Dangerous elections
A study on electoral violence and clientelism
Erik Forsberg Bachelor thesis
Department of Peace and Conflict Research Fall semester 2018
Supervisor: Espen Geelmuyden Rød
Abstract
Why do some elections spark violence whilst others do not? That is a question that has gained increased interest from scholars during the last few years. However, because of the field’s relative novelty, and despite the vast literature on democratization and civil war, it is still a question that is not fully comprehended. In this thesis, a theory claiming that clientelism should increase the risk of electoral violence is presented. It is argued that clientelism increases the stakes of elections by increasing the costs of losing and the rewards of winning them. This should also increase the risk that electoral violence is employed as a strategy in elections. It is further argued that this relationship should be present both when an incumbent is partaking in the election and when no incumbent does so. It is further argued that violence both prior to and after elections should correlate positively with clientelism. The theory is tested by a series of regression models. It is found that clientelism only has a consistently positive and statistical significant relationship with post-election violence.
Furthermore, evidence is found disproving the hypothesis that electoral violence is positively correlated with clientelism regardless of whether an incumbent partakes in the election or not. On the other hand, evidence is found that a condition for the proposed theoretical mechanism is that an incumbent is running for office. The thesis contributes to the knowledge about electoral violence in general, but also to the vast literature on democratization in Africa.
Content
1. Introduction ... 4
2. Previous research ... 5
Definitions: ... 5
Literature review: ... 6
3. Theory ... 9
Electoral violence: ... 9
Clientelism: ... 10
Incumbency: ... 11
4. Research Design ... 11
Case selection: ... 12
Data and source criticism: ... 12
Operationalizations, validity and reliability: ... 14
Control variables: ... 15
Statistical models: ... 16
Scope and limitations: ... 16
5. Results and analysis ... 17
Alternative explanations: ... 22
6. Conclusion ... 23
7. References ... 25
8. Appendix ... 27
Figures and Tables
Table 1: Hypothesis 1... 18Table 2: Hypothesis 2... 20
Table 3: Distribution of electoral violence, pre- and post-election violence ... 21
Table 4: Pre-election violence... 27
Table 5: Post-election violence... 28
Table 6: Collinearity test... 29
4
1. Introduction
Since 1990, there has been a remarkable transition from single-party rule to multiparty elections in African politics. This shift has been celebrated by many as an indicator that Africa is on a path towards democratization, liberalization and eventually economic and social development. However, the path has not been problem free. Most pressing of these problems is arguably electoral violence of which, there have been several noteworthy instances. The election in Kenya in the summer of 2017, is the most recent one of these. These instances do not only cost lives and damage property, it also undermines the legitimacy of elections and thereby democracy itself. To understand what causes electoral violence is thus of upmost importance, both to minimize violence in general, but also for the democratization process in Africa. Yet, studies on the subject are quite few. Therefore, this thesis aims at answering the question: why do some elections spark violence, whilst others do not?
There are a few previous studies that examine this exact question, both qualitatively and
quantitatively. It has been argued that election fraud, election observers, high probability of regime change, low GDP per capita, illiberalism and majoritarian electoral rules cause a higher risk of electoral violence. However, a factor that has received limited attention is clientelism, even though it has been argued to be the dominant institutional form in Africa (Taylor et al, 2017; Van de Walle, 2003). Clientelism is an informal political institution, consisting of interpersonal relationships and exchanges between patrons and their clients (Van de Walle, 2003). Taylor et al (2017) examine the subject theoretically, arguing that clientelism should cause electoral violence if an incumbent is running for office, however they do not analyze opposition actors, nor supporters of either incumbents or opposition actors. Furthermore, they do not measure clientelism in their empirical analysis and it is therefore not possible to know whether their theorized causal mechanism is true or not. This thesis will therefore examine how clientelism affects the risk of electoral violence both theoretically and empirically. This will be done both by examining how clientelism affects the risk of both incumbents and opposition actors, using violence, as well as how clientelism affects their supporters’ incentives to use violence. The study will contribute empirically, by testing the relationship through a measurement of actual clientelist practices.
It is argued in this thesis that since electoral violence is a particularly costly strategy to use in
elections, it will only be employed if the stakes of the election are particularly high. Strong clientelist institutions should increase the stakes of elections for all parties, by increasing the costs of losing power and increasing the rewards of winning it. Thereby, electoral violence should be more common
5 in states with strong clientelist institutions. It is further argued that the risk of both pre- and post- election violence should increase, since violence can be used both before and after elections to affect the outcome of it. Lastly, it is argued that the mechanism should present, both when an incumbent is partaking in the election as well as when no incumbent is, since states with strong clientelist
institutions, usually experience it on all levels of society. There should thereby be existing patron- client relationships between others than incumbents and their clients.
Cote d’Ivoire experienced great levels of electoral violence. This violence has included, amongst other things, violence used to drive away people from their land and reallocate it to other groups to gain their support in the election. However, the allocation of land is not the only clientelist practice that has been present in Cote d’Ivoire (Boone and Krieger, 2012). Between its independence and 1987 the amount of cabinet positions increased by a factor 2.5, indicating that allocation of
government positions was also used to gain loyalty (Bratton and van de Walle, 1997, p.66). It could thus be that a lot of people has been dependent on a certain outcomes in the elections to secure the positions they held.
In this thesis, it is found that clientelism only causes electoral violence when an incumbent is running for office, thus supporting the theorized causal mechanism proposed by Taylor et al (2017). It is also found that clientelism does not seem to covary with pre-election violence in itself, but have a positive effect on post-election violence. These results are somewhat contradictory to the theory and
hypotheses of this thesis.
This thesis will proceed firstly with definitions of the two main concepts, clientelism and electoral violence and a review of the previous literature concerning these two concepts. The theory and hypotheses will then be presented, followed by a description and analysis of the research design, including the operationalizations of the variables, case selection, data collection and methods. Lastly, the results will be presented and analyzed, followed by a discussion on the implications of the study for the academic field.
2. Previous research
Definitions:
Clientelism is in this thesis defined as a system where a patron buys support by his or her clients by providing various forms of patronage to set clients. It means that the patron invests resources to keep the loyalty and support needed to remain in power or to gain further power, and that his clients
6 are dependent on him to remain in power, in order to secure the patronage they have gained (Van de Walle and Bratton, 1997, p.87-91). Clientelism can also be defined from the clients’ perspective, as an informal institution where clients express loyalty to a patron in exchange for various forms of
patronage (Hyden, 2013). A clientelist practice can for example be a president offering cabinet seats to the leader of a group in society, in exchange for support from that group.
Electoral violence will in this thesis be defined, in accordance with Höglund (2009), as violence committed to influence electoral outcomes. It thus includes violence that occurs before, during and after elections and it can be committed by a variety of actors, such as state actors, rebels, opposition parties, militias et cetera, as long as it is meant to influence electoral outcomes (ibid).
Literature review:
The literature on reasons, consequences and dynamics of armed conflict is huge. However, the literature on electoral violence is limited and the research examining electoral violence as a phenomenon separated from other forms of political violence, is slim (Bekoe, 2012; Taylor et al, 2017; Höglund, 2009). Elections overall and electoral processes in conflict countries has on the other hand gained huge academic
attention. Furthermore, the processes leading to, and enhancing, democratization, have been studied extensively (Taylor et al, 2017). Considering the impact of electoral violence on democratization and its fatal consequences for electoral credibility, this research gap is surprising. However, since 2009 the body of literature on electoral violence has increased steadily. Both books and articles have been written, examining the subject theoretically, as well as through both case studies and quantitative analyses. In this section, the literature on electoral violence will first be reviewed separately, after which the literature on clientelism is reviewed. Lastly, the literature concerning both phenomena is presented.
Höglund (2009) is one of the first articles examining electoral violence from a theoretical perspective. She argues that it is vital for our understanding of electoral violence to conceptualize it as a specific sub- category of political violence. She writes that in order to understand the causes and consequences of violence, we have to be able to distinguish it from other forms of political violence. What electoral
violence is should therefore, as with other forms of violence, be determined by its actors, timing, activities and motives. She distinguishes between electoral violence committed by the government and by
opposition actors, as well as between pre- and post-election violence and she identifies three enabling conditions. Heavy reliance on patron-client relationships, particular electoral institutions and democracy in itself, can all affect the incentive structure, making violence more profitable as an electoral strategy (Ibid).
7 Since 2009 a few articles have been written, examining electoral violence, quantitatively. Strauss and Taylor (2012) have created a dataset on electoral violence between 1990 and 2008 in sub-Saharan Africa.
In their first analysis of the data, they find that of the 221 elections that are included in the dataset, 42 percent were non-violent, 38 percent produced low levels of violence and 20 percent experienced large- scale violence. These results are consistent throughout the time period examined, even though the level of democracy has risen steadily since 1990. Furthermore, they find that most violence takes place before the elections and that most violence is committed by the incumbent and his supporters (ibid). They argue that the variation between different countries and different elections, is not fully comprehended and that a critical next step is to examine this variation; why some elections cause violence whilst others do not.
This is examined by Daxecker (2012), who claims that elections where election fraud and international election monitoring is present are more likely to produce electoral violence. Hafner-Burton et al (2013), on the other hand, examine what causes incumbents to use violence as an electoral strategy. They argue that the bigger the risk of the incumbent losing power, the bigger risk is it that he or she will resort to pre-election violence since it increases the incumbent’s chance of winning. Salehyan and Linebarger (2015) outline a few structural conditions, under which electoral violence is more common. Additionally, they examine how certain characteristics of the elections themselves affect the risk of electoral violence.
They find that electoral violence is more common in autocracies than democracies, that illiberal elections are more likely to spark violence than free and fair elections and that neither current nor recent armed conflict affects the risk of electoral violence (ibid).Fjelde and Höglund (2016) find that majoritarian electoral rules cause electoral violence and argue that the effect should be stronger if clientelism is
present. They argue that the winner-takes-all nature of elections with majoritarian electoral rules raises the stakes of elections, making electoral violence a more feasible strategy.
While all these possible explanations deserve merit, and must be considered when examining the causes of electoral violence, the focus of this paper is the effect of clientelism on the risk of electoral violence.
Clientelism has gained some attention in the literature on electoral violence and there is a vast literature on the dynamics and consequences of clientelism in African politics overall.
Clientelism is a prevalent political institution in Africa. Taylor et al (2017) writes that clientelistic
presidential systems characterize most African states. The importance of patron-client relationships is also emphasized in Bratton and van de Walle (1997, p.64), Hyden (2013, p.87-91) and van de Walle (2003) amongst others. Bratton and van de Walle (1997, p.65) argue that strongmen in Africa rely upon favors from a patron, a person in power, such as distribution of public resources, and in return they provide
8 political support for that patron. These clientelist practices include systematic tax fraud, where some companies are exempted from tax or import duty, in exchange for loyalty. They continue by arguing that it severely damages economic development and democratic credibility (ibid). Hyden (2013, p.100-115) also argues that clientelist practices in African countries damage their democratic credibility. He claims that rulers in Africa see their interests as tied to local communities, rather than to constitutions and laws.
They depend upon personal relationships and exchanges. Their actions are thus contradictory to the democratic values of accountability and transparency (ibid).
Van de Walle (2003) argues that much of these clientelist practices occur along ethnic lines. Furthermore, he argues that it is often legitimized by claiming that it serves a community purpose and is not meant to yield personal enrichment. A position of power is thus not only valued because of what resources it provides for that person, but also for what resources it can produce for his or her family, clan and ethnic group. However, little of the resources gained through patronage, actually trickles down to the lower levels of clients, but instead stays with the elite. Despite this, many African countries experience a high degree of strategic voting. This is both demonstrated by vote buying, and by voting patterns, where citizens vote for candidates of their own ethnicity (ibid).
A potential consequence of clientelism is electoral violence. A few case studies on electoral violence have been done, arguing that clientelism has caused electoral violence in particular cases. Boone and Krieger (2012) examine the causes of electoral violence in Cote d’Ivoire and Zimbabwe, arguing that incumbents used land rights as patronage to mobilize support in constituencies they risked losing. They used the selective allocation of land, like the selective allocation of government positions, to offer benefits to potential voters in exchange for political support. In both cases, government officials and pro-
government militias used violence to make the land transfer a reality. Boone and Krieger (2012) argue that the conditions which allow for this tactic, are found in many African countries. Mueller (2012) also claims that clientelist practices caused electoral violence in Kenya in elections from 1992 to 2007. She argues that lack of institutional checks on the president and the expectations of benefits from his clients is instrumental in the distribution of public resources. This can explain why political leaders and followers are willing to use electoral violence as a strategy to secure their power (ibid).
Furthermore, Taylor et al (2017) study the relationship between clientelism and electoral violence quantitatively. They find that elections where an incumbent is running for office are more likely to spark violence, and argue that this is due to clientelism. Since clientelism raises the stakes of an
9 election, electoral violence should be more likely in countries with strong clientelistic institutions.
However, in elections where the incumbent is not running for office, existing patron-client
relationships are more uncertain, since none of the candidates have successfully rewarded loyalty in the past. Therefore, the presumed violent actors are less likely to be willing to commit violence for their patron, and the patron is less likely to have enough confidence in his clients to call on them to commit violence (ibid).
As there are noquantitative studies on the relationship between clientelism and electoral violence that measure clientelism, we cannot determine the actual relationship between clientelism and electoral violence. The aim of this thesis is therefore to examine the causal mechanisms proposed by Taylor et al (2017), as well as examine how clientelism affects the risk of electoral violence,
independent from who runs for office, both theoretically and empirically, in a large-N study.
3.
Theory
Electoral violence:
Elections are intrinsically conflictual, as several parties are competing for political power. If one accepts the definition of electoral violence that is used in this thesis, that electoral violence is violence used to in some way influence election outcomes, and that all strategies in elections has the purpose of in some way affecting the results of the election, it follows that violence is a strategy amongst others in elections.
However, the heaviness of the competitive tactics can vary from peaceful and healthy political debate to violence (Salehyan and Linebarger, 2015). One can thereby see violence as the top of a ladder of
increasingly heavy tactics that can be used in elections. It can be an effective strategy to reach the
preferred election outcomes since it can be used to intimidate opposing parties, making them concede or defer from campaigning activities. It can also be a way of proving to citizens that one is the stronger candidate and thereby that one will be able to protect the citizens from internal and external threats.
Furthermore, it can be used to coerce voters into supporting oneself. Lastly, violence can be used in the post-election period, either to protest the election result and thereby weaken the legitimacy of the election which could for example cause reelection, or to crack down on such protest. It can thus be used both to change election results and to minimize the risk of election results to be changed (Taylor et al, 2017).
Electoral violence is however, a particularly costly electoral strategy. Electoral violence overall weakens the legitimacy of an election and thereby the democratic credibility of the subsequent government.
Furthermore, violence committed by a government will decrease the likelihood of international actors
10 supporting the country, both in diplomacy and with aid. Similarly, violence committed by opposition parties, will decrease the likelihood of international actors supporting them. Moreover, violence both by government actors and opposition actors, risks turning the population or parts of it against them, since they might be negatively affected by the violence. Lastly, all forms of violence are costly, both since it requires resources such as weapons and soldiers, and since it costs lives (Bekoe, 2012, p.243).
Assuming that actors in elections are rational, a strategy will only be committed during circumstances where the expected reward is bigger than the expected cost. Since electoral violence is a particularly costly strategy, it is expected that it will only be used, either when the expected reward of using it is relatively high, or when the expected cost of using it is relatively low.
Clientelism:
Clientelism is built upon mutually enforcing actions between patrons and clients. For clientelism to be maintained, the patron needs to be able to provide the patronage he or she has committed to, otherwise his or her clients, need to seek patronage elsewhere (Hyden, 2013). Patrons are thereby dependent on their ability to keep the power that allows them to distribute patronage. However, as long as they can maintain this distribution of patronage they are guaranteed loyalty which in turn makes it even more likely that they are able to keep the power they need. The clients are also dependent on their patron’s ability to distribute patronage to secure the benefits they have gained from the relationship. Thus, if a patron is likely to lose power, making it impossible for him or her to provide patronage, his clients either must look for patronage elsewhere, which would further increase the risk of the patron losing, or they must struggle to keep him or her in power. If a patron gains more power, on the other hand, their ability to provide patronage will increase, making it possible for them to gain more clients. Furthermore, clients to a patron who gains more power will also stand to gain, since they can demand further benefits from their patron (ibid). Thus, if a patron is likely to win new power they are likely to gain more clients, making it more likely that he or she will gain further power, which would also increase the benefits gained from his or her clients.
Clientelism thus causes higher stakes for all actors and their supporters in an election. It makes it costlier to lose and more beneficial to win. Furthermore, it causes even the risk of losing power to be costly, and the mere chance of winning power to be rewarding. Since electoral violence can be an effective tactic to reach one’s goals in an election, both for incumbents and opposition actors, and since it can be used both before and after elections for that purpose, it is likely to be used when it is calculated that the expected
11 rewards of using it is bigger than the expected cost. Thereby, electoral violence should be more prevalent in states with strong clientelist institutions, where both the cost of losing an election and the reward of winning one, should be higher both for an incumbent and his or her clients, as well as for opposition parties and their clients.
H1: The risk of electoral violence, both prior to and after elections, is higher in states with strong clientelist institutions.
Incumbency:
Taylor et al (2017) argue that the risk of electoral violence should be higher in elections where an incumbent is running for office, due to clientelist institutions. In elections where no incumbent is running for office, there is no legacy of any of the competing parties providing patronage to clients, and there is no history of clients providing support to them. They argue that the insecurity of such patron-client relationships should eradicate the effects of clientelist institutions on the risk of
electoral violence. However, in elections where clientelist institutions make the stakes of the election higher, and an incumbent is running for office and there thus is a tested patron-client relationship, the risk of electoral violence should increase (ibid).
However, in states with clientelist institutions, clientelist practices occur on several levels and a position of power can be held by others than the incumbent president. Lower ranking officials and other elites in the country will, like the incumbent president, provide various forms of patronage in exchange for loyalty and support (Van de Walle, 2003; Hyden, 2013). Thus, there should be patron- client relationships between others than incumbents and their clients, that are strong enough for the causal mechanisms discussed above to be valid also when no incumbent is running for office.
H2: The risk of electoral violence is higher in states with strong clientelist institutions, both when an incumbent is running for office and when no incumbent is running for office.
4. Research Design
In this section, the research design is outlined. The sources of data are discussed, the
operationalizations of the variables are presented and the statistical models are introduced. The study is conducted quantitatively, through several logistic regression models on 220 multiparty elections in sub-Saharan Africa between 1990 and 2008, that together will test the three hypotheses. The
independent variable is clientelism, and the dependent variable is electoral violence. Furthermore, five control variables are included in the analysis, as to establish true covariation. These are also discussed later in this section.
12 Case selection:
The cases used in this analysis are 220 elections that have been held in the 48 countries typically considered sub-Saharan Africa. It does not, however, include South Sudan, which was not yet
independent from Sudan in 2008 (Strauss and Taylor, 2012). There are three major reasons for why the study is limited to sub-Saharan Africa. Firstly, these countries have experienced a remarkably similar transition during the time period of the study. Since the end of the Cold War, the region has largely gone from single-party rule to multi-party competition. However, there is great variation in the experiences, particularly regarding electoral violence (Strauss and Taylor, 2012). It is thus possible to hold many things constant in the analysis, whilst at the same time having great variation in the dependent variable of interest. This will make the study more internally valid, since there is less risk that the causal inferences drawn are compromised by factors that is not accounted for. It does however, make the study less
externally valid since the conclusions made are less generalizable to other populations. Secondly, elections in Africa have produced noteworthy instances of electoral violence, both during the years examined and since then, most recently in Kenya the summer of 2017. There are thus, particularly strong practical and policy implications of the study in this context, making the extent to which the result is internally valid in sub-Saharan Africa more important than the ability to generalize to other regions (Strauss and Taylor, 2012). Lastly, the independent variable of interest, clientelism, is by many considered particularly dominant in African politics (Taylor et al, 2017). Bratton and van de Walle (1997, p.63) argue that clientelism is present in all politics. However, it is only a core feature in African politics and a few other states such as Haiti and the Philippines. Studies on the effect of clientelism on the risk of electoral violence in the entire world is thus likely to give biased results.
Data and source criticism:
The data on electoral violence is gathered from The African Electoral Violence Database, or AEVD, (Strauss and Taylor, 2012), which is the only dataset which examines electoral violence specifically. It was established in 2012, but presented the first time in 2009. It includes 221 elections in the 48 countries in sub-Saharan Africa. They record all violence that occurs in the six months prior to an election and in the three months following an election and they distinguish between four levels of violence. No violence, violent harassment, violent repression and highly violent campaign. They further distinguish between violence by the incumbent and the opposition, as well as between pre- and post-election violence. To avoid uneven reporting across countries, the data on violence comes mainly from the US State Department’s annual Country Report on Human Rights Practices. These cover most cases in the sample, and are, according to
13 Strauss and Taylor (2012), sufficiently detailed to code the four levels of violence they have outlined.
Since the State Departments reports have only been issued since 1993, the data on 1990-1992 was collected from Human Rights Watch, Amnesty International and journalism coverage in Africa Report, which together, also are sufficiently detailed for the purpose of the dataset. The source of data on elections is mainly the dataset by Staffan Lindberg (2006), which records all multiparty elections between 1990 and 2003. The rest is hand coded by the authors of the AEVD dataset.
Strauss and Taylor (2012) argues that there is always a risk of measurement bias, as well as arbitrary cutoff points in any coding of cross-national data. However, they have, by consulting additional sources where the data was not consistent, minimized the risk of such biases. They also argue that due to the large sample size, any minor changes in the coding of some cases would not affect the general pattern. One bias that could arise from the coding, is that it risks including violence that is not related to the election, and it risks missing violence related to the elections that occurs further away in time from the election than their timespan.
For the main independent variable, clientelism, data on vote buying from the Varieties of Democracy of dataset, V-Dem, is used. Vote buying is defined by V-Dem as “the distribution of money or gifts to individuals, families or small groups to influence their decision to vote/not vote or whom to vote for”.
Each election included in the dataset is assessed on a scale of zero to four, where zero implies systematic, widespread and almost nationwide vote buying by almost all parties in the election and four means that there was no evidence of vote buying. Since it requires substantial case knowledge as well as subjective evaluation to code the data, there is some risk of measurement bias. However, several precautions have been taken, to minimize this risk. Firstly, the data is coded by a minimum of five country experts for each country-year in the dataset, which allows for inter-coder reliability tests. They further recruit the experts carefully based on five criteria, including their expertise in the country, as well as subject, they are meant to code and their impartiality. Additionally, they use several statistical methods to estimate and
compensate for any biases, as well as to minimize the effect of varying coding practices between the experts (Coppedge et al, 2017). Even though there might still be some faulty measurements in the data, the precautions taken should make these small enough to not affect the general patterns in a large-N study.
For information on whether an incumbent is running for office, I will use the data used in Taylor et al (2017). They have coded a dummy variable for whether or not an incumbent is running for office in each
14 election included in the dataset. Since the coding does not rely upon any subjective evaluation by the researchers and because of the impartiality of the coders, the risk of measurement bias is very small.
For further control variables, data is collected from Taylor et al (2017) and the Ethnic power relations dataset (Wimmer et al, 2009).
Operationalizations, validity and reliability:
As mentioned previously, the dependent variable, electoral violence, is measured through the AEVD.
All four categories of violence are collapsed into a dichotomous variable, where 0 equals ‘no violence’ and 1 equals ‘any violence’. However, for robustness checks, pre- and post-election violence are distinguished between as to be able to draw further inferences from the results.
The intendent concept that is meant to be measured is electoral violence, thus violence that is meant to influence election results in multi-party elections. Therefore, this measurement should be valid. It measures all violence that occur close in time to a multi-party election. It is thus plausible that the clear majority of the data is measuring what it is meant to measure. However, there is a risk that it includes some violence that is unrelated to the election and it is possible that it misses violence that is meant to influence election results, but that occurs outside the timespan set up by Strauss and Taylor (2012). This somewhat weakens the validity of the measurement, but the bias should not affect the results substantially. The measurement is also very reliable since it requires little subjective
assessment by a researcher.
The main independent variable, clientelism, is operationalized as a measurement of vote buying. It is originally coded as a categorical variable with five levels of vote buying, however, a transformed version of the variable that has been transformed into continuous, will be used in this study. This is the most commonly used form of the variable, since it is the form of measurement that compensates most efficiently for measurement biases.
Vote buying is not a perfect measurement of clientelism. Because of the informal nature of clientelism and since it includes many different possible practices, it is difficult to perfectly measure clientelism.
Because of the limited timespan of this study, a qualitative analysis of clientelist practices in all 220 observations is not possible to perform. A proxy, such as vote buying, is thus the closest possible measurement of clientelism that is feasible to use. Since vote buying is a clientelist strategy, it should somewhat capture the intended concept. Vote buying is a way of exchanging patronage for loyalty. Thus, vote buying constitutes an act of clientelism in itself. It is also plausible to believe that if one form of
15 clientelist practices occurs, others do to. Clientelism exists when patrons and clients believe that a pragmatic exchange of patronage and loyalty is the most efficient way of reaching one’s goals. Thus, clientelism exists in contrast to programmatic institutions, where candidates in elections strive to achieve their goals by persuading the citizens that they would be better rulers than their opponents. If vote buying occurs, it is thus likely that the attitudes that causes clientelism exists and thus, that programmatic
institutions have been abandoned in favor of clientelist ones. However, there is a risk, both that vote buying occurs in elections where clientelism in general is not widespread, and, perhaps even more likely, that vote buying does not occur in countries that experience widespread clientelist practices. This could cause variation in the data that does not correspond to variation in actual clientelism. It is, for example, possible that there is something particular about the clientelist practice, vote buying, that causes electoral violence, and not clientelism in general. This must be considered when analyzing the results.
Incumbent running is, in accordance with Taylor et al (2017), operationalized as a dichotomous variable, where 0 means that no incumbent ran for office and 1 means that an incumbent ran for office.
Control variables:
There are five control variables that were chosen because of their potential to influence clientelism and electoral violence. Firstly, incumbent running is included in the analysis, both to examine the second hypothesis, but also as a control variable in the models used to examine the first hypothesis. Secondly, three control variables are chosen to control for structural conditions that could influence the results.
These are GDP per capita, democracy and ethnic fractionalization. Lastly a control variable that control for the presence of election observers is included.
Salehyan and Linebarger (2015) find that electoral violence is more likely to occur in countries with low GDP per capita. Furthermore, a lower GDP per capita should cause a higher risk of clientelist practices.
In a state with low GDP per capita, the possibility to establish effective constraints on power and monitoring practices is limited, which should in turn make resorting toclientelist practices more feasible.
Low GDP per capita also increases the potential benefits of resorting to clientelist practices for clients, since the possibility to lead a pleasant life, without support from a person in power is limited.
Salehyan and Linebarger (2015) also find that truly democratic states are less likely to experience electoral violence. Firstly, since countries with strong clientelist institutions are likely to be classified as less
democratic due to the corrupt nature of clientelist practices. Secondly since, in truly democratic countries,
16 the rulers have more accountability towards their citizens, making clientelist practices riskier and thereby less likely to occur (ibid).
Ethnic fractionalization is also likely to cause higher levels of clientelism. Van de Walle (1997) argues that clientelism often occurs along ethnic lines and that one of the major clientelist practices is ethnically aligned voting. States with high levels of ethnic fractionalization are thus more likely to be clientelistic, since rulers are more likely to premier citizens of his or her own ethnicity and citizens are more likely to be loyal to rulers of their own ethnicity. Ethnic fractionalization should also increase the risk of electoral violence. If the president rewards citizens of his own ethnic group and excludes citizens from other, the cost of losing an election increases. Therefore, a control variable is included which measures the
percentage of the population that is part of an ethnic group that is excluded.
Daxecker (2012) argues that electoral violence is more likely to occur when international election
observers and election fraud is present, since election observers will draw attention to election fraud and help opposition parties and citizens to coordinate resistance against the results. To account for the risk that it is not the clientelist nature of vote buying, but instead that vote buying is seen as election fraud that causes electoral violence, election observers is controlled for. This should reduce the risk of the alternative causal relationship between vote buying and electoral violence.
Statistical models:
The statistical method that is used in all models is logistic regression, thus, measuring the change in risk of electoral violence in logged odds as the independent variable changes. Several models are used to test both hypotheses, with different control variables. Furthermore, robustness checks are performed where the effect is tested separately for pre- and post-election violence.
The first hypothesis is tested by three models, one bivariate regression with only the main
independent and dependent variable, one multiple regression where the control variables concerning structural conditions are included and one multiple regression with all control variables. The second hypothesis is examined by comparing the result of models where only elections with an incumbent running for office is included and models where only elections where no incumbent is running for office is included
Scope and limitations:
The scope of this study has been limited to only examine the effect of clientelism on electoral violence. There are arguably several other factors that affect the risk of electoral violence.
17 Furthermore, the study has been limited to electoral violence in sub-Saharan Africa. The ability to generalize to other regions is thereby limited. One should be cautious about generalizing, even to those states where clientelism is as strong as in Africa, since there might be structural differences between sub-Saharan Africa and the rest of the world, that affects the results. It is also possible that the limited time-period of the study, 1990-2008, causes that study to be less generalizable to other time-periods. It might be that the dramatic shift from authoritarian rule to multi-party elections in Africa during this time-period and the structural shifts in the world order since the end of the cold war, influences the results.
The main weaknesses of this thesis are thus arguably the limited scope of the study and the crudeness of the indicator of the main independent variable. However, even though the
operationalization of clientelism is not perfect, the results of the study should be an indicator of what the true relationship looks like. The study should thereby further increase our understanding of electoral violence and clientelism in sub-Saharan Africa, which is of great importance.
5. Results and analysis
In this section, the results of the various regression models are presented and analyzed. First, the results and analyses concerning both hypotheses are presented separately, as well as an analysis of the goodness of fit of the models based on the Akaike Information Criterion (AIC) values of the
different models. This is a measure of goodness of fit, which allows for a comparison of different models run on the same data. It does, however, not say anything about the absolute level of goodness of fit. Lastly, a summary of the results concerning both hypotheses, as well as a brief analysis of the implication for the two hypotheses are presented. No analysis of substantive
significance is made since the variable used to measure vote buying is a categorical variable that has been converted to a continuous one. To analyze the substantive significance of the results, a reversed conversion would have been necessary, which is not possible within the scope of this study.
18 Table 1: Hypothesis 1
Logistic regression is employed in all models. Standard errors are reported in parenthesis
In model 1, where the effect of vote buying on electoral violence is examined without controlling for any other variables, vote buying has a positive1 and statistically significant effect on electoral violence at 99% confidence level. In model 2, where election characteristics are controlled for, the same relationship is present, and both incumbent running and observers have a positive and statistically significant effect on the risk of electoral violence at 99% respectively 95% confidence level.
However, in model 3, where structural conditions are controlled for, this relationship vanishes. The relationship is not statistically significant in model 4 either, controlling for all variables. However, both incumbent running and excluded population, has a positive and statistically significant on electoral violence at 99% respectively 90% confidence level.
1 The lower the value of the vote buying variable, the more vote buying is present. Thus, a negative coefficient indicates a positive relationship between vote buying and the risk of electoral violence.
19 To check for robustness, the same models as in table 1 were used on pre- and post-election
violence2. The results for pre-election violence follows a similar pattern as the results in table 1.
However, post-election violence does show some different results. Vote buying has a consistent positive and statistically significant relationship with post-election violence, also when controlling for all other variables. Furthermore, both election observers and democracy covaries positively with post-election violence.
The AIC values lessens somewhat when controlling for structural conditions both in the analysis of electoral violence overall and in models with pre- and post-election violence as dependent variables.
This could be due to the lower number of observations in these models. However, the difference is relatively small and should thus not substantially affect the conclusions that can be drawn from the results.
Since there is a positive and statistically significant relationship between vote buying and electoral violence in model 1 and 2, and since this relationship vanishes when controlling for structural conditions, it is plausible to believe that vote buying and these control variables covary. Therefore, a multicollinearity assessment was performed, testing the covariation between vote buying and each of the structural control variables3. Both GDP per capita and democracy have a statistically significant relationship with vote buying at 99% respectively 95% confidence level. However, the R2 of the different regressions were all below 0.25. This indicates that it is not multicollinearity that causes the statistical significance to vanish when controlling for the control variables controlling for structural conditions. Thus, it is not possible to reject the null hypothesis, that clientelism does not influence electoral violence. However, it seems as if clientelism in fact has a relationship with post-election violence. Since there is a bivariate covariation between vote buying and electoral violence, and because of the relationship between post-election violence and vote buying, it is plausible that there is some relationship between clientelism and electoral violence overall, however, how this
relationship looks is not possible to determine from this analysis.
Regarding the control variables, it is noteworthy that neither democracy nor GDP per capita has a statistically significant effect on electoral violence, however, excluded population does. This indicates that it might be that neither democracy nor GDP per capita has a causal relationship with electoral
2 These can be seen in table 4 and 5 in the appendix.
3 These can be seen in table 6 in the appendix.
20 violence, but that the relationship found by Salehyan and Linebarger (2015), could be caused by some missed factor, such as ethnic fractionalization.
Table 2: Hypothesis 2
Logistic regression is employed in all six models. Standard errors are reported in parenthesis. In model 1, 3 and 5, only observations where no incumbent was running for office, are included. In model 2,4 and 6, only observations where an incumbent was running for office, are included.
Table two displays the models that distinguishes between elections where an incumbent is running for office and elections where no incumbent is running for office. The relationship between vote buying and electoral violence is consistently positive and statistically significant, at 90 percent confidence level, in model 2, 4 and 6, which were all run on the subset of observations where an incumbent was running for office. However, the results were not statistically significant in model 1, 3 and 5, that were run on the subset of observations where no incumbent was running for office. In model 6, which controls for all variables, vote buying is the only statistically significant variable. In model 5, which also includes all control variables, no variables are statistically significant. The
21 coefficient for vote buying lessens somewhat when controlling for structural conditions, however, it stays almost exactly constant when controlling for election observers.
The AIC values of the models where an incumbent is running for office are slightly higher than those where no incumbent is running for office; the difference is however small. The AIC values also lessen somewhat when including more control variables. This could be due to the lesser number of observations. Furthermore, all models in table 2 have a significantly lower AIC value than those in table 1. The number of observations are also significantly lower. This makes the results of these regressions less trustworthy, since there might well be some overlooked aspect that intervenes in the regression.
Since there is a positive and statistically significant relationship between vote buying and electoral violence in model 2, 4 and 6, whilst this relationship is not present in model 1, 3 and 5, it is plausible to believe that the proposed causal mechanism proposed by Taylor et al (2017) is true. When an incumbent is running for office, clientelism causes a higher risk of electoral violence, however when no incumbent is running for office, clientelism does not in this analysis seem to have that effect.
Furthermore, one can conclude that the presence of election observers does not affect the
relationship between vote buying and electoral violence. The risk that the independent variable, vote buying, captures the alternative explanation, that vote buying is seen as a form of election fraud, rather than the theoretical mechanism proposed in this thesis, is low.
Table 3: Distribution of electoral violence, pre- and post-election violence
NO VIOLENCE VIOLENCE
ELECTORAL VIOLENCE 91 129
PRE-ELECTION VIOLENCE
98 122
POST-ELECTION VIOLENCE
182 38
Distribution of the variables electoral violence, pre-election violence and post-election violence. All three are dummy variables, where no violence is coded as 0 and violence is coded as 1.
22 To conclude, the results do not show conclusive support for any of the two hypotheses. Based on this analysis it is not possible to reject either the first hypothesis, nor the null hypothesis. The results do however indicate that there is some relationship between, at least, post-election violence and clientelism. As can be seen in table 3, only 7 elections experienced post-election violence without experiencing any pre-election violence, whilst 98 elections experienced pre-election violence without any post-election violence. This means that the results for the main models are probably driven mainly by the cases with pre-election violence.
The results regarding the second hypothesis, that higher levels of clientelism increase the risk of electoral violence both when an incumbent partake in the election and when no incumbent does so, shows more conclusive results. They indicate that clientelism only causes a higher risk of electoral violence when an incumbent is running for office, which supports the causal mechanism proposed by Taylor et al (2017). However, since the goodness of fit of the models concerning the second hypothesis is relatively low, it would be unwise to infer to much from these results.
Alternative explanations:
There are some methodological weaknesses that might have affected the results. Firstly, the
operationalization of clientelism does not capture the entire concept. The levels of clientelism should therefore generally be underestimated in this study. This could both cause an exaggerated and an understated variance in the level of clientelism between the observations. It might also be that vote buying has certain characteristics that other forms of clientelism does not. For example, vote buying might be seen, not merely as a clientelist practice, but also as an act of election fraud. If so, it is possible that this caused violence as a protest against the results. The inclusion of the control variable, election observers, was an attempt to minimize this risk. However, there is still a risk that this mechanism is present in the results. This would also explain why vote buying only had a statistically significant effect on post-election violence. Furthermore, the covariance between the main independent variable and some of the control variables might cause collinearity, which in that case could decrease the statistical significance of the coefficients. This could be one explanation for why the statistical significance vanishes when including the control variables concerning structural conditions. It might also be that clientelism only causes a higher risk of electoral violence during some circumstances. One such circumstance is when an incumbent is running for office. However, Fjelde and Höglund (2016) argues that majoritarian electoral rules cause higher stakes in elections
23 and that this mechanism is aggravated by clientelism. It might thus be that clientelism only causes electoral violence in elections that employ majoritarian electoral rules.
6. Conclusion
In this thesis, the variation in electoral violence in different elections in sub-Saharan Africa has been examined in order to understand what causes violence in some elections whilst not in others, despite them being very similar in many aspects. It is hypothesized that clientelist institutions cause higher stakes in elections, making electoral violence more profitable and thereby more common. By examining this subject theoretically as well as empirically using multivariate logistic regression, this thesis aims at contributing to the cumulative theoretical knowledge of both clientelism and electoral violence in sub-Saharan Africa, and thereby also to the vast literature on democratization in Africa.
Furthermore, the thesis aims at making an empirical contribution to the field, by being the first study that tests the effect of clientelism on the risk of electoral violence, through a measurement of actual clientelist practices.
The study shows that there is no obvious relationship between clientelism and pre-election violence.
However, it indicates that clientelism is causally linked to post-election violence. Furthermore, they show that, when an incumbent is running for office, there is a relationship between clientelism and vote buying. Furthermore, ethnic fractionalization seems to have a positive and causal effect on pre- election violence, whilst both democracy and election observers have positive effects on post-
election violence. It can thereby be concluded that the main theoretical argument of this thesis is not sufficient to explain the relationship between clientelism and electoral violence. The characteristics of the different categories of electoral violence and of the different types of elections needs to be analyzed deeper and more thoroughly distinguished between to fully understand the patterns of the phenomena. However, the results of this analysis correspond well with the theory proposed by Taylor et al (2017), which further strengthens their claim that clientelism causes electoral violence, but only when an incumbent is running for office.
To answer the research question, why some elections spark violence whilst other do not, further analysis of the different categories of electoral violence is needed. Firstly, it seems as if pre- and post- election violence are caused by different things. There is thus need for more sophisticated theory concerning these two different kinds of electoral violence. Secondly, the different possible
perpetrators of electoral violence must be analyzed further, since it seems as if electoral violence is
24 more likely to occur when an incumbent is involved in the election. Lastly, a more thorough analysis and measurement of clientelism is needed, both to eliminate the risk that other theoretical
mechanisms are what causes the relationship and to be able to make further investigations into what the effects and causes of clientelism are.
Clientelism is a complex and multidimensional concept and does seem to have many implications for African politics in general, and so also for electoral violence. However, what these implications are has yet to be established. It seems as if clientelism, causes some forms of electoral violence, during some circumstances. Thus, the closest thing to an answer to the question, why some elections spark violence, whilst others do not, that can be derived from this thesis is that clientelism might.
25
7. References
Bekoe, Dorina, ed. (2012) Voting in Fear: Electoral Violence in Sub-Saharan Africa. Washington. DC:
United States Institute of Peace.
Boone and Krieger (2012) Land Patronage and Elections: Winners and Losers in Zimbabwe and Cote d’Ivoire. In: Dorina Bekoe (ed.) Voting in Fear: Electoral Violence in Sub-Saharan Africa.
Washington, DC: United States Institute of Peace Press, 75-116.
Bratton, Michael & van de Walle, Nicolas (1997) Democratic Experiments in Africa: Regime Transitions in Comparative Perspective. New York: Cambridge University Press.
Coppedge, Michael, John Gerring, Staffan I. Lindberg, Svend-Erik Skaaning, Jan Teorell, David Altman, Michael Bernhard, M. Steven Fish, Adam Glynn, Allen Hicken, Carl Henrik Knutsen, Joshua Krusell, Anna Lührmann, Kyle L. Marquardt, Kelly McMann, Valeriya Mechkova, Moa Olin, Pamela Paxton, Daniel Pemstein, Josefine Pernes, Constanza Sanhueza Petrarca, Johannes von Römer, Laura Saxer, Brigitte Seim, Rachel Sigman, Jeffrey Staton, Natalia Stepanova, and Steven Wilson, 2017, V-Dem [Country-Year/Country-Date] Dataset v7.1, Varieties of Democracy (V-Dem) Project.
Coppedge, Michael, John Gerring, Staffan I. Lindberg, Svend-Erik Skaaning, Jan Teorell, David Altman, Michael Bernhard, M. Steven Fish, Adam Glynn, Allen Hicken, Carl Henrik Knutsen, Kyle L. Marquardt, Kelly McMann, Valeriya Mechkova, Pamela Paxton, Daniel Pemstein, Laura Saxer, Brigitte Seim, Rachel Sigman, and Jeffrey Staton. 2017. “V-Dem Codebook v7.1“ Varieties of Democracy (V-Dem) Project.
Coppedge, Michael, John Gerring, Staffan I. Lindberg, Svend-Erik Skaaning, Jan Teorell, Joshua Krusell, Kyle L. Marquardt, Valeriya Mechkova, Daniel Pemstein, Josefine Pernes, Laura Saxer, Natalia Stepanova, Eitan Tzelgov, Yi-ting Wang, and Steven Wilson. 2017. “V-Dem Methodology v7.1” Varieties of Democracy (V- Dem) Project.
Taylor, Pevehouse and Strauss (2017) Perils of pluralism: Electoral violence and incumbency in sub- Saharan Africa, Journal of Peace Research, 54(3) 397–411
Daxecker, Ursula E (2012) The cost of exposing cheating: International election monitoring, fraud, and post-election violence in Africa. Journal of Peace Research, 49(4): 503–516.
26 Fjelde, Hanne & Kristine Höglund (2016) Electoral institutions and electoral violence in sub-Saharan Africa, British Journal of Political Science, 46(2): 297–320.
Hafner-Burton, Emilie; Susan Hyde & Ryan Jablonski (2013) When do governments resort to election violence? British Journal of Political Science, 44(1): 149–179.
Hyden, Goran (2013) African Politics in Comparative Perspective. New York: Cambridge University Press.
Höglund, Kristine (2009) Electoral Violence in Conflict-Ridden Societies: Concepts, Causes, and Consequences. Terrorism and Political Violence 21(3):412–27.
Mueller (2012) The Political Economy of Kenya’s Crisis. In: Dorina Bekoe (ed.) Voting in Fear:
Electoral Violence in Sub-Saharan Africa. Washington, DC: United States Institute of Peace Press, 145- 180.
Salehyan, Idean & Christopher Linebarger (2015) Elections and social conflict in Africa, 1990–2009.
Studies in Comparative Social Development 50(23): 23–49.
Straus, Scott & Charlie Taylor (2012) Democratization and electoral violence in sub-Saharan Africa, 1990–2008. In: Dorina Bekoe (ed.) Voting in Fear: Electoral Violence in Sub-Saharan Africa. Washington, DC: United States Institute of Peace Press, 15–38.
Van de Walle, Nicolas (2003) Presidentialism and Clientelism in Africa’s Emerging Party Systems.
Journal of Modern Africa Studies 41(29):297–321.
Wig, Tore, Håvard Hegre, Patick M. Regan, 2015, Updated data on institutions and elections 1960–
2012: presenting the IAEP dataset version 2.0, Research and politics 2(2):1-11
Wimmer, Andreas, Lars-Eric Cederman, Brian Min, 2009, Ethnic Politics and Armed Conflict: A Configurational Analysis of a New Global Data Set, American Sociological Review 74(2):316-337
27
8. Appendix
Table 4: Pre-election violence
Logistic regression is employed in all models. Standard errors are reported in parenthesis
In model one and two, vote buying has a positive and statistically significant relationship with pre- election violence. When controlling for structural conditions, in model 3 and 4, the statistical
significance vanishes. When controlling for all variables, incumbent running and excluded population is the only two variables that are statistically significance.
28 Table 5: Post-election violence
Logistic regression is employed in all models. Standard errors are reported in parenthesis
Vote buying has a positive and statistically significant relationship with post-election violence in model 1, 2 and 4. Thus, it is statistically significant when controlling for all other variables, as well as when no other variables are controlled for. Furthermore, incumbent running, observers and
democracy is statistically significant.
29 Table 6: Collinearity test
OLS regression is employed in all models. Standard errors are reported in parenthesis
Both democracy and excluded population has statistically significant relationship with vote buying.
GDP per capita has not. The R2 of the different models are all below 0.25.
