievedon 20.10.14.

58

Marshall, M. G. and K. Jaggers: 2002, ‘Polity IV project: Political regime characteristics and transi-tions, 1800-2002’.

Mason, T. D., J. P. Weingarten, and P. J. Fett: 1999, ‘Win, lose, or draw: predicting the outcome ofcivil wars’. Political Research Quarterly 52(2), 239–268.

Minority Rights Group: 2008, ‘Zimbabwe Overview’. World Directory of Minorities and IndigenousPeoples. Available at: http://www.minorityrights.org/4504/zimbabwe/zimbabwe-overview.html.

Norgaard, A. S.: 1996, ‘Rediscovering reasonable rationality in institutional analysis’. European Jour-nal of political research 29(1), 31–57.

Norris, P.: 2004, Electoral engineering: voting rules and political behavior. Cambridge University Press.

Norris, P.: 2012, ‘Why electoral malpractices heighten risks of electoral violence’. In: APSA 2012Annual Meeting Paper.

Onapajo, H.: 2014, ‘Violence and Votes in Nigeria: The Dominance of Incumbents in the Use ofViolence to Rig Elections’. Africa Spectrum pp. 27–51.

Opitz, C., H. Fjelde, and K. Höglund: 2013, ‘Including peace: the influence of electoral managementbodies on electoral violence’. Journal of Eastern African Studies 7(4), 713–731.

Ostrom, E.: 1991, ‘Rational Choice Theory and Institutional Analysis: Toward Complementarity.’.American political science review 85(01), 237–243.

Pausewang, S., K. Trondvall, and L. Aalen: 2003, Ethiopia Since The Derg: A Decade Of DemocraticPretension And Performance. Zed Books.

Person, T. and G. Tabellini: 2004, ‘Constitutions and economic policy’. Journal of Economic Perspec-tives pp. 75–98.

Poe, S. C. and C. N. Tate: 1994, ‘Repression of human rights to personal integrity in the 1980s: aglobal analysis.’. American Political Science Review 88(04), 853–872.

Poe, S. C., C. N. Tate, and L. C. Keith: 1999, ‘Repression of the Human Right to Personal IntegrityRevisited: A Global Cross-National Study Covering the Years 1976-1993’. International StudiesQuarterly 43(2), 291–313.

Pospieszna, P. and G. Schneider: 2013, ‘The Illusion of Peace Through Power-Sharing: ConstitutionalChoice in the Shadow of Civil War’. Civil Wars 15(sup1), 44–70.

Raleigh, C., A. Linke, H. Hegre, and J. Karlsen: 2010, ‘Introducing acled: An armed conflict locationand event dataset special data feature’. Journal of peace research 47(5), 651–660.

Rapoport, D. and L. Weinberg: 2000, ‘Elections and violence’. Terrorism and Political Violence12(3-4), 15–50.

Reilly, B.: 2002a, ‘Elections in post-conflict scenarios: Constraints and dangers’. International Peace-keeping 9(2), 118–139.

Reilly, B.: 2002b, ‘Electoral systems for divided societies’. Journal of Democracy 13(2).

Reilly, B., A. Reynolds, et al.: 1999, Electoral systems and conflict in divided societies. NationalAcademies Press.

Reynal-Querol, M.: 2002, ‘Ethnicity, political systems, and civil wars’. Journal of Conflict Resolution46(1), 29–54.

Reynolds, A., B. Reilly, and A. Ellis: 2005, Electoral system design: the new international IDEAhandbook. International Institute for Democracy and Electoral Assistance.

59

Richards, D. L.: 1999, ‘Perilous Proxy: Human Rights and the Presence of National Elections.’. SocialScience Quarterly 80(4), 648–65.

Richards, D. L. and R. D. Gelleny: 2007, ‘Good things to those who wait? National elections andgovernment respect for human rights’. Journal of Peace Research 44(4), 505–523.

Saideman, S. M., D. J. Lanoue, M. Campenni, and S. Stanton: 2002, ‘Democratization, PoliticalInstitutions, and Ethnic Conflict A Pooled Time-Series Analysis, 1985-1998’. Comparative PoliticalStudies 35(1), 103–129.

Salehyan, I., C. S. Hendrix, J. Hamner, C. Case, C. Linebarger, E. Stull, and J. Williams: 2012, ‘Socialconflict in Africa: A new database’. International Interactions 38(4), 503–511.

Schedler, A.: 2002, ‘The menu of manipulation’. Journal of democracy 13(2), 36–50.

Schneider, G. and N. Wiesehomeier: 2008, ‘Rules that matter: Political institutions and the diversity-conflict nexus’. Journal of Peace Research 45(2), 183–203.

Schultz, K. A.: 1998, ‘Domestic opposition and signaling in international crises’. American PoliticalScience Review 92(4), 829–844.

Selway, J. and K. Templeman: 2012, ‘The myth of consociationalism? Conflict reduction in dividedsocieties’. Comparative Political Studies 45(12), 1542–1571.

Shepsle, K. A.: 1989, ‘Studying Institutions Some Lessons from the Rational Choice Approach’. Journalof theoretical politics 1(2), 131–147.

Shin, D. C.: 1994, ‘On the third wave of democratization: A synthesis and evaluation of recent theoryand research’. World politics 47(01), 135–170.

Sikkink, K.: 1991, Ideas and Institutions: Developmentalism in Argentina and Brazil. Ithaca: CornellUniversity Press.

Singer, J. D. and M. Small: 1994, ‘Correlates of war project: International and civil war data, 1816-1992 (ICPSR 9905)’. Ann Arbor, MI: Inter-University Consortium for Political and Social Research.

Sisk, T. D.: 2008, ‘Elections in fragile states: Between Voice and Violence’. In: Conference Paper,International Studies Association.

Sobek, D.: 2010, ‘Masters of their domains: The role of state capacity in civil wars’. Journal of PeaceResearch 47(3), 267–271.

Straus, S. and C. Taylor: 2009, ‘Democratization and Electoral Violence in Sub-Saharan Africa, 1990-2007’. APSA 2009 Toronto Meeting Paper.

Sundberg, R. and E. Melander: 2013, ‘Introducing the UCDP georeferenced event dataset’. Journalof Peace Research 50(4), 523–532.

Sundstroem, A. and L. Waengnerud: 2014, ‘Corruption as an obstacle to women’s political represen-tation: Evidence from local councils in 18 European countries’. Party Politics.

Svolik, M. and S. Chernykh: 2013, ‘Third-Party Actors and the Success of Democracy: How ElectoralCommissions, Courts, and Observers Shape Incentives for Electoral Manipulation and Post-ElectionProtests’.

Teorell, J., N. Charron, S. Dahlberg, S. Holmberg, B. Rothstein, P. Sundin, and R. Svensson: 2013,‘The Quality of Government Dataset, Version 20Dec13, University of Gothenburg: The Quality ofGovernment Institute’.

Themner, L. and P. Wallensteen: 2013, ‘Armed Conflicts, 1946-2012’. Journal of Peace Research.

60

Thies, C. G.: 2010, ‘Of rulers, rebels, and revenue: State capacity, civil war onset, and primarycommodities’. Journal of Peace Research 47(3), 321–332.

Tsebelis, G.: 1990, Nested games: Rational choice in comparative politics. University of CaliforniaPress.

Tucker, J.: 2007, ‘Enough! Electoral fraud, collective action problems, and post-communist coloredrevolutions’. Perspectives on Politics 5(03), 535–551.

UNDP: 2011, ‘Understanding electoral violence in Asia’. Technical report, United Nations DevelopmentProgramme.

Van de Walle, N.: 2003, ‘Presidentialism and clientelism in Africa’s emerging party systems’. TheJournal of Modern African Studies 41(02), 297–321.

Vreeland, J. R.: 2008, ‘The effect of political regime on civil war: unpacking anocracy’. Journal ofConflict Resolution 52(3), 401–425.

Wagner, W. and S. Dreef: 2014, ‘Ethnic Composition and Electoral System Design: DemographicContext Conditions for Post-conflict Elections’. Ethnopolitics 13(3), 288–307.

Wantchekon, L. and Z. Neeman: 2002, ‘A theory of post-civil war democratization’. Journal ofTheoretical Politics 14(4), 439–464.

Wasserman, J. and E. Jaggard: 2007, ‘Electoral violence in mid nineteenth-century England andWales’. Historical research 80(207), 124–155.

WB: 2013, World Development Indicators 2013. World Bank Publications.

Weber, M.: 1958, Politics as a Vocation in From Max Weber: Essays in Sociology, ed. HH Gerth,trans. C. Wright Mills. New York: Oxford University Press.

Wig, T., H. Hegre, and P. Regan: 2014a, ‘Updated Data on Institutions and Elections 1960-2012:Presenting the IAEP dataset version 2.00’. Unpublished Manuscript.

Wig, T., H. Hegre, and P. Regan: 2014b, ‘Users’ Manual for the IAEP Dataset V 2.0’. UnpublishedManuscript.

Wilkinson, S. I.: 2006, Votes and violence: Electoral competition and ethnic riots in India. CambridgeUniversity Press.

Young, J. K.: 2009, ‘State capacity, democracy, and the violation of personal integrity rights’. Journalof Human Rights 8(4), 283–300.

Zierhofer, W.: 2005, ‘State, power and space’. Social geography 1(1), 29–36.

61

Appendix

62

Table 8: Election-specific factors and pre-election violence(1) (2) (3) (4)

Governmentviolence

Oppositionviolence

Governmentviolence

Oppositionviolence

1. Majoritarian 1.052∗∗∗ 1.056∗∗∗(0.406) (0.376)

2. Mixed 0.833∗ -0.715(0.428) (0.584)

District size (log) -0.242∗ -0.590∗∗∗(0.136) (0.219)

Civil war -0.407 0.142 -0.527 0.150(0.460) (0.303) (0.547) (0.309)

Population (log, lag) 0.732∗∗∗ 0.721∗∗∗ 0.712∗∗∗ 0.635∗∗∗(0.151) (0.097) (0.149) (0.116)

GDP pc (log, lag) -0.118 0.415∗∗∗ 0.128 0.632∗∗∗(0.244) (0.146) (0.277) (0.234)

Polity (lag) 0.0192 0.00412 0.00940 0.00341(0.030) (0.031) (0.033) (0.029)

Polity squared (lag) -0.00668 -0.0116∗ -0.0120 -0.0170∗∗(0.007) (0.007) (0.009) (0.008)

Ethnic fractionalization -0.180 -0.510 0.194 1.075(0.777) (1.028) (0.842) (0.986)

Ethnic polarization 0.443 0.560 0.648 1.199∗∗∗(0.416) (0.355) (0.534) (0.391)

Number of elections -0.0836∗ 0.0416 -0.0814∗ -0.000383(0.045) (0.035) (0.043) (0.036)

Competitive elections -0.193 0.437 -0.283 0.359(0.496) (0.576) (0.490) (0.577)

GDP growth (lag) -0.0537∗ -0.0945∗∗∗ -0.0577∗ -0.0968∗∗∗(0.030) (0.025) (0.030) (0.019)

Previous gov. violence -0.0129 0.159(0.296) (0.326)

Previous opp. violence 0.149 0.209(0.340) (0.353)

Constant -6.879∗∗∗ -12.16∗∗∗ -7.567∗∗∗ -12.54∗∗∗(2.523) (2.445) (2.793) (3.284)

ln(alpha) 0.189 -2.424 0.350 -1.855(0.455) (2.864) (0.430) (1.655)

Observations 277 277 258 258Clusters 45 45 44 44Robust standard errors clustered by country in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

63

Table 9: Election-specific factors and pre-election violence (cont.)(1) (2) (3) (4)

Governmentviolence

Oppositionviolence

Governmentviolence

Oppositionviolence

Unitary 0.927∗ 0.297(0.517) (0.348)

1. Semi-parliamentary 0.842∗∗ -0.335(0.398) (0.707)

2. Parliamentary -0.867 -0.0131(0.664) (0.620)

1. Majoritarian 1.219∗∗∗ 1.136∗∗∗ 1.401∗∗∗ 0.993∗∗(0.421) (0.439) (0.424) (0.444)

2. Mixed 0.862∗ -0.750 0.998∗∗ -0.696(0.459) (0.589) (0.418) (0.571)

Civil war -0.392 0.149 -0.341 0.143(0.414) (0.285) (0.429) (0.311)

Population (log, lag) 0.857∗∗∗ 0.774∗∗∗ 0.679∗∗∗ 0.729∗∗∗(0.189) (0.135) (0.146) (0.103)

GDP pc (log, lag) -0.124 0.433∗∗∗ -0.155 0.409∗∗∗(0.260) (0.163) (0.248) (0.146)

Polity (lag) 0.0353 0.00778 0.0385 0.000123(0.033) (0.033) (0.033) (0.033)

Polity squared (lag) -0.00892 -0.0122∗ -0.00953 -0.00997(0.007) (0.007) (0.007) (0.007)

Ethnic fractionalization 0.0263 -0.286 -0.270 -0.564(0.696) (1.057) (0.756) (1.020)

Ethnic polarization 0.614 0.623∗ 0.637 0.479(0.413) (0.377) (0.427) (0.376)

Number of elections -0.0677 0.0550 -0.109∗∗ 0.0475(0.046) (0.038) (0.048) (0.039)

Competitive elections -0.185 0.433 -0.145 0.439(0.477) (0.555) (0.485) (0.587)

GDP growth (lag) -0.0511∗ -0.0939∗∗∗ -0.0303 -0.0998∗∗∗(0.029) (0.026) (0.034) (0.024)

Previous gov. violence -0.139 -0.0229(0.279) (0.298)

Previous opp. violence 0.120 0.158(0.357) (0.354)

Constant -9.216∗∗∗ -13.35∗∗∗ -6.277∗∗∗ -12.14∗∗∗(2.700) (3.071) (2.431) (2.502)

ln(alpha) -0.102 -2.734 0.0925 -2.417(0.600) (3.969) (0.467) (2.921)

Observations 273 273 277 277Clusters 45 45 45 45Robust standard errors clustered by country in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

64

Table 10: Colonial origin, electoral systems and pre-election violence(1) (2)

Government violence Opposition violence

1. Majoritarian 0.762∗ 0.702∗(0.457) (0.418)

2. Mixed 0.745∗ -1.399(0.444) (0.892)

Civil war -0.366 0.342(0.368) (0.243)

Population (log, lag) 0.565∗∗∗ 0.496∗∗∗(0.142) (0.116)

GDP pc (log, lag) -0.197 0.348∗∗(0.243) (0.157)

Polity (lag) 0.0185 0.0197(0.032) (0.031)

Polity squared (lag) -0.00931 -0.0193∗∗∗(0.007) (0.007)

Ethnic fractionalization 0.405 -0.495(0.981) (1.038)

Ethnic polarization 0.669 0.626∗(0.487) (0.363)

Number of elections -0.0436 0.0972∗(0.050) (0.054)

1. Former British colony 0.661 1.238∗∗(0.675) (0.485)

2. Former French colony -0.250 0.277(0.772) (0.712)

Previous gov. violence -0.0424(0.284)

Previous opp. violence 0.185(0.348)

Constant -5.877∗∗ -10.23∗∗∗(2.606) (2.205)

ln(alpha) 0.0858 -2.785(0.460) (3.858)

Observations 277 277Clusters 45 45Robust standard errors clustered by country in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

65

Table 11: Election-specific factors and post-election violence(1) (2) (3) (4)

Governmentviolence

Oppositionviolence

Governmentviolence

Oppositionviolence

1. Majoritarian 3.035∗∗∗ 16.72∗∗∗(1.137) (0.632)

2. Mixed 0.543 15.98∗∗∗(0.871) (0.487)

District size (log) -0.890∗∗∗ -0.307(0.299) (0.208)

Civil war -0.103 -1.492 -0.286 -1.105(1.153) (1.161) (1.127) (1.198)

Population (log, lag) 0.00816 0.679∗∗ 0.211 0.526∗∗(0.317) (0.283) (0.294) (0.250)

GDP pc (log, lag) -1.220∗∗∗ 0.452 -1.299∗∗∗ 0.839∗(0.328) (0.396) (0.456) (0.479)

Polity (lag) 0.273∗∗∗ 0.0485 0.157∗∗ 0.108(0.066) (0.075) (0.065) (0.090)

Polity squared (lag) -0.0357∗∗∗ -0.0168 -0.0349∗∗∗ -0.0374∗∗(0.013) (0.016) (0.012) (0.015)

Ethnic fractionalization -7.111∗∗∗ 0.0756 -5.258∗∗ 0.790(2.055) (2.021) (2.095) (2.215)

Ethnic polarization -1.311∗ -0.606 -1.263 -0.812(0.704) (0.826) (1.044) (0.850)

Number of elections -0.108 0.0855 -0.155 0.0660(0.089) (0.088) (0.114) (0.082)

Electoral fraud -0.513 0.322 -0.711 0.279(0.548) (0.584) (0.529) (0.610)

Results canceled -19.89∗∗∗ 1.072∗∗ -15.70∗∗∗ 1.177∗∗(1.000) (0.495) (0.574) (0.493)

Incumbent lost 2.001∗∗∗ -0.873 1.645∗∗∗ -0.629(0.721) (0.727) (0.560) (0.815)

Previous gov. violence 0.928 0.666(0.738) (0.804)

Previous opp. violence -0.956 -0.593(0.600) (0.687)

Constant 9.624∗∗ -28.32∗∗∗ 10.79∗∗∗ -13.21∗∗(4.566) (5.675) (4.075) (6.189)

ln(alpha) -0.119 -16.88∗∗∗ 0.715 -15.57∗∗∗(1.063) (0.649) (0.494) (1.098)

Observations 167 167 150 150Clusters 39 39 38 38Robust standard errors clustered by country in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

66

Table 12: Election-specific factors and post-election violence (cont.)(1) (2) (3) (4)

Governmentviolence

Oppositionviolence

Governmentviolence

Oppositionviolence

State capacity (lag) -8.242 -4.669(6.393) (3.094)

1. Semi-parliamentary 0.879 -14.80∗∗∗

(0.875) (1.069)

2. Parliamentary -1.348∗∗ -15.62∗∗∗

(0.661) (0.714)

1. Majoritarian 3.637∗∗∗ 16.90∗∗∗ 3.352∗∗ 16.15∗∗∗

(1.010) (0.638) (1.358) (0.798)

2. Mixed -13.13∗∗∗ 15.97∗∗∗ 0.885 15.36∗∗∗

(3.804) (0.630) (0.985) (0.572)

Civil war -17.84∗∗∗ -1.211 0.391 -1.237(1.685) (1.246) (1.181) (1.042)

Population (log, lag) -0.129 0.594 0.0579 0.632∗∗

(0.533) (0.470) (0.321) (0.257)

GDP pc (log, lag) -1.334∗ 0.606 -1.167∗∗∗ 0.408(0.762) (0.439) (0.340) (0.380)

Polity (lag) 0.291∗∗ 0.00887 0.336∗∗∗ 0.0493(0.116) (0.077) (0.098) (0.078)

Polity squared (lag) -0.0394∗∗∗ -0.00535 -0.0384∗∗∗ -0.0180(0.014) (0.015) (0.014) (0.016)

Ethnic fractionalization -9.018 0.281 -7.889∗∗∗ 0.0211(6.104) (1.583) (2.462) (1.887)

Ethnic polarization -3.104 0.169 -1.081 -0.593(2.771) (0.901) (0.677) (0.807)

Number of elections -0.297 -0.0101 -0.157∗ 0.0871(0.186) (0.112) (0.091) (0.085)

Electoral fraud -0.250 0.663 -0.587 0.225(0.814) (0.649) (0.598) (0.549)

Results canceled -14.83∗∗∗ 1.456∗∗ -10.98∗∗∗ 1.070∗∗

(1.101) (0.648) (1.138) (0.474)

Incumbent lost 1.880∗ -0.703 2.119∗∗ -0.737(1.041) (0.732) (0.829) (0.692)

Previous gov. violence 0.578 1.049(1.043) (0.757)

Previous opp. violence -1.352∗ -0.906(0.726) (0.554)

Constant 17.61∗ -26.78∗∗∗ 9.269∗∗ -26.84∗∗∗

(9.758) (8.196) (4.522) (5.195)

ln(alpha) -0.790 -66.74 -0.197 -24.96(3.619) (.) (1.099) (.)

Observations 133 133 167 167Clusters 30 30 39 39Robust standard errors clustered by country in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01

67