Coercion,capacity,and coordination: Predictors of political violence

Abstract

Using a risk assessment method developed by Gurr and Moore (American Journal of Political Science 41: 1079–1103, 1997) and applying O’Brien’s (Journal of Conflict Resolution 46: 791–811, 2002) risk assessment metrics, we present a global, comparative, cross-national model predicting the states where political violence is likely to increase. Our model predicts more political violence when governments violate the physical integrity rights of their citizens—especially when they frequently imprison citizens for political reasons or make them “disappear”. These coercive techniques may create more citizen dissatisfaction than other types of violations of physical integrity rights, because citizens perceive political imprisonment and disappearances as the direct result of the deliberate policy choices of politicians. Our model also forecasts more political violence in weak states and states that allow dissatisfied citizens to coordinate their anti-government activities. Specifically, we demonstrate that political violence tends to be higher if governments respect their citizens’ right to freedom of assembly and association and offer widespread use of mobile phone and internet technology.

Keywords

Domestic violence human rights internet mobile phone personal security rights physical integrity rights political violence rebellion state capacity state coercion

What factors make domestic political violence more likely to occur? On 1 October 2010, one day after the start of the recent Ecuadorian protests, this was the question many were asking. After a turbulent decade, Ecuador was an improving state, on the road to economic recovery and democracy (Batty, 2010). However, it was not immune to violent protest; thousands of people took to the street in Quito on 30 September to protest changes in government policy. These protests were violent, well-orchestrated and, according to some, motivated by recent increases in human rights abuses (UN, 2010). At their height, many feared the violence would bring an overthrow to the Correa regime. Ecuador is not alone. Over 1100 events of domestic anti-government violence occurred in 2009. Some of these events were low-scale, involving minor stone-throwing and defacement of public property by groups of unemployed youths. Other events were much more violent, leading to state responses squarely qualified as the beginning of civil wars. What causes this anti-government violence and, moreover, what can be done to limit its magnitude?

We define political violence as organized anti-government aggression that occurs within a state by a domestic population against its own government (Carey, 2006; Gurr, 1968; Moore, 1998).¹ Building on the existing literature, we identify three main conceptual factors that affect the degree of political violence within a state: coercion, coordination and capacity. We argue that coercion, defined as violations of physical integrity rights, makes citizens more willing to commit acts of political violence. Coordination, defined as the availability and ease with which domestic groups can cooperate, organize and mobilize, increases the ability of citizens to commit acts of political violence. Conversely, capacity, typically thought of as the ability of the state to project its power throughout its territory, decreases the opportunity of citizens to engage in political violence by raising the costs for potential rebels. Using the now common distinction between willingness and opportunity, coercion increases willingness and coordination and capacity affect opportunity in opposite ways (Most and Starr, 1989). Our findings suggest that, if governments coerce their citizens by imprisoning many of them for political reasons or making many citizens disappear, political violence will be higher. If citizens with grievances can coordinate their responses, or if they face a weak government, political violence also will be higher.

We make several contributions to the current explanations of political violence. First, some previous scholarship has argued that the use of political violence by citizens leads to the use of coercion by governments (Carey, 2006; 2010; Davenport, 2007; Most and Starr, 1989; Poe and Tate, 1994),² while other scholarship has claimed that the direction of the causal relationship is reversed (Gartner and Regan, 1996; Gurr, 1968; Gurr and Moore, 1997; Opp and Ruehl, 1990; Regan and Norton, 2005; Walsh and Piazza, 2010). This work provides additional evidence that the predominant causal force is increased government indiscriminate coercion, which leads to the increased intensity of political violence. Second, we emphasize that the ability of citizens to coordinate their anti-government activities substantially and independently contributes to the likelihood that those activities will become violent. Third, by focusing on the conceptual distinction between capacity and coercion, we highlight a key distinction between coercion potential, defined here as state capacity, and the actual use of coercion as factors affecting anti-government violence. Attention to this distinction may help explain some of the divergent empirical findings of the current literature on political violence (Bhasin, 2008; Carey, 2006; Regan and Henderson, 2002).

Finally, our work highlights the crucial impact of two types of physical integrity violations—political imprisonment and disappearances—as stimulators of political violence. Physical integrity rights are a subset of human rights that protect citizens from torture, extra-judicial killing, political imprisonment and disappearance by government authorities. Previous research has shown that indices measuring overall respect for these rights are inversely related to the outbreak of civil war (Regan and Norton, 2005) and levels of domestic and international terrorism (Walsh and Piazza, 2010). Our work suggests that some types of physical integrity violations create greater grievances among the citizenry, because they are perceived by citizens as resulting from the deliberate policy choices of politicians. Other types are more likely to be perceived as resulting from the exercise of administrative discretion by police, prison guards and soldiers.

Empirically, we make two principal contributions to the field. First, we introduce a new ratio measure of the intensity of political violence. It is based on events reported by the Reuters Global News Service between 1990 and 2009. Each country’s annual score accounts for all acts of political violence reported by Reuters during that year, weighting each event by its intensity. This new measure does not depend upon arbitrary ordinal thresholds such as numbers of deaths and includes all types of political violence rather than isolating particular types such as civil wars or acts of terrorism as the focus of anlaysis. Second, using our pooled times series results from 1990 to 2009, using a risk assessment method developed by Gurr and Moore (1997) and later used by Poe et al. (2006), we predict the countries where political violence is most likely to increase in subsequent years. Few previous studies of political violence have generated out-of-sample predictions, and the precision, accuracy and recall percentages for our model exceed the current standards for evaluating models of political violence (O’Brien, 2010).

Theory

What makes people take to the street violently? The literature has established, to varying degrees of importance, that both willingness and opportunity are required in order for a population to take violent actions against their government (Gurr, 1968; Oliver and Myers, 2002; Rasler, 1996; Tilly, 1973). Below, we outline these two approaches, incorporating our framework of coercion, coordination and capacity into this literature.

The willingness to commit acts of political violence

The literature assumes that an individual who rebels against her government must be dissatisfied with it. Gurr (1968) argued that feelings of “relative deprivation”, defined as beliefs that what has been provided or obtained is lower than what was expected, is a necessary condition for dissatisfaction, which could lead to dissident action. There are many possible causes of dissatisfaction including poor economic conditions (Richards and Gellany, 2006), the political arrangement within the polity or the government’s treatment of citizens (Elbadawi and Sambanis, 2002; Rasler, 1996), including the violation of their human rights.

We focus here on whether government use of indiscriminate coercion, or repression of human rights, causes more intense domestic political violence, where repression is defined as actions taken by the government which raise the costs of disagreeing with the regime in power (Carey, 2006; Moore, 2000; Tilly, 1978). Typically, the concept of repression refers to all forms of human rights violations perpetrated by the state or agents of the state. According to the bulk of this research, in the long term, government repression causes an increase in the scope and intensity of grievances held by a population, leading to more intense domestic political violence (Gurr and Moore, 1997; Regan and Norton, 2005; Walsh and Piazza, 2010). Previous studies examining this causal relationship have concentrated on the effects of government violation of physical integrity rights, because the norm against violating that class of human rights is the most widely shared among countries and cultures.

Indiscriminate violation of physical integrity rights is expected to increase the size of the targeted group and the intensity of its grievances while decreasing the proportion willing to accept policy accommodation. By banding together, members of the targeted group can offer mutual protection from future violation of their rights by government agents. Thus even without many resources, dissident groups are likely to grow (Clay, 2012). As the dissident group grows, the government is likely to use even greater amounts of coercion, and a positive feedback cycle is created. Over time a dissident group is likely to move from non-violent dissent to political violence (Murdie and Bhasin, 2011; Regan and Norton, 2005). Although repression can increase the potential costs faced by protesters, it can also increase “micro-mobilization”, defined as the provision of rallying cries around which protest groups can recruit new members into their movement (Opp and Ruehl, 1990; Rasler, 1996). As such, even though the use of indiscriminate repression, by definition, raises the costs of protest, it could also spread these costs among a more diverse pool of protesters who see the increased need for their activities (Clay, 2012). If this occurs, repression should be associated with more mobilization and more political violence in the future.

Most research on counterinsurgency reaches conclusions that are consistent with this argument. The literature suggests that, when counterinsurgency forces use brutality against suspected insurgents and civilians, popular support for the insurgency is likely to grow (Joes, 2004). Repression of physical integrity rights also promotes domestic and international terrorism (Walsh and Piazza, 2010). Moreover, when governments violate physical integrity rights, they also risk losing the support of the international community (Hoffman, 2004), and the loss of that support also can help fuel domestic violence (Walsh and Piazza, 2010). Foreign governments that abide by human rights norms may be unwilling to cooperate with counter-terrorism or counter-insurgency efforts and may even be prepared to provide moral or material assistance to insurgent groups as they did in Egypt, Libya, Tunisia and Syria.

Previous research has assumed that all types of physical integrity rights violation have an equal stimulating effect on political violence by focusing on the degree, as opposed to the type, of government violation of physical integrity rights (Regan and Norton, 2005; Walsh and Piazza, 2010). Here we argue that the types of physical integrity violations matter. Politicians may order that the police, soldiers or prison guards commit violations of physical integrity rights, but they rarely commit those violations themselves. Because politicians delegate the responsibility for respecting human rights to bureaucratic agents, this relationship can be usefully viewed as a principal–agent problem (Butler et al., 2007; Conrad and Moore, 2010; Mitchell, 2004, 2012). Viewed in this way, violations of physical integrity rights result in part from orders by political leaders (politicians) and in part from the exercise of bureaucratic discretion.

Violations of physical integrity rights perceived by citizens as the result of deliberate, strategic decisions of politicians rather than from the exercise of bureaucratic discretion by their agents will create the most citizen dissatisfaction and will, therefore, stimulate the most political violence. Violations perceived by citizens as the result of the exercise of bureaucratic discretion will cause less citizen dissatisfaction with the regime and, as a result, stimulate less political violence. Taking and holding political prisoners or making citizens disappear is much more likely to be, and be perceived as, the result of strategic decision-making by politicians, so the impact of these practices on political violence is likely to be greater. Political imprisonment refers to the incarceration of people by government officials because of their speech, their non-violent opposition to government policies or leaders, their religious beliefs, their non-violent religious practices including proselytizing or their membership of a group, including an ethnic or racial group. Disappearances refer to people who have simply vanished, where political motivation appears likely and the victims have not been found. Knowledge of the whereabouts of the disappeared is, by definition, not public knowledge. However, it is typically known by whom they were taken and under what circumstances (Cingranelli and Richards, 2010).

Principals can control the implementation of policies imprisoning people for political reasons or making them disappear more easily than they can control torture or extra-judicial killing. Holding political prisoners is a long-term, calculated decision, not one that results from impulse. For example, political imprisonment and disappearances were the coercive tools of choice for the governments of the former Soviet Union and communist-controlled Central Europe. Implementation of these policies was carefully controlled and documented. Political imprisonment is still widely used against potential insurgents and their leaders. Many states have enacted preventive detention and anti-terrorism legislation allowing the imprisonment of people for long periods without access to legal counsel or the opportunity to have a speedy trial. Politicians commonly defend these policies as necessary to protect public safety, but individuals incarcerated under these laws are political prisoners under international law. There are many examples where holding political prisoners, such as the members of the Muslim Brotherhood in Egypt or the Irish Republican Army, fueled domestic political violence.

Previous research has indirectly supported the claim that there is less agency loss in policies of political imprisonment and disappearances by emphasizing the difficulty politicians have in controlling their agents’ use of torture or extra-judicial killing, especially during periods of intense political violence (Conrad and Moore, 2010; Mitchell, 2004, 2012). Torture refers to the purposeful inflicting of extreme pain, whether mental or physical, by government officials or by private individuals at the instigation of government officials. It includes the use of physical and other force by police and prison guards that is cruel, inhuman or degrading. Extra-judicial killings are deliberate, illegal killings by police, prison guards or soldiers without due process of law. Typically, the victims are criminal suspects, detainees or prisoners.

These violations of physical integrity rights typically result from the behavior of overzealous state agents operating outside of the policies set by politicians. Research shows that, even if agents are told not to use torture, some agents will still use it as an effective means of punishing, gaining compliance or extracting information (Conrad and Moore, 2010). Police and prison guard brutality, including sexual assaults, meet the international definition of torture. Both practices are widespread and difficult to eradicate. In contrast to their usual defense of political imprisonment policies, politicians usually loudly condemn torture and extra-judicial killing when these practices are uncovered.

Hypothesis 1: when politicians decide to use coercion by imprisoning large numbers of citizens for political reasons or by making many of them disappear, there is likely to be an increase in the level of domestic political violence.

The opportunity for political violence

The literature also suggests that opportunity affects the ability to conduct domestic political violence (Regan and Norton, 2005; Tilly, 1978). Repression of freedom of association and assembly rights could be one way to limit opportunities for the organization and coordination necessary for individuals to mobilize domestic anti-government violence. When disenchanted citizens are not able to communicate their frustrations to each other and coordinate their responses, it increases the costs of domestic political violence, making any action less likely. This anti-coordination tactic has been used recently in Russia, where the regime has restricted all foreign non-governmental organizations and severely restricted civil society groups that could aid in organizing anti-government violence (Bogoroditskii, 2010).

Widespread anti-government violence requires coordination among dissidents about locations, tactics and responses to pressure from government agents. It also requires knowledge of some type of government grievance, such as the use of coercive measures mentioned above. The communication of this information requires connections both within the state and to the larger international community. As such, this logic is really an extension of the larger sociological literature on political mobilization, which links international standing and connections to protest (Smith et al., 1997; Tsutsui, 2004).

Given this line of reasoning, it follows that citizens can communicate their grievances about the government’s use of coercion and can coordinate an appropriate response when they have access to internet or mobile phone technology or both. When these technologies are widespread, it is likely that groups of individuals will be able to work together to organize strong protests against the state, perhaps including anti-government violence. Many referred to the “Arab Spring” revolution of 2011 as the “Twitter Revolution” or “YouTube Revolution”. Similarly, in Ukraine in 2004, mobile phones and internet access aided in the diffusion of information concerning election results and coordination about protest activities (Goldstein, 2007). Uses of technology were also present in the recent Iranian elections of 2009; Time Magazine called the internet site Twitter the “medium of the movement” (Grossman, 2009). Additionally, as Gregory (2010) reported, mobile phones with cameras also allowed protesters to capture police brutality and other forms of coercive human rights violations on the scene; these images can then be broadcast to sympathizers, increasing the overall movement size and leading to more violent mobilization.³

Coordination through electronic communication appears to be increasingly important as a tool for protest mobilization. Communication serves as a way both to disseminate information about grievances to the general population and to provide the details of upcoming events to protesters. In a similar vein, coordination concerning non-violent protest could aid in coordination concerning violent anti-government protest. Many groups use similar strategies at different points in their movement (Bartley, 2007; Oliver and Myers, 2002). Coordination resources used by non-violent groups will probably diffuse to violent groups (Haines, 1984). A meeting for a non-violent event could provide the space and connections for a violent offshoot to similarly plan their next move (Haines, 1984). In this way, when coordination resources are high, either because the state protects free assembly and association rights or because of widespread access to mobile phone, internet technology and existing non-violent protest, violent protest is more likely.

Hypothesis 2: The greater the ability of potential rebels to coordinate their activities, the more likely it is that there will be a high level of domestic political violence.

So far, we have outlined two concepts, coercion and coordination, that we argue are related to an increase in the frequency and intensity of domestic political violence. Given these arguments, it could seem that the state is somewhat powerless in the wake of a mobilized population. However, we argue that this is not the case. In addition to the role development can have in diffusing the need for violent protest, as discussed above, we also argue that state capacity, as defined as the ability and compliance of state agents to enact state policy throughout the polity, will lessen domestic violence.

When a state has the potential to respond to protests and has control over its agents, this raises the likely costs of domestic anti-government violence for the protesters without triggering the same micro-mobilization processes that the actual use of coercion does. State capacity serves as a signal to the domestic population that the state is in charge and that violence will be less successful in bringing about the changes in policy that the protesters would like (Boulding, 1962; Kalyvas, 2006). State capacity can also aid in the provision of public goods, lessening, perhaps, the need for protests in the first place (Besley and Persson, 2009).

There are many indicators of state capacity (Hendrix, 2010). For the relationship between state capacity and political violence, there appear to be two major forms of state capacity that diminish conflict: control of territory and a strong military apparatus (Gurr, 1968; Kalyvas, 2006). By control of territory, we are referring to the ability of government agents to spread their power even to the peripheries of the state territory. This ability to effectively project state power throughout a state’s territory has been shown to limit rebellion, one type of domestic political violence. Past research has focused on the presence of adequate roads and the availability of electrical power, for example, as indicators of state capacity (Agnew et al., 2008).

The second pathway through which state capacity could impact civil strife is through the existence of military personnel. Soldiers, as Hendrix (2010) points out, are thought of as a sign of overall military power and state capacity. Military personnel also are typically associated with a thriving bureaucratic structure, another necessary component for a capable state. However, like Gurr (1968), we allow for the possibility of military personnel having a curvilinear impact on civil strife. At low and high levels, military personnel will be associated with less political violence. When military personnel numbers are low, the state has alternative ways of reacting to the demands of its population. When military personnel numbers are high, the state’s leadership is signaling that it could deal with domestic anti-government violence with the use of force. At medium levels, however, military personnel may not be able to effectively control the population and will actually be associated with more political violence. Consistent with this logic, Wintrobe (2006) demonstrated that, in societies with either low or high levels of military personnel, citizens believed that the state was strong, leading to less mobilization against the government. In other words, lower state capacity would only be captured by a medium level of military personnel; sufficiently high and sufficiently low levels of military personnel would indicate a higher level of state capacity.

In addressing the use of military personnel as a proxy for state capacity, we diverge, however, from Gurr’s (1968) contention that the mere presence of military personnel is a proxy for the state’s use of coercion against its citizens. We view the size of the military as an indicator of the potential to use repression against a population, but not as the actual use of repression against a population, a subtle but important difference. As outlined above, it is the actual use of coercion that is associated with micro-mobilization, leading to more violent protest. Military personnel, instead, are associated with coercive potential and, thus, state capacity. The existence of a large army allows for a potentially high level of government coercion, but states with large armies do not necessarily have higher levels of human rights abuses. A large army indicates that the state is capable of dealing with political violence and that, in effect, the state has complete control over its territory. Similarly, when there are extremely low levels of military personnel, the state is also seen as capable, through either norms of engagement or outside support, of dealing with anti-government violence. A capable state limits the opportunity for anti-government violence. It has the potential to respond quickly and effectively to increased mobilization throughout its territory, increasing the cost of protest and, thus, leading to less civil strife. This logic has the following empirical implication:

Hypothesis 3: Higher state capacity will lead to lower levels of domestic political violence.

Building our statistical model

Our analysis includes all available states of the world for the years 1990–2009. In this section, we describe the measures of our dependent and independent variables and our statistical model. In the following sections, we present our in-sample statistical results, and we use that model to make out-of-sample predictions.

Dependent variable: domestic political violence

Our dependent variable is a relatively new measure of the overall level and intensity of domestic anti-government violence within a state in a given year developed by Bhasin (2008) and refined by Murdie and Bhasin (2011). Existing studies of political violence typically rely on the Cross-National Time Series Archive (Banks, 2008). The Banks dataset, coded from New York Times reports, is a yearly count of the number of various protest events within a state. A domestic political violence variable typically combines counts of assassinations, guerilla warfare, riots and revolutions (Murdie and Bhasin, 2011; Schatzman, 2005). Various critiques of Banks (2008) have been provided in the literature, including biased news coverage, arbitrary coding thresholds and the inability to isolate only protest directed at a government or government agent (Nam, 2004).

In light of these critiques, Bhasin (2008) and Murdie and Bhasin (2011) argue for an alternative measure of domestic anti-government violence that relies on the Integrated Data for Event Analysis (IDEA) framework (Bond et al., 2003). IDEA, produced by Virtual Research Associates (VRA), is a data set of all events in Reuters Global News Service. The data are publically available from 1990 to 2004 at the daily level. We obtained the data from VRA through 2009. The events are organized in a “who” did “what” to “whom” manner; this framework allows researchers to isolate events of interest for their particular project. The use of the IDEA data allowed us to make use of a news source with a much broader international presence, to depend less upon arbitrary ordinal thresholds, and to isolate domestic violence against a government. Bhasin (2008) and Murdie and Bhasin (2011) parse the IDEA dataset to only events where a domestic group or individual is the source of an action, various political violence or protest activities are the events, and the target of the action is a government or government agent within the same country as the source of the event. Over 50,000 violent domestic political violence events were identified using this approach.⁴

Murdie and Bhasin (2011) use a country–year count of all domestic anti-government protest events identified in IDEA as their key dependent variable. Although this variable closely follows Banks (2008), a simple count of all political violence does not capture the relative intensity of each event. In other words, this operationalization could not differentiate between 15 blockades of government property and 15 assassination attempts on government leaders. Given that our chief interest is not only in predicting the volume of political violence but also in predicting its intensity, we weight the count used in Murdie and Bhasin (2011) by the intensity or violence of each event. In this way, our key dependent variable allows for differentiation between country–years with high numbers of low-level violence and country–years with high numbers of more extreme political violence.

To capture intensity, we rely on augmented Goldstein (1992) scores developed by VRA. Goldstein (1992) scores are a weighting system developed from a survey of foreign policy officials on the “conflictual” or “cooperative” nature of various political or economic events; smaller numbers correspond to more violent or “conflictual” events. For example, a hijacking of a government vehicle has an augmented Goldstein (1992) score of −7.3755. Although this is still a violent form of protest, it is not as intensely violent as an actual coup (−9.938). The use of this information aids in the differentiation among states with different intensities of violent protest. In creating our final variable, therefore, we weight each event in Murdie and Bhasin (2011) by its corresponding augmented Goldstein (1992) score before aggregating to the country–year. This resulting measure is close to our central concept of interest, namely the overall intensity of violent forms of anti-government behavior within a country in a given year.⁵ We take the absolute value of this measure to allow for a more intuitive interpretation. Higher values on the dependent variable thus equate to greater levels of intense domestic anti-government political violence. There is likely to be bias in the amount of news coverage given by the Reuters News Service to political violence in different countries. Thus, in all results presented below, we include a count of all events of any type reported by Reuters within each country for each country–year. We label this control variable “coverage” and use the natural log of the raw count to account for variance non-stationarity.

In creating the dependent variable, we code country–years without a violent protest event as a zero if there was at least one event of any type in Reuters concerning that country in that year. In an alternative coding of the dependent variable, available on the online Appendix, we only include country–years where Reuters reported at least one event of non-violent or violent protest in the year. Results remain similar in sign and statistical significance across these different specifications and we use results from both approaches in our risk assessment predictions.

Independent variables

We incorporate independent variables that have previously been used to measure our concepts of coercion, capacity and coordination along with new variables that have not previously been specified in the protest literature but that capture components of these three concepts.

Coercion variables

We include a number of variables that represent the coercive or repressive nature of a state. These measures are meant to capture the actual coercive activities that a state engages in, not their capacity to do so. We include measures to capture government use of torture, political killings, disappearances and political prisoners. These are the constituent variables that make up the CIRI index of government respect for physical integrity rights (Cingranelli and Richards, 2010). The range on each variable is 0, 1 or 2, with higher values indicating more respect for that specific human right. So, a 2 indicates that the state does not commit the violation. A 0 indicates that the state committed the violation 50 or more times during the year.

Coordination variables

We include four variables to capture the concept of coordination: freedom of association, mobile phone subscribers, internet users and non-violent protest. These variables capture the potential ability of the population to spread information about grievances against the government, organize and mobilize. The expectation is that increases in any of these variables will lead to increased levels of domestic political violence. The freedom of association measure is also collected as a part of the CIRI data. This measure also varies from 0 to 2, with higher values indicating conditions where individuals are free to associate and assemble with each other. The mobile phone subscribers and internet user measures are both percentages of the population that have access to that technology (World Bank, 2012). Finally, we include the level of non-violent protest because peaceful protest is evidence that the population is able to spread information about grievances against the government, to organize, and mobilize to present their grievances. Other research suggests that non-violent protest can often diffuse coordination resources to violent protest (Bartley, 2007). In the models presented here, we do not lag this variable. In an alternative specification available in the online Appendix, non-violent protest is lagged one year and results are similar substantively and statistically. We also use models with this specification in our risk assessment.

Capacity variables

Finally, we include a series of variables to capture state capacity or “coercive potential”. We use three variables to capture this concept: GDP per capita, military personnel and electric power consumption. These measures come from the World Bank World Development Indicators. Similar to Fearon and Laitin (2003: 80), we see GDP (gross domestic product) per capita as a “proxy for a state’s overall financial, administrative, police, and military capabilities”. We expect that increases in GDP per capita will decrease the magnitude of domestic political violence. The military personnel measure is a count of the number of individuals in a state that are members of the military. As military size increases, a state is in a stronger position to either quell or deter domestic violent actions by the population. This variable provides a specific measure of a tool states can use to control their populations. To account for any nonlinear effects, we include military personnel² in our model. Electric power consumption provides a novel proxy for state infrastructure. In order to have high volumes of power consumed in a state, it is usually necessary to have a government that either builds or protects the generation of power. This capacity is predicted to decrease domestic political violence.⁶

Control variables

We include a number of control variables that are conventionally found in models predicting domestic political protest, civil war and state-directed terrorism, as each of these types of conflict are included in the broad scale of violence that we implement here. We include a measure of ethno-linguistic fractionalization. This index is based on data from Atlas Narodov and gives the probability that two randomly drawn individuals in a country are from different ethno-linguistic groups (Fearon, 2003). This is a ratio level variable, with higher values indicating more ethnic diversity. We include a measure of Regime Type, captured by the Polity2 IV data (Marshall et al., 2009). We use a −10 to 10 ordinal scale. We control for a state’s logged population to account for the different sizes of states.⁷

Model specification

A generalized least squares model with random effects and robust standard errors was estimated, using country–years as the units of analysis. Given our panel data structure, the random effects approach allows us to control for cross-sectional and cross-temporal differences in the level of violent activity. Compared with a fixed effects model, we can accomplish this control without having to introduce a large number of additional variables into the model for each country and year. The degrees of freedom lost through fixed effects would severely damage our ability to draw inferences. Unless noted, the independent variables are all lagged one year to help in assessing the relationship between our independent variables and our dependent variable. This also reflects common practice in the protest literature (Rasler, 1996; Richards and Gelleny, 2006).⁸

Imputed data

Missing data is a serious problem with datasets used to explain state instability and violence. The least stable states are also likely to have the most missing data points. The tradeoff between listwise deletion and multiple imputation is well documented (Honaker and King, 2010; King et al., 2001). Most importantly, if missing data can be systematically predicted, we should expect biases in any model estimation that drops the observations with missing values. For the purpose here, missing data are especially problematic. We are trying to forecast which observations are at the greatest risk of experiencing increases in domestic political violence. If we are missing data on a single variable for high-risk cases, we are unable to make any predictions about political violence for those cases. Therefore, we use the program AMELIA II to conduct multiple imputations (Honaker et al., 2009). We imputed missing values for the following independent variables: ethnic linguistic fractionalization, GDP, electric power consumption, mobile cell subscribers, internet users and military personnel. This allows us to include a total of 155 countries in our baseline models.

Results

Table 1 highlights the results of our statistical models. Column 1 presents the results of a model including just the coercion variables, along with the lagged dependent variable. Column 2 presents the results of the coordination variables and the lagged dependent variable. Similarly, the capacity variables are presented in column 3. Column 4 then includes all of these variables, together with the full control variables.

Table 1.

Predicting domestic violence through 2009, random effects generalized least squares regression; all countries

	1	2	3	4

Variables	Coercion model	Coordination model	Capacity model	Full model
Coercion variables
Low torture_t−1	−5.56799			−13.00912*
	(5.77449)			(7.83633)
Low kill_t−1	−11.77480*			−13.94877
	(6.89110)			(8.62702)
Low disappearances_t−1	−37.09441**			−36.55120**
	(15.76203)			(15.93153)
Low political prisoners_t−1	−11.95904**			−14.05563**
	(5.56284)			(6.01010)
Coordination variables
Freedom of association_t−1		−4.77803		16.12004*
		(10.50516)		(8.58658)
Mobile phone subscribers/100_t−1		0.10127		0.71568**
		(0.16915)		(0.29621)
Internet users/100_t−1		−0.00001		0.00001*
		(0.00001)		(0.00001)
Non-violent protest		4.61407**		5.20230**
		(1.86967)		(2.33325)
Capacity variables
GDP per capita (ln)_t−1			0.99722	−24.41396**
			(5.31230)	(10.76325)
Electric power consumption_t−1			−0.00001	−0.00001*
			(0.00001)	(0.00001)
Military personnel_t−1			0.00031***	0.00013*
			(0.00008)	(0.00007)
Military personnel²_t−1			−0.00001***	−0.00001
			(0.00001)	(0.00001)
Controls
Ethnic linguistic fractionalization_t−1				50.41765
				(31.95562)
Population (ln)_t−1				−40.43182***
				(14.10187)
Polity_t−1				0.67048
				(1.46426)
Coverage (ln)_t−1				25.19235***
				(7.58227)
Lagged DVAR	0.64026***	0.44470***	0.62366***	0.39088***
	(0.03888)	(0.08808)	(0.03953)	(0.08695)
Constant	145.15301***	−5.06962	11.68567	706.36411***
	(41.27587)	(16.69961)	(37.79972)	(270.76844)
Observations	2686	2686	2686	2686
Number of countries	155	155	155	155

Robust standard errors in parentheses.

***

p < 0.01, ** p < 0.05, * p < 0.1.

In line with Hypothesis 1, we find consistently that lower levels of disappearances and political prisoners are associated with lower magnitude of domestic political violence. This holds both without (model 1) and with (model 4) control variables. As hypothesized, these forms of coercion are typically the result of strategic decision by politicians, not the result of bureaucratic discretion, and greater use of these coercive practices is thus more likely to fuel domestic political violence. The evidence for other forms of coercion, such as political killings in model 1, and torture in model 4, is of lesser magnitude and only hold at the p = 0.1 level (two-tailed) of statistical significance. Although there is some support here for other forms of human rights abuses producing violent protest, the types that are most easily connected to government are most consistently and strongly related to violent protest.

The results in Table 1 support our second hypothesis that states where citizens have an easier time coordinating and communicating with each other have a greater potential for violent anti-government activities. In the model without control variables, model 2, we find that more non-violent protest is associated with greater magnitudes of violent anti-government actions. In the fully specified model 4, we find that greater freedom of association (at the p = 0.1 level, two-tailed), mobile phone usage, internet usage (at the p = 0.1 level, two-tailed) and non-violent protest are associated with more domestic political violence.⁹ Overall, the findings on the coordination variables suggest that states allowing easy interaction and mobilization by individuals in a population face greater risks of anti-government violence.

The findings reported in Table 1 also support our third hypothesis, already well supported in the literature, that there is less anti-government violence against highly capable states. With the exception of the military personnel findings, these results are only evident when we control for other possible confounding variables in model 4. To assess the relationship between military personnel and domestic political violence presented in model 3, it is necessary to look at both the constituent and the squared terms that are included in the model to represent the argument made by Gurr (1968). The constituent term is positive and statistically significant in all the specifications, except for the model without a lagged dependent variable. The squared term is negative and statistically significant in the same models. Combined, these results suggests an upside down U-shaped relationship between military personnel and the magnitude of domestic violent protest, with less violent protest at both high and low levels of military personnel and more in the middle. We contend that low state capacity is only captured by this mid-range of military personnel. This finding holds in sign but loses statistical significance in the full specification presented on model 4.

In model 4, the results on the capacity variables GDP per capita (ln) and electric consumption (at the p = 0.1 level, two-tailed) both indicate that, at high levels of state capacity, a population either cannot engage in violent protest or is adequately deterred from doing so. Coupled with the results on coercion, these are especially interesting findings. They suggest that having the capability to control the population through infrastructure (GDP, electric consumption), aid and military personnel can prevent violent protest, but if this capacity or coercive potential is called into action, the aggrieved population is more likely to mount a violent domestic challenge.

As to the control variables, the coefficient of the lagged dependent variable, when included in any model, is positive and statistically significant. The coefficient for the population variable in model 4 is negative and statistically significant, suggesting that states with smaller populations are more prone to domestic violent protest. This is a surprising result in light of past findings on a diverse array of domestic political violence, but could be due to our unique measure that captures the intensity of overall domestic political violence. The coefficient for regime type is also not statistically significant.¹⁰ The measure for ethno-linguistic fractionalization does not attain conventional levels of statistical significance. The coefficient for the coverage variable, the control that we include to account for biases in the amount of media coverage various countries receive, is positive and statistically significant across all models. This should not be surprising: the more media coverage a state receives, the higher its magnitude of domestic violence. It is for just this reason that we include this control variable.

Figure 1 provides the substantive effect of each of our key independent variables that are statistically significant in Table 1, column 4. The x-axis in the figure indicates whether an increase in a variable has a positive or negative effect on the magnitude of domestic political violence and the size of that effect. Each of the variables is varied from its 25% to its 75% in the dataset. They are listed along the y-axis of the figure; the results of the coercion variables are listed first, then the coordination variables, and finally the capacity variables are provided. This figure illustrates the basic findings of the aggrandizing effects of variables associated with coercion and coordination and the dampening effect that state capacity has on domestic political violence. The results show that, while the use of torture and political imprisonment stimulate political violence, when governments stop making their citizens “disappear”, political violence is likely to decrease the most.

Figure 1.

Substantive effects of change from 25 to 75 percentile in the independent variables on expected domestic political violence.

Examination of endogeneity

Since all the independent variables were lagged one year, our results provide some evidence in support of our theoretical framework, which suggests that coercion causes political violence. Still it is necessary to directly address the possibility of an endogenous relationship between the human rights measures and domestic political violence. To do this, we estimate impulse response functions based on a panel vector autoregression program developed by Love and Zicchino (2006). This method uses a generalized methods of moments approach that is “numerically equivalent to equation-by-equation 2SLS” (Love and Zicchino, 2006: 195). The impulse response functions, shown in the online Appendix for both the expanded and restricted datasets, indicate that an orthogonal (exogenous) increase in domestic political violence has no discernible from zero impact on CIRI physical integrity rights. The impulse response functions, however, also indicate that an orthogonal increase in respect for physical integrity has a short-term negative impact on domestic political violence, especially evident in the more restricted coding of the dependent variable. This is consistent with our baseline results and expectations; when focusing on the country–year, human rights abuses increase domestic political violence but domestic political violence has no discernable impact on human rights.

These results, indicating that human rights violations have an exogenous impact on domestic political violence, are also consistent with other recent large-scale cross-national work that assesses whether there is an endogenous relationship between repression and political violence. Carey (2010), for example, finds that only guerilla warfare leads to the onset of repression. Other forms of domestic political violence, like riots and strikes, do not appear to cause repression. Earlier work that did find an endogenous relationship between repression and protest, such as that by Carey (2006) or Moore (1998), differs greatly from our approach in that it is typically very micro-level, capturing a limited number of countries and events at the daily or monthly level. Given that we are looking at a global sample, our data are at the yearly level, and the dependent variable is not a count of one type of protest but the intensity of all types of political violence, perhaps the lack of feedback between domestic political violence and human rights is really not that surprising. Using the above models, we turn to a risk assessment of future increases in domestic political violence in the next section.

Predicting future political violence

Given these results, where should we expect more domestic political violence in the future? The method employed here is borrowed from Gurr and Moore (1997) and was used in Poe et al. (2006). As applied to political violence, when the statistical model presented above identifies a country–year as a negative residual—as having less anti-government violence than one would expect—then we would predict higher levels of political violence in the future. Gurr and Moore (1997) used this method for assessing the future risk of violent conflict between minority groups and their government. Poe et al. (2006) employed this method to assess the risk of higher future levels of human rights violations.

The first step in this process is estimating the above statistical models and generating the residuals from those models. We then examine the residuals to identify all cases where they are negative. This helps in identifying the cases where there is more predicted violence than actual violence. We examine the residuals at time t and examine whether there are increases in violence in the following five years. Similar to the analysis by Gurr and Moore (1997), we have little reason to choose one year as a cutoff point for observing increases in violence. They also look at the following five years. This also reflects practices set out in O’Brien (2002, 2010). What can we actually say about these cases with negative residuals? By definition, these are cases where our model predicts that there should be more violence than there actually is. The interpretation that Gurr and Moore (1997) and Poe et al. (2006) make is that these are cases where the conditions suggest that there should be more of whatever the dependent variable represents in future years (in our case domestic political violence).

This method should tell us which states have the potential for future increases in violent actions by the citizenry. In other words, a negative residual indicates that there is some excess demand for violent protest that we are not observing it in the present year, making it likely that we will see increases in violence in future years. It is also important to note that a negative residual tells us where and when we should expect to see some increase in political violence, not necessarily a big increase. As such, this method is useful in telling us that a country has the characteristics that make future domestic anti-government violence more likely. This is different from predicting that a country will have extremely high levels of domestic anti-government violence in the future.

Risk assessment results

In this section, we use a series of metrics implemented by O’Brien (2002) to assess the quality of both in-sample (1998–2005) and out-of-sample (after 2009) predictions generated from our models. Within the conflict forecasting practitioner literature, O’Brien (2002, 2010) has suggested evaluating the predictive power of a model with three performance metrics: accuracy, recall and precision. Taken directly from O’Brien (2010: 91), these metrics are calculated as follows:

Accuracy = no. of correct predictions/no. of predictions made

Recall = no. of correctly predicted increases/no. of increases occurred

Precision = no. of correctly predicted increases/no. of increases predicted to occur

O’Brien (2010) outlines a forecasting attempt where a guideline of these performance metrics was set at 80% accuracy, 80% recall and 70% precision. It is worth noting, however, that, when the forecasts in O’Brien (2010) focused only on domestic crises, a similar outcome to the low range on our dependent variable of domestic political violence, their recall and precision measures were below 50%.

For the full-specification model presented in Table 1, our in-sample performance metrics are as follows: accuracy is 63%, recall is 57% and precision is 67%. In Table 2, we outline the performance metrics for this model (1), the model with the more restricted data structure (2), these models with lagged non-violent protest (3 and 4), and the restricted model without reliance on a lagged dependent variable (5).

Table 2.

Summary model performance metrics, using O’Brien (2002, 2010)

Model name	Accuracy	Recall	Precision
Expanded data, baseline model	0.63	0.57	0.67
Restricted data and model specification	0.68	0.58	0.68
Expanded data with lagged nonviolent protest model	0.64	0.65	0.69
Restricted data with lagged nonviolent protest model	0.68	0.67	0.71
Restricted data and no lagged DVAR	0.64	0.66	0.65

Our performance metrics are consistent across the different models used, including the model used without reliance on a lagged dependent variable (see O’Brien, 2010). They are also, on the whole, better than the aggregate O’Brien (2010) results. This is especially true given O’Brien’s (2002: 804) statement that “recall and precision scores are generally the most important performance metrics”.

The performance metrics presented in Table 2 are based on a global sample of countries. O’Brien (2010) only focuses on countries in the US Pacific Command. For comparison purposes, as shown in Table 3, we also estimated regional models and calculated performance metrics based on these results, using the full specification expanded data model presented in Table 1, column 4. For these predictions, we estimated the models for each regional subsample and produced the list of states predicted to be at a higher risk of increased violence, and generated the performance metrics. Two things are striking. First, our model is relatively consistent across geographic areas. Actually, when each regional model is estimated separately, the performance metrics increase, while remaining uniform across regions. Second, even when just focusing on the same Pacific Command countries utilized by O’Brien (2010), our results for precision and recall remain superior.

Table 3.

Regional variation in performance metrics, using O’Brien (2002, 2010)

Region	Accuracy	Recall	Precision
O’Brien countries (Pacific Command)	0.66	0.64	0.70
North and South America	0.69	0.63	0.74
Middle East	0.70	0.59	0.75
Africa	0.65	0.66	0.70
Asia	0.68	0.62	0.71
Europe	0.64	0.60	0.65

The out-of-sample results are shown in Table 4. To examine these, we estimate the model presented in Table 1, column 4 up to a given year for which there is existing data that follows that year, then examine whether negative residuals in the last year in the estimated model are followed by increases in violence in the one of the five years following (i.e. in data that were not used in the estimation). For example, to look at whether a negative residual in 1998 provides an accurate risk assessment for the years 1999–2003, we estimate a model that includes all country–years from 1990 to 1998. We estimate models and generate out-of-sample predictions for the years 1998–2005 and produce the three performance metrics for each year. We do not check earlier years because our sample size becomes limited and we avoid later years because there are fewer out-of-sample years that the predictions can be checked against. Since we only have data through 2009, the predictions from 2005 are checked against the following four years.

Table 4.

Out-of-sample model performance metrics, using O’Brien (2002, 2010)

Model year	Accuracy	Recall	Precision
1998	0.82	0.70	0.82
1999	0.81	0.75	0.87
2000	0.78	0.64	0.78
2001	0.75	0.63	0.76
2002	0.70	0.55	0.71
2003	0.73	0.40	0.80
2004	0.64	0.41	0.51
2005	0.55	0.52	0.67

All calculated using baseline model.

On average, the metrics for out-of-sample predictions are either equal to or stronger than the in-sample predictions. In the baseline model, accuracy’s maximum is at 82% and minimum is at 55%. The average across years is 72.5%; this is compared with accuracy of 63% for the baseline in-sample predictions. For recall, the maximum is 75%, with a minimum of 40%, and an average of 57.5%. This is on par with the baseline model in-sample recall of 57%. The maximum statistic for out-of-sample precision is 87%, the minimum is 51% and the mean is 74% across the years. Again, this is on par with in-sample predictions of 67%. We take the results here, that the out-of-sample performance metrics are as strong or stronger than the in-sample predictions, as an encouraging characteristic of this approach.

Table 5 lists the cases that are predicted to have increased violence after 2009, based on the model that includes all years prior to and including 2009. These predictions are based on the model results of the expanded data and restricted data models, both with and without a lagged nonviolent protest variable. The results are organized according to the number of models where we predict an increase. The first column lists the countries we deem at “very high risk”; these are the countries where all four models predicted an increase. The next column lists the countries we deem at “high risk”. We include in this category countries where, if the country was included in all four models, three models predicted an increase or, if the country was only included in the expanded data samples, both of these models predicted an increase. The third column lists the countries we deem at “medium risk” for increases in political violence: half of the models predicted an increase in violence. Finally, the last column lists the countries where there is “some risk”: only one out of the four models predicted an increase. The data are sorted based on the size of the mean residual, so the states with the most pent-up demand for violence are listed at the top. The expectation is that we should observe increases in domestic violence in these states between 2010 and 2014.

Table 5.

Countries where political violence is expected to increase between 2010 and 2014

Very high risk	High risk	Medium risk	Some risk
Russia	Eritrea	Czech Republic	Congo
Sri Lanka	Bahrain	Comoros	Bangladesh
Georgia	Cyprus	Australia	Romania
Iran	Egypt	Mongolia	Angola
Turkey	Bulgaria	Singapore	Guinea
Burundi	Gambia	Ireland	Kuwait
Indonesia	Bhutan	Fiji	Kazakhstan
Chad	Sudan	Myanmar	Ghana
South Africa	Brazil	India	Sierra Leone
Kenya	Macedonia	Belarus	South Korea
Colombia	Latvia	Zambia	Swaziland
Mexico	Mauritius	Nigeria
Israel	Niger	Kyrgyz Republic
Ukraine	Gabon	Nepal
China	Libya	Trinidad and Tobago
Malaysia	Moldova	Armenia
Peru	New Zealand	Venezuela
Italy	Argentina	Cambodia
Democratic Republic of the Congo	Spain	Switzerland
Honduras		Botswana
Chile		Tunisia
Syria		Equatorial Guinea
Jordan		Albania
Ecuador		Austria
Belgium		Slovenia
		Saudi Arabia
		Guyana

Note: we define “very high risk” as countries where the results of all four basic models (models 1–4) predicted an increase in domestic political violence after 2009; “high risk” as countries where three of four models predicted increases, or where only two models were applied (model 3 and 4), two of two models predicted increases; “medium risk” as countries where half of the models predicted an increase but half did not; and “some risk” as countries where only one of the four models predicted an increase. “Arab Spring” 2011 countries are in italics. Countries sorted based on their mean residual, so states with the most pent-up demand for violence are listed at the top. For commentary and 2011–2015 model prediction updates, see http://radicalism.milcord.com/blog

These results have considerable face validity. First, the countries with the highest negative residuals are those that have experienced high levels of domestic political violence in recent years: Iran, Sri Lanka and Russia. Other countries not often thought of as hot-beds of political violence have also experienced recent increases in domestic political violence, including Egypt, Bahrain, Ecuador, Libya, Tunisia, Chad, Belarus and Ireland. In fact, many of the countries involved in the 2011 “Arab Spring” were identified by our approach, as indicated in italics in Table 5. Yemen is conspicuously absent from this list, and is an example of a false negative prediction by our model. Overall, this list of at risk for increased domestic political violence has much face validity.

Conclusion

In this study, we argue and find evidence to suggest that domestic political violence is likely to increase if a government is relatively weak, if it frequently violates human rights that its politicians could easily protect and if it allows citizens to coordinate their actions. Most politicians already recognize the importance of increasing state capacity, and the inverse relationship between state capacity and political violence is well supported in the literature. However, building state capacity is a slow and difficult process. Reducing the use of political imprisonment and disappearances is a rapid and less difficult strategy to reduce the willingness of citizens to engage in political violence. Limiting the opportunity of potential dissidents to coordinate their activities also is a faster and easier way to reduce political violence. These relationships between coercion and coordination and political violence are not well established in the literature.

Our theory suggests that government use of political imprisonment and disappearances creates greater grievances among the citizenry, because these violations are perceived by citizens as resulting from the deliberate policy choices of politicians. Torture and extra-judicial killing, on the other hand, are more likely to be perceived by citizens as resulting from the exercise of administrative discretion by police, prison guards and soldiers. This is the first study to link specific types of physical integrity rights abuse to increases in political violence. Future research exploring the relationship between other types of government human rights violations and political violence may benefit from this insight differentiating types of human rights easily controlled by politicians (e.g. freedom of speech and press and the ability to participate in free and fair elections) from those where politicians are less directly responsible (e.g. respect for worker rights and freedom of religion).

We argue that the ability of citizens to coordinate their anti-government activities also substantially and independently contributes to the likelihood that citizen dissatisfaction will lead to political violence. Our results suggest that the potential for citizen coordination is a necessary condition for political violence. If governments use coercion against their citizens, but citizens cannot communicate easily with one another and cannot form organizations and then assemble to make demands for changes in government policy, then increases in political violence are less likely. These findings are consistent with US counter-insurgency doctrine, but they are the first to demonstrate that respect for freedom of assembly and association and widespread citizen use of advanced communication technology can lead to increased violence against the state. By implication, these findings also suggest that the efforts of relatively repressive governments to block citizen access to the internet and limit cell phone use, while morally repugnant, are likely to be effective.

Methodologically, this work makes several contributions. First, we introduce a new, ratio measure of the intensity of political violence, drawing on earlier work by Murdie and Bhasin (2011). This new measure captures both the frequency and intensity of a wide variety of types of anti-government domestic violence against the state, weighting each event based on the degree to which an event of that type threatens the regime in power. It is a close approximation of our central concept of interest, the overall intensity of violent forms of anti-government behavior within a country in a particular year. Ours is a particularly useful measure if one believes, as we do, that particular types of political violence complement one another in the sense that potentially rebellious groups can substitute one for another depending upon the opportunities available.

Second, using a risk assessment method developed by Gurr and Moore (1997) and later used by Poe et al. (2006), we predict the countries where political violence is most likely to increase in subsequent years. The toughest test for any empirical model is its usefulness in making predictions. Using a risk assessment method developed by Gurr and Moore (1997) and applying O’Brien’s (2002) risk assessment metrics, we presented accurate in-sample “predictions” of political violence and we make out-of-sample predictions to identify the countries where political violence is likely to increase in the next five years. Because all available countries in the world are included in our forecasting model, we are able to make predictions of increased political violence in countries such as Chad and Gabon, which would have been excluded otherwise. The precision, accuracy and recall percentages for our model exceed the current standards for evaluating models of political violence (O’Brien, 2010).

Few previous studies have made an effort to retain cases with missing values on independent variables, developed predictive models, assessed the accuracy of their in-sample predictions or generated and disseminated out-of-sample predictions. These elements of research design comprise the most important empirical innovations of this study. One of the problems in previous research on political violence is that data are often missing for many crucial cases on key independent variables. Governments experiencing political violence or at a high risk of increased violence often fail to report critical statistics and make it difficult for other organizations to collect the information as well. Thus, the missing values on key variables are not randomly distributed across country–years, and the sample of cases with complete information on all independent variables is not representative of the population.

Footnotes

Acknowledgements

The authors wish to thank Colin Barry, Laura Cassani, K. Chad Clay and Richard Frank for comments and suggestions.

Funding

This research was previously presented at the Council on Foreign Relations and the Montesquieu Institute. This work is sponsored by the United States Air Force Research Laboratory under Air Force contract no. FA8750-09-C-0132. The AFRL Program Manager for the effort is Michael Hinman.

Notes

References

Agnew

Gillespie

Gonzalez

Min

(2008) Baghdad nights: Evaluating the US military surge using night light signatures. Environment and Planning 40(10): 2285–2295.

Banks

(2008) Cross-national times series, 1815–2006. Center for Comparative Political Research, State University of New York.

Bartley

(2007) How Foundations shape social movements: The construction of an organizational field and the rise of forest certification. Social Problems 54(3): 229–255.

Batty

(2010) Ecuador: Hope turns sour. The Guardian, 1 October.

Besley

Persson

(2009) The origins of state capacity: Property rights, taxation, and politics. American Economic Review 99(4): 1218–1244.

Bhasin

(2008) Explaining political protest: state institutions & state response. PhD dissertation, Emory University.

Bogoroditskii

(2010) Russia. International Journal of Not-for-Profit Law 12(3): 28–34.

Bond

Jenkins

Taylor

(2003) Integrated data for events analysis (IDEA): An event typology for automated events data development. Journal of Peace Research 40(6): 733.

Boulding

(1962) Conflict and Defense. New York: Harper & Row.

10.

Butler

Gluch

Mitchell

(2007) Security forces and sexual violence: A cross-national analysis of a principal–agent argument. Journal of Peace Research 44 (6): 669–687

11.

Carey

(2006) The dynamic relationship between protest and repression. Political Research Quarterly 59(1): 1–11.

12.

Carey

(2010) The use of repression as response to domestic dissent. Political Studies 58(1): 167–186.

13.

Cingranelli

Richards

(2010) The Cingranelli and Richards (CIRI) Human Rights Data Project. Human Rights Quarterly. 32(2): 401–424.

14.

Clay

(2012) Learning some respect: Foreign events and the international diffusion of human rights practices. Doctoral Dissertation. Binghamton University, SUNY.

15.

Conrad

Moore

(2010) What stops the torture? American Journal of Political Science 54 (2):459–476.

16.

Conrad

Haglund

Moore

(2013) Disaggregating torture allegations: Introducing the ill-treatment and torture (ITT) country–year data. International Studies Perspectives, in press.

17.

Davenport

(2007) State Repression and the Domestic Democratic Peace. New York: Cambridge University Press.

18.

Davidson

MacKinnon

(1993) Estimation and Inference in Econometrics. New York: Oxford University Press.

19.

Elbadawi

Sambanis

(2002) How much war will we see? Explaining the prevalence of civil war. Journal of Conflict Resolution 46(3): 307–334.

20.

Fearon

(2003) Ethnic and cultural diversity by country. Journal of Economic Growth 8(2): 195–222.

21.

Fearon

Laitin

(2003) Ethnicity, insurgency, and civil war. American Political Science Review 97(1): 75–90.

22.

Gartner

Regan

(1996) Threat and repression: The non-linear relationship between government and opposition violence. Journal of Peace Research 33(3): 273–287.

23.

Goldstein

(1992) A conflict–cooperation scale for WEIS data. Journal of Conflict Resolution 36(2): 369–385.

24.

Goldstein

(2007) The role of digital networked technologies in the Ukrainian Orange Revolution. Berkman Center Research Publication no. 2007–14.

25.

Gregory

(2010) Cameras everywhere: Ubiquitous video documentation of human rights, new forms of video advocacy, and considerations of safety, security, dignity and consent. Journal of Human Rights Practice 2(2): 191–207.

26.

Grossman

(2009) Iran protests: Twitter, the medium of the movement. Time Magazine, 17 June.

27.

Gurr

(1968) A causal model of civil strife: A comparative analysis using new indices. American Political Science Review 62(4):1104–1124.

28.

Gurr

Moore

(1997) Ethnopolitical rebellion: A cross sectional analysis of the 1980s with risk assessments for the 1990s. American Journal of Political Science 41(4): 1079–1103.

29.

Haines

(1984) Black radicalization and the funding of civil rights: 1957–1970. Social Problems 32(1): 31–43.

30.

Hendrix

(2010) Measuring state capacity: Theoretical and empirical implications for the study of civil conflict. Journal of Peace Research 47(3): 273–285.

31.

Hoffman

(2004) Human rights and terrorism. Human Rights Quarterly 26(4): 932–955.

32.

Honaker

King

(2010) What to do about missing values in time-series cross-section data. American Journal of Political Science 54(2): 561–581.

33.

Joes

(2004) Resisting Rebellion: The History and Politics of Counter-insurgency. Lexington, KY: University of Kentucky Press.

34.

Kalyvas

(2006) The Logic of Violence in Civil War. Cambridge: Cambridge University Press.

35.

King

Honaker

Joseph

Scheve

(2001) Analyzing incomplete political science data: An alternative algorithm for multiple imputation. American Political Science Review 95(1):49–69.

36.

Love

Zicchino

(2006) Financial development and dynamic investment behavior: Evidence from panel VAR. The Quarterly Review of Economics and Finance 46: 190–210.

37.

Marshall

Gurr

Jaggers

(2010) POLITY™ IV Project Political Regime Characteristics and Transitions, 1800–2009. Dataset Users’ Manual. Available at: http://www.systemicpeace.org/inscr/p4manualv2009.pdf

38.

Mitchell

(2004) Agents of Atrocity: Leaders, Followers, and the Violation of Human Rights in Civil War. New York: Palgrave Macmillan.

39.

Mitchell

(2012) Democracy’s Blameless Leaders: From Dresden to Abu Ghraib, How Leaders Evade Accountability for Abuse, Atrocity, and Killing. New York: NYU Press.

40.

Moore

(1998) Repression and dissent: Substitution, context, and timing. American Journal of Political Science 42(3): 851–873.

41.

Moore

(2000) The repression of dissent a substitution model of government coercion. Journal of Conflict Resolution 44(1): 107–127.

42.

Most

Starr

(1989) Inquiry, Logic, and International Politics. Columbia, SC: University of South Carolina Press.

43.

Murdie

Bhasin

(2011) Aiding and abetting? The impact of human rights INGOs on domestic protest. Journal of Conflict Resolution 55(2): 163–191.

44.

Nam

(2006) What you use matters: Coding protest data. Political Science and Politics 39(2): 281–287.

45.

O’Brien

(2002) Anticipating the good, bad, and the ugly: An early warning approach to conflict and instability analysis. Journal of Conflict Resolution 46(6): 791–811.

46.

O’Brien

(2010) Crisis early warning and decision support: Contemporary approaches and thoughts on future research. International Studies Review 12: 87–104.

47.

Oliver

Myers

(2002) Networks, diffusion, and cycles of collective action. In: Diani

McAdam

(eds) Social Movement Analysis: The Network Perspective. Oxford: Oxford University Press.

48.

Opp

K-D

Ruehl

(1990) Repression, micromobilization and political protest. Social Forces 69: 521–547.

49.

Poe

Tate

(1994) Repression of human rights to personal integrity in the 1980s: A global analysis. American Political Science Review 88(4): 853–872.

50.

Poe

Rost

Carey

(2006) Toward systematic risk and opportunity assessment for physical integrity rights. Journal of Conflict Resolution 50(4): 484–507.

51.

Quinlivan

(1999) Coup-proofing: Its practice and consequences in the Middle East. International Security 24(2): 131–165.

52.

Rasler

(1996) Concessions, repression, and political protest in the Iranian Revolution. American Sociological Review 61: 132–152.

53.

Ray

(2011) The ‘story’ of digital excess in revolutions of the Arab Spring. Journal of Media Practice 12(2): 189–196.

54.

Regan

Henderson

(2002) Democracy, threats and political repression in developing countries: Are democracies internally less violent? Third World Quarterly 23(1): 119–136.

55.

Regan

Norton

(2005) Protest, rebellion, and the onset of civil wars. Journal of Conflict Resolution 49(3): 1–18.

56.

Richards

Gelleny

(2006) Banking crises, collective protest and rebellion. Canadian Journal of Political Science 39(4): 777–801.

57.

Schatzman

(2005) Political challenge in Latin America: Rebellion and collective. Science Review 91(3): 515–529.

58.

Smith

Chatfield

Pagnucco

(1997) Transnational Social Movements and Global Politics: Solidarity Beyond the State. Syracuse, NY: Syracuse University Press.

59.

Tilly

(1973) Does modernization breed revolution? Comparative Politics 5(3): 425–447.

60.

Tilly

(1978) From Mobilization to Revolution. Reading, MA: Addison-Wesley.

61.

Tsutsui

(2004) Global civil society and ethnic social movements in the contemporary world. Sociological Forum 19(1): 63–87.

62.

UN News Center (2010) Ecuador: Recent unrest highlights need for press freedom, accurate reporting – UN expert, 7 October.

63.

Walsh

Piazza

(2010) Why respecting physical integrity rights reduces terrorism. Comparative Political Studies. 43(5): 551–577.

64.

Wintrobe

(2006) Rational Extremism: The Political Economy of Radicalism. Cambridge: Cambridge University Press.

65.

World Bank (2012) World Development Indicators 2012. Washington, DC: World Bank.