Limiting the use of pro-government militias

(1)

Limiting the use of pro-government militias

The effect of media freedom on the impact of pro-government militas on repression

Johannes Blad

Statskunskap C (Examensarbete) Department of Government Uppsala University, Fall 2017 Supervisor: Niklas Bremberg Words: 11905

Pages: 39

(2)

Abstract

Based on previous research on repression, this paper aims to investigate the relationship between pro-government militias and repression. Pro-government militias are armed groups that are either identified as being pro-government or sponsored by governments. Research suggests that governments delegate tasks such as torture, killing and disappearances to these groups in order to avoid accountability costs that comes with violating physical integrity rights. The purpose of this paper is to further examine this relationship by introducing an interaction effect i.e. that the effect of pro-government militias on repression is moderated by another variable. Media freedom is the supposed moderator and the argument is that free media will monitor and expose governments illegitimate behavior, thus leading to a reduction in the effect of pro-government militias on repression. Utilizing multiple regression analysis, the findings in this paper suggest that the effect of pro-government militias on repression decreases in magnitude when journalists are able to freely and safely criticize government behavior.

(3)

International in their annual report for 2016/2017. Despite efforts from NGO’s and the international community, governments continue to suppress and quell oppositional voices, violate basic human rights and infringe on the personal integrity of individuals. While some humans enjoy the protection of and full respect for personal integrity, others live in fear of what the state might do to them. What is it that makes some governments repress their own citizenry while other governments do not?

A wide range of studies on state repression suggest that accountability is one major factor that decrease violations of human rights, political violence and repression. Accountability is primarily related to democracy and is often regarded as a key feature of democracy (Henderson, 1991; Regan & Henderson, 2002; Davenport & Armstrong, 2004). Research inclined towards the relationship of accountability in democracies and repression has mainly converged on electoral or judicial aspects that might restrict executives’ ability and will to choose repression over other options from a repertoire of tactics. Political competition, executive constraints and media freedom are three of several internal accountability, control and monitoring mechanisms that interfere with governments willingness and ability to use repressive tactics (Whitten-Woodring, 2009; Davenport, 2007c; Bueno De Mesquita et al, 2005). Furthermore, countries are more or less subject to external accountability mechanisms through conditionalities and monitoring from the international community (Mitchell et al, 2014:815).

Recent research implies that although accountability mechanisms are in place, governments that face accountability costs can, rather than abstain from repression, delegate repressive tasks to groups outside of the regular security sector in order to bypass the costs of using violence. By shifting the responsibility to an armed group and denying affiliation with it, governments can continue to repress their citizenry (Carey et al, 2015). These groups are called pro-government militias and when they are present the severity and comprehensiveness of human rights violations and repression increase within the country in which they operate (Mitchell et al, 2014). This is the starting point for the purpose of this paper, to find another set of controls which might limit the use of pro-government militias. If governments can

(5)

continue to repress their citizenry even when assumed strong control, monitoring, and accountability mechanisms are in place, what will constrain the use of pro-government militias as agents of repression?

One aspect that has been overlooked in the repression-accountability nexus is the ability of media to monitor government behaviour, and by that expose eventual illegitimate behaviour facilitating the process of accountability. Scholars quantitatively analysing the repression- accountability nexus have for the most part utilized data from the well-known “Polity Project”

designed to code the authority characteristics of states (Marshall et al, 2017:1). The characteristics measured are then either used to operationalize different aspects of

accountability or to act as a measure of democracy. However, media freedom or freedom of the press is not included in the PolityIV scale neither is it rarely included in the literature despite its supposed function in exposing illegitimate behaviour. Jennifer Whitten-Woodring (2009) examines how media affects governments respect for physical integrity rights.

Conclusively, she finds that free media can serve as a watchdog over government behaviour under certain circumstances. Freedom of the press continues to be a vital part of development practices and in the promotion of democracy. Non-governmental organizations and

international organizations such as the UN promotes the development of free and pluralistic media as it is assumed to facilitate good governance, transparency and promote human rights (UNESCO, 2017). Studies on the effect of free media can have implications for policy- making within the international community directed at reducing repression and human rights violations. Can free media deny governments from using pro-government militias to execute repressive tasks?

The main purpose of this paper is therefore to analyse if media can limit the effective use of pro-government militias by governments to execute repressive actions. This has yielded the following research question:

Does free media reduce the effect of pro-government militias on repression?

The three-part relationship is analysed through regression analysis and the findings suggests that higher levels of media freedom reduces the impact that pro-government militias have on repression.

What follows from here is at first a review of previous research which leads into a discussion of the key concepts that make up the theoretical framework of this paper. Secondly, the

(6)

overall research design and choice of method is presented followed by a discussion of the chosen data for the regression analysis. The third section presents the results in the form of tables and plots allowing for easy interpretation. Finally, a discussion and summary of the results is provided and concluded with suggestions for further research.

(7)

2. Theoretical framework

This section outlines the three main components: pro-government militias, media and

repression that constitute the theoretical framework for this paper. Previous research is at first discussed and theories deriving from these three components lay the foundation for the theoretical argument that this sections ends with.

2.1 Previous research on pro-government militias, media and repression

In the field of repression, political violence and domestic human rights abuse¹, much of the literature is devoted to finding and explaining the determinants or causes of repression, although research regarding the consequences of repressions is growing larger. Scholars have actively turned to state-level explanations in an attempt to understand why and what type of governments use repression, often leading to an explanation related to the regime-type of countries. Democracy and its institutions is commonly regarded to be negatively associated with repression and human rights abuse. Research on repression assume that repression is a strategic means of control and that the primary goal for governments is to remain in power, meaning that repression is one type of tool governments can use from their repertoire of the coercive apparatus (deMeritt, 2016:2).

One of the core findings within the field of repression is that internal dissent provide the means for states use of repressive tactics. When the balance of power is agitated i.e. when governments’ power is challenged or threatened, they can respond by meeting the demands of or come to terms with the group challenging the power. The problem is that the option of accommodation and co-optation entails a reduction of power, e.g. the release of political prisoners, or including opposition-members in the decision-making process. For some governments this option might not be viable nor even considerable. An alternative to

accommodation is therefore to subdue or quell the opposition by force which ultimately is one form of repression (deMeritt, 2016:4). In a study conducted by Davenport (1996), he finds that political conflict or as previously expressed, dissent, increases repression. Primarily, governments respond immediately to current dissent with sanctions when regime survival is threatened. Dissent also has a long-term effect on governments use of repression meaning that past experience with political conflict influences governments tendency to use repression

1 Repression, political violence, human rights violations and physical integrity violations are used correspondingly

(8)

(1996:384-385). Ritter (2014) is interested in understanding under what conditions dissent provide the means of repression rather than accommodation. She finds that governments are more likely to use repressive tactics when the cost of using repression is cheaper, both in terms of power and resources, compared to accommodate to the actors expressing dissent.

Furthermore, in developing countries where effective channels of expressing dissent and demanding change are weak or even missing, anticipated threats to the regime survival is found to be the main factor affecting levels of repression (Reagan & Hendersson, 2002:133).

That dissent is so closely related to repressive behaviour and as a result of consistent findings, the relationship is sometimes referred to as the “Law of Coercive Responsiveness”

(Davenport, 2007a:7).

Furthermore, Conrad and DeMeritt (2013) finds that higher levels of repression correlate with how dependent the government is on their citizens. They argue that governments are

dependent on their citizenry in two ways, either “at the bank for financial resources or at the ballot box for political support” (2013:1). This leads us onto another key finding in the repression, human rights violations and political violence literature, namely that regime type matters. Strictly speaking, democracy seems to have an impact on levels of repression. I will discuss this finding and relationship more closely in section 2.3 as it lays the foundation to the theoretical argument of this paper.

Previous research on pro-government militias (PGMs) and repression is sparse. The

qualitative research on the consequences of pro-government militias is primarily in the form of case studies while the quantitative literature has for the most part focused on specific types of violence such as genocide (Ahram, 2013) or specific human rights violations like sexual violence (Cohen and Nordås, 2015). These non-state armed groups are associated with not only different types of violence but also to state failure, both as contributing to state failure and as an effect of state failure (Bates, 2008). Other scholars interested in pro-government militias have taken a different approach, rather than observing and analysing the effects of pro-government militias they have tried to explain the formation of and linkage of pro- government militias. Ahram (2016) shows that pro-government militias are more frequent in low capacity states. High-capacity democracies are less likely to have a PGM present and active because they have more options for managing conflicts through non-violent means (2016:219). An individualist approach taken by Ash (2016) seeks to explain PGM formation as a result of individual leaders decision-making rather than using state-level explanations.

(9)

PGM formation is an option to leaders facing political threats, both externally and internally, in order to increase chances of survival. Threats to leaders’ political survival in terms of coups and international pressure are examples of threats associated with the formation of PGMs (2016:721). The most notable work on PGMs and repression suggests that where there is a presence of a PGM, the level of repression is likely to be higher (Mitchell et al, 2014). This work and their findings will be discussed more thoroughly in the next section.

Research on the impact or effects of media on repression is rather limited but there are some suggestions that lead us to think that free media can lead to reductions in human rights violations. Besides the traditional branches of government: the executive, the legislative and in some cases the judicial, mass media is sometimes referred to as a fourth additional

independent branch. Journalists, news channels and media in full have the possibility to play the role of a watchdog over government behaviour, exposing illegitimate behaviour such as corruption and illicit violence and thus facilitating the process of holding authorities

accountable (Norris, 2008:189). Freedom of the press along with other related freedoms such as freedom of expression is considered to be fundamental to establishing good governance, consolidating democracy and bolster development (UNESCO, 2017; Norris, 2008:186). A considerable amount of the research on press freedom is interested in the relationship between media and corruption, suggesting that press freedom is a key factor in decreasing corruption.

Several studies have tested this relationship. Improved freedom of the press and less

controlled media has shown to decrease levels corruption. Government officials are less likely to use their power for private gain if there is a free media in place that increases the risk of being exposed (Ahrend, 2002; Dutta & Roy, 2016; Kalenborn & Lessman, 2013). According to Amartya Sen (2007:95) media is a significant accountability mechanism that augment and is beneficial to peace and democracy as it can serve to address important issues and report on cases of improper government behaviour. Similarly, O’Donnell points out the importance of encouraging free media to improve accountability in new democracies (O’Donnell, 1998).

2.2 Pro-government militias

States can monopolize the use of violence to ensure stability and order. The monopoly on the use of violence is advocated by scholars, theorists and policymakers because of its crucial effect, and the complications that might follow if the use of violence is distributed to other actors within the border of the state. Indeed, the monopoly of violence is even included as one criteria in the commonly used Weberian definition of the state. Despite the prospected

(10)

beneficial effects of monopolizing the use of violence, some governments choose to delegate the use of violence to other groups outside of official government forces.

The theoretical framework of pro-government militias used in this paper is retrieved from

“The Pro-Government Militia Project” led by Sabine Carey and Neil Mitchell. Their project includes the Pro-Government Militia dataset which will also be employed in this paper. Carey et al (2012:250) define PGMs as groups that are: (1) either identified as pro-government or sponsored by the government, (2) not part of the official security forces, (3) armed and (4) has some sort of organization. Pro-government militias are divided into two types based on their link and affiliation to the government: semi-official and informal. Semi-official PGMs sometimes hold official status by law (Carey et al, 2012:251) and may in some cases be financially rewarded by the government. Informal militias are either armed or supported, though not openly or officially, by the government. Informal groups connection to the government can still however be generally acknowledged (Carey & Mitchell, 2017:130).

What separates the two types is if there is a formalized link, established by the government, between the group and the government (ibid). PGMs are present across all types of regimes, although most frequent in regimes that share both democratic and autocratic characteristics (Carey & Mitchell, 2017:130).

Mitchell et al (2014) argue that governments can use these groups to shift the responsibility that comes with using violence away from the government and towards the militia. As

research on repression suggests, accountability mechanisms are crucial for terminating human rights abuse (Henderson, 1991; Conrad & DeMeritt, 2013; Bueno De Mesquita, 2005;

Davenport, 2007c). However, even when accountability mechanisms are in place, governments can continue violating human rights. As Mitchell et al claims “Rather than forego repression, an alternative for government officials is to seek to evade accountability for this violence (2014:815)”. This can be done by delegating violence to pro-government

militias outside of the states’ security apparatus (ibid). Mitchell et als (2014) argument for why PGMs are likely to increase repression is based on a modified version of the principal- agent model. In this case, the benefits of delegating repression to PGMs is balanced against the costs and problems that might erupt between the government and the PGM (2014:817).

These problems are often related to goal variance and information asymmetry. In principal- agent models, the problem of control is commonly mentioned. Principals must enforce some sort of mechanism to minimize the risk of losing control of the agent. However, when

(11)

governments delegate repression to PGMs they can intentionally refuse to control the PGM as they may benefit from keeping a distance i.e. having less actual contact with the PGM in order to strengthen arguments for their non-association (2014:818).

Furthermore, governments and PGMs can jointly benefit from repression. Governments can utilize the problem of goal variance and information asymmetry. Agents may have different goals compared to the principal and may withhold information. In the regular security forces, the agency problem is terminated or at least diminished through proper training,

indoctrination and compensation to ensure compliance. However, in pro-government militias, these compliance mechanisms are likely to weak or even absent (2014:817). This is one reason for why PGMs are likely to be more violent than regular security forces (2014:816).

The goal of a government (e.g. quelling political opposition) is something that must be done through repression and the members of the militia might have private goals (e.g. monetary interest) that can be met through repression. As a consequence, assuming that the government has a strategic interest in using repressive tactics and the members of the militias has a private interest for using violence, both parties jointly benefit from the repression. The government can then shift the responsibility of violence by claiming either an agency problem, loss of control (mostly in the case for semi-official PGMs) or denying affiliation to the armed group (mostly in the case of informal PGMs) (2014:819). In essence, governments are more likely to use violent tactics if they can avoid being held responsible for it and as a result: the presence of pro-government militias will increase levels of repression (Carey et al, 2015:852).

2.3 State Repression

State repression refers to coercive activities by some political authority targeting those within the authority’s territorial jurisdiction. While repression can take many forms, both covert and overt, ranging from surveillance to police brutality, it is generally divided into two types of activities/violations: civil liberties restrictions and personal/physical integrity violations. Civil liberties restrictions include restrictions in the form of bans and constraints on the freedom of speech, assembly, belief and association. Personal integrity violations are rather actions taken by the state or its agents that directly threaten or target the integrity of individuals such as beatings, torture and political imprisonment (Davenport, 2007b:487). This paper will have its focus on violations of physical/personal integrity rights.

(12)

One of the most consistent findings in the quantitative repression literature is that higher levels of repression is associated with countries under authoritarian rule (Hendersson, 1991;

Poe & Tate, 1994; Davenport, 2004; Richards et al, 2015). Democracies are associated with lower levels of repression but what are the factors within democracies that has this pacifying effect on repression? Taking a closer look at democracies has proven to give better and more refined explanations of the components or characteristics of democracy that decreases repression. In democracies, governments are discouraged to use repressive tactics primarily because of the inherent government-power limiting effect of democracy and the higher levels of accountability to its citizens compared to autocracies. Repression is therefore more costly in democracies because of the populations possibility of removing the government from power in the next election, and the fear of sanctions from other powerholders in the country (Henderson, 1991).

Davenport (2007c) extends this argument by developing a distinction between these two aspects of democracy that can explain the negative impact that democracy has on repression.

Voice are the mechanisms that makes government officials accountable to the citizens subject to their power. This mechanism is usually met by structuring the political system in a way that cherish competition, compelling authorities to compete with each other for power, observed by and later approved by the population i.e. what is usually thought of when thinking about democracy. When this criterion is met, accountability will create incentives for authorities to provide public goods favoured by the citizenry, rather than repression, to remain in power (2007c:22). Veto is the mechanism that makes government officials power constrained by other power-holders. Power-holders such as different political-economic elites, judiciaries and other politicians control and watch the government, through means of checks and balances and executive constraints, in order to make credible threats of removal if unacceptable behaviour is observed (2007c:56). In conclusion, political authorities abstain from repression when they anticipate citizen reaction and uprising (Voice – Vertical Accountability) or they fear action of other powerholders (Veto – Horizontal Accountability). The results show that

“different aspects of democracy (Voice and Veto) promote lower level coercive activity² while discouraging more lethal forms… As found, Voice (Competition/ Participation) is more effective than Veto (Executive Constraints) in reducing state repression” (Davenport, 2007c:132). Similarly, Bueno De Mesquitas et al (2005) findings also demonstrate of the

2 Lower level coercive activity refers to restrictions and infringements of civil liberties.

(13)

importance of accountability, clearly depicted in their results: “accountability appears to be the critical feature that makes full-fledged democracies respect for human rights; limited accountability generally retards improvements in human rights” (2005:439). The data shows that institutional reforms that promote e.g. executive constraints and competitiveness or in other words accountability, is fundamental for democracy to lead to reductions in state repression.

2.4 Media: The fourth estate

In an influential study on the impact of press freedom on corruption, it is argued that press freedom is a mechanism of external control of government officials. Indeed, press freedom might be a stronger and more reliant mechanism of control compared to other alternatives such as the legal system or other government officials, given a high level of press freedom (Brunetti & Weder, 2003:1805). Case studies show the importance of media in monitoring and facilitating accountability (Bonner, 2009). Media does this by continuous investigation of politicians and authorities. The media is therefore central to accountability by informing citizens of any type of illicit behaviour and by keeping officials restrained. Media has shown to be much effective in reporting wrongful behaviour of authorities in Latin America

(Smulovitz & Peruzzotti, 2000:154).

As implicitly understood, freedom from control and from government pressure is conditional for the media to function in a watchdog manner. Freedom House assesses freedom of the press based on three categories: the political environment, the legal environment and the economic environment (Freedom House, 2017). Although, these categories increase the capability of media to monitor government officials and activity, it does not necessarily mean media will perform as the watchdog it should. As Jenifer Whitten-Woodring has noted “While Freedom House focuses on the environment or the structure within which the media function, it fails to consider the professional norms which shape the agency that drives journalistic practices” (Whitten-Woodring, 2009:598). The environment in which media operate also requires a professional environment that promotes the type of journalism that is directed towards monitoring the government. It is therefore the functionality in terms of what the media produces in practice rather than the legal, political and economic structure in which media function that determines whether media can embrace the role of a watch-dog over government behaviour (2009:601). In her study, Whitten-Woodring uses Van Belle (1997) Global Press Freedom data, based on medias ability to criticize government behaviour, which

(14)

she argues captures to a greater degree the direct function of media rather than the environment it operates in. I will employ an updated and slightly modified version of the dataset developed by Whitten-Woodring and Van Belle (2015). However, the updated dataset still uses a conceptualization of media freedom in a similar conduct, journalists’ ability to freely and safely criticize the government.

The reason for introducing media in this paper is because scholars interested in the effect of democracy and accountability on repression have either turned towards specific accountability mechanisms or democracy en masse. Specific accountability mechanisms have generally involved executive constraints or judicial constraints. In that case media has been overlooked although it has the potential to facilitate accountability. In other cases where democracy has either been the independent or the control variable it is operationalized using the PolityIV scale. The PolityIV project however does not include freedom of speech nor freedom of the press as an indicator for democracy.

2.5 Theoretical argument

Repression is a tool for governments to exert control over the population within their territory.

Based on previous research on the effects of mechanisms of accountability on repression, the argument for this paper turns to another aspect of accountability that has the possibility to limit governments effective use of pro-government militias. Free media can work as an additional external type of control when other control measures are not sufficient. When media is not free it might have the opposite effect, rather than facilitating accountability, it can legitimize governments illicit behaviour and serve to protect governments from being held responsible and then held accountable. The argument of this paper is structured around the assumptions that (1) when pro-government militias repress it is for the purpose of the government to evade accountability costs that comes with using violence. (2) Higher levels of press freedom discourage governments from illicit behaviour. Assuming that the purpose of using pro-government militias is to evade accountability my argument is that when media is either free of governmental control, or journalists are able to safely criticize government behaviour, it will affect the relationship between PGMs and repression by undermining the possibility of evading accountability.

When repression is costly i.e. when there are established mechanisms of accountability that can sanction or punish illegitimate behaviour, governments have the option to evade the

(15)

consequences of their behaviour by delegating tasks of illicit violence to a PGM and thereby shifting responsibility to the PGM. However, in line with my argument this option should be less likely (more costly) when there is a free media in place that have the possibility through investigative journalism, to expose governments connection and linkage to an eventual PGM and reduce the extent of repressive behaviour that the PGM can pursue. If governments are found to be responsible by the media, the reason for using PGMs for repression is lost. Carey et al argues that governments take advantage of the PGMs separation from the regular

security apparatus, thus reducing the chance of tracing the repression to back to the government. By delegating repression to PGMs, governments disorient the information of who is responsible for repression (2015:852). However, a free and effective media

environment could in line with my argument clear any confusion of who is responsible. If media exposes governments linkage to the repressive acts of a PGM, the government can no longer credibly deny affiliation with the group or argue for an agency problem. Thereby limiting their ability to keep using the group for repression. Furthermore, the key role of the free media is not to prevent repressive behaviour but rather limit the effectiveness of

governments use of PGMs. Neither will the media prevent the formation of pro-government militias but instead limit the possibility of using them for repression. The hypotheses of this paper are as follows:

(1) Pro-government militias increase levels of repression

(2) The effect of pro-government militias on repression varies across levels of media freedom. The effect declines in magnitude when media is “free” compared to “not free”

The relationship between pro-government militias and repression when free media interacts is illustrated as follows:

Figure 1. The interaction of media in the relationship between PGM and repression Goverment

delegate repressive task

to pro- goverment

militia

Media investigates act of

repression or the pro-government militia prior to act

of repression

Government is found responsible

for act of repression or

planning for repression

Government is reluctant

to use pro- government

militia for repression

(16)

3. Research design

This section outlines the research design of this thesis. It will first present the method chosen for answering the research question in addition with a discussion of the method chosen. This is followed by a presentation of the data used in this thesis. The section ends with a discussion of the limitations and problems in terms of validity and reliability of the data and overall design of the thesis.

3.1 Choice of method

When political scientists are interested in explaining differences in outcome of a certain feature between countries e.g. variances in levels of democracy, corruption or GDP, they can turn to statistical models in order find covariance between the outcome and possible

explanatory factors (Teorell & Svensson, 2007:159). Regression analysis is one type of statistical model that can be used to provide evidence for a relationship between two variables. However, it requires that the dependent variable can be treated as a continuous interval scale (2007:166). Initially, a bivariate regression (OLS) will be employed to test the relationship between PGMs and repression. This might seem superfluous because it has already been done by Mitchell et al (2014) but they use a different operationalization of repression compared to this thesis so in order to test if the relationship still exists, the

bivariate regression is carried out. To rule out the possibility of other explanatory factors that can explain repression, an extension of the bivariate regression is used: multiple regression.

By including other potential predictor variables of repression into the regression, the model gains further explanatory power by distorting the influences of other independent variables on the effect of PGMs (Lewis-Beck, 2017:55). For the purpose of increasing the explanatory power I will therefore employ several multiple regressions and also introduce media freedom as an interaction effect on the relationship between PGMs and repression. The purpose for introducing an interaction effect is to analyse whether the effect of one independent variable is dependent on the value of another independent variable (Lewis-Beck, 2017:70). In this case it would translate into that the effect of PGMs on repression is dependent on whether media is free or not. The interaction effect is measured by multiplying the value of PGMs and the value of media freedom. The regression including the interaction effect is what will ultimately answer my research question and hopefully coincide with the given hypothesis. The

regressions include data for 178 countries from 1981 to 2007 as a result of the data availability for pro-government militias.

(17)

3.2 Data selection

The empirical data analysed in this thesis is gathered from several sources, one of them is commonly used in the academic field while the others are relatively new and has not been frequently employed in the relevant literature. The data on PGMs in this study is retrieved from Carey et al (2013) PGMD Country-Year dataset. The dataset provides information on organized armed groups that are identified by several sources as being pro-government. It contains information on the type of linkage they have with the government and an additional dataset contains more detailed information on each of these groups. The dataset contains information on pro-government militias in 178 countries during the time period from 1981 to 2007.

In choosing data for the dependent variable: repression, there are several options. In the repression literature, there is primarily two separate datasets that are generally used: the CIRI Physical Rights Integrity Index (CIRI PHYSINT) and the Political Terror Scale (PTS). The CIRI Physical Integrity Rights Index measures a country’s level of respect of physical integrity rights based on four variables: extrajudicial killings, political imprisonment, torture and disappearances. The combined score of these four variables then make up the CIRI Physical Integrity Rights Index (Cingranelli & Richards, 1999:409). In determining the level of human rights violations, coders get information from the annual US State Department Country Reports on Human Rights Practices and Amnesty International Annual Reports on human rights (Cingranelli & Richards, 2014). The Political Terror Scale uses a similar coding process and similar sources but differs from the CIRI PHYSINT mainly in terms of what type of violations that are the unit of observation. Carey et al (2014) use the CIRI PHYSINT as their measure of repression while also analysing each variable that comprise the index separately. Repression is constituted by various types of violence and the PTS account for a larger amount of violations that is as much of importance. Pro-government militias have previously been linked to other types of violence and not only the ones used in the CIRI PHYSINT. For the purpose of this thesis, the PTS will be employed to measure repression since it includes other types of violations in its coding process. However, there is no possibility to disaggregate violations in the PTS, although it is not essential for this thesis (Gibney et al, 2017).

When choosing the data for media freedom, there were even more established and commonly used datasets at my disposal. Firstly, there was the Freedom House annual Freedom of the

(18)

Press data which I have already touched upon. Analysts answer questions that are split into subcategories: legal environment, economic environment and the political environment.

Based on the answers of these questions, countries are given points which is at a later stage combined to form the final rating of a country’s press freedom. At first glance, the Freedom of the Press Data seem suitable for measuring media freedom but using it for this thesis has complications. One of the questions that analysts answer when examining the political environment of a country is “Are journalists, bloggers, or media outlets subject to extralegal intimidation or physical violence by state authorities or any other actor as a result of their reporting?”. This question can yield a higher amount of points in general compared to other questions for determining the legal environment (Freedom House, 2017). Although state or state-sponsored violence against actors of the media bears relevance to the level of media freedom, cases of these violations are most likely included in my dependent variable. The second dataset at my disposal was the Reporters Without Borders annual World Press Freedom Index which ranks countries based on the level of freedom available to journalists.

Journalists and experts are given a questionnaire to answer for the respective country in which they operate in. The questionnaire is scored and combined with an additional score on abuses and violations of journalists. Compared to the Freedom of the Press data, it is possible to disaggregate the scoring of the World Press Freedom Index to have it not include abuses and violations. However, data from the World Press Freedom Index is only available for the time- period 2002-2017 (Reporters Without Borders, 2017). In order to not suffer from a case where violations are included in the dependent variable and the interaction variable, this thesis will employ the The Global Media Freedom Dataset that covers a larger time period, does not include violations in its evaluation and has previously been used in the repression literature (Whitten-Woodring & Van Belle, 2015; Whitten-Woodring, 2009).

3.3 Dependent, independent and interaction variables Independent variable: Pro-government militias

Pro-government militias are operationalized using the binary indicator for the presence of a PGM in a given country and year retrieved from the Pro-Government Militias Country-Year Database (Carey et al, 2012). The database records both any identified activity and the presence of pro-government militias separately but assuming that media freedom might inhibit PGMs from performing repressive tasks, I must resort to use the indicator for presence as my independent variable. The indicator for presence ranges from 0 to 1 where 0 indicates that there was no presence of a PGM and 1 indicates that the country had at least one PGM

(19)

present. Where there is uncertainty regarding the presence of the PGMs, the last year of recorded activity is taken as the dissolvement of the specific PGM. Information on pro- government armed groups is gathered by searching for related articles by major international media sources through LexisNexis. Relevant sources are then thoroughly examined (Carey et al, 2012). If there is uncertainty about the relationship between an armed group and a

government, coders proceed by examining country-specific sources and academic papers. The creators of the dataset note that one problem is that all PGMs are most likely not captured in the dataset as a result of the limited availability of sources for the time-period and that the link between some armed groups and governments have yet to be exposed³.

Dependent variable: Repression

Repression can be difficult to measure because it includes violations that is not necessarily known to the individuals targeted. Surveillance is one example of a violations that can bypass not only the targeted individuals’ knowledge but also external actors’ knowledge of its

occurrence. While all types of government illegitimate behaviour might bypass the eyes of NGOs, IOs or other observing actors, severe violations of physical integrity are in most cases more visible to external actors. In an attempt to measure state repression, scholars have turned to the Political Terror Scale. It is designed to measure political terror which they define as

“violations of basic human rights to the physical integrity of the person by agents of the state within the territorial boundaries of the state in question” (Haschke, 2017). Violations of physical integrity rights include:

“torture and cruel and unusual treatment and punishment; beatings, excessive use of force, brutality; rape and sexual violence; killings and unlawful use of deadly force;

summary or extra-judicial executions; political assassinations and murder; political imprisonment, arbitrary arrest and detention; incommunicado and clandestine imprisonment and detention; forces disappearances; kidnappings, forced relocations and removal” (Haschke, 2017:1).

Agents of the state include police, military, deathsquads and militias etc⁴. To test my

argument that the relationship between the presence of PGMs and repression is dependent on media freedom I will therefore use the 5-point Political Terror Scale (Gibney et al, 2017) to

3 For a more detailed description of the coding process, please see the Pro-Government Militia Database Codebook. Found at: http://www.sabinecarey.com/militias-data/

4 For a full list, please see the Political Terror Scale Codebook. Found at:

http://www.politicalterrorscale.org/Data/Download.html

(20)

measure the level of repression committed by a state in a given year⁵. The PTS ranges from 1 to 5 in which 1 indicates the lowest amount of violations and 5 the highest. The coding scheme used is presented as follows:

1. “Countries under a secure rule of law, people are not imprisoned for their view and torture is rare or exceptional. Political murders are extremely rare.

2. There is a limited amount of imprisonment for nonviolent political activity.

However, few persons are affected, torture and beatings are exceptional.

Political murder is rare.

3. There is extensive political imprisonment, or a recent history of such imprisonment. Execution or other political murders and brutality may be common. Unlimited detention, with or without a trial, for political views is accepted.

4. Civil and political rights violations have expanded to large numbers of the population. Murders, disappearances, and torture are a common part of life.

In spite of its generality, on this level terror affects primarily those who interest themselves in politics or ideas.

5. The terrors of Level 4 have been extended to the whole population. The leaders of these societies place no limits on the means or thoroughness with which they pursue personal or ideological goals (Haschke, 2017).”

The PTS has three different indicators of political terror that uses the same coding scheme.

Each indicator is based on information gathered from three different annual sources: the US Department of State Country Reports on Human Rights Practices, the Amnesty International Annual Report (Wood & Gibney, 2010:372), and the recently added Human Rights Watch World Report (Haschke, 2017). A minimum of two professional coders are asked to read each report for each country while ignoring own knowledge and bias in order to assign a score of 1 to 5 for each country. The dataset includes measurements of over 180 countries for the time period 1976 to 2016 (Haschke, 2017). Although there is some evidence that early year reports of the US State department are guilty of bias (Poe et al, 2001) I will rely on the indicator that is based on the reports from the US State department as it allows for more consistent and extensive coverage compared to the other sources and thus generates an increased amount of observations.

5 Treated as a continuous interval scale

(21)

Interaction variable: Media freedom

Data on media freedom is retrieved from the Global Media Freedom Dataset created by Whitten-Woodring and Van Belle (2015). The dataset provides information on the media environment for 196 countries during the the time period from 1948 to 2014. Media freedom is defined as the ability to freely and safely criticize governments. Focusing on the

environment in which media function rather than on the laws and provisions that permit or prohibit media to function freely allow for a better indicator of media functioning as a control of government behaviour (Whitten-Woodring & Van Belle, 2015:180). Each country in a given year is placed in one of three categories:

• “0–Free: countries where criticism of government and officials is a common part of the political dialogue.

• 1–Imperfectly Free: countries where social, legal or economic costs related to the criticism of government or officials limits public criticism, but investigative journalism and criticism of major policy failing can and does occur

• 2–Not Free: countries where it is not possible to safely criticize the

government or officials, and media are either indirectly controlled or directly controlled” (Whitten-Woodring & Van Belle, 2015:180).

The primary sources used for coding the Global Media Freedom Dataset include the US Department of State Country Reports on Human Rights and the archives of the International Press Institute. If those sources were not sufficient, other sources were used to complete the coding process (Whitten-Woodring & Van Belle, 2015:181)⁶.

3.4 Control Variables

The control variables are selected based on the literature on repression. There are many supposed determinants of repression and there is no room to control for all relevant factors in this paper. Therefore, the chosen control variables are factors that have shown to be the most significant in the literature while also of importance to the relationship analysed in this paper.

Armed Conflict

Domestic conflict, civil war and interstate war has consistently been found to have an impact on human rights violations and repression (Davenport & Armstrong, 2004; Poe & Tate, 1994;

6 For a full list of the sources used, please see the online appendix of (Whitten-Woodring & Van Belle, 2015).

(22)

Richards et al, 2015). The incidence of armed conflict is also of special importance to the relationship between PGMs and repression to see whether PGMs affect repression regardless of the incident of conflict. Although, as Mitchell et al (2014:823) remarks it might seem that PGMs are more common during times of conflict, close to half of all PGMs are present during times where no conflict is observed. Armed conflict is measured using a binary indicator for the incidence of intrastate conflict for each country-year from the UCDP Monadic Conflict Onset and Incidence Dataset version of the UCDP/PRIO Armed Conflict Dataset (Allansson et al, 2017; Gleditsch et al, 2002). The variable is coded 1 when at least one conflict is active in the country-year.

Democracy

Another commonly used variable in the quantitative literature on repression is regime type or specifically, democracy. The reason for including democracy as a control variable is to be able to control for the effect of a free media independent of the institutions of institutionalized democracies. As previously discussed, much of the literature on repression is devoted to analysing the effect of democracy and its institutions on repression to an extent that the effect of democracy has been termed the “Domestic Democratic Peace” (Davenport:2007c).

Davenport and Armstrong (2004) suggests, based on the Polity measure, that there is a threshold for when democracy has a negative impact on repression implying that the relationship between democracy and repression is of non-linear nature. Specifically, the domestic democratic peace only comes into effect above a certain level (0.8) on the Polity measure. In order to capture this threshold effect and allow for easier interpretation of the regression model, a recoded variable off the Polity IV scale is borrowed from Mitchell et als (2014:824) study. Observations with value of 8 or higher (institutionalized democracies) are coded as 1 and observations with a value below 8 are coded as 0.

Economic development

The third control variable regards the country’s economy, economic development or wealth depending on how you see it. Following the argument that citizens are more likely to be satisfied with the regime if they enjoy higher levels of economic development which will decrease internal dissent and thus, reducing the need for political authorities to use repression (Henderson, 1991:126). Economic development will be measured using the log of real GDP per capita in constant 2005 U.S. dollars. The data is retrieved from Gleditsch (2002)

Expanded Trade and GDP Data.

(23)

Population

The last control variable concerns the size of the population in the given country-year.

Population is also a control variable commonly used in the repression literature. It is almost regarded as a standard procedure to include the log of population as control variable because it is found to independently have an effect on levels of repression. The data on population is also retrieved from the Gleditsch (2002) Expanded Trade and GDP Data.

3.5 Validity and reliability

The main problem in terms of validity in this paper regards the dependent variable.

Repression is a concept including a large variety of violations. Violations of physical integrity and infringements on civil liberties are usually what scholars mean when they refer to

repression. In this paper, repression has been operationalized as violations of physical

integrity rights. Measuring levels of physical integrity rights is problematic as there is no way to give a perfectly accurate estimate of a given country’s number of violations. Lower level violations are likely to be less visible for coders than for example political assassinations and will therefore not be included in the assessment of the level of repression. Some countries are less transparent than others resulting in the likeliness of missing observed violations of physical integrity rights. Hidden numbers and lack of information is a problem which leads researches to judge countries and estimate repression rather than measuring the exact amount of violations.

Using indexes for measuring repression can also be a problem. For example, some violations should be considered more severe than others. The PTS takes three dimensions of repression into account when scoring countries: scope, intensity and range. The scope of violations refers to the type of repression. This means that more severe violations are regarded as more

repressive. Intensity refers to what extent in terms of frequency repression is used i.e if

repression is common or rare. Range refers to the percent of the population subject to violence (Haschke, 2017:4). These dimensions of repression make the PTS suitable for comparative research as the level of repression is relative to other countries.

The PTS just as other alternative measurements of repression rate countries based of the reports of the US State Department, Amnesty or Human Rights Watch. Not only is possible that the reports are in itself subject to bias. Coders assigned to read and rate countries based of these reports are in the end more or less subjective. There is no guarantee that the coding

(24)

process is completely free of bias. This is also a reliability concern as different measurements of repression might have conflicting results. Furthermore, this paper does not aim to analyse the causal mechanism of how media freedom affects the relationship between pro-government militias and repression. It does however provide evidence for if there is a difference in effect depending on the level of media freedom.

(25)

4. Results and analysis

This section begins with a presentation of the results from the bivariate and multivariate regression analyses including interaction terms (OLS). The results from the regression output is presented in Table 1. As interaction effects can be difficult to interpret from the regression output visual presentations are also presented. Visual presentations of interaction effects allow for easier interpretation. This section ends with a discussion of the results.

Regression Analysis

The dependent variable for all models in Table 1 is Repression measured by the five-point Political Terror Scale. Higher values on this measure indicate higher levels of repression.

Initially, a bivariate regression was carried out in order to test for the relationship between pro-government militias and repression. The results are presented in model (1). The results show a positive statistically significant relationship between Pro-government militias and repression similar to previous research. As hypothesized in section 2.5, the relationship between PGMs and repression remains even when including a larger scope of violations into the operationalization of repression. In model (2); the variable Media Freedom is included in the regression and the previous relationship remains statistically significant, however the coefficient for Pro-government militias is slightly reduced.

Control variables are included in model (3) in an attempt to isolate the effect of Pro-

government militias on repression. The regression coefficient for Pro-government militias is 0.282 and is statistically significant at 99.9% when controlling for Democracy, Armed Conflict, Population and GDP per capita. The reason for controlling for these specifically is because they have previously shown to affect levels of repression. Furthermore, the

coefficients for all control variables in model (3) are highly statistically significant conforming to previous research. The direction of the respective coefficients

(positive/negative) also correspond with previous research. Substantially, Model (3) suggests that the presence of at least one pro-government militia increase levels of repression

regardless of the level of media freedom, regime type, the incidence of armed conflict, population size and GDP per capita. Furthermore, by looking at the adjusted R2 values and the residual standard errors of models (1), (2) and (3), we can conclude that model (3)

provides the best fit out of the three, accounting for approximately 58% of the variance in the dependent variable repression. Conclusively, the results presented in model (3) supports previous quantitative research on repression.

(26)

Table 1. The interaction of media freedom on the relationship of pro-government militias and repression

Dependent variable: Repression (Political Terror Scale)

(1) (2) (3) (4) (5) (6)

Pro-government

militias 1.212^*** 1.011^*** 0.282^*** 0.615^*** -0.004 0.375^***

(0.035) (0.032) (0.032) (0.066) (0.057) (0.061)

Media Freedom 0.579^*** 0.278^*** 0.512^*** 0.225^*** 0.231^***

(0.017) (0.022) (0.020) (0.023) (0.025)

Democracy -0.258^*** -0.258^*** -0.255^***

(0.039) (0.039) (0.043)

Armed Conflict 1.022^*** 1.020^***

(0.037) (0.037)

Population (log) 0.162^*** 0.157^*** 0.184^***

(0.009) (0.009) (0.009)

GDP per capita

(log) -0.231^*** -0.229^*** -0.277^***

(0.012) (0.012) (0.013)

PGM * Media

Freedom 0.277^*** 0.206^*** 0.211^***

(0.041) (0.034) (0.038) Constant 2.101^*** 1.424^*** 2.403^*** 1.503^*** 2.496^*** 2.706^***

(0.019) (0.027) (0.128) (0.029) (0.128) (0.140)

Observations 4,181 4,176 3,870 4,176 3,870 3,870

R² 0.223 0.384 0.576 0.390 0.580 0.497

Adjusted R² 0.222 0.383 0.575 0.390 0.579 0.496

Residual Std.

Error

1.031 (df = 4179)

0.917 (df = 4173)

0.748 (df = 3863)

0.913 (df = 4172)

0.745 (df = 3862)

0.815 (df = 3863)

Notes: Ordinary Least Squares (OLS) regression is employed. Coefficients are reported with standard errors in parenthesis. Significance: *p<0.1; **p<0.05; ***p<0.01

(27)

Hypothesis (1) which suggests that pro-government militias increase repression finds support in model (1), (2) and (3). This is important because of its significance for testing hypothesis (2) and answering the research question of this paper. Seeing that the effect of pro-

government militias is statistically significant in model (3), there is enough evidence to go forth in exploring the potential moderating effect of media. Hypothesis (2) is based on that there is evidence for the argument that the presence of pro-government militias increases levels of repression. In the following models of Table 1 the interaction effect or in other words, the product of pro-government militias and media freedom is introduced. However, raw regression outputs with interaction effects can be tricky to interpret. When including interaction terms in regression the coefficients can no longer be interpreted as in additive regression (Brambor et al, 2006:72). Therefore, I will only briefly comment on the models and then proceed with further and more interpretable presentations of the results.

The coefficient for PGM * Media freedom in model (4) is statistically significant suggesting that there is an interaction effect. Yet, the model cannot show how media freedom interacts with pro-government militias and how the effect varies over different values of media freedom. The only possible interpretation of the model is that the effect of pro-government militias on repression is 0.615 when the value of media freedom is 0 (Free) which in itself is interesting but is not enough to answer the research question of this paper. In model (5), all of the control variables are included. The interaction term remains statistically significant

suggesting that there is an interaction effect on top of controlling for Democracy, Armed conflict, Population and GDP per capita. The coefficient for Pro-government militias is heavily reduced up to a point where it is now negative while also losing its statistical

significance. The loss of significance however, does not mean we reject the results since we are not interested in the “main” effects but rather the effects of the interaction term. Model (6) includes all control variables except for Armed conflict. The reason for this will be evident after the visual presentation of the interaction effect in model (5). Nevertheless, the

interaction term in model (6) is statistically significant just as in previous models.

In order to easily interpret the interaction effects, interaction plots and Johnson-Neyman intervals will be conducted and presented. Johnson-Neyman intervals tell us if the coefficient for Pro-government militias is significant at certain values of Media freedom. Figure 1 provides a visual presentation of the interaction effect in model (4). It presents the regression line for each value of Media freedom i.e. the coefficient of Pro-government militias when Media freedom is “Free”, “Imperfectly Free” and “Not Free”. A steeper line indicates a greater coefficient. The coefficients of Pro-government militias at each value of Media

(28)

freedom are all statistically significant⁷. Hypothesis (2) finds support in figure (2). The effect of pro-government militias on repression varies depending on the level of media freedom.

Figure 2. Interaction plot for Model (4).

Judging by Figure (2), the effect of pro-government militias on repression is decreasing as media becomes more free. The simple slopes analysis provides the coefficients (i.e. the slopes of the regression lines in Figure 2) for pro-government militias at all levels of media

freedom⁸. Initially, the results confirms hypotheses (2) that the the effect of pro-government militias on repression decreases in magnitude when there is higher levels of media freedom.

Nevertheless, these results need to be tested further in order to claim that the effect is moderated by media freedom.

The control variables used in model (3) are also implemented in model (5) in an attempt to isolate the interaction effect. Again, the regression output is not sufficient for a full

interpretation of the interaction effect. A simple slopes analysis of model (5) provided the slope of pro-government militias at all levels of media freedom controlling for democracy, armed conflict, population and GDP per capita. The results showed that the effect was not significant across all levels of media freedom. A visual presentation of this is presented in Figure (3). The Johnson-Neyman plot demonstrates the slope and significance for X on Y when another variable Z is held at some specific values.

7 Full coefficients and their significance in figure (2) can be found in the simple slopes analysis in the appendix

8 Not Free = 1.17***, Imperfectly Free = 0.89***, Free = 0.61***

(29)

Figure 3. Johnson-Neyman plot for Model (5)

As presented in Figure (3), the slope of pro-government militias is only significant when the value of media freedom is either 1 (imperfectly free) and 2 (not free). By observing figure (3), a rough estimation of the slope can be made⁹. Since the slope for pro-government militias is not significant when the value of media freedom is 0 (free), no conclusions can be drawn for the impact of free media on the relationship between pro-government militias and repression.

However, the effect of pro-government militias on repression is greater when media is not free compared to imperfectly free. Substantially, this means that in countries where journalists have limited, albeit existent, possibilities to criticize government behaviour, governments use of militias for repression is not as effective compared to governments in countries where media is fully restricted. This suggest that free media can possibly have a limiting effect on governments use of militias for repressive tasks. Although the results are not conclusive, it opens up for further investigation of the effect of media freedom.

Model (6) does not include the control variable armed conflict. The reason for this is related to the complication of insignificance in the previously discussed model (5). Leaving out armed conflict in the regression model results in a significant slope for pro-government militias when media freedom is held at 0 (free), thus yielding conclusive results for the slope at all levels of media freedom. A visual presentation of model (6) in the form of an interaction plot will not allow for easy interpretation as the difference in the regression lines has been

9 Exact coefficients and their significance in figure (3) can be found in the simple slopes analysis in the appendix

Slope of pro-government militias

Media freedom

(30)

reduced compared to model (4). For that reason, the coefficients for pro-government militias are reported in Table (2).

Table 2. Coefficients of pro-government militias for Model (6) Level of Media freedom Coefficient (slope) for Pro-

government Militias

Free 0.37*** (0.06)

Imperfectly free 0.59*** (0.03)

Not Free 0.80*** (0.04)

Notes: Simple Slopes Analysis. Coefficients with standard errors in parenthesis. Significance: *p<0.1; **p<0.05;

***p<0.01

As seen in Table (2), the effect of pro-government militias on repression is at its lowest when media is free. The results support initial results from model (4) and (5) suggesting that the media environment affects governments use of pro-government militias. Hypothesis (2) finds support in the results of model (6). In terms of interaction, the effect of pro-government militias on repression varies across different levels of media freedom. The effect has decreased in magnitude when media is free compared to imperfectly free and not free. The flaw of model (6) in contrast to model (5) is the loss of explanatory power, based primarily on the reduced adjusted R² value (0.579 compared to 0.496) as a result of excluding armed conflict. The reason for controlling for the incidence of armed conflict in the first place is of great importance to the relationship between pro-government militias and repression.

Repression levels are of course higher in times of conflict and controlling for armed conflict allow us to see if the effect of pro-government militias on repression remains outside of conflict. Consequently, the exclusion of armed conflict undermines the results because of its importance to levels of repression.

5. Discussion

In the previous section it was concluded that hypothesis (1) finds support in the regression analysis. Hypothesis (2) also finds support in the regression analysis but to a lesser extent.

Most notably, the multiple regression analysis excluding the interaction term and the conclusions drawn from it resemble and bolster previous research on the determinants of repression and the relationship between pro-government militias and repression. Secondly, the regression analysis including the interaction term provides less conclusive results although interesting. That the effect of pro-government militias on repression when media is free was

Limiting the use of pro-government militias

Limiting the use of pro-government militias

Table of Contents